The relationship between flow conditions and quality of load prediction is presented.
Blade fatigue loads estimated with neural networks and SCADA data have been assessed.
Inflow conditions impact short-term predictions, long-term ones are accurate.
A reasonable monitoring system can be designed indifferent to inflow conditions.
Lifetime evaluation with fatigue loads is commonly used in the design phase of wind turbines, but rarely during operation due to the cost of extra measurements. Fatigue load prediction with neural networks, using existing SCADA signals, is a potential cost-effective alternative to continuously monitor lifetime consumption. However, although assessments for turbines in wind farm flow have been pointed out as deficient, the evaluations were limited to single cases and the implication for the design of a monitoring system was not discussed. Hence, we proposed metrics to evaluate prediction quality and, using one year of measurements at two wind turbines, we evaluated predictions in six different flow conditions. The quality of fatigue load predictions were evaluated for bending moments of two blades, in edgewise and flapwise directions. Results, based on 48 analyses, demonstrated that prediction quality varies marginally with varying flow conditions. Predictions were accurate in all cases and had an average error below 1.5%, but their precision slightly deteriorated in wake flow conditions. In general, results demonstrated that a reasonable monitoring system can be based on a neural network model without the need to distinguish between inflow conditions.
- Wind turbine;
- Wake effects;
- Fatigue damage;
- Lifetime extension;
- Condition monitoring;
- Neural networks
© 2017 Elsevier Ltd. All rights reserved.