Data-driven modeling of hydroelectric turbine startup fatigue load spectra

• Strain signals during startup transients enhances fatigue assessment of hydroelectric turbines. • Combining a signal envelope model and rainflow reconstruction technique improves strain signal estimation. • Training the model with measurements on a turbine in operation ensures practical relevance. • Providing accurate estimations of the signal extreme values with minimal calibration. Startup transients significantly impact hydroelectric turbine runner fatigue. Due to the high cost and extreme conditions associated with experimental measurements, the number of startup schemes that can be tested is limited, restricting optimization and fatigue assessment capability. Moreover, the complexity of dynamic strain behavior during startup presents a significant challenge for modeling such signals. This paper proposes a methodology aimed at preserving fatigue loading cycles, represented specifically as a rainflow-based loading spectrum. The approach integrates the rainflow reconstruction technique with a signal envelope estimator, enabling the generation of strain signals from vane opening and rotational speed signals collected during startups. Data from an actual hydroelectric prototype was used for training and evaluation, resulting in accurate estimations with minimal calibration, even for extreme values associated with the highest fatigue damage.