Research On Mechanical Fault Diagnosis Based On Adaptive Chirplet Atomic Decomposition Method

Equipment failures occurred in the production process will lead to downtimes ordamages for equipments, and result in production disruption. Thereby, condition monitoringand fault diagnosis for mechanical equipments have significant theoretical meaning andpractical merit in manufacturing industry, process manufacturing industry particularly. As weall know, extracting the fault feature from the vibration signals is the key of fault diagnosis.Under varying rotate speed, vibration signals from mechanical equipments often containmore operation and fault information about the equipments. Yet, the vibration signals whichmeasured at equal-time-interval are the multi-component non-stationary signals and have thelow signal-to-noise ratio, which makes it is very difficult to extract the fault features from thevibration signals with the existing signal processing techniques. This thesis, funded by project“Sparse Signal Decomposition Based on Multi-scale Chirplet and Its Application toMechanical Fault Diagnosis”(Project’s Serial Number:50875078) supported by NationalNatural Science Foundation of China, and project “Research on condition monitoring andfault diagnosis of large wind turbine”(Project’s Serial Number:20090161110006) supportedby The National High Technology Research and Development of China863Program,proposes an adaptive chirplet atomic decomposition (ACAD) method to overcome the issuesexisting in the present signal processing methods, and applies the ACAD method to the gearand rolling bearing fault diagnosis under varying rotate speed. The main researches and theacquired innovative achievements are as follows：(1) Aiming at the issues of low algorithm efficiency and amplitude distortion of themulti-scale chirplet spare decomposition method, we propose an adaptive chirplet atomicdecomposition method. This method can decompose in adaptive mode a complex signal intoseveral PF components, each of which is the product of an envelope signal and a purefrequency-modulated signal. Our study with numerical experiments showing that ouralgorithm boasts of many advantages, such as favorable time–frequency resolution, strongnoise immunity, good decomposition precision and acceptable algorithm efficiency. It ispreponderant to decompose multi-component non-stationary vibration signals.(2) Aiming at the problem that the modulation frequencies of fault gear vibration signalunder varying rotate speed are hard to be extracted, the time synchronous averaging (TSA)method based on ACAD is proposed. The ACAD method can effectively extract meshfrequency of gears under variable rotating speed conditions, and accurately estimate the mesh frequency curve and the rotating speed curve. The stationary requirement is satisfied for TSAmethod by resampling the signal in equal angle according to the rotating speed curve got bythe ACAD method. The SNR is improved by processing the resampled signal with TSAmethod. The modulation orders of gears can be shown in the order spectrum clearly with theFFT transform, and it reveals the types of a gear’s fault. Simulation and application examplesprove the effectiveness of the method.(3) Aiming at the problem that the modulation sidebands of vibration signals of the gearunder variable rotating speed are difficult to identify, the envelope order and cycle frequencyspectrums based on the ACAD is proposed. First, resample the signal in equal angle accordingto the rotating speed curve got by the ACAD method, translating the time domainnon-stationary signal into stationary one in angle domain, then the faults will be revealed byextracting the modulation orders from the envelopes and phases of the resampled signals byutilizing the envelope order and cycle frequency methods. Simulation and applicationexamples prove the effectiveness of the method.(4) A roller bearing fault diagnosis method based on the ACAD method and neuralnetwork is proposed. First, ACAD method is applied to decompose the envelope signalsobtained by carrying out Hilbert transform onto the original signals, to get the roller bearingfault characteristic frequency components, then time domain statistics feature parameters,such as energy, extracted from the fault characteristic components could be served as inputparameters of neural networks to identify fault patterns of roller bearing. The analysis resultsof roller bearing signals with inner race and out race faults show that the roller bearing faultdiagnosis method based on adaptive chirplet atomic decomposition and neural network canidentify roller bearing fault patterns accurately and effectively.In this thesis, a suitable method for dealing with the multi-component non-stationarysignals by atomic decomposition method based on adaptive chirplet is introduced. Based themethod, we propose the time synchronous averaging method, the envelope order and cyclefrequency spectrums, and the neural network based on the ACAD method. These methods canbe effectively applied to the fault diagnosis of gears and roller bearings under time-varyingrotational speed condition. Simulation and application examples show that ACAD method hasgood application prospect in fault diagnoses.