The structural analysis and safety assessment of safety-critical components requires the transfer of standardized laboratory tests to structural application for each considered material. For example, the fracture mechanical evaluation is based on macroscopic concepts and the comparison of global load and materials fracture toughness (resistance to crack propagation). This work investigates the implementation of supervised, semi-supervised and unsupervised machine learning (ML) based computer vision techniques into fracture surface analysis, for fracture toughness evaluation and notched bar impact testing. In this interdisciplinary approach, computer vision-driven ML models are used to extract and evaluate complex failure mechanisms for two material classes commonly used in nuclear safety components. However, the developed methodology is not limited to these materials and encourages the transfer to other fields of application. Fracture surfaces represent a materials loading history and the materials correspondence in terms of fracture and deformation behaviour. Fracture surfaces can therefore be seen as a unique identifier or “fingerprint”. The deployed models demonstrate different aspects of fracture surface evaluation. For fracture toughness testing, macroscale measurement procedures are facilitated with increased reproducibility and efficiency, while the subjective bias of manual measurements is eliminated. Microstructure parameters are evaluated locally, revealing the features directly involved in the fracture process, enhancing microstructure modeling. Fracture surfaces originating from notched bar impact tests are used to predict the materials force-displacement behaviour generating added value to non-instrumented testing. By providing model interpretability and statistical validation, the trust in the presented techniques is increased. This accounts for macro- and microscale related tasks, different material classes and test scenarios. The work lays the base for data-driven, interpretable and efficient fracture surface evaluation and advances the methodologies crucial for enhancing safety and reliability in critical applications and promotes the use of ML in materials science for fracture surface analysis in general.
Johannes Rosenberger (Wed,) studied this question.