Research Of Fault Prediction Methods Based On Time Series Analysis And Intelligent Algorithm

Fault trend prediction is regarded as a key technique of fault diagnosis, for ensuring the safe operation of mechanical equipment. By analyzing the past and current running status, it can reveal state trend and characteristics in the future, provide basis for machine’s maintenance strategy. This paper introduced different classification and methods about fault prediction, studied on the prediction methods of rolling bearing based on time-series analysis, support vector machine (SVM) and genetic algorithm (GA), the main contents are as follows:It is particularly introduced about the principle and modeling process of auto-regression (AR) model and grey model. Studying on the vibration signal of rolling bearing, root-mean-square (RMS) and frequency band energy after wavelet packet decomposition were calculated as features to constitute a time-series, in order to get its state thresholds, the relative standard of rolling bearing was combined. Fault prediction based on time-series was carried out using data of the state that needs attention. The result showed, AR model got good dynamic tracking performance and could alert in advance successfully with appropriate training data; Grey model grasped main trend of the time-series, but bad dynamic tracking performance, its prediction result was not ideal and strongly influenced by training data.Learned the advantage of support vector machine (SVM) which can solve the limit sample learning problems, this paper studied on a rolling bearing fault prediction method based on support vector regression (SVR). The result showed that with appropriate selection of training data SVR could predict effectively, it has good performance in short-time fault prediction, and the model’s parameters affect its accuracy badly.Since training data affects model’s parameter and prediction result, considering the good optimization ability of genetic algorithm (GA), a new prediction method based on GA was proposed. This method was proved effective using leaner regression as a model. After applying it in the method based on SVR and Grey model, the accuracy of prediction was effectively improved, and fault early warning was implemented as well.