Surrogate-assisted parametric calibration using design of experiment platform within digital twinning

Abstract

The process of developing a virtual replica of a physical asset usually involves using the best available values of the material and environment-related parameters essential to run the predictive simulation. The parameter values are further updated as necessary over time in response to the behaviour/condi- tions of physical assets and/or environment. This parametric calibration of the simulation models is usually made manually with trial-and-error using data obtained from sensors/manual survey readings of designated parts of the physical asset. Digital twining (DT) has provided a means by which validat- ing data from the physical asset can be obtained in near real time. However, the process of calibration is time-consuming as it is manual, and as with each parameter guess during the trial, a simulation run is required. This is even more so when the running time of a single simulation is high enough, like hours or even days, and the model involves a significantly high number of parameters. To address these shortcomings, an experimental platform implemented with the integration of a simulator and scientific software is proposed. The scientific software within the platform also offers surrogate building support, where surrogates assist in the estimation/update of design parameters as an alternative to time-consum-ing predictive models. The proposed platform is demonstrated using BEASY, a simulator designed to predict protection provided by a cathodic protection (CP) system to an asset, with MATLAB as the scientific software. The developed setup facilitates the task of model validation and adaptation of the CP model by automating the process within a DT ecosystem and also offers surrogate-assisted optimisation for parameter estimation/updating.