Deformation-induced damage investigated through deep learning and microscale observations

The occurrence of microstructural damage in the form of deformation-induced voids affects the properties and performance of a formed part on multiple scales. This process begins with the formation of microscale voids and ends with macroscopic failure due to accumulated damage during plastic deformation. It is because of this multi-scale effect, that the initiation and evolution of these deformation-induced voids needs to be viewed and analysed from a multi-scale point of view. Such an approach is expected to start with the mechanisms of plasticity in a mechanically heterogeneous microstructure, at the nanoscale, leading to the mechanisms of damage formation. To gain further insights about the formation of these types of deformation-induced damage, however, the viewpoint on these microscale events needs to be enlarged, gathering information about a whole plethora of damage sites, in order to achieve statistically sound assertions. Ideally, the scale can be enlarged in such a way that correlations to global parameters of the deformation process are possible, leading to a bridging of the scales from microstructural mechanisms to the order of magnitude required to engineer complete forming processes with respect to damage formation. To achieve this bridging of scales between nanoscale investigations on damage mechanisms and damage quantification on the scale of a formed component, this work presents a complementary approach consisting of the characterisation of local plastic properties via nanomechanical experiments, and new methods for automated characterisation and quantification of deformation-induced voids, that has been developed as part of this dissertation, utilising automated SEM imaging and deep neural networks. This dissertation consists of four research publications, three of which are concerned with the characterisation and quantification of emerging microstructural voids and their mechanisms of initiation and evolution in dual-phase steel DP800. A scale-bridging experimental and computational method that spans from automated imaging, void detection and mechanism classification utilising deep neural networks is introduced in publication #1. It therefore presents an approach to gain statistically relevant information about damage incidents and their originating mechanisms in a way that bridges the gap between the microscale mechanism and the global state and quantity of deformation-induced damage on the scale of a formed part. As such a method can be applied to larger scale specimens to quantify damage behaviour, it is used in publication #2 to unravel the effects of the globally applied stress state in a systematic study on the influence of stress triaxiality on damage formation. Due to the large field of view, the previously assumed dependency could be proven for a real, heterogeneous microstructure, in which the evolution of a single void is dominated by the local stress state, exerted by local microstructural properties and morphology. Viewed in the statistically relevant ensemble of voids reached with this new approach, the expected dependency could be visualised. Having applied the approach to such a systematic study, it is applied to a technological forming process in publication #3, namely bending. Here, the effect of reduced triaxiality, due to superposed compressive stresses, on void formation could be proven due to the microscale measurements of single voids on an area large enough to be relevant to the scale of a sheet metal forming process. Publication #4 is thematically concerned with characterising the plasticity of α-manganese sulphide (MnS), by applying the methods of nanoindentation and micropillar compression. As in many steels besides dual-phase steels, foreign phases like inclusions are a major cause for damage formation, the plasticity of such inclusions is of high interest to research and modelling on this mechanism of damage. MnS is a common inclusion phase in many case-hardened steels, which is why the quantitative data on crystal plasticity gathered by this study is a vital addition to the research on how the plasticity of inclusions like MnS effect damage initiation. As a whole, this work presents an approach to quantify deformation-induced damage on all scales, bringing a scientific benefit to the research on damage formation and its mechanisms through its considerably increased efficiency. It is applicable for researchers working on all steps in the chain of development for damage-tolerant materials and processes, and considers all length scales from the fundamentals of plasticity, applicable for crystal-plasticity simulations ultimately simulating the process of damage formation, to the research on the damage mechanisms itself - through its findings on the dominance of damage mechanisms and the observations on stress and strain dependence of initiation and growth. Finally, on the macroscopic scale of formed components, this approach originating from the microscale can be applied to make the quantification of damage in these parts more accurate and representativ