Equipment Health Condition Recognition and Prediction Based on CNN-LSTM Deep Learning

With the advancement of intelligent manufacturing, the sensor installed on the electromechanical equipment can easily obtain a large amount of field data. In this paper, we establish a pre-warning model of equipment health recognition based on deep learning. The model firstly uses the eddy current sensor to collect the original vibration signal during the operation of the equipment and obtains the time-domain feature sequence as the CNN-LSTM input sample set; The input samples extract robust information from feature learning capabilities of the Convolutional Neural Network (CNN) for device health status recognition; At the same time, multi-layer short and long-time memory neural network (LSTM) coding feature information is introduced to reconstruct the sequence space to predict the operating status. The model is applied to the Bentley rotor test bench of normal, touch grinding, unbalanced three states to identify and predict. The experimental results show that the model has strong feature extraction ability and generalization ability to the original vibration signal, the correct rate of the training samples is 99.16% and the correct rate of the test samples is 98.61%; LSTM coded feature information for target state prediction with fast response and tracking performance. At the same time, three kinds of state recognition and warning of the spindle characteristic test stand, the spindle and the support end, and the crack of the bearing are identified and verified. The feasibility of the CNN-LSTM network is verified and has certain practical value.