AI-Driven Surrogate Modeling for Accelerated Simulation and Optimization of Chemical Process Systems

Abstract

Modern chemical process system design and optimization is more and more based on high-fidelity simulations, including computational fluid dynamics (CFD) and first-principles-based models, to describe complex physicochemical interactions. Although these methods give the correct representations, their high computational cost limits them to iterative optimization, quantification of uncertainty, and real-time digital twin applications. To overcome this problem, this paper constructs an artificial intelligence (AI) based surrogate modeling system that integrates both deep learning systems and state-of-the-art regression techniques to provide high accuracy with low computational cost estimates of rigorous simulations. Two case studies have been used to illustrate the proposed methodology: (i) a multi-stage distillation column is modeled in Aspen Plus; (ii) a catalytic reactor is simulated by nonlinear kinetics in CFD. High-fidelity simulation data were preserved to train deep neural networks (DNNs) and Gaussian process regression (GPR) models and then test them on unseen situations. Findings indicate that the surrogate models can predict with 95 percent precision and can save up to 90 percent or so computational times when compared with traditional simulations. Moreover, incorporation of these surrogates in evolutionary optimization models hastened convergence by almost five times without the cost of decreasing the quality of solutions. The results indicate the power of AI-based surrogate modeling to revolutionize the process systems engineering by allowing the scalability and cost-effectiveness of workflows that are energy-efficient. Future directions will also involve the use of physics-informed neural networks, cross-process generalization through transfer learning, and real-time use in industrial digital twin settings.