Text-to-video diffusion models have made significant progress. However, there is still a lack of dedicated research on generating fire scene videos with physical realism and visual fidelity. To address this gap, we propose text-to-video fire (T2VFire) scene generation. T2VFire uses GPT-4o as the core engine, which is integrated with an external fire-related knowledge base and a retrieval-augmented generation (RAG) mechanism that can be dynamically updated based on prompts. With the support of this knowledge, the system first expands the user’s initial text description and generates a keyframe image. Then, through iterative prompt optimization, it guides a pretrained video diffusion model to generate fire scene videos with physical consistency. Experimental results show that T2VFire improves upon the physical consistency and visual realism of fire scene videos generated by current video generation models. This method provides a solid foundation for future smart firefighting and digital twin systems in building fire safety management.

Since the 1970s, researchers have developed semi-empirical models to describe fire smoke movement inside buildings, but there are three major issues. Firstly, several plume models are available to estimate the smoke production rate and the capacity of smoke extraction fans, but their discrepancy or accuracy is unclear. Secondly, the phenomenon of stratification affects the vertical transportation of smoke, and influences the activation time of the detectors and the efficiency of the smoke extraction system. A stratification model is available in the literature to calculate the maximum height that smoke can rise, but it cannot cover all design scenarios. Thirdly, the size of the smoke reservoir has been regulated in fire regulation. The regulation does not consider the factors that strongly affect the movement of smoke in the reservoir, such as the ceiling height, reservoir shape, smoke temperature, etc. These models are difficult to directly apply to a practical design project, and some clauses of the fire regulation do not address the requirements correctly and become a hurdle of design. This paper depicts the cases encountered during the design over the past decades and provides detailed processes of solving these issues. The approach of the design process demonstrates how fire engineers further develop the fire models and fill the gap between research and engineering practice. This paper systematically examines fire smoke models for the plume, vertical transportation of the smoke, the ceiling jet, and smoke spreading underneath the flat ceiling, and provides practical solutions for each of the smoke development stages.

In recent years, fires in tall buildings have become more frequent, which costs billions of dollars each year and the loss of many human lives. The façade fire in the Grenfell tower made the structure uninhabitable, and the collapse of the three World Trade Center (WTC) towers is the total structural failure caused by fire. Despite such events, no well-defined methodology exists to reconstruct both fire and structural behaviors and carrys out the forensic investigation of a building fire. This Part Ⅰ paper collects the evidence of the Plasco Building fire and generates a coherent timeline to reconstruct the fire processes. The vertical and horizontal fire spread of the building is reconstructed using computational fluid dynamics (CFD) fire modeling and calibrated against the evidence library. The spatio-temporal temperature history from the fire modeling provides realistic fire scenarios to simulate the structural response. The fire simulation results are used as boundary conditions to be transferred to a finite element analysis tool for a detailed structural analysis to determine the likely collapse mechanism of the Plasco Building in Part Ⅱ. The methodology presented in this paper to reconstruct the fire can also guide the structural fire safety engineers to improve the building fire-safety and life-safety strategies.

The extreme temperature induced by fire and hot toxic smokes in tunnels threaten the trapped personnel and firefighters. To alleviate the potential casualties, fast while reasonable decisions should be made for rescuing, based on the timely prediction of fire development in tunnels. This paper targets to achieve a real-time prediction (within 1 s) of the spatial-temporal temperature distribution inside the numerical tunnel model by using artificial intelligence (AI) methods. A CFD database of 100 simulated tunnel fire scenarios under various fire location, fire size, and ventilation condition is established. The proposed AI model combines a Long Short-term Memory (LSTM) model and a Transpose Convolution Neural Network (TCNN). The real-time ceiling temperature profile and thousands of temperature-field images are used as the training input and output. Results show that the predicted temperature field 60 s in advance achieves a high accuracy of around 97%. Also, the AI model can quickly identify the critical temperature field for safe evacuation (i.e., a critical event) and guide emergency responses and firefighting activities. This study demonstrates the promising prospects of AI-based fire forecasts and smart firefighting in tunnel spaces.