The AI-Reasoner proposed in this paper serves to transparently reason the predictions of AI models by extracting morphological characteristics of defects from images and utilizing decision trees. It provides insight into AI outputs through visual and textual explanations, aiding in the understanding and improvement of AI predictions. The AI-Reasoner’s evaluation on a masked-based defect detection model shows promising results in making the AI’s decision process more accessible and trustworthy.

• AI-Reasoner reasons predictions of AI models transparently.
• Extracts morphological characteristics from images involving defects.
• Provides insights using visual and textual explanations to aid in model understanding.
• Demonstrates effectiveness in explaining predictions of a masked-based defect detection model.
• Contributes to creating more transparent and reliable AI in industrial settings.

The emphasis on explainability and reliability of AI predictions in industrial applications, particularly defect detection, is of paramount importance. This research marks an advancement towards building trust in AI systems through verifiable and interpretable reasoning processes.