Fault detection method for belt conveyor roller

In view of problems that existing fault detection methods for belt conveyor roller use contact measurement, are not easy to install and operate, and are not suitable for underground large-scale fault detection, a roller fault detection method based on wavelet denoising and BP-RBF neural network is proposed. Audio signal during the operation of the roller is collected and denoised by a compromise method which combines soft threshold method and hard threshold method; energy sum of the wavelet decomposition signal of each layer is used as feature value of the layer, and low-frequency feature values are converted by processing coefficients to reduce their proportion in the total energy and make the fault feature more obvious;the extracted feature vectors are input into BP-RBF neural network model for fault detection. The test results show that fault recognition rate of the method reaches 96.7% for three cases of normal roller signal, crack failure on the roller surface, and wear failure on the roller surface. Compared with the traditional spectrum analysis and diagnosis technology, the proposed method requires less workload and has higher accuracy; compared with fault detection technologies based on temperature detection and other technologies, the proposed method uses a non-contact installation method, which is more convenient to install and has larger detection range and good application prospect.