Anomaly Detection Of Bolt Tightening Quality Based On Big Data Analysis In Car Production

Bolted connections are widely used in automobile manufacturing.The quality of bolted connections directly affects the performance of products.With the rapid popularization of industrial Internet of Things,tightening equipment with monitoring function is widely used in enterprises.The tightening process involves many kinds of massive data,such as torque,rotation angle,time,etc.By using big data analysis method and combining with enterprise bolt tightening process,the original data are digested,the anomaly detection model is established,the pattern recognition of potential anomalies is realized,and the accurate quality management decision is provided for enterprises.Starting from the actual production of enterprises,this paper analyses the main control methods and theories of bolt tightening,studies the bolt tightening process and stages,analyses the quality control technology of automatic tightening equipment applied in enterprises at present,and points out the defects of quality evaluation.Based on the analysis of the original data characteristics of bolt tightening process,several problems of data set are summarized.In view of the imbalanced data characteristics of bolt tightening,the SMOTE algorithm is improved,and the effectiveness of the improved SMOTE algorithm is verified by experiments.On the basis of the analysis of tightening technology,the big data operation framework is used to complete the pre-processing of original data,and further extract the direct and indirect characteristics of bolt tightening process.Aiming at the characteristics of bolt tightening business,an anomaly detection model based on CART decision tree is established to mine the feature information of bolt tightening process,diagnose tightening quality,and find the global optimal solution of tree depth and Gini impurity threshold parameters by combining grid search and cross validation,which improves the generalization ability and diagnostic accuracy of the model.The performance of the model is evaluated by using the AUC value of the area under the learning curve and ROC curve,and the model tree is interpreted visually.On this basis,based on KMeans clustering analysis of abnormal causes and abnormal pattern recognition classification of tightening process,the "elbow method" is used to determine the optimal number of clusters,and the abnormal situation of tightening process is classified and identified by synthesizing bolt tightening process.Finally,through experiments,the performance of parallel clustering algorithm in Spark platform is measured from three aspects: speedup,scaleup and sizeup.