Smart performance-based design for building fire safety: Prediction of smoke motion via AI

Abstract

The performance-based design (PBD) has been widely adopted for building fire safety over the last three decades, but it requires a laborious and costly process of design and approval. This work presents a smart framework for fire-engineering PBD to predict the smoke motion and the Available Safe Egress Time (ASET) in the atrium by Artificial Intelligence (AI). A CFD database of visibility profile in atrium fires is established, including various fire scenarios, atrium volumes, and ventilation conditions. After the database is trained with the transposed convolutional neural network (TCNN), the AI model can accurately predict the smoke visibility profile and ASET in the atrium fire. Compared to conventional CFD-based PBD by professional fire engineers, AI method provides more consistent and reliable results within a much shorter time. This research verified the feasibility of using AI in fire-engineering PBD, which may reduce the time and cost in creating a fire-safety built environment.