Data- Driven Quality Control and Process Planning for Continuous Carbon Fibre Reinforced Thermoplastic Composite 3D Printing

Abstract

Continuous carbon fibre-reinforced thermoplastic (CCFRTP) 3D printing enables lightweight, high-strength structures but remains highly sensitive to deposition parameters, leading to defects such as under-extrusion, over-extrusion, and poor layer adhesion. This study presents a data-driven Prescriptive Quality Control (PQC) framework integrating machine learning and explainable artificial intelligence (XAI) for predictive quality classification and parameter optimization in co-extrusion-based continuous fibre printing. A physically grounded dataset was established by systematically varying print speed, layer height, PLA feed rate, and thermal conditions, supported by optical and scanning electron microscopy (SEM)-based defect labelling across four quality classes. To address data sparsity, a domain-constrained Conditional Tabular Generative Adversarial Network (CTGAN) was employed for data augmentation, followed by supervised learning using Multi-Layer Perceptron (MLP), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) classifiers. XGBoost achieved 98.1% classification accuracy and was selected for its robustness and compatibility with SHapley Additive exPlanations (SHAP) analysis. Feature attribution identified print speed (mean SHAP = 0.351) and layer height (0.272) as dominant quality drivers, consistent with extrusion stability and interlayer fusion mechanics. Decision-boundary analysis revealed feasible parameter regions yielding >0.85 probability of Good-quality prints. Independent forward printing validation on unseen parameter combinations demonstrated 87.5% prediction agreement and 87.5% optimization success, confirming practical generalization of the framework. The proposed approach provides a scalable and interpretable pathway for data-driven process planning in continuous fibre composite additive manufacturing without reliance on complex in-situ sensing.