Adaptive Simulation Modelling Using The Digital Twin Paradigm.

Abstract

Structural Health Monitoring (SHM) involves the application of qualified standards, by competent people, using appropriate processes and procedures throughout the struc- ture’s life cycle, from design to decommissioning. The main goal is to ensure that through an ongoing process of risk management, the structure’s continued fitness-for-purpose (FFP) is maintained – allowing for optimal use of the structure with a minimal chance of downtime and catastrophic failure. While undertaking the SHM task, engineers use model(s) to predict the risk to the structure from degradation mechanisms such as corrosion and cracking. These predictive models are either physics-based, data-driven or hybrid based. The process of building these predictive models tends to involve processing some input parameters related to the material properties (e.g.: mass density, modulus of elasticity, polarisation current curve, etc) or/and the environment, to calibrate the model and using them for the predictive simulation. So, the accuracy of the predictions is very much dependent upon the input data describing the properties of the materials and/or the environmental conditions the structure experiences. For the structure(s) with non-uniform and complex degradation behaviour, this pro- cess is repeated over the life-time of the structure(s), i.e., when each new survey is per- formed (or new data is available) and then the survey data are used to infer changes in the material or environmental properties. This conventional parameter tuning and updat- ing approach is computationally expensive and time-consuming, as multi-simulations are needed and manual intervention is expected to determine the optimal model parameters. There is therefore a need for a fundamental paradigm shift to address the shortcomings of conventional approaches. The Digital Twin (DT) offers such a paradigm shift in that it integrates ultra-high fidelity simulation model(s) with other related structural data, to mirror the structural behaviour of its corresponding physical twin. DT’s inherent ability to handle large data allows for the inclusion of an evolving set of data relating to the struc- ture with time as well as provides for the adaptation of the simulation model with very little need for human intervention. This research project investigated DT as an alternative to the existing model calibration and adaptation approach. It developed a design of experiment platform for online model validation and adaptation (i.e., parameter updating) solver(s) within the Digital Twin paradigm. The design of experimental platform provided a basis upon which an approach based on the creation of surrogates and reduced order model (ROM)-assisted parameter search were developed for improving the efficiency of model calibration and adaptation. Furthermore, the developed approach formed a basis for developing solvers which pro- vide for the self-calibration and self-adaptation capability required for the prediction and analysis of an asset’s structural behaviour over time. The research successfully demonstrated that such solvers can be used to efficiently calibrate ultra-high-fidelity simulation model within a DT environment for the accurate prediction of the status of a real-world engineering structure.