AI-Driven Predictive Maintenance in Infrastructure and Facilities Management

This study addresses the critical challenge of transforming traditional reactive maintenance approaches within infrastructure and facilities management into proactive, data-driven strategies leveraging advancements in artificial intelligence. Conventional maintenance, often reliant on fixed schedules or post-failure interventions, falls short in mitigating unexpected downtimes and escalating costs. To overcome these limitations, this research deploys multiple machine learning algorithms, namely, K-Nearest Neighbors (KNN), Support Vector Classifier (SVC), Random Forest Classifier (RFC), and Extreme Gradient Boosting (XGBoost), applied to a comprehensive synthetic predictive maintenance dataset. This dataset encapsulates key operational metrics including temperature, torque, rotational speed, and tool wear across diverse failure modes. The comparative analysis reveals that XGBoost substantially outperforms alternative models, achieving a remarkable accuracy of 98.9% in multiclass failure prediction, supported by an AUC nearing 1.0 and F1 scores above 0.98 in both validation and test sets. RFC and SVC closely follow, each delivering precision and recall rates exceeding 95%. Notably, KNN provides rapid inference, facilitating real-time applications despite slightly lower accuracy metrics (~96.6%). Advanced preprocessing techniques, such as feature scaling, label encoding, and synthetic minority oversampling (SMOTE), enhanced model robustness amid inherent class imbalances. Critical features, including torque and tool wear, exhibited pronounced predictive importance, aligning with known mechanical failure signatures. The findings underscore AI’s potential to revolutionize maintenance by offering granular failure diagnostics and enabling timely interventions, thus significantly reducing operational costs and preventing catastrophic infrastructure failures. This research advances the state-of-the-art by integrating interpretability, cross-model benchmarking, and practical scalability considerations, positioning AI-driven predictive maintenance as a cornerstone of modern infrastructure resilience and sustainability.