Title: Deep Learning Models with Hard Physical and Logical Constraints
Abstract: Modern deep learning models have achieved remarkable success in computer vision, natural language processing, and a growing range of scientific applications, from image analysis to molecular and materials modeling. However, their direct application to chemical engineering problems remains challenging because engineering systems are governed by strict physical and logical constraints and are often characterized by sparse, expensive data.
Standard neural networks may produce accurate predictions on average, yet still violate fundamental principles such as mass and energy balances, operational limits, or logical rules, making them unreliable for scientific and industrial use. In this talk, I will present a framework for embedding hard physical and logic constraints directly into deep learning models using ideas from constrained optimization. Instead of enforcing constraints through penalty terms or post-processing, we introduce differentiable projection layers inspired by KKT conditions and convex optimization that guarantee constraint satisfaction by construction.
These layers are model-agnostic, computationally efficient, and compatible with standard training via backpropagation. I will illustrate the approach through chemical engineering case studies, including surrogate modeling of chemical processes with exact mass balance enforcement, as well as broader applications involving inequality and logic constraints. Results show improved data efficiency, stronger generalization, and inviolable feasibility compared to unconstrained and softly constrained models. Overall, this work highlights how optimization-inspired neural architectures can bridge the gap between data-driven learning and first-principles modeling, enabling reliable machine learning tools for safety-critical chemical engineering applications.
Bio: Can obtained his bachelor’s degree from Tsinghua University, China, in Chemical Engineering. He completed his PhD in Chemical Engineering at Carnegie Mellon University. His PhD research is focused on stochastic mixed-integer nonlinear programming and long-term expansion planning of power systems. Can did a one-year Postdoc at Polytechnique Montreal on using machine learning techniques to accelerate optimization algorithms. He joined the Davidson School of Chemical Engineering at Purdue University as an assistant professor in the Fall of 2022. His research group is focused on optimization, machine learning, and applications in sustainable energy systems. His group won Air Liquide’s global scientific challenge on data sharing for decarbonization in 2023, the Amazon Research Award in 2024, and the NSF CAREER Award in 2025.