Challenges and Opportunities in Credibility Assessment of Computational Models: A Case Study in Rapid Discovery and Testing of Metallic Alloys

Abstract

Abstract Advances in high-resolution, high-throughput measurement techniques, coupled with powerful recent developments in computational modeling, simulation, and artificial intelligence (AI), have enabled breakthroughs in rapid discovery and testing of novel materials. These possibilities raise new questions pertaining to the credibility of the underlying models for real-world use, especially when these models are adopted to inform engineering or regulatory decisions in applications that deal with human safety and life. We present a risk-informed credibility assessment framework, that has been applied in adjacent domains to support engineering and regulatory decision-making based on the use of computational (including AI) models, and demonstrate its application in the context of rapid discovery and testing of metallic alloys. The framework consists of a number of steps and best practices to define a question of interest, establish a context of use, assess the model risk, develop and execute a plan to validate and verify the model’s credibility within that context, document the results and deviations from the plan, determine adequacy of the models based on such results, and iterate over these steps until the required adequacy is achieved. We present a few case studies using a number of models developed to support rapid discovery and testing of novel metallic alloys in DARPA METALS. We highlight challenges, gaps, and opportunities for future directions that are generalizable to broader engineering applications.

Ähnliche Arbeiten