Abstract Additive manufacturing, particularly Fused Filament Fabrication (FFF), increasingly benefits from machine learning to predict failures and optimize print quality. This study investigates a supervised learning approach that integrates design parameters, printing conditions, and image-based features to predict potential failures before the additive manufacturing process begins, with a specific focus on identifying whether a print will be successful or not. An arch-shaped model was selected due to its geometric complexity, and its sensitivity to unsupported printing conditions; notably, the arch was printed without the use of any support structures, making the process more prone to deformation and failure. This geometry enabled a systematic examination of four key design parameters: the distance between the supporting columns of the arch, the slope of the arch, the overall arch height, and the layer thickness, all of which directly influence print stability in FFF processes.. Image processing techniques were employed to compare CAD models with the corresponding printed outputs, allowing the extraction of various features that enhanced classification accuracy. Nine different classifiers were implemented across three modeling strategies, to evaluate the effectiveness of various machine learning approaches, with the Artificial Neural Network (ANN) achieving the highest accuracy of 0.948. Multiple k-fold cross-validation experiments confirmed model robustness, while evaluation metrics highlighted the impact of dataset imbalance—Support Vector Machine (SVM) achieved perfect sensitivity, and Naïve Bayes (NB) demonstrated high precision. Overall, this work demonstrates that combining multimedia data, geometric design variables, and machine learning techniques significantly improves predictive reliability and decision-making in FFF-based additive manufacturing. Graphical Abstract
Tzimtzimis et al. (Mon,) studied this question.