Model based statistical process control for continuous processes

Statistical process control (SPC) is a set of methods for detecting abnormal behavior in a manufacturing process. SPC is generally used as a tool for the assurance of product quality in the discrete event manufacturing industries. This work considers a method for the application of SPC to continuous chemical process systems which are subject to dynamic effects not found in the discrete event case. Development of the method started with a study of existing SPC schemes for discrete and continuous processes. Based on this work a model based approach for SPC for continuous processes has been developed. This thesis describes a new method for updating the structure and parameters of stochastic models based on observed process data which can be used for SPC applications. The updating of model structure and parameters is carried out through the use of hypothesis tests. The updating scheme is designed so that only significant parameters are included in the model and that parameter updates occur only when a significant change in parameter values has occurred. The class of models considered is a class of time series models called autoregressive with exogenous inputs models. The performance of the method for detection of abnormal conditions is evaluated through computer simulations and analytical means. An example involving application of the method to an ammonia synthesis loop is presented. Extension of the method to vector valued models is discussed.