Due to the ongoing rise in demand for materials with high strength coupled to great ductility in the automotive industry, DP steel as one of the most widely used Advanced High-Strength Steels (AHSS) has experienced a large research interest. This interest mainly concerns its damage resistance criteria and nucleation mechanisms of ductile damage – leading to the goal of reducing safety factors and thus, increasing efficiency by weight reduction.
Classically applied methods for the analysis of deformation-induced porosities on the microscale, however, deliver an incomplete picture. Post-mortem analysis of damage sites can only unravel the nucleation mechanism for a very limited fraction of the overall porosities, as the individual damage sites evolve with further deformation. More modern in-situ methods deliver a temporal evolution but are very limited in terms of statistical relevance as a spatial limitation to the efficient tracking of only a small number of damage sites is given at high magnification.
For a human, a realistically achievable magnitude of evaluating damage porosities from a multi-stage in-situ deformation experiment covering a large deformed area does not scale up to such a number that a statistically relevant analysis is possible. To enable the processing and evaluation of the whole of porosities detected in a large-area in-situ observation, an algorithm based on machine learning is implemented. Its purpose is to identify the damage sites and simultaneously attribute them to the individual nucleation mechanisms. Furthermore, porosities that are not caused by ductile deformation, such as non-metallic inclusions, are identified and taken out of the calculations for ductile damage statistics. In combination with this advanced computational method, a drastically higher throughput in experimental data, and thus statistics, is realised. As the system is capable of returning site-specific images with their evolution over the distinct deformation steps, it incorporates the spatial resolution as well as the temporal evolution, leading to a highly efficient method in reliably characterising a maximum amount of damage sites from each sample and deformation step and delivering a complete picture of ductile damage evolution.