Multi-view deep information fusion framework for online quality monitoring and autonomous correction in material extrusion additive manufacturing

Fused Filament Fabrication (FFF) faces challenges in maintaining consistent print quality, particularly in detecting and preventing defects like over-extrusion and under-extrusion. Conventional monitoring systems, relying on a single camera, often struggle with complex geometries. This paper presents an innovative multi-view deep information fusion framework for online defect detection and autonomous correction in FFF printing. The system integrates three strategically positioned cameras for synchronised multi-view data acquisition, ensuring comprehensive visual coverage. To enhance robustness across varying conditions, we develop an automated data acquisition and labelling pipeline for high-quality dataset generation. A custom-designed Multi-View Deep Information Fusion Network (MVDIF-Net) integrates complementary information from different perspectives, significantly improving defect detection accuracy. Additionally, we introduce a dual-strategy control mechanism, combining short-term sliding window analysis for rapid responses with long-term trend validation for robust parameter adjustments. The proposed system achieves 97.67% detection accuracy, with F1-scores consistently exceeding 97% across all defect categories. It demonstrates strong online correction capabilities by dynamically adjusting printing parameters under challenging conditions, including severe over-extrusion and under-extrusion. Experimental results highlight significant improvements in defect detection and correction over single-view approaches, addressing the critical need for more reliable and adaptive FFF processes.