Application of Artificial Intelligence (AI) In Electrical Quantity Takeoff From Housing Construction Drawings

Abstract

Quantity takeoff for electrical systems in housing construction projects is typically performed manually, which is prone to errors. This led to the idea of applying artificial intelligence (AI) technology, especially convolutional neural network (CNN) models, to enhance the efficiency and accuracy of the quantity takeoff process. Among various CNN models, Mask R-CNN, YOLO,

and U-Net have been increasingly used for object detection and classification in images. This research presents a comparative study on the application of three AI models Mask R-CNN, YOLOv9, and U-Net for extracting room areas and identifying number of electrical devices including light fixtures, switches, and receptacles from architectural and electrical construction drawings of residential buildings. The models were trained using a dataset of 100 to 300 construction drawings, with variations in training epochs, image resolution, and the number of training images of 100 to 300 images. The results showed that, when trained with 300 images at a resolution of 1700×1200 pixels and 80 epochs, YOLOv9 achieved the highest accuracy (90%) in detecting electrical devices such as lights, switches, and receptacles. Meanwhile, U-Net outperformed others in room area segmentation, achieving an Intersection over Union (IOU) score of 83%.