A Multisensor IoT and ConvLSTM Framework for Intelligent Fire Monitoring in Resilient Built Environments

Abstract

Enhancing the safety and resilience of built environments is an important component of sustainable urban development, particularly because fire incidents can result in human casualties, property loss, service disruption, and substantial economic costs. Conventional fire alarm systems primarily provide binary event detection and offer limited information regarding fire location, intensity, and evolution. This study proposes a multisensor Internet of Things (IoT) framework that integrates distributed environmental sensing, cellular communication, cloud-based monitoring, and a Convolutional Long Short-Term Memory (ConvLSTM) model for spatiotemporal fire characterization. Fire scenarios were generated using the Fire Dynamics Simulator (FDS), with measurements spatially mapped to a  sensing configuration. The completed dataset comprised 31 scenarios, including 21 single-fire and 10 dual-fire cases, yielding 3,689 temporal samples. Scenario-level partitioning was adopted to prevent information leakage between temporally correlated windows. On unseen validation scenarios, the model achieved  values of 0.834 and 0.995 for the heat release rates of the first and second fire sources, respectively. The first fire source was localized with a median error of 0.97 m and a 90th-percentile error of 2.39 m, whereas the second source achieved a median error of 3.12 m. Comparison with naive, threshold-based, Random Forest, and LSTM baselines showed that data-driven multisensor approaches substantially outperform conventional rule-based estimation, although Random Forest remained competitive for some outputs. Chamber-scale prototype tests further demonstrated the feasibility of local and remote fire notification. By integrating intelligent sensing, fire-state estimation, and remote monitoring, the proposed framework provides a technological basis for improving fire-risk awareness and operational resilience in smart and sustainable built environments. Further validation using larger scenario sets and full-scale physical environments is required before practical deployment.   Keywords: smart fire detection; sustainable built environment; building resilience; internet of things; ConvLSTM; fire localization; smart cities.