Research On Fault Prediction Algorithm In Edge Computing

With the rapid development of manufacturing industry,the problem of failure maintenance of system equipment comes with it.Equipment failure prediction and health management have become areas of extensive research by many scholars.Among them,data-based research methods have been rapidly developed with the development of the Internet of Things and sensor technology.The data on the edge of the Industrial Internet of Things is growing in a blowout.The previous way of uploading all data to the cloud and processing by the cloud center gradually exposed the problems such as high cost and large delay.The edge computing method came into being.The advantage of fault prediction at the edge is to quickly analyze and process data while reducing the problems of insufficient bandwidth,excessive delay,and high cost during the transmission of large amounts of data.The combination of edge computing and artificial intelligence further strengthens the application scenario of edge fault prediction.The fault prediction is mainly composed of data collection,data processing,condition detection,health prediction,warning assessment and suggestions.This paper focuses on the intelligent algorithm for the equipment failure prediction stage at the edge.The main work and research results are as follows:1)This paper proposes an improved extreme learning machine algorithm for edge fault prediction according to the needs of edge fault prediction.Model compression is used to reduce the complexity and calculation of the algorithm.The redundant nodes in the hidden layer of the model are clipped,and the model is updated after the real-time monitoring data is added.The accuracy of the prediction is improved under the premise of limited model size.Finally,the validity of the model is verified by the data of gear working conditions and it is suitable for fault prediction at the edge.2)Due to the large amount and variety of data collected at the edge,the fault status of the device can be characterized by multiple types of data,and the realtime monitoring of the edge often lacks full life cycle data.Therefore,a K-means clustering method is proposed to divide the degradation phases represented by various data into failure stages,and to complete real-time failure warning under the condition of limited prior data,so as to provide a reliable basis for system equipment maintenance strategy formulation.Finally,the multi-type data of the gear test bed is used to verify the feasibility of the method proposed in this paper,and it is proved that the algorithm can accurately divide the fault stage and predict the occurrence of the fault.