Heterogeneity: The key to forecasting material failure?

Abstract:

Empirical mechanistic models have been applied to the description of the stress and strain rate upon failure for heterogeneous materials. The behaviour of porous rocks and their analogous two-phase viscoelastic suspensions are particularly well-described by such models. Nevertheless, failure cannot yet be predicted forcing a reliance on other empirical prediction tools such as the Failure Forecast Method (FFM). Measurable, accelerating rates of physical signals (e.g., seismicity and deformation) preceding failure are often used as proxies for damage accumulation in the FFM. Previous studies have already statistically assessed the applicability and performance of the FFM, but none (to the best of our knowledge) has done so in terms of intrinsic material properties. Here we use a rheological standard glass, which has been powdered and then sintered for different times (up to 32 hours) at high temperature (675°C) in order to achieve a sample suite with porosities in the range of 0.10-0.45 gas volume fraction. This sample suite was then subjected to mechanical tests in a uniaxial press at a constant strain rate of 10^-3 s^-1 and a temperature in the region of the glass transition. A dual acoustic emission (AE) rig has been employed to test the success of the FFM in these materials of systematically varying porosity. The pore-emanating crack model describes well the peak stress at failure in the elastic regime for these materials. We show that the FFM predicts failure within 0-15% error at porosities >0.2. However, when porosities are <0.2, the forecast error associated with predicting the failure time increases to >100%. We interpret these results as a function of the low efficiency with which strain energy can be released in the scenario where there are few or no heterogeneities from which cracks can propagate. These observations shed light on questions surrounding the variable efficacy of the FFM applied to active volcanoes. In particular, they provide a systematic demonstration of the fact that a good understanding of the material properties is required. Thus, we wish to emphasize the need for a better coupling of empirical failure forecasting models with mechanical parameters, such as failure criteria for heterogeneous materials, and point to the implications of this for a broad range of material-based disciplines.