Mining implicit 3D modeling patterns from unstructured temporal BIM log text data

Abstract

Abstract Building information modeling is instrumental in documenting design, enhancing customer experience, and improving product functionality in capital projects. However, good building models do not happen by accident, but rather as a result of a managed process that involves several participants from different disciplines and backgrounds. Effective management of this process requires an ability to closely monitor the modeling process and correctly measure modelers' performance. Nevertheless, existing methods of performance monitoring in building design practices lack an objective measurement system to quantify modeling progress. The widespread utilization of Building Information Modeling (BIM) tools presents a unique opportunity to retrieve granular design process data and conduct accurate performance measurements. As a building's 3D model is gradually developed, model generation software packages, such as Autodesk Revit, automatically create log files that record design activities. This paper investigates what information these log files contain and how one can extract and further analyze the information to provide insight into the design modeling process. The specific objectives of this study were to: (1) investigate the presence of implicit patterns in 3-D design log files; and (2) to empirically characterize the performance of modelers based on the time it takes them to execute similar modeling tasks. To fulfill these objectives, design log files provided by an international architecture and design firm were analyzed. Using a tailored text file parser, user-model interaction data including modeler characteristics, command type, and command time were extracted from the journal files. To identify implicit command execution patterns, a sequence mining algorithm based on Generalized Suffix Trees (GST) was implemented. It was shown that there is a statistically significant difference between the average time it takes modelers to execute each command sequence. This study extends the existing knowledge by proposing a novel methodology to extract meaningful patterns from time-stamped unstructured design log data. This research contributes to the state of practice by providing a better understanding of information embedded in design log files.