Synthetic Simulated Data for Construction Automation: A Review

The integration of deep learning (DL) technologies into construction offers great potential for promoting the level of automation in construction. However, the implementation of the DL model requires the acquisition of substantial data, which is error-prone and time-consuming. Additionally, due to safety and privacy concerns, not all real-world data can be retrieved. To address these issues, synthetic simulated data have emerged as promising alternatives, and various methods have been developed to generate such data. However, currently there is neither a summary of synthetic simulated data generation methods nor unified metrics to evaluate these methods. In this paper, a comprehensive review of 129 scholarly articles from Web of Science is conducted. Based on the source of data assets and the techniques employed for their combination, we categorize synthetic simulated data generation methods into seven distinct categories. Furthermore, we summarize seven metrics for evaluating these methods and consolidate the evaluation results in a table. The provided table serves as a reference for practitioners in identifying and selecting suitable synthetic simulated data generation methods for their applications. © 2024 ASCE.