This paper proposes a novel Bayesian network-based methodology for anomaly detection in fabricated artifacts using unstructured 3D point cloud data. The approach addresses challenges such as nonexistence of global coordinate ordering, variant anomaly patterns, and the presence of outliers.