Research On Industrial Equipment Fault Prediction Based On Multi-sensor Information Fusion

Along with the development of industrialization and technology,the structure of industrial equipment tends to be precise,moreover,the coupling between components increases.When industrial equipment in operating status,a failure occurs in a certain part or components may break the entire production chain,it could cause huge economic loss even personal safety problems.The industrial equipment failure prediction can identify failure in advance,therefore,fault prediction of industrial equipment has practical engineering significance.But traditional fault prediction model requires a large amount of data,and only using single sensor's information to establish failure prediction model which cannot use all the valid information during the operation of industrial equipment.Therefore,a multi-sensor layered information fusion based on grey prediction model,neural network algorithm,and evidence theory is proposed to realize fault prediction of industrial equipment,relevant work is as follows:First,for some issues in in the gray prediction model,such as the detection of data sequence for modeling is not smooth,construction method of background value for gray prediction model is too simple,and the traditional prediction model cannot meet the needs of the dynamic development of the system,by studying the relationship of the model's prediction accuracy,the type of original sequence and function transformation,the smoothing of oscillating sensor monitoring sequences with an inverse sine function is proposed;An integral method is proposed to instead traditional construction method of background value;The metabolism is combined with the traditional prediction model to meet the dynamic development needs of the system.The final numerical experiments respectively verify the feasibility of the improved method.Second,taking the gray model prediction sequence as the input training set of the BP neural network fusion algorithm,the actual feature value as expected input,after getting the BP neural network with expected error,taking the predicted value of the grey prediction model as input for the trained fusion,it will get the local fusion prediction results of industrial equipment based on feature layers of single sensor multi-feature prediction.Finally,the measurement data of one sensor in the gear box was verified.Third,in view of the fact that the accuracy of fusion results using single-sensor's multi-feature predictive values is not high,the single sensors cannot provide comprehensive information for the target system operating status,and the operational status of unmonitored location on the target system is uncertain,the multi-sensor information fusion technology is introduced,a decision-making layer's multi-sensor local conclusion fusion fault prediction method based on D-S evidence theory is proposed,that is,using D-S evidence theory to fuse single-sensor local conclusions of the feature layer.This can improve the stability of the recognition.The gearbox fault test verifies the effectiveness of the fusion prediction method.