Research On Application Of Fault Diagnosis Technology For Petrochemical Rotary Machinery Bearings

Fault diagnosis is one of the key technologies to ensure the "safe,stable,long,full,and optimal" operation of petrochemical units.As petrochemical units are becoming larger and more integrated,the failure of a certain component will cause concurrent failures of petrochemical units.This paper takes the widely used and fault-prone components of petrochemical rotating machinery—bearings as the research object.In view of the serious noise interference of bearing vibration signals,which brings difficulties to bearing fault diagnosis,the goal is to improve the accuracy of bearing fault diagnosis.The research on bearing signal noise reduction,fault feature extraction and bearing fault identification has important engineering significance to ensure the stable operation of petrochemical units and prevent the occurrence of accidents in the petrochemical industry.The specific research contents are as follows:Through the time domain analysis method and the envelope spectrum analysis method to analyze the characteristics of the bearing signal,combined with the bearing failure mechanism,it is found from the time domain waveform analysis that the bearing vibration signal has the characteristics of nonlinearity and serious noise interference;from the envelope spectrum diagram,the characteristic frequencies and their multipliers corresponding to different fault types of the bearing can be analyzed.Aiming at the problem that the bearing signal has serious noise interference,an ICEEMDAN-wavelet threshold noise reduction method is proposed.ICEEMDAN is an improvement on the basis of CEEMDAN,which can overcome the pseudo-modal problem caused by the residual signal of the CEEMDAN method.Combining ICEEMDAN and wavelet threshold denoising method,the ICEEMDAN-wavelet threshold denoising model was constructed,and compared with the traditional wavelet threshold denoising method to denoise the bearing signal at the same time.The noise reduction effect is evaluated,which proves the effectiveness of the ICEEMDAN-wavelet threshold noise reduction method.Aiming at the problem of difficulty in extracting fault features of bearing signals,the denoised signals are further processed to each other dimensionless to construct a mutual dimensionless evidence body that is sensitive to bearing fault features,which not only completes the fault feature extraction of bearing signals,but also improves the The training efficiency of the model.Aiming at the difficulty of establishing a reliable bearing fault identification model,a multi-population genetic algorithm(MPGA)optimization support vector machine(SVM)was studied.Using the global parallel search optimization ability of MPGA,the penalty factor C and the kernel parameter γ that affect the SVM classification performance are optimized,and the optimal combination of C and γ is searched to construct the optimal MPGA-SVM bearing fault identification model.It has been verified that the bearing fault identification accuracy of the MPGA-SVM model is as high as 95.3%,which is 5.4% higher than that of the traditional GA-SVM.Finally,the algorithm proposed in this paper is fitted and integrated,and the humancomputer interface is designed to realize the visualization of the bearing fault diagnosis process.