Transfer Learning Enabled Process Recognition for Module Installation of High-rise Modular Buildings

Abstract

High-rise modular buildings (HMB), based on the advanced approach of modular construction, have gained momentum in practice due to their offered benefits in accelerated construction, improved quality, reduced health and safety risks, and enhanced productivity. Modular construction with standard design of modules and repetitive processes of module installation is in favor of the development of construction automation. As module installation is one of the critical activities in the delivery of HMBs, it is important to recognize the module installation process automatically so as to facilitate automation in modular construction. However, there is no detailed phase-division of module installation process. Also, little research has been carried out on intelligent process recognition for module installation due to the limited amount of images of real-life projects. To fill in the knowledge gaps, this paper aims to build a transfer learning enabled process recognition model using convolutional neural network (CNN) for module installation of HMBs. The study first divided the module installation process into three stages: hooking, lifting and positioning, with a comprehensive literature review. Then the recognition model for module installation process was created and trained with the adoption of CNN-based transfer learning, and verified with images taken from real-life projects. The results show that the three stages of module installation process are effectively recognized with the proposed model. The transfer learning enabled image recognition model for module installation process accelerates automation in the construction of HMBs for enhanced productivity and accuracy.