With the increasing adoption of wind energy,monitoring systems are crucial to ensuring optimal turbine operation and reliability.This work contributes a methodology to detect rotor imbalance in wind turbines using out...With the increasing adoption of wind energy,monitoring systems are crucial to ensuring optimal turbine operation and reliability.This work contributes a methodology to detect rotor imbalance in wind turbines using output-only vibration data.Furthermore,an extended isolation forest algorithm is proposed for anomaly detection,which is particularly effective in identifying rare and irregular patterns that indicate rotor imbalance.The experimental setup features the Enair E30Pro wind turbine(3 kW)adapted for laboratory use and equipped with triaxial accelerometers to capture comprehensive vibration data in various scenarios of induced imbalance.A dataset is acquired from 65 experiments that simulate different levels of rotor imbalance(used exclusively to test the methodology)and healthy operating conditions(used for training,validation,and testing).Data reshaping and feature engineering techniques,such as permutation entropy,fractal dimension analysis,and kurtosis calculations,are utilized to extract meaningful insights from vibration signals.On 1 min windows,the method achieved a precision of 100%,recall ranging from 60%to 100%,and F1-scores between 75%and 100%.Extending the aggregation to 5 min windows resulted in perfect classification across all categories,with precision,recall,and F1-scores at 100%.To evaluate the effectiveness of the proposed strategy for detecting progressive rotor imbalance,a comparison is made against two widely used literature approaches:the One-Class support vector machine and the traditional isolation forest,and,in each test configuration,the extended isolation forest surpassed these baselines.These results demonstrate the ability of the methodology to effectively distinguish between normal and imbalanced conditions,offering a proof-of-concept methodology for early detection and maintenance planning in wind turbine operations.Importantly,the proposed strategy requires only data from healthy operational states for its development,thus extending its applicability to any wind farm,regardless of the availability of data specifically associated with rotor imbalance conditions.展开更多
基金grant PID2021-122132OB-C21 funded by MCIN/AEI/10.13039/501100011033 and by“ERDF A way of ing mak-Europe”,by the“European Union”grant TED2021-129512B-I00 funded by MCIN/AEI/10.13039/501100011033 and by the“European Union NextGenerationEU/PRTR”grant 2021-SGR-01044 funded by the Generalitat de Catalunya,Spain.
文摘With the increasing adoption of wind energy,monitoring systems are crucial to ensuring optimal turbine operation and reliability.This work contributes a methodology to detect rotor imbalance in wind turbines using output-only vibration data.Furthermore,an extended isolation forest algorithm is proposed for anomaly detection,which is particularly effective in identifying rare and irregular patterns that indicate rotor imbalance.The experimental setup features the Enair E30Pro wind turbine(3 kW)adapted for laboratory use and equipped with triaxial accelerometers to capture comprehensive vibration data in various scenarios of induced imbalance.A dataset is acquired from 65 experiments that simulate different levels of rotor imbalance(used exclusively to test the methodology)and healthy operating conditions(used for training,validation,and testing).Data reshaping and feature engineering techniques,such as permutation entropy,fractal dimension analysis,and kurtosis calculations,are utilized to extract meaningful insights from vibration signals.On 1 min windows,the method achieved a precision of 100%,recall ranging from 60%to 100%,and F1-scores between 75%and 100%.Extending the aggregation to 5 min windows resulted in perfect classification across all categories,with precision,recall,and F1-scores at 100%.To evaluate the effectiveness of the proposed strategy for detecting progressive rotor imbalance,a comparison is made against two widely used literature approaches:the One-Class support vector machine and the traditional isolation forest,and,in each test configuration,the extended isolation forest surpassed these baselines.These results demonstrate the ability of the methodology to effectively distinguish between normal and imbalanced conditions,offering a proof-of-concept methodology for early detection and maintenance planning in wind turbine operations.Importantly,the proposed strategy requires only data from healthy operational states for its development,thus extending its applicability to any wind farm,regardless of the availability of data specifically associated with rotor imbalance conditions.
