Re-definición de procesos de fabricación en convergencia

Re-definición de procesos de fabricaciónen convergencia hacia Industria 4.0

Ped

ro O

rgei

ra C

resp

o . T

ese

de d

outo

ram

ento

. 202

0

Pedro Orgeira Crespo 2


D. Fermín Navarro Medina, como director de la tesis doctoral titulada “Redefinición de

procesos de fabricación en convergencia a Industria 4.0” realizada por Pedro Orgeira

Crespo autoriza la presentación de la misma.

Fecha: 09/10/2020

Firma:


Los siguientes artículos publicados en revistas de investigación han sido incluidos en la

elaboración de esta tesis doctoral:

Orgeira-Crespo, P., Ulloa, C., Núñez, J. M., & Pérez, J. A. (2020). Development of a Transient

Model of a Lightweight, Portable and Flexible Air-Based PV-T Module for UAV Shelter

Hangars. Energies, 13(11), 2889.

Orgeira-Crespo, P., Ulloa, C., Rey-González, G., & Pérez, J. A. (2020). Methodology for

indoor positioning and landing of an unmanned aerial vehicle in a smart manufacturing

plant for light part delivery. Electronics

… con autorización de los coautores, y con un 100% de participación en las tareas de

concepción, investigación, experimentación y escritura de los documentos.


Plaza…, ¡o plomo!

-Pablo Escobar-



Agradecimientos

Aunque suene a tópico, el presente trabajo no es sino el fruto del esfuerzo de muchas

personas que, bien han aportado una parte sustancial de lo que en este documento se

expone, bien han sido unos “posibilitadores”, consiguiendo que el proceso en su conjunto

fuera viable. Es imposible expresar con palabras el agradecimiento debido, pero no menos

cierto es que, cuando menos, hay que intentarlo:

A Pedro, Loli y Raque, por una cuenta en el balance que ni en siete vidas podría saldar.

A Paty, Tibo y Pabochiño, porque fueron el aliento y las fuerzas, el motivo y el consuelo.

A Guillermo y Carlos….y no habría espacio en el documento para incluir todos los motivos.

A José Luis, por no cerrar las puertas a un imposible.

A Fernando, por esa mente tan brillante sobre un corazón tan grande.

A Fran y Adrián, por su inestimable ayuda.

…y en general, a todas las personas con las que me he encontrado en la vida. De todas

aprendí, y todas me dieron algo bueno.


Índice

Agradecimientos .................................................................................................................. 7

Chapter 1. Resumen .................................................................................................... 14

Chapter 2. Summary ................................................................................................... 23

Chapter 3. Introduction .............................................................................................. 30

1. Industry 4.0 ..................................................................................................................... 30

2. Internal logistics within manufacturing plants ......................................................... 32

3. Unmanned aerial vehicles ............................................................................................ 33

3. General overview of the objective ............................................................................... 33

Chapter 4. State of the art ........................................................................................... 37

1. Unmanned aerial vehicles in the logistic context ..................................................... 37

2. Indoor positioning ......................................................................................................... 38

2.1 Introduction .............................................................................................................. 38

2.2 Positioning techniques ............................................................................................ 40

3. Computer vision landing ............................................................................................. 41

4. PV-T as hangar conditioning solution ........................................................................ 42

Chapter 5. Proposal ..................................................................................................... 44

1. Business problem ........................................................................................................... 44

2. Mechanical design of the UAV .................................................................................... 48

3. Standard component selection .................................................................................... 50

4. Mechanical design of carrying case using FEM ........................................................ 51

5. Aerodynamic design of a wing to improve autonomy ............................................ 58

5.1 Introduction .............................................................................................................. 58

5.2 Initial calculations .................................................................................................... 59

5.3 Determination of the airfoil.................................................................................... 60

5.4 3d wing design ......................................................................................................... 65

5.5 Flow simulation ....................................................................................................... 67

5.6 Determination of the autonomy ............................................................................ 69

6. Computer vision positioning system.......................................................................... 70

6.1. Introduction ............................................................................................................. 70

6.2. Avionics ................................................................................................................... 70


6.3. Flight mathematical model ................................................................................... 71

6.4. Positioning ............................................................................................................... 73

6.4.1. Computer vision positioning ............................................................................. 73

6.4.2. Kalman filter ......................................................................................................... 74

6.5. Algorithm ................................................................................................................ 76

7. Landing system using computer vision and fiducial markers ............................... 81

7.1 Introduction .............................................................................................................. 81

7.2 Finding the opening at the corridor to begin descent ........................................ 82

7.3 Finding the short-range markers ........................................................................... 83

9. PV-T as hangar conditioning solution ........................................................................ 86

9.1. Experimental Setup ................................................................................................ 86

9.2. Thermal Model........................................................................................................ 90

9.3. TRNSYS Simulation ............................................................................................... 95

10. Wireless network for message communications ..................................................... 97

Chapter 6. Results ....................................................................................................... 98

1. Wing design.................................................................................................................... 98

1.1 Simulations ............................................................................................................... 98

1.2 Autonomy analysis ............................................................................................... 100

3. Positioning system ...................................................................................................... 102

4. Landing system ............................................................................................................ 104

5. PV-T system .................................................................................................................. 108

Chapter 7. Conclussions ........................................................................................... 114

Chapter 8. Future lines of research ......................................................................... 116


List of figures

Figure 1 Different topologies for node association ...................................................... 40

Figure 2 Overview of the manufacturing plant. ........................................................... 45

Figure 3 Detailed view of the manufacturing plant, showing the zones reserved for

the aerial vehicle. ............................................................................................................... 47

Figure 4 Modified design, with the 14-inch propellers and 595 diagonal structure 49

Figure 5 Cockpit with the bay of the elements needed for flight ............................... 49

Figure 6 Standard 14-inch propeller mounted on the engine ..................................... 50

Figure 7 Assembly showing the legs and the carrying case ....................................... 52

Figure 8 After applying a 30 N load, von Mises analysis show acceptable values . 53

Figure 9 Maximum tension values in the light weighted structure........................... 54

Figure 10 Analysis of the loads of the carrying case .................................................... 54

Figure 11 Deformations for a 1 kg load inside the carrying case ............................... 55

Figure 12 Interior reinforcement ..................................................................................... 55

Figure 13 Carrying case .................................................................................................... 56

Figure 14 The structure containing the legs, 162 g ....................................................... 57

Figure 15 Legs and carrying case, assembled, with the dimensions capable of

carrying a payload up to 1 kg .......................................................................................... 57

Figure 16 Design of the light part delivery aerial vehicle, with its adapted

dimensions, and with the new wing to provide extra autonomy (next section) ..... 58

Figure 17 Comparative study of different airfoils for slow speed flight [118] ......... 59

Figure 18 Coordinate points in the simulation software for studied airfoils ........... 60

Figure 19 Software simulation for different speed ranges .......................................... 60

Figure 20 Behavior of the different airfoils under analysis ......................................... 61

Figure 21 Simulation shows the minimum drag coefficient ....................................... 62

Figure 22 Obtaining the optimum gliding distance ..................................................... 62

Figure 23 Angle of attack obtained in XFLR ................................................................. 63

Figure 24 Maximum angle of attack at which lift is properly generated .................. 63

Figure 25 Airfoil comparison for Re=75,000 .................................................................. 64

Figure 26 Designing the 3d finite wing under XFRL5 ................................................. 65

Figure 27 Comparing wing performances under XFRL5 ............................................ 66

Figure 28 Wake simulation for the selected configuration ......................................... 66

Figure 29 Polar comparison: WACO and 5409 ............................................................. 67

Figure 30 CFD simulation ................................................................................................ 68

Figure 31 The need to keep a constant angle of attach [118] ....................................... 68

Figure 32 Confined flying corridor at the perimeter of the plant (left), and confined

transversal corridor (right) ............................................................................................... 70


transversal corridor (right) ............................................................................................... 71


Figure 34 Computer vision algorithm to obtain UAV’s positioning ......................... 73

Figure 35 ArUco markers deployment in flying corridors ......................................... 77

Figure 36 Navigation workflow ...................................................................................... 78

Figure 37 Navigation strategy. ........................................................................................ 78

Figure 38 Mapping ArUco markers as nodes. .............................................................. 79

Figure 39 Mapping ArUco markers as nodes. .............................................................. 80

Figure 40 Markers on landing table: four 50x50 mm ArUco markers from standard

dictionary (#12, #13, #14 and #15) inside a 260 mm diameter circle ........................... 82

Figure 41 Markers are rotated to use OpenCV functions that provide their top left

corner, calculate de diagonals and the intersection, defining UAV’s goal as landing

destination. ......................................................................................................................... 84

Figure 42 Pnp model procedure [82]. ............................................................................. 85

Figure 43. Detailed manufacturing blueprints for experimental setup..................... 86

Figure 44 Detail on the cooling duct. .............................................................................. 87

Figure 45 Detail on the fans once installed. ................................................................... 87

Figure 46 Manufactured experimental setup ................................................................ 88

Figure 47 Manufactured experimental setup ................................................................ 88

Figure 48. (a) test bench; (b) test bench, fan view......................................................... 89

Figure 49. Test bench with different measuring probes highlighted. ....................... 89

Figure 50. Measurements of collector outlet versus inlet temperature during one

week. .................................................................................................................................... 90

Figure 51. PV-T operating scheme: the system is provided with (cold) tent air and

solar radiation, supplying heated air and electricity. .................................................. 91

Figure 52. PV-T thermal model. ...................................................................................... 93

Figure 53. Block heat transfer scheme for the design of the thermal simulation. .... 95

Figure 54. Thermal model TRNSYS implementation. ................................................. 96

Figure 55 Wireless communication module (ESP8266) for mesh network ............... 97

Figure 56. Simulation of incident flow at nominal speed for theoretical angle of

attack (left), and considering effective angle of attack due to UAV’s pitch (right) . 99

Figure 57. Designed rotatory mount with servomotor to keep ideal angle of attack

of the wing. ....................................................................................................................... 100

Figure 58 Autonomy test ................................................................................................ 101

Figure 59 Autonomy increase under test ..................................................................... 102

Figure 60 Positioning laboratory test ........................................................................... 103

Figure 61 Positioning process: a) Captured frame; b) Greyscale image; c) Binarized;

d) Pose estimation ........................................................................................................... 103

Figure 62 Calculated distance using the vision algorithm (Filtered) versus ultrasonic

measured distance (Sensed) ........................................................................................... 104

Figure 63. Finding landing table from transversal corridor; from left to right: a)

Original image; b) Greyscale modification; c) Binarized ........................................... 105


Figure 64. Bounding box area and detected circle. The ArUco markers are still not

within range, but the circle helps obtaining a reference from long-range distance.

............................................................................................................................................ 105

Figure 65. Descent control while long-range and short-range are available. a) ArUco

markers first detection; top left corners are detected, and the dictionary entries are

identified; the rotation of the markers generates two diagonals whose intersection

indicates landing destination. b) At this flight level, the circle is still visible and

allows confrontation of destination point coordinates between the two methods.

............................................................................................................................................ 106

Figure 66. Discrepancy between center coordinates evaluated via circle versus via

ArUco markers. ................................................................................................................ 106

Figure 67. ArUco markers close range detection. ....................................................... 107

Figure 68. Actual landing spot versus landing goal. ................................................. 108

Figure 69. Experimental actual temperature (Treal) versus simulated (Tsim) during

simulation period. ........................................................................................................... 109

Figure 70. Output’s simulated temperature (Tsim) versus intake temperature (Tin)

in °C. .................................................................................................................................. 110

Figure 71. Actual output temperature (Treal) versus input temperature (Tin) versus

simulated lump mass temperature (°C). ...................................................................... 112

Figure 72. Incident radiation (SI) compared to thermal gains (HG1 and HG2). .... 112


List of tables

Table 1 Drone’s initial weight ......................................................................................... 50

Table 2 Selected parts for previously estimated ones .................................................. 51

Table 3 Updated simulated parameters ........................................................................ 58

Table 4 Obtained lift forces .............................................................................................. 69

Table 5 Obtained drag forces .......................................................................................... 69

Table 6 Expected autonomy ............................................................................................ 69

Table 7 UAV onboarded electronics .............................................................................. 71

Table 8 Instruments used in test bench ......................................................................... 90

Table 9 Lift values experimented at the wing for the original scenario and the

adjusted one. ...................................................................................................................... 99

Table 10 Flight time comparison, without airfoil, and with airfoil ......................... 101

Table 11 Average relative error and RMSE values .................................................... 104

Table 12 Temperature increases during peak hours. ................................................. 111

Table 13 Heating net time .............................................................................................. 111


Chapter 1. Resumen

Las industrias aeronáutica y espacial no han estado tradicionalmente muy

ligadas a los procesos productivos de las industrias de fabricación de otros sectores

ajenos al mundo aeroespacial. Industrias de producción manufacturera como la de

vehículos, artículos electrónicos, o el textil (por citar sólo algunos), no gozan en

general de la presencia de técnicas y tecnologías aeroespaciales dentro de sus

procesos. Sí existe en muchos casos transferencia de tecnología, ya que mucha de la

investigación desarrollada dentro del ámbito aeroespacial es de aplicabilidad a otros

sectores, pero efectivamente no son muchos los casos de uso de utilización de

artefactos aéreos dentro de una fábrica de producción industrial.

El concepto de Industria 4.0, más allá de haberse convertido en una

denominación “de moda”, responde al paradigma de la conjunción de una serie de

tecnologías introducidas dentro de un proceso industrial (los habilitadores) que, bien

ya existían en otras industrias como la Informática (como el Big Data), bien son

nuevos en su uso generalizado (como la impresión 3d). En los últimos años se ha

experimentado una creciente utilización de muchos habilitadores de Industria 4.0 en

diferentes sectores, con una demostrada mejora en indicadores de producción, y la

posibilitación de existencia de nuevos productos.

La presente tesis doctoral pretende ofrecer una visión acerca de cómo es posible

obtener una redefinición de ciertos procesos industriales en una fábrica genérica de

producción manufacturera, bajo la óptica del paradigma de Industria 4.0, y en

convergencia a la cuarta revolución industrial. Se utilizará una fábrica concreta de

reacondicionamiento de ordenadores sobre la que desarrollar el proyecto, que

consiste en la utilización de medios y tecnologías aeroespaciales para proveer los

servicios de aprovisionamiento interno de piezas ligeras, por medio de dron. El caso


de estudio contemplará la solución al problema de negocio para solventar la

necesidad de transportar piezas de ordenador desde un almacén hasta la zona de la

fábrica donde es necesaria, fruto de una incidencia en las operaciones de

reensamblaje de ordenadores. Se resolverán los retos a los que la utilización de este

tipo de solución aeroespacial da lugar, siempre bajo una óptica tecnológica del

paradigma de Industria 4.0

El punto de inicio es, como no podía ser de otra manera, una fábrica de

producción industrial, en la cual se desarrollan procesos de manufactura. En este

caso, se ha optado por una fábrica dedicada a la reparación y reacondicionamiento

de ordenadores (PC y portátiles), que provienen de operaciones de renting en

grandes clientes del sector de la banca. Retornan para recibir una limpieza,

reparación (en su caso), y reacondicionamiento según unas nuevas características,

con la sustitución de ciertas piezas y la adición de otras, de cara a cumplir con los

requisitos explícitos en los pedidos de nuevos clientes, a un precio competitivo

(típicamente en PYMES, cooperativas, etc.) Dicho proceso se organiza por medio de

unos preparadores de material, que preparan pallets conteniendo los ordenadores a

reacondicionar, con todas las piezas que es necesario incorporar de cara a cumplir

las especificaciones solicitadas en los pedidos de cliente. El material se mueve en

pallets utilizando los habituales transpaletas desde el almacén (donde se preparan

los pallets) hasta la mesa de trabajo del operario al que corresponde trabajar sobre

dicha orden de fabricación, según organiza el responsable de producción en cada

momento (quien distribuye las órdenes de fabricación entre los operarios

atendiendo a criterios de plazo, experiencia, y demás).

Cuando no hay ninguna incidencia, el operario trabaja sobre los ordenadores y

piezas que recibe sin tener que moverse de su mesa de trabajo; al finalizar con una

orden de trabajo, continúa trabajando sobre la siguiente orden mientras un operario

de movimiento logístico interno recoge el pallet recién terminado y lo mueve, bien

a su siguiente etapa, bien a la zona de embalaje y expedición, para su entrega a

cliente final. Pero en el proceso productivo no es infrecuente encontrarse con

incidencias, de origen múltiple: ordenadores a los que falta alguna pieza, o que ha

sido sustituida en el proceso de renting por otra, o que tenga algún defecto, o que la

pieza nueva a aportar esté defectuosa, no sea compatible, falte, y otros casos

similares. Las incidencias tienen que ser resueltas, generalmente atendidas por el

jefe de turno, y en general conllevan el traslado de una nueva pieza desde el almacén

hasta la mesa de trabajo del operario. La incidencia suele dar lugar a una parada de

trabajo del operario mientras se gestiona la reposición de la pieza, una búsqueda de

pieza por parte del jefe de turno; en ocasiones, es incluso del propio operario, si

sucede más de una incidencia simultánea, el que se desplaza al almacén para obtener

la pieza con el fin de evitar la espera por el jefe de turno (algo que no es anormal).

En definitiva, una rotura de la continuidad del proceso, que además provoca


retrasos significativos, cuando las dimensiones de la planta son tales que las

distancias recorridas por los operarios son importantes.

Las soluciones de aprovisionamiento de materiales y piezas existentes en la

industria en general pueden no ser óptimas para el caso de aprovisionamiento de

pieza ligera, como es el caso, dado que, con carácter general, suelen tener un tamaño

considerable (lo cual es costoso en términos de aprovechamiento de superficie de

planta), tener proyectos de desarrollo costosos, y costes de mantenimiento no

despreciables. Para casos en los que las piezas a transportar son pesadas, las

tradicionales cintas de movimiento, mesas de aire, o incluso los rodillos, suelen ser

una buena solución; sin embargo, cuando la pieza es ligera (un procesador, una

memoria, una tarjeta gráfica), la inversión, el mantenimiento, y el espacio perdido,

tienen un impacto elevado.

En ese contexto, se buscan alternativas para el traslado de pieza ligera en el

interior de la planta, por un mecanismo que sea autónomo, rápido, de un coste

moderado o bajo, y que tenga un pequeño impacto sobre el layout de la planta de

producción, a nivel suelo; el espacio en la superficie de la planta es un bien preciado

y costoso, de forma que minimizar el impacto de utilización de suelo de fábrica en

cualquier proyecto es interesante y necesario. Es aquí, precisamente, donde tenemos

el punto de encuentro entre la necesidad de negocio en el sector industrial, y las

tecnologías aeroespaciales. Los sistemas aéreos no tripulados (SANT) resultan una

atractiva opción por su rapidez, por sus capacidades autónomas, su coste moderado

(o bajo, en algunos casos), y por su capacidad de utilizar espacios de la fábrica para

su vuelo que habitualmente no tienen uso, o al menos no provocan interferencia con

el trabajo en planta: las alturas perimetrales. Es cierto que un SANT, al que por

deformación coloquial denominaremos dron, no dispone en general de unas altas

capacidades de transporte de peso, e incluso no son muchos los casos de uso en los

que se les ve como transportadores de material; no es menos cierto, en cualquier caso,

que para situaciones en las que la pieza a transportar no supera unos cientos de

gramos, o incluso en otros no llega ni a la centena, resultan una alternativa muy

tentadora.

En ese sentido, pues, se realizará un diseño de la solución de negocio en el que

los sistemas aéreos no tripulados se conviertan en ese sistema autónomo que

transporte la pieza ligera, por el interior de la fábrica, desde el almacén hasta la mesa

del operario que declare una incidencia sobre el ERP. A lo largo del proyecto se van

dando solución a los retos que una propuesta como ésta genera, desde la solución

organizativa a la problemática del posicionamiento.

Cada uno de esos retos es abordado bajo el paraguas del paradigma de Industria

4.0, utilizando tecnologías como visión artificial para la maniobra de aterrizaje, la


hiper conectividad a través de una red inalámbrica mallada preparada para evitar

las interferencias electromagnéticas de un ambiente industrial, la robótica móvil

colaborativa, y la digitalización de los procesos productivos. El espíritu de

transformación de la fábrica de producción hacia una digitalización de procesos es

la guía que lidera las soluciones tecnológicas que se han diseñado para cada reto,

dentro del proyecto conjunto, híbrido entre el mundo productivo industrial, la

digitalización por medio de las tecnologías de la información y comunicaciones, y el

mundo aeroespacial.

El primer reto a abordar es precisamente la redistribución en planta, y el

rediseño del flujo de trabajo para tener en cuenta el aprovisionamiento aéreo. Se

proporcionará una mesa de aterrizaje para que la aeronave de transporte pueda

aterrizar de forma segura, verticalmente, desde un pasillo confinado transversal, y

proporcionar la pieza requerida al operario que la ha solicitado por medio de una

incidencia en el sistema de información.

El segundo reto a abordar ha sido precisamente el diseño del SANT (dron), tanto

desde el punto de vista mecánico, como de selección de componentes. Este diseño

es el fruto de un proceso de varios pasos. Como punto de partida se ha utilizado la

idea de un cuadricóptero, por disponer de una combinación ideal entre velocidad,

maniobrabilidad y peso.

En primer lugar, se ha estimado la masa del dron completo, como suma de la

masa del dron vacío, de la cubeta (o caja de transporte, que alojará la pieza a llevar

desde el almacén hasta el operario), y la masa de la pieza precisamente a transportar.

La masa de los elementos a transportar se consideró como el peor caso de la pieza

con mayor tamaño, y de mayor peso, más dos piezas extra (de peso la media de los

restantes tipos de piezas). Con respecto al chasis, se realizó un diseño minimalista

obteniendo la inspiración de uno comercial, y teniendo en cuenta los tamaños de los

elementos de control y piezas a tener que transportar. Utilizando el software e-calc,

y a través de una serie de iteraciones, se procedió a determinar los elementos

estándar que podrían mantener, bajo un peso del orden de los 2 kg, una

configuración capaz de maximizar el rango de vuelo, con capacidad de transporte

según las especificaciones. Se ha mantenido una relación empuje/peso de al menos

dos, de forma que la aeronave tuviese una cierta agilidad para sus maniobras dentro

del espacio de vuelo.

En segundo lugar, se realizó el diseño mecánico de la cubeta o caja de transporte.

Partiendo de las dimensiones máximas de la pieza más grande a transportar, se


diseñó un mecanismo basado en dos piezas que encajan para proporcionar cubeta

de transporte y patas del dron. El diseño está pensado para ser ejecutado en

impresión 3d, lo cual facilita el reemplazo sencillo en la propia fábrica ante roturas

o deformaciones fruto de potenciales “accidentes” aéreos que pudieran suceder. Por

otra parte, la impresión 3d permite una serie de mejoras, como el tapado de los

huecos de aligerado de material (consiguiendo la reducción de peso en la estructura

reticulada sin deformar el flujo aerodinámico en las patas), y está totalmente

alineado con el paradigma de Industria 4.0. Otro aspecto a destacar es la inclusión

de un refuerzo en la zona central de la estructura de las patas, para soportar

adecuadamente las tensiones fruto del peso del conjunto, y mantener las

deformaciones en un nivel razonable dentro de su límite elástico; se utilizó la

herramienta FEM de SolidWorks para conseguir un diseño óptimo entre una

resistencia razonable para las solicitaciones esperadas, y un peso no excesivo como

para ser aerotransportado. La cubeta fue ensayada desde el punto de vista

aerodinámico bajo SolidWorks Flow Simulation para mejorar el paso del flujo que

inyectan los rotores sobre el par patas-cubeta, incluyendo unas aberturas laterales

para el desalojo de dicho flujo. El ensamblado final de patas y cubeta de transporte

responde a las necesidades de soportar las cargas esperadas, bajo un peso admisible,

y ofreciendo una reducida resistencia aerodinámica.

En tercer lugar, se buscó aumentar la autonomía de la aeronave por medio de la

inclusión de un ala. El objetivo era claro: mantener la maniobrabilidad y las

capacidades de despegue y aterrizaje vertical (proporcionadas por la configuración

de cuadricóptero), pero a la vez obtener las ventajas de una aeronave de ala fija en

cuanto a la sustentación obtenida por el perfil aerodinámico, lo cual redunda en un

aumento del tiempo de vuelo, una mejora de autonomía. Se han estudiado una serie

de perfiles aerodinámicos muy conocidos dentro del mundo del aeromodelismo,

para encontrar uno que sirviese de base para el ala, proporcionando unas

características aerodinámicas óptimas. Se ha utilizado el software de simulación

XFLR5 para, dentro del perfil de vuelo a velocidad nominal, revisar el

comportamiento de cada uno de los perfiles bajo estudio. Por medio de un batch

analysis en XFLR5 Xfoil Direct Analysis, se obtuvieron las polares de los perfiles

Reynolds, revisando las evoluciones de los coeficientes de drag (resistencia) y lift

(sustentación) frente al ángulo de ataque. La idea subyacente era la de obtener los

cuatro puntos básicos de mínima resistencia, máxima esbeltez, mínima velocidad de

descenso, y el ángulo de ataque límite, o de entrada en pérdida. Bajo esa óptica, el

perfil WACO fue seleccionado como base para el ala 3d. Dicha ala fue de nuevo

simulada en XFLR5 utilizando el módulo Wing and Plane Design para determinar la

sustentación esperada en el mejor de los casos, bajo los diferentes perfiles de

velocidades.


Una vez estimada por simulación la sustentación que podría generar el ala, se

procedió a realizar la simulación CFD del conjunto completo utilizando la

herramienta de Solid Works, para estimar por software cuál podría ser la mejora en

sustentación esperada. Los resultados inicialmente fueron inferiores a los esperados

(previamente proporcionados por XFLR5), lo cual se determinó debía a dos motivos.

Por un lado, que en XFRL5 no es viable simular piezas como la cubeta y patas, que

tienen un impacto importante en la reducción de la aerodinámica del conjunto; este

factor ya era esperado, y no dio lugar a cambios de diseño. Por otro lado, se observó,

por medio del Flow Analysis cómo las partículas de aire no evolucionaban en la salida

del perfil como se esperaba, generando una sustentación extra. Se determinó que el

motivo era que el ángulo de ataque efectivo que “veían” las partículas de flujo

entrante era inferior que el calculado, dado el natural cabeceo de la aeronave, debido

a su condición de cuadrimotor. Para solventar esta problemática, ya detectada en

investigaciones anteriores, se procedió a mantener el ángulo de ataque constante por

medio de un montaje rotatorio con servomotor, pilotado por la controladora de

vuelo.

En cuarto lugar, se abordó la problemática del posicionamiento. Como es

conocido, la tradicional señal que suele realizar el guiado de aeronaves de esta

tipología en exteriores (GPS), no es, en general, utilizable en interiores (donde la

señal puede no llegar, o recibirse bajo condiciones inutilizables). Es por ello que fue

necesario diseñar un sistema de localización de la aeronave en entornos de interiores.

Con la idea de reducir costes, además, se ha buscado una solución que no tuviese un

impacto presupuestario alto, recordando además que debía tratarse de una solución

embarcable.

El vuelo de la aeronave se ha previsto dentro de un pasillo confinado, para

cumplir con las expectativas al respecto de este tipo de aeronaves dentro de una

fábrica industrial según los habituales requerimientos de los departamentos de

prevención de riesgos laborales. Así, se ha evitado el sobrevuelo de drones sobre los

operarios utilizando un pasillo de vuelo confinado según el perímetro de la fábrica;

dicha zona se utiliza en general para bandejas de cables, conductos de aspiración,

impulsión, etc., habiendo en general siempre espacio para un pasillo confinado de

las características del diseñado. Por medio de un elemento en “L” girado 90° a

derechas hecho en chapa de aluminio (como los habituales “pantalones” de las

conducciones de aire acondicionado), se consigue confinar el vuelo por la parte

superior y lateral izquierda. La propia pared de la nave industrial proporciona el

límite por el lateral derecho; en la zona inferior, una red con ganchos desmontables

garantiza, por una parte, el hecho de que un fallo en el vuelo no provoque un

accidente; por otra parte, provee un mecanismo para recuperar el dron ante una

pérdida de control, con una relativa accesibilidad. La nave estaría recorrida también

de forma transversal por una serie de pasillos a los que se accede desde diferentes


puntos del pasillo perimetral, y que además disponen de una serie de aberturas en

las verticales de las mesas de aterrizaje de dron para pueda descender y entregar el

material.

El mecanismo de localización se basa en la utilización de un sistema de visión

artificial, por medio de una cámara de bajo coste, una estrategia de navegación

basada en recorrer pasillos perimetral y transversales, y una serie de marcadores. En

las paredes de los pasillos confinados perimetral y transversal se situaron, a

distancias oportunas obtenidas por experimentación, marcadores ArUco impresos,

ubicados en posiciones conocidas y almacenadas en la base de datos de navegación.

El sistema aéreo no tripulado, utilizando una ligera cámara embarcada y un

computador basado en el conocido system-on-chip Raspberry Pi, va “detectando”

los marcadores dentro del plan de vuelo que esperar recorrer entre origen y destino,

como una sucesión de nodos. El ordenador de a bordo, una Raspberry Pi, ejecuta

una variación del algoritmo de Dijkstra para obtener la ruta desde la posición de

espera del dron, al lado del almacén de piezas, hasta la mesa del operario que ha

generado la incidencia. Recorrerá el pasillo perimetral hasta alcanzar la embocadura

que conduce al pasillo transversal que conduce a la mesa del operario, atravesará

dicho pasillo transversal, y al llegar a la vertical sobre la mesa de aterrizaje de destino,

se detendrá, para realizar la maniobra de aterrizaje.

En quinto lugar, se ideó una solución para el aterrizaje, de nuevo bajo el reto de

no poder contar con una señal GPS válida dentro de las instalaciones donde se

produce el vuelo. Además, se tuvo en cuenta que la solución de localización no

podría proporcionar la precisión que se requiere para aterrizar con seguridad en la

mesa provista al efecto al lado de la estación de trabajo del operario de ensamblaje.

De hecho, la idea es que el sistema de localización guiase a la aeronave hasta la

abertura en la red de seguridad del pasillo perimetral que está sobre la mesa de

aterrizaje, de forma que, al detectar la abertura, el sistema de aterrizaje tomase el

control, realizase el ajuste de aproximación fino, y luego iniciase el descenso. Se

decidió utilizar un sistema de visión artificial empleando una cámara de bajo coste,

ligera, pero de una definición y capacidad de integración con el computador más

que suficiente (la cámara oficial de Raspberry). El sistema basaría su guiado en unos

marcadores de referencia que permitirían no sólo detectar la mesa de aterrizaje en

su ubicación precisa, sino también proporcionar una estimación de la pose.

Dadas las importantes distancias a las que los pasillos de vuelo pueden

encontrarse (las naves industriales pueden tener una altura respetable), se detectó

experimentalmente que una solución que garantizase correctamente la precisión

requerida en el punto de aterrizaje, podría no ser detectable con una cámara

económica y ligera a la altura de pasillo transversal; por otra parte, unos marcadores

de referencia que fueran fácilmente y de forma segura detectados desde la altura de


pasillo de velo, quedarían fuera del campo de visión de la cámara a partir de cierta

distancia en el descenso. De esa manera, el sistema completo está compuesto por un

mecanismo de largo alcance (utilizando una marca circular de gran tamaño y

perfectamente visible desde la altura de vuelo), y un sistema de corto alcance

(empleando marcadores ArUco, y un doble sistema de determinación del punto de

aterrizaje). El sistema de largo alcance comenzará a buscar, en cuanto el sistema de

RFID indique cercanía a la zona de abertura, el punto adecuado de descenso. Se

obtendrán fotogramas sobre los que se, por medio de unas transformaciones

morfológicas, se buscará el contorno circular que tiene el marcador de larga distancia.

Una vez localizado, comenzará el descenso, que será controlado por el marcador de

larga distancia mientras se comienza a buscar los marcadores de corto alcance. En

cuanto se detectan los cuatro marcadores, el guiado pasa a estar proporcionado por

los mismos, ya que en breve el tamaño del círculo exterior pasará a dejar de ser

visible. Los marcadores se han rotado para que, detectando sus esquinas superiores

izquierda, se calcule el centro deseado de aterrizaje como la intersección de las dos

diagonales que forman, resolviendo el modelo de cámara pinhole a través del

algoritmo EPnP.

El tercer reto consistía en proporcionar una conectividad a todos los sistemas

previamente citados. La aeronave de transporte no tripulada, por su propia

definición requiere que dichas comunicaciones se produzcan de forma inalámbrica:

la recepción de las órdenes de gestión de incidencia desde el ERP de la compañía, el

envío de la telemetría, estado y localización del dron de vuelta al ERP, y los

comandos de interfaz “hombre-máquina” (HMI, como “dron a la espera de pieza”,

“pieza entregada a dron”, “dron en ruta”, “dron en destino”, “dron a la espera de

que le retiren pieza”, “dron de vuelta”, “dron retornado”), son mensajes que tienen

que ser intercambiados entre el ERP y el SANT.

Se han evitado soluciones tradicionales de Wifi industrial, por ser habitualmente

costosas en ambientes fabriles con naves de tamaño medio y grande. El habitual

ruido electromagnético de la maquinaria presente, la gran cantidad de conducciones

presentes, las distancias que comúnmente superan las distancias permisibles para

UTP, dan lugar en muchas ocasiones a tiradas de fibra, racks intermedios con

bandejas, transductores, GBIC, switches, tiradas de cobre, y un largo etcétera que

suelen elevar los costes radicalmente. Dadas las bajas necesidades de ancho de

banda, el objetivo de controlar los costes, y las habituales configuraciones de puesto

de trabajo en industrias como la presentada, se ha optado por crear una red mesh de

reenvío de mensajes ESP8266 para la comunicación con un nodo router que resultaría

el único elemento cableado, directamente vinculado a la LAN.


El cuarto reto era el de proveer un ambiente adecuado para el hangar de las

aeronaves no tripuladas, por medio de tecnologías verdes. En una instalación

industrial habitual las temperaturas suelen ser a primera hora de la mañana muy

parecidas a las del exterior en la estación fría, dado que los sistemas de aislamiento

no están habitualmente pensados para proporcionar confort. Las temperaturas

extremas son habitualmente un problema para la duración de las baterías, de forma

que intentar obtener una solución no contaminante, sin consumo de energía eléctrica,

y que provocase una mejora en el potencialmente frío ambiente, era un complemento

imprescindible para la viabilidad de la solución. Se ha utilizado un panel PV-T

flexible sobre una estructura a colocar en la cubierta de la nave, de forma que una

pequeña extracción de aire generaba, por convección, una corriente de aire que se

había calentado en el panel; todo el sistema es controlado por un microcontrolador

Arduino para ver las evoluciones de temperaturas interior y exterior, a lo largo de

todo un año.

Los resultados de los experimentos realizados han demostrado la viabilidad de

este tipo de soluciones que, si bien se ha pensado sobre un caso de uso concreto,

podría ser fácilmente sustituida por otra industria en la cual se desarrollasen

procesos productivos con carácter general.


Chapter 2. Summary

The first challenge to be addressed is precisely the redistribution in the plant,

and the redesign of the workflow to take into account air supply. A landing table

will be provided so that the transport aircraft can land safely, vertically, from a

transverse confined aisle, and provide the required part to the operator who has

requested it through an incident in the information system.

The second challenge to be addressed has been precisely the design of the UAV

(drone), both from a mechanical point of view, as well as the selection of components.

This design is the fruit of a multi-step process. The idea of a quadcopter has been

used as a starting point, as it has an ideal combination of speed, maneuverability

and weight.

First, the mass of the complete drone has been estimated, as the sum of the mass

of the empty drone, of the bucket (or transport box, which will house the part to be

carried from the warehouse to the operator), and the mass of the part precisely to


transport. The mass of the items to be transported was considered as the worst case

of the piece with the largest size, and with the greatest weight, plus two extra pieces

(weighing the average of the other types of pieces). With regard to the chassis, a

minimalist design was made, drawing inspiration from a commercial one, and

taking into account the sizes of the control elements and parts to be transported.

Using the e-calc software, and through a series of iterations, we proceeded to

determine the standard elements that could maintain, under a weight of the order

of 2 kg, a configuration capable of maximizing the flight range, with transport

capacity according to specifications. A thrust / weight ratio of at least two was

maintained, so that the aircraft had a certain agility for its maneuvers within the

flight space.

Second, the mechanical design of the transport bucket or box was carried out.

Starting from the maximum dimensions of the largest piece to be transported, a

mechanism was designed based on two pieces that fit together to provide a transport

bucket and legs for the drone. The design is intended to be executed in 3d printing,

which facilitates simple replacement in the factory itself in the event of breakage or

deformation resulting from potential air "accidents" that may occur. On the other

hand, 3d printing allows a series of improvements, such as the covering of the gaps

of lightened material (achieving a reduction in weight in the reticulated structure

without deforming the aerodynamic flow in the legs), and is fully aligned with the

paradigm of Industry 4.0. Another aspect to highlight is the inclusion of a

reinforcement in the central area of the structure of the legs, to adequately withstand

the stresses resulting from the weight of the assembly, and to maintain the

deformations at a reasonable level within its elastic limit; The SolidWorks FEM tool

was used to achieve an optimal design between a reasonable resistance for the

expected loads, and a weight not excessive to be airborne. The bowl was tested from

the aerodynamic point of view under SolidWorks Flow Simulation to improve the

flow path that the rotors inject on the pair of legs-bowl, including some lateral

openings to evacuate said flow. The final assembly of the legs and the transport

bucket responds to the needs of supporting the expected loads, under an admissible

weight, and offering reduced aerodynamic resistance.

Third, it was sought to increase the autonomy of the aircraft through the

inclusion of a wing. The objective was clear: to maintain the maneuverability and

the vertical take-off and landing capabilities (provided by the quadcopter

configuration), but at the same time obtain the advantages of a fixed-wing aircraft

in terms of lift obtained by the aerodynamic profile, which results in an increase in

flight time, an improvement in autonomy. A series of aerodynamic profiles well

known within the world of model aircraft have been studied to find one that would

serve as the basis for the wing, providing optimal aerodynamic characteristics. The

XFLR5 simulation software has been used to review the behavior of each of the


profiles under study within the flight profile at nominal speed. By means of a batch

analysis in XFLR5 Xfoil Direct Analysis, the polars of the Reynolds profiles were

obtained, reviewing the evolutions of the drag (resistance) and lift (lift) coefficients

against the angle of attack. The underlying idea was to obtain the four basic points

of minimum resistance, maximum leanness, minimum descent speed, and the limit

angle of attack, or stall. From this perspective, the WACO profile was selected as the

basis for the 3d wing. This wing was again simulated in XFLR5 using the Wing and

Plane Design module to determine the expected lift in the best of cases, under the

different speed profiles.

Once the lift that the wing could generate had been estimated by simulation, the

CFD simulation of the complete assembly was carried out using the Solid Works

tool, to estimate by software what the expected lift improvement could be. The

results were initially lower than expected (previously provided by XFLR5), which

was determined for two reasons. On the one hand, that in XFRL5 it is not feasible to

simulate parts such as the bowl and legs, which have a significant impact on

reducing the aerodynamics of the assembly; this factor was already expected, and

did not lead to design changes. On the other hand, it was observed, through Flow

Analysis, how the air particles did not evolve at the exit of the profile as expected,

generating extra lift. The reason was determined to be that the effective angle of

attack that the incoming flow particles "saw" was lower than that calculated, given

the natural pitch of the aircraft, due to its four-engine condition. To solve this

problem, already detected in previous investigations, the angle of attack was kept

constant by means of a rotary assembly with a servomotor, piloted by the flight

controller.

Fourth, the positioning problem was addressed. As is known, the traditional

signal that is usually carried out by the guidance of aircraft of this type outdoors

(GPS), is not, in general, usable indoors (where the signal may not reach, or be

received under unusable conditions). That is why it was necessary to design an

aircraft location system in indoor environments. With the idea of reducing costs, in

addition, a solution has been sought that did not have a high budgetary impact, also

remembering that it should be a shippable solution.

The flight of the aircraft has been planned within a confined corridor, to meet

the expectations regarding this type of aircraft within an industrial factory according

to the usual requirements of the occupational risk prevention departments. Thus,

the drone overflight over the operators has been avoided by using a confined

corridor according to the perimeter of the factory; This area is generally used for

cable trays, suction and impulsion ducts, etc., in general, there is always space for a

confined corridor of the design characteristics. By means of an "L" element turned

90 ° to the right made of aluminum sheet (like the usual "pants" of the air


conditioning pipes), it is possible to confine the flight at the top and left side. The

wall of the industrial building itself provides the limit on the right side; in the lower

area, a net with removable hooks guarantees, on the one hand, that a failure in the

flight does not cause an accident; on the other hand, it provides a mechanism to

recover the drone from a loss of control, with relative accessibility. The ship would

also be crossed transversely by a series of corridors accessed from different points of

the perimeter corridor, and which also have a series of openings in the verticals of

the drone landing tables to be able to descend and deliver the material.

The location mechanism is based on the use of an artificial vision system, by

means of a low-cost camera, a navigation strategy based on traveling perimeter and

transverse corridors, and a series of markers. Printed ArUco markers were placed

on the walls of the perimeter and transverse confined corridors, at appropriate

distances obtained by experimentation, located in known positions and stored in the

navigation database. The unmanned aerial system, using a light on-board camera

and a computer based on the well-known Raspberry Pi system-on-chip, “detects”

the markers within the flight plan that it expects to travel between origin and

destination, as a succession of nodes. . The on-board computer, a Raspberry Pi,

executes a variation of the Dijkstra algorithm to obtain the route from the drone's

waiting position, next to the parts warehouse, to the table of the operator who

generated the incident. It will travel the perimeter corridor until it reaches the mouth

that leads to the transverse corridor that leads to the operator's table, it will cross

said transverse corridor, and when it reaches the vertical on the destination landing

table, it will stop, to perform the landing maneuver.

Fifth, a landing solution was devised, again under the challenge of not being

able to have a valid GPS signal within the facilities where the flight takes place. In

addition, it was taken into account that the locating solution could not provide the

precision required to land safely on the table provided for this purpose next to the

assembly operator's workstation. In fact, the idea is that the location system would

guide the aircraft to the opening in the safety net of the perimeter corridor that is on

the landing table, so that, upon detecting the opening, the landing system would

take control. , perform the fine approach adjustment, and then start the descent. It

was decided to use an artificial vision system using a low-cost, lightweight camera,

but with more than enough definition and capacity for integration with the

computer (the official Raspberry camera). The system would base its guidance on

reference markers that would not only detect the landing table at its precise location,

but also provide an estimate of the pose.

Given the important distances at which the flight corridors can be found

(industrial buildings can have a respectable height), it was experimentally detected

that a solution that correctly guarantees the required precision at the landing point


could not be detectable with an inexpensive camera at the transversal corridor

height; on the other hand, reference markers that were easily and safely detected

from the height of the veil corridor, would be out of the field of view of the camera

from a certain distance on the descent. Thus, the complete system is made up of a

long-range mechanism (using a large circular mark that is perfectly visible from

flight height), and a short-range system (using ArUco markers, and a double

determination system landing point). The long-range system will begin to search, as

soon as the RFID system indicates closeness to the opening area, the appropriate

point of descent. Frames will be obtained on which, by means of morphological

transformations, the circular contour that the long-distance marker has will be

sought. Once located, the descent will begin, which will be controlled by the long-

distance marker while searching for the short-range markers. As soon as the four

markers are detected, the guidance becomes provided by them, since shortly the size

of the outer circle will cease to be visible. The markers have been rotated so that, by

detecting their upper left corners, the desired landing center is calculated as the

intersection of the two diagonals they form, solving the pinhole camera model

through the EPnP algorithm.

The third challenge consisted in providing connectivity to all the previously

mentioned systems. The unmanned transport aircraft, by its own definition, requires

that these communications take place wirelessly: the receipt of incident management

orders from the company's ERP, the sending of telemetry, status and location of the

drone back to ERP, and the “man-machine” interface commands (HMI, such as

“drone waiting for part”, “part delivered to drone”, “drone en route”, “drone at

destination”, “drone waiting of having part removed ”,“ drone back ”,“ drone

returned ”), are messages that have to be exchanged between the ERP and the SANT.

Traditional industrial Wi-Fi solutions have been avoided, as they are usually

expensive in manufacturing environments with medium and large warehouses. The

usual electromagnetic noise of the machinery present, the large number of conduits

present, the distances that commonly exceed the permissible distances for UTP,

often lead to fiber runs, intermediate racks with trays, transducers, GBICs, switches,

cable runs. copper, and a long etcetera that tend to raise costs dramatically. Given

the low bandwidth needs, the objective of controlling costs, and the usual

workstation configurations in industries such as the one presented, it has been

decided to create a ESP8266 message forwarding mesh network for communication

with a router node. which would be the only wired element, directly linked to the

LAN.


The fourth challenge was to provide a suitable environment for the unmanned

aircraft hangar, through green technologies. In a common industrial installation,

temperatures are usually very similar to those outside in the cold season in the early

morning, since insulation systems are not usually designed to provide comfort.

Extreme temperatures are usually a problem for the life of batteries, so trying to

obtain a non-polluting solution, without consuming electricity, and causing an

improvement in the potentially cold environment, was an essential complement for

the viability of the solution. A flexible PV-T panel was used on a structure to be

placed on the roof of the ship, so that a small air extraction generated, by convection,

an air current that had been heated in the panel; The entire system is controlled by

an Arduino microcontroller to see the evolution of interior and exterior

temperatures, throughout a year.

The results of the experiments carried out have shown the viability of this type

of solution, which, although it has been thought about a specific use case, could

easily be replaced by another industry in which production processes are developed

in general.



Chapter 3. Introduction

1. Industry 4.0

Industry 4.0 has become a buzzword for at least the last ten years. The term was

initially used by the German government to describe how the processes in the

industry can be orchestrated using certain technologies that interact between each

other autonomously through the whole business workflow. It not only introduces

new technologies that were present in other areas (like Cloud Computing or

BigData, coming from the IT world), but also focuses on the use of more traditional

technologies in a connected industry: everything and everyone within a process is

connected and interacting.

In fact, this revolutionary paradigm has been considered a new industrial

revolution. Every industrial revolution has had its own enabler, catalyzers, or “new

technologies” that introduced a radical change from how products or services were

developed before, and how the same products or services were

developed/manufactured/performed after the introduction of that enabler. At the

beginning of the 19th century, the introduction of the mechanic power loom, and the

steam locomotive generated a flow of changes that were considered the first

industrial revolution; the world moved from the artisan hand-made products to the

development of an industry to manufacture products up to a bigger scale. At the

beginning of the 20th century, the second industrial revolution was promoted by the

use of the electric energy, and the advent of massive production at assembly lines.

By the end of the 20th century, approximately round 1980, a third industrial


revolution was called, influenced by the massive automation in manufacturing

plants, were the automate systems, the electronics, and the introduction of the IT

changed how industrial processes were developed.

Industry 4.0, and its fourth industrial revolution, comes hand in hand with

cyber-physic systems, with the Internet of things, with the extension of the

connectivity of the systems through the hyper connectivity, Bigdata, and Cloud

Computing. The key aspect is to obtain a hybrid industrial process, made of the

physical and the digital words, enabling systems to bind to devices, machines,

operators, products and so on. It is the means to change from an third revolution

plant, to an intelligent industry. There are many examples of how Industry 4.0

impacts the three key areas in industry: the business model, the process, and the

product.

The introduction of the technologies involved in Industry 4.0 enable (and that is

why they are called enablers) the emerge of new business models that were not

possible before since the technology was not ready. Even “traditional” business

models like having a market place on the Internet has also evolved since the dawn

of the e-commerce, in the 90’s: new predictive models based on customer’s data

collection allow big e-commerce sites to perform logistic operations from warehouse

to warehouse based only on the evolution of the shopping trends (based on what

customers that have already bought after buying a specific item will help the

commerce sites to predict what other customers will demand).

As well, the new technologies have a deep impact on the process, now with the

capability of being more flexible, more efficient, or just for begin now possible, like

the revolution that 3d printing is having in many sectors, including the aerospace

one. Manufacturing processes, from design to distribution, including the real

manufacture and logistics, are challenged because of these new enablers as well. The

industry must adapt to digitalization of the design sub-process, using collaborative

method; the smaller time to market implies the combination of flexibility and

efficiency in the plants, with reduced manufacturing slots, using intelligent logistic

solutions, inside and outside plant’s boundaries. Manufacturers perform deep

specialization since interaction is easier due to digitalization; communication and

interaction between business customer and its suppliers, have breached traditional

customs and traditions, thanks to the digitalization capabilities. Everything is

connected, not only in the business to business world, but also in the business to

customer one. Customer’s hyper connectivity is also challenging the industry to

perform better, to execute better, to provide traceability through all the chain. Being

transparent to customers, allowing them to know details on the manufacturing

process, or even interact is now more a need rather than a marketing pose.


Finally, the product itself is experimenting a new life since new functionalities

are being added, especially digital and connected products. Another area is the

extreme customization that products are suffering as a result of customer’s needs,

without increasing cost, and having no undesired effect on quality. The production

lot is dimensioned so as to have better time to market and provide that

customization for the customer. The product is as well to be adapted to a

digitalization context, where traditional products (like clothes, for instance), are

changed to a “digital version” with advanced functionalities (activity sensors, health

sensors, for instance).

2. Internal logistics within manufacturing plants

In the context of the Industry 4.0 change and digitalization initiatives, a key

aspect is how the internal processes related to the material motion with factories are

developed.

From a traditional point of view, internal logistics within a manufacturing plant

are those necessary material transportation operations performed to let materials

and products evolve along the factory workflow. Using different manual or

automatic tools and machines, like conveyor belts, or other advanced systems,

different industries have adopted several solutions to the problem of moving

materials and semi-elaborated products from process A to process B, from one

machine to another. Whether the human was assisting the process, or it was

automatic, the solutions used from the dawn of the industry up to recent times have

usually involved terrestrial devices that carried the materials through their

workflow stages. In many cases, pre-defined paths were reserved for these carrying

devices, where humans were not allowed to interfere. Even very recent solutions,

like the ones deployed in several Amazon warehouses, need specific areas to work

on, where humans are not allowed for safety reasons. In this environment, the

aerospace technologies are a promising asset, where unmanned aerial vehicles can

provide logistic services internally, within a manufacturing plant, providing

automatic means for the delivery. Where a safe flying corridor can be provided, a

properly separated path that does not interact with humans is feasibly; in many

cases, this is possible, since many other conducts are already present in the top of

the plant’s walls (HVAC systems, vacuum suction, electric and communication

trays, as part of a long etcetera). Those systems, very common to many industries,

have usually down pipes for workstations, or specific parts of the plants; these down


pipes may become ascent and descent paths within the flying corridors for

autonomous flying devices to perform their duties.

3. Unmanned aerial vehicles

Unmanned aerial vehicles are very common nowadays, and present in many

areas of our life, although they have been among us for several years now. It may be

considered that the Kettering, at the end for the First World War, could possibly be

the first air vehicle considered as UAV. The plane was to be guided to the objective,

loaded with bombs, so that it would stop flying at the precise moment and hit the

target. It was never used during the war but made its tests in 1917. After that, many

other attempts to improve the technology came. In 1924, Professor Low’s UAV made

a safe flight using a plane with a radio control. During the 30’s, systems were fairly

improved, and during Second World War many radio planes were created, mainly

as target UAV to test weapons and train airmen. During the war at Vietnam had

another push, especially in the area of reconnaissance, with the special case of the

Fireflys, by Northrop; more than three thousands of those air vehicles were using in

the Southern East during war, with a great success in the intel obtained, and the

number of them that came back safely to base. After the Vietnam war, UAV lost

some interest, and it wasn’t until the Aquila development, by Lockheed, that the

interest came back again. The Dessert Storm operation showed the big public what

that type of aircrafts were capable of, and the rest is History and very well known to

the general public.

3. General overview of the objective

Generally speaking, this doctoral dissertation focuses on showing how a process

within a manufacturing plant for a generic industry can be reinterpreted from an

Industry 4.0 point of view, using aerospace technologies. It is a crossroad between

manufacturing plant processes, the digitalization and hyper-connectivity that comes

in hand with the fourth industrial revolution, and the use of aerospace technologies

to improve performance, and take advantage of the Industry 4.0 paradigm.

To show that, a specific manufacturing plant will be chosen, in the computer

refurbishing sector. The actual process to be redefined is the internal logistics, that

can be improved for light parts delivery within the manufacturing work cycle.

Computers coming from renting, mainly in corporate customers, come to the


manufacturing plant to be refurbished, and its configuration altered, for a second

life in different customers, under different hardware configurations. The

refurbishing manufacturing plant performs all cleaning, checking and reconfiguring

to adjust PC and laptops characteristics to those required by their new customers.

Although a computer should travel along the manufacturing plant with all its

required parts, it is very common that for several reasons, some part is missing, or

some new parts should be added, or the ones that were to be added and were carried

together with the PC were faulty or wrong. A quick delivery system is needed to

provide the workstation with the right parts in a short period of time, to prevent the

production workflow to be stopped. The right part, required by the operator

performing the manufacturing process, should be transported as quick as possible

from the warehouse to the workstation where the operator is working. The internal

logistic operation should be autonomous (no human taking part in the material

movement action) and should not interfere with the rest of the operations that

happen in the plant. Traditional conveyors, or similar solutions, are really expensive,

they need periodic maintenance (preventive and corrective, especially after some

years of service), and they take a lot of room in the manufacturing plant (which is

very expensive on its own).

UAV meet all unmanned requirements since they are precisely designed to be

autonomous, are reasonably quick (at least, faster than most of the terrestrial internal

logistic solutions) and are a promising solution for indoor internal logistics for light

parts delivery within a manufacturing plant. Many challenges have to be addressed:

the need of a cuvette or transportation box where the light part has to be while

transported, the need for an indoor positioning system (since inside the

manufacturing plant no GPS signal is expected to be received to guide the UAV), the

need for an automatic landing system (since, again, GPS cannot be used), and the

need to provide air conditioning to the small hangar where the UAV will be resting

until operation is needed. Finally, another key issue involved with UAV (specially

multirotors, that have the VTOL capability required in this type of environments),

is the very low autonomy they have, and this is something to also be addressed, so

that the number of deliveries between battery charge is increased.

The document is organized as follows:

Chapter four provides a vision of the state of the art of the several aspects to be

considered within this project. The UAV will be presented in the logistic context.


The indoor positioning challenge is presented, with traditional positioning

techniques and methods covered in the literature. The computer vision system for

landing is also addressed, covering the techniques already used in previous

research. A PV-T system will be investigated to show how it can provide an efficient

and clean conditioning solution. The means to provide identification for parts and

the operators taking part in the manufacturing process are also revised, as well as

the communication system to provide a message forward mechanism.

Chapter five depicts the proposal, beginning by the description of the business

problem, the proposed solution, the mechanical design of the UAV, the selection of

the standard components (engines, propellers, etcetera), and the design of the

transportation box using FEM to provide a reasonable means to transport the light

part to be delivered. A wing will be proposed, based on the study of several airfoils,

and using CFD simulation. The location system will be designed as well, as a hybrid

solution made of a proper design of the flying corridor and horizontal and vertical

flight profiles, and a computer vision system. The computer vision landing system,

based on fiducial markers is also depicted in this section, using the combination of a

long-range system and a precise short-range one. The PV-T system is also addressed

in this chapter and, finally, the wireless mechanism to forward messages is also

designed.

The last chapters of the document, six seven, and eight, show the results,

discussion, conclusions and future lines of work for this project.



Chapter 4. State of the art

1. Unmanned aerial vehicles in the logistic context

Unmanned aerial vehicles have incremented their usage in the last years,

spreading their presence in many areas in military and civil applications. In the

logistics area, extensive research has been made for package delivery UAV [1],

especially after the Amazon and Google attempts [2], not only for individual drones

performing delivery, but also optimizing and providing intelligence to a delivery

platform [3]. Their sustainability for such usage has also been investigated [4] as well

as its performance [5] and their inevitable issues [6]. The critical success factors for

the implementation of an UAV in general logistics have also been identified, where

the technological advancement and government regulations emerge as the most

influential [7]. There are interesting research on how to obtain a specific delivery

time from the intermediate warehouses, calculating the number of steps from the

main warehouse to the end-customer [8], while other focus on the mathematic

determination of the right paths [9]. Finding the optimal routes that match the best

delivery time considering origin and destination, and the necessary intermediate

points has received intense attention [10]. So far, unmanned aerial vehicles have

demonstrated their value in the logistic arena with notorious use cases for long

distances [11], and it has been studied how this alternative could have a great impact


in cities [12]. Not only in the urban area, but also in the countryside, it has been

shown that the impact on the environment would be lesser than other traditional

solutions for delivery [13]. The scientific literature has demonstrated the capacity of

integration of UAV with lorries for the last step of the delivery [14], [15], proving

their economic viability in some cases [16], especially when the design is adapted

[17]. An interesting approach if the case when the UAV might be used in an urgency

environment [18], especially to deliver medicines in cases of disaster [19], or when

the terrestrial mode is not an option for medicine logistics [20]. For our particular

case, there are few studies where unmanned aerial vehicles are used in the logistic

context within an industry [21].

2. Indoor positioning

2.1 Introduction

As for indoor use cases, extensive research is focused in provide indoor

navigation in these GPS-denied environments; several positioning solutions have

been investigated: UWB [22], Wi-Fi [23], LIDAR [24], computer vision [25] and

mixed solutions using data fusion from different sensors and techniques [26]-[28] .

To the best of our investigations, indoor logistic delivery has not received intense

focus throughout literature.

Getting data from manufacturing plants is a very common and necessary

practice [29], and not only under the Industry 4.0 paradigm [30]. Whether process

depends on physical values (temperature, pressure, flow...), or when plant events

need to be reported for making business decisions, information is gathered. This

process has evolved historically. First, presented where obtained: a simple

thermometer, a PTC/NTC connected to a scale, or wired to a 7-segment display, gets

the information and presents it locally. Then, historically, the information was

distributed: with the 4-20 mA analog signal, with field-buses like CAN or Profibus,

with industrial Ethernet [31]. Historically used SCADA systems not only get data:

standing for supervisory control and data acquisition systems, they can display (by

using sensors) and even control (with actuators) information from the plant, locally,

or even remotely [32]-[34]. When local to the plant, SCADA systems present

information gathered by sensors on HMI [35], and sends to server’s rack through

PLC’s wired through Ethernet [36]. Lately, SCADA has also been taken to the cloud

[37]. The key point about IIoT is that everything (any person, machine, sensor, …) in

the plan is something that can be networked to send data to the information system

(ERP, MES, ….); consequently, a massive amount of information is now available


(total coverage is possible), converging the operational and the information

technology networks [38]. M2M (Machine to machine communication) can

traditionally communicate sensors and machine modules inside industrial elements

present in the plant) [39], [40], and precisely, IoT (as a result of the exponential

growth of wireless networks, [41]), focuses in sensor networks communication

wirelessly, [42], [43] (specially, to the cloud [44]).

The key idea is to always generalize the solution designed, to be able to adopt it

in different plants, and even in other manufacturing or services sectors. The

manufacturing plant should be abstracted as certain inputs (purchases of articles or

raw material), and outputs. In the middle, human and/or machine processing are

applied, as well as any sort of quality inspection and packaging [45]. Physical

variables and events (operator and material identification as well as position) are to

be captured by a IIoT [46]-[49] wireless network, reporting to storage system

(traditional database system, no-SQL, or Bigdata). Through the necessary

firewalling and security, the information may be made world-wide accessible (using

the right access control lists to specific company users), in a private, public or hybrid

cloud. Machine learning algorithms can be used to generate a model that will allow

extracting business information and predict future events (object of a future project).

Information should be abstracted as the values of variables and events that

happen in locations/machines of interest. For the events going on (an operator

identified, an article located in a station, drone’s position), the information would

come directly from sensors and machines reporting relevant actions happening in

the plant instead of being polled. A common representation of information is the

JSON notation [50], [51].

The communication solutions that are based on wires are not very common

nowadays and cannot apply to a use case employing aerial solutions. Besides, this

type of wired solutions are not aligned with the Industry 4.0 paradigm, while

wireless systems help the extension of the internet of things [52], spreading the

wired networks over traditional wired solutions, or data buses. The Industrial

Internet of Things (or IIoT) deploys wireless sensors that report the sensed data

wirelessly, so that they can be spread in manufacturing plant where, when, and in

the number they are needed [53]. In fact, a not-so-new wave of wireless sensors are

spreading widely in several industries, providing not only sensing and sending, but

also some processing capabilities, or even some storage capacity [42].


To create a network of sensors, with the wireless transmission capacity, there

are different topologies that can be used. Every option consider each sensor as a

node, and they differ on the topology, quality of service, jitter, error correction

mechanisms and transmission capabilities [54]:

Bus

TreeMesh

Star Ring

Figure 1 Different topologies for node association

For our particular case, mesh is a very interesting case, for its capabilities of

forwarding data through the node that is closest, along the network. A mesh can be

formed using an all-connected series of nodes, or just having scattered connectivity.

Some of the nodes in the mesh may act just as individual nodes, with forward

capabilities, or routers (that may connect the mesh with other networks, as a LAN).

Mesh networks have the capability of reconfiguring themselves in case some of the

nodes are lost and can rearrange routing in real time. Finally, the message can only

be forwarded one at a time (it is sent just to one node), or to every node in the

network to provide full guarantee that the message is going to arrive its destination

[55].

Every wireless network alternative is defined in its own official standard.

According to different distances, to the topology of the network, to the energy needs

the network needs, and the available bandwidth, there are different alternatives [56].

2.2 Positioning techniques

In previous research [57], widely used technologies that in some contexts can be

used for positioning have been studied. The more relevant, including UWB, ZWave,

Bluetooth, NFC, Wi-Fi and RFID have been reviewed [58]-[63]. Moreover, the

techniques, common to every reviewed technology have also been depicted,

including proximity techniques, the scene analysis option, and triangulation.

Triangulation is relevant for this case since it can be split into lateration and

angulation [64]. In [57], the strategy followed (according to the point of view of the


use of hardware), was a range-based method, using RSSI (received signal strength

indicator) [65]. The key was to obtain distance according to the reduction of the

power received by an antenna, using the Friss equation:

𝑃𝐿 = 𝑃𝑡

𝑃𝑟= (

2𝜋

𝑐)

2

𝑓2𝑑𝑛 (1)

The idea, thus, was to obtain positioning once two or more markers where

visible (which had known positions) [66] using the strength of the received signal,

as opposite to other alternatives as time difference of arrival (TDoA [67]), or angle

of arrival (AoA [68]). In our particular case, given a known a defined by design

height of flight, the 3d positioning problem is simplified to a bidimensional case.

3. Computer vision landing

With regards UAV’s autonomous landing, computer vision has been covered in

the literature through two main approaches:

a) The detection of natural environment (using line features detection of natural

scenes [69] or natural landmarks [70])

b) The use of artificial markers, where an element with a specific imagen pattern

is placed in the landing region to be discovered and provide positioning and

orientation (traditional “H-shape” [71], square-shaped [72], specially ARTag [73],

ApriTag [74] and ArUco [75]-[77]).

Indoors, especially in manufacturing plants, require artificial markers to be

deployed to allow pattern recognition and support landing. In our research, a

conventional camera is selected to create a simple and affordable solution, and the

recognition algorithm is simplified to reduce computing needs, and permit

onboarding the system.

ArUco markers, a synthetic square marker made of a wide black border and an

inner binary matrix that determines its unique identifier within a dictionary, has

been successfully used for object tracking [78] and landing purposes [79].


4. PV-T as hangar conditioning solution

Nowadays, photovoltaic and thermal panels are both used separately in new

construction buildings, and solar electrical power is a consolidated source of energy.

Their combined use is not new, where photovoltaic panels act as thermal absorbents

into a photovoltaic–thermal system (PV-T).

Recent advances in the PV-T field have resulted in the analysis of different

technologies (as flat plane, or those based on solar concentrators), or in the study of

the efficiency of several configurations [80], [81]. Other research investigated new

configurations, with special emphasis on plane and parabolic PV-T collectors [82]-

[86].

Studies on PV-T focused on working fluids are also important, especially those

that make the segmentation on air or water types: according to the nature of the

working fluid, both system configurations and applications are different. Water

cooling PV-T systems are more suitable for thermal energy storage, but their

complexity and fluid leaks risk prove to be a major disadvantage. Air cooling

systems are simpler but have their own drawbacks as well: less efficient heat transfer

and a more difficult thermal storage [83]-[87]; in addition, their combination might

be risky, as shown in [88].

An interesting approach for flat plate is found in [89], where TRNSYS is used to

model that type of solar photovoltaic systems, as in this research. It is very relevant

since it summarizes the applications of PV-T, including their advantages and

limitations.

Research such as [90]-[94] is also significant in that the different energy exchange

models are analyzed, especially those air-type systems relevant for this paper. The

most recent research in this field are focused on developing domestic applications

for this type of systems [95], comparative studies of flat plate PV-T for different

configurations [96], [97], development, new technologies and applications of PV-T

systems [98], [99], and the analysis of PV-T arrangements from the point of view of

exergy [100].

Experimental data is crucial to validate theoretical models for these systems, as

in [101], with an especial focus on PV-T system’s performance [102]-[106]. The

results of the research carried out in [104], [107], [108] are specially related to the


discussion of our analysis, since they validate experimental data with the simulation

of the models they introduce.

PV-T devices are energy multigeneration systems, since they generate two types

of energy at the same time: both electric and thermal. There share certain similarities

with cogeneration systems, where thermal engines are used to generate electricity,

restoring residual heat. As more research has focused on multigeneration, different

validation models have been proposed in the literature [109]-[111].

Most models have been developed under stationary state conditions, as in

Stirling’s engine-based micro-cogeneration system [112], but transient responses are

needed for any HVAC (heating ventilation air conditioning) system simulation,

since they very frequently turned on and off throughout a year.

Although annual system efficiency of a Stirling Engine (SE) facility can be

estimated using permanent state models, dynamical models are more suitable for

these types of simulations, despite the fact that they present certain drawbacks

depending on the type of system under study.

A complex system model definition generally requires a comprehensive analysis

of design parameters, that sometimes are unknown (such as heat exchange surfaces,

heat transfer coefficient, and material characteristics), which implies that such

models have limited results. According to this, a valid approach is the use of a

lumped mass, in order to model a Stirling engine-based micro-cogeneration system:

two lumped (inertial) masses allow decoupling transient thermal model from

system’s geometry [113].

Throughout the scientific literature, different operative SE parameters have been

extensively studied using concentrated mass transient models, and their influence

on engine behavior has been researched [114]-[116]. These grouped parameter

models constitute a valid method to successfully simulate different thermal machine

behavior [117].


Chapter 5. Proposal

1. Business problem

The context for this research uses a manufacturing plant to the address the

challenge of providing an unmanned aerial solution as a light part logistic feeder

inside an industry. The business problem has already been stated in [57], with a plant

within the refurbishing sector; the goal of the plant where the project is to be

delivered is to refurbish computers coming from renting operations, modifying their

configuration to fit new orders from customers.

The layout is split in the following zones: a warehouse (where received

computers and parts are stored until needed), the cleaning area (to remove any

prove of previous use of the computer, or repair any cosmetic damage), the zone

where the pallets with computers and their parts are prepared to feed the

assemblage operators, a zone where computers wait until they can be taken to an

assembly operator (buffer), the area where the assemblers work, the packaging area,

and the zone for deliveries:


Figure 2 Overview of the manufacturing plant.


In the following picture a detailed vision of the plant is shown, where the wait

zone for the UAV can be seen. Next to it, the team of 4 people that prepare the pallets

with the computers to be refurbished and the necessary parts. The big area

corresponds to the assembly zone where operators perform their duties. The flying

area is reserved as a perimetral corridor (under the top of the plant, in its perimeter)

and several transversal corridors, that would take the UAV from the perimetral

corridor to the top of the landing table of the operator that requested help. Every

assembly operator would be equipped with a landing table for the drone to perform

a secure landing and take-off.


Figure 3 Detailed view of the manufacturing plant, showing the zones reserved for

the aerial vehicle.

Focusing in the part of the workflow that needs the aerial solution, the plant has

several workstations where operators perform the assemblage of the computers.

They receive not only the computers to be reassembled, but also the necessary parts.

An incident occurs whenever the production has to be stopped, since the assembly

cannot progress (missing parts in the original computer, wrong or non-working

parts received, etcetera); the idea is that a feeder would attend the incident, the UAV

would fly to its table through the perimetral corridor, and the part was put inside

the carrying case. The drone would then fly through the perimeter of the plant until

the orthogonal corridor was found, to lead to the table for landing, next to the

workstation of the operator with the issue.


2. Mechanical design of the UAV

In previous research [118] an initial design for some specific conditions was

performed, using the eCalc software. The average weight of the elements was

studied, ranging from the lightest part (memory, 10 g), to the heaviest (rotational

disk, 725 g). The total weight must be initially estimated, to provide the software its

boundary conditions. In this case, a worst-case scenario was considered as carrying

a payload of 1 kg, as a combination of the heaviest element and two times the

average weight of the rest of the potential parts.

The reference framework for the initial estimation was a Tarot 650, rounding a

mass of 395 g. The rest of the onboarded systems, as the flight controller (250 g), the

system-on-chip computer (113 g), the camera (3 g), and the ultrasonic sensor (6 g).

As for the rest of the structure (carrying case and drone legs), an initial estimation of

400 g was used. The total weight as the result of the UAV components was 1,167 kg,

plus the extra kilo given by the payload, for a total of 2.167 kg. All this information

is used in eCalc to obtain an estimation of the autonomy, and a proposal for the

standard components.

eCalc was given two extra conditions as well in order to keep a reasonable

performance and flight capabilities: this is, providing thrust up to two times the

weight of the UAV when unloaded (so that it can perform vertical climbs when the

payload is present), and focusing in obtaining the maximum autonomy.

The initial design was consequently modified to cope with these new

requirements, obtaining a 595 mm diagonal frame, provided the necessary room

required by the propellers, now 14-inch ones. A light structure was maintained with

a minimalist design, to provide the minimum possible drag, while keeping a

compact structure.


Figure 4 Modified design, with the 14-inch propellers and 595 diagonal structure

Inside the box, in the center of the UAV, an aerodynamic design of a cockpit

contains the necessary parts that will be depicted later, including the flight

controller, the system-on-chip computer, etc.

Figure 5 Cockpit with the bay of the elements needed for flight

Engine and propeller are standard, as seen in the next section:


Figure 6 Standard 14-inch propeller mounted on the engine

3. Standard component selection

The e-calc software was used to obtain a valid configuration for the propulsive

elements, once the initial weight was estimated. The process of selecting a

combination of ESC, engines and propellers was performed several times to find a

valid configuration that would adjust to requirements, and provided the best flight

time, with the maximum autonomy. The initial weight estimation was updated from

the previous research [118], as shown on the following table:

Element Part Weight (g)

Flight controller Pixhawk 4 250 g

Onboard computer Raspberry Pi 4 113 g

Positioning electronics Five Maxbotix MB1232

I2CXL 6 g

Camera Raspberry Pi Camera 3g

Payload transport basket and

legs

Ad-hoc design 400 g

Chassis Ad-hoc design 395 g

Motors - 340 g

ESCs - 130 g

Propellers - 72 g

Battery - 860 g

Table 1 Drone’s initial weight


Total estimated mass for the UAV is 2,167 g, where 863 g are per design, and the

rest estimated. As for the payload, the worst-case scenario was considered, that was

carrying the heaviest load of 1 kg (heaviest part plus two times the average weight

of the rest of the elements), to the far end of the plant. Final achieved configuration

is shown on the following table:

Element Part Weight (g)

Motors 4 Turnigy Aerodrive SK3 3548 700 KV 310 g

ESCs 4 BLHeli S 35A 131

Propellers Multirotor Carbon 14x4.7 CW/CCW 78 g

Battery 18650 Li-ion 6000mAh 14.8V 4S3P 600 g

Table 2 Selected parts for previously estimated ones

4. Mechanical design of carrying case using FEM

The mechanical design of the carrying case to hold the payload is an evolution

of a previous design [119], that had to be adapted to the new conditions: increase of

the potential payload up to 1 kg, as a result of carrying several heavy parts and not

just one. It is a two-pieces part that contains the legs for the UAV and the carrying

case itself.

The smooth aerodynamic design has been kept reducing the drag, while keeping

the structural strength so as not only to carry the payload, but also to withstand the

weight of the flying vehicle.


Figure 7 Assembly showing the legs and the carrying case

The dimensions to carry the payload in the carrying case were 32x294x165 mm,

for an initial payload of 1 kg, after analyzing the different dimensions of the

designed elements. The criteria of maintain as low profile in dimensions as possible

was kept, so that drag was as reduced as possible (and the bigger the structure, the

heavier, also unwanted); moreover, the shape should be as regular as possible, to

keep the right balance during flight; finally, the structure should withstand the

maximum load (UAV plus its payload), so a 30 N force was applied for simulation:


Figure 8 After applying a 30 N load, von Mises analysis show acceptable values

The most critical tension obtained rounds 7.26 MPa, corresponding to the areas

where the tension is higher, shown in red areas in the figure. This maximum tension

values are way below the maximum stress of the ABS plastic material used to

manufacture the structure, using a 3d printer.

In the following picture we may also see that the holes created in the structure

to lighten it, also have their maximum tension values within acceptable values,

according to the plastic used to manufacture it:


Figure 9 Maximum tension values in the light weighted structure

With the loads we are considering now, it is necessary to strengthen the interior

of the carrying case, according to the loads it is withstanding:

Figure 10 Analysis of the loads of the carrying case


After analyzing the tensions, it can be seen that the problem to address is the

deformation generated by the payload, that goes beyond 12 mm, and it is not

acceptable, as seen on the next figure:

Figure 11 Deformations for a 1 kg load inside the carrying case

The solution is, again, applying a reinforcement with a ‘v’ shape that reduces

the deformation by using the areas where the tension has lesser values. In this case,

an expected value of 3.1 mm is obtained, for the worst-case scenario of 1 kg of

payload.

Figure 12 Interior reinforcement


The final mass of the carrying case with the dimensions that can hold the worst

case of the three elements, counting up to 1 kg, is finally 192 g.

Figure 13 Carrying case

Once we add the structure holding the legs of the air vehicle, we find that the

total weight of the structure containing the legs, and the carrying case, totalize 354

g.


Figure 14 The structure containing the legs, 162 g

Figure 15 Legs and carrying case, assembled, with the dimensions capable of

carrying a payload up to 1 kg

We may find a photorealistic representation of the cockpit containing the

electronics, the legs, the carrying case, standard components, and a wing to increase

the autonomy.


Figure 16 Design of the light part delivery aerial vehicle, with its adapted

dimensions, and with the new wing to provide extra autonomy (next section)

As in previous research [118], a mount with rotating capability was added to

maintain the proper AoA (angle of attack). It will be described in the next section.

5. Aerodynamic design of a wing to improve autonomy

5.1 Introduction

Once the mechanical design has been updated from the previous configuration

[57], and the standard components have been changed to increase its carrying

capability up to 1 kg, the whole air vehicle is simulated again on eCalc to determine

its new parameters, as seen on the following table:

Parameter Value Unit

Total weight (fully loaded) 3.286 kg

Push/weight ratio 2.2 []

Hovering time 11 minutes

Table 3 Updated simulated parameters

With that in mind, the idea was to increase the autonomy of the UAV, while

keeping the vertical take-off and landing that the quad rotor provides.


5.2 Initial calculations

From previous research [118], we had a preliminary design of a wing based on

a NACA airfoil for the 750 g payload, and a flight speed of 10 m/s. In this case, the

analysis has to be repeated to address, especially, the new speed, adapted to the

computer vision positioning algorithm.

Again, the XFLR5 software was used to determine the drag and lift coefficients,

as well as their relationship with the angle of attack. Moreover, the airfoils used for

the initial comparison were the same GOE, NACA and WACO Cootie, very common

in slow speed aero models.

Figure 17 Comparative study of different airfoils for slow speed flight [118]

To generate the new polar graphs, in this case a chord of 0.23 m was initially

assumed, as per previous experiences. Determining again the Reynolds number

using:

𝑅𝑒 = 𝜌𝑣𝑠𝐷

𝜇 (2)


with a speed interval that is within the [0;5] m/s for this case, to allow the computer

vision system performing its recognition operations for positioning.

Consequently, in this case the range of Reynolds numbers is between 40,000

and 90,000, with an AoA ranging between -1° and 9°.

5.3 Determination of the airfoil

The airfoils were again inserted in the simulation software with all the points

that define the shapes:

Figure 18 Coordinate points in the simulation software for studied airfoils

A batch analysis for the different Reynolds numbers was run:

Figure 19 Software simulation for different speed ranges


With this simulation, the information to compare the behavior of the different

airfoils for the new conditions is obtained, and represented as in the following

picture:

Figure 20 Behavior of the different airfoils under analysis

Thanks to XFLR5, the lift and drag coefficients with respect to the AoA are

obtained, and the polar curves as well, CL over CD. As in previous research [118] the

four representative points for every airfoil can be obtained (minimum resistance,

maximum finesse, minimum descent speed, and the stall point).

In this case, the minimum resistance point, related to the drag coefficient, is

found at an angle of attack of 3°.


CD versus AoA

Figure 21 Simulation shows the minimum drag coefficient

As for the maximum finesse location, we may find it in the following figure. It

represents the maximum flight distance in the case of a fixed wing.

CD/CD versus AoA

Figure 22 Obtaining the optimum gliding distance

The third key element for the analysis is the minimum descent speed, also

simulated in XFRL5. As in previous research, the angle of attack at which this point


is found, using the well-known 𝐶𝐿3/2

/𝐶𝐷 formula is bigger than the maximum

finesse point:

(CL)3/2/CD versus AoA

100

120

80

60

40

20

Figure 23 Angle of attack obtained in XFLR

Finally, we found the stall point, very well known for being the point from

which no more lift is generated although the angle of attack is increased. Generated

lift from that point on is smaller than the drag generated.

CD versus AoA

2.00

2.50

1.50

1.00

0.50

0.00

Figure 24 Maximum angle of attack at which lift is properly generated


Consequently, finding the airfoil that, for the selected nominal speed, provides

a better behavior is the key criterion that will be used to choose the wing that

increases autonomy. High lift values and small drag values is the searched

combination.

In this case, the nominal speed of 5 m/s is considered, that corresponds to a

Reynolds number of 75,000. For that speed, the WACO Cootie is the one that

performs better in the terms and conditions stated before. For this new

configuration,

Figure 25 Airfoil comparison for Re=75,000

In this case, the WACO provides a better performance because of the slow speed.

Although in previous research [118] it was not the option since it is not a normalized

airfoil, in this case it was the selection, based on the much better performance that

were showing at small and relative big angles of attack, near 9°.

Indeed, a 9°angle of attack was chosen as showing the best performance for the

selected airfoil, and it will be the one that will be desired to be kept, independently

of the pitch of the aerial vehicle.


5.4 3d wing design

The design of the wing begun by the selection of the right airfoil for the designed

flight speed. In previous sections, the wing was considered as infinite, as airfoil

analysis estimates. In this section, the effects that happen when the wing is

considered with a finite span are taken into consideration.

As in previous research [118] an XFLR5 module called Wing and Plane Design

was utilized, now injecting the real dimensions of the designed wing:

Figure 26 Designing the 3d finite wing under XFRL5

The span is, according to the extra lift needed to carry a heavier load, designed

to be 1.2 m, that for the chord of 23 mm, yields an aspect ratio of 5.2, no sweep, no

twist, according to the low speeds.

In this case, the performance of the wing with the airfoil selected in the previous

section was compared to the wing in the previous research [118], again using XFLR5,

for the 5 m/s, using a Fixed Speed simulation type:


Figure 27 Comparing wing performances under XFRL5

Lift distribution and wake are very similar to the ones previously found:

Figure 28 Wake simulation for the selected configuration

The polar comparison was performed between the previous wing (NACA), and

the wing with the selected airfoil (Cootie), finding a better performance for the

chosen airfoil.


Figure 29 Polar comparison: WACO and 5409

In this case, for the designed speed of 5 m/s, we obtain lift forces of up to 13 N

at an angle of attack near 9°.

5.5 Flow simulation

A CFD simulation was performed with all parts assembled together, to find out

about performance in flight. Moreover, predicted lift values obtained at the wing

were compared to the simulated ones for the wing, without the chassis and cockpit.

They were finally compared to ones obtained in previous research [118].

The nominal speed of 5 m/s was speed as per design incident flow under

SolidWorks Flow Simulation:


Outlet velocity5 m/s

Figure 30 CFD simulation

As in previous cases [118], lift values were under expected ones, since air

particles do not follow the flow as designed: the angle of attack that the incident flow

receives is different from the one the airfoil has, because of the pitch of the drone.

Figure 31 The need to keep a constant angle of attach [118]

The lift force value obtained for each of the components at the nominal speed is

displayed in the following table:


Lift force (N)

Speed

(m/s)

Airfoil Carrying case + legs Total

5 12.92 -1.7 11.22

Table 4 Obtained lift forces

The drag force created at the wing and the carrying case is shown in the

following table:

Drag force (N)

Speed

(m/s)

Airfoil Carrying case + legs Total

5 -3.51 -0.62 -4.13

Table 5 Obtained drag forces

5.6 Determination of the autonomy

Once the lift forces created at the wing were obtained via flow simulation, they

were injected on the eCalc software again to simulate again the expected autonomy,

according to the new configuration, weight, size and flight speed.

In the following table we may observe the estimated flight time for the 5 m/s

speed, according to the worst-case scenario of a payload of 1 kg.

Flight time (minutes)

Speed (m/s) Flight time

5 16.6

Table 6 Expected autonomy


6. Computer vision positioning system

6.1. Introduction

The flight must be accomplished through a security flight corridor that is in the

perimeter of the plant and in transversal corridors, for operator’s safety. It consists

of an aluminum “L” (1.5 meters width, 0.15 meters tall) attached to the wall of the

plant (perimetral corridor) or hanging from plant’s deck (transversal), as displayed

on the following figure:


transversal corridor (right)

Under the “L” confined passage, a protection net is issued to prevent any

accident in case there is a problem with the UAV’s flight. The net is attached to the

“L” and to the wall (in the perimetral corridor), or to the grid structure hanging from

the plant’s roof (in the transversal case), having openings in the vertical of the

landing tables in both cases.

6.2. Avionics

The UAV is onboarded with the following electronics to provide positioning and

flight control, as shown on the following table:

Description Item

Camera 2 x Raspberry Pi NoIR v2 (8MP)


Flight controller Pixhawk 4

Computer Raspberry Pi 4 8GB

Table 7 UAV onboarded electronics

Two cameras provide positioning: one in the forward direction (for when the

UAV is flying through the corridors), and another pointing downwards (for take-off

and landing procedures). Selected camera is official Raspberry Pi NoIR v2 camera,

attached to the Raspberry Pi via its dedicated CSi interface, with a total weight under

5g, and supporting up to 1080p30 resolutions; it provides the frames used to

calculate distance and estimate the pose of the UAV. Flight controller is well-known

Pixhawk 4, connected via its serial connection with the on-boarded computer. The

Raspberry Pi system on chip provides the computational capabilities to analyze the

frames provided by the cameras and satisfies the communication with the flight

controller.

6.3. Flight mathematical model

Quadcopter can be considered as a rigid body to analyze flight dynamics. The

problem has been researched in the literature [120], [121], using the four rotors to

control the six degrees of freedom, and described by a static inertial reference frame

(ground), plus a dynamic reference frame (body), as shown in the following figure:


transversal corridor (right)


The attitude can be represented as vector 𝛥 = (𝑅𝐺 , 𝜃𝐺) = (𝑋, 𝑌, 𝑍, 𝜑, 𝜃, 𝜓) ,

where 𝜑, 𝜃, 𝜓 are the rotations around the ground reference system (yaw, pitch,

roll). The static and the dynamic reference systems are related by a translation

followed by a rotation using the Euler angles, yielding to:

𝑅𝜃

= [

cos 𝜓 cos 𝜃 − sin 𝜓 cos 𝜑 + cos 𝜓 sin 𝜑 sin 𝜑 sin 𝜓 sin 𝜑 + cos 𝜓 sin 𝜃 cos 𝜑sin 𝜓 cos 𝜃 cos 𝜓 cos 𝜑 + sin 𝜓 sin 𝜃 sin 𝜑 − cos 𝜓 sin 𝜑 + sin 𝜓 sin 𝜃 cos 𝜑

−sin 𝜃 cos 𝜃 sin 𝜑 cos 𝜃 cos 𝜑]

(3)

Representing the linear velocity as 𝑈𝐵 = (𝑢, 𝑣, 𝑤) and the angular velocity as

𝑊𝐵 = (𝑝, 𝑞, 𝑟), both in the dynamic frame, yields to:

�̇�𝐺 = 𝑅𝜃 𝑈𝐵 (4)

�̇�𝐺 = 𝑅𝜃 𝑊𝐵 (5)

Considering F1, F2, F3 and F4 the thrust forces generated by the propellers

perpendicular to them, the force vector with respect the body reference system can

be written as 𝐹 = (0 0 −𝐹1 − 𝐹2−𝐹3 − 𝐹4 ); adding the thrust generated by rotors

and gravity force 𝐹𝑔𝑟𝑎𝑣 = (0 0 𝑚𝑔), we finally get [122], [123]:

�̇� = 𝑝 + (𝑞. sin 𝜑 + 𝑟. cos 𝜑) tan 𝜃 (6)

�̇� = 𝑞. cos 𝜑 − 𝑟. sin 𝜑 (7)

�̇� = (𝑞. sin 𝜑 + 𝑟. cos 𝜑). sec 𝜃 (8)

𝑝 = �̇� − �̇�. sin 𝜃̇ (9)

𝑞 = �̇�. cos 𝜑 + �̇�. cos 𝜃 . sin 𝜑 (10)

𝑟 = �̇�. cos 𝜑 . cos 𝜃 − �̇�. sin 𝜑

(11)


6.4. Positioning

ArUco markers provided positioning for the UAV, using computer vision to

detect distance and to generate pose estimation as well. There are two different

scenarios: first, when the UAV is flying through a corridor; second, when the UAV

is landing. In both scenarios, the same algorithm and filtering was used for detection,

distance calculation and pose estimation; slight differences, depicted in the

following sections, address the differences between the two cases.

6.4.1. Computer vision positioning

The general algorithm followed is depicted on the following picture:

Figure 34 Computer vision algorithm to obtain UAV’s positioning

A traditional loopback algorithm is used, where the reference is provided by the

node navigation algorithm, that finds the path to UAV’s destination, as indicated

below. The reference is combined with pose estimation given by the computer vision

algorithm, and then injected in a PID to act on UAV’s rotors. Once the camera

provides a frame, it is adapted to obtain distance and pose, according to [124]. The

first step is transforming the image to grayscale (color does not provide useful

information according to selected marker nature); selected algorithm is Luminance,

as described in [125]. Next step is contour extraction, where the outline of the object

to be detected (the marker) is obtained to separate it from its background; this step

is divided in two: first, a Canny edge detector is issued [126], and the Suzuki and

Abe algorithm is performed [127]. Afterwards, only the outermost rectangle contour

is left as a polygon made of four vertexes, using the polygonal approximation given

by [128]. After that, four steps are performed to obtain the identification code from


the detected rectangular candidates: first, perspective distortion is eliminated (to

provide a perpendicular image of the potentially detected marker), calculating the

homography of the frame; second, the proper threshold value to perform the

partition of the digital image into different segments is determined and applied, by

using the method originally proposed by Otsu [129]; third, the transformed image is

split into a grid so that each point can be characterized with a “0” or “1” value;

fourth, resulting binary matrix is compared to the dictionary containing the subset

of markers used for this project, to evaluate which id corresponds to the candidates

found. When more than one marker is visible, the first one that has been detected is

selected for pose estimation, while the other will be used when the active one is not

visible. Next step uses nonlinear optimization for pose estimation according to

Levenberg-Marquardt algorithm [130]; the result is the obtention of the rotation and

translation vector to the marker. Finally, upper-left corner is detected by using the

method initially proposed by Tomasi and Kanade and refined by Tomasi and Shi

[131], setting the maximum number of corners according to their likelihood.

6.4.2. Kalman filter

By using previous algorithm, we may find not only the location of the UAV

according to the known positions of the markers, but also estimate its pose.

Nevertheless, the camera may suffer vibrations due to flight, and consequently it

might take some time to detect some markers. A Kalman filter is added to predict

the future state of the system by using previous states and new measured values. It

is a two-step estimation problem: first, a prediction on the state of the system is

performed, according to previous position and error estimate; second, new

measurements help updating error estimation for future predictions. For every step

at the process the state is predicted by applying:

�̂��̅� = 𝐴 . 𝑥𝑘−1 + 𝐵 . 𝑢𝑘 (12)

where �̂�𝑘 is the prediction of the state at step time k, A is the transition matrix of the

process, 𝑥𝑘−1 denotes state at previous step, B transforms 𝑢𝑘 control vector into

state space, and 𝑢𝑘 is the actual control vector [132].

The projection of the covariance of the estimated error is obtained by adding two

terms: the first contains the matrix representing the covariance of the error at step

time k (𝑃𝑘−1), and the second is the covariance of the process noise (Q) [133]:


𝑃�̅� = 𝐴 . 𝑃𝑘−1. 𝐴𝑇 + 𝑄 (13)

For this case, images are bidimensional, and hence state vector has four

components (x and y coordinates of the marker center, and dx and dy, velocities in

both dimensions). A and B matrixes can be chosen, for easiness [134]:

𝐴 = [

1 0 1 00 1 0 10 0 1 00 0 0 1

] (14)

𝐵 = [1 1 0 0]

(15)

and the process noise covariance matrix Q is selected upon previous experiences,

showing the accuracy of the method, as:

𝑄 = [

0.2 0 0 00 0.2 0 00 0 0.2 00 0 0 0.2

] (16)

Once a prediction is performed, it is updated by new observation to update the

prediction. Measurement update is achieved by updating variance and state using a

combination of the predicted state and the observation [135]:

𝐾𝑘 = 𝑃�̅� . 𝐻𝑇 . (𝐻. 𝑃�̅� . 𝐻𝑇 + 𝑅𝑘)−1 (17)

having measurement matrix H, that translates state space to measurement space; in

our case, measurements are consistent with predictions, and thus:

𝐻 = [1 0 0 00 1 0 0

] (18)

In (16), as well, 𝑅𝑘 stands for the measurement noise covariance matrix,

representing the uncertainty of the measurement process; the following was initially

chosen:


𝑅𝑘 = [0.1 00 0.1

] (19)

And finally, we may update the new estimated value, and update error

covariance:

�̂�𝑘 = �̂��̅� + 𝐾𝑘 (𝑧𝑘 − 𝐻 . �̂��̅�) (20)

𝑃𝑘 = (𝐼 − 𝐾𝑘 . 𝐻) . 𝑃�̅� (21)

The new state vector is calculated using previous estimated one, plus a value

affected by Kalman filter gain, 𝐾𝑘, multiplying 𝑧𝑘 (calculated coordinate values of

the center of the marker, according to our algorithm) minus previously estimated

state vector times the measurement matrix. Error covariance is updated with respect

previous value using again Kalman filter gain and previous covariance (here, I is the

identity matrix).

6.5. Algorithm

Positioning while flying through perimetral and transversal corridors is

achieved using the localization of the markers; the AruCo markers were deployed

as seen on the next figure, separated 1 meter, after checking visibility from different

distances during previous tests.


Figure 35 ArUco markers deployment in flying corridors

The fiducial marker is hanging from a rod attached to the wall. Dictionary used

is original mip_36h12, using A4 printed markers. According to previous tests,

marker separation, speed and camera angle enable at least one ArUco to be detected

from UAV’s camera at every time; a second marker is detected before first one

becomes invisible (because of the angle). Markers’ coordinates are consequently

known so that distance and angle with respect them can provide real time location

of the UAV. The general strategy for flight inside the plant is based on abstracting

corridors and workstation landing tables (destination point) as nodes, performing

navigation as shown on the following figures:


Figure 36 Navigation workflow

Figure 37 Navigation strategy.

To perform UAV’s navigation safely inside manufacturing plants, confined

flying corridors with net protections are proposed. For the UAV to get to its

destination carrying the load, it will climb up vertically to the flying altitude, and

traverse the perimetral corridor (step 1). When it finds the transversal corridor

corresponding to the landing table of the operator that raised the incident, it will

perform a turn (step 2). Next steps are finding destination, perform a vertical land,

wait until the operator confirms reception, climb again to cruise altitude, and finally

get back to its wait zone (4 and 5). The UAV will keep a centered flight inside the

confined corridor, detecting markers during its course and deciding directions

according to tags layout, stored on its Raspberry Pi computer. Every marker is

defined as a node, to map their arrangement in memory as shown on the following

figure:


Figure 38 Mapping ArUco markers as nodes.


Once a command is received to fly to a specific destination table, the path to

reach it is calculated; a vertical take-off is performed until flight cruising altitude is

achieved; then, the flying corridor is traversed, by means of the markers to confirm

the progression of the UAV. The sequence of the markers will also indicate the UAV

when to reduce speed and finally stop progression at destination’s vertical; landing

procedure is executed using computer vision, payload is fetched by workstation

operator, and a vertical take-off is done again until flight cruising altitude, for the

UAV to return back it is wait zone.

To allow navigation, the locations of very landing table and marker are stored

on the memory of the UAV, as depicted on next figure:

Figure 39 Mapping ArUco markers as nodes.

General flight strategy is based on traversing the perimeter until the transversal

corridor corresponding to the destination point is found, and then perform ninety

degrees turn to continue flying until the landing table of the operator who needs the

part is reached. The algorithm is based on following the sequence of nodes through

the perimetral corridor, where the nodes are materialized as fiducial markers; once

found the marker at the crossing between perimetral and transversal corridors, a


decision must be taken about turning or not according to UAV’s final destination.

To obtain the shortest path to a specific goal we follow the Dijkstra’s algorithm [136].

7. Landing system using computer vision and fiducial markers

7.1 Introduction

Once the UAV reaches the opening area in the flying corridor, marked by its

tags, it will activate the camera to begin recognition. Proposed markers on landing

table include a circle for long-range recognition and a set of four ArUco marker for

short range, as depicted on the following picture:


Figure 40 Markers on landing table: four 50x50 mm ArUco markers from standard

dictionary (#12, #13, #14 and #15) inside a 260 mm diameter circle

The dimensions of the circle allow its visibility from the flying corridor, and the

long-range part of the descent; when the UAV approaches the landing table, the

circle is not visible, and the ArUco markers provide control at this stage. The

following steps are defined for landing.

7.2 Finding the opening at the corridor to begin descent

At this step at transversal corridor, first requirement is to determine the point

where descent must begin. While progressing through the corridor, sonar is not

operative in the forward direction and RFID suffers from certain error. Fiducial

markers in the landing pad of the workstation table, vertically aligned with descent

point will be looked for and used to determine where UAV should begin descent;

consequently, we need to control the x axis. The OpenCV library is used to alter the

image from RGB to greyscale using discrete values from 0 to 255, and then convert

it to a black and white binary image. It is morphologically transformed then to

eliminate image noise and extract the contours and silhouettes of the shapes found

in an opening operation (erosion and dilation processes).


Next steps are aimed to determine whether detected elements in the image

correspond to a circle, as described in [137]: aspect ratio (AR) and solidity (SL)

should have values similar to 1, and extension similar to 𝜋4⁄ . The aspect ratio is the

quotient between the width and the height of the shape; the solidity is the quotient

between the areas of the contour and the convex hull (the convex perimeter around

the points of the circle); finally, the extension refers to the quotient between the areas

of the contour and the bounding box around it. When the circle shape is confirmed,

the center is obtained, and the distance to it (x,y) is determined; given that left and

right sonars are keeping the UAV centered in the corridor, a value of 𝑥 = 0 distance

will determine descent point. We must transform the distance in pixels to meters,

using the pinhole camera model [138]:

𝑑𝑥 = 𝐾 · 𝑥 = 𝐷

𝑑· 𝑥 (22)

where D is the diameter of the real circle and d is the diameter in pixels; dx is the

distance to the descent point.

7.3 Finding the short-range markers

The UAV performs descent controlling the z variable with sonar until gets out

of the flying corridor; when the ArUco markers become visible, they will be used for

short-range descent, where the size of the outer circle gets out of sight; after some

tests, the sizes of the markers and the circle are chosen to have constant positioning.

Whilst one ArUco marker can provide pose estimation for the UAV [139], in this

research four are used to: a) use a fusion algorithm that conveniently combines the

information from them; b) provide another mechanism to obtain desired landing

point, as the intersection of the two diagonals that form the top-left squares of the

markers, as shown in the following figure:


Figure 41 Markers are rotated to use OpenCV functions that provide their top left

corner, calculate de diagonals and the intersection, defining UAV’s goal as landing

destination.

In this case, the image is transformed to a greyscale, detection of its contour is

performed, and a polygonal approximation is made [140], resulting in every item to

have a binary value due to the threshold limit. Marker’s unique identifier confirms

that it is one of the four selected ArUco items, and consequently recognition is

confirmed.

Next step is to solve the well-known PnP problem (pinhole camera model), to

describe the projection of a point from the 3d world coordinate system to a 2d

imagen model [141]:


Figure 42 Pnp model procedure [82].

There are several solutions in the literature for the PnP problem, and the Efficient

PnP (EPnP) [142] was selected for its efficiency that allows performing the algorithm

with four ArUco markers at a time. A linear system is generated as:

𝑀 𝑥 = 0 (23)

where x is the transposed vector of unknows and M is the matrix that combines

camera intrinsic calibration matrix, the 2d projections of the reference points, the

scalar projective parameters, the 3d coordinates of the n control points; the method

simplifies the complex problem by expressing the 3d points as a weighted sum of

four virtual control points [142].

Since four markers are used, fusion estimation is performed to combine the four

pose estimations into a more accurate one. The problem is within a general

multisensor linearly weighted estimation fusion case, that extends the Gauss-

Markov estimation to the random parameter under estimation [143]; in our case,

given that mj are the pose estimations (for j in the 1-4 interval) and r’ is the unbiased

estimate for m as:

𝑟′ = 𝑤1𝑚1 + 𝑤2𝑚2 + 𝑤3𝑚3 + 𝑤4𝑚4

(24)


where wj represent the weights to be calculated. Using the Lagrange multipliers, the

variance is minimized when [144]:

𝑤𝑗 =

1𝑉𝑎𝑟(𝑚𝑗)

∑1

𝑉𝑎𝑟(𝑝𝑖)4𝑗=1

(25)

8. PV-T as hangar conditioning solution

8.1. Experimental Setup

To accomplish the validation of the theoretical model for the PV-T based on a

flexible thin layer with experimental data, a workbench was implemented on the

roof of a building. It was designed using SolidWorks, and manufactured for this

project, as seen in the next figure:

Figure 43. Detailed manufacturing blueprints for experimental setup.

It is made of a rollable Powerfilm R21 panel (fiberglass substrate), with a textile

envelope wrapping a metallic structure, as developed for previous research [111]. A

150-mm long insulated cooling duct was connected on one side to extract heat; it is


shown in the following figure, and its diameter is 60 mm. Fluid inside the PV-T

increases temperature because of panel heat that cannot be extracted through the

other surfaces (that are adiabatic).

Figure 44 Detail on the cooling duct.

Air is extracted with the aid of two 140 mm diameter NF-A14 fans (next figure)

that can provide a maximum flow of 182.5 m3/h per unit; they operate under PWM,

so that flowrate can be controlled, up to a maximum of 8.2 g/s and 4.3 Watts. Next

figure shows the fans integrated in the test bed:

Figure 45 Detail on the fans once installed.

The experimental setup was manufactured and then tested in the laboratory.

The following figure shows the test bed with the PV-T panel:


Figure 46 Manufactured experimental setup

And the next figure shows the data monitor station that was getting recorded

information:

Figure 47 Manufactured experimental setup

A specific mounting base was designed with a slope angle equal to test location’s

latitude (42.4° N), with clamping plates for the different probes, as seen on figure 68.

PV’s backsheet cooling duct is completely isolated from the environment and

includes safety screens in both input and output that provide protection against Sun

radiation effect on the sensors. The test bed includes a data acquisition system,

specifically designed as a flexible, reliable and scalable solution to record and

transmit remotely the data captured through its sensors. The core of the acquisition

system is an Arduino board, chosen since it is a low-cost open source well known to

authors, already used on similar cases [145]-[148]. Next figure shows test bed

deployed in the top of a building:


(a) (b)

Figure 48. (a) test bench; (b) test bench, fan view.

The next figure shows measuring point locations in the PV-T panel. For the

thermal model, input (Tin) and output (Tout) refrigeration air temperatures, and airflow speed (FR) are logged. Environmental conditions are also logged, inasmuch as they are also relevant; wind speed is obtained using a meteorological station, and the incident radiation on PV-T panel (SI) is measured on workbench.

Figure 49. Test bench with different measuring probes highlighted.

This setup has been collecting data since January 2019, having an insight on a

whole year of information (8760 h). On that dataset, a process aimed at purging outliers was carried out; moreover, we focused on the coldest days (worst case scenario), specifically eight days without erroneous or incomplete data. Recorded data are input and output air temperature, that can be found in the following figure, where average hourly data obtained from two digital thermometer IC are displayed over that period of time (the raw data is captured every ten seconds and it is averaged every hour to be displayed in the figure):


Figure 50. Measurements of collector outlet versus inlet temperature for one

week.

The following table shows the parameters obtained on the workbench, the

sensors used for every measurement, the measurement range, the type of output

data, and the required power voltage. Instruments and actuators were integrated on

the Arduino system by using different input mechanisms, as shown in the table. To

analyze the different airflow conditions in the panel, the speed of the fans that push

the air in the cooling duct are controlled by two industrial fans having flow

regulation capability: Noctua NF-A14, regulated by a digital PWM signal that the

Arduino system provides, for each fan. Since the dynamic of the plant is relatively

“slow”, logging frequency is not critical, having thus reduced the sensing period to

every 10 seconds.

Parameter Number

of Sensors

Sensor Range I/O Type Power

Supply

Temperature 2 DS18B20 Thermometer −55–125 °C Digital

(One-Wire)

5 VDC

Flow rate 1 Bosch HFM 5 Air-mass

meter

8–370 kg/h Analog

(0–5 V)

8–17 VDC

Solar

irradiation

1 Kipp & Zonen SMP10

Pyranometer

0–1600

W/m2

Analog

(4–20 mA)

5–30 VDC

Table 8 Instruments used in test bench

8.2. Thermal Model


For the purpose of the validation of this heating system based on flexible PV-T on any facility, a thermal model is designed, as shown in the following figure. As may be seen, the air used to warm the small facility is the one from thermal system output.

To estimate system output temperature, a heat exchange model is proposed. The

panel surface is being heated by Sun’s radiation, that is logged through our instrumentation system described in the previous section. Obviously, output variables will depend on system location and incident radiation angle. Moreover, panel losses (physically due to radiation and convection) are computed and classified in internal losses (the one from panel surface to the cooling duct, due to convection), and external losses (due to radiation and convection); internal losses are responsible for air heating.

Figure 51. PV-T operating scheme: the system is provided with (cold) tent air and solar radiation, supplying heated air and electricity.

The proposed thermal model for the system will be represented as a lumped

mass in Trnsys, as shown in figure 71. This solution simplifies the heat transfer issue without accuracy loss. The main ideas behind lumped mass are: first, to assume that panel temperature is unique, constant through the whole panel; second, transfer coefficients are also constant and temperature independent. Consequently, we assume that photovoltaic panel temperature is constant through the panel surface and will vary according to its thermal capacity (calculated as the product of multiplying the mass per its specific heat).

PV-T panel absorbs solar radiation and sets up a heat transfer with the

environment at the same time. The system reaches a temperature that allows a convection towards passing fluid (air) passing through the system, provided that the temperature gradient allows heat transfer: energy losses suffered by lumped

Room temperature

ElectricityHeated air


mass to the system will equal passing fluid absorbed energy, having only to perform

an energetic balance to reach the desired output temperature.

Next equation depicts heat transfer rate, that will be equaled to convection losses

as shown in this section. This is the total heating power received by the air that

passes through PV-T pipe:

𝑄[̇ J s⁄ ] = �̇�𝑐𝑝∆𝑇 (26)

In this equation, �̇� is air’s mass flow in cooling duct expressed in [kg/s], 𝑐𝑝 is

specific heat of air expressed in [J kg · K⁄ ] , and ∆𝑇 is the gradient temperature

between input and output expressed in [K].

Thermal energy undergoes different changes in its process from being radiated

heat to becoming a means to condition the temperature of the facilities. First,

irradiation “𝐺” is defined as the quantity of incident radiation per unit of surface

and time. The ratio of total irradiation absorbed by panel surface is obtained by the

following equation:

𝛼𝐺 (27)

where 𝛼 is solar absorptivity, and 𝐺 is incident irradiation.

As shown in figure 74, the part of the losses named “Losses 1” is divided in two

parts: radiation losses and convection losses. As for exchanged heat, it can be defined

according to the following equation:

�̇�𝑟𝑎𝑑[Wm2⁄ ] = ε𝜎(𝑇𝑠

4 − 𝑇𝑠𝑘𝑦4) (28)

where ɛ is panel emissivity, 𝜎 Stefan-Boltzmann’s constant ( 5.67 ∙

10−8 W m2 ∙ K4⁄ ),𝑇𝑠 the temperature in K of the surface in touch with the fluid, and

𝑇𝑠𝑘𝑦 surrounding surface temperature in K. The latter is estimated through

environment temperature 𝑇𝑎𝑚𝑏 according to Swinbank’s proposed model [46]:

𝑇𝑠𝑘𝑦 = 0.0552 𝑇𝑎𝑚𝑏1,5 (29)

Not only panel surface temperature but also internal and external temperatures

are computed by using the same parameter, 𝑇𝑠, since we are modeling PV-T panel


as lumped mass. Moreover, convection losses to the outside can be defined using

the following equation:

�̇�𝑐𝑜𝑛𝑣[W m2⁄ ] = ℎ(𝑇𝑠 − 𝑇𝑎𝑚𝑏) (30)

where ℎ is convection heat transfer coefficient expressed in W

m2·K, that will be

estimated using recorded wind speed (in m/s), according to the classic model by

Nusselt-Jürges [149]:

ℎ = 5.8 + 3.95 𝑉𝑤 (31)

As a result, heat flow perpendicular to panel surface can be written according to

the following equation.

−𝑘𝑑𝑇𝑠

𝑑𝑥= ℎ(𝑇𝑠 − 𝑇𝑎𝑚𝑏) + 𝜀𝜎(𝑇𝑠

4 − 𝑇𝑠𝑘𝑦4) − 𝛼𝐺 (32)

Once losses to the outside have been defined, heat transfer from panel surface

to the inside has to be identified. Being again a convection loss, it will follow the

same equation 19, only instead of using 𝑇𝑎𝑚𝑏, the temperature to be used is the one

the fluid is achieving along the cooling duct of the system. To perform this task an

integration through the whole length of the pipe is necessary, since the air is going

to increase temperature on its way. That gain results in a convectional power loss as

the fluid progresses, the higher the temperature gradient, the faster the heat

exchange; the gain is getting smaller, anyway, since inertial system temperature is

the same regardless of its location, while that of air is increasing.

Figure 52. PV-T thermal model.

Incident radiationLosses 1

Losses 2

Air flow Output


Consequently, to simplify calculations, an average temperature for the fluid in

the pipe will be used. That assumption is valid since this heating system does not

experiment great temperature gradients, and it will be calculated in the following

section in a series of iterations (it is dependent on fluid’s temperature at system’s

output, that is in fact unknown, and the value we are seeking).

Moreover, in this section the idea of using an energetic balance to obtain

𝑇𝑜𝑢𝑡 was introduced. Up to this point, we already know that convectional energy

losses through the panel corresponds to the energy that air flow is obtaining, so by

using equations 15 and 19 we may calculate the balance:

ℎ𝑖𝑛𝐴𝑃𝑉(𝑇𝑙 − 𝑇𝑚𝑎) = �̇�𝑐𝑝(𝑇𝑜𝑢𝑡 − 𝑇𝑎𝑚𝑏) (33)

where 𝐴𝑃𝑉 is panel area, 𝑇𝑙 lump mass temperature, and 𝑇𝑚𝑎 is the average

temperature of the fluid inside. The temperature at the surface of the lump mass 𝑇𝑠

is the temperature of the lump mass 𝑇𝑙, as per its definition, as shown in equation

22.

𝑇𝑠 = 𝑇𝑙 (34)

The reason for defining in the same manner received radiation and panel’s losses

to the outside is that both directly affect the temperature that the lump mass

achieves, so output temperature depends on both, as will demonstrated in this

section.

As for the simulation, it will follow the layout shown in figure 76, where the

equations and parameters for lump mass temperature 𝑇𝑙 are depicted and the

relationship between that temperature and output’s temperature get into a circular

argument. 𝑇𝑜𝑢𝑡, thus, will be obtained as result of a series of iterations, and will be

the temperature of the facility we are conditioning.


Parameters

Equation 1

Equation 2

Lump

Heat gain 1

Heat gain 2

Lump’s Temperature

Lump’s Temperature

Output’s Temperature

Q[J/s] = mcpΔT. .

Figure 53. Block heat transfer scheme for the design of the thermal simulation.

Obviously, this heating system is only useful as long as the output temperature (of facilities intake temperature) is higher than the temperature inside the facility. According to this, a sensor must be installed to compare both temperatures, activating a controller that stops air flow when heating system is really cooling it, instead of providing heat.

8.3. TRNSYS Simulation

Lumped systems are those where body temperature varies over time, but in which that body temperature is also constant throughout its whole volume. The PV-T panel is not a true lumped system but, due to its slenderness, it can be modelled as such. Once this hypothesis has been formulated, PV-T thermal capacitance is then defined as the product of multiplying its mass by its specific heat.

The proposed model was implemented on TRNSYS, using the lump mass

(Type963) as the core simulation concept. The great unknown when it comes to implement lump is the definition of its capacitance. That capacitance will be obtained by minimizing the gap (or error) between simulated data and the dataset obtained after selecting an average week. To solve this, a series of iterations are performed taking the capacitance as a changing variable and simulating iteratively


until the minimum possible error is achieved. To do so, GenOpt software is used in

conjunction with TRNSYS.

Previously shown heat transfer equations for the model are inserted as TRNSYS

Equations in that software. For each module, energetic balances are computed. By

applying collected climatological data to the model, the PV-T system output

temperature is obtained.

TRNSYS implemented model can be seen in the following figure:

Figure 54. Thermal model TRNSYS implementation.

Heat gain 1 (HG1) computes energetical balance in the external surface,

according to Equation 21. Heat gain 2 (HG2) performs energetic balance calculations

according to the first term of Equation 22. The signs of both equations have to be

changed, since for the lump mass energetic balance is not to be considered as a loss,

but a gain.

The main parameters used to perform simulation are solar absorptivity and

panel emissivity, both fixed to 0.7. Experimental values applied to the model are

obtained for an air volumetric flow of 8.2 gr/s , pumped by the three fans the

workbench has at the entrance of the pipe. Selected meteorological data are for the

second week of January in Marin, Pontevedra.


9. Wireless network for message communications

In this research, a wireless solution has been designed to provide the ERP system

the capability of sending commands to the UAV (to attend incidents raised by the

assembly operators), or for the aerial vehicle to send telemetry back to the system

(speed, positioning, alerts, etc.) Since standard industrial wireless solutions are quite

expensive as the manufacturing plant dimensions grow, a specific solution to obtain

information without wires was developed.

The key concept was to use an inexpensive network mesh of ESP8266 boards in

the assembly operator’s workstations. They average distance is relatively small,

inside its ranges, and there is direct line of sight between them. Moreover, a specific

protocol focused on the IoT (MQQT, or message queuing telemetry transport) was

used in the application layer, so that messages were forwarded from node to node.

Figure 55 Wireless communication module (ESP8266) for mesh network


Chapter 6. Results

1. Wing design

1.1 Simulations

The whole set was then simulated to study the interaction of the individual parts

once mounted. It was observed that the lift generated by the wing was under

expected values, explained primarily for two reasons: first, the interaction of the

flow suctioned by the propellers, that “steal” air particles from the incident flow;

second, the effective incident angle of attack on the wing, due to the pitch that the

UAV experiments in its flight.

The scenarios were simulated again: first, for the ideal angle of attack (AoA),

that is effectively reduced by the pitch experimented at nominal flight speed; second,

increasing the angle of the attack of the wing according to the pitch of the UAV, to

maintain the AoA in its designed value. Both scenarios are displayed on the

following figure:


Figure 56. Simulation of incident flow at nominal speed for theoretical angle of

attack (left), and considering effective angle of attack due to UAV’s pitch (right)

Next table displays lift values generated at nominal speed by the wing at both

scenarios, showing the importance of keeping the ideal angle of attack of the wing

with respect incident flow, independently of the pitch of the UAV:

Scenario Lift (N)

Original (theoretical) -2.60

Adjusted 13.41

Table 9 Lift values experimented at the wing for the original scenario and the adjusted

one.

Keeping the wing in its optimum angle of attack was achieved by a rotatory

mount (as seen on the figure with an anchor point for a servomotor, connected to

one of the PWM outputs of the flight controller; according to the pitch angle detected

by the IMU, the servomotor adjusts wing’s angle of attack to keep it in optimal

position:


Figure 57. Designed rotatory mount with servomotor to keep ideal angle of attack

of the wing.

1.2 Autonomy analysis

Some flight tests were performed outside the laboratory to determine the

influence of the wing on the autonomy.

A circuit in a football field was prepared, with a perimeter of approximately 308

meters, that was done 60-62 seconds, for an average nominal speed of 5 m/s


Figure 58 Autonomy test

The process was performed with five different battery packs under the

configuration with airfoil, or without it, and the values obtained are displayed in the

following table:

Test Flight time no wing

(minutes)

Flight time with

wing (minutes)

%

1 12:23 14:09 14.3

2 11:46 14:01 19.1

3 13:11 15:24 16.8

4 11:58 13:32 13.1

5 12:39 15:36 23.3

Average 17.3

Table 10 Flight time comparison, without airfoil, and with airfoil

In every case there was an increase of the autonomy, that in average drops a

17.3%, quite reasonable.


Figure 59 Autonomy increase under test

3. Positioning system

The UAV was provided with three sensors to perform laboratory test flights in

order to check positioning: two Maxbotix MB1242 I2CXL-EZ4 were installed

pointing forwards (longitudinal, in the direction of travel) and downwards

respectively, and one Ultrasonic HC-SR04 was installed pointing to the wall

(transversal), on the right side of the UAV; the three sensors are connected to the

Raspberry Pi using I2C (MB1242) and available GPIO pins (HC-SR04). The following

figure shows the markers deployed and the wall used to test positioning:

0,0%

5,0%

10,0%

15,0%

20,0%

25,0%

0

100

200

300

400

500

600

700

800

900

1000

1 2 3 4 5

Au

ton

om

y in

crea

se (

%)

Flig

ht

tim

e (s

)

Test

Autonomy increase

No wing (s) Wing (s) %


Figure 60 Positioning laboratory test

Next figure illustrates captured frames from the camera and the process to

obtain distance and pose estimation, as described in the previous section:

Figure 61 Positioning process: a) Captured frame; b) Greyscale image; c)

Binarized; d) Pose estimation

In the experiment, the UAV flied 5 meters until forward ultrasonic sensors

detected a distance of 0.6 m to the front wall, where it stopped. Forward and

transversal distance sensors recorded distance to their respective walls; the

positioning algorithm was providing distance as well according to the well-known

position of the markers, and the captured data was filtered as described in the

previous section. The experimented was repeated ten times; Figure 26 displays

averaged ultrasonic distance (sensed) and the average distance obtained using the

computer vision system (filtered), for the straight flight depicted in Figure x7 from

x = 5.6m until x = 0.6m, trying to keep a constant distance with respect the wall of

0.6m:


Figure 62 Calculated distance using the vision algorithm (Filtered) versus

ultrasonic measured distance (Sensed)

Ultrasonic readings (sensed distance measures) were considered to be providing

the true distances to evaluate the algorithm. Table 7 displays average relative errors

and RMSE values for the x and y coordinates:

Coordinate Relative error (%) RMSE

x (forward) 7.5 0.24

y

(transversal)

7.2 0.05

Table 11 Average relative error and RMSE values

The average of the obtained values by the computer vision algorithm stay

reasonably close to real values provided by the ultrasonic sensors. Despite the

relative error, flight is kept relatively steady thanks to the PID controller that, with

the flight dynamic, softens the trajectory.

4. Landing system

The landing procedure was tested in the laboratory floor using a 0.9x0.9 m MDF

black board as landing table on a 3x3 m white bed, and the proposed landing pad

marker in the center of the table. It was repeated 15 times and each iteration was

split into three steps:

The first step is long range landing pad detection, as shown in the following

picture:


Figure 63. Finding landing table from transversal corridor; from left to right: a)

Original image; b) Greyscale modification; c) Binarized

Reasonable results were obtained for the location of the landing center as

displayed on the next figure; the circle was recognized and kept with visual range

in every iteration, and every captured frame obtained its bounding box, ellipse

approximation, and center calculation:

Figure 64. Bounding box area and detected circle. The ArUco markers are still not

within range, but the circle helps obtaining a reference from long-range distance.

The second step is to obtain positioning from both methods for when distance

to the landing pad keeps the circle in visual range and the inner markers enter visual

range as well. Once the UAV leaves transversal corridor height, the ArUco markers

will become visible, as shown in the following figure:


Figure 65. Descent control while long-range and short-range are available. a)

ArUco markers first detection; top left corners are detected, and the dictionary

entries are identified; the rotation of the markers generates two diagonals whose

intersection indicates landing destination. b) At this flight level, the circle is still

visible and allows confrontation of destination point coordinates between the two

methods.

Consistency between the two previous methods was checked to evaluate the

discrepancy of the (x,y) coordinates location given by both. Average absolute

calculated error among the 15 sets of values where 7 mm for the x coordinate, and 8

mm for the y coordinate. Coordinates are represented on the following figure,

showing an acceptable discrepancy between the values:

Figure 66. Discrepancy between center coordinates evaluated via circle versus via

ArUco markers.


The third step is for close range positioning, when UAV’s height is low enough

as to lose sight of the outer circle and perform pose estimation and locate landing

point just with ArUco markers. Algorithm stops circle detector after 100 continuous

frames with no circle obtained. Next figure displays OpenCV’s coordinate reference

systems for each of the markers, according to their rotation:

Figure 67. ArUco markers close range detection.

Finally, once landing was finished, landing spot was compared to the real

intersection between ArUco marker’s top left corner; as landing spot it was

considered the vertical from the camera, and the distance in (x,y) coordinates was

measured as displayed on the following figure. The average absolute error in the x

coordinate was 19 mm, and 23 mm por the y coordinate:


Figure 68. Actual landing spot versus landing goal.

A certain interference with landing table is affecting last centimeters approach,

as a ground effect, generating horizontal x and y displacements when the UAV is

about to finish landing; ground effect is expected to be affecting the dynamic of the

operation as it has been already discussed in the literature [150]. Since laboratory

results are within acceptable error, this issue has not been addressed within this

research.

5. PV-T system

In the first place, experimental- and TRNSYS-simulated data are compared to

find the minimum error between those values, according to the previous section. To

perform the optimization, we try to find the mean square error (RMSE), according

to:

RMSE = √1

𝑛∑(𝑦𝑖 − 𝑥𝑖)2

𝑛

𝑖=1

(35)

The value obtained for the lump mass’ capacitance after minimizing RMSE up

to 1.1 °C is 9.0134 KJ K⁄ , with a 0.98 regression coefficient. This RMSE value is

obtained for an average experimental temperature of 7.8 °C and is within the

validated values systems with similar characteristics throughout the literature. For


instance, in [107], a validation of the experimental data obtained from a PV-T is

performed, achieving a 1.11 °C value, for an actual temperature of 30 °C. Other

similar RMSE results are also obtained in other scenarios where lumped mass is also

applied. In [114], a Stirling engine block is modeled as a lump mass; the output

temperature of the refrigeration water achieves a RMSE value of 1.0 °C, for a 45 °C

block temperature. In that research, other different experimental scenarios for 60 and

70 °C block temperatures are also tested, resulting in RMSE values of the same order

of magnitude (1.4 and 4.0 °C, respectively). In [104], an analysis of the regression

coefficient is performed, achieving 0.95 and 0.98 values for output temperature, in

the order of magnitude of those in our research.

The regression coefficient value obtained is within expectations and aligned

with similar experimental validation tests. For example, in [108] a validation of

experimental results for a PV-T panel is performed. In that research, the PV-T used

water as refrigeration fluid and the value of the regression coefficient between actual

output temperature and the simulated one is 0.99, very similar to our case (0.98).

The next figure shows actual temperature (red color), while simulated data are

represented in the blue color. As can be appreciated in the figure, there are eight

peaks corresponding to the eight simulated days, and maximum achieved radiation.

The simulated data matches actual data reasonably well to expectations.

Figure 69. Experimental actual temperature (Treal) versus simulated (Tsim) during

simulation period.


The following figure represents intake temperature versus output temperature

obtained through simulation. According to the size of the gap between input and

output, the system will have further thermal conditioning capabilities.

Figure 70. Output’s simulated temperature (Tsim) versus intake temperature

(Tin) in °C.

The proposed system is validated, given its capability of increasing the

temperature of the facilities to be conditioned. The following table shows the

simulated output temperature peaks according to time, input temperature, and

achieved gradient temperature. A maximum temperature increase of 9.2 °C is

reached. This value allows for a generic room air conditioning up to a 19.9 °C limit,

since that is the temperature of the air at the outlet of the duct. It is important to note

that this system was originally designed to fulfil the demand of UAV shelter

hangars; those are operational facilities (not for leisure) and, consequently, a small

temperature increase immediately generates a thermal comfort sensation.

Simulation Step (h) Tout(℃) Tin(℃) ∆T (℃)

16 17.1 10.3 6.7

34 19.9 10.8 9.2

62 18.4 10.8 7.6

87 21.9 13.7 8.1

111 18.9 12.7 6.3

135 20.9 12.4 8.4

159 21.2 12.7 8.5


183 21.7 13.7 7.0

Table 12 Temperature increases during peak hours.

Another interesting analysis is the one focused on nightly periods where output

temperature is below the one at the input. Considering the whole period of

simulation time, we obtain an average temperature increase of 1.2 °C. If we limit

analysis period to the phases where Tout > Tin, it happens for 72 out of the 192 h in

total, which represents 37.5% of the whole time. Under these conditions, the average

temperature increase reached is 4.8 °C, as displayed in the following table:

Period T̅out(℃) T̅in(℃) ∆T̅ (℃)

Simulation time (100%) 7.7 6.6 1.2

Heating time (37.5%) 14.9 10.0 4.8

Table 13 Heating net time

It is reasonable to have a system working in a 37.5% cycle for heating purposes

for a full year simulation, since nightly period takes most of the daytime during

winter (when there are few hours of solar radiation).

As represented in the following figure, during the night phase of the day, the

temperature inversion phenomenon occurs: the input temperature is the one higher,

followed by the output temperature, and finally, the lump mass temperature is the

lowest. This is due to radiation heat exchange with the environment together with

quite low sky temperatures during winter nights.


Figure 71. Actual output temperature (Treal) versus input temperature (Tin) versus

simulated lump mass temperature (°C).

Finally, thermal gains HG1 and HG2 are shown. The next figure plots heat

transfer rates 1 and 2 versus total incoming radiation. The scales are adapted to

undoubtedly show that PV-T’s performance as thermal collector is relatively small.

Figure 72. Incident radiation (SI) compared to thermal gains (HG1 and HG2).

As it is displayed, HG2 is under zero values for most of the simulation time; the

HG2, heat transfer rate for the lump mass, actually represents a heat loss towards

the air inside the duct. This happens during daylight, but the opposite occurs at


night; during the night, it is the lump mass that is heated from the duct, since the air

is still flowing inside the test bed, and ambient temperature is higher than that of the

lump mass at that moment. As for the negative values of HG1, the reason is the heat

loss that the lump mass suffers, interchanging that heat with the sky at night by

means of radiation.


Chapter 7. Conclussions

In this research, a positioning system and a landing system are defined for an

indoor light part delivery UAV within a manufacturing plant, providing a mesh

network for WIFI communication with the back-end software that manages the

operation. The overall control of the UAV is provided by an onboarded Raspberry

Pi, performing localization, computer vision for landing and communication. The

System on Chip computer is also providing nodes for the mesh network on the

ground as well forwarding incoming command messages for the UAV and outgoing

location telemetry from the flying vehicle.

As for localization, a combination of RFID, sonar, and a proper definition of

flying corridors with operator’s safety in mind is done. The conjunction of the three

elements provide a solution to avoid the well-known problem of RSS readings.

Sonars provide accurate distance to the confined flying corridor and keep the UAV

centred in the right track. An improved mechanism to deploy tags, and the use of a

specific coating to prevent reflections help evading the multipath problem;

consequently, the whole solution the initial lack of accuracy of RSS positioning

method, and the average error and RMSE are kept within acceptable values.

Manufacturing plant layout is abstracted by representing it as a series of nodes, and

an improved performance algorithm is used to allow finding operator’s workstation.

Horizontal flight planning has been simplified as a graph of perimetral and


transversal corridors, what also allows providing required safety network for

operators’ safety. Nodes graph and tag distribution are kept in the onboarded

Raspberry Pi computer to provide autonomous flight. Laboratory experiments

showed reasonable results for keeping UAV’s location under back-end software

control.

As for autonomous landing, an affordable computer vision is designed to

provide long-range and short-range localization of the landing pad and pose

estimation. Descent is split into three different steps. First, finding the point where

transversal flying corridor must be left; a long-range detection of a circle helps

determining when the UAV should stop going forward, and begin descent; descent

begins and continues until the short-range detection mechanism, based on four

ArUco markers, is within reach. Second, performing most of the descent using the

short-range algorithm, while keeping the long-range still active to double-check

destination point; fusion estimation is used to leverage the existence of more than

one element and provide a better estimation; we even take advantage of marker’s

arrangement and orientation, to obtain another reference (diagonal intersection);

laboratory experiments provided acceptable discrepancy between these

complementary method. Finally, the last stage uses only the ArUco markers to

perform short-range approximation to the landing table; experiments show a

reasonable difference between landing goal and actual landing spot.

According to the obtained results, we conclude that the proposed model is

validated through experimental results. PV-T’s thermal modelling as lump mass is

not only possible, but also a great concordance between the workbench’s data and

simulated values. Root mean square error achieved, 1.1 °C, and regression coefficient

(0.98) are considered reasonable for the use of the model to predict system’s behavior

under different climatological conditions.


Chapter 8. Future lines of research

There are still many relevant fields to be further investigated and improved.

First, a single UAV is considered to be flying in this research, but it is expected that,

according to the number of incidents to be attended concurrently; more than one

flying vehicle needs a suitable mission control and the management of path

intersection, as well as a system to regulate which UAV is attending each of the

incidents raised. Second, the capability of the vehicle to attend more than one

incident in a flight; in this research one flight was considered to attend one incident,

but it could be improved by delivering to more than one workstation in a single run;

payload’s characteristics should be taken into consideration so that the system could

decide when two parts could be delivered together. Third, ground effect should be

addressed; it was visually observed in the experiments that when the UAV is just a

few centimetres far from ground, horizontal displacements were suffered, resulting

in worst results than the ones achieved in the rest of the descent; the particular

dynamics when the vehicle is about to perform landing should be added to improve

landing accuracy.

It is expected that these next steps can be taken in a near future to extend the

coverage of the research already performed.


References

[1] J. Grzybowski, K. Latos and R. Czyba, "Low-cost autonomous UAV-based solutions to package delivery logistics," in Advanced, Contemporary ControlAnonymous Springer, 2020, pp. 500-507.

[2] K. Han and H. Jung, "Trends in Logistics Delivery Services Using UAV," Electronics and Telecommunications Trends, vol. 35, (1), pp. 71-79, 2020.

[3] K. Kuru et al, "Analysis and optimization of unmanned aerial vehicle swarms in logistics: An intelligent delivery platform," Ieee Access, vol. 7, pp. 15804-15831, 2019.

[4] J. Eun et al, "Mathematical Investigation on the Sustainability of UAV Logistics," Sustainability, vol. 11, (21), pp. 5932, 2019.

[5] A. Solomasov, "Analysis of supply chain operational performances using vehicle routing with UAV delivery in city logistics," 2019.

[6] Y. Qi and C. Wang, "Research on the problems and countermeasures of intelligent logistics system in china," in 2020 International Conference on Computer Information and Big Data Applications (CIBDA), 2020, pp. 116-119.

[7] A. Raj and B. Sah, "Analyzing critical success factors for implementation of drones in the logistics sector using grey-DEMATEL based approach," Comput. Ind. Eng., vol. 138, pp. 106118, 2019.

[8] H. Y. Jeong, B. D. Song and S. Lee, "The Flying Warehouse Delivery System: A Quantitative Approach for the Optimal Operation Policy of Airborne Fulfillment Center," IEEE Transactions on Intelligent Transportation Systems, 2020.

[9] I. K. Nikolos et al, "Evolutionary algorithm based offline/online path planner for UAV navigation," IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), vol. 33, (6), pp. 898-912, 2003.

[10] B. D. Song, K. Park and J. Kim, "Persistent UAV delivery logistics: MILP formulation and efficient heuristic," Comput. Ind. Eng., vol. 120, pp. 418-428, 2018.

[11] B. H. Kim et al, "Real-time counter-UAV system for long distance small drones using double pan-tilt scan laser radar," in Laser Radar Technology and Applications XXIV, 2019, pp. 110050C.

[12] S. Ortiz et al, "A UAV-based content delivery architecture for rural areas and future smart cities," IEEE Internet Comput., vol. 23, (1), pp. 29-36, 2018.


[13] P. Shao, "Risk Assessment for UAS Logistic Delivery under UAS Traffic Management Environment," Aerospace, vol. 7, (10), pp. 140, 2020.

[14] A. M. Ham, "Integrated scheduling of m-truck, m-drone, and m-depot constrained by time-window, drop-pickup, and m-visit using constraint programming," Transportation Research Part C: Emerging Technologies, vol. 91, pp. 1-14, 2018.

[15] D. Wang et al, "Routing and scheduling for hybrid truck-drone collaborative parcel delivery with independent and truck-carried drones," IEEE Internet of Things Journal, vol. 6, (6), pp. 10483-10495, 2019.

[16] J. Aurambout, K. Gkoumas and B. Ciuffo, "Last mile delivery by drones: An estimation of viable market potential and access to citizens across European cities," European Transport Research Review, vol. 11, (1), pp. 30, 2019.

[17] X. Wang and L. Duan, "Dynamic pricing and capacity allocation of UAV-provided mobile services," in IEEE INFOCOM 2019-IEEE Conference on Computer Communications, 2019, pp. 1855-1863.

[18] P. Wrycza, M. Rotgeri and M. ten Hompel, "Spielzeitreduktion autonomer Drohnen für den Transport eiliger Güter durch den Einsatz automatisierter Lastaufnahmemittel im Kontext eines ganzheitlich automatisierten Gesamtsystems," Logistics Journal: Proceedings, vol. 2017, (10), 2017.

[19] N. H. Motlagh, M. Bagaa and T. Taleb, "UAV-based IoT platform: A crowd surveillance use case," IEEE Communications Magazine, vol. 55, (2), pp. 128-134, 2017.

[20] W. G. Patrick, J. R. Burgess and A. Conrad, Mechanisms for Lowering a Payload to the Ground from a UAV, 2016.

[21] J. Kim, S. Ha and Y. Moon, "A study on automatic precision landing for small UAV's industrial application," Journal of Convergence for Information Technology, vol. 7, (3), pp. 27-36, 2017.

[22] A. Masiero et al, "A COMPARISON OF UWB AND MOTION CAPTURE UAV INDOOR POSITIONING." International Archives of the Photogrammetry, Remote Sensing & Spatial Information Sciences, 2019.

[23] F. Liu et al, "Survey on WiFi-based indoor positioning techniques," IET Communications, vol. 14, (9), pp. 1372-1383, 2020.

[24] C. Xin et al, "Research on indoor navigation system of UAV based on LIDAR," in 2020 12th International Conference on Measuring Technology and Mechatronics Automation (ICMTMA), 2020, pp. 763-766.


[25] I. Hatem et al, "Low-cost quadcopter indoor positioning system based on image processing and neural networks," in Mechanism, Machine, Robotics and Mechatronics SciencesAnonymous Springer, 2019, pp. 243-257.

[26] W. You et al, "Data Fusion of UWB and IMU Based on Unscented Kalman Filter for Indoor Localization of Quadrotor UAV," IEEE Access, vol. 8, pp. 64971-64981, 2020.

[27] E. Martín-Gorostiza et al, "An Indoor Positioning Approach Based on Fusion of Cameras and Infrared Sensors," Sensors, vol. 19, (11), pp. 2519, 2019.

[28] C. Debeunne and D. Vivet, "A Review of Visual-LiDAR Fusion based Simultaneous Localization and Mapping," Sensors, vol. 20, (7), pp. 2068, 2020.

[29] E. Hozdić, "Smart factory for industry 4.0: A review," International Journal of Modern Manufacturing Technologies, vol. 7, (1), pp. 28-35, 2015.

[30] B. G. Fahey, Method and Apparatus for Industrial Data Acquisition and Product Costing, 1999.

[31] R. H. Caro, "Field bus applications," ISA Trans., vol. 28, (4), pp. 65-70, 1989.

[32] E. H. Chan et al, "A coordinated statewide load management and SCADA system," IEEE Transactions on Power Apparatus and Systems, (11), pp. 3496-3501, 1983.

[33] G. Meyer, "SCADA-a 10-year look into the future," Pipe Line Ind.;(United States), vol. 62, (4), 1985.

[34] C. Roy, "Technological growth and future development will influence SCADA systems design," Pipeline News;(United States), vol. 56, (2), 1984.

[35] U. RMS, "1 Advanced control methods," 1984.

[36] K. Mak and B. Holland, "Migrating electrical power network SCADA systems to TCP/IP and Ethernet networking," Power Eng J, vol. 16, (6), pp. 305-311, 2002.

[37] H. Tan, Z. Huang and M. Wu, "An Interactive Real-Time SCADA Platform With Customizable Virtual Instruments for Cloud Control Systems," Journal of Dynamic Systems, Measurement, and Control, vol. 141, (4), pp. 044501, 2019.

[38] H. Boyes et al, "The industrial internet of things (IIoT): An analysis framework," Comput. Ind., vol. 101, pp. 1-12, 2018.

[39] A. Nigam and A. Agiwal, Machine to Machine Communications in a Multiple M2M Device Group Zone Environment, 2018.


[40] S. Singhal, A Method A Server and A Client for Policy Based Control of M2M Devices, 2018.

[41] L. Thames and D. Schaefer, Cybersecurity for Industry 4.0. Springer, 2017.

[42] I. F. Akyildiz et al, "Wireless sensor networks: a survey," Computer Networks, vol. 38, (4), pp. 393-422, 2002.

[43] T. Chikaraishi, T. Minato and H. Ishiguro, "Development of an android system integrated with sensor networks," Geminoid Studies: Science and Technologies for Humanlike Teleoperated Androids, pp. 1-19, 2018.

[44] S. Corbellini et al, "Cloud based sensor network for environmental monitoring," Measurement, vol. 118, pp. 354-361, 2018.

[45] S. Li and L. Qiao, "Ontology-based modeling of manufacturing information and its semantic retrieval," in Proceedings of the 2012 IEEE 16th International Conference on Computer Supported Cooperative Work in Design (CSCWD), 2012, pp. 540-545.

[46] A. Jayaram, "Lean six sigma approach for global supply chain management using industry 4.0 and IIoT," in 2016 2nd International Conference on Contemporary Computing and Informatics (IC3I), 2016, pp. 89-94.

[47] L. Thames and D. Schaefer, "Software-defined cloud manufacturing for industry 4.0," Procedia CIRP, vol. 52, pp. 12-17, 2016.

[48] L. Barreto, A. Amaral and T. Pereira, "Industry 4.0 implications in logistics: an overview," Procedia Manufacturing, vol. 13, pp. 1245-1252, 2017.

[49] J. Wan et al, "Software-defined industrial internet of things in the context of industry 4.0," IEEE Sensors Journal, vol. 16, (20), pp. 7373-7380, 2016.

[50] T. W. Moore, P. R. Wardrop and M. McIntosh, Method and System for Signing JavaScript Object Notation (JSON) Messages, 2012.

[51] C. Ortega-Corral et al, "A “Lighter” JSON for message exchange in highly resource constrained wireless sensor network applications," Congr Int Ing Electron Mem Electro, vol. 34, pp. 73-78, 2012.

[52] C. Wu et al, "Real-time wireless routing for industrial internet of things," in 2018 IEEE/ACM Third International Conference on Internet-of-Things Design and Implementation (IoTDI), 2018, pp. 261-266.


[53] M. Sha et al, "Empirical study and enhancements of industrial wireless sensor–actuator network protocols," IEEE Internet of Things Journal, vol. 4, (3), pp. 696-704, 2017.

[54] W. Dargie and C. Poellabauer, Fundamentals of Wireless Sensor Networks: Theory and Practice. John Wiley & Sons, 2010.

[55] R. Faludi, Building Wireless Sensor Networks: With ZigBee, XBee, Arduino, and Processing. " O'Reilly Media, Inc.", 2010.

[56] D. Slama et al, Enterprise IoT: Strategies and Best Practices for Connected Products and Services. " O'Reilly Media, Inc.", 2015.

[57] P. Orgeira-Crespo, "AUTONOMOUS AERIAL SYSTEM FOR LIGHT PARTS PROVISIONING IN AN INDOOR MANUFACTURING PLANT," Tfg, 2020.

[58] T. de Almeida Oliveira and E. P. Godoy, "ZigBee wireless dynamic sensor networks: feasibility analysis and implementation guide," IEEE Sensors Journal, vol. 16, (11), pp. 4614-4621, 2016.

[59] B. Klauer et al, "Wireless sensor/actuator device configuration by NFC with secure key exchange," in 2017 Ieee Africon, 2017, pp. 473-478.

[60] S. Mejjaouli and R. F. Babiceanu, "RFID-wireless sensor networks integration: Decision models and optimization of logistics systems operations," J. Manuf. Syst., vol. 35, pp. 234-245, 2015.

[61] A. Pavan, "A Survey of Z-wave Wireless Sensor Network Technology," 2018.

[62] J. F. Schmidt et al, "Experimental study of UWB connectivity in industrial environments," in European Wireless 2018; 24th European Wireless Conference, 2018, pp. 1-4.

[63] J. Yang et al, "A real-time monitoring system of industry carbon monoxide based on wireless sensor networks," Sensors, vol. 15, (11), pp. 29535-29546, 2015.

[64] J. Du, Indoor Localization Techniques for Wireless Sensor Networks, 2018.

[65] R. F. Brena et al, "Evolution of indoor positioning technologies: A survey," Journal of Sensors, vol. 2017, 2017.

[66] M. Khan, N. Salman and A. Kemp, "Cooperative positioning using angle of arrival and time of arrival," in 2014 Sensor Signal Processing for Defence (SSPD), 2014, pp. 1-5.


[67] X. Li et al, "Contributed Review: Source-localization algorithms and applications using time of arrival and time difference of arrival measurements," Rev. Sci. Instrum., vol. 87, (4), pp. 041502, 2016.

[68] M. Li and Y. Lu, "Angle-of-arrival estimation for localization and communication in wireless networks," in 2008 16th European Signal Processing Conference, 2008, pp. 1-5.

[69] J. Wubben et al, "Accurate landing of unmanned aerial vehicles using ground pattern recognition," Electronics, vol. 8, (12), pp. 1532, 2019.

[70] T. Yang et al, "Monocular vision SLAM-based UAV autonomous landing in emergencies and unknown environments," Electronics, vol. 7, (5), pp. 73, 2018.

[71] S. Lin, M. A. Garratt and A. J. Lambert, "Real-time 6DoF deck pose estimation and target tracking for landing an UAV in a cluttered shipboard environment using on-board vision," in 2015 IEEE International Conference on Mechatronics and Automation (ICMA), 2015, pp. 474-481.

[72] S. M. Chaves, R. W. Wolcott and R. M. Eustice, "NEEC research: Toward GPS-denied landing of unmanned aerial vehicles on ships at sea," Nav. Eng. J., vol. 127, (1), pp. 23-35, 2015.

[73] M. Fiala, "ARTag, a fiducial marker system using digital techniques," in 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05), 2005, pp. 590-596.

[74] J. Wang and E. Olson, "AprilTag 2: Efficient and robust fiducial detection," in 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2016, pp. 4193-4198.

[75] R. Muñoz-Salinas et al, "Mapping and localization from planar markers," Pattern Recognit, vol. 73, pp. 158-171, 2018.

[76] F. J. Romero-Ramirez, R. Muñoz-Salinas and R. Medina-Carnicer, "Speeded up detection of squared fiducial markers," Image Vision Comput., vol. 76, pp. 38-47, 2018.

[77] S. Garrido-Jurado et al, "Generation of fiducial marker dictionaries using mixed integer linear programming," Pattern Recognit, vol. 51, pp. 481-491, 2016.

[78] R. Jiménez Bravo, "Sistema de seguimiento de objetos usando OpenCv, ArUco y Filtro de Kalman extendido," 2018.


[79] M. F. Sani and G. Karimian, "Automatic navigation and landing of an indoor AR. drone quadrotor using ArUco marker and inertial sensors," in 2017 International Conference on Computer and Drone Applications (IConDA), 2017, pp. 102-107.

[80] T. T. Chow, "A review on photovoltaic/thermal hybrid solar technology," Appl. Energy, vol. 87, (2), pp. 365-379, 2010.

[81] M. A. Hasan and K. Sumathy, "Photovoltaic thermal module concepts and their performance analysis: a review," Renewable and Sustainable Energy Reviews, vol. 14, (7), pp. 1845-1859, 2010.

[82] R. Kumar and M. A. Rosen, "A critical review of photovoltaic–thermal solar collectors for air heating," Appl. Energy, vol. 88, (11), pp. 3603-3614, 2011.

[83] B. Parida, S. Iniyan and R. Goic, "A review of solar photovoltaic technologies," Renewable and Sustainable Energy Reviews, vol. 15, (3), pp. 1625-1636, 2011.

[84] R. Daghigh, M. H. Ruslan and K. Sopian, "Advances in liquid based photovoltaic/thermal (PV/T) collectors," Renewable and Sustainable Energy Reviews, vol. 15, (8), pp. 4156-4170, 2011.

[85] A. Ibrahim et al, "Recent advances in flat plate photovoltaic/thermal (PV/T) solar collectors," Renewable and Sustainable Energy Reviews, vol. 15, (1), pp. 352-365, 2011.

[86] V. Tyagi, S. Kaushik and S. Tyagi, "Advancement in solar photovoltaic/thermal (PV/T) hybrid collector technology," Renewable and Sustainable Energy Reviews, vol. 16, (3), pp. 1383-1398, 2012.

[87] D. Kamthania and G. Tiwari, "Photovoltaic thermal air collectors: A review," Journal of Renewable and Sustainable Energy, vol. 6, (6), pp. 062701, 2014.

[88] S. A. Hamid et al, "An overview of photovoltaic thermal combination (PV/T combi) technology," Renewable and Sustainable Energy Reviews, vol. 38, pp. 212-222, 2014.

[89] J. J. Michael, S. Iniyan and R. Goic, "Flat plate solar photovoltaic–thermal (PV/T) systems: A reference guide," Renewable and Sustainable Energy Reviews, vol. 51, pp. 62-88, 2015.

[90] K. Sopian et al, "Performance analysis of photovoltaic thermal air heaters," Energy Conversion and Management, vol. 37, (11), pp. 1657-1670, 1996.

[91] M. Y. H. Othman et al, "Performance analysis of a double-pass photovoltaic/thermal (PV/T) solar collector with CPC and fins," Renewable Energy, vol. 30, (13), 2005.


[92] A. Tiwari and M. Sodha, "Parametric study of various configurations of hybrid PV/thermal air collector: experimental validation of theoretical model," Solar Energy Mater. Solar Cells, vol. 91, (1), pp. 17-28, 2007.

[93] J. Tonui and Y. Tripanagnostopoulos, "Air-cooled PV/T solar collectors with low cost performance improvements," Solar Energy, vol. 81, (4), pp. 498-511, 2007.

[94] J. Tonui and Y. Tripanagnostopoulos, "Performance improvement of PV/T solar collectors with natural air flow operation," Solar Energy, vol. 82, (1), pp. 1-12, 2008.

[95] V. P. Katekar and S. S. Deshmukh, "A review on research trends in solar still designs for domestic and industrial applications," J. Clean. Prod., vol. 257, pp. 120544, 2020.

[96] W. Pang et al, "A comparative analysis on performances of flat plate photovoltaic/thermal collectors in view of operating media, structural designs, and climate conditions," Renewable and Sustainable Energy Reviews, vol. 119, pp. 109599, 2020.

[97] Y. Jia, G. Alva and G. Fang, "Development and applications of photovoltaic–thermal systems: A review," Renewable and Sustainable Energy Reviews, vol. 102, pp. 249-265, 2019.

[98] F. Yazdanifard and M. Ameri, "Exergetic advancement of photovoltaic/thermal systems (PV/T): A review," Renewable and Sustainable Energy Reviews, vol. 97, pp. 529-553, 2018.

[99] S. M. Sultan and M. E. Efzan, "Review on recent Photovoltaic/Thermal (PV/T) technology advances and applications," Solar Energy, vol. 173, pp. 939-954, 2018.

[100] A. Chauhan, V. Tyagi and S. Anand, "Futuristic approach for thermal management in solar PV/thermal systems with possible applications," Energy Conversion and Management, vol. 163, pp. 314-354, 2018.

[101] S. Solanki, S. Dubey and A. Tiwari, "Indoor simulation and testing of photovoltaic thermal (PV/T) air collectors," Appl. Energy, vol. 86, (11), pp. 2421-2428, 2009.

[102] A. Joshi et al, "Performance evaluation of a hybrid photovoltaic thermal (PV/T)(glass-to-glass) system," International Journal of Thermal Sciences, vol. 48, (1), pp. 154-164, 2009.

[103] F. Sarhaddi et al, "An improved thermal and electrical model for a solar photovoltaic thermal (PV/T) air collector," Appl. Energy, vol. 87, (7), pp. 2328-2339, 2010.

[104] A. Shahsavar and M. Ameri, "Experimental investigation and modeling of a direct-coupled PV/T air collector," Solar Energy, vol. 84, (11), pp. 1938-1958, 2010.


[105] R. Kumar and M. A. Rosen, "Performance evaluation of a double pass PV/T solar air heater with and without fins," Appl. Therm. Eng., vol. 31, (8-9), pp. 1402-1410, 2011.

[106] D. Kamthania, S. Nayak and G. Tiwari, "Performance evaluation of a hybrid photovoltaic thermal double pass facade for space heating," Energy Build., vol. 43, (9), pp. 2274-2281, 2011.

[107] R. Daghigh and Y. Khaledian, "Design and fabrication of a bi-fluid type photovoltaic-thermal collector," Energy, vol. 135, pp. 112-127, 2017.

[108] C. J. Kuo et al, "The photovoltaic-thermal system parameter optimization design and practical verification," Energy Conversion and Management, vol. 180, pp. 358-371, 2019.

[109] S. Thiers, B. Aoun and B. Peuportier, "Experimental characterization, modeling and simulation of a wood pellet micro-combined heat and power unit used as a heat source for a residential building," Energy Build., vol. 42, (6), pp. 896-903, 2010.

[110] K. Lombardi, V. Ugursal and I. Beausoleil-Morrison, "Proposed improvements to a model for characterizing the electrical and thermal energy performance of Stirling engine micro-cogeneration devices based upon experimental observations," Appl. Energy, vol. 87, (10), pp. 3271-3282, 2010.

[111] G. Magri, C. Di Perna and G. Serenelli, "Analysis of electric and thermal seasonal performances of a residential microCHP unit," Appl. Therm. Eng., vol. 36, pp. 193-201, 2012.

[112] C. Ulloa et al, "Development of a transient model of a stirling-based CHP system," Energies, vol. 6, (7), pp. 3115-3133, 2013.

[113] C. Ulloa et al, "Application model for a stirling engine micro-generation system in caravans in different european locations," Energies, vol. 6, (2), pp. 717-732, 2013.

[114] F. Keynejad and C. Manzie, "Cold start modelling of spark ignition engines," Control Eng. Pract., vol. 19, (8), pp. 912-925, 2011.

[115] A. Hasan, M. Vuolle and K. Sirén, "Minimisation of life cycle cost of a detached house using combined simulation and optimisation," Build. Environ., vol. 43, (12), pp. 2022-2034, 2008.

[116] F. Yan and J. Wang, "Pressure-based transient intake manifold temperature reconstruction in diesel engines," Control Eng. Pract., vol. 20, (5), pp. 531-538, 2012.

[117] X. Wang and H. Chua, "Two bed silica gel–water adsorption chillers: an effectual lumped parameter model," Int. J. Refrig., vol. 30, (8), pp. 1417-1426, 2007.


[118] A. Rosales-Pousada, "Estudio aerodinámico de un perfil alar para dron de aprovisionamiento. Estudio de autonomía," Tfm, 2020.

[119] F. Fernández-Rodriguez, "Diseño mecánico de la estructura de un UAV con cubeta de transporte para el aprovisionamiento de pieza ligera en interiores de plantas industriales," Tfm, 2020.

[120] D. Gheorghiţă et al, "Quadcopter control system," in 2015 19th International Conference on System Theory, Control and Computing (ICSTCC), 2015, pp. 421-426.

[121] M. R. A. Bhosale, D. Padmanabhan and S. Kadam, "Mathematical modelling and simulation of duadcopter UAV using PID controller," IOSR J.Eng., pp. 29-36, 2018.

[122] P. Wang et al, "Dynamics modelling and linear control of quadcopter," in 2016 International Conference on Advanced Mechatronic Systems (ICAMechS), 2016, pp. 498-503.

[123] Z. Benić, P. Piljek and D. Kotarski, "Mathematical modelling of unmanned aerial vehicles with four rotors," Interdisciplinary Description of Complex Systems: INDECS, vol. 14, (1), pp. 88-100, 2016.

[124] F. J. Romero-Ramirez, R. Muñoz-Salinas and R. Medina-Carnicer, "Speeded up detection of squared fiducial markers," Image Vision Comput., vol. 76, pp. 38-47, 2018.

[125] W. Pratt, "Digital image processing Wiley-Interscience," 2007.

[126] J. Canny, "A computational approach to edge detection," IEEE Trans. Pattern Anal. Mach. Intell., (6), pp. 679-698, 1986.

[127] S. Suzuki, "Topological structural analysis of digitized binary images by border following," Computer Vision, Graphics, and Image Processing, vol. 30, (1), pp. 32-46, 1985.

[128] D. H. Douglas and T. K. Peucker, "Algorithms for the reduction of the number of points required to represent a digitized line or its caricature," Cartographica: The International Journal for Geographic Information and Geovisualization, vol. 10, (2), pp. 112-122, 1973.

[129] N. Otsu, "A threshold selection method from gray-level histograms," IEEE Trans. Syst. Man Cybern., vol. 9, (1), pp. 62-66, 1979.

[130] D. W. Marquardt, "An algorithm for least-squares estimation of nonlinear parameters," Journal of the Society for Industrial and Applied Mathematics, vol. 11, (2), pp. 431-441, 1963.


[131] J. Shi, "Good features to track," in 1994 Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, 1994, pp. 593-600.

[132] S. Haykin, Kalman Filtering and Neural Networks. John Wiley & Sons, 200447.

[133] H. Sriyananda, "A simple method for the control of divergence in Kalman-filter algorithms," Int J Control, vol. 16, (6), pp. 1101-1106, 1972.

[134] L. E. Cannon, A Cellular Computer to Implement the Kalman Filter Algorithm, 1969.

[135] I. Reid and H. Term, "Estimation ii," University of Oxford, Lecture Notes, 2001.

[136] E. Dijkstra, "Dijkstra’s algorithm," Dutch Scientist Dr.Edsger Dijkstra Network Algorithm: Http://En.Wikipedia.Org/Wiki/Dijkstra'S_algorithm, 1959.

[137] A. Rosebrock, Deep Learning for Computer Vision with Python: Starter Bundle. PyImageSearch, 2017.

[138] A. Kaehler and G. Bradski, Learning OpenCV 3: Computer Vision in C with the OpenCV Library. " O'Reilly Media, Inc.", 2016.

[139] C. Open, "OpenCV documentation," Obtained from Http://Opencv.Org/Documentation.Html, 2019.

[140] D. Liu and J. Yu, "Otsu method and K-means," in 2009 Ninth International Conference on Hybrid Intelligent Systems, 2009, pp. 344-349.

[141] X. X. Lu, "A review of solutions for perspective-n-point problem in camera pose estimation," in Journal of Physics: Conference Series, 2018, pp. 052009.

[142] V. Lepetit, F. Moreno-Noguer and P. Fua, "Epnp: An accurate o (n) solution to the pnp problem," International Journal of Computer Vision, vol. 81, (2), pp. 155, 2009.

[143] Y. Zhu, X. Li and J. Zhao, "Linear minimum variance estimation fusion," Science in China Series F: Information Sciences, vol. 47, (6), pp. 728-740, 2004.

[144] X. Liu et al, "An Onboard Vision-Based System for Autonomous Landing of a Low-Cost Quadrotor on a Novel Landing Pad," Sensors, vol. 19, (21), pp. 4703, 2019.

[145] L. Petru and G. Mazen, "PWM control of a DC motor used to drive a conveyor belt," Procedia Engineering, vol. 100, pp. 299-304, 2015.

[146] F. Salamone et al, "An open source low-cost wireless control system for a forced circulation solar plant," Sensors, vol. 15, (11), pp. 27990-28004, 2015.


[147] E. Avallone et al, "Electronic multiplex system using the Arduino platform to control and record the data of the temperatures profiles in heat storage tank for solar collector," International Journal of Energy and Environmental Engineering, vol. 7, (4), pp. 391-398, 2016.

[148] L. J. Claros-Marfil, J. F. Padial and B. Lauret, "A new and inexpensive open source data acquisition and controller for solar research: Application to a water-flow glazing," Renewable Energy, vol. 92, pp. 450-461, 2016.

[149] W. Nusselt and W. Jürges, "The cooling of a flat wall by an airstream (Die Kühlung einer ebenen wand durch einen Luftstrom)," Gesundh-Ing, vol. 52, (45), pp. 641-642, 1922.

[150] D. D. C. Bernard et al, "A dynamic analysis of ground effect for a quadrotor platform," IFAC-PapersOnLine, vol. 50, (1), pp. 10311-10316, 2017.

Documents

Re-definición de procesos de fabricación en convergencia