Batch Process Monitoring Method Based On Higher Order Partial Least Squares

In recent years, with the continuous changes in social market demands, batch process, who are suitable for the production of small volume, variety and high-value-added products, have been more and more widely used in industrial production. Due to the complexity of batch process, small disturbances during operation may lead to reduce the quality of the final product, or even cause production failure. At the same time, with the development of computer technology and the distributed control system, large amounts of data of the industrial processes have been preserved. Consequently, data-driven methods of batch process modeling and monitoring have become the foucs of industrial sector.Different from the continuous process, batch process data is three-dimensional in nature, which makes the data processing and monitoring of batch process more complex than a conventional continuous process. How to handle three-dimensional data of batch process to achieve more precise process monitoring has become an emerging research hotspot. Currently, most batch process morniting methods are based on unfold method, which is to unfold the three-dimensional data to a two-dimensional matrix, after that, modeling, detecting and diagnosing of the batch process are based on this matrix. However, these kinds of methods will inevitably lead to information loss since the data structure was destroyed by unfolding procedure. In this thesis, the existing modeling methods are compared, and a more accurate method for batch process modeling is proposed, based on the characteristics of batch process data, a new batch process monitoring method is developed. The main work of this paper includes the following aspects:(1) This paper makes an intensive study of the existing batch process modeling methods, then, compares the advantages and disadvantages of these methods, after that, summarizes the application scopes of different methods.(2) In order to compensate the shortcomings of existing batch process modeling methods, a new generalized linear regression model-Higher order partial least squares (HOPLS) is introduced to handle three-dimensional data of batch process. This method can overcome the information loss, computational complexity and poor fitness problems of traditional modeling methods. Two traditional modeling methods are compared with the proposed modeling method, the fed-batch penicillin fermentation process and Dupont process are applied to illustrate the efficient of the proposed method. (3) The HO-SPE and HO-T2statistics are developed based on HOPLS for batch process monitoring to improve the reliability, accuracy of batch process. The efficiency of the proposed method is compared with MPLS based on the fed-batch penicillin fermentation process and Dupont process.Finally, some conclusions and future research directions are discussed.