(29f) Multi-Model Based Model Predictive Control of a Rotational Molding Process

Rotational Molding, also known as rotomolding or rotational casting, is a batch process used for the processing of plastics through distinct heating and cooling phases [1]. In this process a mold filled with a powdered charge is slowly rotated in a heated oven, resulting in the softened material being dispersed and sticking to the walls of the mold. In order to maintain an even wall thickness, the mold is continuously rotated at all times both during the heating and cooling phase. One of the control challenges is to obtain consistent high-quality products from one batch to another while avoiding unfinished parts with incomplete curing or degradation due to prolonged overheating.

In a recent work [2], a batch subspace identification [3] based state-space modeling and control approach is used for rotational molding to achieve desired quality specifications. The modeling approach facilitates the accommodation of variable duration batches without the need for alignment of batch lengths as demonstrated in several other applications as well [3-6]. The results demonstrate the merits of the approach in terms of improved quality, variability rejection across difference batches and the ability to achieve the specified grade of product. In that work, however, the data-driven model is built only for the heating phase. Furthermore, while the implementation tests its merit to achieve new quality targets, the ability of the modeling and control framework to reject raw material variability is not explicitly demonstrated.

Motivated by these considerations, this work presents a multi-model-based model predictive control of a uni-axial rotational molding process to achieve the desired quality specifications. The model consists of two deterministic models for the heating phase and two stochastic models for the cooling phase, identified using deterministic and stochastic batch subspace identification respectively. The four models corresponding to these sequential stages in the process are connected together using the least squares model between the terminal states of the previous model and initial states of the next model. This modeling approach results in improved accuracy of the model predictions enabling improved control of the product properties. In particular, in the previous work [2], the cooling cycle was terminated upon reaching a predefined internal mold temperature. In contrast, in the present work the cooling cycle is terminated based on the real-time predictions such that the final product of desired specifications is achieved. Further, to evaluate the robustness of the proposed approach, we blend the original polymer resin with a similar resin with slightly different melt index and demonstrate the meeting of product specification.

[1] Gomes, F. P. C., Garg, A., Mhaskar, P. and Thompson, M. R. Thompson, Quality monitoring of rotational molded parts using a nondestructive technique, ANTEC (2018).

[2] Garg, A., Gomes, F. P. C., Mhaskar, P., & Thompson, M. R. (2019). Model predictive control of uni-axial rotational molding process. Computers & Chemical Engineering, 121, 306-316.

[3] Corbett, B., Mhaskar, P., Subspace identification for data-driven modelling and quality control of batch processes, AIChE Journal 62 (2016) 1581-1601.

[4] Garg, A., Mhaskar, P. (2017), Subspace Identification Based Modeling and Control of Batch Particulate Processes, Industrial and Engineering Chemistry Research, 56 (26) 7491-7502.

[5] Garg, A., Corbett, B., Mhaskar, P., Hu, G., Flores-Cerrillo, J. (2017), High Fidelity Model Development and Subspace Identification of a Hydrogen Plant Startup Dynamics, Computers and Chemical Engineering, 106 183-190.

[6] Rashid, M. M., Mhaskar, P., & Swartz, C. L. (2017). Handling multi‐rate and missing data in variable duration economic model predictive control of batch processes, AIChE Journal, 63(7), 2705-2718.