AI-Enabled Predictive Maintenance for Fire Alarm and Smoke Management Systems: A Systematic Review of Models

Abstract

This study presented a quantitative systematic review of AI-enabled predictive maintenance models applied to fire alarm and smoke management systems, with the objective of evaluating their effectiveness in improving system reliability, operational efficiency, and fault detection accuracy. The analysis synthesized data from 62 peer-reviewed studies covering diverse infrastructure environments, including hospitals, airports, commercial buildings, industrial facilities, and high-rise structures. The findings indicated that predictive maintenance models achieved a high overall mean accuracy of 91.6%, with deep learning and ensemble models outperforming traditional approaches by reaching accuracy levels of 94.2% and 93.5%, respectively. The results further demonstrated a significant reduction in false alarm rates by an average of 37.4%, alongside improvements in detection sensitivity, with recall values increasing to 90.8%. The study also revealed that dataset size and sensor integration significantly influenced model performance, as large datasets exceeding 200,000 observations achieved mean accuracy levels of 93.8%, compared to 87.2% in smaller datasets. Multi-sensor IoT-based systems demonstrated superior anomaly detection performance at 92.9%, highlighting the importance of integrated data environments. Detection latency improved by approximately 24.3%, indicating faster system responsiveness, while overall system reliability scores increased by 28.6% following predictive maintenance implementation. Statistical analysis confirmed that these improvements were significant, with large effect sizes observed in false alarm reduction and fault detection performance. Subgroup analysis showed variability across infrastructure types, with commercial buildings achieving the highest accuracy at 94.1%, while more complex environments such as industrial facilities reported lower performance at 88.9%. Hybrid models demonstrated consistent improvements across all environments, particularly in dynamic conditions. The findings also indicated strong positive correlations between dataset size, sensor integration, and predictive accuracy. Overall, the study provided comprehensive quantitative evidence that AI-enabled predictive maintenance significantly enhances the performance and reliability of fire safety systems, supporting its application as an effective and scalable solution for safety-critical infrastructure management.