Research On Key Technologies Of Health Management For Complex Equipment Core Components Based On MTS

Over the last few years,with the development of precision instruments and high-end manufacturing,complex equipment has played an increasingly important role in weapons,aviation,aerospace,production,engineering,and other fields.Its safety and reliability have become the focus of attention.The equipment is powerful,complex,and constructed with a diverse set of core components of varying levels and structures.In actual operation,the core components have poor reliability under the influence of various factors and easily become“short slab of cask”.Once a failure happens,it will cause immeasurable losses.If users can be provided with timely and reliable health status assessment,fault diagnosis,and conditionbased maintenance strategy before failure,it can help them eliminate the fault in the bud,reduce maintenance costs,and improve work efficiency.Therefore,service-oriented complex equipment prognostics and health management(PHM)will become a critical technical problem to be solved in the transformation and upgrading process of complex equipment manufacturers.In this paper,the core component of rotating equipment,the rolling bearing,is taken as the research object.Starting from the monitored non-stationary vibration signal,this research is targeted at improving the accuracy of health status assessment and multi-failure mode diagnosis.After thoroughly studying the feature extraction and feature selection methods,research on the improved Mahalanobis-Taguchi system(MTS)and its application in the PHM of core components are further studied.The main works are as follows:(1)Investigation of vibration signal multi-domain feature extraction and feature selection methods.To accurately describe the operation status of the core component and identify failure modes,the multi-domain feature extraction and construction scheme of time domain,frequency domain,Hilbert envelope spectrum,and energy entropy of vibration signals is studied.Considering the influence of noise in the signal on the assessment and diagnosis,a sensitive intrinsic mode function(IMF)energy entropy feature extraction method combining correlation and Kullback-Leibler divergence is proposed based on the complete empirical mode decomposition of adaptive noise(CEEMDAN).Also,a feature reduction and fusion method based on MTS is employed to reduce the dimensionality and filter out the features relevant to the core component’s failure in each domain.Data experiments of rolling bearing vibration signals verify that the MTS can accurately identify the important features while maintaining the features’ classification ability.(2)Investigation of two-stage multi-domain feature selection algorithms based on feature sorting and MTS.To effectively extract relevant features for core component health assessment and fault diagnosis,two-stage feature selection methods based on feature sorting algorithms and MTS are investigated.The primary goal of the first stage is to reduce the redundancy among nonlinear features.Combining the advantages of the Laplace score(LS)and Relief F algorithm in feature selection performance,the ranking of feature importance is obtained by the two algorithms,respectively,and a feature clustering algorithm based on mutual information is adopted to obtain the subset with the lowest redundancy and lowest correlation.In the second stage,MTS is used to select important features to effectively avoid the influence of redundancy on relevant screening.The rolling bearing data experiments verify that the two-stage feature selection algorithms are effective in reducing redundancy,identifying the initial fault,and recognizing the fault location and degree,which can improve the efficiency of the follow-up evaluation and diagnosis.(3)Investigation of health status assessment methods based on improved MTS.To realize service-oriented health management and implement active maintenance of the core component,a health index model and threshold determination method based on Mahalanobis distance(MD)are proposed.From the shortcomings of the MTS,robust Mahalanobis space construction and improved WMD method are described.Function xe and cumulative sum function are utilized to transfer the WMD to the health index,which is established as a quantitative indicator of the core component’s health status.Following that,a health index threshold is provided to identify the initial failure using the control chart and Johnson transformation.The performance of the improved MTS based health status assessment method is compared through experimental research on rolling bearing life cycle data,The results showed that this method is sensitive to early failures and can accurately identify different degradation stages to provide a basis for health management.(4)Investigation of the multi-fault pattern recognition methods using multi-class MTS.To address the shortcomings of the MTS in multi-class classification and improve the accuracy of fault diagnosis,two improvement methods are proposed: constructing a multidimensional binary classifier and building a hierarchical architecture classifier.On the one hand,the multi-domain features are classified into multiple dimensions to build multidimension MTS(MD-MTS),and the classification threshold space and classification criteria are used to determine the sample category.On the other hand,the multi-class classification problem is transformed into multiple binary classification problems.The directed acyclic graph MTS(DAG-MTS)method is constructed around class separability,and the threshold of MTS in each node is used for layer-by-layer classification.The experimental data of rolling bearings confirmed that both the two multi-class MTS methods are superior to the traditional MTS method and binary tree MTS(BT-MTS)in terms of fault location and fault degree,effectively expanding the MTS research and multi-fault diagnosis method.