Research On Prediction Method Of Mechanical Equipment Status Based On LSTM

The prediction technology of mechanical state can combine the actual operation of mechanical equipment,predict the development trend of equipment operation and discover the deterioration behavior of equipment early,It also can provide reasonable basis for the maintenance of equipment.As the state’ parameters of the device show an exponential explosion,the traditional prediction methods often have some problems such as inaccurate prediction accuracy and low efficiency in the face of large amount of mechanical state data,it’s also difficult to find degradation and long-term prediction.In order to solve above problems,this paper uses the Long-Short Term Memory Neural Network(LSTM)to construct an autoregressive prediction model.After comparing this method with the traditional prediction method through rolling bearing experiment,the method of autoregressive prediction is verified its advantage in accuracy.However,the LSTM-based autoregressive prediction method has poor accuracy and performance in the long-period prediction of mechanical equipment state.A multi-regressive trend prediction method based on stacked GRU is proposed for this problem.The observation model and the middle layer are used.And the system model is constructed to predict the multi-regressive long-period trend prediction model.The method is validated by the tool wear state experiment.After comparing with the traditional method and the auto-regressive trend prediction method,the effect of the proposed method on the long-term prediction of the mechanical state is verified.The research carried out is as follows:1)The autoregressive prediction method based on LSTM is studied.The autoregressive prediction model is built with LSTM as the core component.After the feature extraction,optimization,model training and model testing of mechanical equipment state parameters,the relevant life cycle data of rolling bearings are related.Experimental comparison shows that the proposed LSTM-based mechanical state autoregressive prediction model not only improves the accuracy of prediction,but also greatly reduces the time and forecast cost of online calculation,thus improving the accuracy and efficiency of mechanical equipment state prediction.2)The multi-regression prediction method based on stacked GRU neural network is studied.The proposed model includes the observation model,system model and the middle layer.These three parts interact to realize the long-term memory of the mechanical equipment time series signal.The proposed model method is validated by tool wear experiment,and the advantages of the proposed model in predicting the running state of mechanical equipment are proved.3)An intelligent prediction software system based on LSTM was developed to carry out signal monitoring functions,feature extraction and optimization,state prediction and maintenance recommendations for rolling bearing equipment.The mechanical equipment status prediction module in the software system,as the core module of the system,mainly trains the collected characterization signals capable of characterizing the running raw data of the mechanical equipment,and mines the implicit relationship in the signal characteristics to realize the mechanical mechanism.The future operation status of the equipment is predicted accordingly.