Tire pressure monitoring system (TPMS) is a very essential technological advancement in the automotive industry that works using pressure sensors, radio transmitters, and onboard batteries. While there exist two types of tires that are pneumatic and non-pneumatic, the current study focuses on the nitrogen filled pneumatic variant tires that are found superior in terms of shock absorption, durability, and noise reduction. In general, tires are prone to high level of fault occurrence due to their prolonged usage, dynamic road conditions, changing climatic scenarios and variable operation modes. Fault intrusion in tires can impact the safety, reliability, and comfort of the vehicle and occupants. To delimit such situations, the present study proposed a fault diagnosis strategy that involves voting algorithm, a machine learning based technique wherein the tire fault conditions can be assessed. In the present study, vibration analysis played a pivotal role in capturing vibration signals corresponding to various tire conditions. Statistical features were then extracted from the vibration data collected by the accelerometer. Subsequently, the J48 decision tree was employed to identify the most significant and influential features capable of enhancing classification accuracy. Initially, several standalone base classifiers, including sequential minimal optimization (SMO), instance-based k – nearest neighbor (IBK), logistic model tree (LMT), random forest (RF), multilayer perceptron (MLP), J48, and Naïve Bayes (NB) were employed to evaluate the classification performance of each classifier. Subsequently, the classifiers were ranked in the descending order of classification performance and the top five classifiers were considered for formulating the ensemble classifier models to implement voting. The study adopted two, three, four, and five classifier ensembles along with five different voting principles. The results obtained detailed that IBK algorithm produced the maximum accuracy of 89.36%. However, on application of voting technique, the ensemble of two classifiers with IBK and RF produced an improved classification accuracy of 91.66%.
Mathew et al. (Fri,) studied this question.