Investigation On Multivariate Statistical Process Monitoring

Statistical process control (SPC) is one of the most powerful tools for process monitoring and product quality improvement. When applied to prompt detecting, diagnosing and appropriately coping with process malfunctions, it will be of considerable benefit. Statistical process monitoring (SPM) is based on multivariate statistical projection methodologies for online process malfunctions detection and diagnosis through analysis and interpretation of the collected measurements. Unfortunately, these conventional methods behave unsatisfactorily when applied to industrial processes. In this dissertation, this problem is addressed accordingly and the related achievements mainly include:1 the similarity metric of any two principal component analysis (PCA) models adopting the concept of principal angles, which is further extended to the case of more than two PCA models. A multiple PCA model based process monitoring methodology and its online applications are then presented, which include the algorithms for developing multiple PCA models, criterion for online models selection and methods for models updating and incorporation of new PCA models.2 the similarity metric of any two partial least squares (PLS) models, which is further incorporated into the framework of process monitoring using multiple PLS models. Issues including online application of the developed PLS models, updating and adding new models are discussed. This approach can be applied to product quality estimation and malfunction detection of processes with multiple operating modes.3 an equivalent presentation of PLS where PLS algorithm is achieved through two sequential but relatively independent steps including projection and reconstruction. A novel nonlinear PLS algorithm is then proposed where both nonlinear latent structures and nonlinear reconstruction are obtained straightforwardly through two consecutive steps. It is also compared with existing nonlinear PLS algorithms with respect to the consistency with PLS, the orthogonality of the latent structures and the computational requirements and accuracies.4, a general-purpose multivariate SPM software platform, where the multiple PCA based methodology is successfully applied to the monitoring of some industrial process including atmospheric distillation unit, fluidized analytic and cracking unit anda large-scale ethylene pyrolysis furnace.