The Research Of Quality Related Statistical Process Monitoring And Fault Diagnosis Method

It is extremely important in the modern process industry for process monitoring and fault diagnosis because of the increasing complexity of the operation with the advent of advanced automation, control and optimization systems and their frequent replacement. Now most of the monitoring technologys in the field application are focus on monitor the abnormal changes of the process variables due to the quality output data are difficult to obtain or acquire extremely expensive. But when checked a fault occurs in the process variables that does not expose the fault whether or not related to the output quality data. The purpose of the production is the final indicator for product quality. It expects that by monitoring the status of the entire process variables to ensure quality product. Therefore, it has more practical significance to take the measured variables that related to output quality into account and by monitoring the system process variables that affect the the final quality product.A novel quality-related method based on global and local partial least squares projection (QGLPLS) is proposed in this paper to obtain the correlation between the quality variables and process variables. The main idea of QGLPLS is to combine the advantages of locality preserving projections (LPP) and partial least squares (PLS). It can extract meaningful low-dimensional structure information to represent high-dimensional process variables and quality data and is able to utilize the potential geometry structure both contain a global and local information to interpret the sample data of process variables and quality variables. It is known that statistical process monitoring model of PLS to find the global structural variety information based on the direction for maximizing the covariance. However, it can not well extract the local adjacent structural feature of sample data. The manifold learning method of LPP can make up for this shortcoming. LPP uses a linear approximation technology to achieve the purpose of nonlinear mapping, while maintaining the local adjacent manifold. But LPP does not consider the implicit information in the quality product. Then, QGLPLS-based method builds a unified optimization framework, that is, maximize the global covariance and minimize local adjacent structure in the process variables and quality variables spaces. QGLPLS can consider the manifold structure of process variables and quality variables at the same time and extract correlation between them effectively. Finally, QGLPLS establishes two statistical models of T2 and SPE to implement online process monitoring.However, it does not provide a relatively better visualization graphic for online process monitoring and fault diagnosis in the above-mentioned study.Therefore, a novel quality related visualization monitoring method based on self-organization map (SOM) is proposed in this paper for process monitoring and fault diagnosis. Firstly, it uses QGLPLS method to predict the quality variables which are difficult to measure. Then, all the stored process measurement data and the predicted quality variable data are composed of input matrix to SOM network so that the high-dimensional data matrix projected to a two-dimensional topological space. Finally, it utilizes the strong clustering ability and visualization capability of SOM for fault diagnosis, and a dynamic trajectory gragh is used to implement online process monitoring that assist the operator to monitor the site conditions better.