Autonomous Fire Resistance Evaluation

Abstract

The structural fire engineering community has been slowly evolving over the past few decades. While we continue to favor a classical stand toward evaluating fire resistance of structures through fire experimentations, a movement toward developing numerical assessment tools is on the rise. A close examination of notable works shows that the majority of these tools continue to have limited scalability, lack standardization, and thorough validation. Perhaps two of the main challenges of adopting such tools can be summed by their need for collecting true representation of response parameters (e.g., temperature-dependent material properties, etc.), and necessity to carry out resource-intensive two-stage thermo-structural analysis. In order to overcome such challenges, and in pursuit of modernizing fire resistance evaluation, this paper introduces a new generation of fire-based evaluation tools that capitalize on perception rather than imitation. More specifically, this paper explores how automation and cognition (A&C), realized through machine learning (ML), can be applied to comprehend structural behavior under fire conditions. To achieve this goal, genetic programing (GP) and computer vision (CV) are used to assess fire response of structural members. The outcome of this study demonstrates that A&C can accurately evaluate fire resistance and identify damage/spalling magnitude in reinforced concrete (RC) structures; thus, paving the way to realize autonomous fire-based evaluation and inspection.