In the following work, a generalizable approach for robotically aided quality inspection is proposed and applied in the field of plastic injection molding of small parts. While artificial intelligence has shown promising results for quality inspection, it often requires large training datasets, which are impractical for industrial applications. Defective, or anomalous parts are heavily underrepresented in normal production in comparison to expected parts which makes it a poorly posed problem for supervised learning approaches. In the proposed concept of this work, this issue is tackled by combining an automated inspection procedure with an anomaly detection approach for defect detection. A 7-DoF robotic manipulator was used to automate the part handling in front of an industrial optical camera sensor. Sampling was done on multiple days with varying surrounding conditions to ensure a heterogenous dataset. The captured images were used to train a PaDiM anomaly detection network to reconstruct a normal image of the part. To determine the capabilities of the developed model, different aspects were evaluated: the amount of training data, the output heatmap resolution and the anomaly decision approach. The results for the best combination of parameters show that various defects, such as particles, scratches, missing structures and deformations can be detected with defect detection rates up to 100% while maintaining approximately 91% true negative rates using a small dataset, consisting of 117 parts and low-resolution anomaly heatmaps to enable faster processing times. Furthermore, the study also found that the anomaly decision approach had minimal impact on prediction quality, whereas reducing training samples below 100 and heatmap resolutions below 1024x1024 significantly decreased prediction accuracy. The combination of robotic automation and anomaly detection is thus suitable for quick adaptation to multiple use cases.