At present, the common mining method in coal mine production is the comprehensive mining method mainly based on mechanized mining. Shearer plays an important role in the mining process. Due to the difficult mining problems caused by the complex geological environment, the working efficiency and reliability of the shearer have been seriously disturbed, thus affecting the coal production efficiency, causing damage to the economic benefits of the enterprise, and bringing potential production hazards to the enterprise.
Therefore, the maintenance and fault prediction of the shearer is very critical, especially for the gear fault of the shearer cutting part that is often damaged. The enterprise needs to deploy maintenance personnel to repair such key parts to ensure the normal operation of the shearer.
For the fault analysis of shearer cutting part, scholars and enterprises at home and abroad have carried out relevant research. Abroad, the high-performance shearer produced by Eickhoff Company in Germany in 2004 has a centralized control system. At the same time, its distributed input/output function combined with circuit carrier signal can monitor real-time status information. The shearer produced by JOY Company of the United States is integrated with a long wall display, which can analyze and display the status operating parameters of the cutting department in real time, and classify the faults of the cutting department.
Foreign scholars have also made innovations in fault prediction and diagnosis theory. TETSURO M et al. put forward a method based on probabilistic logic neural network for bearing operation detection, and the detection effect of this method is very obvious. By using wavelet theory, SANDHYAS et al. intelligently filter the background noise in fault prediction, effectively enhancing the ability to collect fault information.
Although the research on fault prediction and classification of shearers started late in China, some achievements have also been made. Jia Junming used envelope demodulation analysis method, combined with the fault characteristics of gear, studied the vibration fault of shearer, and designed the corresponding vibration detection instrument. Xie Minmin and others put forward a fault feature extraction method based on wavelet packet, analyzed the gear fault of the cutting part of the shearer, designed a fault diagnosis system, and effectively diagnosed the shearer fault. In order to enhance the extraction ability of bearing fault features, Zhong Dawei adopts a fault diagnosis method based on convolutional neural network to solve the bearing fault diagnosis problem under the conditions of unbalanced samples and variable working conditions, which provides a reliable basis for reasonable arrangement of production and equipment maintenance. Bao Congwang et al. put forward a fault diagnosis method for gear reducer of shearer cutting section based on convolutional neural network. The fault recognition rate of the reducer is more than 95%, which solves the problem of tedious feature extraction in traditional methods, and provides a new idea for fault diagnosis of gear reducer of shearer cutting section. In order to quickly diagnose rolling bearing faults, Gong Wenfeng et al. proposed a new method to improve CNN. This method introduced the global mean pooling technology to replace the full connection layer of traditional CNN, and solved the problem of large parameters of traditional CNN models. Liu Hongjun et al. proposed a fault diagnosis model based on Gram angular difference field (GADF) combined with convolutional neural network, which decomposed and reconstructed the data to eliminate the noise signal, and realized fault feature extraction more quickly and accurately.
To sum up, although scholars at home and abroad have studied the fault diagnosis of shearers, there is a lack of research on the fault prediction and classification of shearer cutting gear.
Based on the analysis of the above documents and based on the method of deep learning, the prediction of gear failure in the cutting part of the shearer is studied. From the time domain vibration data of key parts, the deeper characteristics of the data are learned independently, so as to achieve fault diagnosis.
This method does not need to select features manually. In the process of model learning, features are automatically selected internally to extract classification features; In view of the over fitting problem of the model, the optimization method of adding the drop out layer is proposed in order to improve the fault prediction technology of the shearer.