Bearing Fault Diagnosis Based On Improved Stochastic Resonance Method

Rolling bearings are critical components in rotating machinery,its early damage detection is crucial to ensure the steadiness of rotating machinery.Thus,developing accurate models that permit identifying the presence of fault signals with low-signal-to-noise ratio(SNR)at its initial stage is necessary.Stochastic Resonance(SR)method changed the traditional bearing fault diagnosis,which by adding proper amount of noise to the signal the target signal can be extracted.Traditional SR methods have some deficiencies such as:limited filtering performance,low frequency input signal and previous knowledge of the signal characteristic frequency(FC).The objective of this thesis is to improve the SR method and to further apply in bearing fault diagnosis,which includes the following contents.1.In traditional SR methods,the SR phenomenon governed by the first order differential equation cannot provide good filtering performance.Therefore,we use a second order system based on Duffing oscillator(DO)-Gauss potential to extract the weak signal that comes from a previous filtered signal,which Bayes-Segmentation combined with Peak Energy index,are used to identify the optimum resonance bandwidth(BW).Moreover,to deal with the small parameter limitation,we guide the parameters search by using the tuning method.For this first approach,that combines BS-PE for the BW detection,the results indicated that the optimum resonance BWs are found for all the analysis cases.Moreover,in comparison with the traditional SR model the proposed SR method based DO-Gauss potential provides better target signal extraction under low SNR.2.We introduce the improved SR(ISR)method,a novel method based in the SR polynomial index(SRPI),which uses four indexes combined into a polynomial,to identify the periodic impact fault signal.The proposed ISR method mainly overcame the FC requirement;it also provides an accurate target signal extraction under low SNR.3.An improved adaptive SR(IASR)method is proposed.In this method,the sliding window function applied in the frequency spectrum to identify the FC of the signal,and then the maximum weighted power spectral kurtosis(WPSK)index is obtained by analyzing the output signals of the underdamped-second order stochastic resonance(USSSR)method,then the mode value obtained from the collection of FC values of each sliding window corresponds to the FC of the target signal.The IASR as the ISR method also overcame the FC prior knowledge.Additionally,once the FC is known,the sliding window mechanic finds the BW that provides the signal with the best quality;thus,compared with the USSSR output signal the IASR output signal has better quality.Both simulated signals,and two sets of real bearing vibration signals which includes:inner race,rolling element and outer race faults,verify the effectiveness of the proposed SR methods in realizing the fault diagnosis of rolling bearings based on improved stochastic resonance method;The results show:1)the proposed DO-Gauss SR method outperforms the traditional SR method in the weak signal extraction;2)the ISR and IASR methods overcame the traditional SR requirement,the FC,and therefore the bearing fault diagnosis is achieved;and 3)the IASR method improved the WPSK index,consequently outperformed the ASR method,thus,the well target signal extraction is achieved;moreover,by using the sliding window function,it provides better USSSR output signal.