Damage Mechanics Challenge: Simulation of Failure Load and Crack Geometry
Abstract
As artificial intelligence methods advance in the detection of anomalous signals in data from sensors, methods are needed to link these signatures to the underlying physics/mechanics of failure to determine if failure is imminent. This requires robust computational methods that capture the physics of failure and identify the measurable signatures of failure. While there are many computational approaches for simulating damage, few have been ground-truth tested with either known experimental data or with blind data sets.
We will present a benchmark laboratory data set to initiate a damage mechanics challenge to compare computational approaches on damage evolution in brittle-ductile material. The experimental design was developed as a community effort at a Damage Mechanics Workshop held at Purdue in February 2019 which included lead computational scientists and engineers in the field of damage mechanics. The consensus, at the workshop, was to: (1) determine the state of the art and future directions to improve the community's ability to simulate crack formation and evolution in natural and engineered brittle-ductile materials; (2) identify the information provided by the different simulation approaches that gives insight into the prediction and interpretation of failure in rock; (3) identify model parameters that are currently not measured or cannot be measured in the laboratory; and (4) determine whether there are other experimental measurements that are needed or better methods of performing measurements to monitor damage evolution. The benchmark laboratory datasets include spatial and temporal measurements from traditional digital load-displacement sensors, 3D digital image correlation to map surface deformations, 3D X-ray microscopy to ground-truth the crack-failure geometry, and laser profilometry to capture surface roughness. The samples were fabricated through additive manufacturing methods (e.g. 3D printing) to produce repeatable specimens designed to fail in controlled ways. These methods were selected to ensure that participant-defined repeatable and unbiased metrics were available to quantitatively assess and measure the quality of the theoretical and data-driven models, given the significant influence of inherent uncertainty and variability on the onset and modes of failure.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFMMR23C0110B
- Keywords:
-
- 3909 Elasticity and anelasticity;
- MINERAL PHYSICS;
- 8010 Fractures and faults;
- STRUCTURAL GEOLOGY;
- 8020 Mechanics;
- theory;
- and modeling;
- STRUCTURAL GEOLOGY;
- 8164 Stresses: crust and lithosphere;
- TECTONOPHYSICS