Recognition of Pick Wear State Based on Deep LSTM Neural Network

Picks are the vital components used in mining machinery for cutting coal-rock mass, and their wear state significantly influence cutting efficiency during the cutting process. Currently, manual experience serves as the primary basis for replacing worn-out picks in most coal mines. However, to improve efficiency, picks are often replaced collectively, resulting in premature replacement of a considerable number of picks that have not reached their expected lifespan, which leads to substantial economic wastage. In this study, real pick replacement data from H coal mine in China is leveraged that classify the wear state of picks into four categories: new picks, slight wear, moderate wear, and severe wear. Due to the purpose of identify picks wear state, a method based on wavelet packet decomposition of the vibration signals generated by cutting drums when cutting coal-rock is proposed. The method utilizes feature vectors composed of the standard deviation of decomposed wavelet packet coefficients and establishes a identification model based on deep Long Short-Term Memory (LSTM) neural networks using the MATLAB platform. The analysis reveals that the deep LSTM network outperforms the single-layer LSTM neural network in capturing the nonlinear variation characteristics of the data. Moreover, the deep learning model achieves more accurate recognition results compared to the basic Back-Propagation (BP) network. This method presents a novel approach to realize intelligent recognition of the pick wear state and to implement an optimal pick replacement strategy.