Key Parameters Extraction And Outlier Correction In A Kind Of Complex Product And Its Application

Complex product often requires high assembling accuracy and contains lots of assembling parameters.So the successful rate of the assembly is low and need to repeatedly alignment to meet the accuracy of assembly.At present,manual or computer aided alignment is often used but the alignment is inefficient.To improve the alignment efficiency of complex products,this paper focuses on the selection of key assembly parameters,assembly outlier detection and amendment.The first chapter introduces the current research of digital alignment in complex product,feature selection technology and outlier data detection technology.Then the shortcomings of the above technology are analyzed,the research content of this paper is put forward and the significance of this research is analyzed.Finally,the structure of the paper is introduced.In the second chapter,the extraction method of assembly performance sensitive parameters based on statistical classification is proposed.The kernel density estimation of assembly performance data is built.After classify the assembly performance data into clusters by mean-shift algorithm,we select the important features of assembly data by the sparse learning algorithm put forward in this paper call weighted L(2,1)norm.At last,we verify this algorithm by experiment.In the third chapter,the extraction of outlier parameters of assembly process based on one-class classification is proposed.A novel algorithm called dual decision boundary one-class support vector machine is constructed.By combining Boosting algorithm,a new unsupervised feature selection framework is proposed,and the outlier parameters is selected with this framework.Finally,the UCI standard data set is used to analyze the proposed feature selection framework.In the fourth chapter,guidance of complex product alignment based on outlier correction is proposed.Firstly,exclude negative samples by the methods proposed in second and third chapter,and the assembly data are simplified with the assembly performance sensitive parameters and the outlier parameters of assembly process,and the outlier detection model is built.Then,the gradient of one-class support vector machine decision boundary is used to amend the outliers and the result of outlier amendment is optimized by kernel density estimation.Finally,the experimental verification of the outlier amendment method is carried out.The fifth chapter develops a digitized guidance system of complex product alignment,which implements data management module,key parameter selection module,outlier detection module,outlier amendment module,and finally applies the system to the assembly and alignment of a complex product.The sixth chapter summarizes the full text and analyzes the shortcomings of this study and looks forward to the future work.