| The 135 series of internal combustion engine unit are used as the major power devices for our national signal corps. Once fault occurs, it will impair the communication between armies, resulting in potential grievous losses. With the army's increasing requirement for reliability of unmanned diesel engine unit used for generating electricity, we are urged to develop a fault diagnosis system that should be precise, practical and have multi-parameters for resolving the problems when the army uses this type of diesel engine. On the basis of developing unmanned automation monitoring system for a internal combustion engine unit, the diesel engine unit is chosen as a specific research object from the point of view of industrial application. The technologies, such as signal processing, wavelet analysis, rough sets theory and neural network, etc., are used to monitor and diagnose this unit. The methods of diesel engine fault diagnosis and character abstraction are thoroughly researched.The reason, purpose, contents and practical meanings of the subject have first discussed briefly. After analyzing several popular diagnosis methods which are often used in the process of diesel engine fault diagnosis, we point out the exist problems about fault diagnosis and possible development. The technologies, such as signal collection and processing, artificial intelligence, are also depicted in brief. Because the diesel engine is a system with many vibration sources, there are abundant messages of vibration signals. After the common fault diagnosis technologies are analyzed, the technical view is given. The vibration diagnosis method, primarily with additional methods such as wavelet analysis, is used to realize the goal of fault diagnosis for diesel engines. The necessary principles of neural network, expert system and wavelet analysis involved in this paper are described. The algorithms and procedures related are also given. In order to diagnose complex system, because there exists too many characteristic parameters, the problems, such as over-large scale of neural network, unduly long train time, and redundant rules in expert system's rule base, will result in the reduction of whole system's practical performance. Then the rough set theory that received focus attention in recent years is led into the internal-combustion engine fault diagnosis work. The application of this theory in the attributable optimization of fault diagnosis characteristic parameters is explored. After a brief description of concepts involved in the paper, such as rough set theory and the indiscernibility relation, the methods of how to preprocess the loss value of data and how to discrete the data are discussed. The algorithms of attribute reduction and value reduction in rough set theory are researched in depth, and then the ordinary reduction algorithms are given. The improved algorithms based on discernibility matrix and decision matrix are put forward. The application of rough sets theory and methods in neural network technology is studied. After combined rough sets theory with the neural network technology, the neural network recognition system based on rough set theory is advanced, in which rough sets theory is used to determine the number of neural network's input nodes. The complexity of neural network's structure is reduced. The detailed procedure of algorithm is given. On this foundation, the concept of layered mining rough set fault diagnosis network is submitted. The user can use with diagnosis of different layer according to his specific requirement.The traditional expert system is introduced and their shortcomings are pointed out. Then the necessity of building the intelligent diagnosis expert system is put forward. With the method of wavelet decomposition and attribute reduction, the fault diagnosis sub-system of diesel engine's valve clearance is set up. During the period of fault diagnosis research, it is found that although the characteristic wave band can be known directly after vibration signal is characterize...
|
PDF Full Text Request