A Data-Driven Framework for Machine Learning- Based Recommendation of Climate-Responsive Building Envelope Assemblies

Introduction: The building envelope plays a critical role in regulating thermal performance and energy consumption in buildings, particularly in regions with diverse climatic conditions. In practice, however, the selection of envelope construction assemblies such as multilayer wall, roof, and floor systems is often based on prescriptive standards, conventional experience, or time-consuming simulation processes. With the increasing availability of simulation data and material performance information, there is an opportunity to transform conventional envelope design into a data-driven and intelligent decision-support process. Problem Statement: The main research problem addressed in this study is the lack of machine-learning-based frameworks capable of supporting early-stage building envelope design at the level of multilayer construction assemblies. While many existing studies apply artificial intelligence to predict building energy consumption or optimize operational systems, few focus on translating simulation-informed envelope detailing into structured datasets suitable for machine learning. This gap limits the use of AI in guiding practical construction decisions during the conceptual design phase. Methodology: To address this challenge, the research develops a structured database derived from simulation-based evaluations of building envelope assemblies across multiple Iranian cities. The database includes detailed layer-by-layer material definitions along with thermophysical properties and climate-related performance indicators. A Python-based computational framework was created to process the multilayer construction data, aggregate thermophysical characteristics into numerical features, and generate a machine-learning-ready dataset. This framework represents a transition from a rule-based selection tool toward a supervised learning model intended to recommend climate-appropriate envelope assemblies. Results and Findings: The results demonstrate that complex construction detailing can be successfully translated into structured numerical representations without losing thermophysical meaning. The study establishes the methodological foundation for an AI-driven envelope recommendation system that can support early-stage design decisions, reduce dependence on repetitive simulations, and bridge the gap between building performance knowledge and intelligent design assistance. These findings highlight the potential of integrating building physics, simulation data, and machine learning to advance climate-responsive architectural design.

کلیدواژه ها:

Building Envelope Design ، Machine Learning in Architecture ، Climate-Responsive Construction ، Building Performance Simulation ، Data-Driven Design Support