Research And Application Of Equipment Remaining Life Prediction Algorithm Based On Deep Learning

At present,traditional manufacturing enterprises are on the road to high-quality and high-efficiency development.More and more enterprises realize that effectively reducing equipment maintenance costs is critical to the long-term development of enterprises.In recent years,Predictive maintenance(PdM)technology has developed rapidly.Its goal is to use real-time monitoring data to predict the remaining useful life(RUL)and potential failures of equipment to avoid unexpected equipment downtime.Maximize the use of equipment residual value to reduce maintenance costs.Therefore,the premise and primary task of predictive maintenance is the RUL prediction of the equipment.The choice of RUL prediction method is related to the amount of data and data type available of monitored equipment.For the equipment of varying degrees of complexity,the types of data that can be monitored vary,and the applicable RUL prediction method is also different.In this thesis,the data obtained by a single sensor is classified as low-dimensional monitoring data,and the data obtained by multiple sensors is classified as multi-dimensional monitoring data,and the RUL prediction algorithm of the equipment in these two cases is studied separately.The main research content of this paper can be summarized into the following three parts:(1)For a simple equipment that supports low-dimensional monitoring data represented by bearings,a RUL prediction model based on a multi-feature fusion time convolution network(Temporal Convolutional Network,TCN)is used.The model uses a convolutional neural network(Convolutional Neural Networks,CNN)extract the abstract features in the signal and combine the selected time and frequency domain features to achieve feature expansion,and use the TCN network to predict the remaining life of the bearing.(2)For complex equipment that support high-dimensional monitoring data,a RUL prediction model based on Convolutional Long Short-Term Memory(ConvLSTM)and attention mechanism is used.The model uses the ConvLSTM network to extract the spatiotemporal features in the sensor data,and adds the attention mechanism in human vision to weight the feature factors that significantly affect the prediction results,further improving the prediction accuracy of the model(3)Taking a small discrete manufacturing demonstration production line as the object,comprehensively using the RUL prediction algorithm theory,a set of predictive maintenance system based on the "Edge-Cloud collaboration" architecture is designed,which realizes the status monitoring of the production line equipment and the RUL prediction of the key components of the equipment.