A physics-informed machine learning framework for constitutive model development

Created by W.Langdon from gp-bibliography.bib Revision:1.8276

@PhdThesis{Garbrecht:thesis,

• author = "Karl Michael Garbrecht",
• title = "A physics-informed machine learning framework for constitutive model development",
• school = "Department of Mechanical Engineering, The University of Utah",
• year = "2024",
• address = "USA",
• month = may,
• keywords = "genetic algorithms, genetic programming",
• URL = "https://www.proquest.com/dissertations-theses/physics-informed-machine-learning-framework/docview/3112739837/se-2",
• size = "185 pages",
• abstract = "This work aims to advance machine learning-based homogenization methods for development of constitutive equations (e.g., equations relating stress and strain that are used in constitutive models). The macroscopic constitutive behavior (i.e., the mechanical stress-strain response) of a material volume element is governed by the cumulative contributions of the local constitutive behavior within that volume element. One approach to incorporating local considerations into higher length scale constitutive equations is micro-mechanical analysis. This work presents novel methods for micromechanical analysis that combine data-driven and analytical methods. Specifically, the data-driven methods consist of a physics-informed genetic programming based symbolic regression (P-GPSR) framework developed herein. The framework is such that P-GPSR is applied to address analytically intractable considerations to produce free-form data-driven equations that are embedded within what would other otherwise be analytical derivations. The capability to strongly enforce known terms in a P-GPSR produced solution reduced the search space and resulted in the data-driven equations (i.e., place holder functions) being constrained precisely where needed. In combination with tailored data sets and solution requirements, theoretical consistency and the a priori physical significance of the data-driven components can be ensured. Therefore, the P-GPSR framework is suitable for use in scientific and engineering applications. Overall, the P-GPSR framework advanced the state of P-ML by introducing a comprehensive method for identifying and implementing solution requirements for the development of accurate, theoretically consistent, and transparent data-driven models.",
• notes = "Order No. 31149397. Available from ProQuest Dissertations and Theses Global",

}

Genetic Programming entries for Karl Michael Garbrecht

Citations