UC Berkeley

In modern manufacturing, optimizing chemical properties, material composition, and processing parameters is essential for achieving desired performance benchmarks given manufacturing and design constraints. Traditional methods often rely on iterative trial-and-error or brute-force design of experiments (DOE), where materials and operating parameters are selected based on intuition and experience. This process is typically repeated until critical benchmarks are met or resources are depleted. Recent advancements in computing are beginning to transform this approach, enabling rapid multi-physics simulations and efficient machine learning/optimization algorithms that offer significant advantages over traditional DOE methods. These simulations are faster, more cost-effective, and environmentally friendly, reducing engineering time and manufacturing resources while minimizing overall development risk.

This work presents an integrated approach that combines experimentation, multi-physics modeling/simulation, numerical optimization, and machine learning techniques. These components are integrated into a cohesive, simulation-informed optimization framework for designing materials in advanced manufacturing applications. This dissertation demonstrates how these components interact and inform each other in the context of designing acrylate-based UV-curable inks for additive manufacturing processes. Specifically, it illustrates how multi-physics modeling provides a virtual environment, and how Evolutionary Strategies and Bayesian Optimization accelerate the search for optimal input parameters within experimentally determined constraints. This comprehensive approach not only offers a more efficient method for addressing formulation strategies in additive manufacturing but also paves the way for general material development across various industrial applications.