Study On Applications Of Process Monitoring Based On Multivariate Statistical Methods Of Factor Analysis

Security and stability of the production process and the product quality guarantee have always been the two main objectives which enterprises attach great importance on especially under the furious circumstance. Process monitoring and fault diagnosis technology are keys to ensure the production safety and improve product quality. However, the traditional process monitoring method depends too much on the mathematical model and experience. With the development of modern tracking control technology and the application of distributed control system, lots of process data can be stored and used. How to make use of information to enhance monitoring capability has become the hotspots of industrial and academia.Data-driven multivariate statistical process control (MSPC) method doesn't require a complex mechanism model, and it can extract important information from process data only by statistical methods and quantify it to the monitoring index to complete the process monitoring. MSPC takes the Gaussian distribution and time sequence independence on the promise, however, actual production process often doesn't meet these constraints limits.This paper mainly discusses the application of Factor Analysis (FA) in process monitoring and proposes modified algorithm and solution to solve the monitoring difficulties that the variables are non-Gaussian and the process is dynamic. The major work and contributions are as follows:1. Firstly, The theoretical background of FA and the differences between PCA,PPCA and FA are described basically; secondly, the model parameters calculation method based on EM algorithm (expectation-maximization) is put forward and the probability model is established; lastly, the process monitoring method based on FA will be introduced which mainly includes the following steps: data standardization, factor number selection and the monitoring index determination. The simulation results verify the efficiency and superiority of the method.2. As for dynamic process monitoring, the dynamic nature of industrial processes are verified. In order to overcome the shortcomings of the expansion method based on auto-regression, two modified dynamic factor analysis methods correlation analysis based on autocorrelation analysis are proposed. Therefore the static FA method is promoted to dynamic multivariate process. Simulation results of TE process demonstrate that this method can improve the performance of process monitoring.3. For Non-Gaussian process monitoring, the independent factor analysis(IFA) method is introduced into the process monitoring to overcome the shortcomings of ICA and the method based on Gaussian mixture model, because ICA model is noiseless model and the GMM method lets the PCA and PPCA as the basis. IFA combines the advantages of FA, ICA and other methods and application in chemical adsorption and separation shows its effectiveness.4. A dynamic process monitoring method based on independent factor analysis is proposed to solve the problem that industrial process is non-Gaussian and dynamic. Superiority is proved by the comparison with the single dynamic monitoring and non-Gaussian monitoring method.