Design of Fault Detection System for Automobile Power Train Using Digital Signal Processing and Soft Computing Techniques

Abstract

The increasing dependence of internal combustion engine in multitudes of application has mandated a detailed study on most of its subsystems. This paper focuses on predictive maintenance using machine learning based models. The transmission system of any power pace is often challenged due to sudden variation in applied load. Any fault in the transmission system could lead to the catastrophic failures hence need for this work. This paper deals with the identification of various fault conditions that happen in a transmission system using vibration signals acquired by an accelerometer. The acquired signals are processed to extract the statistical and spectral features. These features are used to build a machine learning model using decision tree or Random forest algorithm. The best combination of features and algorithm is evaluated and the results are presented.

Iterative learning based estimation of periodically occurring faults

“…In [23], a robust state estimation as well as FD, isolation, and sensor FE observer for descriptor‐linear parameter varying (D‐LPV) systems with unmeasurable gain scheduling functions have been proposed; for estimating the faults, the states of the D‐LPV system have been augmented by adding the fault as auxiliary state variables; finally, the proposed method has been verified by a realistic model of an anaerobic bioreactor. Unlike the above mentioned model‐based approaches, in recent decades, the analytical and soft computing approaches [24] have been applied to the research in FD and FE [25, 26]. The neural network (NN), as a typical soft computing technology, has been widely applied to detecting and estimating faults in systems, see, for example [27, 28] and the references therein.…”