Construction progress monitoring of masonry walls using BIM-computer vision models interaction
Date
2022
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
The construction industry is expanding since it significantly contributes to the economy. At the same time, the skilled labor shortage in the United States makes it challenging to manage the high demands of the construction process. According to the Associated General Contractors of America (AGC) construction industry creates nearly $1.4 trillion worth of structures each year; therefore, to accompany the enormous budget, the shortage of skilled labor, and the limited construction durations, it is proposed to provide an automated method in the construction process that mainly utilizes Mask R-CNN models in detecting the activity of masonry walls construction. ☐ Quantitative analysis and conclusive research methodology were performed to train, test, and evaluate the developed computer vision models (i.e., YOLOv4, YOLOv4-Tiny, and Mask R-CNN) using Google Colab Pro. The models were trained and tested on forty masonry walls with more than two thousand bricks annotated that differ in pattern and location, in addition to physical and environmental conditions. The bricks’ masks were combined in a 3D matrix and converted to the surface area representing the masonry wall progress ratio. The monitoring progress percentage is translated to the Building Information Modeling (BIM) through Dynamo. Experimented data were used to validate the performance of the adopted Mask R-CNN model in different scenarios using JMP Pro 16. A structural equation model (SEM) was proposed to show the causal relationship between the variables. ☐ The developed computer vision models had a maximum accuracy of 84%, recall of 95%, mean average precision of 96%, intersection over union of 72%, and errors of 11% in detecting bricks in masonry walls. The SEM had a root mean square error of approximation of 0.1134 and a comparative fix index of 0.9674. Three hypotheses were developed and tested; the results showed that the only significant relationship on the estimated area through the Mask R-CNN model was the actual area of the wall. An IDEF0 was proposed to demonstrate the concept of construction progress monitoring using up-to-date inspection technologies. The IDEF0 process was developed to show the flow of stages during construction progress monitoring. ☐ The research integrates Dynamo with the Mask R-CNN model to detect and reflect BIM 3D models' progress by summating generated masks’ matrices. It shows the interaction of BIM and drones in masonry wall progress monitoring, data collection, and the possibility of detecting brick elements using their location when following the IDEF0 process.
Description
Keywords
Building information modeling, Computer vision, Construction monitoring, Masonry walls, Structural equation modeling, Visual programming