US 12,394,037 B2
Physics-informed anomaly detection in formed metal parts
Baris Erol, Rochester Hills, MI (US); Jason Dube, Windsor (CA); and Yuan Zi, Houston, TX (US)
Assigned to Siemens Aktiengesellschaft, Munich (DE)
Filed by Siemens Aktiengesellschaft, Munich (DE)
Filed on Jan. 4, 2023, as Appl. No. 18/149,686.
Claims priority of application No. 22151483 (EP), filed on Jan. 14, 2022.
Prior Publication US 2023/0230224 A1, Jul. 20, 2023
Int. Cl. G06T 7/00 (2017.01); G06T 7/33 (2017.01); G06V 10/25 (2022.01); G06V 10/74 (2022.01)
CPC G06T 7/0008 (2013.01) [G06T 7/001 (2013.01); G06T 7/337 (2017.01); G06V 10/25 (2022.01); G06V 10/761 (2022.01); G06T 2207/10024 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/30136 (2013.01); G06V 2201/07 (2022.01)] 15 Claims
OG exemplary drawing
 
1. A computer-implemented method for detecting defects in a formed metal part, comprising:
locating one or more regions of interest in a synthetic image of a part manufactured by a forming process, the synthetic image being informed based on a physics-based simulation of material state resulting from the forming process, the one or more regions of interest indicative of a higher risk of having a defect resulting from the forming process than other regions in the synthetic image,
registering a set of training images comprising real images of actual manufactured parts with the synthetic image, and overlaying the one or more regions of interest on each of the training images to extract patches from the training images that correspond to high-risk regions, and
training a first anomaly detection model on the patches extracted from the training images, wherein the first anomaly detection model is executable by a processor to detect a defect in a formed metal part in a manufacturing line from an acquired image of the formed metal part by detecting an anomaly in a patch extracted from the acquired image that corresponds to a high-risk region.