Fault detection method for belt conveyor idler

The existing fault detection methods for belt conveyor idler have the problems of low recognition precision, poor anti-interference capability and inability to operate stably over a long period of time. In order to solve the above problems, a fault detection method for belt conveyor idler based on time-frequency-MFCC(TFM) and multi-input one-dimensional convolutional neural network (MI-1DCNN) is proposed. Firstly, the pickup collects the audio signal of the coal conveyor idler running along the line. The dB 4 wavelet unbiased risk estimation threshold noise reduction method is used to preprocess the signal to eliminate the background noise and improve the signal-to-noise ratio. Secondly, the time domain, frequency domain and Mel frequency cepstrum coefficient (MFCC), and the first and second order difference coefficient of the noise reduction audio signal are normalized respectively, and finally assembled to obtain the feature TFM. Finally, that TFM signals are input into a MI-1DCNN model with a multi-scale convolution kernel. The feature fusion is carried out at the end of a network channel. The classification and identification of the normal idler and the fault idler are completed through a Softmax function. The TFM-MI-1DCNN model is tested with the audio signal samples of coal conveyor idler collected in a coal mine. The results show that the average recognition accuracy of the fault idler is 98.65%. The average recognition accuracy is improved by 1.50% and 1.03% compared to the improved wavelet threshold denoising-backpropagation-radial basis function network and MFCC-K-nearest neighbor algorithm-support vector machine. The false detection rate is only 0.194%. The results of field application show that the average dentification accuracy of the proposed method is 98.4%, indicating that the proposed method is suitable for field application.