Precise regulation of extrusion flow remains a major barrier preventing fully autonomous 3D printing, as existing vision-only or sensor-specific approaches fail to capture the complex thermo-mechanical interactions that drive defects such as warping, stringing, layer shifting, and improper fusion. Prior studies using CNNs, Vision Transformers, and foundation models report strong accuracy yet remain constrained by single-modality dependence, limited temporal reasoning, and poor adaptability to new materials making them unsuitable for real-world dynamic environments. To overcome these limitations, this study introduces M-PACT 3D, a multimodal temporal contrastive and physics-guided meta-adaptive framework designed to unify visual cues, multi-sensor signals, and physically meaningful flow dynamics within a single intelligent control architecture. Implemented entirely in PyTorch, the proposed method integrates a Multimodal Spatio-Temporal Fusion Transformer (MST-Transformer), Temporal Contrastive Representation Learning (TCRL), Physics-Guided Neural Dynamics (PGND), Meta-MAML++ rapid material adaptation, and a physics-aware SAC+MPC reinforcement controller. Using three complementary datasets FDM Defect Images (1912 samples), PBF Layer-Wise Thermal Images (1500 + layers), and a full multi-sensor IMU-thermal-force dataset (58 sequences) the system demonstrates superior generalization and physical consistency compared to state-of-the-art baselines. The M-PACT 3D model achieves a Defect Detection Accuracy (DDA) of 96.85%, representing an improvement of 2.15% over CNN-based, 8.4% over ViT-based, and 7.45% over VFM-based methods. The results confirm that multimodal physics-aligned learning significantly enhances robustness, stability, and real-time flow correction. Overall, M-PACT 3D establishes a scalable and adaptive foundation for next-generation autonomous 3D printing systems, positioning this work as a compelling advancement that encourages deeper exploration into physics-aware, meta-adaptive additive-manufacturing intelligence.
Natrayan Lakshmaiya (Sat,) studied this question.