Product Quality Detection Strategy Based On Multidimensional Data Model

In traditional manufacturing,most of the qualified rates of products are detected by sampling.This method not only requires a lot of manpower,but also easily leads to batches of products being rejected,which brings huge economic losses to the manufacturing industry.In order to ensure the qualified rate of manufactured products and improve the intelligent level of manufacturing product quality detection,In this thesis,the process parameters and quality index data in the product production process are used as the input data of the model to determine the normal range of the data,and detect the product quality through the quality abnormality boundary,two methods for detecting abnormal product quality based on multi-dimensional data models are proposed.A product quality detection method based on the clustering hyper-rectangle model is proposed: firstly,perform kernel density estimation on the training data,the distribution characteristics of the data are extracted,and on this basis,the clustering parameters of the k-means clustering algorithm are designed;secondly,perform k-means clustering processing on the data,split the data into different area blocks;then perform kernel density estimation again on the different attribute characteristics of each area block,establish the density limiting threshold of different quality characteristic indicators,and determine each area statistically.The quality detection width of each dimensional feature data to forms a clustering hyper-rectangle detection boundary model;finally,the distance between the tested sample and the center of each hyper-rectangle is calculated,and it is compared with the corresponding hyper-rectangle width to detect the quality of the product.A product quality detection method based on the clustering hypersphere model is proposed: firstly,perform k-means clustering on the training data and divide it into k subsets;secondly,perform data description for each subset to determine the hypersphere center and radius;then respectively look for the mirror center of each hypersphere in the feature space,and calculate the exact original image of the mirror center in the original data domain,optimize its detection function;finally calculate the tested sample and each hypersphere sphere,and compare it with the radius of the corresponding hypersphere to detect the quality of the product.This thesis uses two-dimensional simulation data to verify and analyze the model detection boundary of the proposed method,and uses the relevant multi-dimensional data set of the UCI database to compare the accuracy and detection speed.According to the performance characteristics of the two methods,suitable methods can be selected to detect the quality of the product on different occasions.