Nonlocal Means Algorithm And Its Application Research On Fault Diagnosis For Rolling Bearing

Nowadays, rolling bearing is widely used in rotating machinery. Owing to the poorand complicated working condition bad, the bearing fault occurs frequently. Once faultyhappens certain functions of the system will be decreased or lost, more serious, it mayleads to disastrous accident. The operation status of rolling bearing impacts theperformance of the whole machinery. Thus, it’s of great significance to study the faultdiagnosis technology of the rolling bearing.Processing and analysis of vibration signals for fault diagnosis is based on a directimpact on the vibration signal noise reduction, noise reduction is good or bad to theeffect of the fault diagnosis. In this article the nonolcal means (NLM) denoisingalgorithm was studied and improved and combines it with the conventional time-frequency domain transform methods, some new bearing fault diagnosis method wasproposed.In this thesis, the characteristic of the vibration signal of the rolling bearing wasanalyzed proceed from the fault mechanism of the rolling bearing. We conducted thesimulation of the defect bearing vibration signals with inner defect and outer defectrespectively.In this thesis an innovative patch-based method was proposed which was callednonlocal means (NLM) based on the abundant signal redundant information and the richsimilar patches. And we introduced the fast NLM methods approach significantlyspeeds NLM by reordering operations to eliminate a nested loop. Meanwhile a newnoise variance estimation method was proposed according to the characteristics ofGauss noise and the bearing fault signals, which was consistent with the patch idea ofNLM.In order to achieve the low frequency extraction of nonlocal means denoisingsignals combined with the common demodulation analysis methods, and combined withthe spectrum analysis method to realize the time-frequency conversion, extraction offault information accurately, a new fault diagnosis of bearing method was proposedwhich was based on NLM algorithm and demodulation analysis.In this thesis a new method for rolling bearing signals processing based on NLMand Local mean decomposition (LMD) was proposed in order to solve the problem thatthe NLM algorithm combined demodulation analysis method may obtain an accurate result when signals have a complex modulation patterns. LMD is an effectivetime-frequency for decomposing a multi-component AM-FM signal into a number ofmono-component AM—FM signals. Combining with the teager energy operatordemodulation, this method can achieve fault diagnosis accurately.At the end, the necessary optimization was implemented, so a new double scaleadaptive NLM method was proposed based on renewed weights in order to improve theeffect of filter.Theoretical analysis and experimental results verification show that the rollingbearing fault diagnosis methods based on NLM can extract the fault characteristicinformation of the rolling bearing effectively and provide a new method to the rollingbearing fault diagnosis.