Machine Learning-Enabled Fire Performance Diagnostics BIM Solution (FireBIM) for Building Design and Construction Compliance Check

不幸的格伦费尔火灾造成72人丧生，成为英国现代最严重的灾难之一。作为防止此类悲剧再次发生的一种方式，政府委托报告（Hackett,2018）的主要建议之一建议需要智能监管和执行系统，以促进遵守一系列法规和标准，包括批准文件B、HTM05-02、BS9999、BS9991和BB100等。几项研究（例如Noren等人，2018年；Sun，2020年）也加强了这一点，这些研究建议将消防性能诊断与建筑信息模型（BIM）相结合，作为减轻火灾的最佳方法。然而，这些领域的发展受到阻碍，因为法规和标准是用自然语言编写的，只有领域专家才能理解，通常不适合机器处理。这意味着遵守重要的安全标准容易出现错误、主观性、偷工减料，以及在纠正通过手工过程确定的不符合方面花费额外的时间。机器学习和分析领域的进步为自动化建筑安全要求合规性检查提供了机会，这是拟议项目寻求实施的。该项目旨在创建一个智能创新系统（FireBIM），作为基于bim的解决方案，由以下两个关键要素组成：平台1—防火性能符合性设计诊断平台（BIMFire-Diagnostics）；该平台通过以下方式评估建筑设计是否符合消防安全标准和法规，包括批准文件B、HTM 05-02、BS 9999、BS 9991和BB100:(i)自动诊断建议/建成设计是否符合目标法规/标准的规定。查明不符合条例和标准的领域。（iii）突出显示未认证/未担保/未保险的区域，要求详细说明以供批准。（iv）提出改进和优化的设计方案。平台2——材料和构件防火性能决策支持（BIMFire-DSS）：基于建筑法规和标准的要求，来自行业和供应商的最新信息，该平台将通过以下方式为设计师、承包商和建筑检查员提供决策支持：(i)突出建筑构件的防火等级要求。（ii）将大量有效的解决方案数字化，以达到合规要求。（iii）建议特定的材料/细节/规格以达到性能标准。（iv）对不安全的材料/组件/设计方案进行标记。发展经检验和认证的材料/部件数据库。该项目采用机器学习来训练智能模型，该模型将识别和预测建筑设计中违反建筑法规/不符合规定的情况。将实现文本挖掘，将标准/法规文档转换为机器可理解的格式，并将自动推理和推理机制应用于机器可处理的信息，以进行自动遵从性诊断。

The unfortunate Grenfell fire disaster, where 72 lives were lost, has gone down as one of the UK's worst disaster of modern time. As a way of preventing future occurrence of such tragedy, one of the key recommendations of government's commissioned report (Hackett,2018) suggests the need for intelligent systems of regulation and enforcement to facilitate compliance with arrays of regulations and standards, including Approved Document B, HTM05-02, BS9999, BS9991 and BB100, among others. This is also reinforced by several studies (e.g. Noren et al.,2018; Sun,2020) which suggested integration of fire performance diagnostics with Building Information Modelling (BIM) as optimal approach for mitigating fire disasters. However, development in such areas is hindered, as regulations and standards are written in natural languages, only comprehensible by domain experts, and not usually for machines to process. This implies that compliance with vital safety standards is prone to error, subjectivity, corner-cutting, and additional time in correcting non-conformances identified through manual processes.Advances in the field of machine learning and analytics offers opportunities for automating building safety requirement compliance check, which the proposed project seeks to implement. The project aims to create an intelligent innovative system (FireBIM), as BIM-based solution, consisting of two key elements as follows:Platform 1--Design Diagnostics Platform for Fire Performance Compliance (BIMFire-Diagnostics): This platform evaluates building design for compliance with fire safety standards and regulation, including Approved Document B, HTM 05-02, BS 9999, BS 9991 and BB100, by:(i)Automatically diagnosing proposed/as-built designs for compliance with provisions of targeted regulations/standards.(ii)Identifying areas of non-conformances with regulations and standards.(iii)Highlighting uncertified/unwarranted/uninsured areas, requiring detailing for approval.(iv)Suggesting improved and optimal design solutions.Platform 2--Materials and Components Decision Support for Fire Performance (BIMFire-DSS): Based on the requirements of building regulations and standards, up-to-date information from the industry and suppliers' information, this platform will provide decision supports for designers, contractors and building inspectors by:(i)Highlighting fire rating requirements for building elements.(ii)Digitising plethora of valid solutions for attaining compliance.(iii)Suggesting specific materials/detailing/specification for meeting performance standards.(iv)Flagging up materials/components/design solutions when unsafe for purpose.(v)Developing database of tested and certified materials/components.The project employs machine learning to train intelligent model that will identify and predict building regulations infringements/non-conformances in building designs. Text mining will be implemented to translate standards/regulation documents into machine-understandable formats, with automated reasoning and inference mechanism applied to the machine processable information for automated compliance diagnostics.