Acoustic Based Fire Event Detection System in Underground Utility Tunnels

• Published in: Fire (Basel, Switzerland) Vol. 6; no. 5; p. 211
• Main Authors: Lee, Byung-Jin, Lee, Mi-Suk, Jung, Woo-Sug
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.05.2023
• Subjects: acoustic sensing; Acoustics; anomaly detection; Artificial neural networks; Automobile safety; Cables; Communication; Corrosion; Deep learning; Dehumidification; Electric cables; Electric sparks; Fire prevention; Fire protection; Fire safety; Humidity; Infrastructure; Insulation; Methods; Neural networks; Noise reduction; Safety and security measures; Sensors; Sound; Tunnels; Underground acoustics; Underground utilities; underground utility tunnel; Ventilation; Waterproofing; Wiring; United States
• DOI: 10.3390/fire6050211
• ISSN: 2571-6255

Underground utility tunnels (UUTs) are convenient for the integrated management of various infrastructure facilities. They ensure effective control of underground facilities and reduce occupied space. However, aging UUTs require effective management and preventive measures for fire safety. The fundamental problems in operating UUTs are the frequent occurrence of mold, corrosion, and damage caused to finishing materials owing to inadequate waterproofing, dehumidification, and ventilation facilities, which result in corrosion-related electrical leakage in wiring and cables. To prevent this, an abnormal sound detection technology is developed in this study based on acoustic sensing. An acoustic sensor is used to detect electric sparks in the moldy environments of UUTs using a system to collect and analyze the sound generated in the UUTs. We targeted the sound that had the highest impact on detecting electric sparks and performed U-Net-based noise reduction and two-dimensional convolutional neural network-based abnormal sound detection. A mock experiment was conducted to verify the performance of the proposed model. The results indicated that local and spatial features could capture the internal characteristics of both abnormal and normal sounds. The superior performance of the proposed model verified that the local and spatial features of electric sparks are crucial for detecting abnormal sounds.