This paper explored the integration of AI with 3D printing technology to address two key challenges: predicting surface roughness and detecting printing defects. A full factorial design of experiments was conducted using a Bambu Lab P1S printer and PLA material, varying three process parameters: temperature, layer height, and filling density, across 27 combinations. Surface roughness (Ra) was measured with a TR200 profilometer, and the data was used to train and evaluate five machine learning models: ANN, SVM, DT, RF, and KNN. Among these, SVM demonstrated the best generalization performance, while ANN achieved the highest R² on the training set. Meanwhile, a custom CNN was trained on a publicly available dataset containing 1,912 labelled defect images to classify five common FDM defects. The CNN achieved an accuracy of 98.6% along with excellent ROC performance, confirming its reliability for real-world defect detection. The results validate the effectiveness of AI-driven approaches for additive manufacturing process optimization and quality assurance. The study demonstrates how machine learning and deep learning models can improve the intelligence, accuracy, and efficiency of FDM systems, thereby helping to achieve smart manufacturing and Industry 4.0 goals.
Liu et al. (Wed,) studied this question.