1832

IEEE SENSORS JOURNAL, VOL. 11, NO. 9, SEPTEMBER 2011

Virtual Sensors for Automotive Engine Sensors Fault Diagnosis in Second-Order Sliding Modes Qadeer Ahmed, Student Member, IEEE, Aamer Iqbal Bhatti, Senior Member, IEEE, and Muhammad Iqbal

Abstract—Automotive engine functions are entirely dependent on installed sensors performance. Any malfunction in the sensors can lead to degraded engine efficiency. This manuscript presents a novel scheme to devise virtual sensors for health monitoring of engine air intake path sensors. The proposed scheme assists in: Sensing of critical immeasurable parameters like Volumetric and Combustion efficiency, development of Virtual Sensor from manifold pressure dynamics for rotational speed and vice versa, Health Monitoring of the manifold pressure and crankshaft sensor. For the suggested scheme, two robust second-order sliding mode observers are employed that require two state mean value engine model based on inlet manifold pressure and rotational speed dynamics. The proposed methodology has the potential of online implementation on any automotive engine after minor tuning. In this paper, the procedure is customized for 1.3 L gasoline engine sensors: Manifold Pressure and Crankshaft sensor. The implementation results demonstrate that all the three mentioned tasks are accomplished efficiently. Index Terms—Second-order sliding mode observers and automotive engine, sensor fault diagnosis, sensor redundancy, virtual sensors.

TABLE I ACRONYMS USED IN THE PAPER

TABLE II SYMBOLS, DESCRIPTION AND THEIR VALUES FOR MEAN VALUE ENGINE MODEL

I. INTRODUCTION

S

ENSORS are integral part of modern ECU equipped vehicles. These sensors play vital role in on-board condition monitoring of vehicle and its engine. This is because of strict legislative regulations defined in OBD-II [1] and EOBD. In order to meet these demands automotive industry is constantly working to incorporate efficient on-board fault diagnosis methodologies. Besides legal requirements, end-user satisfaction is also a major concern of automotive industry. Fault free, fuel efficient and environmental friendly vehicles are always preferred. One of the main requirements of ODB-II for SI engine is to monitor the health of its AIS. Any OBD-II compliant vehicle will be equipped with MAF/MAP, throttle position and engine speed sensor. These sensors provide useful information to examine the health of air intake system and to diagnose malfunctioning of the components involved in AIS. The condition monitoring is totally dependent on the proper functioning of these sensors. Any malfunction of these sensors will lead to poor engine performance. A brief discussion will give us an overview Manuscript received July 13, 2010; revised November 22, 2010; accepted December 19, 2010. Date of publication January 13, 2011; date of current version July 29, 2011. This work is supported by ICT R & D Fund, Pakistan. The associate editor coordinating the review of this manuscript and approving it for publication was Prof. Okyay Kaynak. The authors are with the Control and Signal Processing Research Group, Mohammad Ali Jinnah University, Islamabad 111-87-87-87, Pakistan (e-mail: [email protected]; [email protected]; [email protected]). Digital Object Identifier 10.1109/JSEN.2011.2105471

of the malfunctions due to MAP and crankshaft sensor and how researchers have diagnosed the faults in these sensors. See Table I for acronyms used in this paper and Table II for symbols, description, and values for the mean value engine model.

1530-437X/$26.00 © 2011 IEEE

AHMED et al.: VIRTUAL SENSORS FOR AUTOMOTIVE ENGINE SENSORS FAULT DIAGNOSIS IN SECOND-ORDER SLIDING MODES

A. Air Intake System Sensors Fault Diagnosis Engine fuel efficiency depends on the AFR of in-cylinder mixture that is required to be as close to its stoichiometric proportions as possible. The standard stoichiometric ratio can be maintained if AIS performance is ensured. Any malfunction in AIS will affect the amount of oxygen required for complete combustion, thus disturbing stoichiometric proportion. One of the main factors that retards fuel efficiency is malfunctioning of MAP and crankshaft sensor. A faulty MAP sensor may let an end-user experience the following. • Exhaust Gases and Gas Smell. • Rough Idling. • Poor Gas Mileage. • Hesitation/Poor Pick Up. However, in actual, due to various nature of sensor faults, MAP sensor generate false measurements that can result in the following. • Deviation in AFR, that may cause emissions to increase (Pollution). • Lean or rich air fuel mixture, that may cause misfire (Hesitation). Similarly, a crankshaft sensor is used to measure angular speed. If it has failed or is failing, certain timing problems will arise in engine function. The engine may start normally in some cases, but will cut off after a few minutes (or seconds) of operation. More than likely, the engine will be unable to start at all. Since the crankshaft sensor is responsible for engine timing, the driver may experience engine backfire or irregular angular speed function, if the vehicle starts at all. These symptoms are visible only after fault in the sensors has occurred. These faults can be due to bias or drift in sensor outputs. Other causes may include loss in effectiveness, damaging of sensors or frozen sensor outputs. In order to avoid such situations, early diagnosis of sensor health becomes inevitable. Currently, automotive industry utilizes many experience and lookup tables-based techniques to monitor sensor faults. However, in public literature, various researchers have attempted to diagnose sensor faults in automotive engine. Sensor fault diagnosis has been carried out with various algorithms. Balaban et al. [2] employed neural network-based classifier to figure out different faults in sensors with its application in aerospace systems. Similarly, Partap et al. [3] developed an artificial neural network-based virtual fault detector for detection and identification of faults in Wheatstone bridge-oriented transducers of a computer-based measurement system. In automotives, Capriglione et al. [4] discussed sensors fault detection, isolation, and accommodation procedure for public transportation. The approach was to develop analytical redundancy for installed sensors that can diagnose the faults efficiently. Jacob et al. [5], analyzed various signals of vehicle sensors and control module to diagnose its faults. The procedures of pattern recognition were employed to classify different faults. Rizzoni et al. [6] proposed detection filters to diagnose sensor failures in automotive engine control systems. The detection filter utilized analytical redundancy within a dynamical system to isolate the cause and location of abnormal behavior. Pau-Lu et al. [7] suggested a hexadecimal decision table to relate all possible failure patterns to the residual code. The residual

1833

code was obtained through simple threshold testing of the residuals, which were the output of a general scheme of residual generators. The proposed diagnostic system was applied to automotive engine sensors and actuators. Nyberg [8] developed a model-based fault diagnosis algorithms to detect air intake system faults: sensor faults and manifold leakage. Choi et al. [9] suggested fusion of classifiers to diagnose various faults in automotive systems. Capriglione et al. [10] described the hybrid solution, based on artificial neural networks, and the production rule adopted in the realization of sensor fault detection, isolation, and accommodation scheme for automotive applications. However, fault diagnosis of automotive engine sensors with robust second-order sliding mode observers is proposed in this paper. SOSM observers provide platform to devise virtual sensors that results in sensor redundancy. This redundancy can be utilized to monitor manifold pressure and crankshaft sensors health. B. Sliding Mode Observers and Fault Diagnosis Linear and nonlinear observers are extensively used to develop fault diagnosis methodologies. These observers provide a facility of virtual sensors development. Virtual sensors measure a phenomena through another interrelated process, thus providing a redundancy in measurements. This redundancy can be utilized for the diagnosis of system components. Like, Heidtmann and Soffker [11] proposed proportional integral observer-based approach for diagnosis and prognosis tasks of elastic mechanical structures. SMOs are widely used for fault diagnosis of dynamical systems. The key attribute of nonlinear SMO is its robust nature. A Super twisting-based SOSM Observer inherits the same robustness as first-order sliding mode and ensure reduced chattering phenomenon [12]. These observers make the states observation without filtering and parameters estimation with just one first-order low-pass filter. Whereas, in SMOs, multiple filters are used for state/parameter estimation, that may lead to corruption in results [13]. Similarly, an anti peaking high gain SOSM observer may or may not require linear or nonlinear model for state observation. Unlike proportional (Luenberger Observer), proportional integral observers that only work with linear models which makes parameter estimation of highly nonlinear systems very difficult. On the other hand, KF-based robust state estimation heavily depends on linear initial system information and noise distribution for its convergence. Moreover, the nonlinear framework of SOSM-based technique is efficient, computationally cheap and online implementable as it is free of multiple matrices evaluation and multiplication like in prediction stage of KF. In addition, it does not involve the system to be Jacobian linearized as required in EKF. This linearization can be inaccurate for a system with large nonlinearities for a short time period [14]. Similarly, the SOSM observer is free of transformations, computation of sigma points, Cholesky factor updating, efficient least squares, and decompositions as mandatory in UKF and its variants [15]. Therefore, one can implement the SOSM observer using a low-cost embedded system very easily. Tan et al. [16] proposed SMO-based sensor fault detection for chemical plants. The same authors in [17], proposed a fault tolerant control methodology after detecting sensor faults

1834

IEEE SENSORS JOURNAL, VOL. 11, NO. 9, SEPTEMBER 2011

using sliding mode nonlinear observers. Rafael et al. [18] presented a SMO for condition monitoring of chemical reactors. The designed SMO is used for online estimation of heat in continuous stirred tank reactors, that facilitates in its FD. Jiang et al. [19] exploited geometric conditions such that the original nonlinear system can be transformed into two different subsystems with uncertainty. The first is in the generalized observer canonical form, which is not affected by faults. The second, whose states can be measured, is affected by the faults. The faults in the second system were estimated by SMO designed for the first subsystem. Similarly, Keng et al. [20] explored sliding mode techniques for fault diagnosis of diesel engine coolant system. Daniele et al. [21] proposed SOSM-based FD scheme for rigid manipulators. Castillo et al. [22] used SMOs to estimate the system state vector, from this the FD signal-residual is generated by the comparison of measured and estimated output. Wu and Saif [23] proposed a super twisting-based SOSM observer for FD of a class of uncertain dynamical systems. The observer input was designed by using a PID-type iterative learning algorithm to detect, isolate, and estimate faults. Keeping in view the attributes of SOSM observers and its authentic applicability for FD, the authors propose a novel scheme for the development of virtual sensors to monitor manifold pressure and crankshaft sensor health. The proposed scheme provides a methodology for estimating immeasurable parameters: Volumetric and Combustion Efficiency, Sensor Redundancy by facilitating with Virtual Sensors and Health monitoring of sensors involved in AIS. For all mentioned tasks, successful implementation has been performed on commercial vehicle engine. The proposed methodology has the potential of online implementation on any SI engine after minor tuning. The rest of this paper is arranged as follows. Section II explains the mean value engine model required for SOSM observer. After the validation of engine model, Section III contains the methodology for the development of virtual sensors via robust SOSM observers. Section IV describes the experimentation and validation of the proposed methodology for sensor health monitoring. The concluding remarks are in Section V followed by the Appendix and References.

Fig. 1. Air intake system of spark ignition engine.

Fig. 2. Validation of MVEM with 1.3 L engine. Angular speed and manifold pressure of MVEM and engine with the same inputs.

brake torque is the difference of torque produced by combustion and other agents. These agents include pumping and load torque. Load torque includes torque induced by loading elements, friction in between engine components, etc. Finally, a four stroke engine can be expressed with inlet manifold pressure and engine angular speed dynamics as

(1) where

II. MEAN VALUE ENGINE MODEL Air intake system of automotive engine is shown in Fig. 1. The air flow across the AIS can be modeled on basic fluid dynamics laws. The adiabatic flow across the throttle body/butterfly valve can be modeled as air flow from orifice. The assumptions are: one-dimensional compressible flow has no friction and inertial effects in the flow and there is no change in temperature and pressure (lumped parameter approach) during the flow [24], [25]. The manifold dynamics can be modeled on the basis of filling and emptying of air behaving as prefect gas. The mass and energy of the air serves as inputs and outputs of the receivers. It is assumed that no substantial changes occur in energy and no mass and heat transfers through the manifold walls [24]. The engine itself acts as a volumetric pump [25], [26], a device that enforces a volume flow approximately proportional to its speed. The rotational speed dynamics are developed from combustion process modeled by Otto cycle [25], [27]. The available

and other expressions can be seen in Appendix A. More details of engine modeling can be seen in [24]–[26]. Remark 1: Validation of the model in (1) was performed against a commercial gasoline engine under uniform temperature and steady-state conditions. The throttle was manipulated at 110, 170, and 230 s and kept constant for some time. Engine operating temperature remained at 89 C and unwanted loads were avoided during the validation experiment. The mathematical model in (1) and the commercial vehicle engine exhibit same response to same steady-state inputs, as shown in Fig. 2. Therefore, the model can be used efficiently to develop virtual sensors and proposed fault diagnostic scheme.

AHMED et al.: VIRTUAL SENSORS FOR AUTOMOTIVE ENGINE SENSORS FAULT DIAGNOSIS IN SECOND-ORDER SLIDING MODES

III. SCHEME FOR VIRTUAL SENSORS DEVELOPMENT As discussed in Section I-B, SOSM observers provide platform to devise virtual sensors for automotive engine. In this section, two second sliding mode observers will be designed for manifold pressure and angular speed dynamics. For SOSM observers, the states of engine inlet manifold and rotational speed dynamics can be redefined as . Based on these states, the validated engine dynamics can be redefined as

(2) The input rotational speed.

is the throttle angle. The outputs are the engine inlet manifold pressure and and are given in Appendix A.

1835

are the observer gains. The above OBD-II kit. The gains second-order sliding mode observer inherits anti-peaking structure [29], where and reach the sliding manifold one by one in a recursive way. The error dynamics before converging to the comes out to be surface, i.e., (9) where

1) Boundedness Analysis: The convergence of the proposed observer is guaranteed provided the manifold states and parameters governing states equations in (3) are bounded in nature. As ECU equipped SI engine is operating under closed loop and bounded input , the pressure dynamics will always remain bounded. This characterizes the system as Bounded Input and Bounded Output (BIBO) system. Therefore, the uncertainty function will always satisfy Assumption 1. satisfies Assumption 1: The uncertainty function

A. SOSM Observer for Inlet Manifold Pressure Dynamics The inlet manifold pressure dynamics can be written as

(10) (3)

where the known dynamics of manifold are (4) and the dynamics to be estimated are (5) The dynamics in and are Lebesgue-measurable in a compact region of . The detailed analysis shows that the dynamics of inlet manifold of automotive engine are observable and identifiable as the respective matrices are nonsingular [28]. In order to estimate the state vector and extraction of unknown variables, the observer based on [29] is proposed for inlet manifold dynamics

, and and as two such that positive numbers [29]. 2) Convergence Analysis: In order to analyze the convergence of the proposed observer for inlet manifold, , and and , such that Assumption 2 is satisfied. Assumption 2: There exist two positive constants and such that

(11) Based on Assumption 1 and 2, , , we have

, (12)

(6)

(13)

represents the corresponding observer states and are the observer injectors based on super twisting algorithm. These injectors model the uncertainty in the system in order to eliminate the error between the estimated states and the actual states, i.e., and . These injectors are defined as [29]

Theorem 1: If the condition (12) holds for engine manifold dynamics in (3), and the parameters of the proposed observer in (6) are selected according to following criteria [29]:

(7)

will converge to in finite time. The The constants and can be chosen as explained in convergence proof [29]. Remark 2: The choice of and depends on the uncertainty bound and the initial state estimation error in the worst case [29]. As is bounded for engine pressure dynamics, therefore in order to avoid chattering and ensure convergence, values of and are chosen according to Table III.

where

and if if

& & (8)

The state is available for measurement from the sensor installed in the engine inlet manifold that can be accessed through

(14)

1836

IEEE SENSORS JOURNAL, VOL. 11, NO. 9, SEPTEMBER 2011

B. SOSM Observer for Engine Rotational Dynamics

TABLE III SOSM OBSERVER PARAMETERS FOR SPEED VIRTUAL SENSOR

Similarly, the engine rotational speed dynamics can be written as

(20) where the known rotational dynamics are When becomes

converges to sliding manifold, the dynamics of (15)

in can be chosen in a similar way as The constants in and convergence of will be ensured discussed for in similar fashion [29]. 3) Virtual Sensor for and Estimation: After the second-order sliding mode has been achieved, the term approaches to zero. When approaches to zero, also vanishes, and from (9) we are left with

(21) and the dynamics to be estimated are (22) The dynamics in and are Lebesgue-measurable in a compact region of . The detailed analysis shows that engine rotational dynamics are observable and identifiable as the respective matrices are nonsingular [28]. For the estimation of and extraction of unknown variables, a similar observer can be defined according to rotational speed dynamics as

or (16) As the convergence of is achieved, the convergence of is also guaranteed. Under convergence, i.e., and the following sliding mode dynamics from (15) will be available:

or (17) Remark 3: It may be kept in mind that and are the low-pass filtered versions of switching functions and , respectively. These first-order low-pass filters are employed to cater for switching effects of discontinuous injectors in estimation results. The time constant of low-pass filters was chosen as 0.2 s for reduced time delays in estimation results. It can be observed that the only unknown in (17) is , that can be calculated as

(23) where represents the observer states and are the observer injectors. These injectors can be calculated with the same methodology, that was followed to design and in (7) and (8). The aim of these injectors is to eliminate the error defined as and . The state is available for measurement from the crankshaft sensor installed inside the engine housing that can be accessed from OBD-II kit. The above second-order sliding mode observer inherits anti-peaking structure [29], where and reach the sliding manifold one by one in a recursive way. The error dynamics before converging to the comes out to be surface, i.e.,

(24) where

(18) Once has been estimated, it can lead to the development of virtual sensor for angular speed of engine from (16) as (19) are defined in Appendix A. where Remark 4: The quadratic nature of (17) for may lead to complex or real solution, however, restricts the solution to be real. Remark 5: Equation (17) will have two solutions for in general; however, the negative values will not be of interest, as it has no physical significance.

1) Boundedness and Convergence Analysis: The same procedure is adopted to check the boundedness of the rotational dynamics of engine to ensure the observer convergence. As the states and parameters involved in ECU controlled engine anand will always gular dynamics are bounded in nature, exist and satisfy Assumption 1. Similarly, and for angular dynamics can be calculated to satisfy Assumption 2 as discussed in [29]. As the boundedness is ensured the convergence is guaranteed by following the aforementioned procedure. The observer gains and are chosen with the same predefined procedure and are given in Table IV. As the convergence of is achieved, the

AHMED et al.: VIRTUAL SENSORS FOR AUTOMOTIVE ENGINE SENSORS FAULT DIAGNOSIS IN SECOND-ORDER SLIDING MODES

1837

TABLE IV SOSM OBSERVER PARAMETERS FOR PRESSURE VIRTUAL SENSOR

dynamics of to be

before converging to sliding manifold comes out

(25) 2) Virtual Sensor for and Estimation: After the second-order sliding mode has been achieved, we are left and . From (24), the dynamics during with second-order sliding mode can be written as (26) As the convergence of is achieved, the convergence of is also guaranteed. Under convergence of , the following sliding mode dynamics from (25) will be available: (27) The unknown engine efficiency

in (27) can be calculated as (28)

has been estimated it can lead to the development of Once virtual sensor for manifold pressure from (26) as (29)

IV. EXPERIMENTAL RESULTS The method outlined for the development of virtual sensors in the previous section has been validated for the same 1.3 L commercial vehicle engine. The vehicle engine is equipped with an ECU which is compliant to OBD-II. The data was logged through OBD-II data scanner and logging software. OBD-II allowed to monitor engine dynamics in detail. The pressure and rotational speed values were accessed through the sensors installed inside the engine. Similarly, manipulation in throttle valve was also logged in OBD-II software. Any variation in throttle position can be monitored by the sensor installed inside the vehicle. In order to estimate immeasurable parameters and validate virtual sensors, an experiment was conducted with idle gear in steady-state conditions. By steady-state conditions it was meant that: throttle valve and load on engine were kept constant through out the experiment. Under steady-state conditions, throttle value remained at 9.8 . The nominal load on engine was contributed by: engine rotating mass, e.g., crank shaft, fly wheel, power train components engaged. As no extra load

Fig. 3. SOSM observers convergence during the experiments.

is induced, load torque remained around 29 Nm. With these inputs to the engine, the pressure and rotational speed remained in steady-state conditions. The data acquired from OBD-II kit served as input to proposed estimation, virtual sensor development and fault diagnosis scheme. In the first experiment, the measurements from installed manifold pressure sensor and the designed SOSM observer in (6) were used to estimate volumetric efficiency and to develop virtual sensor for the angular speed. The second experiment involved measurements from installed crankshaft sensor and the SOSM observer in (23) to estimate combustion efficiency and to monitor manifold pressure via virtual sensor. This practice provided the facility to: • Estimate/Sense Immeasurable Parameters. • Create Redundancy for Available Sensors. • Effectively Monitor Sensors Health. The prime requirement of the proposed scheme was the convergence of observers in (6) and (23). Fig. 3 shows the 1.3 L engine and observers output. It can be seen that throughout the experiments, SOSM observers efficiently track the actual values of manifold pressure and angular speed. The estimators error remain near to zero. The robustness of the scheme can be observed when the observers are tracking the sensors reading even in the presence of variations in engine behavior. Once the observers convergence was achieved, following tasks can be done proficiently.

A. Estimation/Sensing of Immeasurable Parameter Estimation of immeasurable parameters provides useful information about system functioning. Like, volumetric efficiency unveils the pumping capacity of SI engine that is impossible to measure from any sensor. Ideally, a mass of air equal to the density of atmospheric air times the displacement volume of the cylinder should be ingested in each cycle. However, because of the short cycle time available and the flow restrictions caused by air filter, intake manifold and intake valves, less than ideal is the measurement of amount of air enters the cylinder.

1838

Fig. 4. Volumetric efficiency  and combustion efficiency  estimated by the proposed scheme. The zoomed versions of results from 25 s onwards show the variations in estimation results.

how close the actual volumetric flow rate is to the theoretical volumetric flow rate [27]. Similarly, combustion efficiency reveals the quantity of energy released by fuel burnt during combustion process. The time available to an engine for combustion process is very brief and not all the fuel molecules may find an oxygen molecule to combine. Similarly, 100% combustion will not occur if the combustion chamber temperature is not favorable for reaction. These reasons may cause incomplete combustion of fuel. Consequently, a small amount of fuel remains unburnt and exits with the exhaust flow. This results in inefficient fuel performance and air pollution. The amount of fuel burnt can be explained by . The combustion efficiency steadily decreases as the air fuel mixture becomes richer, provided the engine combustion process remains stable [27]. These immeasurable parameters were successfully estimated by the proposed scheme. Fig. 4 demonstrates the estimated and . It can be observed that the volumetric efficiency estimated from (18), remains around 70% and combustion efficiency estimated from (28), shows that around 80% of the fuel energy is being utilized. These parameters values indicate that the engine exhibits degraded performance, as ideally the volumetric efficiency should remain near 80% and combustion efficiency should remain near 95%. The degraded performance can be due to leakage in inlet and exhaust valves, piston rings, etc. B. Virtual Sensors The second outcome of the proposed methodology is the redundancy of the sensors installed in AIS. These sensors include: manifold pressure and crankshaft sensor. Equations (19) and (29) serve as virtual sensors for the measurement of angular speed and manifold pressure, respectively. Fig. 5 shows the measurements from actual and virtual sensors. It can be observed that virtual sensors efficiently track actual sensors as the measurements are with less than 10% error of the actual sensors readings. The tracking was achieved after the estimated efficiencies acquire their nominal values in less than 5 s. These minor delays in estimation results are due to reachability phase and do not influence the effectiveness of proposed fault diagnostic methodology.

IEEE SENSORS JOURNAL, VOL. 11, NO. 9, SEPTEMBER 2011

Fig. 5. Measurements by actual and virtual crankshaft and pressure sensors, serving as redundancy.

Fig. 6. Sensor fault diagnostic methodology for SI automotive engine.

C. Sensor Fault Diagnostics The third and major outcome of the proposed scheme is the fault diagnosis of the installed sensors. Fig. 6 describes the fault diagnosis methodology for pressure and angular speed sensor. The measurements from pressure sensor were used to measure angular speed of engine and vice versa. In normal circumstances, the residuals in (30) should remain within a bound (30) These bounds are defined under for pressure sensor and for crankshaft sensor (ideally, and can be zero). Under faulty condition, the residuals will exceed the defined bound and sensor fault can be detected well in time. All the sensor faults: bias, drift, scaling, freeze, and dropout, will certainly let the residuals to cross the defined bounds. The moment the residuals

AHMED et al.: VIRTUAL SENSORS FOR AUTOMOTIVE ENGINE SENSORS FAULT DIAGNOSIS IN SECOND-ORDER SLIDING MODES

1839

engine efficiencies, redundancy for the installed sensors by virtual sensors, and health monitoring scheme of the respective sensors. The three attributes have been successfully validated on 1.3 L Honda city engine. The efficient fault diagnosis of sensors involved in air intake system depicts the authenticity of the proposed scheme, as ECU was unable to detect sensor faults. APPENDIX A

Fig. 7. SI engine intake manifold pressure sensor fault diagnosis, when the ruptured diaphragm froze the pressure sensor measurements at 93 kPa after 78 s.

cross the bound, respective sensor fault will be announced on time. To diagnose AIS sensor faults, the engine was manipulated in such a way that the faults get revealed. This practice is termed as Active Diagnosis [30]. Therefore, an experiment for pressure sensor fault diagnosis was performed that replicated pressure sensor diaphragm damage. When the diaphragm gets ruptured, instead of measuring actual pressure the sensor starts sensing normal air pressure of 93 kPa. This replicates freeze sensor fault in nature. Fig. 7 demonstrates the effectiveness and sensitivity of the proposed methodology, as the fault was detected in less than 1 s. It can be seen that the virtual sensor was informing that the pressure is 30 kPa even after the fault has occurred in pressure sensor. Thus, the difference between virtual sensor and actual sensor informs that a fault has occurred in the manifold pressure sensor. Remark 6: It may be kept in mind that engine operations during the sensor fault diagnosis experiment were working properly, only the pressure sensor started malfunctioning. This is the reason why virtual pressure sensor revealed healthy engine operations. Remark 7: When the power of sensor was interrupted, ECU at once announced the fault code for pressure sensor, however, for the experiment conducted, as shown in Fig. 7, ECU did not announce any fault code, this depicts the effectiveness of the proposed algorithm. Remark 8: The same methodology can be applied for crankshaft sensor, any fault in the sensor can be successfully detected by the proposed algorithm. The proposed methodology is readily applicable for sensor fault diagnosis of OBD-II compliant engines. The scheme is computationally cheap as it updates the system status on each acquired sample therefore the scheme can be implemented online using any low cost OBD-II kit. V. CONCLUSION This paper presented a SOSM observer-based scheme for virtual sensors to monitor AIS sensors health. The main attributes of proposed methodology are: sensing of immeasurable

ACKNOWLEDGMENT The authors would like to thank the respected reviewers, the members of the Control and Signal Processing Research Group, and Mr. Yaseen for their ideas and suggestions to improve the quality of this paper. REFERENCES [1] P. E. P. Agency, “National environmental quality standard for motor vehicle exhaust and noise,” Islamabad, Pakistan, 2009. [2] E. Balaban, A. Saxena, P. Bansal, K. Goebel, and S. Curran, “Modeling, detection, and disambiguation of sensor faults for aerospace applications,” IEEE Sensors J., vol. 9, pp. 1907–1917, Dec. 2009. [3] A. Singh, T. Kamal, and S. Kumar, “Development of ann-based virtual fault detector for wheatstone bridge-oriented transducers,” IEEE Sensors J., vol. 5, pp. 1043–1049, Oct. 2005.

1840

[4] D. Capriglione, C. Liguori, and A. Pietrosanto, “Analytical redundancy for sensor fault isolation and accommodation in public transportation vehicles,” IEEE Trans. Instrum. Meas., vol. 53, pp. 993–999, Aug. 2004. [5] J. Crossman, H. Guo, Y. Murphey, and J. Cardillo, “Automotive signal fault diagnostics—Part I: Signal fault analysis, signal segmentation, feature extraction and quasi-optimal feature selection,” IEEE Trans. Veh. Technol., vol. 52, pp. 1063–1075, Jul. 2003. [6] G. Rizzoni and P. Min, “Detection of sensor failures in automotive engines,” IEEE Trans. Veh. Technol., vol. 40, pp. 487–500, May 1991. [7] P.-L. Hsu, K.-L. Lin, and L.-C. Shen, “Diagnosis of multiple sensor and actuator failures in automotive engines,” IEEE Trans. Veh. Technol., vol. 44, pp. 779–789, Nov. 1995. [8] M. Nyberg, “Model-based diagnosis of both sensor-faults and leakage in the air-intake system of an si engine,” presented at the SAE, 1999, Paper no. 1999-01-0860. [9] K. Choi, S. Singh, A. Kodali, K. Pattipati, J. Sheppard, S. Namburu, S. Chigusa, D. Prokhorov, and L. Qiao, “Novel classifier fusion approaches for fault diagnosis in automotive systems,” IEEE Trans. Instrum. Meas., vol. 58, pp. 602–611, Mar. 2009. [10] D. Capriglione, C. Liguori, C. Pianese, and A. Pietrosanto, “On-line sensor fault detection, isolation, and accommodation in automotive engines,” IEEE Trans. Instrum. Meas., vol. 52, pp. 1182–1189, Aug. 2003. [11] F. Heidtmann and D. Soffker, “Virtual sensors for diagnosis and prognosis purposes in the context of elastic mechanical structures,” IEEE Sensors J., vol. 9, pp. 1577–1588, Nov. 2009. [12] A. Levant, “Higher-order sliding modes, differentiation and outputfeedback control,” Int. J. Control, vol. 76, pp. 924–941, 2003. [13] N. M’sirdi, A. Rabhi, L. Fridman, J. Davila, and Y. Delanne, “Second order sliding mode observer for estimation of velocities, wheel sleep, radius and stiffness,” in Proc. Amer. Control Conf., 2006, pp. 3316–3321. [14] D. Caveney, “Cooperative vehicular safety applications,” IEEE Control Syst. Mag., vol. 30, pp. 38–53, Aug. 2010. [15] R. Van der Merwe and E. Wan, “The square-root unscented kalman filter for state and parameter-estimation,” in Proc. IEEE Int. Conf. Acoustics, Speech, and Signal Process. 2001 (ICASSP’01), 2001, vol. 6, pp. 3461–3464. [16] C. P. Tan and C. Edwards, “Sliding mode observers for detection and reconstruction of sensor faults,” Automatica, vol. 38, no. 10, pp. 1815–1821, 2002. [17] C. Edwards and C. P. Tan, “Sensor fault tolerant control using sliding mode observers,” Control Engineering Practice, vol. 14, no. 8, pp. 897–908, 2006. [18] R. Martínez-Guerra, R. Aguilar, and A. Poznyak, “A new robust sliding-mode observer design for monitoring in chemical reactors,” J. Dynamic Syst., Meas., Control, vol. 126, no. 3, pp. 473–478, 2004. [19] B. Jiang, M. Staroswiecki, and V. Cocquempot, “Fault estimation in nonlinear uncertain systems using robust/sliding-mode observers,” IEE Proc. Control Theory Appl. , vol. 151, pp. 29–37, 17, 2004. [20] K. B. Goh, S. K. Spurgeon, and N. B. Jones, “Fault diagnostics using sliding mode techniques,” Control Eng. Practice, vol. 10, no. 2, pp. 207–217, 2002. [21] D. Brambilla, L. Capisani, A. Ferrara, and P. Pisu, “Fault detection for robot manipulators via second-order sliding modes,” IEEE Trans. Ind. Electron., vol. 55, pp. 3954–3963, Nov. 2008. [22] B. Castillo-Toledo and J. Anzurez-Marin, “Model-based fault diagnosis using sliding mode observers to Takagi-Sugeno fuzzy mode,” in Proc. IEEE Int. Symp. Mediterranean Conf. Control and Autom., Intelligent Control, 27–29, 2005, pp. 652–657. [23] Q. Wu and M. Saif, “Robust fault diagnosis of a satellite system using a learning strategy and second order sliding mode observer,” IEEE Systems J., vol. 4, pp. 112–121, Mar. 2010. [24] L. Guzzella and C. Onder, Introduction to Modeling and Control of Internal Combustion Engine Systems. Zurich, Switzerland: Springer, 2004, ETH. [25] I. H. Kazmi and I. A. Tirmizi, “Modeling and simulation of internal combustion engines for design of model-based control methods,” M.S. thesis, Ghulam Ishaq Khan Inst. Eng. Sci. Technol., Topi, Pakistan, 2007.

IEEE SENSORS JOURNAL, VOL. 11, NO. 9, SEPTEMBER 2011

[26] J. Jongeneel, H. Nijmeijer, C. Manzie, and D. Nesic, “Input redundant internal combustion engine with linear quadratic Gaussian control and dynamic control allocation” Eindhoven Univ. Technol., Edinhoven, The Netherlands, 2009, Internal Rep., DCT doc. no. 2009.023. [27] W. W. Pulkrabek, Engineering Fundamentals of the Internal Combustion Engine. Englewood Cliffs, NJ: Prentice-Hall, 2003. [28] I. Muhammad, A. Bhatti, S. Ayubi, and Q. Khan, “Robust parameter estimation of nonlinear systems using sliding mode differentiator observer,” IEEE Trans. Ind. Electron., no. 99, p. 1, 2010, vol. PP. [29] M. Saif, W. Chen, and Q. Wu, High Order Sliding Mode Observers and Differentiators Application to Fault Diagnosis Problem, Modern Sliding Mode Control Theory. Berlin, Germany: Springer-Verlag, 2008. [30] K. Uwe and N. Lars, Automotive Control Systems: For Engine, Driveline, and Vehicle. Berlin, Germany: Springer-Verlag, 2005.

Qadeer Ahmed (S’08) received the B.S. degree in mechatronics and control engineering from the University of Engineering and Technology, Lahore, Pakistan, in 2007 and the M.S. degree in control systems from Mohammad Ali Jinnah University, Islamabad, Pakistan, in 2009. Currently, he is working towards the Ph.D. degree at Mohammad Ali Jinnah University. He is the first author or coauthor of more than ten international publications. His research interests include robust control, estimation, and condition monitoring of engineering systems.

Aamer Iqbal Bhatti (M’05–SM’09) received the B.S. degree in electrical engineering from the University of Engineering and Technology, Lahore, Pakistan, in 1993, the M.S. degree in control systems from Imperial College of Science, Technology and Medicine, London, U.K., in 1994, and the Ph.D. degree in control engineering from the University of Leicester, Leicester, U.K., in 1998. He is now with Mohammad Ali Jinnah University, Islamabad, Pakistan, as a Professor of DSP and control systems with the Department of Electronic Engineering. He is also Head of the Control and Signal Processing Research Group (CASPR). He has authored more than 70 refereed international papers, including eight journal publications. His research interests include sliding-mode applications and radar signal processing.

Muhammad Iqbal received the M.S. degree in physics from the University of the Punjab Lahore, Lahore, Pakistan, in 1991, the M.S. degree in nuclear engineering from Quaid-e-Azam University, Islamabad, Pakistan, in 1993, and the M.S. degree in computer engineering from the Center for Advanced Studies in Engineering (CASE), Islamabad in 2004. Currently, he is working towards the Ph.D. degree at the Center for Advanced Studies in Engineering, Islamabad. Since 1993, he has been working in the field of electronics and computers systems engineering at NESCOM, Islamabad. He is the first author and coauthor of more than 16 refereed international publications. His research interests are controls and DSP application emphasizing fault diagnosis of uncertain nonlinear dynamic systems.