| Author: | Ye, Zhongnan |
| Title: | Dynamic rescue in flames : real-time structural fire response prediction and adaptive indoor rescue path planning |
| Advisors: | Hsu, Shu-chien Mark (CEE) Wei, Hsi-hsien (BRE) Li, Heng (BRE) Usmani, Asif (BEEE) |
| Degree: | Ph.D. |
| Year: | 2023 |
| Subject: | Fire extinction Fire prevention Fire prevention -- Data processing Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Civil and Environmental Engineering |
| Pages: | xxvi, 145 pages : color illustrations |
| Language: | English |
| Abstract: | Fire emergencies in buildings present substantial challenges to firefighters and rescue teams due to their rapidly evolving nature and potential compromise of structural safety. In such high-risk environments, timely and accurate information about the building’s structural safety and efficient rescue path planning is crucial for the success of firefighting operations and the safety of both occupants and firefighters. However, the dynamic behavior of fire propagation and the intricate processes involved in fire-induced structural degradation present substantial challenges in predicting structural fire responses and determining corresponding safe and efficient rescue paths. Traditional methods, such as finite element models (FEM) and computational fluid dynamics (CFD), may provide accurate predictions of structural responses to fire, but their computational complexity and time-consuming nature make them unsuitable for real-time applications. Additionally, existing path planning algorithms may not be well-adapted to the dynamic and uncertain conditions of building fire environments, potentially leading to suboptimal or even dangerous route recommendations for firefighters. While significant progress has been made in the field of machine learning (ML), computational modeling, and numerical simulation techniques for fire emergencies, efficient and effective solutions remain limited for real-time structural fire response prediction and adaptive fire rescue path planning. This study aims to facilitate efficient and safe fire rescue operations under time-varying building fire environments by exploring the dynamics of structural fire responses and corresponding fire rescue paths with higher efficiency but lower risk for firefighters. A data-driven approach is employed to address the challenges of dynamic behaviors of fire propagation and fire-induced structural degradation, by integrating data analytics, computational simulation, and mathematical modeling within the context of building fire emergencies. The primary research objectives are twofold: (1) to develop an ML-based model for real-time prediction of structural fire responses by integrating FEM and CFD databases, and (2) to develop an adaptive indoor fire rescue path planning model that considers the time-varying structural safety conditions. The models developed through this research are designed to provide reliable and timely predictions of structural responses and suggest efficient and safe rescue paths, ultimately contributing to the safety and effectiveness of firefighting operations in dynamic building fire environments. In the realm of real-time structural fire response prediction, a FEM-based ML framework is developed to predict structural displacement using temperature field data as input. This framework combines the precision of FEM with the efficiency of ML techniques to enable real-time predictions of structural responses during fire emergencies. Utilizing a virtual case of a single-story, two-bay, 3D steel frame structure, four distinct ML models are trained based on a FEM numerical database of structural responses under various fire scenarios derived from parametric fire curves. Among these models, the Random Forest (RF) and Gradient Boosting (GB) models demonstrate superior performance in terms of predictive accuracy and model robustness. The Coefficient of Determination (R2) value of the predictive mid-span displacement reaches up to 0.99 when 1000 fire scenarios are included in the training dataset. Additionally, all models exhibit robustness against noise in temperature data when the signal-to-noise ratio exceeds 15. The developed FEM-based ML framework shows considerable promise for real-time structural response prediction during building fires. To enhance the realism of the fire scenarios in the numerical database, the FEM-based ML structural response prediction framework is augmented by incorporating CFD for fire simulation. The integration of CFD and FEM enables the CFD/FEM-based ML framework to capture the intricate and dynamic interactions more comprehensively between fire-induced temperature fields and structural deformation, resulting in more precise and reliable predictions of the structural behavior during fire emergencies. With numerical cases of an 8m×8m×0.6m steel roof structure under 1200 virtual fire scenarios, various ML models are developed to predict the vertical displacement in real-time and in the near future (e.g., 10 seconds) using temperature field data as input. The RF and GB models demonstrate considerable effectiveness in predicting real-time displacement, with R2 values reaching up to 0.97 when the number of fire scenarios in the training dataset is 200 or more. In contrast, the Long-Short Term Memory (LSTM) model displays competitive performance in predicting displacement for the subsequent 10 seconds, with a maximum Mean Squared Error (MSE) less than 0.05 when the epoch exceeds 200. The proposed CFD/FEM-based framework can offer reliable and precise predictions of structural responses in a real-time manner, aiding firefighting teams in making informed decisions and optimizing their response strategies. Building upon reliable and timely predictions of structural fire responses, an adaptive path planning model has been developed to recommend efficient and safe rescue paths under dynamic building fire environments. The model considers time-varying structural safety conditions as a dynamic risk map, optimizing rescue paths to minimize overall cost in terms of time, distance, and risk. A dynamic grid-based search algorithm is specifically developed to generate a sequence of path segments connecting the firefighting agent’s current location to the destination while considering time-varying structural safety conditions incorporated into a dynamic risk map. The model is tested using a case of a single-story office building, measuring 30m×30m, and involving 50 random fire scenarios. The rescue paths suggested by the proposed model are found to be adaptive to the updated risk map of the environment, attempting to avoid paths with a high risk of structural component failure or collapse. In comparison to the traditional non-adaptive path suggested by the Dijkstra algorithm, the proposed adaptive model results in an average of 12.94% longer travel time. However, the adaptive model demonstrates a significant advantage in reducing the time duration that firefighters are exposed to high-risk areas with severe damage or worse. Specifically, the rescue paths suggested by the adaptive model decrease such exposure by 45.45% compared to those suggested by the non-adaptive model, which is of critical importance in ensuring the safety of firefighters during fire rescue operations. Furthermore, with a noise level of 20% in the risk map, the proposed adaptive model remains robust enough to suggest rescue paths with substantially lower risk exposure durations than the non-adaptive model. The proposed model helps firefighters in fire rescue tasks by recommending optimal paths that adapt to dynamic building fire environments, addressing both efficiency and safety concerns. The main contribution of this thesis lies in developing data-driven methodologies for real-time structural fire response prediction and adaptive fire rescue path planning in dynamic building fire environments. By providing robust and efficient frameworks and models, this research addresses critical gaps in the available tools and methodologies for firefighting teams during building fire emergencies. Furthermore, the integration of FEM, CFD, and ML techniques offers valuable insights and advancements for future research in structural fire safety, smart firefighting, and emergency response planning. Ultimately, the developed frameworks and models serve as a prototype for practical tools to enhance the safety and effectiveness of firefighting operations, reducing the risks to firefighters during building fires. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/12771