Data-Driven Design of Engineering Material Systems: from Artificial Intelligence to Physical Intelligence

Liwei Wang (CMU)

Abstract:

Modern machines often rely on complex electronics for sensing, actuation, and control. Recent advancements, particularly in 3D/4D printing, introduce a new paradigm by embedding intelligent functions directly into materials and structures. This integration gives rise to the concept of physical intelligence, enabling multifunctionality, remote control, and autonomous operation across various scales. The potential of such systems is vast, offering groundbreaking applications in fields such as biomedical engineering, manufacturing, and robotics.However, modeling and designing these systems presents significant challenges, as they involve multiple scales, intricate physical interactions, mixed-type variables, and complex interrelationships between materials, structures, and external stimuli.

In this talk, we will explore how computational intelligence—including artificial intelligence (AI) and computational optimization—can be leveraged to develop design principles and frameworks for engineering material systems that embody physical intelligence. The first part of the talk will focus on the integration of Gaussian processes, active learning, and deep generative modeling to optimize heterogeneous metamaterial systems across multiple scales. The proposed approaches enable efficient design to meet customized performance requirements, leading to applications such as vibration control and mechanical cloaks. The second part will discuss the integration of metamaterials with responsive materials to enable system programmability and active control. Using differentiable simulation and physics-informed machine learning, we will demonstrate efficient multi-physics optimization techniques that simultaneously design structures, materials, stimuli, and manufacturing processes, enabling systems to dynamically transition between multiple functional states in response to various stimuli.