A regularized concrete model for high strain rates with a FAIR parameter estimation framework

点击翻译按钮获取中文摘要

The investigation of concrete behavior under blast loads is of utmost importance for industries and government agencies handling high explosives or seeking protection against intentional blast events. While field experiments offer insights, they are resource-intensive. Therefore, numerical simulations provide a cost-effective alternative with adjustable parameters. For the simulation of the complex behavior of concrete under high strain rates, several local models exist that consider different phenomena (nonlinear equation of state with history variable for pore-crushing, strain-rate-dependency, hardening, softening, pressure-dependent yield surfaces, residual strength in compressive states, etc.). This complexity also results in a large parameter space, making model calibration a non-trivial task. In recent years, gradient-enhancement for explicit simulations has been introduced. However, these models only partly cover the complexity of the failure mechanisms observed in reality. Therefore, a regularized extension of the RHT model is proposed that simulates the complex material behavior while being mesh-independent. This includes a gradient-enhancement with an inertia term to model the strength increase of the concrete in high strain rates. Additionally, a viscosity term is introduced to account for low strain rates. In contrast to a description of the strain-rate dependent strength using phenomenological approaches in state-of-the art models, the new model is based on physical assumptions. The gradient enhancement is further combined with a decreasing length parameter, mitigating a spurious expansion of the damage zone. Integrating gradient enhancement introduces additional parameters to an already complex model complicating the parameter determination even further. To leverage prior research, a database of experimental data with corresponding metadata schema is developed and shared with the research community according to the FAIR principles. Due to the regularization, constitutive parameters are mesh-independent material parameters which are determined using Bayesian inference. In practical applications, acquiring the complete parameter set from limited data without conducting experiments is desirable. Hence, the parameter space is explored using machine learning and dimensionality reduction techniques to identify possible relationships among the parameters and the concrete mix composition. This approach enables the determination of reliable approximations for arbitrary concrete mixtures. In order to ensure the reproducibility of the entire methodology, including the data, the simulation models and the determination of the material parameters, all software components are developed as open source and linked using automation tools. The aim is to develop a methodology that makes it possible to use complex numerical models with quality assurance for industrial applications.