A comparative study of deep learning and Evac4BIM in building evacuation design

The integration of pedestrian simulations into the design process of a building is an inevitable step to improve its evacuation performance. This becomes particularly important during the initial design stages, when countless building configurations are explored and evaluated through interdisciplinary discussions. However, conventional simulators typically operate as stand-alone tools, necessitating several manual export and conversion steps and time-consuming runtimes. To address these challenges, recent developments have led to the integration of Building Information Modeling (BIM) with the widely-used pedestrian simulator Pathfinder in an open source tool called Evac4BIM. In this study, we explore the potential for further accelerating the generation and integration of evacuation simulation results through neural networks, trained on a comprehensive, synthetic dataset to enable real-time predictions. Subsequently, we analyze and compare the applicability of Evac4BIM and our deep learning approach in terms of practicality and accuracy.