Gear Fault Diagnosis Based On Hidden Markov Model

The gear transmission has been applied widely to the mechanical device; itsstate directly determines the performance of the entire device. Therefore, monitoringand fault diagnosis of the running gear is particularly important. When themechanical device is working, motion and force is transmitted through engagementbetween the driving gear and driven gear in geared system, and the mesh vibrationmust be generated. The mechanical vibration can be changed by manufacturingassembly error, wear, cracks as well as others defaults of gear. Meshing vibrationsignal contains a wealth of gear status information; so, the analysis of meshingsignals is the most effective way to gear fault diagnosis.If the faults can be found and fault types can be isolated during gear operation,it would be possible to use the gear in reason or develop maintenance plans in orderto improve the utilization and prevent unexpected accidents. Based on the above, bytaking gear as the research object, the method and the technology based hiddenMarkov model (HMM) for gear fault diagnosis has researched by theoreticalanalysis and experiment. Several main works are as follows:1. The significance of gear fault diagnosis were analyzed, the development ofdiagnostic techniques gear fault diagnosis and fault diagnosis based on HMM werereviewed, Expounded gear failure modes, common faults vibration mechanism andcharacteristic of the gear meshing signal.2. The basic theory of HMM and especially continuous HMM is studied. Theissues of overflow and parameter initialization in HMM algorithm are discussed andsolutions are given. Finally, the basis ideas and processes of fault diagnosis based onHMM are established.3. A gear fault feature extraction method based on zoom spectral analysis wasproposed and combined with the use of discrete hidden Markov model. Firstly,Zoom Spectrum is analyzed using meshing signal extraction using time-domainsynchronous average of an interested gear from original signal. Then, theside-frequency bands of fundamental frequency and its harmonious amplitude areprocessed as a feature vector, finally, the models were trained and the features were classified after quantization. The performance of the method is checked usingexperimental data collected from gearbox.4. Continuous HMM based on AR coefficients feature extraction method wasintroduced for gear fault diagnosis and state identification. A gear full life cycle testwas conducted in the research of state identification, with the cross-validation tofind the optimal number of states and the K-means clustering algorithm is used toinitialize the state of the model, Three states of the full life cycle were identifiedsuccessfully.The recognition results showed that the proposed method of staterecognition based on CHMM using AR coefficients is feasible and effective．...