Research On Fault Prediction Based On Fusion Method

The modern maintenance philosophies developed from Condition-based Maintenance are all centered on the reliability theory.Among them,Predictive Maintenance based on fault prediction technology can greatly improve the safety of equipment operation,and is an important method to reduce maintenance costs.However,with the rapid development of science,technology and manufacturing industry,more and more equipment parameters can be monitored,analyzed,and processed,the traditional single method for fault prediction shows its shortcomings for that usually the accuracy of fault prediction is not high enough to meet the needs of maintenance.In order to overcome the defects of single method for fault prediction,a fusion method based on fuzzy granulation is proposed in this dissertation.In order to verify the validity of the fusion method,the following steps have been taken in this research.First of all,as the theoretical basis for constructing a physical simulation platform,the AMESim software platform is used to simulate the internal leakage fault of the plunger pump.After that,by using the physical simulation experimental platform for plunger pump internal leakage fault,the hydraulic parameters such as output flow,pressure and temperature are monitored in real time.And the experimental data of plunger pump with different internal leakage levels are recorded to support the subsequent study of fault prediction technology.Then,on MATLAB software platform,three models for fault prediction based on support vector machine,wavelet neural network and BP neural network are individually established.After that,the three models are trained by using part of the experimental data,and the accuracy of each method is verified by the rest of experimental data.By comparing the prediction error of the three methods,it is found that the fault prediction method based on wavelet neural network is superior to the other two methods.Finally,in order to further improve the accuracy of fault prediction,the three methods are adapted based on fuzzy granulation theory,which means that the part of the experimental data for training the model should be fuzzy granulated in advance.The results show that the fault prediction method based on wavelet neural network and fuzzy granulation is the best.