J. Jewaratnam

Batch-to-batch iterative learning control using linearised models with adaptive model updating

This paper presents batch-to-batch iterative learning control (ILC) of a fed-batch fermentation process using batch-wise linearised models identified from process operation data. The newly obtained process operation data after each batch is added to the historical data base and an updated linearised model is adaptively identified. In an effort to adapt to the current process environment, the updated model is identified from a moving window of the most recent historical batches. The new model is used to compute control policy for the next batch. The control policy at different batch stages are generally correlated as the overall control policy is obtained to maximize the amount of product at the end of a batch. To address the colinearity issue, partial least square (PLS) is used in estimating the linearised model parameters. The proposed strategy is applied to a simulated fed-batch fermentation process and the performance is evaluated. The effect of window sizes was studied. Simulation results show that the proposed approach improves the batch-to-batch ILC performance.

Batch-to-Batch Iterative Learning Control of a Fed-batch Fermentation Process Using Incrementally Updated Models

Abstract Batch-to-batch iterative learning control of a fed-batch fermentation process using batchwise linearised models identified from process operation data is presented in this paper. Due to model-plant mismatches and the present of unknown disturbances, off-line calculated control policy may not be optimal when implemented to the real process. The repetitive nature of batch process allows information from the previous batches being used in modifying the control policy of the next batch in the framework of iterative learning control. In order to cope with nonlinear behaviour of batch fermentation processes, the model is lineariesed using the immediate previous batch as a reference batch and the model is updated from batch to batch. The control policy (feed rates) at different batch stages are generally correlated as the overall control policy is obtained to maximize the amount of product at the end of a batch. To address the colinearity issue of the control variable, principal component regression and partial least squares regression are used in estimating the linearised model parameters. Application results on a simulated industrial scale fed-batch fermentation process demonstrate that the proposed strategy is effective.

Reliable Batch-to-Batch Iterative Learning Control of a Fed-batch Fermentation Process

Abstract A reliable batch-to-batch iterative learning control (ILC) strategy is proposed. A batch-wise linearised model of the batch process is obtained from historical process operation data and is used to derive the ILC strategy. In order to enhance the reliability of ILC, model prediction confidence is incorporated in the ILC optimisation objective function and control policies leading to wide model prediction confidence bounds are penalised. In order to cope with nonlinearities, the batch-wise linearised model is re-identified after each batch run with the immediate previous batch as the reference batch. The proposed method is applied to a simulated fed-batch fermentation process and the results demonstrate that the proposed reliable ILC strategy is very effective.