Abstract:Fires in high-voltage cable ducts on bridges are characterized by rapid spread and difficulty in extinguishing, which can lead to significant economic losses and pose threats to the safety of inspection personnel. Early identification of the fire's location is crucial for effective rescue operations. Therefore, researching the intelligent identification and prediction of ignition points in the early stages of fires in bridge cable ducts is of great importance. A simulation model for the early spread of smoke in bridge box girder cable ducts was established using PyroSim analysis software, yielding the CO diffusion characteristics. An artificial neural network (ANN) model was designed and trained for data stratification and ignition point identification at each layer. Experiments on ignition point identification were conducted based on simulation data, and an ignition point identification system was designed and test-ed on-site in a simulated cable duct. The results showed that: (1) In the ignition point identification experiments based on simulation data, the ANN model exhibited a maximum error of 0.98 meters and a minimum error of -0.32 meters for identifying the ignition point in a 50-meter, single-layer cable duct. For the smoldering ignition point in a three-layer cable duct, the maximum error was 1.53 meters and the minimum error was -1.26 meters. (2) In the on-site testing of the ignition point identification system, the maximum error was 0.68 meters, and the minimum error was -0.27 meters. This level of precision meets the requirements for intelligent identification and prediction of ignition points in the early stages of fires in bridge box girder high-voltage cable ducts. The findings hold potential for improving the accuracy and timeliness of fire alarm systems in practical applications for bridge box girder cable ducts.
