Research On Quality Monitoring Method Of Production Process With High Dimensional Data

With the rapid development of intelligent manufacturing in the world,the continuous introduction and wide application of intelligent equipment,the complexity of manufacturing process and the diversity of sensor types,the data used to monitor the operation status of production process has the characteristics of non-linear,non normal,high-dimensional and so on.Under the background of intelligent manufacturing,the quality control of high-dimensional data production process has attracted more and more scholars' attention.In the past,most of the production process monitoring research is to use the control chart method to achieve the quality monitoring of the production process.When faced with the complex production process of high-dimensional data,these data often show nonlinear characteristics,and the distribution characteristics of the data are difficult to estimate.Most of the statistics constructed by the control chart quality monitoring technology need to obey a certain probability distribution,which limits the applicability of the control chart quality monitoring method.Therefore,how to put forward effective monitoring methods according to the characteristics of high-dimensional data production process has become an important issue in the research of production process quality monitoring.On the basis of collecting a large number of domestic and foreign research literature on process monitoring,this paper systematically studies the production process quality monitoring method for high-dimensional data based on the theory of data dimensionality reduction method and machine learning method.First of all,through the analysis of the quality control research status of high-dimensional data production process,the methods of high-dimensional data dimensionality reduction,control chart quality monitoring and machine learning quality monitoring are studied.Secondly,according to the characteristics of high-dimensional data production process and the comparative study of data dimensionality reduction methods,the locally linear embedding(LLE)dimensionality reduction method and support vector data description(SVDD)method are introduced into the monitoring ofhigh-dimensional data production process.Using the locally linear embedding method,the dimension of high-dimensional data is reduced,remove redundant information and extract key information.Then,the quality monitoring model of production process based on SVDD is built by using the dimension reduced data.Finally,the simulation experiment and semiconductor production process data are used to analyze the high-dimensional data production process monitoring method.The results show that the LLE-SVDD model proposed in this paper has a good monitoring effect on the production process of high-dimensional data,and verifies the validity and applicability of the model.The research features and innovations of this paper are as follows: 1.aiming at the problem of "dimension disaster" in the production process of high-dimensional data,the locally linear embedding method is used to reduce the dimension of high-dimensional data,which can improve the efficiency of quality monitoring.2.Support vector data description method is used to identify the reduced dimension data and particle swarm optimization is used to optimize the parameters of SVDD model.3.The effectiveness of LLE-SVDD quality monitoring method for high-dimensional data quality is demonstrated.This study not only provides a set of operable high-dimensional data production process quality monitoring methods,but also provides new ideas and implementation approaches for other production process quality monitoring.