Optimized Production Scheduling for Modular Construction Manufacturing

Modular construction is becoming a viable construction method in North America due to its support of the concept of circular construction and its inherent ability to provide a faster return on investment. The process of modular construction manufacturing (MCM) operates as a production line, where the number of module components (e.g., wall, roof, and floor panels) with different design specifications and their sequence (i.e., order of prefabricating these module components) dictates the productivity of the production line. This variation in design specifications and impractical sequences of module components leads to imbalanced production lines and prolonged makespan (i.e., total completion time) of prefabricating module components at workstations. To address these challenges, this paper proposes a method that utilizes deep neural network and genetic algorithm (GA) techniques to solve the modular construction manufacturing scheduling problem (MCMSP). The method consists of two processes: (i) developing a deep neural network model based on the historical time data and later hyperparameter tuning using a GA in order to select the optimal neural network configurations; and (ii) subsequently using the predicted process times as input in the optimization model in order to schedule the sequences of module components (e.g., wall panels). The proposed method is implemented in a wood-based wall panel production line of a modular fabricator in Edmonton, Canada. This developed method can assist production managers in efficiently forecasting process times and developing production line schedules.