A Digital Twin Environment Designed for the Implementation of Real Time Monitoring Tool

Abstract

Real time models and digital twin environments represent a new frontier that allow the development of supplemental data analytics of measurable and unmeasurable parameters for a variety of power plant configurations. Performance prediction, monitoring of degradation effects, and a faster recognition of anomalous events during power plant load following operations and/or due to cyber-attacks can be easily detected with the support of digital twin environments. In the research work described in this article, a digital twin environment was developed to capture the dynamics of a micro compressor-turbine system modified for hybrid configuration at the Department of Energy’s National Energy Technology Laboratory (NETL). The innovative approach for the development of the digital twin environment was based on creating a compressor-turbine physics-based model using a stateless methodology generally utilized for microservices architectures. The advantage of using this approach was represented by modeling individual or a group of power plant components on distributed computational resources such as clouds in a lightweight and interchangeable manner. Supplemental data analytics were performed using an online system identification tool developed in previous work and applied to an unmeasurable parameter only available in the digital twin system. This work demonstrated the ability to train a recursive algorithm to predict a supplemental parameter for faster anomaly detection or for replacing the physics-based model during design or monitoring of operational systems. The thermodynamic compressor-turbine net power was the unmeasurable parameter only available in the digital twin model, which was predicted with the online system identification tool. Results showed that the online system identification algorithm predicted the dynamic response of the thermodynamic net power based on a set of experimental data points at nominal operating conditions with an absolute mean percentage error of ∼0.644%.