(140g) Multiscale Optimization of Integrated Process-Product Systems: Application to Polymeric Coating Curing

A comprehensive and accurate multiscale characterization of product formation is very challenging, because of the existence of numerous interconnected physical and chemical phenomena at different scales. Over recent years, various multiscale modeling and simulation techniques have emerged for characterizing process-product behavior in a comprehensive way. Continuum methods, such as computational fluid dynamics (CFD) based, have been used to describe macroscopic dynamic processing condition featuring mass, momentum and heat transfer. On the other hand, product micro-meso-scale structural evolution can be described using molecular simulation methods, such as Monte Carlo (MC) and molecular dynamics (MD). However, application of multiscale models poses various challenges in a systematic derivation of optimal operational conditions necessary for achieving the best process and product performance. Note that it is computationally prohibitive to accomplish even a partial exploration of a huge solution space.

In this work, we introduce a multiscale optimization framework. This framework consists of three essential components: integrated process-product system multiscale characterization, multiscale model simplification, and optimal solution derivation. First, a multiscale target-oriented method is introduced to guide system model development. By this method, the key process-product variables directly and/or indirectly affecting the interested system performance need to be identified with appropriate time and length scales. Complex inter-correlations among those variables are then established using multiscale models, a scale-coupling method, and performance evaluation models. Then, multiscale models are simplified to achieve a desirable balance between the prediction precision and computational efficiency. In order to simplify CFD models, simple first-principles-based models are adopted to capture dominant system dynamics in lumped areas/volumes, and lumped parameters in those models will be obtained by a data fitting approach. In addition to coarse-graining the system and decreasing the simulation system size, a special technique for simulation task division (into online and offline parts) is also pursued. The resulting multiscale system is then optimized to achieve simultaneously the best possible product and process performance. In this effort, the simplified multiscale models are formulated into an optimization problem, whose solution identification is accomplished by a population-based probability distribution estimation (PPDE) dynamic optimization method.

The efficacy of the generic multiscale optimization framework is demonstrated by a case study on automotive coating curing, which is characterized by coupled CFD and lattice MC models. The multi-zone oven operational settings are successfully derived that can ensure simultaneously the best possible product quality (i.e., solvent resistance and inter-coat adhesion determined by coating's microstructure at a nanometer scale) and process performance (i.e., curing energy consumption at macroscale).