Application Of HHT And Fuzzy Neural Network To The Fault Diagnosis Of Rolling Bearings

Rolling bearing is an important part in mechanical system. Because of runningfrequently, rolling bearing’s potential fault will appear. When fault occur, rolling bearingalways accompany with the vibration noise. In order to keep mechanical system safe andreliable operation, it is very important to diagnose the rolling bearing fault. According to thevibration of rolling bearing operation, such as the vibration amplitude, vibration frequencyand so on, to identify the fault types of rolling bearing. In order to research the rolling bearing,the6205-2RS JEM deep groove ball bearing is chosen, it is produced by SKF company inSweden. Diagnosing three types of rolling bearing fault, such as inner fault, outer fault andballs fault.Because rolling bearing runs frequently, it always accompany with the vibration noise. Iffailures occur, the noise will be more seriously. In order to diagnose rolling bearing faultaccurately, firstly, the fault signals should be denoised. In order to achieve denoised signals,the new digital signal processing technology empirical mode decomposition (EMD) combinedwith wavelet packet is used to denoise signals, EMD can process nonlinear and nonstationarysignals well. Then the denoised signal can be got and they are decomposed by wavelet packet.At last, the energy eigenvectors are got, they are imported to adaptive network based fuzzyinference system (ANFIS) and diagnose the rolling bearing fault types. The results show thatthe wavelet packet denoising based on EMD can improve the signal to noise ratio (SNR)effectively. After signals are preprocessed, the result of ANFIS analysis shows that averageerror is low. ANFIS can diagnose the three fault types effectively. Using this method issuitable for rolling bearing fault diagnosis.Because the major drawback of EMD is mode mixing problem, it always gets uselessresult when complex fault signals are decomposed by EMD. In order to overcome thisshortcoming and useful intrinsic mode function (IMF) can be obtained, a method calledensemble empirical mode decomposition (EEMD) and distributing fitting testing for rollingbearing fault diagnosis is proposed. Because of stochastic vibration process, EEMD canalleviate the mode mixing problem perfectly. For achieve useful IMFs and abandoninsignificant IMFs, after using EEMD method, distributing fitting testing is applied to chooseuseful IMFs and abandon noise IMFs. Using two tests method, one is normal probability plot,the other is Jarque-Bera test. They can improve the correctness of the test, then the denoisedsignals can be got and the envelope spectrums are also obtained. Envelope spectrums candiagnose different failure frequency and different types of fault. The results show that EEMDand distributing fitting testing can choose useful IMFs effectively, then envelope spectrums can diagnose the denoised signals effectively. The analyzed results demonstrate that theproposed method is able to identify rolling bearing multi-fault diagnosis effectively.