Research On Fault Diagnosis For A Class Of Nonlinear System Based On Deterministic Learning

With the rapid development of the science and technology, the requirements of reliabilityand safety about the system operation become stricter, so we pay more and more attentions onthe system fault diagnosis. The traditional fault diagnosis methods are divided into two parts:the method based on system mathematics model and the method based on signal processing.The former method need to build accurate system mathematical model, which is not suitablefor complicated nonlinear system especially; the latter method avoid the mathematic model ofthe system, but missed a lot of informations of diagnosis in the fault process. In recent years,the method of the nonlinear system fault diagnosis based on the system knowledge hasattracted much attention. Combining the deterministic learning theory, this paper studied faultdiagnosis for a class of nonlinear system based on deterministic learning, including thefollowing three aspects:First, we introduce the deterministic learning theory and rapid pattern recognitionmechanism. By choosing the radial basis function (RBF) and meeting the conditions forpersistent of excitation, we present system modeling method with local RBF neural networklearning, and then take the modeling of the information has received and similarity judgmentof nonlinear systems into consideration, so as to realize the rapid pattern recognition betweencurrent unknown dynamic pattern and dynamic pattern from the modeling of the information.Second, a fault diagnosis scheme is proposed for a class of nonlinear system. Firstly, theunknown fault dynamics are identified through deterministic learning. The knowledge on thefault dynamics is stored in a space distribution bank of invariable neural networks. Secondly,based on the bank which has been established, a series of estimators are constructed tocompare with the test monitored system and a set of residual can be generated. So, the faultcan be detected and isolated by the residual.Third, we study the fault diagnosis of rubbing rotor and basic loose rotor by usingdeterministic learning theory. Also, we first set up Jeffcott rotor system model includingrubbing rotor and basic loose rotor, and then establish the bank of fault rotor system includingdifferent fault degree and different fault types of the rotor in the form of constant RBF neuralnetwork. We realize the fast estimate of rubbing degree and recognition of different fault rotorfaults by the the bank of fault rotor system.Compared with the intelligent fault diagnosis method, the method in this paper highlightsprocessing the system modeling information and reusing in diagnosis, which set up the bankof fault system and mechanism of fault diagnosis based on rapid pattern recognition. It’s sensitivity to the different faults, quickness in the diagnosis processing and wide suitability. Atlast, simulation studies of fault rotor system are included to demonstrate the effectiveness ofthe proposed approach.