Model Libraries for Rapid Design and Prototyping
Abstract
Over the past decade, significant investments have been made in high performance computing. These investments have resulted in revolutionary advances, allowing for larger models to be analyzed in less time. Design engineers now routinely evaluate models with over one million degrees offreedom. However, a significant challenge remains. This challenge is to extract and use knowledge from models, so that every design exercise is not evaluating models from scratch but rather informed by knowledge from model libraries. Currently, many model calculations are redundant in that similar model calculations have been performed previously. The research problem addressed by this work may therefore be summarized as follows: How do we leverage advances in high performance computing to extract qualitative and quantitative knowledge essential to the rapid design and prototyping of new dynamic systems from model libraries? The objective of the proposed work isto produce theories, algorithms, and databases that derive design knowledge from models without relying on human discovery. The hypothesis of the proposed work is that such knowledge is best derived by creating libraries of models specifically for this purpose and using existing algorithmsfrom computer science to analyze them. Previous work by our group has demonstrated dramatic performance increases using high performance models to optimize the distribution of damping in a structure, and has distributed those tools to the Navy research community. The proposed work represents a revolutionary step in this direction. We have established collaborations with computer scientists at Boston University who will support us in the selection and application of algorithms. The methodology proposed here is to create libraries of results from models specifically designed for that purpose using a knowledge of structural acoustics to guide selection and evaluation of models. Next, we will consider the estimation of performance metrics needed for rapid designand prototyping. State-of-the-art algorithms from machine learning and artificial intelligence will be applied to determine how these system characteristics vary with design. The proposed work is a significant improvement over simple repeated evaluations of similar models. The proposed work will produce estimates of performance criteria that are sufficiently accurate to design andprototype a fundamentally new structure. It will also provide design trends and tradeoffs that will dramatically accelerate the design and prototyping process.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 08, 2020
- Source ID
- N000141912100
Entities
People
- James Mcdaniel
Organizations
- Boston University
- Office of Naval Research
- United States Navy