Real-time combustion torque estimation and dynamic misfire fault diagnosis in gasoline engine

• Published in: Mechanical systems and signal processing Vol. 126; pp. 521 - 535
• Main Authors: Zheng, Taixiong, Zhang, Yu, Li, Yongfu, Shi, Lichen
• Format: Journal Article
• Language: English
• Published: Berlin Elsevier Ltd 01.07.2019; Elsevier BV
• Subjects: Angular speed; Artificial neural networks; Automotive engines; Back propagation networks; Combustion; Crankshafts; Engine combustion torque; Engine cylinders; Fault diagnosis; Gasoline engine; Gasoline engines; Luenberger sliding mode observer; Misfire fault diagnosis; Neural networks; Sliding mode control; Spark ignition; State observers; Support vector machines; Torque; Misfire fault diagnosis; Engine combustion torque; Luenberger sliding mode observer; Gasoline engine
• ISSN: 0888-3270; 1096-1216
• DOI: 10.1016/j.ymssp.2019.02.048

•An automated dynamic misfire fault diagnosis system is developed.•An optimized Luenberger sliding mode observer is used to estimate the engine combustion torque.•The artificial neural networks is used to diagnosis the misfire fault. In this research, an innovative state observer of gasoline engine based on the combination of Luenberger and sliding mode technique is proposed. This state observer is designed to track crankshaft angular speed and estimate engine combustion torque based on the experimental crankshaft angular speed of a four-cylinder Spark Ignition (SI) engine. Then, a new advance in the application of Artificial Neural Networks (ANNs) based on the estimated results of automated dynamic misfire fault diagnosis both under steady state and non-stationary condition is discussed in detailed. In order to effectively obtain data for network training, the estimated engine combustion torque is segmentally preprocessed according to the crank angle displacement of automobile engine. Furthermore, a series of experiments are carried out under normal and a variety of misfire conditions. The ANN systems are trained and tested using prepared cases. Finally, the Back-Propagation Neural Network (BPNN), Elman Neural Network (ENN), and Support Vector Machine (SVM) are applied to diagnose misfire fault, the effectiveness of each is evaluated respectively. Based on the estimated engine combustion torque, the experimental results show that the designed ENN is able to correctly diagnose misfire fault with a running time of 0.6 s, including single misfire, intermittent double-cylinder misfire, and continuous double-cylinder misfire in transient working condition.