Scribd Logo
Upload

Welcome to Scribd!

  • PHD Dissertation 1997 Williams

    Uploaded by

    Javier Bush
    12 views157 pages
    engine tesis

    Original Title

    PhD Dissertation 1997 Williams

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document
    engine tesis

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    12 views157 pages

    PHD Dissertation 1997 Williams

    Uploaded by

    Javier Bush
    engine tesis

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    of 157
    MISFIRING CYLINDER DIAGNOSIS THROUGH CRANKSHAFT TORSIONAL VIBRATION MEASUREMENT A dissertation submitted to the Division of Research and Advanced Studies of the University of Cincinnati in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY in the Department of Mechanical, Industral, and Nuclear Engineering of the College of Engineering 1997 by Jeremy Williams B.Sc. ME., Virginia Polytechnic Institute and State University 1987 MSc. ME., Virginia Polytechnic Institute and State University 1989 Committee Chair: Dr. David L. Brown MISFIRING CYLINDER DIAGNOSIS THROUGH CRANKSHAFT TORSIONAL VIBRATION MEASUREMENT A dissertation submitted to the Division of Research and Advanced Studies of the University of Cincinnati in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY in the Department of Mechanical, Industrial, and Nuclear Engineering of the College of Engineering 1997 by Jeremy Williams B.Sc. MEE, Virginia Polytechnic Institute and State University 1987 MSc. ME, Virginia Polytechnic Institute and State University 1989 Committee Chair: Dr. David L. Brown ABSTRACT When each cylinder in an internal combustion engine fires, a torque pulse is applied to the crankshaft. These torque pulses contain significant energy for the first twenty four harmonics of engine cycle speed, resulting in dynamic torsional excitation and response of the crankshaft and drivetrain in both rigid body and flexible modes. Ifa cylinder misfires, the excitation torque for that cylinder is altered, resulting in an altered dynamic torsional system response. ‘This research presents a new method for diagnosing and identifying misfiring cylinders in an internal combustion engine. A knowledge of the system dynamic model, system boundary conditions, and measured crankshaft torsional response are used to characterize the dynamic input at each engine cylinder. Current methods of misfire diagnosis using crankshaft torsional response as an indicator rely on several key assumptions to form a solvable equation set relating measured response to the system input at cach cylinder. Chief among these assumptions are non-overlapping firing pulses and rigid crankshaft behavior. These assumptions limit diagnostic success with current methods to engines with six or fewer cylinders operating at low to moderate engine speeds. ‘The proposed diagnostic method does not rely on these assumptions. Rather, a set of linear constraint equations are added to the otherwise unsolvable equation set that relates the measured torsional crankshaft response (0 individual cylinder input. The constraint equations relate the magnitude and phase of the harmonics of the forcing function for each cylinder to the peak pressure and location of peak pressure for each cylinder pressure curve, Thus, the harmonics ofthe excitation torque for each cylinder fare no longer independent. ‘The constraint equations are introduced into the diagnostic equation set through the simultaneous consideration of multiple frequencies of excitation and response. AA series of test cases are presented which sbow that the diagnostic process should work with engines with overlapping firing pulses operating with significant crankshaft deflection. An error sensitivity analysis shows the process is viable with reasonable levels of noise and error associated with all key assumptions and measurements, In addition, a new signal processing technique is proposed which promises to significantly reduce torsional vibration measurement error compared to current state ofthe art techniques. ACKNOWLEDGMENTS | would like to thank my committee chairman, Dr. David Brown, for his advice and support, and {for creating an environment in which I could pursue my research goals. I would also like thank my other ‘committee members, Dr. Randall J. Allemang, Dr. Robert W. Rost, Dr. Swart J. Shelley, and Dr. Robert H. Wynn Tam especially grateful to Dr. Stuart Shelley for his invaluable advice and feedback in the last two years and to Dr. Robert W. Rost for coordinating this project with the University. I would like to thank Dr. Allyn Phillips for his help in developing the new digital torsional vibration measurement techniques. | would also like to thank Mr. Thomas Reinhart and Mr. Alan Zhao of Cummins Engine Co, Inc. Mr. Reinhart’s group at Cummins has sponsored this work in its entirety for which I am very grateful. ‘Tom Reinhart and Alan Zhao have been very supportive throughout the research process. Without the support of Tom and Alan, this work could not have been completed. I look forward to a continuing, relationship with each of them. Last, but by no means least, I would like to thank John Schultze, Walter Lynge, Imtiaz Ali, Ming Yang, Cindy Adelman and all the other members of the Structural Dynamics Research Laboratory for their friendship. Working within the SDRL has been an enriching experience for me on a personal as well as professional level. Ihave made friendships over the last three years that will last a lifetime, TABLE OF CONTENTS 1. INTRODUCTION 13 2, LITERATURE REVIEW, 2.1 Introduction : 2.2 Overview And Comparison Of Current Methods. 2.2.1 Introduction. - 2.2.2 Engine Torsional Vibration Analysis vs. Engine Diagnostic AnalySis...u.n. 22.3 System Model 2.2.4 Analysis Methods. 2.2.5 Misfire Detection And Location. 2.2.6 Assumptions. 2.2.7 Remaining Challenges. 2.3 Summary By Author Of Relevant Literature, 2.4 Summary 3. INTERNAL COMBUSTION ENGINE TORSIONAL VIBRATION ANALYSIS, 3.1 Introduction 3.2 Forcing Function. 3.3 System Models. 3.4 Boundary Conditions 3.5 Analysis Techniques... 3.5.1 Overview. 3.5.2 Order Domain Techniques... 3.5.3 Time Domain Techniques 3.54 Summary of Analysis Techniques, 4, TORSIONAL VIBRATION MEASUREMENT... 4.1 Introduction 4.2 Overview Of Measurement Methods... 4.2.1 Analog Methods... 4.2.2 Digital Methods. 4.3 Analysis Of Digital Measurement Techniques.. 4.3.1 Simulation Code. 43.2 Aliasing 4.33 Leakage.. 4.3.4 Measuring Tooth Pass Times.. 43.5 Demodulation Techniques. 43.6 Tooth Spacing Variation . 4.3.7 Magnetic Pickup Vibration... 44 Summary And Conclusions. seen 6 |. OVERVIEW OF DIAGNOSTIC TECHNIQUE. 5.1 Problem Definitio 5.2 Solution Procedure... 5.3 Two Degree of Freedom Proof of Concept Study se DIAGNOSTIC ALGORITHM 0. 6.1 Overview 6.2 Constraint Equation Formulation... 6.3 System Matrix Equation Formulation, 6.4 Generalization of Equations.. ALGORITHM VERIFICATION... 7.1 Overview 7.2 Forward Simulation 7.2.1 Forward Simulation COSC nn 7.2.2 System Model 7.23 Forcing Function 7.3 Misfire Indicator Function... 110 7.4 Baseline Diagnostic Case. . ERROR SENSITIVITY ANALYSIS. 8.1 Introduction .. 8.2 Number And Choice Of Engine Orders.... 8.3 Choice OF Solution Parameters. 8.4 Use Of Multiple Engine Speeds. 8,5 Boundary Condition Assumptions... 8.6 Sensitivity To Noise.. 86.1 Sensitivity to System Model Noise.. 114 14 14 17 119 120 121 1D 8.6.2 Sensitivity to C 8.6.3 Sensitivity to Angular Velocity Measurement Error.. 8.64 Test Cases with Random Noise.. 122 130 131 134 8.7 Summary. 9, SUMMARY AND CONCLUSIONS 10. RECOMMENDATIONS FOR FURTHER RESEARCH... 10.1 Misfiring Cylinder Diagnostics 10.2 Torsional Measurement Methods snes 135 137 137 138 138 10.3 Force Reconstruction with Constraint Equations. 1, BIBLIOGRAPHY 0139 APPENDIX A: DIAGNOSTIC TEST CASES . . 142, LIST OF FIGURES Figure 1. Diagnostic Process.. Figure 2. Diagnostic System Models. Figure 3. Order Domain vs. State Space Deconvolution Figure 4, Illustration of Diagnostic Equation Indeterminacy . Figure 5. Rigid Crankshaft and Constant Load Torque Assumptions Figure 6, Slider Crank Mechanism Model Figure 7. Single Cylinder Excitation Torque. Figure 8. Typical Torsional Mass Elastic Mode! Figure 9. Torsional Vibration Analysis Techniques.. Figure 10, Holzer Analysis Forced Response. Figure 11. Transfer Matrix Analysis Figure 12, Impedance Model Analy Figure 13, Standard Digital Measurement Method. Figure 14. Effect of Counter/Timer Sample Rate on Measurement Noise Floot Figure 15. Digital Data Acquisition of Magnetic Pickup Signal. Figure 16, Effect of A/D Sample Rate on Measurement Noise Floor... Figure 17. Frequency Response of Standard Demodulation Method .. Figure 18, Frequency Response Function Correction Window... Figure 19. Corrected Frequency Response Function Figure 20. Effect of Gear Tooth Spacing Errors on Measurement Noise Floor. Figure 21. Cylinder Pressure Curve Parameterization Scheme Figure 22. Diagnostic Solution Procedure... Figure 23. Two Degree of Freedom System... Figure 24. Two Mass Diagnostic Algorithm. Figure 25. Diagnostic Algorithm Flowchart. Figure 26. Misfire Indicator: Baseline Case Figure 27. Iteration Time vs. Number of Engine Orders. Figure 28. Engine Order Selection.. Figure 29. Sensitivity wo System Model Nois Figure 30. Peak Pressure Balance Sensitivity to Cylinder Pressure Shape Random Noise... Figure 31. Excitation Using Different Cylinder Pressure Curves... Figure 32. Diagnostic Sensitivity t Cylinder Pressure Shape Bias Errors. 129 130 Figure 33. Diagnostic Results Solving for Peak Pressure Alone. Figure 34, Sensitivity to Angular Velocity Measurement Noise, LIST OF TABLES Table 1, System Model Complexity... Table 2. Diagnostic Analysis Method wn. ‘Table 3. Misfire Detection and Location Methods. ‘Table 4. Common Assumptions... ‘Table S. Comparison of Torsional Analysis Methods. ‘Table 6, Analog Torsional Measurement Methods. Table 7. Digital Measurement Errors... Table 8. Six liter Engine Mass Elastic Mode! ‘Table 9. Six liter Engine Forcing Function Table 10. Diagnostic Test Cases 132 LIST OF SYMBOLS T™ sPcP ApCP Ari ee ees eur Input force ‘Number of eylinders Number of engine orders used in the diagnostic system matrix equation Minimum number of engine orders needed for a solvable diagnostic equation set Linear displacement Piston top area After top dead center Parameter characterizing forcing function in proof of concept study Connecting rod length Reciprocating inertia for an engine cylinder Cylinder pressure Peak eylinder pressure (Crankshaft throw Torque ‘Top dead center ‘Torque applied to crankshaft by a cylinder Boundary condition torque atthe flywheel Cylinder torque due to gas pressure Cylinder torque due to inertia effects Dyn: Incremental positive increase in peak cylinder pressure ic timing value Incremental positive increase in dynamic timing Peak cylinder pressure correction Dynamic timing correction Coefficient for peak pressure in constraint equation Coefficient of dynamic timing in constraint equation Nominal value of excitation torque in constraint equation Constraint coefficient in proof of concept study Angular displacement or crankshaft angle Angular velocity Angular acceleration Frequency of excitation and response [4] [2] {a [Kea] [eveq] [4] [xe] [Hu {fy {Fea} {F°eq} {s} {TIN} {x} {Xeq} fra} {a} {o} {ok} {ou} Matrix of or coefficients Matrix of coefficients Matrix of 7 coefficients Equivalent complex system stiffness matrix Equivalent real system stiffness matrix ‘System impedance matrix Portion of system impedance matrix corresponding to known torsional displacements, Portion of system impedance matrix corresponding to unknown torsional displacements, force vector Equivalent complex system force vector Equivalent real system force vector Coefficient vector for unknown boundary condition torque ‘Vector of inertia torque excitations Complex system unknowns vector Equivalent complex system unknowns vector Equivalent real system unknowns vector ‘Vector of unknown peak pressure and dynamic timing parameters ‘Vector of torsional displacements ‘Vector of known torsional displacements. ‘Vector of unknown torsional displacements GLOSSARY OF TERMS “This dissertation presents a method for diagnosing and locating misfiring cylinders in an internal combustion engine using measured crankshaft torsional vibration response as an indicator of the applied torque for each cylinder. ‘The body ofthe report presupposes that the reader has a working understanding of internal combustion engine dynamics and engine torsional vibration analysis and measurement, An overview of each these topics is included in the body of the report. The following glossary of terms is included for those are not familiar with the field of engine dynamics and torsional vibration analysis and to minimize possible confusion, ‘The terms are grouped by subject matte. The lst is not exhaustive, but rather includes the terms most likely to need clarification. Internal Combustion Engine Terms Crankshaft A shaft that runs the length of the engine that transmits the engine torque to the flywheel, Crankshaft Nose The portion of the crankshaft that extends out ofthe front ofthe engine. The front of the engine is the end where the radiator fan is located. The rear of the engine is the end where the flywheel is located. Clutch A device attached to the flywheel that is used to transmit torque to the drivetrain, The clutch usually has a low torsional stiffness, a maximum torque slip device, and a disconnect capability. Cylinder Pressure Curve ‘The pressure waveform in the combustion chamber Driven Inertia The equivalent inertia ofthe drivetrain components behind the engine flywheel. Dynamic Ti ring The crankshaft angle where peak pressure occurs, usually defined in degrees after top dead cemter (degrees ATDC). Flywheel A large inertia disk attached to the rear of the crankshaft. The flywheel acts as the attachment point for the clutch, Flywheel Ring Gear A toothed ring around the circumference of the crankshaft flywheel, The flywheel engine starter motor engages the ring gear to rotate the crankshaft for engine starting. The ring gear also provides a location to measure torsional vibration atthe engine flywheel. Peak Pressure ‘The maximum pressure inthe cylinder pressure curve. TDC Top Dead Center. The crankshaft position where the crankshaft and connecting rod for a given cylinder are aligned in a straight line so that the piston is atthe top of its stroke. ‘Signal Processing Domain Terms Crankangle Domain Analysis performed with crankshaft angle as the independent variable, Data is sampled at constant increments of crankshaft angle. If the angular velocity of the system is known, time ‘domain data can be converted (o the crankangle domain using the relation, @ = at and then re-sampling the data at constant increments of crankshaft angle. Frequency Domain Analysis performed with frequency as the independent variable. Analysis is typically performed at discrete, equally spaced frequencies within some frequency band. Order Frequency referenced to a multiple ofthe mean shaft speed. For a shaft turning at 1000 rpm, 1* ‘order is 1000 rpm, 2% order is 2000 rpm, 3" order is 3000 rpm, etc. For a four stroke engine, one engine cycle corresponds to two complete revolutions of the crankshaft. The fundamental frequency of the excitation and response for the engine, therefore, is one half engine speed, or Ys order. Order Domain Frequency analysi in which the analysis frequencies are orders. Performing a discrete Fourier transform on crankangle domain data that represents exactly one revolution of data results in harmonic frequency components at integer engine orders. Time Domain Analysis performed with time as the independent variable. Data is sampled at constant increments of time, Temporal Filtering Partitioning a time domain waveform into a series of contiguous waveforms. This process is used to partition the applied torque waveform for a crankshaft into waveforms separating the ‘input for individual cylinders. 10 ngine Torsional Vbration Analysis Single Cylinder Excitation Torque The torque applied to the crankshaft by a single cylinder during an engine cycle, Single cylinder excitation torque can be expressed in the time, crankangle, frequency, or order domains. Engine Torsional Vibration Analysis The analytical procedure of predicting the torsional vibration free and forced response of an internal combustion engine and connected drivetrain, Firing Frequency The frequency of combustion events in an engine. In a four stroke engine, each cylinder fires every two revolutions of the crankshaft, resulting in a firing frequency (in orders) equal to the number of cylinders divided by two. For example, the firing frequency of a four stroke, six cylinder engine is 3 onder. Torsional Vibration Variation in shaft speed about some mean speed. Torsional vibration includes both rigid body speed fluctuation and dynamic torsional deformation. ‘Torsional Vibration Measurement Magnetic Pickup A variable reluctance sensor (VRS) used to generate a periodic waveform whose frequency is equal to the tooth passing frequency of a magnetic gear. Shaft Encoder An optical disk used to produce a periodic waveform whose frequency is proportional to the angular velocity of the attached shaft. Misfiring Cvlinder Diagnostic Terms Misfiring Cylinder cylinder with a cylinder pressure curve peak pressure that is significantly lower than normal. Common causes of misfiring cylinders include faulty fuel injectors, faulty valves, and faulty piston rings Misfire Detection ‘The detection of the presence of a misfiring cylinder. Misfire Identification/Location The identification of the misfiring cylinder(s). 1. INTRODUCTION ‘A misfiring cylinder in an internal combustion engine results in decreased engine performance, increased exhaust emissions, and increased powertrain noise and vibration. Also, a misfiring cylinder is ‘often symptomatic of a more serious engine problem that can lead to catastrophic engine or drivetrain failure if not corrected. For engines with high cylinder counts such as sixteen cylinder engines used in locomotives and off road mining trucks, the overall power drop associated with a misfiring cylinder is small and the problem is often not apparent to the operator until a catastrophic failure occurs. With such engines, once a misfire is diagnosed, the identification of the offending cylinder is often a long process involving the removal of individual cylinder heads (these engines have a separate cylinder head for each cylinder) in succession until the offending cylinder is found, Current and proposed legislation requiring ‘on-board detection of engine misfire in automotive passenger vehicles has created an immediate industry need for reliable engine diagnostic systems. Existing angular velocity based diagnostic systems only work ‘with engines with six or fewer cylinders operating at low to moderate engine speeds. When each cylinder in an internal combustion engine fires, a torque pulse is applied to the crankshaft throw. This torque contains significant energy for the first 24 harmonics of cycle speed (see Section 3.2). The resulting periodic excitation causes the engine and driveline angular velocity to fluctuate in both rigid body and flexible modes. When a cylinder misfires, the applied torque to the crankshaft is altered, resulting in an altered engine and driveline angular velocity response. The objective of this doctoral research is to develop a novel and improved technique to use the measured angular velocity response of the system to diagnose and locate misfiring cylinders, Current angular velocity based misfire diagnostic techniques include threshold criteria, pattern recognition, and model based deconvolution, All current techniques require the assumption of non- overlapping firing pulses and a rigid crankshaft. These assumptions have limited misfire diagnostic suocess so far to engines with six or fewer cylinders operating at low to moderate engine speeds. 13 ‘The goal of the doctoral research is to develop a diagnostic system which will work for engines with overlapping firing pulses and with significant crankshaft deflection. The proposed diagnostic method will, forthe fist time, enable 100% detection of misfire in passenger vehicles, and also allow misfire detection and faulty cylinder identification in engines with high cylinder counts and significant crankshaft deflection. The fundamental process in the diagnostic method is to deconvolve measured angular velocity atthe crankshaft nose and flywheel through a dynamic model forthe engine and powertrain to arrive at the unmeasured cylinder pressure input for each cylinder. Additional equations required to solve the ‘otherwise indeterminate equation set are generated by forming constraint equations between the harmonic components of the forcing function at each cylinder. The research results in several improvements in state of the art techniques: 1. A new diagnostic method allowing misfire detection and location for engines with flexible ‘crankshaft behavior and overlapping firing pulses. 2. Anew, more accurate method of measuring rotating shaft angular velocity fluctuation. ‘A new method of input force reconstruction by forming constraint equations between harmonics. ‘Chapter Two reviews the literature on angular velocity based mistire diagnostic methods. The relative merit of each ofthe existing diagnostics methods are presented. The choice and consequences of the underlying assumptions in the various methods are reviewed in detail. The remaining challenges for a comprehensive, accurate and reliable diagnostic system are explained. A summary of relevant literature is then listed by author. Finally, the current state of the artis summarized. Chapter Three provides an overview of engine torsional vibration analysis techniques. Engine torsional vibration analysis refers to the process of using a dynamic system model, system boundary conditions, and an applied forcing function to calculate the dynamic torsional response of an engine and powertrain system. ‘There are many assumptions in this process that is the accuracy and applicability of the results. The consequences of each of these assumptions is explained in detail, A thorough understanding of torsional vibration analysis techniques, assumptions, capabilities, and limitations is 4 necessary since many of the same techniques and assumptions are used in the proposed diagnostic algorithm. Chapter Four explores torsional vibration measurement techniques. The proposed diagnostic method relies on measured flywheel and crankshaft angular velocity as key inputs 10 the diagnostic algorithm, Consequently, errors in the measured angular velocity result in errors in the diagnostic process. An overview of available measurement methods is presented. Current measurement methods include analog and digital techniques. Existing measurement methods are marginally accurate for the proposed diagnostic method. A new, digital measurement technique is proposed which significantly increases measurement accuracy over curtent state ofthe art. Chapter Five is an overview of the proposed diagnostic technique. The diagnostic problem is explained and defined. The solution procedure is outlined, explaining how the solution is made possible through the generation of Tinea constraint equations between separate harmonics inthe forcing function. A two degree of freedom system is analyzed using the proposed diagnostic method as a proof of concept study. Chapter Six 1 detailed description of the engine diagnostic algorithm. The algorithm is an extension of the procedure summarized in Chapter Five. An overview of the diagnostic algorithm including a flow chart of the program structure is presented. The derivation of the constraint equations relating the harmonics of the forcing functions is presented. The main matrix equation set relating the measured system response to the unknown inputs is derived for the most complicated case, The resulting equation set is then generalized to include all other cases. Chapter Seven presents analytical verification of the diagnostic algorithm. A torsional vibration simulation code is used to simulate the measured torsional response at the crankshaft nose and flywheel for a known set of eylinder pressure inputs. These simulated measurements are then used as input to the diagnostic algorithm. ‘The diagnostic algorithm develops constraint equations between the harmonics of the forcing function at each cylinder, enabling calculation of the original individual cylinder inputs. ‘Chapter Eight contains an error sensitivity analysis for the diagnostic algorithm. The diagnostic process relies on several key assumptions regarding the system dynamic model, the system boundary 15 conditions, and the nature of the applied torque at each engine cylinder. The process also relies on measured angular velocity data for input. Error in the key assumptions and the measured data is inevitable, ‘This error results in error in the diagnostic process. The error in the diagnostic method resulting from errors in each key assumption and measured input is quantified. A series of test cases are presented in which expected levels of noise are added to all key assumptions and measurements. The test ‘cases include a six liter, six cylinder engine, anda fifty lite, sixteen cylinder engine, ‘Chapter Nine presents the summary and conclusions. The diagnostic method promises to enable diagnosis of misfire and identification of the misfiring cylinders in engines with large cylinder counts ‘operating with significant crankshaft deflection, This conclusion is based on verification of the method With realistic errors on the measured data and Key assumptions. ‘Chapter Ten contains recommendations for further research. The next step in the diagnostics research is to evaluate the proposed diagnostic procedure by taking data on a real engine. Each of the key assumptions such as the model forthe eylinder pressure curve, the model for the dynamics of the system, the boundary condition model, and the assumed torsional measurement accuracy can be evaluated with ‘measured data, Finally the diagnostic process should be implemented on an engine. The ideas for new signal processing techniques for digital measurement of torsional vibration and the idea of using constraint equations between harmonics as a forcing function reconstruction technique also deserve further study. ‘The Appendix contains the results of a of diagnostic test cases. Diagnostic results are presented for a six cylinder and a sixteen cylinder engine. The test cases are run with expected levels of noise and error added to all key assumptions and measured input. 16 2. LITERATURE REVIEW Note: A glossary of terms relevant (o internal combustion engines, torsional vibration analysis and ‘measurement, and misfiring cylinder diagnostics is included at the beginning of this dissertation. 2.1 Introduction ‘The technical literature relating to engine misfire detection and faulty cylinder identification using crankshaft angular velocity measurements is reviewed. Reference dates range from 1983 to the present, Much of the current research is in response to the California OBDII (California On Board Diagnostics I) 1994 requirements for on-board misfire detection in passenger vehicles. Detecting the presence of engine misfire using crankshaft speed fluctuations is practical with current methods for certain road and engine conditions. Identifying specific faulty cylinders is more difficult, especially at high speeds, low loads, and with dynamic road inputs. The presence of crankshaft torsional vibration and overlapping firing pulses complicates the diagnostic process. Consequently, most current diagnostic methods are aimed at four and six cylinder engines operating at low to moderate engine speeds. An overview and comparison of current, crankshaft speed based, diagnostic methods is presented. The relative advantages of each method are discussed. The effect of various underlying assumptions are explored. Remaining challenges are identified. A summary of relevant literature is then listed by author. Finally, the current state of the artis summarized. 2.2. Overview And Comparison Of Current Methods 2.2.1 Introduction ‘When each cylinder in an internal combustion engine fires, a torque pulse is applied to the ‘crankshaft throw. This torque contains significant energy for the first 24 harmonics of cycle speed (cycle ” speed is half engine speed for a four stroke engine). The resulting periodic excitation causes the engine ‘and driveline angular velocity to fluctuate in both rigid body and flexible modes. When a cylinder misfires, the applied torque (0 the crankshaft is altered, resulting in an altered engine and driveline angular velocity response. ‘The goal ofall the reviewed engine diagnostic methods is to use the measured angular velocity response of the system to develop insight into the nature of the forcing torque applied by the cylinders, The applied forcing torque is then, in tur, related to individual cylinder contributions, All engine speed related misfire diagnostic methods incorporate some common téchiniques. First, the angular velocity of one or more components inthe drivetrain system is measured. In most cases, there isa single measurement location at the flywheel ring gear. Additional measurements, sometimes taken at the crankshaft nose or driven inertia, allow a more complete characterization of the system response. The measurement must have sufficient bandwidth to accurately represent the angular velocity fluctuation ‘within an engine cycle, Definitions of the crankshaft nose, flywheel ring gear, and driven inertia are included in the Glossary. Once the angular velocity response is known, analytical or experimental threshold criteria for speed fluctuations can be used to determine the presence of engine misfire. Experimental or analytical databases of driveline responses can also be used with pattern recognition techniques to identify misfiring cylinders. Alternatively, the measured response can be deconvolved through an analytical system model to reconstruct the applied forcing torque. Finally, the system inputs can be guessed and then refined in an iterative fashion until the assumed inputs and system model produce a simulated output that matches the measured angular velocity response. The accuracy of the individual approaches is strongly affected by the assumptions that must be made to perform the analysis. ‘The difference between normal torsional vibration analysis and engine diagnostic analysis is explained, Current diagnostic methods are categorized according to system model complexity, analysis method, misfire detection and location method, and assumptions employed. Each combination of approaches offers distinct advantages and drawbacks. An overview and comparison of current methodology is reviewed. Remaining challenges are also discussed. 2.2.2. Engine Torsional Vibration Analysis vs. Engine Diagnostic Analysis Engine torsional vibration analysis refers to the process of applying known cylinder torque inputs toa known system model to calculate the torsional vibration response of the system, This is referred to as the “forward solution”. Engine diagnostic analysis refers the process of using the measured torsional response of the system and the known system model to calculate the unknown system inputs for each cylinder, This is referred to as the “reverse solution”, The two processes are illustrated in Figure 1 ‘The forward solution is essentially a matrix muliplication procedure, while the reverse, diagnostic solution involves a matrix inversion process. ‘The matrix inversion required for the reverse, gnostic procedure is only possible if the number of independent torsional vibration response measurements is qual to or greater than the number of independent inputs, The numberof unknowm inputs is equal to the number of cylinders. The number of available equations is equal to the number of torsional vibration ‘measurement response locations. Since the number of unknown inputs exceeds the number of response measurements, the reverse diagnostic process is not possible without the introduction of additional assumptions. For example, a six eylinder engine has six unknowm inputs (one unknown input at each engine cylinder). Torsional vibration measurements taken at the flywheel and crankshaft nose give rise 10 two independent equations. Since the number of unknowns is six, and the number of equations is two, the ‘equation set cannot be solved without making assumptions to reduce the number of unknowns or increase the number of equations. The problem i illustrated for a six cylinder engine in Figure 4. The assumptions made in all current diagnostic methods reduce the number of unknown inputs to the number of measured outputs. This is accomplished by making simplifying assumptions about the system model. The assumptions and ther effect onthe diagnostic process are presented in detail in this chapter. The diagnostic method proposed inthis dissertation relies on a separate set of assumptions which increases the number of independent equations in the system, rather than reducing the number of unknowns, The diagnostic procedure proposed is in this dissertation is outlined in Chapter Five. The agnostic algorithm is developed in Chapter Six. An algorithm verification is presented in Chapter Seven. An error analysis is presented in Chapter Eight, 19 Forward Solution Diagnostic (Reverse) Solution | Inputs {x} Inputs { x} “ Outputs { b} Outputs { b} [Alx} = fo} {a}= [AP fo} + Number of Eqns is number of outputs + Number of unknowns is number of inputs + Equation set is solvable only if: Number of Outputs 2 Number of Inputs Figure 1. Diagnostic Process 20 2.2.3. System Model ‘A system model is used to relate the known torsional system response to the unknown system inputs (the individual cylinder contributions), Table 1 categorizes system models according to increasing complexity. Corresponding system models are shown in Figure 2, ‘= No model «Single inertia for engine (rigid crankshaft) ‘© Rigid crankshaft with driveline model ‘ Flexible crankshaft with driveline model Table 1. System Model Complexity ‘The simplest method is to use no model at all, im which case no attempt is made to solve explicitly for the system inputs. This option is available for threshold criteria and pattern recognition techniques that rely solely on the measured angular velocity data. To reconstruct applied torque from the measured angular velocity data, a system model is required. Models range from a single inertia for the engine to a ‘complete flexible crankshaft and driveline model. For each of the diagnostic methods, the number of unknown inputs that can be calculated is equal to the number of independent response measurements that are taken. If only one response measurement is taken (e.g. velocity fluctuation at the flywheel ring gear), only one independent input can be calculated. 21 System Model Complexity Single Inertia for Engine Rigid Crank with Driveline Model T T. T, 1 2 Oveesuret "| = Osea Flexible Crank with Driveline Model Ta a eT Ts H 9, Measured e. ‘Measured Figure 2. Diagnostic System Models 2 ‘System models employing a rigid crankshaft allow the input from individual cylinders to be lumped into a single unknown composite torque, since there is no dynamic effect between engine cylinders. This allows computation of the composite applied torque from a single speed measurement. If, in addition to assuming a rigid crankshaft, the firing angles are well separated, then the calculated applied torque ‘waveform can be partitioned in the time domain into individual inputs from each cylinder. System ‘models incorporating a flexible crankshaft require the treatment of each cylinder input as a separate, independent unknown. For applications where torsional driveline response is significant, the model must also include driveline components. Additionally, significant dynamic input from the road will introduce another system input that must be dealt with, A simple model allows a more complete reconstruction of the applied torque in the sense that the ‘number of unknowns is reduced. However, assumptions that must be made in developing the simple ‘model (e.g. rigid crankshafi, constant load torque, constant clutch stiffness), often limit the engine type ‘and operating range where the model is valid. A rigid crankshaft assumption is not valid under operating ‘conditions where crankshaft torsional vibration is significant. Also, temporal filtering of a calculated ‘composite torque to arrive at individual cylinder contributions is not valid for engines with overlapping firing pulses 2.2.4 Analysis Methods Table 2 lists analysis methods used to reconstruct the system forcing function from the measured system response. Forcing function reconstruction techniques are based on order domain or state space system deconvolution. Order domain frequency response functions are evaluated at discrete engine orders to take advantage of the periodicity of the combustion process. The differences between order domain and state space deconvolution methods are outtned in Figure 3. 2B + Order domain deconvolution + Multiples of firing frequency « Firing frequency only ‘* Multiples of half engine speed «State space deconvolution Table 2. Diagnostic Analysis Methods ‘Order domain methods are limited to a steady state analysis. Inputs for the deconvolution process are the magnitude and phase of the measured torsional system responses. The outputs of the deconvolution are the applied order domain system torques. Most current order domain diagnostic ‘methods limit analysis to the order corresponding to firing frequency. The firing frequency in orders for 1 four stroke engine is the number of cylinders divided by two since each cylinder fires once every two revolutions of the crankshaft. For example, the firing frequency of a six cylinder, four stroke engine is 3° ‘order. Some researchers have included multiples of frequency in the analysis to obtain additional information, Ac firing frequency and multiples of firing frequency, excitation from all the cylinders is in phase, ‘This means there is no excitation that is trying to “twist” the crankshaft at these frequencies. This condition is important if the engine model includes a rigid crankshaft assumption. In an evenly firing engine with a torsionally rigid crankshaft, the net torsional excitation for all engines orders other than multiples of firing frequency will sum to zero, Ifa flexible crankshaft is modeled, measured responses at ‘multiples of half engine speed become significant, and can be employed to generate additional information. Order Domain Deconvolution Frequency Ow Response (@) Pareto Tio) FRF(@) steady state, time invariant Independent Variable: Excitation Order Algorithm Input: Torsional Displacement Algorithm Output: Applied Torque State Space Deconvolution ew P(t) steady state or transient, time invariant or variant Independent Variable: Time Algorithm Input: Torsional Displacment Algorithm Output Cylinder Pressure Curve Figure 3. Order Domain vs. State Space Deconvolution State space deconvolution is valid for transient solutions. Input for the deconvolution process is the time domain measured system response. Unlike the order domain method, the slider crank mechanism ccan be included in the system model by linearizing the slider crank model and updating the system dynamic model at each time step. As such, the applied cylinder pressure, rather than the applied crankshaft torque can be the output of the deconvolution process. A drawback with this method is the tunobservability in the system at cylinder top dead center (TDC). At TDC, the cylinder pressure has no effect on the applied crankshaft torque, hence cylinder pressure at TDC cannot be reconstructed from the known system response at TDC (sce Figure 6). Since determining the input cylinder pressure is most critical around TDC, the unobservability at TDC is the most serious problem with reconstructing cylinder pressure directly rather than reconstructing the applied crankshaft torque, 2.2.5 Misfire Detection And Location Misfire detection refers to detecting a significant unevenness in cylinder power contribution, while misfire location refers to determining which cylinder or cylinders are at fault. Table 3 categorizes misfire detection and location methods. Methods include threshold criteria, pattern recognition, temporal filtering (see Glossary) of the applied torque profile, and iterative techniques. calculated composite torque. This assumption is marginal for eight cylinder engines and not valid for engines with more than eight cylinders. 2.2.7 Remaining Challenges Current diagnostic methods range from relatively simple thresbold criteria to diagnose misfire, to sophisticated model based deconvolution to locate misfiring cylinders, The common threads among all methods are the assumptions that reduce the number of unknowns to the number of measured outputs. ‘Most significant among the assumptions are the rigid crankshaft, constant load torque, and non- ‘overlapping firing pulse assumptions. Diagnostic methods relying on rigid crank assumptions break down at high engine speeds where crankshaft torsional vibration is significant. The constant load torque assumption also creates problems for automotive diagnostic systems when the vehicle encounters rough roads. The non-overlapping firing pulses assumption is not valid for high cylinder count engines. ‘The remaining challenge is to find a way around the restrictions imposed by the current set of assumptions. Only then will misfire detection system be capable of nearly 100% reliability for automotive and heavy duty applications. 2.3 Summary By Author Of Relevant jterature Sood et a. (1985, [1,21 propose three statistical methods for engine fault analysis and an iterative parameter estimation approach. All methods are based on a rigid crankshaft model and flywheel angular speed measurements, Statistical detection methods include: cross correlation of flywheel speed signatures ‘with known fault patterns, similarity measures between flywheel speed signature and known patterns, and pattern recognition using the coefficients of a polynomial representation of the measured flywheel speed signature. The parameter estimation approach involves linearizing the engine model around an assumed operating point, and then iterating on the applied inputs until the measured speed waveform is matched. ‘The model states are angular velocities. The model input is the net applied crankshaft torque for all the 32 cylinders. Faulty cylinders are isolated by looking at the relative net torque contribution during each combustion event, Mihele and Citron (1984, (3)) develop an engine roughness criteria based on flywheel speed fluctuations, The contributing torque for each cylinder is represented as an equivalent inertia multiplied by the change in flywheel speed over the elapsed time between successive TDC’s. The engine roughness criteria relates individual cylinder torque variation to mean engine torque. The roughness criteria suocessfully identifies misfiring engines and over lean air-fuel mixtures. Citron, Higgins, and Chen (1989, (4)) develop a system for determining individual cylinder pressure contributions based on flywheel speed measurements. The system model includes a torsional stiffness between the rear cylinder and the flywheel. The crankshaft model is rigid between cylinders. The system model, measured flywhee! velocity, and known inertia torque are used to determine the net applied crankshaft torque waveform. All analysis and data is performed at the firing frequency. ‘The torque contribution for each cylinder is determined by partitioning the applied torque waveform into individual input waveforms for each firing pulse. Mean engine torque is determined as a linear function of the pressure torque fluctuation, Rezeka and Henein (1987, (5) develop three methods for the identification of misfiring cylinders based on measured flywheel angular velocity. The three methods are based on criteria for mean cyclic acceleration, cyclic period of acceleration, and maximum variation in angular velocity Mueller, Hadad, and Marinc (1988, [6)) develop a spark ignition (SI) engine misfire detection scheme by measuring the time between ignition pulses in free acceleration tests. Net applied crankshaft torque is calculated from a knowledge of the crankshaft accelerations, system inertia and the characteristics ofthe friction and brake torque. The system is most effective at low to moderate engine speeds. Maver and Watts (1989, (7)) develop a method to identify misfiring cylinders by measuring flywheel speed. A linear regression is performed on the speed signal upward and downward trend during each cylinder firing. Faulty cylinders are identified by comparing regression results for each cylinder. The method is most successful under free acceleration (versus steady state) conditions. Mauer and Watts 3 (1989, [8}) also develop a misfiring cylinder identification method by measuring flywheel and crankshaft nose speed simultaneously. The relative motion between the front and rear of the crankshaft is used as an indicator of applied crankshaft torque. Cylinders low on power show up as reduced torque amplitudes. “This method is least effective when significant crankshaft torsional vibration is present. Mauer and Watts (4990, (9}) demonstrate that measurement of flywheel and crankshaft nose speed can be used to identify a failed crankshaft damper. Mauer (1991, (10) develops a method to determine the mean available torque in an engine by ‘measuring flywheel speed during a free acceleration test. A linear regression is performed on the flywheel speed over a fixed number of engine cycles. The slope of the regression line is proportional to mean engine torque. Mauer (1989, (11) also develops a model based method to determine individual cylinder misfiring through measurement of flywheel and crankshaft nose speed. The model reduces to a lumped inertia for the crankshaft nose and lumped inertia for the flywheel, An inertialess station for the firing cylinder is cradled between the two inertias via, torsional springs. The value of the torsional springs varies with the particular cylinder that is firing. The model and measured velocities are used to calculate applied torque during each firing. A coefficient of performance for each cylinder is determined using the calculated torque. A significant negative coefficient of performance indicates a faulty cylinder Experimental results are presented for a V-8 diesel engine operating with various cylinder faults under steady state and free acceleration conditions. Plapp, Klenk, and Moser (1990, {12)) investigate a misfire detection method based on flywheel angular velocity measurements. An engine roughness index is calculated based on changes in successive angular accelerations of the flywheel. An algorithm was developed to distinguish speed fluctuations due to engine misfire from speed fluctuations due to dynamic road input to the drive fine, The method is most

    You might also like