Trustworthy and resilient AI: a case study in remote ship inspection

Deploying AI in real-world environments often leads to performance degradation due to data drift, concept shift, and unforeseen anomalies. This paper presents a scalable approach to building trustworthy and resilient AI for remote ship inspection. Using a defect segmentation AI system as a case study, we demonstrate how continuous monitoring detects early signs of failure, data intelligence identifies high-impact retraining data, and model adaptation efficiently restores performance. We also introduce a collaborative visual inspection portal that enables data upload, AI-driven detection, expert review, and continuous model improvement through feedback. The proposed framework supports long-term reliability and adaptability of AI systems in complex, real-world inspection scenarios.