BATCH PROCESS IMPROVEMENT USING LATENT VARIABLE METHODS

Abstract

<p>This thesis deals with the following four topics: 1. Multivariate statistical methods are used to analyze data from an industrial batch drying process. Principal Component Analysis (PCA) and Partial least-squares (PLS) methods were able to isolate which group of variables from the initial conditions and the process variables were related to a poor-quality product. The use of a novel approach to the time warping of the trajectories for batches, and the subsequent use of the time-warping information, is presented. 2. In the procedure to monitor a new batch using the method proposed by Nomikos and MacGregor (1994), an assumption about the unknown future samples in the batch has to be taken. This work demonstrates that using the missing data (MD) option and estimating the score with an appropriate method are equivalent to the use of an adaptive-expansive multivariate time series model in the forecasting for the unknown future samples. The benefits of using the MD option are analyzed on the basis of (i) the accuracy of the forecast, (ii) the quality of the score estimates, and (iii) the detection performance during monitoring. 3. laeckle and MacGregor (1998) introduced a technique to estimate operating conditions in order for a process to yield a product with a desired set of characteristics. This thesis provides a detailed study of the application of such technique in designing the operation of a batch process. The original technique is modified to include constraints and other optimal criteria onto the desired quality and the trajectories. A parallel approach based on derivative-augmented models is proposed to avoid the analysis of the null space. 4. An extension to the work by laeckle and MacGregor (2000) in solving the product transfer problem is proposed. The early technique does not consider all the data structures involved in the problem and particularly the operating conditions from the source plant. The Joint-Y PLS model is presented as an alternative to solve this problem using all the available data.</p>