AI Ladder-Based Intelligent Enterprise Resource Planning for Predictive Railway Track Maintenance from UAV Imagery

Abstract

Railway safety relies on the early detection of track defects that can lead to derailments and service disruptions. Traditional inspections are labor-intensive and error-prone, whereas many vision-based studies only focus on detection and fail to link predictions to maintenance execution. This study addresses this perception-to-action gap. In the Collect phase, unmanned aerial vehicles (UAVs) acquire high‑resolution images of track segments. In the Organize phase, images are standardized, binary‑masked to generate ground truth, and embedded into fixed‑length feature vectors. In the Analyze phase, four classifiers including Artificial Neural Networks (ANNs), Support Vector Machines (SVMs), Random Forest (RF), and K-Nearest Neighbors (KNNs) are compared using Area Under the Curve (AUC), accuracy, F1‑score, and Matthews Correlation Coefficient (MCC). In the Infuse phase, the optimal model is integrated into an enterprise resource planning (ERP) maintenance module to support real‑time defect flagging, automated work orders, and dashboard visualization. ANN model achieves the highest performance (AUC = 0.935; accuracy = 0.884; F1‑score = 0.884; MCC = 0.768). The AI Ladder-guided machine-learning (ML)-ERP pipeline demonstrates a practical pathway from aerial sensing to actionable maintenance, aligning with Sustainable Development Goals (SDGs) 9. By directly embedding classification into ERP workflows, operators can transition from periodic, manual inspections to continuous, predictive maintenance, featuring automated scheduling, notifications, and auditable condition histories.