Research On Fault Diagnosis Algorithm Based On Unsupervised Learning

Currently,in energy,transportation,metallurgy,chemical,aviation and other industries.mechanical equipment has been widely used and plays an increasingly important role in modern industry and production.The mechanical equipment structure is developing rapidly in the direction of heavy-duty,large-scale,high-speed,precise,and complicated.The function of mechanical equipment is becoming more and more perfect,and the structure is becoming more and more complex,which makes it more difficult to monitor the health states of the mechanical equipment.Once the mechanical equipment fails,it will cause serious consequences.Reliable monitoring of mechanical equipment and the development of effective fault diagnosis technology can not only ensure the safety of mechanical equipment operation,but also obtain obvious economic benefits.Thanks to the continuous progress of modern science and technology,the field of mechanical equipment fault diagnosis has absorbed more and more advanced technologies,and many excellent research results have been obtained.In recent years,with the rapid development of big data,data-driven fault diagnosis methods have received more and more attention and research.The data-driven fault diagnosis methods can process and analyze the relevant operating data of the mechanical equipment without the need of the precise system model and expert knowledge,so as to realize the fault diagnosis of the mechanical equipment.At present,the data signals reflecting the states of machinery equipment obtained in many industrial application scenarios are non-stationary,non-linear,noisy,and signal modulation may also occur.Many data signals cannot be accurately labeled,and large amounts of unlabeled data have been accumulated.Analyzing large amounts of unlabeled data is a laborious and timeconsuming task for fault diagnosis personnel.Unsupervised learning algorithms can directly use unlabeled data for analysis and research,and have strong applicability.In recent years,they have become a new research hotspot in the field of fault diagnosis.In this thesis,based on unsupervised learning,the research on the fault recognition methods of mechanical equipment is carried out,and a reasonable solution is provided for making full use of the unlabeled vibration data of mechanical equipment for fault diagnosis.The main research contents of this thesis are as follows:(1)In this thesis,stacked sparse autoencoders(SSAE)is selected to unsupervisedly extract the highly abstract features of the original data.Aiming at the problem that the SSAE network cannot be fine-tuned in the traditional supervised way when facing untagged data,through indepth analysis of the SSAE model structure and algorithm principle,the traditional supervised fine-tuning method is improved,and a two-stage unsupervised fine-tuning strategy is proposed to fine-tune the SSAE.which realizes the completely unsupervised feature extraction of SSAE.The feature extraction ability of SSAE with unsupervised fine-tuning strategy is verified through experiments on a public dataset.Compared with several other common unsupervised feature extraction algorithms,SSAE has the best feature extraction effect.(2)Aiming at the problems of traditional fuzzy clustering algorithms,such as the need to set the cluster number before clustering,the sensitivity to the selection of the initial cluster center,and the inaccurate clustering results caused by outlier sample points,through in-depth understanding of the causes of these problems,an adaptive weighted fuzzy clustering algorithm(AWGG)is proposed.AWGG algorithm selects the appropriate initial clustering centers according to the data density of the samples.In the process of clustering training.AWGG algorithm obtains the weighted clustering centers by weighting the samples.AWGG algorithm introduces the PBMF clustering evaluation index,which can adaptively obtain the optimal number of clusters and clustering results without presetting the number of clusters.The AWGG algorithm effectively solves the shortcomings of the traditional fuzzy clustering algorithms,and can stably and automatically obtain the optimal clustering result without setting the number of clusters.The performance of the proposed AWGG algorithm is verified through experiments,and the experimental results prove the effectiveness of the proposed AWGG algorithm.Compared with other common fuzzy clustering algorithms,the AWGG algorithm has obvious advantages.(3)Aiming at the scene where there are a large number of unlabeled vibration data in reality,a completely unsupervised fault recognition framework—deep adaptive fuzzy clustering algorithm(DAFC)is proposed.DAFC uses SSAE to extract highly abstract features layer by layer from the original data set.The extracted features can effectively represent the basic characteristics of the sample and can reduce the impact of noise.The highly abstract features extracted by SSAE are used as the input of the AWGG algorithm.The AWGG clustering algorithm performs unsupervised adaptive fuzzy clustering,and automatically obtains the optimal clustering result.DAFC organically combines SSAE and AWGG through the unsupervised fine-tuning strategy of SSAE to form a completely unsupervised fault recognition framework,which is the first time to realize completely unsupervised fault recognition in the field of fault diagnosis.DAFC algorithm can directly use unlabeled vibration data for clustering analysis and automatically obtain the optimal clustering results,which greatly saves the time cost of fault personnel.Experimental results on two datasets prove the effectiveness of DAFC in unsupervised fault recognition.