ABSTRACT

Within the scopes of a current EC-funded research project, Acoustic Emission (AE) monitoring has been extensively applied during laboratory testing of various Wind Turbine Blades. Baseline static and fatigue full-scale testing up to failure was performed on a series of small-scale blades. Loading included several load-and-hold periods at intermediate stages prior to failure, and has enabled the development of a fast and effective "damage grading" system for the blades (from "minor damage" to "severe damage"). Grading was based on the amount of critical AE hits appearing during the hold periods. Such hits are identified and sorted out automatically, after each period, using trained Supervised Pattern Recognition algorithms on the acquired AE data, incorporated into specially developed (within the scopes of the Project) software, while, furthermore, the blade is graded using the software’s predefined grading criteria. The whole procedure is applied on load-hold data, at anytime in-between testing stages and is implemented during testing of large, commercial scale blades. Ultimately, the grading technique will, also, be used for the evaluation of installed blades, with proof tests and on-site monitoring.

1. INTRODUCTION

Operating Wind-Turbine (W/T) blades are subjected to complex loading sequences, due to the stochastic nature of wind conditions on wind-turbines sites. The experimental (or numerical) simulation of such loading conditions is rather complicated and the suitability of a W/T blade to operate on a specific site is, currently, verified through a certification procedure, which entails the conduction of a series of static and fatigue laboratory tests on the W/T blade. The purpose of such tests is to ascertain that the blade can survive the applied (static and fatigue) loads as per the applicable design standards [1], [2]. The applied static loads aim to simulate the site and W/T-specific extreme loaded cases (e.g. the "1-in-50-years gust") and are applied on the blade for ten seconds. Subsequently, the same blade is subjected to an accelerated 20-years fatigue lifetime test.

Fig :

In the aforementioned procedures the pass/fail criteria for W/T blades are based on deflection and strain measurements on the blade’s surface, during loading. Damage characterisation is, usually, performed by visual inspection for evaluating surface or visible damage, or through post-mortem examination, by cutting the blade, for assessing internal damage. Audible sound emissions heard during such certification tests are considered as potential damage indications, however, the location of the source of such noises is not always possible.

In this respect, Acoustic Emission (AE) monitoring was applied during both static and fatigue testing of W/T blades, within the scope of a current, EC-funded research project (AEGIS), aiming to provide a more reliable technique for the assessment of W/T blades’ structural integrity. A series of ten (10) similar, small-scale, 4.5m-long blades (Fig. 1), made of Fibre Reinforced Plastic (FRP, Glass/Polyester material), specifically manufactured for the project, were loaded up to final failure using various static and fatigue loading envelopes. Additionally, one commercial-scale blade (of Glass/Epoxy material) has very recently (June 2002) been loaded to failure by static testing and a similar one is being presently fatigue-tested and expected to fail by autumn 2002. The applied static testing loading envelopes varied from simple, monotonic, stepwise load increases (Fig. 2) up to failure with ten-minute load holds at various load levels, to more complex loadings were applied, incorporating a mixture of "certification-type" loadings (ten seconds load increase, ten seconds load hold, load decrease) to gradually increasing loads until failure and ten-minute load-hold loadings (AE "proof-type" tests) at, increasing load levels, performed in-between the certification-type tests (Fig. 4, upper part).

With the use of AE monitoring during testing of these small-scale blades [3], [4] damage occurring during the certification testing (both static and fatigue) can, in most of the cases, be well located with AE, and "weak" areas of the blade are indicated at early loading stages by high rates of AE. Additionally, the AE "proof-type" tests (with ten-minute load-holds) applied before and after certification-type static tests, as well as before and during the fatigue tests (at various times during testing) provide indications of the criticality of the damage (if any) introduced by the test, based on "emission during load-hold" criterion. During testing of the small-scale blades, it was commonly seen that the AE features of the recorded AE signals during load-holds (e.g. Amplitude, Duration, etc.) exhibited similar "patterns" close to failure loads. In general, the intensity of the recorded signals increased as the damage was becoming more critical and as the loads were approaching the failure loads. The consistency in the presence of this family of "critical" AE data right prior to failure has enabled the formulation of criteria for the assessment of the blade’s ability to withstand specific loads. Association of such data with a specific FRP failure mechanism (e.g. fibre breakage or matrix cracking etc.) is under investigation. With the use of a Pattern Recognition and Blade Grading software (the "AEGIS" software), specially created for the Project, critical AE data can be automatically identified and quantified, for any given AE data set, and each section of the tested blade (i.e. the area around each sensor) can be graded from "A" (minor damage) to "E" (severe damage) for a specific load, based on the number of critical AE data recorded per AE sensor (AE channel) during the load-hold.

In the present paper, results from the test to failure of two different blades will be presented; AEGIS Project’s "Small Blade #7" (small-scale blade, manufactured by Geobiologiki S.A., see photo on the left) and "Large Blade #1" (commercial-scale blade) that were both statically loaded to failure. The derivation of the grading technique, based on results of "Small Blade #1" is, also, presented. The tested blades presented herein were tested at the Centre for Renewable Energy Sources, while additional blades (not presented) were loaded at Delft University of Technology. For all tests presented, multi-channel SPARTAN-2000 Acoustic Emission system by Physical Acoustics Corporation (PAC) was used, with PAC-R6I (60kHz resonant) AE sensors with 40dB integral preamplifier.

DERIVATION OF THE GRADING METHOD

Small Blade #1 (Figure 1) was loaded to failure with the loading envelope shown in Figure 2. The blade final failure area was at 2,300mm from the root, close to sensor 8. Delamination was observed at the bottom section of the root area (sensor 2, not shown). The blade failed by buckling at 7KN load.

Fig 1: AE sensor positions, AE channel numbers, and load application point of Blade #1.

Fig 2: Cumulative AE hits vs. time, with load superimposed.

Unsupervised Pattern Recognition (UPR) was applied on the data of the last load-hold (at 6.5KN) prior to failure using the "AEGIS" Pattern Recognition and Blade Grading software [5] especially developed for the Project. A "first-hit" analysis was followed, i.e., only the first hit of each AE event was used. Before application of UPR an appropriate set of AE features was selected comprising Counts to Peak, Energy (MARSE), Duration, Amplitude and Average Frequency. All selected features were normalised individually to a range from 0 to 1. UPR was performed using the K-Means [6] algorithm and yielded three classes of AE data.

Figure 3 refers to the data from the load-hold at 6.5KN, classified into three classes by UPR. A thorough examination of the data of Class "0" reveals that this class has an intense rate of AE Energy during load-hold, as opposed to the other classes (Fig. 3, top left). Additionally, AE hits of this class have high Amplitudes in the load-hold (Fig. 3, bottom left), and high Counts values (Fig. 3, bottom right). Finally, the vast majority of hits of this class appear mainly in the channels close to the failure area (Fig. 3, top right). Class "0" was characterized as "critical" AE class and was used as the basis for the blade grading method. It is worth mentioning that there is overlapping of the individual AE signal features between classes, indicating that similar classification results could not be achieved by conventional 2-dimensional analysis of the data. For example, statistical analysis by the AEGIS software revealed that Energy values of class "0" (critical class) range from 11 to 4000 while for class "2" they range between 6 and 561, whereas the mean values are 270 and 30, respectively.

Fig 3: Referring to the 6.5KN load-hold (as in Fig.2): Cumulative AE Energy vs. time (top left), AE Amplitudes vs. Time (bottom left), AE Amplitudes vs. channel (top right) and Counts vs. Amplitude (bottom right ).

Following the clustering performed by UPR on the 6.5KN load hold AE data, a "k-Nearest Neighbour" (k-NNC) Supervised Classifier [7] was trained to correspond AE hits, from any given data set, to one of the three classes based upon the values of their AE features. Subsequently, data from each one of the load-holds of the loading envelope of Blade #1 was classified separately and the amount of data falling into the critical class was observed.

As a result, a colour-coded grading strategy was formulated which grades the damage of the blade (for a specific load) based on the number of AE hits which are classified with the critical class. E.g. for a channel to get a "B" grade, 10-25 critical class hits must have been recorded by this channel during the ten-minute load-hold. This same grading strategy was applied on subsequent blades’ load holds, (at various loading stages) and results from Small Blade #7 Large Blade #1 are presented below.

ANALYSIS OF SMALL BLADE #7

Fig 4: Loading envelope of Blade #7 indicating loading stages (top), blade and sensors layout and damage information.

The blade layout, and sensor positions are presented in Figure 4 (bottom). Load application point was 3,000mm from the blade root. The applied loading envelope is shown in Figure 4 (upper part), where the loading stages and the sustained damage are, also, indicated. It can be observed that the loading envelope included both certification-type tests (e.g. the MTL11 test) and AE proof-type tests (e.g. tests AE7a to AE7d). Apart from an artificially imposed skin delamination which was located between 2m and 2.2m from the root, on the sensors’ line, the blade sustained damage during the test at various positions. Visual inspection during and after the test revealed (see also Fig. 4 bottom):

• Delamination at the root area, visible at later stages of the test,
• Two cracks at the compressive side, perpendicular to the blade axis, at 1.1m and 1.2 m from the root, visible after the MTL11 certification-type test, and,
• One crack at the compressive side, perpendicular to the blade axis, at 2.2m from the root, developed after the MTL11 certification-type test.

The blade failed at the position of the crack at 2.2m, during the MTL21 certification-type test at 22.8 KN, slightly before this load was achieved.

Each AE proof test was graded using the AEGIS software and the grading strategy defined in the previous paragraph, herein. Grading of some characteristic periods is presented in Table 1. It is worth noting that, prior to loading MTL5, the grading did not reveal any grades other than "A" or "no-grade" and the blade did not exhibit any visible damage. The grading of sensor 1 (closest to the Delamination area) and sensor 3 (closest to the crack formed during MTL11 test) gradually increases from "A" (period AE7D) to "E" (period AE11E) giving a very good warning of the impending damage, i.e. the crack first seen after period MTL11. During the periods AE21A to E, sensor 5 was graded "E", while the blade failed catastrophically during the next test (MTL21) exactly in the area of sensor 5. Grading of the rest of the periods is not presented due to space limitations.

ANALYSIS OF LARGE BLADE #1

The blade layout, and sensor positions are presented in Figure 5 (bottom), where sensors 1 to 11 lie on the 25% chord line. The blade was loaded with two hydraulic actuators and the load application points were at 8,000mm and 10,000 mm from the blade root. The applied loading envelope (percentage of the actual ultimate failure load during this test) is shown in Figure 5 (upper part) for each actuator, where the loading stages and the sustained damage are, also, indicated. The loading envelope included both certification-type tests (e.g. the MTL3 test) and AE proof-type tests (e.g. tests AEL3A to AEL3B). The blade sustained very minor visible damage during the test at various positions. Visual inspection during and after the test revealed (see also Fig. 5 bottom):

Fig 5: Loading envelope of Large Blade #1 indicating loading stages (top), blade and sensors layout and damage information (bottom).

• Two very small cracks (5cm – 8cm length), at 470mm and 720mm from the root, that did not propagate, thereinafter,
• A small area of surface bubbles at 1,980mm from the root and 400mm from the 25% chord line, which, also, did not propagate thereinafter.

The blade failed near the 1st saddle area, in an area spanning from about 6,900mm to 9,000mm from the root, during the 5th MTL test at 103.9KN, immediately when this load was achieved.

In Figure 6, only results from the certification type tests are being presented. As can be observed, most of the AE activity emanated during these tests was linearly located, using sensors from 1 to 11, in the areas close to the root and close to the first actuator, where final catastrophic failure occurred. Such locations were observed already from the first test. During the last certification type test, significant AE is observed in the area between the two actuators, right before failure. During the AE proof tests (Fig. 7) the AE activity was located mainly in the area of the first actuator.

Fig 6: Linearly located AE events' axial position on blade (y-axis) vs. time (top), and loading envelope for the MTL tests only. Blade superimposed for reference.

Fig 7: Linear location history (x-position vs. time, top), and loading envelope for the AE proof tests only. Blade superimposed for reference.

The grading strategy derived from Small Blade #1 and applied on Small Blade #7, was, also, applied on Large Blade #1. Indicative grading results are presented in Table 2, below. The grading did not reveal severity of the loads applied during the AE proof tests, which were performed at a maximum of 64% of the ultimate failure load. An "A" grade was given to channels close to the hydraulic actuator, where the final failure occurred, and to channel 14 (during the last AE proof test only).

CONCLUSIONS

Acoustic-Emission-based criteria and procedures for the assessment of the structural integrity of wind turbine blades and of their ability to withstand specific load levels have been defined and incorporated within specialized software, created for the scope of an EC funded R&D Project. The procedures involve static loading of the blade and load-hold at the load of interest. AE activity during the load-hold is classified by a trained supervised classifier, and each zone of the blade is, subsequently graded, from "A" to "E" based on the number of AE hits classified with a critical class (revealed from loading to failure of a small blade) from each corresponding AE sensor. A grade "A" indicates that there is minor or non-critical damage on the blade, while an "E" grade indicates that the applied load causes severe damage on the blade. The grading method has been calibrated during testing of small-scale blade and it has been verified in a series of static and fatigue tests to failure of similar small-scale wind-turbine blades, made from Glass/Polyester material. In Small Blade #7, an "E" grade was assigned right prior to ultimate failure (or extreme damage). The same grading strategy was applied on a large, commercial-scale blade, made from Glass/Epoxy and statically tested to failure. The blade failed rather suddenly, by buckling-type mechanism, close to the first load application point during a certification-type test. Location of AE throughout testing indicated that the area around the first load application point and certain areas in the root section were quite emissive, already from early load stages, and, thus, may have been characterized as weak areas. Visual inspection of the large blade after each certification-type and AE-proof test revealed no significant visible damage prior to failure. Grading of the intermediate AE proof tests conducted during the test revealed only "A" grades (minor damage) for the AE proof test loads, which went up to 64% of the ultimate failure load. Apart from the apparent absence of severe damage on the blade prior to failure, the low number of "critical" AE data could, also, be attributed to behavioural differences of the Glass/Epoxy material compared to the Glass/Polyester material or the larger size of the blade compared to the small-scale blades.

Future work will include application of the same grading strategy during fatigue loading of a similar, commercial-scale blade, made from Glass/Epoxy material in order to assess whether the method can properly identify criticality of propagating damage, or if further calibration may be needed.

REFERENCES

1. IEC 61400-1, Wind turbine generator systems - part 1: safety requirements, 1998.
2. IEC 61400-23 TS Ed.1: Wind turbine generator systems - part 23: Full scale structural testing of rotor blades for WTGS's, draft 1999.
3. Joosse, P., A., Blanch, M., J., Dutton, G., A., Kouroussis, D., A., Philippidis, T., P., Vionis, P., "Acoustic Emission Monitoring Of Small Wind Turbine Blades", 21st ASME Wind Energy Symposium, 14-17 January 2002, Reno, U.S.A.
4. Blanch, M., J., et al, "Analysis of Acoustic Emission, Acousto-Ultrasonic and Infra Red Thermography Data from Ten Small Wind Turbine Blade Tests", Global Windpower, 2-5 April 2002, Paris, France.
5. Nikolaidis, V., N., Kouroussis, D., A., Anastassopoulos, A., A., "AEGIS Software Manual", Envirocoustics A.B.E.E. Technical Report, Ref:TR-015-10/2000, Athens, October 2000.
6. Philippidis, T. P., Nikolaidis, V. N., Anastassopoulos, A. A., "Damage Characterisation of C/C laminates using Neural Network Techniques on AE signals", NDT&E International, Vol. 31, No 5, Elsevier 1998, pp. 329-340.
7. Anastassopoulos, A. A., Nikolaidis, V. N. and Philippidis, T. P., "A Comparative study of Pattern Recognition Algorithms for Classification of Ultrasonic Signals", Neural Computing & Applications, 1999, Vol. 8, pp 53-61.

© NDT.net - info@ndt.net

|Top|