Quality Classification of Defective Parts from Injection Moulding

This report examines machine learning algorithms for detecting short forming and weaving in plastic parts produced by injection moulding. Transfer learning was implemented by using pretrained models and finetuning them on our dataset of 494 samples of 150 by 150 pixels images. The models tested were Xception, InceptionV3 and Resnet-50. Xception showed the highest overall accuracy (86.66%), followed by InceptionV3 (82.47%) and Resnet-50 (80.41%). Short forming was the easiest fault to identify, with the highest F1 score for each model.