Research On Health Status Assessment Of Freight Vehicles Based On Heterogeneous Integration And Improved DS Evidence Fusion

Under the background of the overall transformation of traditional manufacturing industry towards intelligent manufacturing,the maintenance system of heavy-haul railway freight vehicles has gradually changed from the traditional "planned maintenance" mode to a more intelligent "condition maintenance" mode.With the continuous upgrading and development of on-line monitoring technology for heavy-haul railway freight vehicles,it has become one of the current key development directions to guide more scientific and efficient intelligent assessment of vehicle health status through massive vehicle monitoring data.Traditional equipment health assessment technology faces problems such as huge volume of state data and insufficient data processing capacity.Therefore,how to scientifically and effectively use massive vehicle state data to quickly and accurately judge the health status of heavy-duty freight vehicles has become an important research topic in the intelligent transformation of railway transportation industry.On the basis of existing equipment health assessment methods,combining the advantages of machine learning method and deep learning method in dealing with massive complex data and improving the ability of evidence theory to resolve evidence conflicts,this thesis proposes a health assessment method for heavy-duty freight vehicles based on heterogeneous ensemble learning and improved DS evidence theory.The main research work includes the following contents:(1)Aiming at the problems of huge volume and unbalanced data distribution of heavy-duty freight car state data,a method of heavy-duty freight car state assessment based on heterogeneous ensemble learning(Ham Ster)is proposed.The method combines the excellent ability of XGBoost and Light GBM models to process data with unbalanced data distribution and the self-learning ability of sparse automatic encoder(SAE)to multi-dimensional features,and carries out heterogeneous integration through soft voting method to realize automatic extraction of deep-seated features from massive data and intelligent evaluation of vehicle health status.The experimental results show that compared with SVM,SAE,XGBoost,Light GBM and other common health assessment methods,this method can obviously improve the accuracy of health assessment results of heavy-duty freight vehicles.(2)Aiming at the problem that Ham Ster’s health assessment method can’t solve the serious conflict between the evaluation results of the basic model,the improved evidence theory is used to solve the evidence conflict,and a health assessment method of heavy-duty freight vehicles based on the improved DS theory(DS-Ham Ster)is proposed.The method firstly provides corresponding evidence based on XGBoost model,Light GBM model and SAE model in Ham Ster,then combines the uncertainty of evidence and the conflict measurement factors of evidence based on Nash equilibrium theory from the perspective of modified evidence body,and further puts forward an improved method of heavy-duty truck vehicle health status assessment combined with DS synthesis rules.Finally,this thesis designs several groups of ablation experiments,which verify the effectiveness of this method in dealing with conflict problems,and further improve the accuracy of model health assessment.