Digital Twin Implementation: Fault Signature Development for APR Applications

This report details the current progress in creating an EPRI Digital Twin applicable to gas turbines (GTs). The impetus for this project were issues related to aging GTs with multiple overhauls, including cases where advanced hardware for increased output had been installed. It has been found that many of these GTs, especially those configured as simple-cycle units, have been derated due to elevated combustion dynamics, lean blowout (LBO), and/or lack of emission compliance. The potential application cases for a digital twin are multi-faceted as it can be applied to a wide range of diagnostic, prognostic, and “what-if” problems. Examples include using the digital twin to identify causes of post-outage emissions and performance issues, expected impact of degradation and fault conditions, simulating improvements to operation through part repair and upgrades, and eventually estimating component remaining useful life (RUL). In all these cases, there must be an interface that uses the predictive analytic capabilities of the digital twin in the physical plant.

One way to integrate this information is through a combination with advanced pattern recognition (APR) software by simulating faults that are outside of the traditional training data set. APR software is system agnostic and works by training a predictive, non-physics-based model, against a set of normal, healthy operating data. When used as a diagnostic tool, APR software will compare the actual operating data against that predicted by the analytical model and will calculate a residual based on the difference in the measured parameters such as power and heat rate. If the residual exceeds a user-defined value, an alarm is triggered that alerts the user or monitoring center of abnormal behavior.

Digital Twin Data Flow

Fully integrated physics and APR approaches – The digital twin is used to generate synthetic training data representing specific faults under a range of ambient operating conditions. The APR can then be trained to recognize specific faults, resulting in faster, more accurate diagnostics.

  • Plant performance prediction and trending: The calibrated digital twin model can be used to predict next-day and future behavior of the GT for capacity-driven markets.
  • Health monitoring/fault diagnostics: While APR software is a common diagnostic tool, the digital twin can be used for diagnostics on its own. The calibration process results in isolated health parameters, which can be used to predict the health of the unit.
  • Performance monitoring and prediction at part load: By employing a prediction model, the digital twin can predict performance at part load. Many traditional heat balance packages and correction curve approaches are only useful or accurate at full (or base) load conditions.
  • Outage and repair impacts: A digital twin model can be used to play what-if games to allow the user or operator to understand the impact of hardware changes on unit performance. This can be used to evaluate the performance vs. cost gain maintenance and repair decisions.

Download Report