Scholarlink

<p>Bridge fires pose a significant threat to the safety, serviceability, and resilience of transportation infrastructure. Conventional approaches for assessing bridge performance under fire, including experimental testing, finite element analysis (FEA), computational fluid dynamics (CFD), and simplified analytical methods, provide valuable insights but involve trade-offs between modelling fidelity, computational cost, and practical applicability. These limitations have stimulated growing interest in machine learning (ML) as a complementary tool for rapid and data-driven bridge fire performance assessment. This paper presents a critical review of recent advances in ML for assessing the fire performance of bridges. Existing studies are examined with emphasis on the underlying databases, input features, prediction capabilities, and reported model performance. Current challenges, including the limited availability of bridge fire databases, model generalization, interpretability, uncertainty quantification, and practical implementation, are critically discussed. Emerging developments, such as explainable artificial intelligence, transfer learning, and physics-informed machine learning, are also reviewed for their potential to improve the reliability and engineering applicability of future prediction models. Based on the current state of knowledge, the principal research gaps are identified, and future research directions are proposed to support the development of robust, interpretable, and physics-consistent ML frameworks for bridge fire performance assessment. The review provides a consolidated reference for researchers and practicing engineers working in bridge fire engineering.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine Learning for Bridge Fire Performance Assessment: Methods, Challenges, and Future Directions

  • 期刊论文

  • 发表日期：

    2026年7月30日

查看全文
  • Paul O. Awoyera,
  • Faisal Mukhtar

摘要

Bridge fires pose a significant threat to the safety, serviceability, and resilience of transportation infrastructure. Conventional approaches for assessing bridge performance under fire, including experimental testing, finite element analysis (FEA), computational fluid dynamics (CFD), and simplified analytical methods, provide valuable insights but involve trade-offs between modelling fidelity, computational cost, and practical applicability. These limitations have stimulated growing interest in machine learning (ML) as a complementary tool for rapid and data-driven bridge fire performance assessment. This paper presents a critical review of recent advances in ML for assessing the fire performance of bridges. Existing studies are examined with emphasis on the underlying databases, input features, prediction capabilities, and reported model performance. Current challenges, including the limited availability of bridge fire databases, model generalization, interpretability, uncertainty quantification, and practical implementation, are critically discussed. Emerging developments, such as explainable artificial intelligence, transfer learning, and physics-informed machine learning, are also reviewed for their potential to improve the reliability and engineering applicability of future prediction models. Based on the current state of knowledge, the principal research gaps are identified, and future research directions are proposed to support the development of robust, interpretable, and physics-consistent ML frameworks for bridge fire performance assessment. The review provides a consolidated reference for researchers and practicing engineers working in bridge fire engineering.