A Novel Machine Learning Based Conveyor Belt Health Monitoring Incorporating UHF RFID Backscattered Power

Abstract

Ultra high frequency (UHF) radio frequency identification (RFID) based sensor is recently explored as a robust health monitoring technique for early crack detection of coal mining conveyor belts. However, a crack detection and predicting a crack pattern in terms of its width from the backscattered power of the sensors remain a challenging task. In this paper, an automatic machine learning (ML) based crack detection and crack width prediction is proposed. The representative crack features of the belt are obtained from the backscattered power of UHF RFID sensor based passive resonator. In order to enhance the crack profile, three different types of crack orientations and 6 different crack widths are investigated. A supervised multilayer perceptron neural network and a Naïve Bayes classifier are trained to classify a crack with width of 0.5 mm. Classification evaluation conducted on a data set of 180 features shows that the learned features with both the proposed ML frameworks provide superior performance in detecting a crack and also predicting its width.