Diesel Engine Fault Diagnosis Based On Rough Sets And Neural Network

As a typical reciprocating machinery and equipment, engine plays an important role in the mechanical system. However, the complexity of its structure makes the diesel engine fault showing the diversity and complexity, In addition due to the impact of many aspects, like the work of environmental noise, the accuracy of information collection system, data processing and the methods of fault diagnosis, the accuracy and efficiency of the diesel engine fault diagnosis is low. Therefore, how to find a rapid and effective fault diagnosis method is the main research which domestic and foreign scholars to study.Rough set theory can remove redundant information in the data to achieve optimal testing points and characteristics of the parameters. Artificial neural network has better fault tolerance, is the wider application of intelligent fault diagnosis method. Information fusion technology can combine multiple sensor information to improve the reliability and credibility of the system. In this article, combine the rough set theory, artificial neural network and information fusion technology together, Using their advantages to research the fault diagnosis of the diesel engine to achieve better diagnostic results.First, this paper describes the status of research project at home and abroad, and the feasibility and necessity of the rough set theory and artificial neural networks which combined together. Second, focus on the optimization problem based on rough set theory, analyzed and summarized the present problems that the optimization method of measuring point, attribute discretization and reduction algorithm, there are some problems in this program, the improved method related to is proposed. Depending on the measuring point attribute, the optimization method can effectively distinguish the different points of the property dependence, and various measuring points on the sensitivity of different fault conditions. The improved NS discretization algorithm can achieve the classification of information systems to ensure that capacity does not change under the premise of minimizing the number of breakpoints, to improve noise immunity. The reduction algorithm which Improved and based on attribute dependence, make the attribute dependency and indiscernibility relation as the criteria for judging for classification capability of the system, examples are used to demonstrate the validity and effectiveness of the method, basing on the basic conditions of reduction. Finally, the vibration and noise signal after reduction are integrated, by comparing the training process and output the results of RBF neural network show that the fault diagnosis system can improve the diagnostic accuracy and efficiency.