The Research On A Fault Diagnosis Method Of The Rough Sets-neural Network

In this paper,the general situation of the fault diagnosis technology and the fault diagnosis system in the recent years is introduced,and the several existing fault diagnosis methods are evaluated briefly.The rationale for diagnosis is analyzed and an intelligent fault diagnosis method which integrates rough set with BP neural network is studied.The main works are listed as follows:1.The paper analyzed vibration signals of gearbox from the time domain and frequency domain,and the results fail to meet the requirement of fault diagnosis.So it accomplished the feature extraction of vibration signals according to the sub-band energy feature extraction method.2.It is necessary to discretize the continuous attributes of the original fault data according to the SOM network.The research interest of this paper concerns the attribute reduction based on discernibility matrix,and studied the improved algorithm:dealing with the single attribute element in the matrix and all others which contain the single one,greatly reduced the amount of computation and raised efficiency of training.3.Mainly analyzed the differences between the compatible and the incompatible decision-making table in calculating the discernibility matrix.Considered the possible errors in the calculation of discernibility matrix with the incompatible decision-making table and figured out the key of the problem as well as the possible improvement. Finally, analyzed the algorithm methods on the basis of the discernibility matrix attribute in the incompatible decision-making table.4.Studied the gearbox fault diagnosis,processed the data with the improvement of algorithm on the basis of the discernibility matrix attribute,and applied this improved algorithm in fault diagnosis through the BP network. The diagnosis results showed that this integrated system was very feasible in fault diagnosis fields. Through a comparison study with other methods in gear wheel diagnosis,the results pointed out that not only can this method get a perfect diagnosis,but greatly decreased the computation time and steps.