Automated Rule Interpretation For Automated Rule Checking

Design review is an important step to ensure construction quality and safety.Since traditional manual review methods are inefficient,error-prone,and require high experience,an intelligent method of automated rule checking(ARC)has become a key factor to improve design quality and efficiency.Rule interpretation,which focuses on converting textual rules into a computer-processable format,is the key step in ARC.However,existing rule interpretation methods still rely on manual or traditional pattern matching methods,which face the challenges of large labor input,the limited scope of application,and high maintenance costs.Thus,it is still difficult to achieve efficient rule interpretation of a large number of complex regulations and ARC.To address this problem,this paper utilizes deep learning,natural language processing,formal grammar,and other technologies to propose a novel method for automated rule interpretation.First,by considering the characteristics of regulatory text and building information,semantic labels that can represent the semantic roles and relations of regulatory text are proposed,and a syntax tree structure that can represent the hierarchies and relations of the semantic elements is also proposed.Meanwhile,regulatory documents are automatically crawled,collected,and processed,which yields611 regulatory sentences.These regulatory sentences are manually annotated with semantic labels and thus established the first regulation data in the ARC area.Second,a deep learning model is developed according to the proposed semantic labels,and the regulation dataset is used for model training to realize automated semantic labeling.Then,the context-free grammar(CFG)which has high expressiveness is utilized to develop a syntactic parsing method,for automatically parsing labeled sentences of various complexity into the syntax tree.Finally,taking the Revit BIM model as an example,an automated rule generation method is developed to convert the syntax tree into an executable rule,and actual projects are selected for ARC to verify the proposed method.Results show that the proposed automated rule interpretation method is significantly better than existing methods in both performance and application scope.Specifically,this research establishes a regulatory text labeling method for automated rule interpretation and opens the first large-scale labeling dataset;proposes an automated rule interpretation method that integrates deep learning and CFG,where the semantic labeling reaches an F1 of 86.2%,simple sentence’s parsing has a high accuracy of 99.57%,and complex sentence parsing is supported and achieving an accuracy of 95.26%.This research also supports the improvement of the efficiency of ARC and regulatory text knowledge extraction and reuse.