Studies on machine learning models for misfire detection and vehicle condition monitoring: a low cost approach

Abstract

The rapid growth of transportation systems mostly using internal combustion (IC) engines has led to a wide range of environmental challenges, demanding immediate attention. Misfire in spark ignition IC engine is a major factor leading to undetected emissions and performance reduction. The engine diagnostic system of the vehicle should be designed to monitor misfire continuously because even with a small number of misfiring cycles, engine performance degrades, hydrocarbon emissions increase, and drivability will suffer. There are various misfire detection techniques that are practiced, each having a unique set of merits and demerits. The use of crank angle encoders is one of the most widely reported approaches for misfire detection. The main objective is to develop a low cost alternative to the existing techniques. The current work uses the vibration signature of the engine block for developing a comprehensive vehicle condition monitoring with predominant focus on misfire. The possibility of using the same sensor data for monitoring fuel consumption impacting parameters like air filter choking, gear knock, high engine speed and low tyre pressure is envisaged. The work was carried out in two distinct phases. In the first phase: model design, development and analysis using an engine test bed were done followed by model extension analysis (capability to accommodate additional conditions using the existing signal features itself). Phase I resulted in the formulation of a model capable of identifying misfire accurately on a IC engine test rig. In the second phase, the developed misfire detection model was implemented on a Suzuki passenger car operated on real road conditions. The model was then fine tuned for performance enhancement and extended toachieve the secondary objective of vehicle condition monitoring. A diverse range of features including statistical features, histogram features, discrete Wavelet transforms (Harr and Debauchees), discrete Fourier Transform