The Design And Implementation Of Tank Fault Diagnosis System Based On Decision Tree

The fault diagnosis technology diagnoses the nature and cause of fault, and forecasts the fault tendency, then determines necessary measure by taking advantage of modern monitoring, detecting and computer analysis and other methods. It may discover the fault of equipment promptly, and then avoid the unnecessary loss by using the diagnosis technology, so it is of high economic value and significance to research the diagnosis technology.This paper mainly describes the traditional general structure of the fault diagnosis system, and lucubrates about the knowledge acquisition, the express of diagnosis rules and the mechanism of diagnosis inference. Meanwhile it describes the decision tree classification algorithm in detail, analyzes the ID3, C4.5 and other prevalent decision tree algorithm.Based on the analysis of key technology of the fault diagnosis system and the requirements of some army, this paper designs a tank fault diagnosis system for the army. This diagnosis system finds the fault position according to the fault phenomenon and the characteristic information. This diagnosis system applied the decision tree classification algorithm to the fault diagnosis. Using Decision Tree algorithm find the relationship between the fault phenomenon and the faults. The machine learning module is completed by the Decision Tree. The Decision Tree is saved in Knowledge Database directly. That can record the rules and the same time it keeps the relationship of the rules. This inference machine is based on decision tree, because of rules saving characteristic. Meanwhile the fault diagnosis system is a generic model that can be applied to other mechanism fault diagnosis systems.Comparing with Decision Tree algorithms, this system chooses the C4.5 to realize the self-learning module. Based on C4.5 analysis and research, this paper gives the method of continuous attributes dispersed, that merge interval based on information entropy. It uses the width divided discrete method to classify attribute, and merge intervals by calculating the entropy values before and after merging. The algorithm avoids the problems of original algorithm. Experiments show that the decision tree algorithm is more effective.Finally, this system is implemented on the C++ Builder platform and SQL Server 2000 Database according to the requirements of customer.