Research On Health Status Assessment And Prediction Based On Data And Knowledge Dual-driven For Body-in-white Welding Robot

The application of industrial robots in the automotive manufacturing industry is extensive,and their health status directly affects the production pace and product quality of the manufacturing process.Body-in-white welding production is the most typical manufacturing process where industrial robots are used in the automotive manufacturing process.Conducting health assessment and prediction research on the white body welding robot is a critical aspect of upgrading the white body welding production line and optimizing the prognostics and health management(PHM)system.The body-in-white welding robot system is complex,with slow degradation,time-variant,non-linear characteristics,and limited fault data,posing challenges for the health status assessment and prediction of the white body welding robot.Therefore,studying the health assessment and prediction of the body-in-white welding robot is of significant engineering and theoretical value.Based on the existing PHM system of the white body welding production line of FAW Group,this study integrates the monitoring data collected during the production process with qualitative knowledge such as expert experience.It explores a data and knowledge dualdriven approach for the health status assessment and prediction of white body welding robots and develops a system for this purpose.This provides a decision-making basis and theoretical reference for the health management of white body welding robots and the upgrading and optimization of the white body welding production line.The main research work carried out in this thesis is as follows:(1)Addressing the issues of slow degradation and limited fault data samples for the body-in-white welding robot,a health assessment model based on belief rule base(BRB)has been proposed.After a thorough analysis of the composition and fault mechanism of the body-in-white welding robot system,a fault tree for the body-in-white welding robot is constructed,providing references for the selection of features in its health status assessment and prediction process.By integrating monitoring data and qualitative knowledge,a BRBbased health status assessment model driven by both data and knowledge is built.The model parameters are optimized by the projection covariance matrix adaptive evolution strategy(PCMA-ES)algorithm.Finally,the effectiveness of the proposed model is validated by conducting health assessment on the body-in-white welding robot using the log data from a certain plant of FAW Group.(2)Addressing the efficient and accurate comprehensive prediction of the health status of the body-in-white welding robot system using multiple features representing its health status,a health status prediction method for the body-in-white welding robot system based on multi-layer belief rule base(MBRB)is proposed.Firstly,an improved random forest feature selection method is used to screen the optimal features as health feature set.Next,the fuzzy C-means algorithm is utilized to partition the reference values of premise attributes in the MBRB model,reducing reliance on expert knowledge and minimizing uncertainty,thus enhancing the robustness and adaptability of the model’s health status prediction.Subsequently,Spearman is used to analyze the correlation coefficients between features and input the highly independent feature quantities into the first layer of BRB.Finally,the multi feature health status information obtained from the first layer BRB is fused with the second layer BRB to achieve prediction of the health status of welding robots.(3)Addressing the challenges of insufficient data and the inability of a single indicator to evaluate the overall health degradation trend of the welding robot,a health degradation trend assessment method is proposed based on evidential reasoning(ER)of double layers.First,multi-source data of each joint of the robot are fused by ER＿layer1,and to evaluate the dynamic changes of each joint of the robot.Then,the evaluation values are used as the input of ER＿layer2 to obtain the health state degradation evaluation value of the whole machine,and realize the dynamic evaluation of the overall health state degradation trend of the welding robot.Finally,degradation trend assessment is conducted on the body-in-white welding robot using stage data from the body-in-white welding robot of a certain FAW Group plant.(4)We design and develop a system for health status assessment and prediction of the body-in-white welding robot,including hardware platform,software modules,data flow,and user interface.This system serves as a reference for the optimization and upgrade of the existing PHM system for the white body welding production line of FAW Group.