Intelligence Fault Diagnosis For Equipment With Incomplete And Inconsistent Information

The complexity of equipment structure brings on the diversity between symptom and cause of equipment fault, and fault information turn incomplete and inconsistent which is a huge puzzle in intelligent fault diagnosis of equipment at present. It is important to research fault diagnosis approaches of the equipment with incomplete and inconsistent information. Based on the rough set theory, this dissertation does deep theoretical researchers on attribute reduction, rules acquisition and rules discovery of incomplete and inconsistent decision-making information system. A new method and means is provided for fault diagnosis of the complex equipment by establishing the intelligent diagnosis model.The major innovations of this dissertation are as follows:The uncertainty of equipment fault, which important representation is incomplete and inconsistent, relates to probability of fault. The uncertainty measurement method by self-information of the event and the information entropy of the system are put forward respectively. The heuristic attribute reduction algorithm based on maximum discernibility degree is put forward, and then the optimal attribute reduction of the multiattribute decision-making information system is realized.Aiming at incomplete information system, the completer algorithm based on maximal confidence is studied, the biggest supporting degree of rules which are obtained from incomplete system is ensured. Two algorithms based on rules acquisition in the incomplete decision-making information system are proposed. One is the incremental rules acquisition approach avoiding repeated calculations from the beginning; the other is the hierarchical reduction approach. The first algorithms can not only decrease effectively the computing times of rules acquisition, but also save a mass of computing resources. The second algorithm proposed, enlightened on the recognition laws of human being, divide the whole system into complete attribute level and incomplete attribute level. Firstly, complete attribute level can be reduced using the heuristic attribute reduction algorithm. Secondly, the incomplete attribute can be reduced according to its significance one by one. Lastly, the rules can be extracted rightly. The hierarchical reduction approach can not only decrease the uncertainty because of completeness process, but also not lead the information losing of decision system because of stratifying process.Through further researches on inconsistent decision-making information system, the depending relationship between the conditional attributes equivalence class and the decision attributes equivalence class is described using inclusion degree. According to the diversity of inclusion degree, the algorithms of distribution reduction, maximum distribution reduction and assignment reduction are researched respectively. The rule reliability by distribute reducing is tantamount with original attribute set. The difficulty to discover hidden rule is overcome and the reliability of hidden rule is given via distributing reduces to inconsistent information system. The hidden rules discovery can relax the conflict between great data size and necessitous knowledge. At the same time, the optimum choice of incomplete fault diagnosis system is offered using the maximum distribution reduction method.Based on the decomposing strategy of equipment's structure, the characteristic parameter space and the fault space split multi-subspace can reduced the dimension of input data of reasoning model. The intelligent diagnosis reasoning model is studied using information fuse and the ANN&FR parallel-series hybrid reasoning model by way of changing single-ANN for multi-ANN parallel integration is put up forward also. The learning speed can be improved effectively and the error concussion can be reduced and even be eliminated by adopting improved BP learning algorithm to learn each subnet. The diagnosis weight matrix can be self-adjusted in the process of fuzzy reasoning so as to increase the accuracy of fault diagnosis. At last, the parallel subnet composition principle and the structure optimization means of BP NN are discussed.Based on analyzing the system requirements of remote intelligent fault diagnosis of loader, the general structure, functional and calculating model of fault diagnosis system of loader is proposed and realized. The practical results show this system can adapt the request of different consumers, structure reasonably, work steadily and practicability well.