XEdgeAI: A Human-centered Industrial Inspection Framework with Data-centric Explainable Edge AI Approach

Explainability is essential for industrial AI adoption: The framework demonstrates that providing both visual and textual explanations alongside AI predictions is "crucial for end-user trust and model interpretability," suggesting that workers and inspectors are more likely to rely on AI systems when they can understand why the system made a decision.

AI can work effectively on resource-limited devices: The research successfully deployed semantic segmentation models on mobile devices while maintaining "competitive accuracy" and "significantly reducing model size," proving that responsible AI doesn't require expensive, centralized computing infrastructure—it can operate where inspections actually happen.

Domain expert knowledge improves AI fairness and performance: By incorporating "domain expert knowledge" through XAI-guided data augmentation, the framework enhanced model performance, indicating that involving human experts in AI development leads to more reliable and trustworthy systems rather than purely automated approaches.

Interpretability requires intentional design, not an afterthought: The framework includes six dedicated modules specifically for explanation generation and evaluation, suggesting that explainability cannot simply be added to existing AI systems—it must be built into the development process from the beginning to be effective.

Visual explanations matter for non-technical users: The emphasis on delivering explanations through both visual and textual formats acknowledges that end-users (industrial inspectors) may not understand technical AI concepts, making accessible communication a key component of responsible AI deployment.