Exploring construction process simulation as part of the digital twin concept in real-world contexts

The construction industry faces problems like low productivity, poor resource efficiency, and cost overruns. Several factors cause these problems. Each building is a unique product, constructed by changing teams on temporary sites under constantly changing conditions. Although operating sequences may be similar across projects, the unique aspects of each construction project make planning complex. Therefore, each project needs adaptable planning to respond to specific conditions. However, construction management usually relies on generic information based on assumptions or historical data and inflexible planning methods. These generic approaches cannot reflect the inherent dynamic processes of construction works and hinder successful management. With advancing digitalisation, there is potential to change conventional practices. Methods like discrete event simulation can help manage construction processes by providing insights regarding future developments. However, reliable information regarding the current state of the system is essential for effective use. The digital twin concept aims to continuously collect and analyse data to provide context-specific information. Therefore, the digital twin concept can significantly improve simulation and management practices in the construction industry. This dissertation explores how the digital twin approach can be integrated with process simulation to optimise construction management. We propose a framework to establish a systematic flow of real-time data. This framework includes three advanced analytics-based methods to convert raw data into quantitative key performance indicators. These methods enable the analysis of real-time data, simulation calibration, and simulation-driven decision support. The applicability of each method was validated in a real-world case study to provide evidence for the methods' usage. Activity durations are the most important factor for construction planning. Thus, the lack of reliable activity durations is a significant shortcoming in management practice. Due to advancing digitalisation, data can be collected to track construction progress on-site in real-time, e.g., by smart sensors. The analysis of data collected during construction enables the extraction of data-based activity durations. Previous approaches for data-based extractions of activity durations focused on cycle times or mean durations. However, cycle times and mean durations are simplifications as valuable information can get lost. To this end, in Chapter 2, we propose a method to extract individual activity durations for each repetition. The method applies machine learning to classify in real time collected kinematic data to recognise activities and extract durations automatically. Subsequently, automatic post-processing using logical rules identifies and corrects misclassifications to improve the accuracy of duration extractions. We illustrated the applicability of the method in a case study for the work process of concreting with a tower crane, which consists of four consecutive activities. The case study demonstrated the successful applicability of the method by comparing the data-based and the measured durations. In the case study, the method disproved normality for each of the four sets of durations. Thus, the results emphasise the need to extract the durations of individual activity repetitions. The automated provision of project-specific activity durations provides management with meaningful and reliable information on construction progress. The provision of reliable activity durations enables accurate forecasts and, thus, supports the implementation of simulation in construction planning. By using updated durations, simulation models can be continuously calibrated. However, simulation practice often misses continuous model calibration and neglects real-world dynamics and uncertainties in forecasts. Chapter 3 addresses these issues by developing a method for automated and continuous discrete event simulation calibrations. Based on sets of durations, the method uses stochastic productivity modelling to identify the most suitable probability density functions representing productivity rates. The method uses the identified functions to continuously calibrate a simulation model to reflect the changing conditions of the system. The method validates functions and the simulation model through statistical tests. We illustrated the application of the method in a case study of concrete pouring using a tower crane. The results of the case study revealed that validation of input parameters, i.e. probability density functions, does not necessarily entail validated simulation models. The calibration method facilitates in-depth analysis of how changes in productivity rates affect simulation forecasts aiming to build reliable models. Once reliable models are available, managers can use simulations to assess how different decisions affect construction outcomes. However, simulation modelling of different options is often complex, laborious, and time-consuming. Therefore, managers rarely use simulation for planning and the possible impacts of different management decisions are unknown. To this end, in Chapter 4, we propose a simulation-driven decision support method to optimise on-site construction process management. Building on the previous chapters, Chapter 4 presents the entire envisioned system for applying simulation as part of the digital twin concept by an integrated method. The simulation-driven method automatically models many options based on inputs for resource allocations, delivery scenarios, and constraint settings. The method applies stochastic discrete event simulation to perform a scenario analyses considering uncertainties when calculating key performance indicators. Additionally, the method performs multi-criteria optimisation to identify Pareto-optimal options. A real-world case study involving the construction of an industrial building validated the method. The results revealed that an option that performs best regarding the mean value can underperform in extreme scenarios, such as the best-case. These insights underscore the need for stochastic investigations when using simulation for decision support. The simulation-driven method simplifies modelling and gives managers a comprehensive understanding of potential outcomes to promote successful construction. This dissertation makes several significant contributions. It provides a systematic approach to integrating real-time data with construction simulations. We developed three advanced analytics methods for automated, continuous, and comprehensive processing of real-time data. The methods enable a data stream to convert collected kinematic data into key performance indicators. By using these methods, managers can gain meaningful insights into construction progress, develop reliable simulation models, and better understand the impacts of decisions before implementation. We validated each method and the entire integrated approach through real-world case studies to provide evidence for the potential of simulation as part of the digital twin concept. The proposed approach aims at iterative management that follows the Plan-Do-Study-Act cycle to support decision-making proactively. Our approach offers significant potential to implement data-driven decision-making and improve performance in the construction industry.

Author dc:creator

• Jungmann, Manuel

Advisor dc:contributor.advisor

• Hartmann, Timo

Licence dc:rights.uri

• https://creativecommons.org/licenses/by/4.0/

Language dc:language.iso

en

Identifier URI

https://doi.org/10.14279/depositonce-23760

OAI identifier oai:identifier

oai:depositonce.tu-berlin.de:11303/24944