Anomaly Detection Of Bolt Tightening Process For Imbalanced Data Sets

With the rapid development of automation in the field of industrial manufacturing and the rapid development of information technology,a large number of data are produced in every link of the industrial manufacturing field,which can be used to analyze the problems of the state of the equipment and the quality of the products.However,there are many factors affecting the quality of industrial data because of the industrial production environment.The problem of noise data and the unbalance of data are widespread and seriously affect the reliability of data analysis results.According to the noise characteristics in the experimental data,a preprocessing method for this kind of noise is proposed in this paper.This method can effectively eliminate noise from the random noise and the tightening process in the data,thus greatly improving the quality of the sample data.For unbalanced data problems,data level and algorithm level are generally studied.In this paper,the traditional SMOTE over sampling method is improved.Based on the DBSCAN clustering idea,the shortcomings of the traditional SMOTE method to change the spatial distribution characteristics of the sample are overcome when the unbalanced sample is oversampled.At the same time,the improved method can make data synthesis for the isolated sample with higher value,thus preserving the data.More sample features.In the aspect of sample feature extraction and classification,this paper uses the PCA dimensionality to get the lower dimension data set,and trains the sample classifier through the lifting tree algorithm,thus realizing the bolt tightening process anomaly detection model for the unbalanced data set.Finally,the validity of the test model is verified and simulated by the actual production data of bolt tightening,and the application of the model is studied in real time.