Model Identification And Monitoring Of Batch Cooking Process

Batch cooking is one of widely used methods in pulping industry. To get high quality pulp and save energy, quality control and process monitoring of batch cooking process are needed and so have enormous economic potential. To improve pulp quality is really an integrated work. Besides implementing modern management method, we need to improve control effects, detect process faults effectively to guarantee pulp end-quality up to expectation. Dynamic model of cooking process is needed before quality control and process monitoring could be implemented. The model can be used for predicting pulp quality and designing optimal control strategy. Due to variation of operation conditions, it's not easy task to acquire batch process model, which lead to difficulty in implementing quality control strategies. So modeling process through identification is always a hotspot in control and monitoring field. In practice, it is not uncommon for disturbances to evolve over several cooking batches, leading to a gradual drift of the pulp quality out of acceptable control limits. In addition, small mean shifts in the pulp quality can occur, which may not immediately translate into products outside the normal range but just an increased risk of off-spec products. Conventional Multivariate Statistical Process Monitoring methods may not be very efficient in terms of detecting these types of changes.Our aim is to develop a batch cooking monitoring framework, which can improving the ability to detect abnormal changes of batch process. This paper combines model identification of batch cooking process with process monitoring. State-space model of batch process are identified directly from process data using subspace identification method, which are developed originally for identifying continuous processes. It is proposed to capture multivariate dynamic behavior of process and quality measurements, both in time and batch-to-batch senses, from identified stochastic state-space model of the cooking process. Some research on process modelling and simulations are carried out for quality control and monitoring in batch cooking process.The main contents of this dissertation are organized as follows:1. Firstly, the batch cooking technology and reaction machanism in the cooking process are introduced. An overview on the history and latest achievements in modeling and control of batch cooking process are given. Advantages and problems of quality control and monitoring for batch cooking process are discussed and research direction of this dissertation are pointed out.2. Conventional analytical methods used in Multivariate Statistical Process Control, such as Principal Component Analysis, Multi-way Principal Component Analysis are summarized. Multi-dimension data is transformed to the analysis of statistical variables as T2 statistic, Q statistic and statistical control charts.3. Concepts of Subspace-based State-Space System Identification(4SID) are introduced. General Subspace-based State-Space model Identification method for batch process is proposed. An online predictive model for process variables is developed. An approach for batch process monitoring based on identified state-space model is presented.4. The proposed state-space monitoring approach is tested on a pulp digester. 4SID method is used to develop a state-space model of multi-dimensional batch variables correlation from nomal operating data. Simulation results on pulp quality control and process monitoring shows the effectiveness of the method.Finally, a conclusion on above work is given and future research directions are presented.