Optimizing dual-shim configurations for non-Newtonian coating liquids in the slot coating process of PEM fuel cells using artificial neural networks

초록

We developed a new machine learning-based method to optimize dual-shim configurations, aiming to enhance flow uniformity at the die exit in slot coating processes for proton exchange membrane fuel cells (PEMFCs). Initially, we conducted three-dimensional computational fluid dynamics (CFD) simulations to characterize the internal die flows using both single- and dual-shim configurations. The comparative simulations demonstrated that the dual-shim configuration can effectively regulate the sharp changes in velocity distribution observed with a single-shim configuration, without altering the manifold geometry of the slot die itself. Furthermore, we explored the impact of the geometric structures of the dual shim on the internal die flow and velocity distribution at the die exit through CFD simulations. However, identifying the optimal configuration by scanning multiple geometric variables of the dual shim requires an extremely large number of iterations, which is challenging using CFD simulations alone. To overcome this limitation, we developed a highly accurate artificial neural network (ANN) model using a moderate-sized dataset derived from CFD simulations across just 60 combinations of five geometric variables. Leveraging this computationally efficient ANN model developed as a reduced-order CFD model, we scanned the entire parameter space by discretizing each of the five geometric variables into 20 different levels. This comprehensive simulation enabled us to identify the optimal dual-shim configuration, achieving a 68% reduction in flow non-uniformity (quantified by the relative velocity deviation), compared to the single basic shim configuration. Our results demonstrate the efficacy of the ANN-based reduced-order CFD model in optimally designing dual-shim configurations to enhance the uniformity of the internal die flow.