Integration of Digital Twin Platforms with Machine Learning Models for Early Detection of Project Delays and Cost Overruns

Abstract

The problem of recurring schedule delays and cost overruns has long been one of the biggest challenges in the management of construction projects. Recent technological developments in Digital Twin and Machine Learning technologies provide project managers with promising opportunities to monitor, forecast, and control projects before any potential issues become delayed and more costly. This study employs a systematic literature review and comparative analysis to evaluate the effectiveness of Machine Learning-enabled Digital Twins for early risk prediction in construction projects. Digital Twin technology and Machine Learning models in an early prediction and control of construction project schedules and costs to avoid any potential delays and overspends. A structured search process has been conducted using the VOSviewer tool to identify the 51 most highly qualified studies from a total of 813 search results through the use of the PRISMA guidelines. The findings indicate that there is a rapidly expanding amount of work after 2019 based on developments in data availability, Building Information Modeling (BIM), Internet of Things (IoT), and predictive analytics. Among the paradigms used by the studies surveyed, there are three major types of models used collectively: White Box models, Black Box models using Artificial Neural Networks (ANN), Random Forest, Support Vector Machines (SVM), Long Short-Term Memory (LSTM), and Gray Box models using Digital Twins. White Box models have excellent explainability capabilities. However, black box models have stronger predictive capabilities than White Box models. Gray Box models are promising for achieving interpretability in AI models. Nevertheless, the gaps in the existing research are also recognized in the review, which include the lack of validation of the study on a real-scale project, the lack of application of deep learning in temporal project data, the dispersed nature of the Digital Twin platform, and the challenges associated with data privacy and interoperability. Opportunities for future research are identified in the development of a comprehensive framework for ML-DT, a portfolio-level Digital Twin, a real-time predictive control system, and a privacy-focused data architecture.