Advanced batch process modelling, control and optimization for injection stretch blow moulding

In addition, preform terminal and heating temperature profiles experiments are designed utilising a combination of THERMOscan and a new self-designed thermocouple based equipment respectively. The temperature profiles are obtained from a number of experiments with different infrared lamp power settings. All of the experimental data is analysed, processed and used as the modelling data pool.<br><br>Furthermore, because the ISBM and reheating processes are quite complex, linear models are not deemed to meet the satisfactory accuracy.

Thus a non-linear identification method, radial basis function (RBF) neural networks model selected by two-stage selection (TSS) based on heuristic approaches including particle swarm optimization (PSO) and differential evolution (DE) is proposed. The main advantage of the proposed method is that the non-linear parameters are optimized in a continuous space by heuristic approaches while the hidden nodes of RBF model are selected one by one in a discrete space using the first stage of TSS algorithm.

Due to this recursive updating mechanism, the computational complexity is significantly reduced. In addition, the insignificant hidden nodes of RBF neural networks model are removed by the second stage of TSS which further improves the model accuracy. The proposed modelling method is then employed for the ISBM and reheating processes.

Experimental results confirm the effectiveness of the proposed modelling method.<br><br>Finally, in order to control the preform temperature profiles, a heuristic based norm-optimal terminal iterative learning control (ILC) method is proposed. Since the reheating process is a batch process, ILC can achieve perfect tracking in a fixed time interval for a repeatable control object by its learning property. The terminal ILC strategy can overcome the problem that only the terminal temperature profile can be measured in the reheating process.

In addition, in order to balance the control performance and energy cost, a norm-optimal method is applied. The main advantage of this control strategy is that the requirement of the system knowledge is minimal. Heuristic methods such as PSO, DE and teaching-learning based optimization (TLBO) can be used as computational tools to calculate the sequence of norm-optimal inputs for this non-linear system model.

Simulation results demonstrate the efficacy of this new control strategy. <br>