Deep Learning-Based Rapid Fire Detection in Buildings Using Acoustic Wave Travel Time Information
Year: 2025
Type: article
Abstract: Rapid fire detection in buildings is essential for effective fire emergency management. Moreover, locating the fire source provides critical evidence for subsequent investigations into the fire’s cause. Acoustic tomography temperature measurement technology can be employed to monitor building fires, as fires create anomalies in the temperature field. This study proposes a fully AI-driven tomographic method to achieve rapid temperature field inversion. In the forward problem, we simulate the time of flight (TOF) of sound waves between transceivers using ray tracing techniques with unstructured meshes, which offers higher accuracy than methods that utilize straight-line ray paths between emission and reception points. To address the inverse problem, we employ a deep learning residual neural network (ResNet) named ResNet50 as the backbone of our AI model, using TOF as input parameters and producing the acoustic velocity fields of the monitored area as output. After training, the AI model can perform high-quality predictions in 0.91 seconds for scenarios involving 169 TOF data points and 1700 velocity field meshes. The temperature fields are derived from the velocity fields using an empirical formula that describes their relationship. The fire source is identified as the location with the highest temperature. Our validation dataset indicates that the prediction errors compared to ground truth measurements are acceptable, even with high noise levels (e.g., 25%). The proposed method presents a promising performance in both accuracy and efficiency. This research serves as a preliminary investigation into the feasibility of a one-step, fully AI-driven approach for fire monitoring and detection in buildings.
Source: IEEE Access
Authors: Mengyao Sun, Cewen Liu
Institutions: China People's Police University, Stanford University
Language: English
Subfield: Safety, Risk, Reliability and Quality
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,075
OpenAlex ID: W4408791539