Implementation of K* Classifier for Identifying Misfire Prediction on Spark Ignition Four-Stroke Engine through Vibration Data

Misfire represents a crucial problem for vehicles, adding to the energy depletion in the midst of air pollution such as CO and NOx caused by exhaust gases. Because of a special cylinder, misfire produces a particular vibration pattern. These patterns can isolate and interpret valuable properties to detect misfires. In this paper, a machine learning approach is used as a predictive model for the identification of misfires. In the current research, vibratory signals were taken as a kind of misfire that is unique to each cylinder (acquired with the help of a piezoelectric accelerometer). Statistical characteristics are then extracted and feature selection is applied using the J48 decision tree algorithm from the features obtained. In the classification of misfires in the cylinders, the K* classification was used. The experiment was conducted in Maruti Suzuki Baleno. Every single cylinder was tested on a separate basis. The performance of the classifier was validated with 10-fold WEKA cross-validation and it was determined that K* had a maximum accuracy of 98% for the 0.24s time complexity.