An Explainable Hybrid Deep Learning Framework for Automated Transportation Infrastructure Defect Detection and Predictive Maintenance
Abstract
Transportation infrastructure monitoring is essential for ensuring public safety, improving maintenance efficiency, and extending the operational lifespan of critical assets. Conventional inspection approaches are often labour-intensive, time-consuming, and limited by human subjectivity and environmental constraints. Recent advancements in deep learning and computer vision have enabled automated defect detection; however, existing models frequently suffer from limited generalization, insufficient interpretability, and weak integration with predictive maintenance strategies. This study proposes an explainable hybrid deep learning framework for automated transportation infrastructure defect detection and predictive maintenance. The proposed framework integrates convolutional neural networks (CNNs) for spatial feature extraction with attention-based Transformer mechanisms for capturing complex global feature relationships. An explainable artificial intelligence (XAI) module is incorporated to improve model transparency by identifying the image regions responsible for prediction outcomes. The framework is designed to classify multiple infrastructure defects, including cracks, potholes, corrosion, and surface deterioration, while supporting proactive maintenance decision-making through predictive analysis. Experimental evaluation demonstrates that the proposed approach improves detection accuracy, robustness under varying environmental conditions, and interpretability compared with conventional deep learning approaches. The integration of hybrid learning, explainable AI, and predictive maintenance provides a reliable solution for intelligent infrastructure management. The proposed framework contributes toward the development of safer, sustainable, and data-driven transportation systems by transforming automated inspection from a defect identification process into a comprehensive decision-support mechanism.
How to Cite
Sonika Lakra, Mamta Sharma, "An Explainable Hybrid Deep Learning Framework for Automated Transportation Infrastructure Defect Detection and Predictive Maintenance", Vol. 4, Issue 5, 26-08-2026, pp. 14-35.
