MISFIRE DETECTION IN A MULTI-CYLINDER DIESEL ENGINE: A MACHINE LEARNING APPROACH

Abstract

Misfire is another type of abnormal combustion. When engine misfires, cylinder (or cylinders) is not producing its normal amount of power. Engine misfire also has negative effects on engine exhaust emissions such as HC, CO, and NOx. Engine misfire should be detected and eliminated. Normal combustion and misfire in the cylinder (if any) generates vibrations in the engine block. The vibration characters due to misfire are unique for a particular cylinder. This can be diagnosed by processing the vibration signals acquired from the engine cylinder block using a piezoelectric accelerometer. The obtained signals were decoded using statistical parameters, like, Kurtosis, standard deviation, mean, median, etc. Misfire identification algorithms such as AdaBoost, LogitBoost, MultiClass Classifier, and J48 were used as tools for feature selection and classification. The signals were trained and tested by the selected classifiers. The classification accuracy of selected classifiers were compared and presented in this paper. MultiClass Classifier was found to be performing better with selected statistical features compared to other classifiers.

Keywords

• Engine misfire
• Feature extraction
• Confusion matrix
• AdaBoost
• LogitBoost
• MultiClass Classifier