Predicting Layer Defects During Additive Manufacturing Using Image Processing and Deep Learning

In recent years, the growing demand for 3D printing technology has enhanced the capabilities of remote operation of 3D printers for fabrication of complex-shaped products, which has introduced significant challenges in quality control, particularly in detecting defects in printed objects. Traditional methods often fall short in effectively identifying defects in a real-time and accurate manner. Nowadays, various artificial intelligence (AI) techniques are being utilized to optimize printing parameters and predict mechanical behavior of 3D-printed components. The present work focuses on in situ imaging during printing and defect prediction using image processing and deep learning. It explores the application of deep learning models in automating the detection of 3D printing defects. In the present study, four models have been compared: a custom CNN from scratch, a custom CNN with data augmentation, VGG16, and ResNet50, utilizing transfer learning techniques. The models are trained on a defect dataset and evaluated based on accuracy, precision, recall, F1 score, and confusion matrices. Our findings demonstrate the superiority of transfer learning-based models in achieving higher accuracy, providing a strong foundation for deploying these models in real-world 3D printing systems.