Artificial Intelligence Applications in Corrosion Inhibition: Future Directions

Abstract Corrosion remains a critical threat to industrial infrastructure, contributing to global economic losses exceeding USD 2.5 trillion annually. Traditional detection methods like visual inspection and ultrasonic testing are often subjective, time-consuming, and lack scalability. This study uses deep learning models, YOLOv5 and Mask R-CNN, for automated corrosion detection and segmentation. Both models were trained and evaluated for accuracy and performance using an annotated dataset with bounding boxes and segmentation masks. YOLOv5 achieved faster inference and a high detection accuracy (mAP@0.5 = 0.71), proving effective for real-time applications. Mask R-CNN delivered superior segmentation quality, offering precise localization of corroded regions. The results highlight a trade-off between speed and spatial granularity, suggesting that model selection should depend on deployment context, real-time monitoring versus high-fidelity inspection. These findings demonstrate the potential of deep learning to enhance industrial corrosion management through automation and precision. Trial Registration: Not applicable.