CAREER: Using Physics-Based Machine Learning to Reconcile the Crack Tip with the Plastic Zone during Fracture of Metals

Rutgers University New Brunswick, New Brunswick NJ

Investigators

Abstract

Metal fracture occurs when cracks or other flaws grow, leading to failure of metallic structures such as buildings and aircraft that has been estimated to cost 4 percent of US gross domestic product. Current theories and models used to predict fracture are based on an incomplete picture of the fracture process, which lead to inaccurate predictions of crack growth, making it difficult to design engineering structures and next-generation structural metals. This Faculty Early Career Development (CAREER) award supports research to develop better material models used in the design of fracture-resistant engineering structures relevant to energy, defense, aerospace, and transportation applications, thereby supporting the U.S. economy and defense. The machine learning methods developed under this project will be broadly applicable to many fields of science and engineering. Through a collaboration with local and national science teaching organizations, the research findings from this project will be used to develop a multi-day lesson-plan for the high school science classroom wherein students learn about fracture of metals using computer simulations. This lesson-plan will be made available to science teachers nationwide. Prevailing theories of metal fracture focus on either the crack tip where dislocation nucleation governs the propensity for brittle fracture, or the plastic zone surrounding the crack tip where plastic dissipation governs the fracture toughness. Recent work has shown, however, that there are important interplays between the crack tip and plastic zone which affect various fracture behaviors such as dislocation multiplication and growth of fatigue cracks. This project is to develop a three-dimensional discrete dislocation dynamics model which simultaneously accounts for the crack tip and the plastic zone by capturing all relevant dislocation and bond-breaking processes in one unified model. To achieve this goal, a physics-informed machine learning-based image solver will be developed which utilizes a new convolutional architecture that couples to the finite element method. Atomistic simulations will be used to quantify the dislocation nucleation rate near the crack tip, grain boundary weakening due to dislocation adsorption and/or emission, and grain boundary decohesion. Predictions from the fracture model will be compared with state-of-the-art fracture experiments. Insights gained from this research will directly inform fracture models used in engineering design, prediction of embrittlement, and alloy design. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.