Abstract: In order to solve the problems of delayed status perception, slow fault response and others of equipment in the 5303 working face of Chuangyuan Coal Mine, a set of predictive maintenance technical system deeply integrated with the special working conditions of coal mines is constructed. By deploying edge intelligent nodes with intrinsic safety certification and integrating multi-source sensors, a multi-parameter coupling acquisition of hydraulic - mechanical - electrical is achieved; A multi-model fusion architecture for the degradation characteristics of fully mechanized mining equipment is innovatively proposed, combined with the advantages of random forest, support vector machine, and LSTM network, a model with high adaptive fault identification and remaining life prediction is constructed. The on-site verification shows that the system is better than the traditional scheme in terms of accuracy rate in abnormal early warning, response speed for maintenance, and unplanned shutdown control, etc.