ITcon Vol. 30, pg. 1080-1098, http://www.itcon.org/2025/44

Process time estimation for workstations in modular construction production line

DOI:10.36680/j.itcon.2025.044
submitted:March 2025
revised:May 2025
published:June 2025
editor(s):Frédéric Bosché
authors:Angat Pal Singh Bhatia, Ph.D., Research Associate
Department of Building, Civil and Environmental Engineering, Concordia University, Montréal, QC, Canada
angatpalsingh.bhatia@concordia.ca

Osama Moselhi, Ph.D., Professor
Centre for Innovation in Construction and Infrastructure Engineering and Management (CICIEM), Department of Building, Civil and Environmental Engineering, Concordia University, Montréal, QC, Canada
moselhi@encs.concordia.ca

SangHyeok Han, Ph.D., Associate Professor
Centre for Innovation in Construction and Infrastructure Engineering and Management (CICIEM), Department of Building, Civil and Environmental Engineering, Concordia University, Montréal, QC, Canada
sanghyeok.han@concordia.ca
summary:Modular construction companies produce module components following a make-to-order process to realize client-defined customization requirements. This customization leads to varying process times of prefabricating module components at workstations, making it difficult for production line managers to accurately predict their process times for planning purposes. To address these challenges, this paper proposes a novel method that employs Deep Neural Networks, artificial neural networks, and multiple linear regression models for predicting workstation production process times at a module prefabrication plant. A Genetic Algorithm is employed to refine the structure of the Deep Neural Networks and find a near-optimum number of hyperparameters. In a case study, a wood-based wall panel production line is analyzed to demonstrate the use of the developed method and test its performance. The developed method for process time prediction is found to achieve a mean absolute error of less than 2.50 min for most workstations, with the symmetric mean absolute percentage error ranging between 22% and 28%. The research contributions of this study include the development of prediction models for all the workstations of the production line and the implementation of a Genetic Algorithm to find the near-optimal hyperparameters of Deep Neural Networks. This assists production managers in making data drive decisions and overcomes the reliance on experience-based methods for estimating process times and creating production plans.
keywords:Modular Construction, Production Line, Prediction Model, Deep Neural Network (DNN), Genetic Algorithm (GA)
full text: (PDF file, 1.269 MB)
citation:Bhatia A P S, Moselhi O, Han SH (2025). Process time estimation for workstations in modular construction production line, ITcon Vol. 30, pg. 1080-1098, https://doi.org/10.36680/j.itcon.2025.044
statistics: