An evolutionary method to refine imperfect sensor data for construction simulation

Abstract

Construction simulation is used to analyze uncertainties inherent to project activities and variations in work packages. However, existing simulation systems often fail to meaningfully contribute to the decision-making process due to their inability to evolve with changing project conditions. Equipping simulation models with sensing and reality capture technologies has been investigated as possible remedies to this problem. This, however, requires meticulous effort to procure, set up, operate, synchronize, and calibrate peripheral devices for data collection, transmission, and mining. Furthermore, sensor readings are often noisy and imperfect. The chaos theory explains how small variations in sensor readings used as simulation model input can lead to relatively large volatility in the output even in simple linear systems. This paper investigates a scientific methodology for generating more stable simulation models using an evolutionary algorithm that produces clean datasets by processing and significantly reducing noise in imperfect data obtained from consumer-grade sensors.