Zero-shot Classification of Additive Manufacturing Anomalies

Additive Manufacturing (AM) process involves building the part layer by layer. Depending on the AM process, various defects (Anomalies) are possible. These defects adversely affect the strength characteristics of the manufactured part. For any given set of AM process parameters, the melt pool characteristics should be similar (within a range) at any instance of AM printing process. Hence the images of the melt pool taken during the AM printing process are expected to have similar characteristics at any given point of AM printing. So, a melt pool image of an anomaly is different from melt pool images from normal printing process. Hence a zero shot classification of anomaly is done with a vision transformer model and a vector database of created from melt pool images.