Although the range of applications for image processing has been growing for years, only a portion of industrial visual inspection tasks have been automated to date. Artificial intelligence (AI) methods hold the potential to automate even difficult visual inspection tasks that have been performed manually so far. Since classical deep learning (DL) techniques require a large amount of training data, they are hardly applicable for inspection objects with high variance and small batch sizes. Furthermore, common DL architectures are typically restricted to the evaluation of single image perspectives, have limited explanatory power due to their inherent black-box nature, and lack the ability to incorporate existing factual and rule knowledge. Accordingly, the goal of the present research project is to meet the aforementioned challenges by developing a hybrid, data-efficient AI solution, making it possible to automatically inspect assembly assemblies with a high number of variants in the near future.