Rotating Machinery Fault Diagnosis Based On The Rough Set

Rough set theory is a new mathematic tool for processing uncertainty information after probabilitytheory, fuzzy set theory and D-S theory of evidence. It not only gives a studying method forinformation science and cognitive science, but also offers a practical technique for intelligentinformation processing. The studying on the math theory and the actual applications of rough setis progressing worldwide.Rough set is a tool for processing indefinite, imprecise data, which deals with knowledge in termsof classification. Especially, it is very easy to integrate the theory with almost all othersoft-computing methods such as neural networks, evidential theory, genetic algorithm, fuzzy settheory and case-based reasoning. Integrated systems have the advantage over using a softcomputing simply. Under this background, intelligent decision based on rough analysis forms theforefront part of decision science.Choosing reasonable and effective attributes set is one of the important contents of the research onrough set. How to choose the optimal attributes set is NP-hard question.Fault diagnosis of machinery is a comprehensive science to find the original of fault according tothe machinery work situation information and make corresponding decision. The information fromfault original is transferred through the characteristic and the situation of the system. One of themost difficult and most key problems in fault diagnosis technique is the features extraction, whichrestrict the correction ratio of fault diagnosis and reliability of prediction of machine fault in earlystate. Features extraction is the "bottle-neck" of the fault diagnosis.The signal of interest that can reflect faults are always weaker than that can not, so they are buriedby the signal of non-interest. The signal which can not reflect faults directly must be processed bymeans of some measures in order to get some new arguments (model arguments). The newmethod(wavelet self-information envelope demodulation, WSED)is proposed in this article forextraction fault features more effective and easier.The test rigs of the rolling element bearing are built to simulate the rolling bearing faults. Thecomparison between the Hilbert and WSED are investigated to verify the WSED effective andpractical.A new time-frequency distribution is investigated in this article. This new kernel has better energygather than WVD, and easy to diagnose the faults and monitor the trend of the faults.To diagnose the unbalance and stator/rotor contact fault of rotors, a forward whirls/backwardwhirls is used as conditional attributes, then the RS-maximum entropy theory is used to getoptimal features set, BP, RBF are used afterwards. Aiming at the characteristic of rolling bearing faults, a dynamic modal of rolling bearing fault isestablished .frequency features: frequency entropy, the maxium corresponding characteristicfrequency magnitude, and frequency energy extracted through WSED and time features: averagevalues, virtue values and kurtosis values got through IMF are adopted as condition attributes, theRS-maximum entropy theory is also used to get optimal features set. The self-organization map isused to diagnose the rolling bearing faults.The faults diagnosis dictionary is built in order to provide better tools for engineers. This measureverified by test is a better way to diagnose rolling bearing faults.The training time of ANN which is shortened after using the RS-maximum entropy theory isverified by lots of test, while the result of classification is the same.