(191as) Practical Pathways of Building Machine Learning Models Using CAE Simulation Data: Case Studies for Pharma Unit Operations and Deployment Strategies for Democratization
2025 AIChE Annual Meeting
(191as) Practical Pathways of Building Machine Learning Models Using CAE Simulation Data: Case Studies for Pharma Unit Operations and Deployment Strategies for Democratization
Engineering simulations are rapidly becoming foundational tools for process design, analysis, and optimization. As industries seek faster, more accessible, and real-time insights, the integration of machine learning (ML) with simulation workflows offers compelling opportunities. ML models trained on high-fidelity simulation data can predict either high-level key performance indicators (KPIs)—such as mixing in a blender, tablet temperatures in a coater —or full 3D spatial fields like velocity and stress distributions. These models can incorporate both process variability and geometric changes, supporting faster design iteration, predictive monitoring, and broader access to simulation insights by non-specialists.
This work presents practical case studies demonstrating predictive modeling using ML algorithms trained with physics-based simulation. The use of NVIDIA’s physics toolkit such as PhysicsNemo & Omniverse framework is explored and compared with in-house platforms. Additionally, deployment strategies are discussed—from lightweight web applications for numerical or graphical output prediction, to immersive environments using NVIDIA Omniverse for interactive, 3D visualization and collaborative decision-making. These studies provide a technically grounded framework for when and how to integrate ML with engineering simulations, offering insights into data requirements, training workflows, predictive accuracy, and deployment readiness.
