Deep Learning-Assisted Microscopic Polarization Inspection of Micro-Nano Damage Precursors: Automatic,...

Area of Science:

• Materials Science and Engineering
• Manufacturing Technology
• Artificial Intelligence in Manufacturing

Background:

• Additive Manufacturing (AM) offers advantages for complex component fabrication but faces challenges in industrial-grade quality monitoring.
• Existing systems struggle with multi-scale damage precursors, noise interference, and limited high-quality training data for intelligent transformation.
• Automated non-destructive testing (NDT) is crucial for ensuring the reliability of AM components in critical applications.

Purpose of the Study:

• To develop an innovative intelligent inspection framework for automated NDT of micro-nano damage precursors in Additive Manufacturing.
• To overcome key bottlenecks in AM quality monitoring, including data scarcity and complex feature detection.
• To enhance the accuracy and efficiency of detecting surface damage precursors in industrial AM processes.

Main Methods:

• Proposed an innovative microscopic polarization YOLOv11-LSF intelligent inspection framework.
• Integrated a multi-scale perception module with a Large Separable Kernel Attention mechanism for enhanced feature detection.
• Developed a Slim-neck architecture using the VoV-GSCSP module and GSConv for reduced model complexity.
• Introduced a novel simulation strategy with physical features for damage precursors to create a virtual and real integrated training sample library.

Main Results:

• The YOLOv11-LSF model demonstrated significant improvements over the baseline: accuracy (+1.6%), recall (+1.6%), mAP50 (+1.5%), and mAP50-95 (+2.8%).
• Achieved high detection accuracy: 99% for porosity-related micro-nano damage precursors and 94% for cracks.
• Showcased strong small sample adaptation capability and robustness in complex industrial conditions.

Conclusions:

• The YOLOv11-LSF framework provides a reliable technical solution for industrial-grade AM quality monitoring.
• The study advances smart manufacturing quality innovation through cross-scale micro-nano damage inspection.
• The developed methodology effectively addresses limitations of traditional deep learning approaches in AM quality assessment.