Trinity AI Co-Designer for Hierarchical Oracle-guided Design of Cyber-Physical Systems

Abstract

The design of complex cyber-physical systems entails satisfying several competing performance objectives. In practice, some design requirements are often implicit in the intuition and knowledge of designers who have many years of experience working with similar designs. Designers use this experience to sample a few promising candidates in the design space and evaluate or simulate them using detailed, typically slow scientific models. The goal in design is usually to generate a diverse set of high-performing design configurations that allow trade-offs across different objectives and avoid early concretization. In this paper, we present a demonstration of Trinity AI co-designer that implements a machine learning approach to automate some aspects of system design. Trinity implements an extension of oracle-guided inductive synthesis, where the learning approaches interact with a hierarchy of oracles that range from detailed slow-to-evaluate scientific models to fast but low fidelity deep neural network surrogate models and symbolic rules. The goal is to enable fast design iterations for earlier phases of design. Trinity uses deep generative models to learn a manifold of the valid design space, followed by a joint exploration and optimization of designs over the learned manifold, producing a diverse set of optimal designs with respect to given design objectives. In our demonstration, we will use several case studies including the design of propellers, a ground vehicle, air vehicle and underwater vehicle. Across these case studies, we successfully show how our method generates high-performing and diverse designs.