Rough Set Model Of Hybrid Decision System And Applicaction In Turntable Fault Diagnosis

Machine learning and knowledge discovery is one of the most important research directions in artificial intelligence, and the fuzzy, incompleteness and uncertainty of information under complex environment are the key problems in knowledge discovery. Rough set theory, which simulates the capability of granulation and approximation in human cognition, is an effective mathematical tool to characterize inconsistency in classification data. This theory has been applied in knowledge discovery from symbolic data. However, most of data sets in real-world applications are character, numerical, fuzzy, incomplete or their mixture. Not much work has been devoted on such issues.In this paper, the hybrid decision system including the mixture of character, numerical, fuzzy and incomplete attributes is studied, and proposed the general model of mixed data classification and knowledge discovery based on granulation and approximate of rough set theory. Then studied several attributes reduction algorithm based on bionics. The rough set theory is application to the fault diagnosis of turn table. The faults of turn table are analysed and fault features are extracted by wavelet transform, and finally the rough set-neural network fault diagnosis system is designed. The main contributions of the work are listed as follows.Pawlak rough set is based on strict equivalence relation and the treatment objects are character attributes. But in practice, many information systems are hybrid systems which including numerical attributes, fuzzy attributes and often accompanying the lost of attributes. So the practice application of rough set is difficult. In order to solve the above problems, the characteristics of the hybrid decision system, fuzzy decision system and incomplete decision system are analysed. The rough set models for general incomplete hybrid decision system and general incomplete fuzzy hybrid decision system are developed, and studied the identification method for incomplete information. The influence on classification by neighborhood operators is studied also. The general incomplete hybrid rough set model and general incomplete fuzzy hybrid rough set model are the expanded of Pawlak rough set model, which can solve the practical application of mixed attributes including fuzzy attributes and incomplete attributes effectively.The attributes reduct algorithms are discussed for incomplete hybrid decision system. Fist, the classical reduct algorithms are analysed. Then, genetic algorithm, clonal selection algorithm and niche POS algorithm are used in the attributes reduction of incomplete hybrid decision system, and the realization method is given respectively. The simulation results indicate that the proposed algorithm can overcome the NP-hard problem of finding all reductions of hybrid decision system.The fault feature extraction methods using the wavelet are studied based on the analysis of the faults of turn table. The faults of turn table are classified as follows, soft faults, angular measuring system faults, communication faults, interface board faults, actuator faults and mechanical faults. The causes of the faults are analysed, and the wavelet theory is introduced to detect the signal singularity and remove the high frequency nosie of the signals. For instance, the wavelet theory is used to extract features of discontinuities, amplitude overproof, distortion overproof and saturation of angular measuring system of turn table.The advantages and disadvantages of rough sets and neural networks are analysed in the last part of the paper, and rough set-neural networks fault diagnosis system is designed for turn table. Fist, the faults decision table is found based on rough set theory. Then the faults decision table is reduced by rough set reduct algorithm. The use of rough set theory can delete redundant attributes effectively and simplify the training sample to shorten the training time of networks. Neural network has good fault-tolerance and scalability, as the post system, can enhance fault-tolerance and anti-jamming capability of the fault diagnosis system. The classifier and recognizer of rough set-neural networks can identify and locate the turn table faults effectively. Finally, the hardware design and software implementation method of turn table rough set-neural networks are proposed, experimental results demonstrate the validity and feasibility.In conclusion, this paper studied the rough set knowledge discovery method in hybrid, incomplete and fuzzy information system, and the practice application in fault diagnosis. The application scopes of rough set are expended, and the fault diagnosis and health management system of turn table are found finally.