The digital twin technology will be designed for use in greenfield applications to help simulate operations of bottoming cycles before installation as well as for brownfield cases, where existing bottoming cycles can be run in a more efficient way with higher operational reliability. This can result in more widespread implementation of steam cycles in offshore oil and gas production which will increase energy efficiency and reduce fuel consumption and CO2 emissions by up to 25%.
It is likely, in the future, that transient models will be more relevant, as the cycles may operate in conjunction with intermittent renewable energy sources like wind power. This means that the combined cycle will have to be operated with increased flexibility, including rapid start-up when the wind dies down. Key questions include what are the important KPIs in the future systems. Operational challenges may include increasingly variable heat and power demand which may trigger needs for more estimations of remaining lifetime, current equipment degradation rate and maintenance schedule. Previously cracking of heat exchangers and other issues have led to poor operational reliability of existing offshore steam cycles.
A digital twin will be able to simulate running of the steam cycle to optimize parameters and improve operability and reliability, and will be useful both to the future offshore energy system and those that exist today. A novel early fault detection system will also be developed. At the core of this system will be a novel sensor scheme and a signal processing system. Through close collaboration between researchers and end users as well as potential vendors, a mapping will be performed of the key desired functionalities of the digital twin, with focus on a “Process Digital Twin”.
The project aims at developing a dynamic model based digital twin, with a solid foundation building on physics-based models, to be incorporated into a digital twin of a steam cycle. It will be a process focused Digital Twin, with transient and thermal models at the core. The models shall be able to be utilized for design, commissioning & operation (including monitoring). It shall include the control system and capture relevant aspects of flexible operation, as well as allow for optimization capabilities. Machine learning based Data Driven and Hybrid Modeling methods for system, equipment and components will be investigated. The aim is to investigate and cover aspects of short-term and long-term operation dynamics.
Key functionalities explored in the project:
- Transient modeling and control
- Control structure design/evaluation.
- Surrogate modeling for real-time simulations of system and components (with focus on OTSG and superheater headers).
- Parameter tuning/optimization: Optimization with Model Predictive Control strategies
- Thermal Stress modeling
- Based on EU standards for Headers and piping in OTSG bundles
- FEM models of headers connected to thermal stress models
- Transient investigative models for start-up/shut down evaluations
- Anomaly detection
- Warnings/Detection - Early fault detection prevents unplanned shutdowns, cuts maintenance cost, and enhances safety.
- Labeled fault data in CCGTs are scarce and costly to obtain. Unsupervised methods operate without fault labels. Method development towards unsupervised machine learning as R&D focus in this project.