Generation of construction schedule information through activity unit recognition of 3D models created at the design phase

Since the 3D model in the design phase is composed of the concept of structural elements, and the 3D model in the construction phase requires a model of the activity unit, the utilization of the design-phase 3D model in the construction phase is reduced due to the different LOD of the model. In order to utilize the design-phase 3D model for schedule management in the construction phase, the process of recognizing each element of the model as an activity unit and the process of setting construction sequence between the recognized activities are required manually. To solve these problems, this study presents a methodology that can automatically recognize the 3D model by classifying it into activity units with PointNet++ algorithm and generate an optimized construction sequence with a genetic algorithm. Three bridge models were used to verify the proposed methodology, and after checking the automatic classification accuracy of activity units, the appropriateness of the schedule sequence was verified by 4D simulation. The PointNet++ algorithm used to classify the activity units showed classification accuracy of 0.90, 0.89, and 0.98, and the genetic algorithm results confirmed that the construction sequence was optimized in >94 % of the activity units.