In this article, a nonlinear dynamic multiway partial least squares (MPLS) based on support vector ma- chines (SVM) is developed for on-line fault detection in batch processes. The approach, referred to as SVM-based DMPLS, integrates the SVM with the MPLS model. Process data from normal historical batches are used to de- velop the MPLS model, and a series of single-input-single-output SVM networks are adopted to approximate nonlinear inner relationship between input and output variables. In addition, the application of a time-lagged win- dow technique not only makes the complementarities of unmeasured data of the monitored batch unnecessary, but also significantly reduces the computation and storage requirements in comparison with the traditional MPLS. The proposed approach is validated by a simulation study of on-line fault detection for a fed-batch penicillin production.
In this article, a nonlinear dynamic multiway partial least squares (MPLS) based on support vector machines (SVM) is developed for on-line fault detection in batch processes. The approach, referred to as SVM-based DMPLS, integrates the SVM with the MPLS model. Process data from normal historical batches are used to develop the MPLS model, and a series of single-input-single-output SVM networks are adopted to approximate nonlinear inner relationship between input and output variables. In addition, the application of a time-lagged window technique not only makes the complementarities of unmeasured data of the monitored batch unnecessary, but also significantly reduces the computation and storage requirements in comparison with the traditional MPLS. The proposed approach is validated by a simulation study of on-line fault detection for a fed-batch penicillin production.
To reduce the variations of the production process in penicillin cultivations, a rolling multivariate statis-tical approach based on multiway principle component analysis (MPCA) is developed and used for fault diagnosis of penicillin cultivations. Using the moving data windows technique, the static MPCA is extended for use in dy-namic process performance monitoring. The control chart is set up using the historical data collected from the past successful batches, thereby resulting in simplification of monitoring charts, easy tracking of the progress in each batch run, and monitoring the occurrence of the observable upsets. Data from the commercial-scale penicillin fer-mentation process are used to develop the rolling model. Using this method, faults are detected in real time and the corresponding measurements of these faults are directly made through inspection of a few simple plots (t-chart, SPE-chart, and T2-chart). Thus, the present methodology allows the process operator to actively monitor the data from several cultivations simultaneously.
