Site-specific load assessment for fatigue analysis and life prediction of wind turbine components

An improved-accuracy method for fatigue load analysis of wind turbine gearbox based on SCADA. Renewable Energy, 115, 391-399. The results presented here correspond to the analysis of 4-years of SCADA data of a 1.5MW wind turbine. The power curve in the workflow diagram (Figure 1) shows all the data categorized by the operational states run and idle. An additional category labeled "garbage" is included for data points that were identified as implausible, corrupted, or incomplete. Out of the 4-years, 70% correspond to "run", 22% to "idle", and 8% are dismissed as "garbage". Abstract Results Objectives Conclusions Methods References The work presented here outlines a process for operational states identification and load simulation, required for proper fatigue load assessment. In the involved process for site-specific load calculation, it is very important to curate the data accurately so the inputs to the aeroelastic models are an appropriate representation of the operational history of each asset. Thus, combining the curated data with the turbine aeroelastic models, results on a more realistic load history representation for further reliability assessment. The future work planned based on these results will be to use the load history for RUL prediction of the gearbox components. To build up a realistic fatigue history of the turbine, the first step is to aggregate the historical data of the turbine operation using the SCADA protocol in 10-min statistical samples (mean, standard deviation, max, & min). Figure 1 shows the workflow of the process described below: 1. In most cases historical data are inadequate for RUL prediction due to insufficient information, sampling errors, sensor malfunction, etc. Hence, additional data curation is crucial. This work developed a unique methodology to identify and distinguish turbine operation between run and idle, eliminating any misleading or false data, and detect transient events such as normal and emergency stops for all occurring wind & environmental conditions. 2. The labeled data from the previous step is then used to define the wind conditions of four DLCs (run, idle, normal stop, and emergency stop) and to carry out aeroelastic simulations using a representative model of the wind turbine. The load time series obtained in the simulations is a close representation of the load experienced by the wind turbine in the field. 3. The combination of wind conditions per load case and the operation history of the turbine is then used to build a database of load history for the different events driving the fatigue of the components. The history is then used for RUL prediction. The goal of this project is to define an improved methodology for fatigue load assessment of already commissioned wind turbines without additional sensors installed on the asset. Instead, only already existing sensor information can be utilized that is already available to an operator, such as supervisory control data. The information can be combined with physics-based models to simulate fatigue loads that would have been available only with expensive on-site measurements otherwise. Wind turbine OEMs utilize design load cases (DLC) for the design and certification of wind turbines. These load cases are defined by International Electrotechnical Commission standards [1] to ensure that turbines operate in a stochastic environment for at least 20 years. However, reliability analysis should not stop at the design stage. Once a turbine is commissioned, consequent reliability analyses for all turbine components should also be conducted to estimate risk and lower costs. This requires a very good understanding of the actual occurring shear and bending loads on blades, main shaft, tower, and other components. Recently, Sentient Science partnered with Acciona Energy to expand on current methodologies [2] for post-installation load estimation. In this partnership several analyses were conducted to identify and distinguish operational conditions and their respective fatigue loads over the history of a turbine, using historic SCADA and physics-based models only. The fatigue loads can then be used to determine the reliability of individual components as well as the overall turbine. Figure 2 represents 10min SCADA data over a period of 15 hours. Within this timeframe, the turbine transition from a running state (run) to an idling state (idle due to a normal stop) where it resides for about an hour before transitioning back to a running state (start-up), as indicated by the vertical dotted lines. The bottom plot shows a time-series that is generated by the transient detection algorithm with negative pulses indicating a stop, and positive pulses indicating a start. The algorithm can distinguish between normal and emergency stops.