A detection method for small casting defects based on bidirectional feature extraction.

Area of Science:

• Materials Science and Engineering
• Computer Science
• Artificial Intelligence

Background:

• X-ray inspection is vital for detecting internal casting defects like pores and inclusions.
• Traditional manual inspection methods are subjective, slow, and error-prone, hindering efficiency and accuracy.
• Automated defect detection is crucial for scientific rigor and precision in casting quality control.

Purpose of the Study:

• To develop a deep learning model for accurate and efficient detection of small-scale defects in castings.
• To enhance the model's ability to identify and locate minute flaws using innovative architectural designs.
• To validate the model's performance against existing state-of-the-art object detection techniques.

Main Methods:

• Proposed a deep learning model with an end-to-end network architecture.
• Utilized a Wasserstein distance-based loss function optimized for small defect targets.
• Developed a dual-layer Encoder-Decoder multi-scale feature extraction architecture (BiSDE) using the Hadamard product.

Main Results:

• The proposed model achieved at least a 5.3% improvement in Mean Average Precision (MAP) compared to state-of-the-art models.
• Ablation studies confirmed that each component significantly contributed to the model's overall performance.
• The model demonstrated superior accuracy in detecting small-scale defects in castings.

Conclusions:

• The developed deep learning model effectively improves detection precision for small-scale casting defects.
• The BiSDE architecture and Wasserstein loss function enhance the identification of minute flaws.
• The research offers significant potential for the automation and intelligent development of industrial defect inspection systems.