Regression-based hub load calculation of a tidal turbine with blade root sensor data

Abstract

Tidal turbines offer great potential for base load supply in coastal and remote regions with sufficient tidal flow rates, due to their good predictability compared to established renewable energy sources such as wind and solar. Being still at the prototype stage, tidal turbines are not yet competitive in terms of levelised cost of energy (LCOE). Since there is a lack of experience and data regarding the operation of tidal turbines for long periods of time, there are no sophisticated maintenance strategies in place yet. Thus, currently most tidal turbine drivetrains are designed with high margins of safety to be robust against failures. However, designing with high margins of safety is costly increasing the LCOE. Designing with lower margins of safety requires a condition monitoring system since on the one hand the remaining useful lifetime can be estimated and on the other hand critical conditions (e.g. due to increased flow rates or turbulence) can be avoided by adjusting the control of the turbine.

To calculate the remaining useful lifetime of a component, the component loads during operation must be known. While being able to calculate loads in easily accessible components from measurement data, some components do not offer cost-effective possibilities for sensor application and their loads during operation remain unknown. Therefore, the component loads need to be calculated indirectly based on available sensor data.

Drivetrain component loads depend on complex elastic and dynamic interactions between individual components. To represent these complex interactions, a simulation-based digital twin is developed for the condition monitoring of the tidal turbine drivetrain components. The digital twin is developed using a multi body simulation (MBS) environment with flexible bodies. As the MBS model is not real-time capable but the condition must be monitored in real time, the MBS model is used to generate training data for an efficient, machine learning based surrogate model. To reduce computational time further, critical components are modeled with a high level of detail while other components are represented using simplified models. Therefore, in this paper the critical components are identified to choose appropriate component modeling fidelities.

A load calculation model for the drivetrain component loads is developed. The component loads are calculated analytically based on measurements with existing sensors from field operation. The remaining useful lifetime (RUL) of each component is calculated by comparing the calculated component loads with the design loads. Based on the RULs, critical components are identified in terms of risk of early failure to define the required modeling fidelity of each component for the digital twin.