Research On Fault Status Evaluation Algorithm Based On Machine Learning

The connection and coupling among the components of modern rotatingmachinery are really tight. Once one of the components breaks down, the wholemachinery equipment, even the entire production line would be significantlyimpacted, which will bring incalculable economic loss, and more seriouslyunnecessary casualties. One main component of the rotating machinery is the rollingbearing, whose fault diagnosis and prediction technology involves multiple factorsthat are constrained and connected. Therefore the method to divide the workingstatus of the rolling bearing into “normal” and “fault” is not enough, and the way toevaluate the fault status becomes an urgent problem to be solved.After reading and analysis amounts of fault diagnosis methods of machinelearning, this paper proposes an improved algorithm of BP neural network based onthe filter idea of Kalman. Since the traditional BP neural network is easy to sink intolocal optimum, adding certain number of data that are related to the output layervalues and the expected values to the input layer will achieve the improvement of thestandard BP neural network; Because many samples can not be definitely dividedinto certain categories by SVM (Support Vector Machine). This paper puts forwardsthe soft output method supporting SVM, in other words, the sign function of theoutput layer node is removed. So the samples can be divided into certain categoriesclassify according to their probability. In addition, this paper proposes the concept ofhealth degree based on fuzzy set theory. Fault state evaluation algorithm is obtainedon the basis of membership function fitted these two machine learning methods (theimproved BP neural network and SVM) and health degree.In view of the characteristics of the rolling bearing, this article extracts thefeature parameters through measure the sensitivity based on the simulationexperiment of the MATLAB platform, using the machine learning method (BPneural network or SVM) to calculate the membership degree of the rolling bearing toeach fuzzy set, and then the HD (Health Degree) of the bearing is available by the mapping between the membership degree and the health degree, through which thehealth grade of the bearing is obtained, and the evaluation results of the bearingstatus are given in the paper. It is beneficial to discuss the status and developmenttrend of the rolling bearing and the intervention.