Wavelet Decomposition Driven Machine Learning for Mechanical Vibration Dependent Fault Detection in Automotive Air Circular System

The air-circular system of a vehicle depends on the proper circulation of cold air inside the car. This system should be robust enough to circulate the cold air to maintain the desired temperature inside the vehicle. As the vehicle is expected to run in different road conditions, the air circulation system faces challenge of vibration due to uneven road conditions. This rough road condition often causes a short circuit in supply system feeding power to the dc motor used to circulate air. This short circuit normally has a very short duration and creates transient short-circuit currents; hence, it becomes difficult to detect for its short duration. Here, an attempt has been made to detect these small duration vibration-dependent short-circuit faults that may occur in power circuit feeding power to motor used for air circulation. This has been achieved by monitoring statistical behavior of current. Current signals have been analyzed, at first, by wavelet-based statistical tools and then by various machine learning (ML)-based approaches at normal and fault conditions. The proposed methods have been validated in simulation as well as in real case studies at low and high levels of noise and disturbances and found capable of identifying and classifying faults. Skewness and kurtosis of wavelet coefficients at certain decomposition levels were found effective for detecting this fault. Also, k-nearest neighbor (KNN)-based ML approach shows highest level of accuracy at minimum training time. Thus, this effective approach can help to detect a practical problem of mechanical vibration-dependent short circuits.

I. Introduction

Faults in air-conditioning systems experienced in a vehicle system may broadly be divided into two categories: mechanical fault and electrical fault. Mechanical faults in the air-conditioning system are quite common in automotive approximate coefficients (AC) systems. Faults in the vapor compression cycle, evaporator, filter dryer, hoses, and thermal expansion valve are the sources of failures in car air-conditioning systems [1]. Different thermos-mechanical approaches are found to deal with the performance of vehicle AC systems. Sparse autoencoder-based fault diagnosis for air-conditioning systems extracts fault features to optimize performance [2]. Fault analysis of the air-conditioning system based on machine learning (ML) algorithms has been discussed in [3]. During normal operating conditions, fault detection and diagnosis of the air conditioning have been introduced in [4]. Condition monitoring of the air-conditioning system was done by monitoring refrigerant pressure and outlet vent temperature [5]. The air-conditioning system in an electric vehicle (EV) with a battery pack has critically discussed all solutions of the AC system and covered the vapor compression cycle for EVs in [6]. The performance of the AC system was improved by condition monitoring of the equipment over the operation time and alerting the user for fault occurrence by fault detection and diagnostic (FDD) system [7]. An attempt was made to detect multiple faults in the AC system by computerized diagnostic tools [8]. The fault modeling of the heating, ventilation, and air-conditioning (HVAC) systems was introduced using the current state-of-the-art technique for fault occurrence probability in the simulation platform has been introduced in [9].