Aero-Engine Ablation Defect Detection with Improved CLR-YOLOv11 Algorithm.

Area of Science:

• Aerospace Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Aero-engine ablation detection is crucial for aircraft health management.
• Existing methods struggle with high computational costs and limited local feature extraction.
• Rotation-based object detection faces efficiency and accuracy challenges in this domain.

Purpose of the Study:

• To develop an improved YOLOv11 algorithm for efficient and accurate aero-engine ablation detection.
• To address limitations in computational complexity and local feature extraction in current models.
• To enhance real-time aviation inspection capabilities through optimized object detection.

Main Methods:

• Proposed CLR-YOLOv11 model integrating Context-guided Large-kernel attention and Rotated detection head.
• Implemented a targeted data preprocessing pipeline with geometric and hybrid augmentations, followed by Z-Score normalization.
• Introduced Context-Guided Feature Fusion (C3K2CG) and Efficiency-Oriented Large-Kernel Attention (C2PSLA) modules.

Main Results:

• Achieved 78.5% mAP@0.5:0.95 on a self-built aero-engine ablation dataset.
• Demonstrated a 4.2% improvement over the baseline YOLOv11-obb model without specialized data augmentation.
• The CLR-YOLOv11 model shows significant gains in both detection accuracy and computational efficiency.

Conclusions:

• The CLR-YOLOv11 model offers an effective solution for high-precision, real-time aero-engine ablation detection.
• Synergistic structural optimization, including data preprocessing and attention mechanisms, enhances detection performance.
• This research contributes to advancing aircraft health management through improved visual inspection techniques.