Applications of advanced diagnostic and active control for

Applications of advanced diagnostic and active control for

lean combustion

The Department focuses on new technologies and is devoted to promote and to disseminate technology innovation; is involved in:

• Renewable energies• Materials science and technology• ICT• Bio-applications of materials and ICT• Nanotechnologies• Manufacturing technologies• Robotics• Design and testing in Mechanical and Civil Engineering

Giornata di Studio sui Combustori di Turbina a Gas, 5 Novembre 2018, Firenze De Giorgi Maria Grazia

About 100 of staff people teaching in Engineering faculty and Science faculty.An average of 100 of PhD students and Postdocs.On a regional basis our Department in the last years was capable to win about 30% of all supported research projects, competing against other 3 Universities (two in Bari and one in Foggia). Our facilities are on 1500 square meters of laboratories


Green Engine Lab @ Department of Engineering for Innovation 1 Full Professor

2 Associate Professors

1 Permanent Assistant Professor

1 Temporal Assistant Professor

5 Assistant Researchers

Ph.D Students

LASER DIAGNOSTIC LAB

COMBUSTION LAB

AEROSPACE PROPULSION


PARTNERS AND COLLABORATIONS

https://sites.google.com/site/greenenginelab2/propulsion-and-combustion-lab/partners-and-collaborations/apulian-aerospace-district

https://sites.google.com/site/greenenginelab2/propulsion-and-combustion-lab/partners-and-collaborations/apulian-aerospace-district

https://sites.google.com/site/greenenginelab2/propulsion-and-combustion-lab/partners-and-collaborations/network-of-apulian-scientific-labs

https://sites.google.com/site/greenenginelab2/propulsion-and-combustion-lab/partners-and-collaborations/network-of-apulian-scientific-labs

https://sites.google.com/site/greenenginelab2/propulsion-and-combustion-lab/partners-and-collaborations/vki_logo_blue_rectangular_557x156.jpg?attredirects=0

https://sites.google.com/site/greenenginelab2/propulsion-and-combustion-lab/partners-and-collaborations/vki_logo_blue_rectangular_557x156.jpg?attredirects=0

https://sites.google.com/site/greenenginelab2/propulsion-and-combustion-lab/partners-and-collaborations/its.jpg?attredirects=0

https://sites.google.com/site/greenenginelab2/propulsion-and-combustion-lab/partners-and-collaborations/its.jpg?attredirects=0

https://sites.google.com/site/greenenginelab2/propulsion-and-combustion-lab/partners-and-collaborations/cmcc.png?attredirects=0

https://sites.google.com/site/greenenginelab2/propulsion-and-combustion-lab/partners-and-collaborations/cmcc.png?attredirects=0

GREEN ENGINE- FacilitiesLASER DIAGNOSTIC TECHNIQUES:

• Stereo Particle Image Velocimetry• Laser Doppler Velocimetry• Phase Doppler Particle Analyzer• High speed visualization systems (UV- VIS-

NIR)• Hot Wire measurements• Acetone LIF• Smoke Visualizations

PARTICLE IMAGE VELOCIMETRY & FLOW

VISUALIZATIONS

LASER DOPPLER VELOCIMETRY


GREEN ENGINE- Combustion Facilities

SPRAY CHARACTERIZATIONHIGH-PRESSURE GAS-FUELED COMBUSTOR

LIQUID-FUELEDCOMBUSTOR



LIQUID-FUELED SWIRLING COMBUSTOR @ Green Engine Lab

1. Flame stability monitoring and characterization through VIS, UV-VIS and NIR Flamedigital imaging

2. Flame stability monitoring by statistical and spectral analysis of flame parameters

3. Chemiluminescence images by ICCD

4. Acetone LIF, Particle Image Velocimetry (PIV), Hotwire measurements and LaserDoppler Velocimetry (LDV) measurements

5. Flame diagnostic technique based on Machine Learning Techniques



LIQUID-FUELED SWIRLING COMBUSTOR

UV-VIS-NIR images INTENSIFIED CAMERA Phantom M320 High speed camera MemrecamIntensified camera LaVisionNIR-FLIR camera

OH*/CH* chemiluminescence emissions PMTSS of Thorlabs® equipped with an interference filter at 307 nm for the OH* signal acquisition with 10 nm FWHM (10 KHz)

Pollutant emissions (NOX, SO2, CO, CO2 and O2) Complete analyzer system PG-350E Horiba equipped with gas sampling, sample conditioning, analyzer and system control unit


EXPERIMENTAL SETUP



MAIN AIMS Development of diagnostic techniques “early detection”

and monitoring flow unsteadiness

Implementation of supervision and control algorithms toincrease safety and reliability of combustors

Development of active control of instabilities in leancombustors

Investigation of the effect of injection mode on flame leanblowout : comparison of non premixed and partiallypremixed combustion mode

Validate numerical models of combustion instabilities, fora better description of the phenomena


LBO MARGIN SENSING APPROACHES

Instability precursor sensors for active/reactive control systems safely and more efficiently lean combustors, operational flexibility, reliability and availability (i.e., downtime for inspections, component repairs and unplanned shutdowns).

Sensing methodology identifying LBO precursors

Transition from stable combustion to LBO transient regime with localized flame extinctions and reignitions

Close to LBO large scale flame pulsations

Large scale pulsations noise, changes in radiative emissions and cyclic thermal loads precursors for LBO



GEOMETRICPARAMETERS

LUMINOUSPARAMETERS

TIME AND FREQUENCY

PARAMETERSFlame shapeVolume integralSurface areaCircularity factorLength

BrightnessNon UniformityChemiluminescenceemissions

Flame oscillationfrequency• SPECTRAL OR WAVELET-

BASED TIME FREQUENCY ANALYSIS

• POD DECOMPOSITION TECHNIQUES


COMPARING COMBUSTION MODE CO starts at leaner condition (=0.35) in the partially premixed combustion . The maximum value of CO is about the same for the two combustion modes. NOx emissions are significantly higher in non-premixed mode compared to partially premixed

regime.

Pollutant emissions (NOX, CO)

If NOx aim ofthe lean combustion

When instability occurs CO


HIGH SPEED CCD ACQUISITIONS

VIS

NIR

17Giornata di Studio sui Combustori di Turbina a Gas, 5 Novembre 2018, Firenze De Giorgi Maria Grazia

Example VIS acquisition, non-premixed regime

High φ: flame looks

stable

If φ flame is

unstable and a relevant

non-uniformities as well

as geometrical and

luminous instabilities of

the flame are evident


Example VIS acquisition partially-premixed regime


Normalization

Mean

Variance

TIME SERIES OF IMAGES

TIME SERIES OF PIXEL INTENSITIES

STATISTICAL ANALYSIS

SPECTRAL ANALYSIS

PSD

Wavelet decomposition

Energy content

Wavelet entropy

SHAPE ANALYSIS

Heywood

circularity factor


SHAPE ANALYSIS: HEYWOOD CIRCULARITY FACTOR

NON-PREMIXED MODE PART. PREM. MODE


VARIANCE of VIS acquisition, non-premixed


The instability becomes evident

VARIANCE of VIS acquisition, non-premixed

Flame fluctuations lead to substantial variations in luminosity in the zone further away from the injector where the effects of the mixture fluctuations are more evident


Spatially averaged values of VIS acquisitions variance

Variance for below 0.25 for the non-premixed combustion regime and 0.20 for the partially-premixed mode

The premixing more promising to reduce emission because it permits to reach leaner condition than in non-premixed strategies


One-dimensional Discrete Wavelet Transform decomposition (Daubechieswavelet decomposition filters) was applied to the single-pixel signals and to the PMT signals to analyse different spectral ranges

A 1

O rig ina l s ignal f

D 1

A 2 D 2

A 3 D 3

A 4 D 4

A 5 D 5

A 6 D 6

A 7 D 7

A 8 D 8

A 1


D 1

A 2 D 2

A 3 D 3

A 4 D 4

A 5 D 5

A 6 D 6

A 7 D 7

A 8 D 8


Discrete Wavelet Transform decomposition

It is possible to analyze the energy content of the detail components of the decomposed signal. If Ej is the wavelet energy of the j’s decomposition scale of the signal, the probability distribution of energy for each scale is given by pi:

where Ei is defined as the sum of square of detailed wavelet transform coefficients.

A 1


D 1

A 2 D 2

A 3 D 3

A 4 D 4

A 5 D 5

A 6 D 6

A 7 D 7

A 8 D 8

A 1


D 1

A 2 D 2

A 3 D 3

A 4 D 4

A 5 D 5

A 6 D 6

A 7 D 7

A 8 D 8

m

i

iEE1 E

Ep i

i



Wavelet Energy details contents spatially averaged

values from VIS imagesanalysis

Towards LBO Ei in the ranges 42-84 Hz and 21-41 Hz in both the regimes, but while in non-premixed combustion the range 42-84 Hz is dominant, the frequency range 21-41 Hz is more energetic in the partially premixed combustion.





=0.66 =0.57 =0.40

=0.27 =0.19

WAVELET ENTROPY MAPS of visible images

m

1iii

ppWEE )log(

m

i

iEE1 E

Ep i

i

Variance Non premixed regime

WD Ei components

STATISTICAL AND SPECTRAL ANALYSIS OF THE PMT SIGNAL

If =0.21 Ei mostly distributed in low-frequency spectral ranges.

Towards LBO ( =0.13), the contribution of the lowest frequency range decreases, while the frequencies in the range between 39 and 78 Hz become more relevant


COMBUSTIONPMT acquisition CH*/OH* emissions

CaseOH*

Mean Value

OH* Variance

CH*MeanValue

CH* Variance

CH*/OH*

A 4.903 0.050 23.300 0.065 4.752

B 5.008 0.049 14.810 0.161 2.957

C 4.463 0.062 12.175 0.256 2.728

D 5.429 0.042 20.344 0.090 3.747

E 4.143 0.071 16.184 0.153 3.906

Mean CH* intensity and the variance decrease and increase respectively as the blowout is approached. This trend is not so remarkable in the case of OH* emissions.

It may be observed a high variability of the CH* signals.

CH*/OH* ratio, related to the flame’s global heat release, decreases in particular for the non-premixed cases. At the same fuel/air ratio, the CH*/OH* ratio assumes different values from those in the case of partially premixed flames, where it remains almost constant lowering the fuel/air ratio.


COMBUSTIONSTATISTICAL PARAMETER of UV acquisition, CH* emissions

Variance when blowout is approached.


COMBUSTIONDOMINANT FREQUENCY of UV acquisition, CH* emissions

)()/(= a

a

aa

a

DOM fPSDfPSDff PSD (fa)is the component of the power spectral density of the acquired signal for the single frequency

fDOM when blowout is approached. This trend is evident both in the non-premixed and in partially premixed cases.

fDOM lower for the partially premixed cases.


IDENTIFICATION OF FLAME STRUCTURES AND INSTABILITIES BY DECOMPOSITION TECHNIQUES

Proper Orthogonal Decomposition (POD)

• POD is a statistical method that reduces a set of original data into a set of Eigen bases that contain all of the spatial information and constants that contain all of the temporal information.

• POD is based on energy considerations. If the flame has energetic and periodic structures, they will be captured in the first couple of POD modes.

• Therefore, POD can be used to identify the dominant flow structures.

• Each spatial POD mode represents the fluctuations in the high speed flame images, while eigenvalues represent the respective energies of each mode


IDENTIFICATION OF FLAME STRUCTURES AND INSTABILITIES BY DECOMPOSITION TECHNIQUES

Proper Orthogonal Decomposition (POD)It extracts an orthogonal basis of eigenvalues from a snapshots of the flow data field u(x; t).Spatial and temporal contributions decoupled obtaining two different eigenfunctions

where,• Φi(x) → i-th mode eigenface, able to capture the scales and shapes of its modal coherent

structure• ai(t) → i-th mode temporal eigenfunction, it collects information about the dynamics of each

modal coherent structureFast Fourier Transform (FFT) analysis of ai(t) → frequency characterization of the dynamics associated to the corresponding modal structures

STRENGTHRobust techniques to detect the most

energetic coherent structures!

WEAKNESSLoss of accuracy in characterizing the

dynamics of modal structures!Giornata di Studio sui Combustori di Turbina a Gas, 5 Novembre 2018, Firenze De Giorgi Maria Grazia

Dynamic Mode Decomposition (DMD)– based on the Singular Value Decomposition (SVD) –

It decomposes the flow data field u(x; t) into spatial coherent structures, or wave patterns, and oscillatory modes (frequency and decay/growth rate)

where,• Φi

DMD → i-th spatial DMD mode (aka dynamic mode)• λi

DMD ϵ ₵ → i-th oscillatory mode with– arg(λi

DMD) is the i-th mode frequency– || λi

DMD || is the i-th mode decay rate

STRENGTHOne single modal frequency and

growth/decay rate!

WEAKNESSNo ranking criterion of DMD modes!

Modes are not orthogonal!


POD MODE DECOMPOSITION of VIS acquisition


=0.37



=0.37

First modes shows the regions of high and low intensities around the axis of the fuel nozzle. This indicates the rotation of the flame, which is an inherent nature of the swirl-stabilized flame.



=0.18



=0.18

Mode 2 and Mode 4 give an indication of the blowout and reignitionMode 2 indicates longitudinal oscillations in the flame





=0.37Mode amplitude coefficient represents the time evolution of the

deconstructed flame features in each POD mode



=0.18



=0.18

Knowing, the frequency of the problem, the corresponding dynamic modes and growth rate have been identified using DMD decomposition technique….

DINAMIC MODE DECOMPOSITION (DMD) of VIS acquisition


=0.18



=0.18



=0.18

MONITORING, DIAGNOSTIC AND PROGNOSTIC TECHNIQUES

1. Real-time combustion instability detection by signal

processing techniques suitable for both non-linear and non-

stationary phenomena

2. Real-time Health Monitoring of Gas Turbine Combustor Using

Online Learning and High Dimensional Data


TOOL TRAINING

(ANN, SVM,...)

OPERATING

CONDITION

CONTROLLER

HEALTHY COMBUSTORRESIDUAL

GENERATION

TRAINED TOOL(ANN, LS-SVM,..)

OPERATING CONDITION

CONTROLLER

COMBUSTOR

RESIDUAL THRESHOLD

HEALTHY COMBUSTOR

FAULTY COMBUSTOR

>

NO

YES

TOOL TRAINING

?

TOOL ACTIVE


The technical approach is based on

Experimental testing to gain knowledge of the physical processes associated with instability combustion.

Data-driven modeling and machine learning for development of analytics algorithms.

• Dimensionality Reduction Methods (Principal Component Analysis)

• Data-driven classifier: Logistic Regression, Artificial Neural Networks, Support Vector Machines

• Combination of data-driven and physics-based models (hybrid modeling)


Machine Learning based and statistical tools

• CUSUM CONTROL CHART

• BAYESIAN TOOL

• ARTIFICIAL NEURAL NETWORK (ANN)

• SUPPORT VECTOR MACHINE (SVM)

• GENETIC PROGRAMMING TOOL

Different tool can be applied to the acquired data from the combustor.The aim could be the detection of a change point in the signal trend and/or the monitoring ofthe signal to build a forecasting tool.The main implemented tools are:


Machine Learning based and statistical tools

Statistical parameter

Frequency data

Mode decomposition

data

INSTABILITY REGIME

RECOGNITION

INSTABILITY REGIME

RECOGNITION

SUPPORT VECTOR MACHINE

ARTIFICIAL NEURAL

NETWORK (ANN)

GENETIC PROGRAM


PLASMA ACTUATION TO ENHANCE METHANE-AIR LEAN FLAMEIn collaboration with

CNR Nanotec Bari, Italy

Two different burner configurationsMethane activated (central air jet) or air

activated (central fuel jet)


PLASMA TO ENHANCE THE FLAME STABILITY

12/30

CH4AIRCH4

CH4AIR AIR

IDF NDF

Steel tube connected to the grounded electrode and a copper

tube (80 mm long, thickness 0.6 mm), on the outer surface of

the quartz tube connected to the HV electrode


They operate at wide range ofdriving frequenciesDirect current (dc), alternatingcurrent (ac), radio frequency (RF) and microwave

INPUT VOLTAGE WAVEFORM• kV peak-to peak sinusoidal high voltage• Pulsed power supply with a short nanosecond rise

time and pulse duration of tens and hundreds of nanoseconds (Nanosecond Repetitively Pulsed Discharges)

Dielectric Barrier Discharge (DBD)Two electrodes with at least one dielectric in between


HIGH VOLTAGE GENERATOR1) SINUSOIDAL DBD“PVM500 Plasma Resonant and Dielectric Barrier Corona Driver” (0-40 kV peak to peak; 20-70 kHz)

2) NANOSECOND REPETITIVELY PULSED DISCHARGES NPG-18/3500 of MegaImpulse Ltd® (0-80 kV peak to peak; pulse repetition at 3.5 kHz, energy at 30mJ/pulse)


PLASMA TO ENHANCE THE FLAME STABILITY

12/30

A high voltage probe (Tektronix P6015A), a current probe (Bergoz Current Transformer CT-D1.0-B) and an oscilloscope (Tektronix TDS2024C) were used to retrieve the electrical power dissipation


Flow Characterization by LDV, PIV and hotwireTemperature2D chemiluminescence (CH*, OH*, CO2)High speed visualizationElectrical characterization

COMBUSTION ACTIVE CONTROL

12/30

GE Lab., Lecce FLAME STABILIZATION OF METHANE/AIR LEAN FLAME


EXPERIMENTAL SETUP

COLD EXPERIMENTS


Impact of induced jet velocity

Flow Characterization by LDV, PIV and hot wire

Impact of thermal effectsTemperature measurements

12/30

Test Case (l/min) (l/min)

1 7.5 0.6

2 0.6 7.5

3 0 0.6

minner moter



1 7.5 0.6

2 0.6 7.5

minner moter

RESULTS-I-

Temperature profile

Velocity profiles



1 7.5 0.6

2 0.6 7.5

minner moter

RESULTS-II-

Temperature profile

Velocity profiles



1 7.5 0.6

2 0.6 7.5

minner moter

RESULTS-II- EXPERIMENTS

72

Power spectral density at y / D = 0.5 and different radial positions



1 7.5 0.6

2 0.6 7.5

minner moter

RESULTS-II-

73

• Actuation OFF:

• Spectra of the baseline flow are dominated by a

frequency peak at about 200 Hz ( shear layer

vortex shedding frequency, fs, Re≈ 1750, Str ≈ 0.7)

• Actuation ON:

• the energy contained at the dominant (shedding)

frequency.

• A broad range of high frequencies (>103 Hz)

presented an increased energy content.

• Such energy variation can be associated to the

decrease of the formation of the coherent

structures at the vortex shedding frequency and to

the occurrence of an energy transfer from the large

to the small scales.

Power spectral density at y / D = 3 and different radial positions



1 7.5 0.6

2 0.6 7.5

minner moter

RESULTS-II-

74

Power spectral density at y / D = 0.5 and different radial positions

Energy is not concentrated in a specific

frequency band , not a dominant frequency

peak, most probably due to the forced

laminar/turbulent transition in the mixing

length

Spectra are reasonably well fitted by a -5/3

power law as predicted by the Kolmogorov

theory



1 7.5 0.6

2 0.6 7.5

minner moter

RESULTS-II-

75


ACTUATION OFF:PSD magnitude generally increases

moving along y/D.

ACTUATION ONAt the jet centerline location (x/D =0) the

PSD magnitude of the frequencies

lower than 100 Hz decreases when

moving from y/D=0.5 to y/D=3.

Energy content of the frequencies

higher than 100 Hz increases

There is an energy transfer to the

smaller scales along the jet centerline.



1 7.5 0.6

2 0.6 7.5

minner moter

RESULTS-II-

76


Increase in magnitude together with the

relatively constant slope between the

baseline and the actuated flow

suggested that the vrms was enhanced

by the actuation. This vrms modification

could imply an increase of the jet

mixing..



1 7.5 0.6

2 0.6 7.5

minner moter

RESULTS-II-

77







mixing..



1 7.5 0.6

2 0.6 7.5

minner moter

RESULTS-II-

78







mixing..


RESULTS: TEMPERATURE PROFILES

Temperature profiles taken at y=5 (mm): (a) test case 1, (b) Test case 2, (c) test case 3

)TT(cmQ inoutp

AVm


1 7.5 0.6

2 0.6 7.5

3 0 0.6

minner moter


COMBUSTIONMETHANE ACTIVATED - BLOWOUT LIMIT

Sinusoidal HV (20 kHz) NRPP HV (1.75 kHz)


METHANE ACTIVATED - BLOWOUT LIMITNANOSECOND REPETITIVELY PULSED DISCHARGES

BROADBAND CHEMILUMINESCENCE

Effect of plasma on the flame shape fuel flow rate = 0,2 l/minBaseline air flow rate = 5,2 l/min

Variation of applied voltage fixing the repetition frequency to 1750 Hz


Baseline case IDF: OH* emissions mainly occur in an annular ring around the centerline

Plasma power : Increase of the emission intensity in proximity of the quartz exit. Axial movement of the area with maximum OH* intensity upstream towards the quartz exit

COMPARISONS BETWEEN IDF AND NDF FOR TWO DIFFERENT STANDOFF DISTANCE

12/30


METHANE ACTIVATED SINUSOIDAL PULSED

DISCHARGES

CO2

CHEMILUMINESCENCE EMISSIONS

𝑽 𝒂𝒊𝒓 = 𝟑.𝟓 𝒍/𝒎𝒊𝒏, 𝑽 𝑪𝑯𝟒= 𝟎.𝟔 𝒍/𝒎𝒊𝒏

0.0 W (21.50.9) W (27.21.1) W (19.00.8) W (23.61.0) W s=0 mm s=0 mm s=6 mm s=6 mm

Ve

rtic

al d

imen

sio

n [

mm

]

𝑽 𝒂𝒊𝒓 = 𝟕.𝟓 𝒍/𝒎𝒊𝒏, 𝑽 𝑪𝑯𝟒= 𝟎.𝟔 𝒍/𝒎𝒊𝒏

0.0 W (22.00.9) W (27.41.2) W (19.80.8) W (25.71.1) W

s=0 mm s=0 mm s=6 mm s=6 mm

Ve

rtic

al d

imen

sio

n [

mm

]

𝑽 𝒂𝒊𝒓 = 𝟐.𝟑𝟓 𝒍/𝒎𝒊𝒏, 𝑽 𝑪𝑯𝟒

= 𝟎.𝟒 𝒍/𝒎𝒊𝒏

0.0 W (21.20.9) W (26.61.1) W (20.10.9) W (25.61.1) W s=0 mm s=0 mm s=6 mm s=6 mm

Ve

rtic

al d

imen

sio

n [

mm

]

Horizontal dimension [mm]


PROPER ORTHOGONAL DECOMPOSITION (POD) ANALYSIS

12/30

84

The addition of plasma discharge results in decreased heat release fluctuations, as

shown by the overall reduction in the energy content of the other dominant modes

i.e., Modes 1 - 10.


GREEN ENGINE

85

HIGH PRESSURE CHAMBER (up to 30 bar)

Chamber dimensions 200x120x120, 575 cm3.Optical windows: 70 mm and 90 mm


EXPERIMENTAL SETUP AND ELECTRICAL CONNECTIONS

86

HIGH PRESSURE CHAMBER EXPERIMENTS200 x 120 x 120 mm, internal volume 573 cm3


HIGH PRESSURE CHAMBER EXPERIMENTS: TEST CASES

88

elP Actuator dissipated power (W).af Actuation frequency (kHz). Vpp Peak-to-peak applied voltage (kV). Q Flame’s thermal power (W)


PRESSURE CHAMBER EXPERIMENTS: RESULTS

89

MEAN (a) AND RMS (b) INTENSITY MAPS

elP

Different plasma levels have been compared for the test case at 3 bar absolute and =0.1 (test cases 9-11).

For the highest dissipated power, 0.7 W (test case 11), less than 0.5% of the flame’s thermal power, the plasma effect is quite evident, implying a better anchoring of the flame compared with the not actuated test case (test case 9), even if the RMS intensity increases consistently.


FUTURE INVESTIGATIONS

90

Preliminary literature studies underlined that nanosecond Repetitively Pulsed (NRP) discharges can be used as an actuator for active control of combustion instabilities, without the drawbacks of traditional actuators such as loudspeakers or fuel valves.

NRP discharges provide unsteady heating and species dissociation on the nanosecond timescale. This generates pressure waves propagating from each discharge

Low-frequency sound generation by modulated repetitively pulsed nanosecond plasma discharges Olaf Bölke et al 2018 J. Phys. D: Appl. Phys. 51 305203


91

Thank You for

Attention


Documents

Applications of advanced diagnostic and active control for