Analysing the Correlation Between Blade Noise and Erosion SeverityThe ability to detect and quantify leading edge erosion (LEE) on wind turbine blades is important to improve power efficiency and to develop predictive maintenance strategies that can be used to identify damage early, allowing more efficient, proactive repairs. LEE leads to an increase in roughness, an increase in airfoil drag, local variations in the boundary layer, and, above all, changes in the aerodynamic performance of the blade, which in turn affects the noise produced by the blade as the turbine operates.
By Obdulia Ley, Subject Matter Expert in Acoustic Emission, Mistras Group, USA
Characterisation of leading edge erosion (LEE) is commonly done by visual or drone-assisted inspections, and estimation of the damage severity lacks consensus. When focusing on LEE, Severity 4 is associated with damage that has an effect on, and Severity 3 with damage that might have an effect on, aerodynamic performance. For instance, blades with deep pits in narrow areas are sometimes misclassified and assigned a severity similar to that of blades with a wide area of damage extent involving only superficial layers of the blade, such as the topcoat (Figure 1).
Recently, considerable efforts have been devoted to the analysis of SCADA data to identify LEE severity based on its effect on power efficiency and to the use of different continuous blade monitoring systems using various sensor technologies which provide information about blade aerodynamic performance in real time. In this study, we investigate the use of background noise collected inside the blades as a technique to quantify and track LEE.Blade Monitoring Using Acoustic Signatures
Acoustic signatures from blades were collected using Sensoria, a structural health monitoring system designed to detect damage onset and evolution. In this system, data is collected with a single acoustic sensor installed inside each blade located on the root closeout. The sensor measures the noise inside the blade produced by the turbulence over the blade generated by the presence of external damage, as well as any other noise inside the blade produced by friction or rupture of blade components (Figure 2).
Acoustic signatures from blades were collected using Sensoria, a structural health monitoring system designed to detect damage onset and evolution. In this system, data is collected with a single acoustic sensor installed inside each blade located on the root closeout. The sensor measures the noise inside the blade produced by the turbulence over the blade generated by the presence of external damage, as well as any other noise inside the blade produced by friction or rupture of blade components (Figure 2).
The data collected consists of background noise magnitude or average signal level (ASL), measured in decibels, and quantities used to characterise the blade sound spectrum in predefined frequency bands, referred to as partial powers or PPis (PPi, for i=1, 2, … , 5). ASL and PPis are highly affected by rotor rpm and environmental conditions, such as rain, ice, wind speed and wind gusts, among others. Collection of the turbine rpm and comparison of data from the three blades allows identification of environmental effects and separation from acoustic blade responses due to the presence of blade damage.
Project Objective and Preliminary Observations
The main objectives of this work were to identify whether different levels of LEE in blades could be detected acoustically and to offer a quantitative way to classify/verify LEE damage stage/severity using a continuously measured quantity. The hypothesis that LEE affects the blade noise response branches from previous results using this monitoring system, which showed that blade damage in the tip area, such as trailing edge longitudinal cracks and splits and delaminations and through holes caused by lightning strikes (LSs) (Figure 3A–D) are associated with increasing ASL or background noise level magnitude and the appearance of a distinctive spectral peak in the third spectral band, and an increase in PP3. Finally, all these features disappear after the damage has been repaired (Figure 3E–F).
The main objectives of this work were to identify whether different levels of LEE in blades could be detected acoustically and to offer a quantitative way to classify/verify LEE damage stage/severity using a continuously measured quantity. The hypothesis that LEE affects the blade noise response branches from previous results using this monitoring system, which showed that blade damage in the tip area, such as trailing edge longitudinal cracks and splits and delaminations and through holes caused by lightning strikes (LSs) (Figure 3A–D) are associated with increasing ASL or background noise level magnitude and the appearance of a distinctive spectral peak in the third spectral band, and an increase in PP3. Finally, all these features disappear after the damage has been repaired (Figure 3E–F).
Spectral Effect of Leading Edge Erosion in Blade Background Noise
As the number of monitoring system installations expanded, along with the variety of damage present on the instrumented blades, the appearance of a peak in Band 4 and an increase in PP4 were observed. An example of such a signature is shown in Figure 4. In this case, the drone inspection showed that blade A had severe LEE on the bottom 1/3 of the blade and blade C had low-severity LS damage that produced a low-level peak in Band 3 (as discussed before) and an increase in PP3.
As the number of monitoring system installations expanded, along with the variety of damage present on the instrumented blades, the appearance of a peak in Band 4 and an increase in PP4 were observed. An example of such a signature is shown in Figure 4. In this case, the drone inspection showed that blade A had severe LEE on the bottom 1/3 of the blade and blade C had low-severity LS damage that produced a low-level peak in Band 3 (as discussed before) and an increase in PP3.
To verify such observations, data from 19 turbines (GE 1.7) in a wind farm in the USA was used. The turbines were tracked for over 12 months, and the average of the data collected 3 months before the scheduled drone inspection was used to investigate the correlation between LEE severity and the acoustic background noise spectral response. Only data collected while the turbine was operating at rpm > 3 was used for the analysis. The average acoustic data compared with inspection reports was collected during winter months and tends to show increased background noise level and PPi values across the spectrum due to higher seasonal wind speed.
The spring drone inspection data was processed by identifying the maximum severity of different types of damage to each blade. The damage considered is: LS damage (to pressure or suction side) in the tip area, LEE along the blade, and both delaminations and cracks along the blade. For simplicity, only the severity is listed per blade in Figure 5A. Except for some structural cracks in blade T33-A, most of the instrumented turbines/blades do not have damage of high severity, and several blades have multiple types of damage (like LS and LEE), which will contribute to the acoustic response recorded.
The average PP4 data collected by the blade monitor was combined with the damage severity table from Figure 5A, and the average PP4 is plotted for each damage severity level using a box plot. Box plots help to show distributions of numerical data values, allowing a simple comparison of data between multiple groups, in this case LEE severity levels. Figure 5B shows that PP4 values are almost symmetrically distributed for LEE Severities 3 and 4. There is an outlier in LEE Severity 3, which corresponds to T25-A, but in general it is observed that there is a linear relationship between PP4 and LEE/severity as shown in Figure 5C.
Conclusions
Although the research is in the early stages, the acoustic data collected from wind turbine blades has been proven to provide useful information about the presence of blade damage that affects aerodynamic response. In particular, it is seen that average PP4 data collected while the turbine operates at rpm > 3 correlates with LEE severity. To complete validation it is necessary to obtain data from blades with severe LEE damage (Severity 5) and from ‘pristine’ blades or from blades after LEE repairs have been performed. The results presented here indicate the feasibility of using spectral acoustic signatures to identify blade damage level and type and, more importantly, to show the contribution that a continuous blade monitoring system can have on maintenance activities, as well as optimising power generation.
Although the research is in the early stages, the acoustic data collected from wind turbine blades has been proven to provide useful information about the presence of blade damage that affects aerodynamic response. In particular, it is seen that average PP4 data collected while the turbine operates at rpm > 3 correlates with LEE severity. To complete validation it is necessary to obtain data from blades with severe LEE damage (Severity 5) and from ‘pristine’ blades or from blades after LEE repairs have been performed. The results presented here indicate the feasibility of using spectral acoustic signatures to identify blade damage level and type and, more importantly, to show the contribution that a continuous blade monitoring system can have on maintenance activities, as well as optimising power generation.
Biography of the Author
Obdulia Ley is a subject matter expert in acoustic emission at Mistras Group. She has been involved in the development and deployment of structural health monitoring solutions using acoustic signatures for over 15 years. Her work focuses on wave propagation, signal processing, pattern recognition, data analytics and machine learning. She has a bachelor’s degree in physics and a PhD in mechanical engineering.
Obdulia Ley is a subject matter expert in acoustic emission at Mistras Group. She has been involved in the development and deployment of structural health monitoring solutions using acoustic signatures for over 15 years. Her work focuses on wave propagation, signal processing, pattern recognition, data analytics and machine learning. She has a bachelor’s degree in physics and a PhD in mechanical engineering.
