| Summary: | 碩士 === 中原大學 === 化學工程研究所 === 97 === In today’s rapidly changing market, batch processes play an important role for producing value-added and high quality products of chemical, semiconductor, and biological/biochemical industries. On-line monitoring is essential to the product quality of batch processes. In this paper, the integration of MPLS (Multi-way Partial Least Squares) models and IOHMM ( Input-Output Hidden Makov Models ), referred to as the IOHMM-MPLS model, is proposed. This method unfolds the three-dimensional batch process data, projecting the data from the original data space to each independence subspace under the MPLS structure. It reduces the multivariable dimension. Then under the simple structure, the conditional probability distribution function of each score in the subspace is built. The score probability function of the IOHMM-MPLS model with the simplified structure parameters can help accelerate the parameter convergence speed while improving the efficiency and accuracy of the probability model. The issue of the asynchronous data length in the batch monitoring is also discussed. Subsequently, with the trained distribution model, two simple monitoring charts are presented to track the progress of each batch run and monitor the occurrence of the observable upsets. The case studies of a mathematical problem and a simulated fed-batch penicillin cultivation process are used to demonstrate the power and advantages of the proposed method. In addition, the comparison of the probability model accuracy of HMM, IOHMM, and IOHMM-MPLS models is also shown to highlight the good features of the IOHMM-MPLS model.
|
|---|
