Abstract
Abstract
The tire pressure monitoring systems (TPMS) are dedicated vehicle systems that calculate the tire pressure under various conditions. Proper maintenance of tire pressure can have a significant impact on enhancing vehicle handling, vehicle performance, occupant safety, comfort and fuel efficiency. Recent years have seen a shift in the preference for nitrogen-filled tires over air filled due to the superior thermal stability and uniform pressure management characteristics displayed by nitrogen-filled tires. The present article details the application of machine learning in TPMS to provide more insights into the tire behaviour for four different tire conditions (puncture, idle, high and normal). This paper specifically focuses on the collaborative approach that combines various features extracted with tree-based classifiers. Vertical wheel hub vibrations were captured using an accelerometer from which distinct features like autoregressive moving average (ARMA), statistical and histogram were extracted. With the application of J48, the most significant and contributing features were identified for every feature set extracted that was fed into tree-based classifiers. The best-performing classifier for every feature set was determined to be 95.83% (statistical–random forest), 93.75% (histogram-optimized forest) and 93.75% (ARMA–random forest). Furthermore, an extensive analysis was carried out to determine the impact of the feature fusion approach on feature combinations like statistical-histogram, histogram-ARMA, statistical-ARMA and statistical-histogram-ARMA. The experimental results indicate a commendable classification accuracy of 97.92% for a feature fusion of statistical-histogram-ARMA features with a forest penalizing attributes algorithm.