Research On Method Of Rotating Machinery Fault Diagnosis Based On Hidden Markove Model With Missing Information

For the fault diagnosis of the kid mechanical equipment of nuclear power plant, the situation of a part of fault information is missing will easily appear when a serious emergence is occurred and the situation is partly occurred by part of sensor failure or the acquired signals appear seriously distorted. The missing of fault information will make the fault of equipment hard to be diagnosed. To solve the problem of fault diagnoses of kid mechanical equipment under the situation of a part of fault information is missing, and to offer theory and technology support to emergency decision making during serious emergency, in this thesis we will adopt the fault diagnosis method of Hidden Markov Model which has bigger advantages in the fault diagnosis of nuclear power equipment to realize the aim. But, because the normal Hidden Markov Model is hard to apply to the situation of missing information, improvement is need to the fault diagnosis process of Hidden Markov Model. To realize the aim of processing missing data and realize the fault diagnosis under missing information by using Hidden Markov Model, the thesis will improve the viterbi algorithm during state recognizing.The research will be started based on the project of nature science foundation(project number: S20120100 09505): “research on nuclear power plant fault diagnosis method under missing data occurred by serious emergency”, National Natural Science Foundation of China(project number:No51275099):“Based on two-dimension inhomogeneous Fault Diagnosis of large equipment acoustics microphone Array Technology Research” and Guangzhou yangcheng scholar chief scientist for the project(project number:12A006S):“Research on Large audio equipment fault diagnosis based on microphone Array Technology”. The rotating mechanical equipment is regarded as research object and the theory of HMM( Hidden Markov Model) and normal fault diagnosis are regarded as theoretical basis of this thesis. Then a HMM fault diagnosis method is built on these basis and the method can keep its diagnosis capability in some degree even under the situation of missing some information. At last, the rotating mechanical fault simulation platform is built on the basis of rotor experiment platform produced by Bently corporation. Then the normal failures of rotating machinery are simulated and the fault signals are gathered to test the HMM fault diagnosis method under missing information. The experiments indicate that the HMM fault diagnosis method under missing information can process effectively the problem under missing information, and the method can provide some reference to the actual diagnosis.This thesis will be started based on several parts as below:1.At the basis of researching the HMM basic theory, HMM basic algorithm, and the HMM normal fault diagnosis process, a inverse Viterbi program an algorithm are researched, improved and realized. Then a HMM closed loop training system is built to optimize the HMM training process and provide initial HMM model sets to HMMs model library selecting process. At last a auto selecting method of HMMs model library is built to increase the automatic level and efficiency of selecting HMM model, and then reduce much manual intervention.2.Improve normal fault diagnosis algorithms of Hidden Markov Model and build diagnosis method based on Hidden Markov Model under missing information. At last realizing the function of diagnosis to missing information.3.Simulating the normal failures of rotating machine and acquitting at the basis of Bently rotor experimental platform, then testing the methods provided in this thesis.