Structural vibration monitoring of wind turbines

Transcription

1 Porto, Portugal, 30 June - 2 July 2014 A. Cunha, E. Caetano, P. Ribeiro, G. Müller (eds.) ISSN: ; ISBN: Structural vibration monitoring of wind turbines A. Mostböck, Y. Petryna Department of Civil Engineering, Chair of Structural Mechanics, Technische Universität Berlin, Gustav-Meyer-Allee 25, D Berlin, Germany annabell.mostboeck@tu-berlin.de, yuriy.petryna@tu-berlin.de ABSTRACT: The present work tests two different techniques for monitoring of carrying structure of wind turbines. A traditional one is based on strain measurements in critical cross-sections. The second one is based on use of the GPS sensors for displacement measurements of the tower. In combination with a validated finite element model, it is able to provide information on arbitrary displacements and strains in the carrying structure. A comparison of these two techniques is performed on a real structure under operational conditions and shows a good agreement. The cost problem for the GPS technique is still a challenge. KEY WORDS: Wind turbine, GPS technique; finite element simulation; structural monitoring 1 PREFACE 1.1 Motivation Wind turbines are exposed to extreme dynamic loads over a scheduled lifetime of usually 20 years. To increase the availability and efficiency of the plants and to minimize the operation breakdowns, both machines and supporting structures are generally to be monitored. The aim of monitoring is the early diagnosis and prevention of severe damages or failures. There are several condition monitoring systems (CMS) for the machinery, which are commercially available on the market. For other parts of the wind turbine, such as rotor blades, tower or foundation, different techniques are currently under testing. Especially CMS for the offshore carrying structures is a challenge. Figure 1. Illustration of the main idea 1.2 Main purpose The aim of the present work is to test the suitability of several techniques for monitoring of the carrying structure of wind turbines. The main physical values to be measured are strains and displacements at various locations on the tower. A new aspect in the present study appears due to the possibility to measure tower displacements directly by use of the Global Positioning System, or briefly GPS technique. It is also of interest to check the accuracy of such a technique in comparison with the other available techniques. The main idea of the present study is illustrated in Figure 1. A GPS sensor is placed on the nacelle of the wind turbine. It delivers data on the top tower displacements. The main interest is focused on the horizontal ones. The measured displacements are then applied to the finite element model of the tower in order to calculate the resulting displacements and strains at any location of the tower. They can finally be compared with the actually measured ones, for example, the strains in the bottom cross-section of the tower (Figure 1). Some conclusions on the accuracy, efficiency and costs of the applied techniques shall be made afterwards. 2 WIND TURBINE The wind turbine under consideration is a typical 2 MW onshore wind turbine with a m high tower (Figure 1) that is fixed in a quadratic reinforced concrete slab of 15.6m side length. The steel tower is made of 5 conical factory-made sections of different length, connected with each other and with the basement by the flanges and pre-stressed bolted joints. The wind turbine has a total weight of 335 t without foundation. The turbine house weighs 68 t, the rotor 36 t and the tower itself 231 t. According to the construction data, the soil stiffness under the slab shall provide the rotating stiffness of the tower equal to C ϕ,dyn Nm/rad. The rotor of a 90 m diameter runs at a nominal speed of 8.8 to 14.9 rounds per minute that corresponds to the operational frequency of and Hz respectively. Due to three 3643

2 rotor blades on the rotor, there appear also the triple operational frequencies between and Hz. The current study is focused on structural vibration monitoring of the carrying structure. Nacelle ACC 2 ACC 7 ACC 4 GPS ACC 8 ACC 5 ACC 1 ACC 6 ACC 3 DMS 1 DMS 70 ACC: accelerometer DMS: strain gauges Figure 2. Type and position of sensors on the tower. DMS 35 3 MEASUREMENT TECHNIQUES Wind turbines are usually equipped with their own operational monitoring systems, which are able to measure many functional parameters such as wind speed, rotational frequency of the rotor and some other special parameters such as pitch angle, temperature, power output and so on. There are also several condition monitoring systems installed on the engine and drive train. However, they are not in the focus of the present study. We are interested in the health monitoring of the carrying structure, which mainly consists of the tower and the foundation. We use in the present work typical vibration monitoring techniques such as strain gauges and accelerometers as well as a new technique based on the GPS sensors. The type, position and number of sensors installed on the tower are given in Figure 2. The accelerometers (indicated in Figure 2 as ACC) are used for measurement of structural behavior during operation and for the identification of modal properties. They are placed in 3 cross-sections along the towers height in order to be able to catch spatial displacements of the tower. The arrows of ACC in Figure 2 indicate the measured direction. Since horizontal displacements of the tower are dominant, longitudinal accelerations, i.e. vertical ones, have not been measured. 6 strain gauges (indicated in Figure 2 as DMS) are installed along the bottom cross-section of the tower to measure vertical strains at the tower clamping. Only three of them are necessary to identify the strain state in the cross-section. And only three of them are also depicted in Figure 2. Other ones serve for redundancy of the measurements. Finally, we use a global positioning system (GPS) to measure actual displacements of the towers top. A GPS sensor is installed on the top of the nacelle (Figure 2) and is able to catch all three spatial directions. 3.1 Global Positioning System The global positioning system (GPS) used in this project is of the differential type. The differential GPS (DGPS) can considerably increase the accuracy of the position determination compared to usual GPS systems. They exploit the fact that the position signals are generally received by two devices simultaneously: a user device on the structure and a receiver on a reference station near to the structure. The position of the reference station should be determined precisely in advance. [1]. The operational scheme of DGPS is depicted in Figure 3. It is essential for DGPS that direct communication between the user device and the reference station should be possible. Usually, it is the case if a visual contact between them exists. The advantage of the operation with two devices is that possible deviations of the both received signals can be reduced or eliminated afterwards by special algorithms [1]. The GPS device for this study has been provided by Alberding GmbH. It is a brand Trimble two-frequency device of the type BD982, which operates with a sampling frequency of 1 Hz. The precision of the horizontal positioning is declared to be up to 8 mm + 1 ppm. This equipment was installed in a plastic box on the nacelle s roof. 3.2 Accelerometers In this study, we use industrial accelerometers of the type PCB 393A03 manufactured by PCB Piezotronis, Inc. This equipment has a sensitivity of 102 mv/(m/s 2 ). According to the datasheet, the specified measuring range is equal to ±49 m/s 2 (±5 g). The sampling frequency was 1 khz. The accelerometers have been fixed to the flanges of the tower shell inside the tower by magnets. 3.3 Strain Gauges The strain gauges used are of the type PL produced by Tokyo Sokki Kenkyujo Co., Ltd. Their feature is a k-factor of 2.12 and a resistance of 120±0.3 Ω. They have been applied 3644

3 on the inner surface of the tower shell in vertical direction, approximately 0.1 m above the top edge of the flange. Due to a short measurement time of only about twenty minutes at once, the temperature-dependence of the strain gauges has been ignored in the present study. Since the temperature is permanently measured, it makes no difficulties to take it into account. The measured signals of the accelerometers and strain gauges have been processed by an analogue-digital transducer and recorded by the same portable computer. In this way it is ensured, that the signals have been recorded simultaneously. The recording of the GPS data has been carried out separately. Several trigger signals have been used for synchronization of the recorded GPS data with the signals of accelerometers and strain gauges. Besides, several test measurements have been performed in advance to check the equipment and to identify the modal parameters of the structure. Figure 4. Partition of the acceleration measurement in components of operation and out-of-operation. Figure 4 depicts the measurement signal of an acceleration sensor. The signal is partitioned in eight intercepts according to operation and out-of-operation phases. The operation phases are characterized by high amplitudes between 0.4 and 0.8 m/s². In Figure 4, the uneven numbers were assigned to these interceptions and they are marked by red hatched boxes. The out-of-operation phases are distinguished by small random acceleration amplitudes of about 0.05 m/s². In Figure 4 the even numbers were assigned to these partitions, moreover they are marked by green filled boxes. Figure 3. Operation scheme of the differential global positioning system (DGPS). 4 TEST PROGRAMM The present study is directed first of all towards testing of monitoring techniques. The long-term monitoring of the wind turbine itself is not discussed here. A typical operation regime of the wind turbine includes a short start phase, a long operation phase with permanent control of the whole facility and a short switch-off phase, in the case of unexpected events or scheduled operational breaks. Besides, the facility can rashly be shutdown in the case of critical or safety-relevant events. However, such an operation breakdown can cause severe electrical or mechanical problems and is, therefore, rather exceptional. For testing purposes in this study, we utilize a usual operational sequence containing the start phase, the operation phase and a switch-off phase, which follow after each other four times. Each operation phase and out-of-operation phase lasts approximately two minutes. The whole reference sequence took about twenty minutes. This reference sequence has been recorded by the installed equipment. Figure 5. Measured horizontal nacelle displacements. Figure 5 depicts a horizontal displacement of the nacelle during the whole measurement cycle. One can clearly recognize four different types of behavior. The out-of-operation response is characterized by small random vibrations around a zero position. They are caused by 3645

4 random wind perturbations of the resting facility. This type is indicated in Figure 5 by red hatched boxes. The tower vibration during operation is characterized by strongly irregular random vibrations around some static displacement state depending on the wind speed. In our case, the mean tower displacement under operation varies between 0.3 and 0.8 m. The corresponding displacement intervals are indicated in Figure 5 by green filled boxes. In between the operation and the out-of-operation phases, one can recognize the start and stop phases of the generator. They are characterized by a rapid increase or decrease of the displacement, respectively, which are indicated by green and red arrows in Figure 5. During these phases, we can observe some typical natural vibration cycles of the tower superposed with the operation vibrations due to the triple rotation frequency of the rotor. During the switch-off phases, the tower vibrates in both directions, so that we can observe even negative displacements up to 0.4 m. However, they rapidly diminish and disappear after usually two cycles. In summary, we can observe quite different response vibrations during the reference measurement sequence, both qualitatively and quantitatively. They contain large mean static deflections (of ca. 0.3 to 1.0 m) due to wind pressure, natural vibrations of the tower due to the first eigenfrequency and random operational vibrations (of ca. ±0.1 m). 5 EXPERIMENTAL AND NUMERICAL MODAL ANALYSIS 5.1 Finite element model The carrying structure of the wind turbine consists of the tower and the foundation slab. In order to correctly calculate dynamic properties, the mass of the rotor and the rotor blades is assigned to the nacelle. The latter is modeled as a rigid body with real dimensions and an appropriate mass distribution. The soil is taken into account as elastic foundation. The finite element (FE) model has been developed and applied within the Abaqus software (Figure 6). The main requirements on the FE model concern the modal analysis and static deformation states. Thus, the discretization should be sufficient to calculate natural frequencies and mode shapes, displacements and strains with a required accuracy. The tower is modeled by shell elements to allow for the calculation of strains at arbitrary points over the height and the circumference. The real tower structure is composed of 36 segments with different cross-sections. In the finite element model, the thickness change is taken into account by a linear function over the height, from which a mean shell thickness in totally 12 segments is determined and assigned to the corresponding finite elements. The flanges are modeled as beams associated to the shell structure. Figure 6 shows the FE model in a deformed state according to the first mode shape. Besides, the bottom section of the shell possesses exactly the real thickness value and allows for a direct comparison of the measured and calculated strains. In the present study, linear shell elements of type S4R with reduced integration are applied. The same discretization with 12 elements in the circumferential direction is applied throughout the tower to ensure the mesh consistency. The element size in vertical direction is chosen in such a way that almost square grid occurs. Only the area of special interest in the bottom segment is modeled finely in order to be able to extract strain information at numerous positions according to the measurement program. The foundation slab is modeled by volume element of the type C3D8R. The material properties corresponds to the concrete actually applied, the reinforcement is not taken into account. Since a linear response of the structure is of interest, such a simplification is feasible. The FE mesh of the slab is generated so that the shell nodes are directly merged with the slab nodes. The mesh outside this connection is generated automatically. The bond of the slab to the ground is modeled as elastic foundation with the option provided by Abaqus. The individual sections of the towers shell as well as the flanges were connected with the program instruction tie. The nacelle was fixed to the top flange in the same way. Again only the part closest to the bottom was an exception: This section was joined up to the nearest part with the program instruction merge to part to avoid non-natural changes of strains in the model at the joint in the area where the strains should be explicitly extracted. At other positions of the model such small discrepancies could be tolerated. The connection of the tower and the fundament was realized as rigid joint with the program instruction tie. This assumption was considered to be acceptable because the tower is actually embedded in the basement slab. The nacelle is modeled as a distortion-free shell object with extra high stiffness in order to attach the mass of the rotor and rotor blades at proper locations and exclude undesired deformation. The rotor mass is attached to the nacelle s edge at a reference point. The vibrations of the rotor blades are neglected. Figure 6. Exemplary visualization of the FE-Model in a state of nacelle displacement. 3646

5 5.2 Comparison of modal characteristics The carrying structure of the wind turbine is quite simple. Hence, no complex parameter identification was necessary in this work. The results of the first numerical simulation already provide a satisfactory agreement with the measured natural frequencies. Due to a low scatter of the steel properties as well as a high manufacturing quality of the shell, there were very few model parameters to be adapted. Some improvement of the model, especially for the higher natural frequencies, could be achieved. However, the main dynamic properties should not change significantly due to some parameter variations. Nevertheless, such an improvement has been carried out by variations of the masses and Young s modulus. The first natural frequency of the tower f 0 was examined for two ratios of magnitude for the joint restraint. It was determined in a range of f Hz. The first value, the smaller one, corresponds to the minimum allowable stiffness of the soil, i.e. the minimum value of elastic bedding. The second value corresponds to the rigid foundation or an ideally clamped tower. The real eigenfrequencies of the structure have been measured on site under various temperature and weather conditions. Each time, the wind turbine has been switched off in order to exclude operational vibrations. Figure 7 depicts the frequency spectrum of the tower in an out-of-operation phase. It was calculated by the FFT of the measured acceleration signals. The peaks corresponding to the first and the second eigenfrequency of the tower are indicated by arrows. Along with the natural frequencies, there are some other frequencies, for instance, one at about 1.1 Hz. It might correspond to vibrations of other structural parts, for example, rotor blades. Figure 8 shows the frequency spectrum recorded during the operation phase by the same sensor as used in Figure 7. Besides of the increased response amplitudes, a significant gain of noise can be recognized. The first two tower eigenfrequencies can still be distinguished. There are also several pure operational frequencies, the most significant one lies at about 0.6 Hz (Figure 8). 1 st Mode, Tower 2 nd Mode, Tower Figure 7. Measured frequency response spectrum in the switch-off phase (intercept VI shown in Fig. 4). 1 st Mode, Tower 2 nd Mode, Tower Figure 8. Measured frequency response spectrum during operation (intercept VII shown in Fig.4). The measured and calculated natural frequencies of the structure are given in Table 1 for comparison. Since the tower is rotationally symmetric, there exist two almost equal eigenfrequencies measured in two rectangular directions. The difference between them is caused by the non-symmetrical distribution of the top mass due to rotor and rotor blades. Vibration frequencies of the other structural components appear without pendant, like the 4 th frequency in Table 1. Table 1. Comparison of the measured and calculated natural frequencies (all values in Hz) Natural frequencies Measured,July 2012 Calculated, FEM Direction f 1 f 2 f 1 f DATA PROCESSING AND RESULTS The complete measurement records in Figures 4 and 5 include quite different behavior of the structure. A large part of these records have been used for data processing and displacement or strain identification. At that, several subsets have been selected and processed to evaluate various response types separately, for example, vibrations with small and large amplitudes as well as a few operation cycles with switch-offs. In the present study, we select two data sets for comparison, denoted as FE GPS and SG MAX, respectively. They are generated by different measurement techniques, but for the same time intervals. A description of the applied approaches and data generation are given below. 3647

6 6.1 Data Set: FE GPS The first data set is based on the GPS displacement measurements in combination with the finite element model. The values of the top tower displacement recorded by the GPS device at discrete time points have been applied to the FE model as static boundary conditions. A stress-strain state of the structure due to a unit top deflection can be calculated in advance and, thus, builds a basis for the application of arbitrary top deflections. The only condition to be satisfied for a desired accuracy is that the actual structural state is well described by such a simulation. It is usually the case, if the material response is linear elastic and the second-order effects are negligible. In order to determine any structural displacement or strain, one needs only to multiply the presolution by the value of the actual top displacement. For the purpose of comparison, a special attention is paid to strain locations on the tower that correspond to the installed strain gauges. The measured strains could be compared to the calculated ones, if the direction of the wind coincides with the direction of the calculated tower deflection. Since the wind direction varies, it is not always the case. A rational way is to compare the maximum strains in any cross-section including that one with the strain gauges. The FE model delivers the maximum strains directly. The measured strain values, at least three ones in number, could also be utilized to determine the maximum strain in the cross-section, if a linear strain distribution over the cross-section is assumed. Under assumption of geometrically and physically linear behavior of the structure, we calculate the strain state due to the measured top tower displacement. The maximum strain in the cross-section, where strain gauges are installed, forms then the first data set. At that, a linear relationship between the tower displacement u GPS (t i ) and the local strain of interest ε FE (t i ) can be described by a proportionality factor K ε : ε FE ( ti) = K ε ugps ( ti). (1) Factor K ε depends on the local position and can be identified by FE calculations of the available measurement data. The first data-set FE GPS is visualized in Figure 10 by a red thick line. 6.2 Data Set: SG MAX The second data set SG MAX contains maximum measured strains in the reference cross-section of about 0.1 m above the tower bottom. These strains are not measured directly but calculated from three data sets of strain gauges in the same cross-section. As mentioned above, a linear strain distribution over the cross-section of the tower (due to the Bernoulli hypothesis) allows defining a strain plain by use of at least three discrete strains and elementary rules of linear algebra. Then, the slope of the strain plane and the maximum or minimum strain value in the cross-section can also be determined. Figure 9 illustrates this idea in the vector space. As a result, we obtain a data set of the maximum measured reference strains at the tower bottom. This data set is drawn in Figure 10 by a thin black line. Figure 9. Calculation of the strain plane by means of linear algebra. 6.3 Comparison and analysis Figure 10 shows a comparison of two data sets for maximum strains in a reference cross-section, which are described above. One set is based on the displacement measurement on the top and a calculation of the resulting strains by the finite element method. The second one is directly based on the measured strains in a reference cross-section and their processing with respect to the maximum strain. As can be seen from Figure 10, both sets agree very well. The data sets cover a time interval with two operation and out-of-operation phases with quite different response types and vibration magnitudes. Nevertheless, the agreement is in all phases very well. It can be seen that the measured strains exhibits numerous small fluctuations (black thin line) whereas the displacement-based set shows a smoothed behavior (red thick line). The difference is caused by a limited accuracy of the displacement measurement by the GPS technique, which is unable to catch very small vibrations. It could be a drawback for the GPS technique, if fatigue loading is of interest. Figure 10. Comparison of two different measurement techniques for strain monitoring: red thick line corresponds to the data set FE GPS ; black thin line to the data set SG MAX. 3648

7 7 CONCLUSION AND OUTLOOK The obtained results show that the GPS measurement techniques in combination with validated finite element models are able to provide sufficiently accurate information on the arbitrary displacements and strains of the structure. This presumes, however, that the structural model describes the global and local behavior of the real structure sufficiently well. Under such conditions, GPS techniques could generally be applied in SHM systems. Their success in practical applications depends, however, on the following problems. To achieve the required precision of the tower top displacements in a range less than 0.01 m, a differential technique using a fix reference station close to the wind turbine was applied. Such an approach is generally possible for onshore wind turbines and impossible for the offshore wind parks. The price of the GPS technique is currently too high for wide-spread applications on wind turbines. In contrast, strain gauges represent a low-cost solution for the same problem. As can be seen from the obtained results, strain measurements in combination with a validated FE model are able to provide the same results as the GPS technique. Thus, the price is an economic challenge for the GPS. However, direct displacement measurements on the tower could be still of interest for some applications. The next drawback of the GPS technique concerns the online data processing. Currently, measured GPS data have to be processed afterwards in order to get real displacement values. The corresponding algorithms and software are currently under development. ACKNOWLEDGMENTS The authors gratefully acknowledge permission for measurements and support of the owner of the wind turbine, Notus Energy Plan GmbH & Co. KG (Potsdam, Germany). Special thank goes to H. Ziese, who spent much time on site helping us. This support has made our study generally possible and comprehensive. Besides, we would like to thank Prof. F. Neitzel for cooperation and expertise in GPS techniques. Special thanks goes also to the stuff members of our team from the Technische Universität Berlin: A. Künzel, W. Walkowiak, F. Vogdt, S. Weisbrich and K. Wezka for planning, instrumentation, measurement and data processing. The GPS sensors and equipment has been provided by J. Alberding and Dr. C. Clemen from the Alberding GmbH (Schönefeld, Germany), that is also gratefully appreciated. REFERENCES [1] W. Mahnsfeld, Satellitenortung und Navigation, Vieweg & Sohn Verlag, Wiesbaden, Germany,

8