Speaker
Description
In this talk I will discuss efforts of using AI systems to predict fracture and damage. Being able to predict material degradation, and in particular fracture and failure, has been a challenging endeavor for many decades. Dynamic fracture in brittle materials, in particular, has been shown to exhibit a distinct behavior that differs from what analytical theories predict. For example, Linear Elastic Fracture Mechanics says that dynamic crack branching should happen with a halving of crack’s speed after the branching event, and yet, experiments show that this is not the case: cracks continue to move with about the same speed after the branching event. Novel computational models have been introduced in order to simulate this type of phenomenon and attempt to explain the discrepancies between theoretical predictions and experimental observations. Our group has contributed to peridynamic models for fracture and corrosion damage. These nonlocal models have been able to correctly obtain results that match experimental observations in a variety of conditions. Important challenges, however, remain: the cost of nonlocal simulations is very high, especially in 3D problems. The need for faster computations is essential in allowing us to use such tools for the design of, for example, new architected materials that behave robustly even if damage is initiated. This is where AI systems may be able to help. I will give an overview of difficult problems in fracture and damage of materials, as well as some recent attempts to predict fracture using ML. I will present the specific difficulties AI systems encounter when they are applied to problems in fracture and damage, which are sometimes so sensitive to initial and boundary conditions that they can be classified as “ill-posed problems”.
