The concept of the digital twin can be thought of as a virtual representation of a physical product, engineering system or facility. This paper presents the role of predictive engineering analytics, alongside operating data, in the digital twin. Using case studies, the authors demonstrate how predictive approaches can be developed to provide data where it cannot be measured and predict future operating data to improve performance, life and integrity of equipment, systems and facilities.
The digital twin is fundamentally based on data, larger datasets provide greater insight. Sensor and inspection data are critical. However, there are scenarios in which engineers require data where it cannot be measured or requires data that cannot be measured. Engineers require ways to extract this data, this can be done through predictive engineering analytics. Predictive engineering analytics, in the form of science-based simulations, combines multiple approaches, often based on fundamental principles of physics and engineering. This paper will demonstrate how high-fidelity approaches, such as Finite Element Analysis (FEA) and Computational Fluid Dynamics (CFD), combined with system-level simulation and reduced-order modelling can work together with field data to provide this data in real-time.
Two case studies presented show how combinations of different levels of science-based modelling approaches can help. A subsea thermal digital twin demonstrates how high-fidelity simulations, undertaken during system design, can be the foundation for reduced order system models capable of capturing critical thermal performance in real time; aiding hydrate risk management during operation. The methods used to train the real-time predictive approach are demonstrated. The second case presented focuses on structural integrity of a heat exchanger showing how real-time sensor data can be translated into structural integrity data and insight through simulation. The cases presented demonstrate the value of science-based predictions to generate data that cannot be obtained from operational sensor data alone. The authors aim to show how the predictive element of the digital twin can be first generated during design and evolved into real-time predictive approaches that provide operations with data that cannot be gained from sensors; when and where it is needed.
Digital twins typically use physical data, limiting operators of in-field equipment to make operational decisions based on information from sensor locations and historical data alone. This can limit assessment of operational performance and integrity of complex production and process systems. In this paper the authors aim to show how, by combining predictive approaches, at differing levels of fidelity and based on fundamental scientific principles, it is possible to generate the missing data required, in both space and time. This can inform and guide operations; filling the gaps where and when physical data is not available.