Benchmarking Gas Turbine Outages: Creating an Expectation Model of Performance Recovery

Abstract Gas turbines are the lynchpin of a stable and prospering U.S. power grid and economy. As more forward growth emphasizes renewables, maintaining the existing gas turbine fleet will become even more critical. Gas turbines undergo regular maintenance cycles that inspect, repair, and replace life limited parts. The complexity and invasiveness of maintenance often leads to variability in the resulting performance recovery. For example, parts may not be installed correctly or controls may not be adjusted after the outage as needed. While engineers have a personal ‘expectation model’, this often is incorrect and few organizations rigorously benchmark outage recovery to identify lost performance and incorrect repairs. This paper applies the EPRI Gas Turbine Digital Twin to a historical set of outages for B class gas turbines. A methodology is presented to track outage recovery by comparing model performance to site data pre and post outage and accounting for model and measurement uncertainty through the use of machine learning. This provides a probabilistic estimate of outage recovery which is used to create a fleet outage recovery expectation model. The intent is to use this model to benchmark and score future outages’ effectiveness. The resulting expectation model is also presented here for others to use in their own benchmarking efforts.