Additive manufacturing offers significant design freedom and the ability to selectively influence material properties. However, conventional processes like laser powder bed fusion for metals may result in internal defects, such as pores, which profoundly affect the mechanical characteristics of the components. The extent of this influence varies depending on the specific defect type, its size, and morphology. Furthermore, a single component may exhibit various defect types due to the manufacturing process. To investigate these occurrences with regard to other target variables, this study presents a random forest tree model capable of classifying defects in binary images derived from micrographs. Our approach demonstrates a classification accuracy of approximately 95% when distinguishing between keyhole and lack of fusion defects, as well as process pores. In contrast, unsupervised models yielded prediction accuracies below 60%. The model’s accuracy in differentiating between lack of fusion and keyhole defects varies based on the manufacturing process’s parameters, primarily due to the irregular shapes of keyhole defects. We provide the model alongside this paper, which can be utilized on a standard computer without the need for in situ monitoring systems during the additive manufacturing process.

In this Article, the targeted adjustment of the relative density of laser additive manufactured components made of AlSi10Mg is considered. The interest in demand-oriented process parameters is steadily increasing. Thus, shorter process times and lower unit costs can be achieved with decreasing component densities. Especially when hot isostatic pressing is considered as a post-processing step. In order to be able to generate process parameters automatically, a model hypothesis is learned via artificial neural networks (ANN) for a density range from 70% to almost 100%, based on a synthetic dataset with equally distributed process parameters and a statistical test series with 256 full factorial combined instances. This allows the achievable relative density to be predicted from given process parameters. Based on the best model, a database approach and supervised training of concatenated ANNs are developed to solve the inverse parameter prediction problem for a target density. In this way, it is possible to generate a parameter prediction model for the high-dimensional result space through constraints that are shown with synthetic test data sets. The presented concatenated ANN model is able to reproduce the origin distribution. The relative density of synthetic data can be predicted with an R²-value of 0.98. The mean build rate can be increased by 12% with the formulation of a hint during the backward model training. The application of the experimental data shows increased fuzziness related to the big data gaps and a small number of instances. For practical use, this algorithm could be trained on increased data sets and can be expanded by properties such as surface quality, residual stress, or mechanical strength. With knowledge of the necessary (mechanical) properties of the components, the model can be used to generate appropriate process parameters. This way, the processing time and the amount of scrap parts can be reduced.