Model-Free Optimization For A Type Of Batch Processes With A Short Cycle Time And Low Operational Cost

Nowadays, model-based optimization (MFO) is the predominant methodology on quality control and optimization of batch processes. However, the application of MFO is largely limited because of the high costs associated with model construction and low efficiency due to model mismatch. A novel model-free optimization (MFO) is proposed in this thesis for a type of batch processes with short cycle time and low operational cost. Instead of using a quality model, online experimental approach is adopted to evaluate quality responses. The model construction and the mismatch problem are avoided in the MFO approach.The MFO is a challenging task because of process noises and disturbances in experiments. High efficient "model-free algorithms", "dynamic quality estimation method", and "iteration termination control method" are integrated to construct a systematic model-free optimization method. The main research works of this dissertation are as follows:1. A systematic framework of the MFO has been proposed for quality control and optimization of a type of batch processes with short cycle time and low operational cost.2. Two model-free algorithms, namely, simultaneous perturbation stochastic approximation (SPSA) and simplex search algorithm, are integrated into the MFO framework, respectively. A numerical test of semi-batch reactor optimization demonstrates the good performance of the MFO with both algorithms. Moreover, performance comparison and analyses are given under different noise levels.3. An online quality estimation method based on small sample statistics is presented to balance quality estimation precision and experimental costs. Based on this method, a dynamic quality estimation strategy is developed. Experimental results manifest that this strategy improves MFO methods in terms of its convergence, robustness, and anti-noise performance.4. Iteration termination control is proposed based on the convergence depth control method. This method can detect the progress which has been and will be achieved during optimization process, and accordingly stop the process at proper time.5. MFO has been implemented on the quality control of injection molding. Part weight, focal length and the dimension are chosen as the controlled quality attributes. Experiments show that the MFO is efficient for this kind of batch processes.6. Experiments are conducted on the quality optimization of injection molding. The results indicated that the MFO is efficient as well. Meanwhile, the iteration termination control method has been online tested. The results show this method can intelligently terminate the iteration.Finally, a conclusion of the research work is given. The applicability, limitations, and the perspectives of the MFO are discussed.