Accurate monitoring of the indoor microclimate, including air temperature, relative humidity, and dew point, is crucial for preserving historic buildings and ensuring sustainable conservation. Similarly, forecasting indoor environmental conditions is essential for both improving building performance and making informed decisions about the conservation of historic structures. This study addresses the research gap in predicting the indoor microclimate of structures by developing and evaluating the performance of three Machine Learning (ML) methods for forecasting indoor microclimate in the Kelso House, a low-thermal mass historic building in San Antonio, Texas, USA. From April 2022 to January 2023, indoor and outdoor conditions (air temperature, relative humidity, and dew point) were recorded every 15 minutes using data loggers. The collected datasets were used to train and test the different ML algorithms, namely Multi-Layer Perceptron (MLP), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost). Various predictive models were developed to forecast the 15-minute-ahead values of indoor air temperature, relative humidity, and dew point within the case study building. The accuracy and computational efficiency of the models were evaluated using metrics such as mean absolute error and convergence time. Results showed that MLP and SVR achieved the highest accuracy and effectively detected abrupt fluctuations in temperature and relative humidity, outperforming XGBoost. However, XGBoost demonstrated exceptional computational efficiency in terms of convergence time, making it suitable for forecasting applications as well. This investigation highlights the potential of the developed ML-driven models for accurately forecasting indoor microclimate. Additionally, the proposed methodology is adaptable and can be applied to a wide range of construction across different climate zones globally. By enabling the prediction of indoor environmental conditions critical to historic preservation, this study provides valuable insights to assist experts in making informed decisions about the conservation of historic structures.

This study develops a data-driven framework to predict and analyze hygroscopic deformation in bilayer wood samples under controlled relative humidity (RH) conditions. The anisotropic nature of wood leads to hygroscopic expansion, a phenomenon integral to wood bending techniques. However, accurate prediction of wood's hygroscopicity remains challenging due to the limitations of existing prediction tools. The research integrates experimental monitoring of hygroscopic deformation with machine learning models, leveraging material and environmental features to predict deformation curvature. This framework leverages experimental data and computational tools to provide an effective prediction tool for wood’s hygroscopic deformation. The findings offer robust alternatives for mass timber construction and innovative applications in digital fabrication.

Interoperability, the ability of different software applications to exchange information meaningfully and usefully, is a major concern in digital modeling. This research investigates the epistemological dimensions of interoperability in constructing digital models of existing buildings, specifically focusing on Revit and Rhino—software applications with distinct ontological and epistemological frameworks. Traditionally viewed as a set of technical challenges, interoperability is repositioned in this study as a productive “friction” capable of generating architectural knowledge. In this sense, our title refers to “knowledge effects,” by which we mean novel interpretations, spatial understandings, and conceptual reframings that emerge through the processes of interoperability.

Using a case study of Marcel Breuer’s Alcuin Library, the research examines translations between a Revit (BIM) model and a Rhino (non-BIM) model, analyzing frictional losses (e.g., geometric distortions, semantic flattening) and gains (e.g., novel visualizations and conceptual reframings). The findings suggest that, while information loss is inevitable in translation, friction enables new opportunities for inference-drawing, capable of revealing latent aspects of architecture.

The study applies Mauricio Suárez’s inferential conception of representation, demonstrating how interoperability alters what Suárez refers to as “representational force” and “inferential capacity” in ways that inform architectural understanding. By framing friction as both a practical limitation and a theoretical mechanism for knowledge production, the paper challenges the assumption that interoperability must achieve seamless data exchange. Instead, it advocates for leveraging the distinct strengths of different software platforms in architectural modeling.

This research contributes to the discourse on building information modeling (BIM) for existing buildings, proposing an approach that values epistemological depth over technical efficiency. It suggests implications for professional practice, pedagogy, and future research, emphasizing the generative potential of friction in developing architectural representation and analysis.