| Author: | Zhang, Tianhang |
| Title: | Smart firefighting framework and building fire forecast driven by artificial intelligence |
| Advisors: | Huang, Xinyan (BEEE) Xiao, Fu Linda (BEEE) |
| Degree: | Ph.D. |
| Year: | 2023 |
| Subject: | Fire -- Data processing Artificial intelligence Fire extinction Fire management Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Building Environment and Energy Engineering |
| Pages: | ix, 96 pages : color illustrations |
| Language: | English |
| Abstract: | The mastery and utilization of fire signify the onset of the Stone Age for humanity, making it one of the earliest tools and a significant milestone in the progress of human civilization. However, simultaneously, uncontrolled burning behavior, fires, have also inflicted immense pain and suffering upon humanity. Traditional empirical fire models have limited perception capabilities regarding the fire environment in buildings, which places firefighters at great risk in unknown fire scenarios. Over the past decade, digitalization and artificial intelligence (AI) enable new smart techniques in the building and construction area. The applications of AI in fire detection, risk assessment, and data-driven fire forecast are emerging. The thesis is structured in a manuscript-style format, beginning with an introduction to the research background and motivation in the first chapter. The following chapters each stand alone as an independent paper, which have either been published or submitted to a journal publication. In the final chapter, the overall outcomes are summarized and potential avenues for future research are outlined. Chapter 1 introduces the research background and motivation. As a result of rapid urbanization and globalization, building density, complexity, and population have all dramatically increased, giving rise to a heightened potential for fire risks. The cost of fires on society is estimated to be between 1-2% of global annual GDP and is strongly correlated with both per capita GDP and the Human Development Index. The impact of fires extends beyond economic loss, as they can also result in devastating casualties. The lack of available fire information and the unpredictability of fire development pose serious threats to both individuals trapped in fires and firefighters. Moreover, fire commanders may struggle to make effective and informed firefighting decisions without access to information about the fire's actual state. Consequently, there is a pressing need for a real-time method to forecast and identify building fire scenarios, enabling a fast response time and improving the safety levels in firefighting practices. Chapter 2 reviews the state-of-the-art research progress on the forecasting methods of fire development and critical events in building fire, e.g., Flashover. Flashover marks the ultimate signal of untenable conditions for people to survive in fire, so the forecast of flashover is of great significance in firefighting and emergency response. Studies over the last century have established a physical framework to understand the flashover phenomenon, but current practices of flashover forecast are still empirical and intuitive. This chapter first induces the mechanisms of flashover under different building fire scenarios and the classical empirical correlations. By revisiting past flashover experiments, the critical onset conditions, including ceiling temperature, heat flux, smoke layer, and flame behaviors, are summarized and discussed. Then, we review the methodologies of flashover forecast from the classical Kawagoe’s law to recent data-driven and machine-learning methods. Finally, a framework for applying Artificial Intelligence of Things (AIoT) to forecast flashover is proposed for future smart firefighting, and the technical challenges are discussed in detail. This review aims to bridge the gap between the conventional static and causal understanding of flashover and the emerging dynamic and AIoT-driven fire forecast. Chapter 3 establishes a framework of Artificial Intelligent Digital Twin of Fire (AID-Fire), which can identify building fire information in real-time. The proposed fire digital twin consists of four main parts: Internet of Things (IoT) sensor network, cloud server, Artificial intelligence (AI) Engine, and User Interface (UI), which are applied to collect and transfer, manage, process, and display the building fire information, respectively. A large numerical database containing 533 fire scenarios with varying heat release rates (HRR), fire positions, and fire source numbers is established to train a Convolutional Long-Short Term Memory (Conv-LSTM) neural network. The proposed fire digital twin is demonstrated and validated by laboratory-scale fire tests. Results show that the AI engine performed well in identifying the fire information by learning the spatial-temporal features of the temperature data with a relative error of less than 15% and a delay time of around 1 s. Moreover, critical fire events, including fire development and fire spread, can also be accurately displayed in the fire digital twin. This proposed framework provides valuable information which can be further processed to predict fire development. Chapter 4 proposes an artificial intelligence (AI) system to fast forecast the compartment fire development and flashover in advance based on a temperature sensor network and a deep-learning algorithm. This fire-forecast system is demonstrated in a 1/5 scale compartment with various ventilation conditions and fuel loads. After training 21 reduced-scale compartment tests, the deep learning model can well identify the fire development inside the compartment and predict the temperature 30 s in advance with relative errors of less than 10%. The flashover can be predicted with a 20-s lead time, and the forecast capacity and accuracy can be further improved with additional test data for training. The AI-forecast model performs well for fires with different fuel types and ventilation conditions and has the potential to be applied to fire scenarios with wider conditions. This chapter demonstrates the real-time building fire forecast based on Internet of Things (IoT) sensors and AI systems that can help future smart firefighting applications. Chapter 5 proposes a novel framework for the fusion of fire images and sensor data to forecast the backdraft explosion in a building fire based on the Transformer algorithm. The backdraft forecast model is demonstrated with a large-scale fire test in a 5.5 (l) × 2.4 (w) × 2.4 (h) m3 chamber. After training 2674 fire scenarios with various fire intensities and shooting angles, the Fusion-Transformer model can reasonably identify the fire mode with an overall accuracy of 84 %. Meanwhile, the backdraft can be predicted with the Mean Absolute Error (MAE) of 1.6 s in occurrence time and 0.14 m in explosion range. Compared with the single modal model, the vision-sensor fusion transformer model can reduce the prediction error of backdraft occurrence time and explosion range by 82.6% and 59.1%, respectively. This research has bridged the gap in the application of data fusion methods in fire prediction and represents a significant stride toward the development of universal Artificial Intelligence-based smart fire modeling. Chapter 6 summarizes the overall outcomes of smart firefighting system for building fire environment in current stage. According to the present findings, the challenges the researcher need to overcome in the future are also discussed. |
| 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/200/12726