Flywheel Model

8/12/2019 Flywheel Model

1/77

MODELING FLYWHEEL-SPEEDVARIATIONS BASED ON

CYLINDER PRESSURE

Masters thesisperformed in Vehicular Systems

byMagnus Nilsson

Reg nr: LiTH-ISY-EX-3584-2004

30th March 2004


2/77


3/77

MODELING FLYWHEEL-SPEEDVARIATIONS BASED ON

CYLINDER PRESSUREMasters thesis

performed in Vehicular Systems ,Dept. of Electrical Engineeringat Link opings universitet

by Magnus Nilsson

Reg nr: LiTH-ISY-EX-3584-2004

Supervisor: Mats J argenstedtScania

Associate Professor Lars ErikssonLinkopings Universitet

Examiner: Associate Professor Lars ErikssonLinkopings Universitet

Linkoping, 30th March 2004


4/77


5/77

Avdelning, InstitutionDivision, Department

DatumDate

Sprak

Language

Svenska/Swedish Engelska/English

RapporttypReport category

Licentiatavhandling Examensarbete

C-uppsats D-uppsats Ovrig rapport

URL f or elektronisk version

ISBN

ISRN

Serietitel och serienummerTitle of series, numbering

ISSN

Titel

Title

F orfattareAuthor

SammanfattningAbstract

NyckelordKeywords

Combustion supervision by evaluating ywheel speed variations is acommon approach in the automotive industry. This often involves pre-liminary measurements. An adequate model for simulating ywheelspeed can assist to avoid some of these preliminary measurements.

A physical nonlinear model for simulating ywheel speed based oncylinder pressure information is investigated in this work. Measure-ments were conducted at Scania in a test bed and on a chassis dy-namometer. The model was implemented in Matlab /Simulink andsimulations are compared to measured data. The rst model can notexplain all dynamics for the measurements in the test bed so extendedmodels are examined. A model using a dynamically equivalent model of the crank-slider mechanism shows no difference from the simple model,whereas a model including a driveline can explain more from the test-bed measurements. When simulating the setups used at the chassisdynamometer, the simplest model works best. Yet, it is not very ac-curate and it is proposed that optimization of parameter values mightimprove the model further. A sensitivity analysis shows that the modelis fairly robust to parameter changes.

A continuation of this work might include optimization to estimateparameter values in the model. Investigating methods for combustionsupervision may also be a future issue.

Vehicular Systems,Dept. of Electrical Engineering581 83 Link oping

30th March 2004

LITH-ISY-EX-3584-2004

h ttp://www.vehicular.isy.liu.seh ttp://www.ep.liu.se/exjobb/isy/2004/3584/

MODELING FLYWHEEL-SPEED VARIATIONS BASED ONCYLINDER PRESSURE

ATT MODELLERA SV ANGHJULSHASTIGHET BASERAT P ACYLINDERTRYCK

Magnus Nilsson

combustion supervision, cylinder balancing, physical model, cylinderpressure, ywheel speed, crankshaft


6/77


7/77

Abstract

Combustion supervision by evaluating ywheel speed variations is acommon approach in the automotive industry. This often involves pre-liminary measurements. An adequate model for simulating ywheelspeed can assist to avoid some of these preliminary measurements.

A physical nonlinear model for simulating ywheel speed based oncylinder pressure information is investigated in this work. Measure-ments were conducted at Scania in a test bed and on a chassis dy-namometer. The model was implemented in Matlab /Simulink andsimulations are compared to measured data. The rst model can notexplain all dynamics for the measurements in the test bed so extendedmodels are examined. A model using a dynamically equivalent model of the crank-slider mechanism shows no difference from the simple model,whereas a model including a driveline can explain more from the test-bed measurements. When simulating the setups used at the chassisdynamometer, the simplest model works best. Yet, it is not very ac-curate and it is proposed that optimization of parameter values mightimprove the model further. A sensitivity analysis shows that the modelis fairly robust to parameter changes.

A continuation of this work might include optimization to estimateparameter values in the model. Investigating methods for combustionsupervision may also be a future issue.

Keywords: combustion supervision, cylinder balancing, physical mo-del, cylinder pressure, ywheel speed, crankshaft

v


8/77


9/77

Preface

This work constitutes a masters thesis at the Division of VehicularSystems, Link opings universitet. The work was carried out duringseptember to november 2003 under the supervision of Mats J argen-stedt at Scania, S odert alje. From november 2003 to march 2004, thework was completed at Link opings universitet under the supervision of Assoc. Prof. Lars Eriksson. It was nancially supported by Scaniawhich has been much acknowledged.

Thesis Outline

Chapter 1 gives an introduction to the problem and outlines the objectives of this thesis.

Chapter 2 summarizes the model described in [ 14].Chapter 3 describes measurements carried out at Scania.Chapter 4 discusses results from simulations for various models.Chapter 5 sums up conclusions from the present work.Chapter 6 discusses future work.Appendix A contains information related to the derivation of ex-

tended models.

Acknowledgment

I would like to thank my supervisors, Assoc. Prof. Lars Eriksson at theDivision of Vehicular Systems, Link opings universitet, and Mats J ar-genstedt at Scania. Lars gave me the oppurtunity to work with thisproblem and contributed with well-needed advise. Mats took care of me at Scania and gave me a good introduction to the problem and themajor outline of my assignment. I am also grateful for the assistancegiven at the measurements in the test-bed environment and at the chas-sis dynamometer. I would further like to thank Mats Henriksson forintroducing me to Rotec and Henrik Pettersson for assistance at thechassis dynamometer. General appreciation is directed to all peopleat Scania that assisted me during my stay in S odert alje. A specialthanks to people at NEE, Scania, and the Division of Vehicular Sys-tems, Link oping, for giving up time to answer questions related to mywork. Finally, much thanks to my family for your constant love andsupport.

vii


10/77


11/77

Contents

Abstract v

Preface and Acknowledgment vii

1 Introduction 11.1 Combustion Supervision . . . . . . . . . . . . . . . . . . 11.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 The Model 32.1 Torque due to Cylinder Pressure . . . . . . . . . . . . . 32.2 Torque due to Motion of Masses . . . . . . . . . . . . . 4

2.3 The Torque-Balancing Equation . . . . . . . . . . . . . 52.4 Crankshaft Dynamics . . . . . . . . . . . . . . . . . . . 52.5 A Time-Domain State-Space Mode l . . . . . . . . . . . . 8

3 Measurement s 93.1 Planning the MeasurementsThings to Consider . . . . 9

3.1.1 Measuring Cylinder Pressure . . . . . . . . . . . 93.1.2 Measuring Flywheel Speed . . . . . . . . . . . . 93.1.3 The Importance of a High Sample Rate . . . . . 10

3.2 Using Indiscope . . . . . . . . . . . . . . . . . . . . . . . 113.3 Using Rotec . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.3.1 Measuring Cylinder Pressure With Rotec . . . . 143.3.2 Transforming Domains for Measured Signals . . 143.3.3 A Short Analysis of the Constructed Signals . . . 183.3.4 Reason for the Noisy Speed Signal . . . . . . . . 183.3.5 Speed ComparisonS6 and Rotec . . . . . . . . 19

3.4 The Chassis Dynamometer . . . . . . . . . . . . . . . . 213.4.1 Speed Comparison Mike and Moa . . . . . . . 223.4.2 Speed ComparisonTrucks and Rotec . . . . . . 22

ix


12/77

4 Simulations 274.1 The Simple Mode l . . . . . . . . . . . . . . . . . . . . . 284.2 A Dynamically Equivalent Model . . . . . . . . . . . . . 314.3 Including a Model of the Driveline . . . . . . . . . . . . 324.4 A Truck-Driveline Mode l . . . . . . . . . . . . . . . . . . 35

4.4.1 Speed ComparisonSimple- and Driveline Model 354.4.2 Speed ComparisonDriveline Model and Rotec . 38

4.5 Trial-and-Error Modeling . . . . . . . . . . . . . . . . . 404.5.1 An Alternative Damper Mode l . . . . . . . . . . 404.5.2 Manipulating Stiffness Parameters . . . . . . . . 40

4.6 Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . 44

5 Conclusions 515.1 Conclusions from the Measurements . . . . . . . . . . . 515.2 Conclusions from the Simulations . . . . . . . . . . . . . 52

6 Future Work 53

References 55

A Equations for Extended Models 57A.1 A Dynamically Equivalent Model of the Connecting Rod 57A.2 A Simple Truck-Driveline Model . . . . . . . . . . . . . 59

x


13/77

List of Tables

3.1 Operating PointsFirst Measurement . . . . . . . . . . 123.2 Operating PointsSecond Measurement . . . . . . . . . 143.3 Operating PointsThird Measurement . . . . . . . . . . 21

xi


14/77


15/77

List of Figures

2.1 The Crank-Slider Mechanism . . . . . . . . . . . . . . . 42.2 The Lumped Mass Mode l . . . . . . . . . . . . . . . . . 6

3.1 Circular Disc and Flywheel . . . . . . . . . . . . . . . . 103.2 Schematic Engine Setup in Test Beds . . . . . . . . . . . 113.3 Indiscope Speed Signa l . . . . . . . . . . . . . . . . . . . 123.4 Measured Signals with Rotec . . . . . . . . . . . . . . . 133.5 Adjusting the Measured Pressure Curve . . . . . . . . . 153.6 Cancellation Phenomenon in the Speed Signal . . . . . . 163.7 Transforming Domain for the Pressure Signal . . . . . . 173.8 Cycle-to-Cycle Variations in Pressure . . . . . . . . . . . 183.9 Noisy Speed Signal From Rotec . . . . . . . . . . . . . . 193.10 Speed ComparisonS6 and Rotec, 1500 rpm . . . . . . 203.11 Chassis Dynamometer Setup . . . . . . . . . . . . . . . 213.12 Speed Comparison Mike and Moa , 1500 rpm . . . . . 223.13 Speed Comparison Mike and Moa , 1000 rpm . . . . . 233.14 Speed Comparison Mike and Rotec, 1500 rpm . . . . 243.15 Speed Comparison Moa and Rotec, 1900 rpm . . . . . 25

4.1 SimulationSimple Model, 1000 rpm . . . . . . . . . . 284.2 SimulationSimple Model, 1500 rpm . . . . . . . . . . 294.3 SimulationSimple Model, 1900 rpm . . . . . . . . . . 304.4 ComparisonSimple- and DE Model, 1900 rpm . . . . . 314.5 The Lumped Mass Model Including a Driveline . . . . . 324.6 Transient Response when a Driveline is Included . . . . 33

4.7 SimulationTest Bed Driveline, 1500 rpm . . . . . . . . 334.8 SimulationTest Bed Driveline, Offset Engine . . . . . 344.9 ComparisonSimple- and Malte Model, 1000 rpm . . 364.10 ComparisonSimple- and Malte Model, 1900 rpm . . 374.11 Simulation Malte Model, 1000 rpm . . . . . . . . . . 384.12 Simulation Malte Model, 1900 rpm . . . . . . . . . . 394.13 SimulationAlternative Damper Model . . . . . . . . . 414.14 SimulationModied Stiffness . . . . . . . . . . . . . . 424.15 Sensitivity Analysis, Offset Pressure . . . . . . . . . . . 44

xiii


16/77

4.16 Sensitivity Analysis, Scaled Pressure . . . . . . . . . . . 454.17 Sensitivity Analysis, Translated Pressure . . . . . . . . . 464.18 Sensitivity Analysis, Stiffness Parameters . . . . . . . . 474.19 Sensitivity Analysis, Inertias . . . . . . . . . . . . . . . . 48

A.1 The Crank-Slider Mechanism . . . . . . . . . . . . . . . 58A.2 A Vehicular Driveline . . . . . . . . . . . . . . . . . . . 59A.3 Free-Body Diagram of a Driveline . . . . . . . . . . . . . 59

xiv


17/77

Chapter 1

Introduction

1.1 Combustion Supervision

The electronic control system which covers at least the functioning of the fuel injection and ignition is called the Engine Management Sys-tem. One objective for the system is to supervise cylinder combustionin order to avoid undesirable vibrations in parts of the engine. Undesir-able vibrations in the crankshaft can arise from defective fuel injectors.There have been many articles and books written on how to discovercontrol this phenomenon [ 5, 6, 9, 15].

A common approach on combustion supervision and cylinder bal-ancing demands properties of the transfer function from cylinder pres-sures to ywheel speed. Thus, the method involves preliminary mea-surements on the engine.

A general physical model that can simulate ywheel speed usingcylinder pressure as input can facilitate to gain more information onthe transfer function. The focus of this thesis has been to implementand examine the properties of one such physical model.

1


18/77

2 INTRODUCTION

1.2 Objectives

The general objectives were to

Implement a basic mathematical model in Simulink that can sim-ulate ywheel speed, given cylinder pressures.

Carry out measurements that hold enough data to test and verifythe mathematical model.

If those objectives were met, the next step would be to

Compare the basic mathematical model with some extended mod-els with a focus on examining torque contributions from drive-lines.

or

Investigate algorithms for cylinder pressure supervision.

The rst two objectives were reached, although it was hard to judgethe validity of the model. Then, comparisons with other models weremade with an emphasis on driveline models. No time was spent oninvestigating cylinder pressure supervision.

Much work was spent on programming and a toolbox evolved asa corollary of working with the models in Matlab . See [12] for more

details.


19/77

Chapter 2

The ModelThis chapter summarizes the model developed by Schagerberg andMcKelvey [14], and it follows their presentation to a high degree. Thechapter may be skipped if familiar to the reader.

The model is a physical nonlinear lumped mass model which com-prises the damper, the crankshaft, the cylinders and the ywheel. Givenpressures as a function of crank angle, an equation for the equivalenttorque on the crankshaft may be set up, based on cylinder and crank-slider information. Together with a model of the crankshaft, a torquebalance equation is used to obtain a differential equation that may beimplemented in e.g. Simulink.

2.1 Torque due to Cylinder Pressure

We dene the differential gas pressure , pg (), as the difference betweenthe absolute pressure inside the combustion chamber and the counter-acting pressure on the back side of the piston. When the gas pressureis multiplied by the piston area, A p , we get the force that acts on the

piston along the cylinder axis. The gas pressure is further transformedinto the gas torque , T g (), on the crankshaft by the crank-slider mech-anism,

T g() = pg ()A pdsd

(2.1)

where denotes the crank angle and s denotes the piston displacement see gure 2.1. The derivation of dsd may be found in e.g. [9].

3


20/77

4 THE MODEL

s

l

A

Br

r + l

Figure 2.1: The crank-slider mechanism.

2.2 Torque due to Motion of Crank-SliderMechanism Masses

The moment of inertia is a function of both mass and position. Inthe case of a crank-slider mechanism where the geometry changes, themoment of inertia will also vary. The term varying inertia will be usedfor this effect. The piston motion is assumed to be purely translationalalong the cylinder axis, why it dynamically can be described by a single

point mass along this axis. Describing the motion of the crank-slidermechanism is more complex since it undertakes both translational androtational motion. For a dynamically equivalent model of the connectingrod, three requirements must be satised:

1. The total mass of the model must be equal to that of the originalbody.

2. The center of gravity must be the same as for the original body.

3. The moment of inertia must be equal to that of the original body.

A common approximation is to consider a statically equivalent model

inwhich the last requirement is not fullled. A statically equivalent modelis used in most of this work. Thus, the piston and connecting rod areapproximated by two point massesone reciprocating, mA , and onerotating, mB placed at the centers of the piston pin and crank pinrespectively as in gure 2.1. A dynamically equivalent model is brieyexamined in section 4.2.

The mass torque may be derived in different ways. In [ 9] it is doneconsidering the kinetic energy of the two point masses, mA and mB .


21/77

2.3. THE TORQUE-BALANCING EQUATION 5

The resulting equation may be expressed as

T m (, , ) = (J A () + mB r 2 ) 12

dJ A ()d

2 , (2.2)

where the varying inertia of mA with respect to the crankshaft axis,J A (), and its derivative with respect to are

J A () = mAdsd

2(2.3)

dJ A ()d

= 2mAd2sd2

dsd

. (2.4)

The expressions for the piston displacement may be found in [ 9].

2.3 The Torque-Balancing Equation

Summing up the torque contribution from a single cylinder gives ascalar differential equation, often referred to as the torque-balancingequation

J = T g() + T m (, , ) + T f () + T l (), (2.5)

where J is the crankshaft inertia, T f () the friction torque and T l ()the load torque. The mass torque, T m (, , ) is given in equation ( 2.2).

The instantaneous friction torque, T f (), is modeled as viscous

dampers. Other torques that contribute to the crankshaft torque, suchas the driving of auxiliary systems of the engine, are neglected in themodel.

2.4 Crankshaft Dynamics

The main reason for using a lumped mass model of the crankshaft asthe physical model in this work, is because such models are developedas standard procedure in the design phase of the crankshafts. Theobjective is then however to calculate the maximum torsional stressesin the shaft to mitigate crankshaft failure and for Noise Vibration andHarshness issues.

A multi-body extension of the torque-balancing equation ( 2.5) maybe expressed as

J + C + K = T g( ) + T m ( ) + T f ( ) + T l ( ), (2.6)

where is now a vector. In equation ( 2.6), J , K and C are symmetricmatrices and referred to as the inertia-, stiffness- and damping matricesrespectively. These matrices are all of size N N , where N is the num-ber of lumped masses. The damping elements are modeled as viscous


22/77

6 THE MODEL

damping . Stiffness and damping elements interconnected between adjacent lumped masses are here referred to as relative , and if connectedbetween an inertia lump and a non-rotating reference absolute . Formodeling of an engine crankshaft, typically both absolute and relativedamping elements are used but only relative stiffness elements since thecrankshaft is free to rotate about its axis.

In gure 2.2, such a model for the engine is outlined. Each relative

00110011 001100110011 0011001100110011001100110011001100110011 001100110011

J 1 J 2J 3 J 4 J 5 J 6 J 7 J 8

J 9

c1,2 c2,3 c3,4 c4,5 c5,6 c6,7 c7,8 c8,9

k1,2 k2,3 k3,4 k4,5 k5,6 k6,7 k7,8 k8,9

c3 c4 c5 c6 c7 c8

DamperFree end

Cyl 1 Cyl 2 Cyl 3 Cyl 4 Cyl 5 Cyl 6

Flywheel

Figure 2.2: The lumped mass model with interconnected stiffnesses and damping. Also absolute damping is included.

stiffness- or damping element adds a 2 2 block matrix in the diagonalof K and C respectively. The block has the stiffness- or damping coef-cient on the diagonal and the negative stiffness- of damping coefficienton the anti-diagonal. The load torque is assumed to be constant withrespect to the time scale considered in this work. It is also assumed tobe applied on the last mass in the crankshaft model. This simpliesthe load torque into the constant vector,

T l = (0 0 . . . 0 T l )T . (2.7)

Friction torque is only modeled as viscous absolute damping elements.It is incorporated in the damping matrix C and consequently T f ( ) =0 .

To describe the cylinder positions in the model, dene the selection matrix S of size N N c , where N is the number of masses in thecrankshaft model and N c is the number of cylinders in the engine. Thematrix S has ones in positions ( nmass , n cyl ), for ncyl = 1 , . . . , N c . For asix-cylinder engine with cylinders in positions 38 in a nine-mass model,


23/77

2.4. CRANKSHAFT DYNAMICS 7

S becomes

S = ( s 1 . . . s N c ) =

0 0 0 0 0 00 0 0 0 0 01 0 0 0 0 00 1 0 0 0 00 0 1 0 0 00 0 0 1 0 00 0 0 0 1 00 0 0 0 0 10 0 0 0 0 0

, (2.8)

where s i denotes the i:th column of S . To simplify notation, the expres-sions for the gas and mass torques of a single cylinder (equations ( 2.1)and ( 2.2)) are used to dene the following three geometrical functions,

g1() = A pdsd

(2.9)

g2() = 1

2J a () = mA

d2sd2

dsd

(2.10)

g3() = J A () = mAdsd

2. (2.11)

In a multicylinder engine the cylinder events are phased by the dif-ference in ring angle between the cylinders. For instance, in the six-cylinder engines examined in this work the ring sequence is 153624 and the crankshaft turns 120 between each ring. Thus dene thephasing vector as

= ( 1 . . . N c )T . (2.12)

The angles of the cranks reect this phasing. For the multicylindercase, the geometrical functions gi (), i = 1 , 2, 3 in (2.92.11) may beused to dene diagonal matrix functions,

G i (S T ) = diag gi (s T 1 1), . . . , g i (sT N c N c ) . (2.13)

The gas torque in the multi-body model may now be written

T g ( ) = SG 1 (S T ) p g (S T ). (2.14)

Here, p g () is an N c 1 vector function of the individual gas pressures.See equation ( 2.1) for the denition of gas torque in the single cylin-der case. The multi-body extension of the single-cylinder mass torquedened in equation ( 2.2) becomes

T m ( , , ) = SG 3(S T )S T + mB r 2 SS T

SG 2(S T )S T , (2.15)

where the symbol means elementwise multiplication.


24/77

8 THE MODEL

2.5 A Time-Domain State-Space Model

The system of N second-order differential equations in equation ( 2.6)may be transformed into a system of 2 N rst-order differential equa-tions e.g. by dening the state vector consisting of angles and rotationalspeeds,

x = ( x T 1 xT 2 )

T = T T T . (2.16)

For convenience, the state vector is partitioned in position states x 1and speed states x 2 .

Now, to get to the state-space description of the model, we splitthe mass torque T m ( , , )see equation ( 2.15)in the two partsmultiplying angular acceleration and speed squared respectively, anddene

T m, 1( , ) = SG 3(S T )S T + mB r 2 SS T (2.17)

T m, 2( , ) = SG 2(S T )S T . (2.18)

We also dene the varying inertia by collecting parts multiplying an-gular acceleration in equation ( 2.6), which give

J ( ) = J + mB r 2 SS T + SG 3(S T )S T . (2.19)

The torque-balancing equation may now be reformulated as

J ( ) = K C + T m, 2( , ) + T g ( ) + T l . (2.20)

Stacking the identity equation = I together with equation ( 2.20)premultiplied by J ( )

1 , the dynamics equation of the state-spacemodel becomes

x = 0 I

J (x 1) 1K J (x 1)

1C x 1x 2

+ 0

J (x 1) 1

SG 2(S T x 1 )S T x 2 x 2

+ 0

J (x 1) 1

SG 1(S T

x 1 ) p g(S T

x 1) + T l. (2.21)

Equation ( 2.21) is divided in three terms to visualize the different con-tributions. The rst term represents the crankshaft dynamics, the sec-ond term the piston-crank inertia effects, and the last term the inputsignals. Equation ( 2.21) and variations of it are used for simulations inthis work. See section A.1 for more details about variations of equa-tion ( 2.21).


25/77

Chapter 3

Measurements

We want to examine if a driveline has to be included to correctly sim-ulate ywheel speed. In order to investigate this, measurements weremade on heavy trucks mounted on a chassis dynamometer. However,another series of measurements was rst conducted in a test bed toacquire pressure curves with synchronized speed signals.

3.1 Planning the MeasurementsThings to Consider

3.1.1 Measuring Cylinder Pressure

We want to use measured or simulated cylinder pressure curves as inputdata to our model. This assumes that cycle-to-cycle variations aresmall, which should hold well for diesel engines, or that the model isrobust with respect to input data. This should be veried to as a largeextent as possible.

3.1.2 Measuring Flywheel Speed

There are two ways to measure ywheel speed in the test-bed environ-ment:

1. The ywheels in Scanias heavy trucks have 60 bores 1 equidis-tantly drilled in the rim. A built-in hardware is used to measuretime differences between the holes as they pass a sensor. An arrayof time differences may be extracted, where the rst element cor-responds to the time difference between top dead center (TDC)for cylinder 1 and the hole 6 thereafter. From this time-stamp

1 One hole is virtual to allow for TDC localization.

9


26/77

10 MEASUREMENTS

array, ywheel speed as a function of ywheel position may becalculated. The built-in hardware is called S6 and samples at arate of 5 MHz.

2. There is a circular disc attached to the ywheel in the test bed.This circular disc, which is depicted in gure 3.1, holds 720 mark-ers, or teeth as we will call them. The test beds are setup suchthat a transistor-transistor-logic (TTL) pulse-train signal may beextracted where each pulse correspond to a passed tooth. 2 Thedisc also holds a trigger tooth that (when passed) can be usedas a reference to keep track of what angle we are on. 3 In thesame way as S6 does, we may create an array of time differencesbetween adjacent teeth and thereafter calculate ywheel speed asa function of ywheel position.

In the test bed, speed from both the ywheel and the circular discare measured. One thing that should be investigated is if there aredynamics to be considered between the two rotating wheels.

Driveline

Circular discFlywheel

Crankshaft

Bore

Figure 3.1: A schematic view of the circular disc and ywheel in the test beds. Flywheel speed may be extracted from the built-in hardware (S6) and/or by measuring on the circular disc.

3.1.3 The Importance of a High Sample Rate

We want to measure in-cycle speed variations which require a high sam-ple rate on the TTL pulse-train signal extracted from the circular disc.How high can be roughly estimated using some simple calculations.

2 This is not the whole truth. There is equipment in the test bed that extrapolatespulses to the TTL signal. This is discussed in section 3.3.4.

3 Note that we can not be sure which phase in the cycle we are in when we passthe trigger tooth by only looking at the speed signal. The phase in the cycle isfound out by also looking at a synchronized pressure curve.


27/77

3.2. USING INDISCOPE 11

Assume that we are running the engine at the true speed

= t

, (3.1)

that our circular disc has teeth with adjacent spacing , and that weuse the sample time T s . Since the time samples may differ at most by T sfrom the actual time that the sensor passes a tooth in the circular disc,the calculated speed of the circular disc based on two consecutive timestamps may be as high as / ( t T s ) or as low as / ( t + T s ),depending on where the samples hit. We will denote the differencebetween these two values by the speed uncertainty , . The formula forspeed uncertainty at speed and angle resolution can be expressedas

= t T s

t + T s

= 2 T s 2

( )2 T 2s 2. (3.2)

We want to keep small, which requires a small value of T s .

3.2 Test-Bed Measurements Using Indis-cope

FlywheelCylindersCircular disc

DynamometerDamper

Engine block

Figure 3.2: Schematic engine setup in test beds.

A D12 engine was mounted in a test bed. A schematic view of how the engines are mounted in test beds is shown in gure 3.2. Thestandard test bed equipment (Indiscope) was used to extract cylinder

pressure and speed from the circular disc. Flywheel speed from S6was not extracted. A map corresponding to table 3.1 was used for themeasurements.

The importance of a high sample rate discussed in section 3.1.3 wasoverlooked. Indiscope samples at 500 kHz which gives a speed uncer-tainty = 54 rpm at 1500 rpm and = 1 using formula ( 3.2)seealso gure 3.3. Even a resampled signal with = 10 would still havea speed uncertainty of 5.4 rpm. This is not feasible for analysis orverication.


28/77

12 MEASUREMENTS

Load 1000 rpm 1500 rpm 1900 rpm25% B B B50% B B ,+1, 6 B100% B B B

Table 3.1: Map of operating points used in the rst series of measure-ments. A B means that the engine was kept balanced. A + followed

by a number j , means that cylinder j was programmed to give a higher peak pressure than the rest of the cylinders. A means that lower pressure was given by the cylinder compared to the others. The pres-sure was measured in cylinder 6.

0 100 200 300 400 500 600 7001440

1460

1480

1500

1520

1540

1560Indiscope Speed Signal at 1500 rpm

l

l y w

h e e

l s p e e

( r p m

Figure 3.3: The time stamps from Indiscope give a speed signal withlow speed uncertainty. It can be seen in the gure that the speed resolution is approximately 54 rpm.


29/77

3.3. USING ROTEC 13

3.3 Test-Bed Measurements Using Rotec

Frequency scaler?

TTL (720 pulses/rev)

TTL (1800 pulses/rev)

To S6

Pressure sensor

1 GHz50 kHz

Voltage, u

Rotec Data Acquisitor

Charge Amplier

Charge signal, q

u= kq +m

Figure 3.4: Signals measured with Rotec in the second series of mea-surements.

Rotec RAS 5.0 is a data aquisition tool that can sample at a rate of up to 1 GHz for digital signals 4 (such as the speed signal) and 50 kHzfor analogue signals (such as the pressure signal). It gives a speeduncertainty = 0 .027 rpm at 1500 rpm and = 1 .

Software was created for extracting ywheel speed data from S6.Using formula ( 3.2) we see that S6 with its six-degree spacing has aspeed uncertainty = 0 .9 rpm at 1500 rpm. Table 3.2 shows themap of operating points used in the second series of measurements.

4 Rotec only saves time samples when it passes teeth (for digital signals) whichmeans it uses small memory space as well.


30/77

14 MEASUREMENTS

Load 1000 rpm 1250 rpm 1500 rpm 1700 rpm 1900 rpm100% B B B ,-5 B B50% B B B , 5, 1,(+3-1) B B25% B B B B B8% B B B B B

Table 3.2: Operating points used in the second series of measurements.The same notation as in table 3.1 is used. The parentheses indicate thatmore than one cylinders pressure was altered in the same measurement.The pressure was measured in cylinder 5.

3.3.1 Measuring Cylinder Pressure With RotecSignals with fast dynamics (apart from S6 data) were collected by Rotecto be able to synchronize pressure and speed signals. 5 The pressuresensor (from Kiestler) mounted on the cylinder is precise, and its signalvalue, q = ap + b, is approximately linear to cylinder pressure p. Thesignal q from the sensor is a charge which is transformed into a voltage,u = kq + m, through a linear charge ampliersee gure 3.4. Thus,we have

u = kap + kb + m ,

where the values of k and m are chosen such that the voltage rangeswithin the capability of Rotecs A/D converter. The linear constant k

remains xed once the charge amplier is calibrated, while the constantm drifts slowly with time.It was not feasible to recalibrate the charge amplier once the mea-

surements had started so it is assumed that the lowest pressure in thecylinder within a measurement is close to the exhaust pressure which isone of the measured mean-value signals. Using this signal, it is possibleto offset the pressure curve to its (assumed) right position as shown ingure 3.5. Section 4.6 shows that the model output should not be sig-nicantly affected even if this would differ from the right lowest pressureby a few bars.

3.3.2 Transforming Domains for Measured Signals

All signals are measured in the time domain while the model uses theangular domain as a base for all signals. We therefore transform themeasured signals onto the angular domain.

The only feasible way to get raw data from Rotec RAS 5.0 is tocreate DIA dem data sets and export them, 6 so a Matlab function

5 Additional interesting signals, such as torque load and exhaust pressure weremean value signals that could be collected from the measurement report, [ 3].

6 The structure of DIA dem les can be found in e.g. [ 2].


31/77

3.3. USING ROTEC 15

Corrected pressure curve

Measured pressure curve

Exhaust pressure

Figure 3.5: Adjusting the measured pressure curve by offsetting itwith the help of measured exhaust pressure.

was created to parse simple DIA dem data les. Rotec hands over rawdata in oating-point format. This results in cancellation when largertime-stamp values are used for calculating ywheel speed, 7 shown ingure 3.6. The interesting exported raw data are the time array t =(t0 . . . tN )T from the speed signal, the pressure array p = ( p0 . . . pM )T and its corresponding time array = ( 0 . . . M )T .

Figure 3.7 depicts our method of transforming time- and pressuresamples onto the angular domain. Assume that the rst sample fromthe speed signal occurs at time t0 and angle 0 . If different spacing

distances between teeth in the disc are neglected and Rotecs samplerate is considered as high, we can approximate a bijective discrete timefunction

t = t ( ) = (t ), (3.3)

where = ( 0 + n )N n =1 . The measured discrete pressure signal

p = p ( ) (3.4)

is also a function of time where every time stamp from the pressuresignal is wedged in between two time stamps from the speed signal.Consider an arbitrary time stamp k from the pressure signal that iswedged in between two time stamps t j = t(j ), and tj +1 = t(j +1 ) fromthe speed signal such that

k = (1 )t j + t j +1 , [0 1]. (3.5)

The distance between j and j +1 is small so we use linear interpo-lation to dene the bijective function

= ( ) = ( ) (3.6)7 This is because the resolution of the oating point number becomes lower as

the exponent gets larger.


32/77

16 MEASUREMENTS

0 200 400 600 800 1000960

970

980

990

1000

10101020

1030

1040

flywheel angle (deg)

f l y w

h e e

l s p e e

d ( r p m

)

Early cycle, flywheel speed based on floating point time stamps

0 200 400 600 800 1000960

970

980

990

1000

1010

1020

1030

1040


f l y w

h e e

l s p e e

d ( r p m

)

Early cycle, flywheel speed from DIAdem channel

0 200 400 600 800 1000940

960

980

1000

1020

1040

1060


f l y w

h e e

l s p e e

d ( r p m

)

Later cycle, flywheel speed based on floating point time stamps

0 200 400 600 800 1000960

970

980

990

1000

1010

1020

1030

1040


f l y w

h e e

l s p e e

d ( r p m

)

Later cycle, flywheel speed from DIAdem channel

Figure 3.6: Cancellation phenomenon due to oating-point-number approximation. The two gures on top are ywheel speed based onoating point time stamps. The two bottom gures are from a speed channel in the DIAdem data setit is not known how it was created.The two left gures show two cycles at the start of the measurement.The two right gures show cycles later in the measurement.


33/77

3.3. USING ROTEC 17

where = (0 . . . M )T and k = (1 )j + j +1 see gure 3.7.To transform the pressure signal onto the angular domain we write

p = p ( ) = p ( ) . (3.7)

p(t) p

pk

(t)

jk

j +1

t j k

t j +1t

Figure 3.7: Mapping pressure samples from the time domain onto the angular domain. Linear interpolation is used between j and j +1 to dene k .

The speed signal is constructed using central approximations of thederivative,

gi (t) = 1 t i

(t + t i / 2) (t t i / 2)

Gi (s) = 1s t i / 2

es t i / 2 e s t i / 2

2 L(s)

Gi ( j ) = sin( t i / 2)

t i / 2 L( j ), (3.8)

i.e. no frequency is phase shifted compared to the exact derivative, andthe magnitude is not signicantly changed for lower frequencies. Whentransforming the constructed speed signal

(t ) = g0(t0 + t0 / 2), . . . , g n 1(tn 1 + tn 1 / 2)T (3.9)

onto the angular domain, 8 it is assumed that

(t i + t i / 2) (t i ) + / 2 (3.10)

is a good approximation.8 The array is one sample shorter than the time array t since we do not have the

value of t n .


34/77

18 MEASUREMENTS

3.3.3 A Short Analysis of the Constructed Signals

The results in gure 3.8 show that cycle-to-cycle pressure variations aresmall. However, different cylinders or engines have not been examined.

0

rotec1000varv100proc.mat

radians

p r e s s u r e

( P a

)

0


radians

p r e s s u r e

( P a

)

0


radians

p r e s s u r e

( P a

)

0


radians

p r e s s u r e

( P a

)

Figure 3.8: Cycle-to-cycle variations of cylinder pressure near TDC at various operating points. The data shown is from the second series of measurements in a test bed, and 10 consecutive cycles are displayed in each plot, even if they are hard to discern. The pressure curves fromthe rst series of measurements give similar plots.

The speed signal that is extracted from Rotec is not smooth whichcan be seen in gure 3.9. This also occur after compensating for non-equidistant spacing in the circular disc. The noisy signal may be due

to imperfections in measurement equipment used.

3.3.4 Reason for the Noisy Speed Signal

There are only 720 markers in the disc, but the TTL signal gives 1800pulses per revolutionsee gure 3.4. The equipment in the measurechain that scales the frequency must extrapolate time stamps in orderto enhance resolution in real time. We can use formula ( 3.2) to roughlyestimate that the equipment needs a sample rate of at least 10 MHz


35/77

3.3. USING ROTEC 19

720 8401400

1420

1440

1460

1480

1500

1520

1540

p l

l

l


Figure 3.9: Non-smooth speed signal from Rotec. The gure shows the spike in ywheel speed that the rst cylinder gives rise to at1500 rpm and 100% load.

to avoid losing information. It should be investigated if the equipmenthas any positive effects.

3.3.5 A Comparison Between the Speed Signal fromS6 and the Speed Signal from Rotec

The main reason for comparing the signal from S6 with the signal fromRotec is to examine if any dynamical differences can be noted. 9 If nomajor differences are seen, data from Rotec may be used as vericationdata for the model. The noisy speed signal is not a major problem sincethe purpose of the model is to correctly model lower order frequencies. 10

Results from speed comparisons show that the signals are not muchdifferent in their dynamical properties at 1500 rpm. The signal fromRotec has a slightly larger magnitude for the third-order frequency,as seen in gure 3.10. Measurements from S6 was only extracted at1500 rpm. 11 Greater differences at higher speeds are expected. How-ever, this may not be necessary to examine in the future since thesignals are cycle-periodic and it is possible to use speed data from S6and pressure data from Indiscope in the model.

9 Remember from gure 3.1 that Rotec gets its signal from the circular discattached to the ywheel, whereas S6 measures directly on the ywheel.

10 It is common practice in automotive engineering to speak about frequency or-ders. The engine-speed frequency is dened as the rst order frequency, the secondorder frequency is twice as high as the engine frequency, and so on.

11 This was due to problems with the software used for extracting signals from S6.


36/77

20 MEASUREMENTS

0 120 240 360 480 600 7201480

1485

1490

1495

1500

1505

1510

1515

1520rotec1500varv50procS6norm.mat

Flywheel angle (deg)

r p m

RotecS6

0 120 240 360 480 600 72010

5

0

5

10Residual


r p m

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g

)

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

1

2

3

4

5Harmonic analysis

Engine order

S p e e

d ( r p m

)

RotecS6

Figure 3.10: A comparison between signals from S6 and Rotec in the test bed environment at 1500 rpm and 100% load. A difference in the third-order magnitude is seen.


37/77

3.4. THE CHASSIS DYNAMOMETER 21

3.4 Measurements on the Chassis Dynamo-meter

To examine if different truck drivelines can give rise to different speedresponses on the ywheel, speed measurements were carried out on thechassis dynamometer on two different trucks, Mike and Moa . Fig-ure 3.11 depicts the measurement setup. The trucks have the same en-

Truck

Rollers

Figure 3.11: A schematic view of how the chassis dynamometer works.

The rollers are connected to dynamometers to provide load torque onthe driveline.

gine type, D12, but different drivelines. 12 Test-bed pressure curves foran engine with a similar conguration as the trucks were found. Hopewas that the same injection time, , and injection angle, , would recre-ate the pressure curves with good accuracy. This would allow to simu-late the trucks ywheel speeds later, if driveline models were available.Table 3.3 shows the map used in the measurements for both trucks.

Load 1000 rpm 1500 rpm 1900 rpm100% B B B75% B B ,+1,-1 B50% B B B25% B B B

Table 3.3: Operating points used in the third series of measurements,using the same notation as in table 3.1.

12 One difference is the length of the propeller shafts.


38/77

22 MEASUREMENTS

3.4.1 A Comparison Between the Speeds for theTwo Trucks

Speed comparisons show that there is no signicant difference in y-wheel speed between the two trucks at most operating points, repre-sented by gure 3.12. The only operating point where the trucks showa difference in ywheel speed dynamics is at 1000 rpm and 100% loadwhere Moa has a half-order contributionsee gure 3.13. One reasonfor this difference could be that a near resonance frequency for Moa sdriveline is excited, which then gives a signicant torque response backto the ywheel.

0 120 240 360 480 600 7201480

1490

1500

1510

1520

1530chDynGear10MOA1500varv75procPlus50mgCyl1.mat


r p m

0 120 240 360 480 600 72012

10

8

6

4

2Residual


r p m

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

90

0

90

180

Engine order

p h a s e

( d e g

)

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

1

2

3

4

5

6

7Harmonic analysis

Engine order

S p e e

d ( r p m

)MOAMIKE

MOAMIKE

Figure 3.12: A comparison between speeds from the two trucks Mike(gear 10) and Moa at 1500 rpm and 75% load with a positive offsetof 50 mg injected fuel in cylinder 1. This example represents how the

speeds differ at most operating points.

3.4.2 A Comparison Between Speeds for the Trucksand Speeds in the Test Bed

Slight differences are seen for most measurements when speed signalsfrom measurements in the test bed and speed signals from measure-ments at the chassis dynamometer are compared. The difference at


39/77


0 120 240 360 480 600 720960

970

980

990

1000

1010

1020

1030

1040chDynMOA1000varv100proc.mat


r p m

0 120 240 360 480 600 7208

6

4

2

0

2

4

6Residual


r p m

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g

)

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

5

10

15Harmonic analysis

Engine order

S p e e

d ( r p m

)MOAMIKE

MOAMIKE

Figure 3.13: A comparison between speeds from Mike and Moa at1000 rpm and 100% load. A signicant difference of the half order magnitude can be seen.


40/77

24 MEASUREMENTS

half the engine order at 1500 rpm shown in gure 3.14 probably derivesfrom an excited resonance frequency for the driveline used in the testbedsee chapter 4. However, the higher order differences seen in g-ure 3.15 at 1900 rpm are probably not only due to driveline differences.It is possible that they result from dynamics between the ywheel andthe circular disc in the test bed.

0 120 240 360 480 600 7201460

1470

1480

1490

1500

1510

1520

1530rotec1500varv100proc.mat


r p m

0 120 240 360 480 600 72025

20

15

10

5

0

5

10

15Residual


r p m

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g

)

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

2

4

6

8

10

12Harmonic analysis

Engine order

S p e e d

( r p m

)

RotecMIKE

RotecMIKE

Figure 3.14: A comparison between speeds from Mike and Rotec at1500 rpm and 100% load. The half-order magnitude in the test-bed signal probably derives from a resonance in the test-bed driveline.


41/77


0 120 240 360 480 600 7201870

1880

1890

1900

1910



r p m

0 120 240 360 480 600 72025

20

15

10

5

0

5

10Residual


r p m

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g

)

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

1

2

3

4

5Harmonic analysis

Engine order

S p e e

d ( r p m

)

Rotec

MOA

RotecMOA

Figure 3.15: A comparison between speeds from Moa and Rotec at1900 rpm and 100% load. The speeds differ somewhat at higher orders which indicate dynamics between the ywheel and the circular disc inthe test bed.


42/77


43/77

Chapter 4

Simulations

Some variations of the model described in chapter 2 were implementedin Matlab /Simulink using S -functions written in C . A Matlab classand a number of M -les were constructed to facilitate when runningdifferent setups, and when analyzing the simulationssee [ 12] for moredetails. The following model variations are discussed in this chapter:

The simple model The model described in chapter 2.

The dynamically equivalent model This model uses a dynamical-ly equivalent model of the connecting rod.

The test-bed driveline model The simple model extended with asimple test-bed driveline.

The truck-driveline model The simple model extended with a sim-ple truck driveline.

Trial-and-error models The simple model with adjusted parame-ters.

Parameters for the models were fetched from [ 7, 8, 10] and [1].There are some uncertainties in the engines congurations, 1 and thevalues of many parameters differ slightly between the reports (a relativedifference of about 0.15).

Parameter values were fed into the models and simulations weremade for all operating points. To estimate the initial state, x 0 , apreliminary simulation was run to the beginning of a cycle where itwas assumed that the transient response had died out. The frictiontorque (modeled as absolute-damping elements) is unknown and es-timated from an energy-balance equation at steady-state conditions.

1 For instance, it is unknown whether Holset- or Hasse & Wrede dampers wereused.

27


44/77

28 SIMULATIONS

The absolute-damping elements in the model are set thereafter. Theabsolute-damping elements ranges between 0.06 and 0.8 [Nms/rad] forthe 65 possible measurement setups used in this work.

4.1 The Simple Model

The model with nine inertias described in chapter 2 will be referredto as the simple model . A simulation at 1000 rpm and 100% load iscompared with actual measured data in gure 4.1.2 It can be seen thatthe simple model follows the measured data well. Interesting to noteare the depths of the dips between the ring of cylinders. The ringsequence of the engine is 153624, where cylinder 1 is the furthestfrom the ywheelsee gure 2.2. The trend is that the dips are deeperwhere cylinders that are far from the ywheel re.

0 120 240 360 480 600 720960

970

980

990

1000

1010

1020

1030

Flywheel Angle (deg)

r p m


measuredsimulated

0 120 240 360 480 600 7208

6

4

2

0

2

4

6

8


r p m

Residual

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g

)

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

2

4

6

8

10

12

14Harmonic analysis

Engine order

S p e e

d ( r p m

)measuredsimulated

Figure 4.1: Simulation with the simple model at 1000 rpm and 100% load. The model works well at this operating point.

The simple model has trouble to follow some dynamics when sim-ulating at 1500 rpm and 100% load as seen in gure 4.2. The largest

2 The time-discrete ywheel speed signal that Simulink returns is not equidistantin angle. Thus, before transforming the signal onto the frequency domain equidis-tant interpolation is done in the angular domain.


45/77

4.1. THE SIMPLE MODEL 29

differences occur for lower-order frequencies where the measured speedhas a signicant contribution. Section 4.3 indicates that the modelneeds a driveline to explain the lower-order contributions at this oper-ating point.

0 120 240 360 480 600 7201460

1470

1480

1490

1500

1510

1520

1530


r p m


measuredsimulated

0 120 240 360 480 600 72010

5

0

5

10

15

20

25


r p m

Residual

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g

)

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

2

4

6

8

10

12Harmonic analysis

Engine order

S p e e

d ( r p m

)measuredsimulated

Figure 4.2: Simulation with the simple model at 1500 rpm and 100% load. The measured speed has lower-order contributions thatthe model can not explain.

Finally, a simulation is made at higher engine speed. It can beseen in gure 4.3 that the model has trouble to follow the measureddata here too. However, note that the simulated speed correspondsomewhat better to the ywheel speed from Mike seen in gure 3.15.Again, one reason for the differences can be that there are dynamics tobe considered in the test bed between the circular disc and ywheel at

higher speeds.The results from the simulations show that the simple model does

not cover all important phenomena that give rise to different vibrationson the ywheel in the test bed.


46/77

30 SIMULATIONS

0 120 240 360 480 600 7201870

1880

1890

1900

1910

1920

1930


r p m


measuredsimulated

0 120 240 360 480 600 72015

10

5

0

5

10


r p m

Residual

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g

)

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

1

2

3

4

5

6Harmonic analysis

Engine order

S p e e

d ( r p m

)measuredsimulated

Figure 4.3: Simulation with the simple model at 1900 rpm and 100% load. The simulation differ from measured data. Remember that the measured data is from the circular disc in the test bed.


47/77

4.2. A DYNAMICALLY EQUIVALENT MODEL 31

4.2 A Dynamically Equivalent Modelof the Crank-Slider Mechanism

The simple model uses a statically equivalent model of the crank-slider mechanism. Schagerberg and McKelvey [ 14] refer to papers byHafner and Shiao respectively where error analysis between statically-and dynamically equivalent models are performed. Still, a dynami-cally equivalent model was implemented and compared to the simplemodel. The equations for a dynamically equivalent model are derivedin appendix A.1.

An approximate value of the connecting rod was used. A com-

parison between models at 1900 rpm is seen in gure 4.4. No majordifferences are seen between simulations with the simple model and thedynamically equivalent model, and it is from here on assumed that thereis no need for a dynamically equivalent model of the crank-slider mech-anism, even though an approximate value for the moment of inertia of the connecting rod was used.

0 120 240 360 480 600 7201870

1880

1890

1900

1910

1920



r p m

0 120 240 360 480 600 7201.5

1

0.5

0

0.5Residual


r p m

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g

)

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

1

2

3

4

5

6Harmonic analysis

Engine order

S p e e

d ( r p m

)dynstat

dynstat

Figure 4.4: A comparison between statically- and dynamically equiv-alent models at 1900 rpm shows that a statically equivalent model of the connecting rod is sufficient for simulation.


48/77

32 SIMULATIONS

4.3 Including a Model of the Driveline

The simple model was not accurate at operating points other than atlow engine speeds (1000 rpm and 1250 rpm), and there were no improve-ments when using a dynamically equivalent model for the crank-slidermechanism. The half-order frequency at 1500 rpm may be due to re-sponses from the driveline in the test bed environment, so the next stepwill be to include a simple driveline in the model.

The major thing that changes in the model when a driveline isincluded is the number of elements included in the lumped-mass model,see gure 4.5. Conversion ratios for the transmission and nal drive areincluded by manipulating the damping- and stiffness matrix, C and K ,as described in section A.2.

Parameters for a simple test-bed driveline can be found in [ 7]. Ittakes many cycles for the transient to die out when we include thisdriveline model as can be seen in gure 4.6. In all simulations in thissection, 30 cycles were simulated before an initial state was estimated.

010101010101 010101010101 010101010101 DrivelineFigure 4.5: The lumped mass model including a simple driveline.

Simulations with the driveline model show no major differences fromthe simple model at 1000 rpm and 1900 rpm. This is probably becausethe driveline is not forced near a resonance frequency at these oper-ating points. On the other hand, at 1500 rpm the driveline modeldiffers signicantly from the simple model. In gure 4.7, a simulationat 1500 rpm is compared with actual measured data. This is an operat-ing point where the simple model had trouble following the lower-order

frequencies. It can be seen that the extended model follows the mea-sured data better which indicates that the test-bed driveline respondssignicantly near 1500 rpm. The parameters used also show that thedriveline has a resonance frequency with low damping near 750 rpm,which is the half order frequency of 1500 rpm.

Another operating point at 1500 rpm is simulated to further inves-tigate the driveline model. The simulation is compared with actualmeasured data where cylinder 5 has a positive injection offset. Thismeans that we are forcing the driveline at its resonance frequency. It


49/77

4.3. INCLUDING A MODEL OF THE DRIVELINE 33

0 0.5 1 1.5 2 2.5

4

1460

1470

1480

1490

1500

1510

1520

1530

1540Transient response with cell driveline and untwisted crankshaft as initial state.


F l y w

h e e

l s p e e

d ( r p m

)

Figure 4.6: It takes a number of cycles for the transient response to die out when including a test-bed driveline. This gure shows the rst30 cycles from a simulation.

0 120 240 360 480 600 7201460

1470

1480

1490

1500

1510

1520

1530


r p m


measuredsimulated

0 120 240 360 480 600 72015

10

5

0

5

10

15


r p m

Residual

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g

)

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

2

4

6

8

10

12Harmonic analysis

Engine order

S p e e

d ( r p m

)measuredsimulated

Figure 4.7: Simulation with the driveline model at 1500 rpm and 100 % load. This model can explain the lower-order contribution.


50/77

34 SIMULATIONS

can be seen in gure 4.8 that the simulated data is nowhere near themeasured data. This may be due to limitations in the driveline model.It is for instance unlikely that the test-bed engine actually has a tran-sient response such as in gure 4.6.

0 120 240 360 480 600 7201460

1480

1500

1520

1540

1560

1580

1600

1620


r p m

rotec1500varv50procPlus50mgCyl5.mat

measuredsimulated

0 120 240 360 480 600 720100

80

60

40

20

0

20


r p m

Residual

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g

)

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

5

10

15

20

25

30

35Harmonic analysis

Engine order

S p e e

d ( r p m

)measuredsimulated

Figure 4.8: Simulation with the driveline model at 1500 rpm and 100 % load with a positive pressure offset present in cylinder 5.


51/77

4.4. A TRUCK-DRIVELINE MODEL 35

4.4 A Truck-Driveline Model

Section 3.4 shows no indications of signicant torque responses fromtruck drivelines. To further investigate this matter, a simple truckdriveline was implemented and simulated at various operating points.Parameters for the driveline are estimated for the truck Malte byBerndtsson and Uhlin [ 4]. The value of the ywheel inertia was adjustedto get better simulation results.

The lowest resonance frequency in the driveline model is higherthan 1900 rpm, and more damped than the resonance frequency forthe test-bed driveline model. It is hard to estimate if the driveline willcontribute to dynamics at higher orders, but there should not be anymajor contributions at lower orders.

4.4.1 Simulations with a Truck Driveline Comparedto Simulations Without a Driveline

Simulations with and without a driveline in the model were performed,and comparisons are shown in gures 4.9 and 4.10. Differences are mostsignicant in simulations at the third-order frequency. This can be dueto a wrongly estimated ywheel inertia. The value of a signicant lower-order difference is not known. However, section 4.4.2 shows thatthe results should be questioned.

One thing that still should be investigated is the half-order contri-

bution for Moa at 1000 rpm and 100% loadsee section 3.4.1. It isthe only measured operating point that indicates a signicant torquecontribution from a trucks driveline.


52/77

36 SIMULATIONS

0 120 240 360 480 600 720970

980

990

1000

1010

1020

1030

1040



r p m

0 120 240 360 480 600 7205

0

5

10

15

20Residual


r p m

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g )

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

2

4

6

8

10

12

14

16

Harmonic analysis

Engine order

S p e e

d ( r p m

)

simpleMALTEsimple

MALTE

Figure 4.9: A simulation at 1000 rpm with a truck-driveline model for Malte compared to one with no driveline.


53/77


0 120 240 360 480 600 7201870

1880

1890

1900

1910

1920

1930



r p m

0 120 240 360 480 600 7201.5

1

0.5

0

0.5

1

1.5

2Residual


r p m

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g )

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

1

2

3

4

5

6

7

Harmonic analysis

Engine order

S p e e

d ( r p m

)

simpleMALTE

simpleMALTE

Figure 4.10: A simulation at 1900 rpm with a truck-driveline model for Malte compared to one with no driveline.


54/77

38 SIMULATIONS

4.4.2 Simulations with a Truck Driveline Comparedto Measured Data

Figures 4.11 and 4.12 show comparisons between simulations and mea-surements carried out at the chassis dynamometer.

0 120 240 360 480 600 720960

980

1000

1020

1040

1060


r p m

chDynMOA1000varv100proc.mat

measuredsimulated

0 120 240 360 480 600 72025

20

15

10

5

0

5

10

15


r p m

Residual

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g

)

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

2

4

6

8

10

12

14

16

18Harmonic analysis

Engine order

S p e e d

( r p m

)

measuredsimulated

Figure 4.11: A simulation at 1000 rpm with a truck-driveline model for Malte compared to measured ywheel speed from Moa .

The speeds from the trucks are signicantly phase shifted comparedto the simulations. This phenomenon also occur when the simple modelis used for simulating the trucks. One suggested reason is uncertaintiesin TDC position when the pressure curves for the measurements was

made, but this is not conrmed when the pressure curves are plotted. 3Another reason could be that the time stamps do not begin exactly atTDC for cylinder 1, but this should still not result in such a large phaseshift. The discussion in section 4.4.1 should be read critically since themodel can not follow measured speed.

3 Remember that the pressure curves for the truck simulations are from storeddata, see section 3.4.


55/77


0 120 240 360 480 600 7201880

1890

1900

1910

1920

1930


r p m


measuredsimulated

0 120 240 360 480 600 72020

15

10

5

0

5


r p m

Residual

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g )

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

1

2

3

4

5

6

Harmonic analysis

Engine order

S p e e

d ( r p m

)measuredsimulated

Figure 4.12: A simulation at 1900 rpm with a truck-driveline model for Malte compared to measured ywheel speed from Moa .


56/77

40 SIMULATIONS

4.5 Trial-and-Error Modeling

All recent models have trouble at higher engine speeds. In this sectionparameters are altered manually to see if a better model can be found.

4.5.1 An Alternative Damper Model

As seen from gures in previous sections, each power stroke from thering of the cylinders creates a torque spike on the crankshaft. Thesetorque spikes cause the crank journal to twist slight and spring back,causing torsional vibrations in the crankshaft. A vibration damper ismounted to the front of the crankshaft to reduce the damaging effectof these vibrations.

Dampers in Scanias trucks use viscous uid to dampen torsionalvibrations. In general, the motion of uids is complicated which makesit reasonable to believe that the model of the damper can be improved.

In the simple model, the damper is modeled as two lumped massesconnected with a linear spring and damper. A more general model of the damper torque is not restricted to linear dependencies only. In onetrial-and-error model of the damper, the function

T d (1 2 , 1 2) = k11 2 + c1 1 2 + k2 1 2 + c2 21 2 (4.1)was used for the damper torque T d , where the parameters were setby trial and error. 4 The arguments 1 2 and 1 2 are the angular-and speed difference respectively between the two lumped masses inthe damper model. Simulations were carried out at various parameterchoices and one result is shown in gure 4.13.

It is not known what constitutes a reasonable model of the damper,but different setups in damper parameters result in slightly differentresults in simulations compared to the simple model. Yet, it has notbeen shown that the problem at higher speeds derives from limitationsin the damper model, and the trial-and-error model is not signicantlybetter at any tried parameter setup.

4.5.2 Manipulating Parameters Based on Simula-tion Experience

In the simulation shown in gure 4.14 the crankshaft stiffness was mul-tiplied by 1.3 compared to the crankshaft stiffness in previous models.Simulations follow dynamics at 1900 rpm better in this case. 5 Still, thesimulated data is phase shifted compared to measured data. This can

4 The aim of the function was to let torque due to difference in position be lesssignicant at larger angle differences, whereas torque due to speed difference shouldbe more signicant at larger speed differences.

5 In fact, the only signicant difference between the signals is a phase shift.


57/77

4.5. TRIAL-AND-ERROR MODELING 41

0 120 240 360 480 600 7201880

1890

1900

1910

1920

1930


r p m


measuredsimulated

0 120 240 360 480 600 72020

15

10

5

0

5

10


r p m

Residual

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g

)

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

1

2

3

4

5

6Harmonic analysis

Engine order

S p e e

d ( r p m

)measuredsimulated

Figure 4.13: The results turn out somewhat different when another model for the damper is used. The linear parameters k1 and c1 are setto zero in this particular simulation, while k2 = 5000 [Nmrad 1/ 2 ] and c2 = 100 [Nms 2rad 2 ].


58/77

42 SIMULATIONS

0 120 240 360 480 600 7201880

1890

1900

1910

1920

1930


r p m


measuredsimulated

0 120 240 360 480 600 72025

20

15

10

5

0

5


r p m

Residual

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g )

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

1

2

3

4

5

Harmonic analysis

Engine order

S p e e

d ( r p m

)measuredsimulated

Figure 4.14: A simulation with modied stiffness parameters. The only signicant difference is a phase shift.


59/77

4.5. TRIAL-AND-ERROR MODELING 43

be corrected to a certain degree by for instance weakening the stiffnessbetween the last cylinder and the ywheel, but this will at the sametime slow down dynamics. By comparison, this model is in some sensethe closest to measured data. However, it is not known if the parametervalues are reasonable.

Optimization of parameters based on measured data from S6 maylead to an acceptable model in the future. The drawback with thismethod is that it may require repeating the optimization procedure fornew engine congurations.


60/77

44 SIMULATIONS

4.6 Sensitivity Analysis

Simulated data has not corresponded to measured data at higher speeds.Could uncertainties in model parameters contribute to the errors? Toexamine this, simulations at 1900 rpm for models with perturbed pa-rameters were made.

First, cylinder pressures that are used as input in the model areperturbed. This is done in three ways. The pressure curve is rst scaled,then it is offset and last it is translated in to see if it makes a bigdifference in the model output. 6 These perturbations mimic possibleerrors when measuring pressure.

After perturbing the pressure curves we alter other parameters inthe model. First we alter the inertias and then we alter the stiffnessparameters. The parameters have been altered by a relative fraction of 0.2. Results are seen in gures 4.154.19.

We altered parameters for the damper in section 4.5.1. It is un-known if those changes are reasonable.

0 120 240 360 480 600 7201880

1890

1900

1910

1920

1930unknown signals


r p m

0 120 240 360 480 600 7200.7

0.6

0.5

0.4

0.3

0.2Residual


r p m

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g

)

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

1

2

3

4

5

6

7Harmonic analysis

Engine order

S p e e

d ( r p m

)

nominalperturbed

nominalperturbed

Figure 4.15: A comparison between simulations where one simulationhas its pressure curve offset by 5 bars.

6 The absolute damping are set different in the two models, to compensate fordifferent work contributions from the pressure curves.


61/77

4.6. SENSITIVITY ANALYSIS 45

0 120 240 360 480 600 7201870

1880

1890

1900

1910

1920

1930

1940

unknown signals


r p m

0 120 240 360 480 600 72012

10

8

6

4

2

0

2Residual


r p m

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g )

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

1

2

3

4

5

6

7

89

Harmonic analysis

Engine order

S p e e

d ( r p m

)nominalperturbed

nominalperturbed

Figure 4.16: A comparison between simulations where one simulationhas its pressure curve scaled by a relative fraction of 0.1.


62/77

46 SIMULATIONS

0 120 240 360 480 600 7201880

1890

1900

1910

1920

1930

unknown signals


r p m

0 120 240 360 480 600 7201.5

1

0.5

0

0.5

1Residual


r p m

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g )

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

1

2

3

4

5

6

7

Harmonic analysis

Engine order

S p e e

d ( r p m

)

nominalperturbed

nominalperturbed

Figure 4.17: A comparison between simulations where one simulationhas its pressure curve translated 1 to the left.


63/77


0 120 240 360 480 600 7201870

1880

1890

1900

1910

1920

1930unknown signals


r p m

0 120 240 360 480 600 7206

4

2

0

2

4Residual


r p m

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g

)

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

1

2

3

4

5

6

7Harmonic analysis

Engine order

S p e e

d ( r p m

)

nominalperturbed

nominalperturbed

Figure 4.18: A comparison between simulations where one simulationhas its stiffness parameters of the crankshaft and driveline perturbed by a relative fraction of 0.2.


64/77

48 SIMULATIONS

0 120 240 360 480 600 7201870

1880

1890

1900

1910

1920

1930unknown signals


r p m

0 120 240 360 480 600 7201

0.5

0

0.5

1

1.5

2Residual


r p m

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .5

180

90

0

90

180

Engine order

p h a s e

( d e g

)

0 .5 1 1 .5 2 2 .5 3 3 .5 4 4 .5 5 5 .50

1

2

3

4

5

6

7Harmonic analysis

Engine order

S p e e

d ( r p m

)

nominalperturbed

nominalperturbed

Figure 4.19: A comparison between simulations where one simula-tion has its inertias of the crankshaft and the driveline perturbed by a relative fraction of 0.2.


65/77


The sensitivity analysis indicates that the model is most sensitiveto scaled pressure and offset stiffness parameters. More experience isneeded to judge if the model is robust enough to nd adequate param-eter values, but nothing yet indicates the opposite.


66/77


67/77

Chapter 5

Conclusions

5.1 Conclusions from the Measurements

Pressure curves from the measurements do not differ much from cy-cle to cycle and there does not appear to be signicant contributionsof frequencies lower than half order at steady-state conditions. This,together with the sensitivity analysis in section 4.6 indicate that it ispossible to nd a model that can facilitate to gain understanding of thetransfer function discussed in chapter 1.

Measurements in the test-bed environment indicate that the test-bed driveline has a resonance frequency near 750 rpm that affects y-wheel speed variations when excited. The two speed signals from S6and Rotec are not signicantly different at 1500 rpm, but could differmore at higher speeds .1 Measurements from the chassis dynamometersupport this, if we assume that speeds from trucks are similar to speedsin the test-bed environment. This may not be necessary to investigatesince the speeds are cycle-periodic and it is possible to use speed fromS6 and pressure curves from Indiscope.

Measurements on Mike and Moa show that both ywheels havesimilar speed variations at similar operating points, indicating that adriveline could be omitted in the model when simulating trucks. Nev-ertheless, at 1000 rpm and 100% load there is a slight difference.

Flywheel speed uctuations in the test bed look similar to uctu-ations in the chassis dynamometer. Still, they differ more at higherspeeds. This may be due to different engine congurations, but it ismore likely due to dynamics between the ywheel and the circular discin the test-bed environment.

More accurate signals may be obtained with Rotec if the frequency1 Remember that we were only able to extract the S6 signal at 1500 rpm in the

test-bed environment.

51


68/77

52 CONCLUSIONS

scaler is omitted in the measure chain. It should be examined if thefrequency scaler used in test beds have any positive effects. Still, thesignals obtained are accurate enough to test and verify the model.

5.2 Conclusions from the Simulations

The simple model follows the measured speed signal well at low speeds,while it has trouble at higher speeds. Simulations with a dynamicallyequivalent model for the crank-slider mechanism do not differ from sim-ulations with the simple model, indicating that a statically equivalentmodel of the crank-slider mechanism is sufficient for simulation.

An extended model with a simple test-bed driveline supports themeasured indication of a resonance frequency near 750 rpm. However,the model has limitations as shown in gure 4.8.

Simulations with driveline parameters optimized for Malte differsomewhat from simulations with the simple model. But since the simplemodel follows measured data better than the model for Malte theresults should not be treated seriously.

Other models were investigated using parameters set after experi-ence. A different model for the vibration damper was used in simula-tions, giving slightly different results in ywheel-speed variations. Yet,it was not shown that the problems derive from an inaccurate dampermodel. A model with a stiffer crankshaft was also investigated, showing

promise at higher engine speeds, but its speed signals are still phaseshifted compared to measured data.Auxiliary systems, such as the cam shaft, are not taken into con-

sideration in the model. The model is also limited by the number of lumped masses included. It may turn out that this is a fundamentallimitation, but it is hard to investigate.


69/77

Chapter 6

Future Work

Gathered measurements can be used together with optimization to ndbetter parameters for the model. If more measurement data is needed,it is suggested that extracted data from S6 are used together withpressure curves from Indiscope. 1 However, operating near 1500 rpm intest beds should be avoided, or measurements should be conducted ontrucks. If the model continues to have problems at higher engine speedsit is suggested that an alternative model for the damper is investigated.Measuring speed at the damper case could then be of importance.

Another thing to investigate is how much pressure curves differ for

different engine congurations and if this has to be taken into consider-ation in simulations. Examining effects from Turbo Compound 2 is alsoa future issue.

The present work can not answer with certainty if drivelines affectvariations in the speed signal for trucks or not. The driveline model forMalte indicates differences but then the simple model is better thanthe model for Malte when simulations are compared with measure-ments. Finding better parameters in the driveline models may be anissue for the future.

The code in the S -functions can be updated to improve simulationspeed. A model for this purpose can be found in [ 14] and general issueson speeding up S -functions can be found in e.g. S -function templatesin Matlab .

If an adequate model is found in the future, it can be made modularand incorporated in a larger model that includes auxiliary systems of the engine. It may also be used to investigate methods of discoveringunbalanced cylinders.

1 This is only because it is simpler than using Rotec.2 Turbo Compound uses exhaust gas ow to provide extra torque on the

crankshaft.

53


70/77


71/77

References

[1] Parameters for Dampers Used in Scania 12 L Engines. Not Pub-lished. Received from Johan Lundqvist.

[2] Description of the Data formats and Data set Proper-ties of DIAdemThe PC-Workshop. Downloadable fromwww.sesystem.co.kr/diadem/image/dmheader.pdf , 2003.c GfS mbH Aachen, Germany.

[3] Protokoll: 17-414-029. Internal protocol at Scania, October 2003.

[4] M. Berndtsson and E. Uhlin. Skattning och aktiv d ampning avdrivlinesv angningar i lastbil. Masters thesis, LuTH, 2003.

[5] L. Jianqiu and Y. Minggao. Analysis of the Inuence of Crankshaft

Vibration Upon the Instantaneous Flywheel Speed. FISITA World Automotive Congress , 2000.

[6] L. Jianqiu, Y. Minggao, Z. Ming, and L. Xihao. Individual Cylin-der Control of Diesel Engines. Society of Automotive Engineering ,(2002-01-0199).

[7] E. Jorpes. Ber akningsrapport: Torsionssimulering av DL. Internal Scania Report , 2001.

[8] E. Jorpes. Ber akningsrapport: Utredning av torsionsfenomen i nymotorprovcell. Internal Scania Report , 2002.

[9] U. Kiencke a

Documents

Flywheel Model