METHOD FOR DETECTING A SENSOR FAULT IN A LOAD SENSOR OF A WIND TURBINE

Description

[0001] The invention relates to a method for detecting a sensor fault in a load sensor of a wind turbine. The invention is also directed to a wind turbine in which such a method is implemented.

[0002] Wind turbines are designed to generate electrical energy from wind. The wind exerts enormous forces on the rotor and rotor blades of the wind turbine. This exposes the rotor blades to extreme stress. The rotor blades' own weight also exerts additional forces on the rotor blades.

[0003] A wind turbine, and especially its rotor blades, must therefore be designed to withstand these loads. In particular, for wind turbines whose rotor blades can be individually adjusted in their angle of attack, measurements of the resulting loads may be necessary. In wind conditions that could lead to high loads, it must at least be ensured that the wind turbine can be operated safely and shut down if necessary.

[0004] Therefore, it is necessary to measure the loads acting on the rotor blades.

[0005] Typically, load sensors can be attached to the rotor blades to measure the loads.

[0006] However, the unfortunate case may occur that at least one or all of the sensors fail or exhibit another malfunction, such as a functional freeze. In this case, the load calculation is no longer guaranteed. If the sensor freezes functionally, i.e., outputs constant signals, there is a risk that the sensor malfunction will not be detected, because a lack of change in the sensor value for a while could also correspond to the actual situation.

[0007] This problem could be avoided, for example, by installing additional redundant sensors on the rotor blade. The resulting loads could then be determined redundantly, and any deviations could be detected as an error. However, this has the significant disadvantage of being very costly, especially when multiple wind turbines have to be installed, as is common in wind farms.

[0008] The document DE10219664 discloses that sensor signals from different rotor blades can be compared to identify a sensor fault.

[0009] The object of the invention is therefore to solve one of the problems mentioned and, in particular, to detect a sensor malfunction in a load sensor cost-effectively. At the very least, an alternative solution to previously known solutions is to be proposed.

[0010] According to the invention, a method according to claim 1 is proposed for this purpose. Thus, a method is proposed for detecting a sensor fault in a load sensor of a wind turbine having a rotor and at least one rotor blade. The load sensor is designed to detect a load variable of one of the rotor blades.

[0011] The load sensor can thus be a measuring device or a measuring instrument that detects a load variable. Such a load variable can be a tensile force acting on the rotor blade, e.g., a weight force, or an extension of the rotor blade, from which the tensile force can be derived. This, in turn, can be used to derive a load acting on the rotor blade. If a large tensile force or extension occurs, or if other forces and / or moments are present, then the rotor blade is subject to a large load.

[0012] In particular, the load sensor can be a strain gauge that measures the strain or torsion of the rotor blade.

[0013] The sensor malfunction occurs when a sensor signal from the load sensor is functionally frozen, so that it remains constant over time. The sensor then outputs a constant measured value.

[0014] A measured value or sensor signal is almost never completely constant. Slight fluctuations are to be expected, for example, due to signal interference or noise caused by the electronics. The sensor, or a connected signal processing unit such as an A / D converter, will therefore not output an exact value of, for example, zero, or a precisely constant value, even though the sensor is frozen.

[0015] The sensor signal of the load sensor is thus functionally frozen when the load quantity that physically acts on the rotor blade changes, but this change is not detected by the load sensor; thus, the sensor signal does not change over time except for some noise, although a change would be expected.

[0016] A particularly problematic aspect is that the sensor signal can be almost constant without any disturbance being present, because at the moment there is actually no, or only a negligible, change in the load magnitude physically occurring. It is therefore not trivial to distinguish between a truly constant sensor signal, for which no physical change is to be expected, and a frozen signal with noise behavior.

[0017] Nevertheless, the load sensor signal is recorded and checked for its temporal stability. Furthermore, at least one load variable on the rotor blade is estimated. In particular, a thrust force on the rotor blade can be estimated from a measured wind speed, or a gravitational force from the known weight of the rotor blade and its current position.

[0018] If the sensor signal is constant over time, a comparison test is performed. During the comparison test, depending on at least one estimated load variable, it is checked whether the sensor signal is expected to be inconsistent. A sensor fault is identified based on the results of the comparison test.

[0019] The purpose of the comparison test is to determine whether the constant sensor signal was unexpected. If the sensor signal is constant even though it was unexpected, this indicates a sensor malfunction.

[0020] The sensor fault can thus be identified if the sensor signal is constant, although a change is expected due to the comparison test.

[0021] An estimated load value is a load that is not directly measured but derived from other values. The first step is to estimate the actual load acting on the rotor blade. The estimated load value does not necessarily have to correspond exactly to an expected sensor signal. The purpose here is simply to verify the general functionality of the sensor.

[0022] Therefore, it is not important to estimate the load precisely; what is important is to check whether a load change is present at all. Measurement sensors that only measure loads in their measuring direction can, however, measure a load component in their measuring direction for loads with a load direction deviating from the measuring direction. Such a load component can be sufficient to verify whether load changes are present, without knowing the exact load direction.

[0023] According to one aspect, the load sensor is designed as a strain gauge sensor. The load sensor is arranged on the rotor blade to detect a blade load via a strain of the rotor blade. In particular, the load sensor can be mounted on a blade surface.

[0024] The load sensor is firmly attached to the rotor blade, specifically in the area of ​​the rotor blade root. If the rotor blade bends, this also causes the sensor to stretch. The load sensor can measure the resulting bending and record the applied load. Conventional strain gauge sensors, also known as strain gauges, can be used. Thus, the proposed method can also be used to test a conventional strain gauge sensor for malfunctions, particularly without additional hardware or sensors.

[0025] According to a further aspect, the at least one estimated load variable is a gravitational load acting on the rotor blade. A blade load results from a weight force acting on the rotor blade. In particular, the estimated gravitational load is determined as a function of a rotor position of the rotor.

[0026] The rotor blade's own weight creates an acting force, which leads to a bending moment on the rotor blade, known as a gravitational load. However, the bending moment changes as the rotor blade's position changes.

[0027] For example, if the rotor blade is at a 6 o'clock or 12 o'clock position, i.e., vertical, the bending moment, i.e., the gravitational load, is zero. On the other hand, the gravitational load is maximum when the rotor blade is at a 3 o'clock or 9 o'clock position, i.e., horizontal.

[0028] The rotor position can also be referred to as the rotor angle. The rotor position thus indicates where the rotor blade is located in the rotor plane and can be expressed as an angle.

[0029] Typically, the estimated gravitational load curve follows a sinusoidal curve depending on the rotor angle or rotor position. For a more precise determination of the gravitational load, the angle of attack of the rotor blade can also be taken into account. The angle of attack influences the direction in which the load acts on the rotor blade, i.e., the load direction. Each load sensor is permanently assigned to a direction, which can be referred to here as the sensor direction or measurement direction. If the load direction changes due to an adjustment of the angle of attack, e.g., away from the sensor direction, the amplitude of the load component pointing in the measurement direction decreases. The load sensor then detects a reduced amplitude.

[0030] The angle of inclination of the rotor plane relative to a vertical plane or the angle of inclination of a rotor axis relative to a horizontal axis can also be taken into account. The angle of inclination also influences the direction of the load.

[0031] According to a further aspect, the at least one estimated load quantity is a wind load dependent on a wind speed acting on the rotor blade.

[0032] The wind's effect on the rotor blade depends on the wind speed. The higher the wind speed, the stronger the effect on the rotor blade.

[0033] Wind load is therefore understood to mean forces, in particular thrust forces, and / or moments, such as bending moments, which act on the rotor blade due to the thrust exerted by the wind.

[0034] In particular, the rotor blade experiences lift due to the wind, which depends on the wind speed. Loads caused by lift, such as lift force and / or bending moments, are also considered wind loads on the rotor blade. In this sense, the wind acts on the rotor blade.

[0035] The wind load also depends on the wind direction. In particular, for the purposes of this application, a wind load is understood to be a wind load in the wind direction.

[0036] If the estimated load is a wind load, this determines which loads act on the rotor blade due to the wind, depending on the wind speed. A wind load is a load that is caused by the wind and acts on the rotor blade. It can occur or be recorded as a bending moment, i.e., a torque, in the area of ​​the blade root. This bending moment or torque can constitute this load variable. For this purpose, the wind speed can be measured, e.g., using a LiDAR measurement and / or an anemometer measurement. LiDAR measurements or anemometer measurements provide a simple way to determine the wind speed. The load variable, i.e., the wind load, can then be derived from the wind speed, if necessary, taking into account the angle of attack of the rotor blade and / or the angle of inclination as well as the rotor speed.

[0037] The wind load is preferably determined as a function of a state variable that depends on the wind speed. In particular, such a state variable includes generated power, acceleration power, and / or a measured generator torque. The angle of attack can also be a state variable.

[0038] To estimate the wind load, at least one state variable is used. This state variable can itself be dependent on the wind speed.

[0039] The state variable can be generated power and / or acceleration power. The acceleration power of the rotor is calculated taking into account the rotor speed, rotor acceleration, and a rotor moment of inertia and is representative of the power absorbed from the wind. In particular, a power representative of the wind speed is considered as a state variable. Thus, the power absorbed by the rotor as a function of the wind speed is representative of the wind speed. The wind speed and / or the load variable can also be determined, in particular estimated, taking into account the generated electrical power and the rotor acceleration. The wind load can also be determined or estimated directly from the power. For the wind load to be estimated M and the power P as a state variable, the following relationship can be assumed: M ∝ P .

[0040] The wind load M is therefore modeled as proportional to the square root of the power P. From the recorded power, in particular from the recorded output power of the wind turbine, the load size can then be directly determined using a proportionality factor, which can be known from measurements or empirical values.

[0041] According to a further aspect, the comparative test checks whether the sensor malfunction is present depending on at least two estimated load variables. The first load variable is the estimated gravitational load. The second load variable is the estimated wind load.

[0042] It is particularly advantageous to consider both the estimated wind load in the wind direction and the estimated gravitational load. These two load variables are perpendicular to each other, so that for each rotor blade angle of attack, at least one load in the measurement direction can be estimated. If only load variables in one direction are estimated, in some situations, depending on the angle of attack and the sensor, it would not be possible to determine whether a variable sensor variable is to be expected.

[0043] By taking both the gravitational load and the wind load into account, it is ensured that, regardless of the angle of attack of the rotor blade, at least one load component in the measuring direction of the load sensor is present at all times and can be taken into account, if a corresponding load is present at all.

[0044] During the comparison test, each load variable, especially the change in each of the load variables, is taken into account. It is sufficient to determine, based on one load variable, that the sensor signal is not constant, i.e., a change in the sensor signal is to be expected.

[0045] According to a further aspect, a projected load magnitude is determined as the estimated load magnitude projected onto a measuring direction of the load sensor. The projected load magnitude is thus the estimated load magnitude projected onto a measuring direction of the load sensor.

[0046] The load sensor measures acting forces or other quantities in its measuring direction. If the load is not directed exactly in its measuring direction, it can only detect a portion of it, namely a projected portion, i.e., the projected load.

[0047] Due to the measuring direction, occurring loads may only be measured partially. To assess whether a sensor malfunction is present, it is therefore advantageous to project the estimated load magnitude onto the measuring direction of the load sensor. The projected load magnitude is thus the load that the load sensor should measure.

[0048] The projected load quantity can be either an estimated wind load projected onto the measuring direction of the load sensor or an estimated gravitational load projected onto a measuring direction of the load sensor.

[0049] The measuring direction of one load sensor can, for example, correspond to a swivel direction and that of another load sensor to a flapping direction of the rotor blade, at least when the angle of attack is 0°. Such an arrangement is particularly useful when only two load sensors or two pairs of load sensors are used. For this purpose, the load sensors are usually arranged on the blade surface in the blade root area in the longitudinal direction of the blade, but offset by 90° from each other around the blade root. Exact positioning is not important, however. The method can also be used to detect a sensor fault in a load sensor with a different sensor arrangement. This can be taken into account in the corresponding projection.

[0050] During the comparison test, depending on the projected load value, it is checked whether the sensor signal was expected to be constant or whether it was expected that it would not be constant.

[0051] In particular, the comparative test takes into account the change in the projected load magnitude. If the sensor signal is constant, this may also be due to the load sensor not detecting any component of the load magnitude due to its measuring direction. The decisive factor in determining whether the sensor signal should have changed is therefore the proportion of the estimated load magnitude in the measuring direction, i.e., the projected load magnitude. In particular, the comparative test examines, depending on both projected load magnitudes, whether it was to be expected that the sensor signal would not be constant.

[0052] Depending on the measurement direction and projection, it is possible that a first load variable would not be noticeable in a sensor. The estimated first load variable would therefore lead one to expect a constant signal. However, a different second load variable would be noticeable in the same sensor, and thus also any changes in the load. If this second load variable changes, one would expect the sensor signal to be non-constant, which would lead to a frozen sensor. Therefore, it depends on whether it was to be expected that the sensor signal would not be constant. By checking whether it was to be expected that the sensor signal would be constant, one also indirectly checks whether it was to be expected that the sensor signal would not be constant.

[0053] At least two load sensors are connected to the rotor blade. The load sensors have different measuring directions. The load sensors are located primarily on the rotor blade surface near the blade root and measure strain or compression in the longitudinal direction of the rotor blade. This strain or compression is attributed to a bending moment, or a bending moment is determined from it.

[0054] It is particularly well-suited to use the load sensors in pairs, with two load sensors of a pair being arranged around the rotor blade at a 180° angle to each other. They are essentially arranged opposite each other. Put simply, one sensor is at the top, one at the bottom, or one is at the front and one at the back. With this paired arrangement, the bending moment can be derived from the difference between the strains of the sensors in the same pair, considered with the correct sign. With two sensor pairs, two bending moments are then recorded, each with a bending axis around which the bending is directed. These two bending axes are ideally perpendicular to each other. In this respect, the bending directions, and thus the measuring directions, are perpendicular to each other.

[0055] It is therefore particularly advantageous if the measuring directions of the load sensors are perpendicular to each other.

[0056] According to a further aspect, if it has been detected that the sensor signal is constant over time, it is determined during the comparison test depending on the at least one estimated load variable, in particular depending on the at least one projected load variable, whether it was to be expected that the sensor signal would not be constant over time. The sensor fault is identified depending on whether it was to be expected that the sensor signal would not be constant over time. As a precaution, it should be noted that an actual load variable is never exactly constant, but it can exhibit fluctuations that are so small that they lie below a limit value that is so small that values ​​below it can be assumed to be constant. This is what is meant by constant here.

[0057] First, it is checked whether the sensor signal is constant over time. If this is not the case, the sensor is obviously not frozen. However, if it is determined that the sensor signal is constant over time, it is checked whether it was expected that the sensor signal would not be constant, i.e., whether a change was to be expected. For this purpose, the estimated load value is examined during the comparison test. If this is also constant over time, the constant sensor signal is explicable. Otherwise, it can be assumed that the sensor is frozen.

[0058] It is particularly advantageous to consider the projected load magnitude to assess whether the sensor signal was expected to be non-constant over time, i.e., whether it was expected to change. Since the sensor signal only measures loads proportionally due to the measurement direction, it is advantageous to project the estimated load magnitude onto this measurement direction, i.e., to use the projected load magnitude.

[0059] If the sensor signal is constant over time and, depending on the estimated load value, a temporally constant sensor signal can also be expected, no sensor fault is identified. On the other hand, a sensor fault is identified if the sensor signal is constant over time but the estimated, projected load value is not. It is conceivable that the load value is not constant, but its change after projection in the measuring direction has such a small amplitude that a temporally constant sensor signal is nevertheless plausible and can therefore be expected. This is the case, for example, if the measuring direction of the load sensor is approximately perpendicular to the load and the proportion that the load sensor can measure is negligible.

[0060] Whether the sensor signal was expected to be inconstant over time may also depend on both the estimated wind load and the estimated gravitational load.

[0061] If, for example, there is no wind load that leads to a bending moment with a direction that does not correspond to the measuring direction of a load sensor, this measuring device can still detect part of the wind load as long as the wind load is not perpendicular to the measuring direction of the load sensor. However, if this load sensor cannot detect the wind load or can only detect it too weakly because it is generally small and / or is completely or almost perpendicular to the measuring direction, it can still be recognized that the sensor signal should have changed if a change is expected due to changing gravitational loads. Such gravitational loads are essentially perpendicular to the wind load in the wind direction and therefore at least one of these two loads, i.e. wind load or gravitational load, will be detected by the load sensor in question due to its direction.

[0062] Gravitational loads in particular can be estimated precisely, so that the estimated gravitational loads are preferred for the assessment, provided they are suitable in terms of their direction and amplitude.

[0063] According to a further aspect, a test deviation is calculated during the comparison test to assess whether the sensor signal was expected to be non-constant over time. The test deviation represents the difference between the projected load magnitude at a current point in time and the projected load magnitude at a point in time when the sensor signal was last detected as not being constant over time.

[0064] The test deviation is compared with a specified limit value. If the test deviation does not exceed the limit value, i.e., if the change is very slight at most, then it can be expected that the sensor signal is constant over time, or at least within a tolerance. In particular, different limit values ​​are specified for the different load variables considered. In particular, a different limit value is specified for the test deviation for the gravitational load and for the test deviation for the wind load.

[0065] The projected load value is thus continuously determined and stored, e.g., in discrete time steps. As soon as it is detected that the sensor signal is constant over time, the projected load value is no longer stored. The test deviation represents the difference between the projected load value at the current time and the last saved projected load value. The test deviation thus represents the expected, but not yet occurred, change in the sensor signal.

[0066] As long as the test deviation does not exceed a specified limit, a change in the sensor signal is not to be expected. Consequently, a constant sensor signal does not lead to the identification of a sensor fault. The test deviation therefore remains within a specified tolerance range.

[0067] By specifying different limit values ​​for the test deviation for different loads taken into account, it is taken into account that the lower the specified limit value can be chosen, the more accurately a load can be estimated.

[0068] According to a further aspect, if the sensor signal was detected as temporally constant, but after the comparison test, it was not expected to be temporally constant, a counter is incremented. When the counter reaches a predetermined threshold, the sensor fault is identified.

[0069] This checking of whether the sensor signal has been detected as being constant over time, incrementing the counter depending on this and checking whether the predetermined threshold value has been reached is repeated successively, in particular continuously.

[0070] The advantage of such a counter is that the sensitivity to sensor failure detection can be adjusted. A sensor signal that remains constant over time, which was unexpected, initially only results in the counter incrementing. This prevents individual outliers in the estimation from falsely detecting a frozen sensor, without having to set the limit for the test deviation so high that errors remain undetected.

[0071] According to a further aspect, the counter is reset if it is detected during the comparison test that the sensor signal has changed. The counter is thus reset if a sensor signal initially detected as constant is later detected as not constant. The counter is thus reset to an initial value, in particular to zero.

[0072] The idea behind this is that a single change in the sensor signal proves that the sensor is not frozen. The counter, and thus the entire test routine, can then be completely reset.

[0073] If, however, the sensor signal is detected as being constant over time and at the same time it is expected that the sensor signal is constant, the counter is left at a current count.

[0074] In both cases, the first step is to check whether the sensor signal has changed. If the sensor signal has changed, the counter is reset, because in this case the sensor is obviously not frozen.

[0075] If the sensor signal has not changed and is recognized as constant over time, the comparison test checks whether the sensor signal should have changed.

[0076] If a change in the sensor signal is also not expected, the counter is left at the current count, because this does not indicate whether the sensor is frozen or not.

[0077] If it is detected that, despite constant behavior, the sensor signal should have changed, e.g., due to the test deviation exceeding the limit, the counter is incremented, as this indicates a frozen sensor. The criterion for incrementing the counter, described above, is therefore still applied.

[0078] Thus, according to the aspects that use the counter, the counter is evaluated in successive time steps. For this purpose, a threshold value is specified for the counter. The sensor fault is detected, or the sensor fault is identified, when the counter reaches or exceeds the threshold value.

[0079] The counter evaluation is repeated successively. Thus, the comparison test is used to repeatedly check whether the sensor signal is not constant over time. In particular, the test is repeated until the counter has reached the threshold.

[0080] The counter is thus evaluated iteratively. Fixed time steps can be specified for the evaluation. For example, the counter can be evaluated every 10 ms. Only when the counter reaches or exceeds the threshold is the sensor fault identified. However, if the sensor signal changes prematurely, the counter is reset and no sensor fault is identified.

[0081] The process then starts again, with the counter set to zero, as soon as a temporally constant sensor signal is detected again.

[0082] According to a further aspect, during the comparison test, if the sensor signal is constant, a gravity load comparison is performed as a comparison based on an estimated gravity load, and a wind load comparison is performed as a comparison based on an estimated wind load. For each test, an identical counter is incremented by an increment if it has been determined that the sensor signal is not constant. A sensor fault is assumed when the counter has reached a predetermined threshold. Preferably, a different increment is used for the gravity load comparison than for the wind load comparison.

[0083] The comparison test is therefore carried out using both an estimated gravitational load and an estimated wind load. To check whether the sensor signal is expected to be inconsistent, the respective projected load magnitude can be used.

[0084] For this purpose, the test deviation can be calculated for both the gravity load comparison and the wind load comparison to determine whether the sensor signal should have changed.

[0085] The gravity load comparison and the wind load comparison are performed independently of each other. However, the same counter is always incremented as a result, preferably with different increment values. If a change is expected as a result of a comparison test, the counter is not incremented by one across the board, but by the increment value, a variably specified constant. Therefore, it is sufficient if one of the comparison tests results in the sensor signal being changed to cause the counter to increment.

[0086] If, based on an estimated load variable, e.g., the gravitational load, the comparison test results in a constant sensor signal being expected, the counter would initially be left at its current count. However, if the other comparison test—in this example, the wind load comparison—results in a change in the sensor signal being expected, the counter is incremented by the increment corresponding to the wind load or wind load comparison. It is therefore sufficient if one load variable is suitable for testing, especially if it is suitable due to its direction relative to the sensor's measuring direction.

[0087] However, if both loads are suitable for testing, the counter may be incremented more quickly, allowing a fault to be detected more quickly. Considering a single counter also has the advantage of allowing a sensor fault to be identified particularly quickly.

[0088] The increment value itself can be dependent on an operating point of the wind turbine for each comparison test. In particular, it is proposed to set the increment value depending on the speed (of the rotor and / or the generator) and / or the power of the wind turbine, which can individually or together characterize the operating point. In particular, the increment value for gravitational loads can be selected as a function of the speed, i.e., depending on the speed of the wind turbine, i.e., the rotor and / or the generator of the wind turbine. The counter will then increase faster during the gravitational load comparison, the higher the speed.

[0089] The increment value can be selected to predetermine how quickly the sensor fault will be identified. The "how quickly" can refer to a specific time, a comparison event, or a comparison behavior.

[0090] For the gravity load comparison, this could be, for example, a rotation of the rotor by a certain angle. The increment value can therefore be selected so that after a rotation of the rotor of, say, 90°, the threshold value would be reached. If the counter weren't also incremented by the wind load comparison, it would be even faster.

[0091] With wind load comparison, however, the increment value is selected so that the threshold is reached after a predetermined period of time at the latest, e.g., after 4 seconds. Thus, if the sensor signal remains constant for 4 seconds, while the wind load comparison does not predict a constant value over time, the counter reaches the threshold and the sensor fault is detected. If the counter were not also incremented by the gravitational load comparison, this would occur even faster.

[0092] By selecting the increment values, you can specify how quickly a sensor malfunction is identified, should it occur. This has the advantage that errors in the comparison test do not immediately lead to the identification of a sensor malfunction, thus avoiding misidentification.

[0093] If several loads occur, the different increase values ​​lead to the same counter counting up, whereby the threshold value is reached particularly quickly and the sensor fault is quickly identified.

[0094] According to a further aspect, the sensor signal is assumed to be noisy. The sensor signal is recognized as temporally constant if the sensor signal fluctuates or varies by a signal value and a predeterminable fluctuation amplitude or predeterminable fluctuation amplitude is not exceeded in magnitude.

[0095] It was recognized here that an exactly constant signal cannot be expected, even if the sensor is frozen. Therefore, a change is only detected as a change if the sensor signal changes by more than a threshold. The fluctuation amplitude or variation amplitude forms such a threshold. In order to still be able to detect whether the sensor signal is frozen, a fluctuation amplitude or variation amplitude is permitted; both can also be referred to as permitted noise amplitude. This means that a fluctuation by a signal value is permitted, which is still assumed to be a temporally constant sensor signal. The threshold can also be specified. If the sensor signal changes by less than the specified threshold within a time step, a constant signal is detected.

[0096] According to the invention, a wind turbine comprising a load sensor, a rotor, and at least one rotor blade is further proposed. The wind turbine is configured to implement a method for detecting a sensor fault according to one of the preceding aspects. In particular, the wind turbine comprises a measuring and control device on which the method for detecting the sensor fault is implemented.

[0097] The present invention will be explained in more detail below by way of examples with reference to the figures. Like reference numerals represent the same or similar elements. Figure 1 shows a perspective view of a wind turbine. Figure 2 shows two load sensors mounted on a rotor blade. Figure 3 shows the method according to the invention. Figure 4 shows a diagram for determining the test deviation. Figure 5 shows the incrementation of a counter due to an expected gravitational load. Figure 6 shows the incrementation of a counter due to an expected wind load. Figure 7 shows timing diagrams illustrating the incrementation of a counter.

[0098] Fig. 1shows a schematic representation of a wind turbine according to the invention. The wind turbine 100 has a tower 102 and a nacelle 104 on the tower 102. An aerodynamic rotor 106 with three rotor blades 108 and a spinner 110 is provided on the nacelle 104. During operation of the wind turbine, the aerodynamic rotor 106 is set in rotation by the wind and thus also rotates an electrodynamic rotor or rotor of a generator, which is directly or indirectly coupled to the aerodynamic rotor 106. The electrical generator is arranged in the nacelle 104 and generates electrical energy. The pitch angles of the rotor blades 108 can be changed by pitch motors at the rotor blade roots 109 of the respective rotor blades 108.

[0099] A load sensor is connected to at least one of the rotor blades 108. In particular, each rotor blade 108 has at least two, in particular two pairs of, load sensors.

[0100] Figure 2 shows a rotor blade 108 to which a first and a second load sensor 112, 114 are attached. The load sensors are arranged such that they are offset by 90° around the rotor blade.

[0101] If the rotor blade bends, the load sensors 112, 114 measure a strain. The strain measured by the first load sensor 112 is converted into a first bending moment 132, and the strain measured by the second load sensor 114 is converted into a second bending moment 134. For this purpose, a first bending axis 122 is defined for the first load sensor 112, and a second bending axis 124 is defined for the second load sensor 114, around which the bending moments 132 and 134, respectively, are calculated.

[0102] Both bending axes 122, 124 are perpendicular to each other, and thus also the bending moments, namely in the sense that the bending moments 132, 134, which are to be determined with the two load sensors, have mutually perpendicular bending axes 122, 124. The directions in which the bending moments point can be regarded as measuring directions, so that in this sense the measuring directions are also perpendicular to each other. This results from the offset arrangement of the load sensors 112, 114 by 90°. The first load sensor 112 is therefore mounted in a first strain direction 142 along the longitudinal direction of the rotor blade, and the second load sensor 114 is mounted in a second strain direction 144, also along the longitudinal direction of the rotor blade.

[0103] Figure 3 shows the method according to the invention.

[0104] First, a sensor signal from the load sensor is recorded.

[0105] In a constancy test step 210, a check is then made to determine whether the sensor signal is constant over time, i.e., whether the sensor signal is functionally frozen. Fluctuations, e.g., caused by evaluation electronics, are also recognized as constant over time.

[0106] If the constancy test step 210 detects that the sensor signal is constant over time, an expectation test step 220 is performed. The comparison test is then performed in the expectation test step 220. A further check is performed to determine whether the temporally constant sensor signal was to be expected, or whether it was to be expected that the sensor signal would not be constant. This is done depending on at least one estimated load variable of the rotor blade.

[0107] If it is determined in the expectation step that a temporally constant behavior is not to be expected, a counter is incremented in an increment step 240.

[0108] Otherwise, if it is determined in the expectation step 220 that the time-constant sensor signal was to be expected, the counter is left at the current count.

[0109] This is followed by an evaluation step 250, which is carried out in any case, even if the counter was incremented in the increment step 240.

[0110] If, instead, it is already determined in the constancy test step 210 that the sensor signal is not constant over time, or that the sensor signal has changed, a reset step 230 is performed. In the reset step 230, the counter is reset, ie, set to zero.

[0111] Following the reset step 230, the evaluation step 250 is also carried out.

[0112] In evaluation step 250, the counter is evaluated. Depending on the count, a sensor fault is detected.

[0113] If a sensor malfunction is detected, a warning message is generated. If necessary, the wind turbine is shut down or switched to a safe operating mode.

[0114] If no sensor fault is identified based on the count in evaluation step 250, the process is restarted at the next time step, where a sensor signal from the load sensor is again acquired and, in constancy test step 210, it is checked whether the sensor signal is constant over time. The process is thus continuously repeated.

[0115] The procedure is therefore largely based on the result of the expectation step 220.

[0116] For this purpose, a load value on the rotor blade is estimated in the expectation step. A distinction is made between a gravitational load and a wind load. In the expectation step 220, a gravitational load and a wind load are estimated.

[0117] Both the estimated gravitational load and the estimated wind load are projected onto the load sensor's measuring direction. A test deviation is determined for both projected load variables. The test deviation is the difference between the projected load variable at a current point in time and the projected load variable at the point in time at which the sensor signal was detected as temporally constant.

[0118] Any test deviation that exceeds a predetermined limit, whether for the estimated gravitational load or for the estimated wind load, results in an increment of the counter in the increment step 240. However, different increment values ​​can be assigned to the different load variables, by which the counter is then incremented.

[0119] Figure 4illustrates the determination of the test deviation of a projected load quantity. The figure shows the coordinate system 300, in which the load quantity L is plotted on the ordinate against time t on the abscissa.

[0120] The measured sensor signal 310 is shown as a solid line. Also shown, as a dashed line, from time t 1 onward, is the projected load magnitude 320, i.e., the expected load magnitude projected onto the measuring direction of the load sensor.

[0121] In the range from time t0 to time t1, the sensor signal still changes with a significant amplitude. At time t1, the sensor signal is detected as temporally constant. Time t1 is therefore the time of the last signal change.

[0122] At time t 1 , i.e., at the time of the last sufficiently large signal change, the projected load variable 320 has the value P 1 . In subsequent time steps, the test deviation is then successively determined as the difference between the current projected load variable and the projected load variable P 1 at the time of the last signal change t 1 . As long as the test deviation does not exceed a limit value G in terms of magnitude, the sensor signal is expected to be constant over time.

[0123] At time t 2 , the projected load quantity 320 reaches the tolerance range and has a value of approximately P 2 = P 1 + G. For larger values ​​of t, the projected load quantity lies outside the tolerance range. Consequently, the test deviation from time t 2 is greater in magnitude than the limit value G.

[0124] Starting at time t2, a change in the sensor signal is therefore expected. In the next expectation step 220, it is consequently recognized that a constant sensor signal is not to be expected, and the increment step 240 is executed. Any number of time steps can occur between t1 and t2. However, time t2 can theoretically also be the immediately next time step following time t1.

[0125] This example also shows a high degree of absolute agreement between the sensor signal and the projected load magnitude. However, since only the presence of changes above thresholds is considered, this is not necessary. The same result would also be obtained if the sensor signal 310 and the projected load magnitude 320 were shifted relative to each other in the ordinate direction.

[0126] Figure 5shows, as an example, the counter incrementing when only gravitational loads act on the rotor blade, or at least only these loads change. In coordinate system 400, the test deviation for the gravitational load ΔP G is plotted on the ordinate as a function of time t on the abscissa.

[0127] The determined test deviation 410 is plotted as a time course, as is a limit value 420 with the value GG , above which a change in the sensor signal is expected due to the gravitational loads occurring.

[0128] At time t 2 , the test deviation 410 exceeds the limit value 420, so that a change in the sensor signal is expected.

[0129] Coordinate system 400 also has an additional ordinate for the change in rotor position Δϕ. Starting at time t 2 , the change in rotor position is determined, i.e., the number of degrees by which the rotor blade continues to rotate. Curve 450 shows an example of a change in rotor position. A signal change is permitted until the rotor has rotated a predetermined value, e.g., 90°. In the example, this is reached at time t 3 .

[0130] The incrementation of counter Z is shown on the same time axis below the test deviation curve. The counter, or rather the selected increment value, together with the threshold value, is dimensioned such that a signal change for one rotor rotation of 90° is permitted before a sensor fault is identified. At time t 3 , the test deviation 410 is above the limit value 420 for the time required for one rotor rotation of 90° from time t 2 .

[0131] Coordinate system 470 thus shows the counter reading Z against the same time axis t. Counter 480 is standardized such that the sensor fault is identified when the value reaches 1.

[0132] From time t 2 , the test deviation 410 is above the limit value 420, so that the counter 480 is incremented. The increment by which the counter 480 is incremented is proportional to the change in the rotor position and is dimensioned such that the counter 480 reaches the threshold value 1 just after one rotor rotation of 90°, i.e., at time t 3 , and the sensor fault is identified.

[0133] Figure 6 shows a case in which only wind loads act on the rotor blade, or at least only these loads change. In diagram 500, the test deviation ΔP W , which results from estimated wind loads, is plotted as a curve 510 against time t.

[0134] The determined test deviation 510 is plotted as a curve, as is an associated limit value 520 with the value GW. If the test deviation 510 exceeds this limit value, a sensor change is expected. At time t 2 , the test deviation 510 exceeds the limit value 520, so a change in the sensor signal is expected.

[0135] The increment of the counter by the increase value due to wind loads is calculated in such a way that an expected but non-occurring sensor change is permitted for a duration of 4 s, which is indicated by the time difference 550. At time t 3 , the test deviation 510 was above the limit value 520 for a duration of 4 s, thus identifying the sensor fault.

[0136] Furthermore, the coordinate system 570 also shows a counter reading Z versus time t. Counter 580 is again standardized such that the sensor fault is identified when threshold 1 is reached. At time t 2 , test deviation 510 exceeds limit 520, which causes counter 580 to increment. The increment by which counter 580 is incremented due to wind loads is dimensioned such that counter 580 reaches threshold 1 after 4 s, and the sensor fault is identified.

[0137] Figure 7 illustrates the incrementing of a counter, especially the one in Figure 6 shown counter 580, and identifying the sensor disturbance as a function of both estimated load variables, namely the gravitational load and the wind load, wherein the simulation is based on a real situation.

[0138] This shows Figure 7three coordinate systems 600, 602, 604 for the same time, which is plotted on the abscissa of the coordinate systems 600, 602, 604.

[0139] In the topmost coordinate system 600, the recorded sensor signal 610 is shown as a solid line. The expected load values ​​from wind load 620 are also shown as dashed lines and gravitational load 630 as dotted lines. The load unit is kNm.

[0140] At time t 2 , a constant sensor signal 610 is detected. As a result, the test deviation for the gravitational load is determined and evaluated. If the test deviation for the gravitational load is below the limit value and thus does not cause the counter to increment, the test deviation for the wind load, namely in the wind direction, is determined and evaluated instead. For this purpose, the expected gravitational load and the expected wind load are projected onto the measuring direction of the load sensor, and the difference to the projected load value at time t 2 , the last signal change, is calculated.

[0141] The middle coordinate system 602 shows the determined test deviation ΔP in kNm for the wind load 622 (dashed line) and for the test deviation of the gravitational load 632 (dotted line). If the test deviation does not exceed the corresponding limit value, i.e., if it does not cause the counter to increment, the test deviation is set to zero. However, this is purely illustrative, showing when which criterion causes the counter to increment.

[0142] The coordinate system 602 thus shows only the time-proportionate test deviation which lies above the respective associated limit value and which leads to the counter counting up due to the operating point.

[0143] Different criteria or designs apply for the gravitational load and the wind load. For the gravitational load, the counter is incremented by an increment proportional to the rotor rotation, i.e., the amount by which the rotor blade has rotated since the last signal change. The increment is chosen so that a sensor fault is detected after a 90° rotation at the latest, if the gravitational load would be expected to change continuously.

[0144] The occurrence of wind loads causes the counter to increment by a different value. This value is selected so that, without the occurrence of gravitational loads, i.e., without changing gravitational loads, the sensor fault is detected after 4 seconds at the latest.

[0145] Since both mechanisms result in the incrementing of the same counter, simultaneous changes in wind loads and gravity loads can lead to faster identification of a sensor failure.

[0146] The lower coordinate system 604 shows the counter value Z versus time t. The counter 680 is standardized so that a value of 1 identifies a sensor fault and shuts down the wind turbine.

[0147] From time t 2 to time t 3, the test deviation 622 due to wind loads causes the counter 680 to increment, and between t 3 and t 4, the test deviation 632 due to gravitational loads causes the counter 680 to increment.

[0148] At time t1, a constant sensor signal was also briefly detected, but a change was expected due to test deviation 622. This caused counter 680 to increment. However, the threshold was not reached, and no sensor fault was identified. Afterward, the sensor signal changed again, so the counter was reset and therefore returned to the value zero until time t2.

Claims

1. Method for detecting a sensor malfunction of a load sensor of a wind power installation (100) having a rotor (106) and at least one rotor blade (108), wherein: - the load sensor (112, 114) is configured to detect a loading variable of one of the rotor blades (108), and - a sensor malfunction is such that a sensor signal (310, 610) of the load sensor (112, 114) is functionally frozen, such that it is temporally constant, wherein: - at least one loading variable (620, 630) of the rotor blade is estimated, - if the sensor signal (310, 610) is temporally constant, a comparability test is executed wherein, according to the at least one estimated loading variable (620, 630), a check is executed as to whether a non-constant sensor signal (310, 610) is to be anticipated, and - a sensor malfunction is identified by reference to the comparability test.

2. Method according to Claim 1, characterized in that the load sensor (112, 114) is configured in the form of a strain measurement sensor, and is arranged on the rotor blade (108), in order to permit the detection of a blade load by reference to strain in the rotor blade (108).

3. Method according to Claim 1 or 2, characterized in that the at least one estimated loading variable (630) is a gravitational load acting on the rotor blade, wherein a blade load is generated by a weight force acting on the rotor blade and wherein, in particular: - the estimated gravitational load is determined by reference to a rotor position of the rotor.

4. Method according to one of the preceding claims, characterized in that the at least one estimated loading variable (620) is a wind load acting on the rotor blade which is dependent upon a wind speed, and - the wind load is determined by reference to a measured wind speed, and the measurement of wind speed particularly comprises: - a LiDAR measurement, and / or - an anemometer measurement, or - the wind load is determined in accordance with a state variable which is dependent upon the wind speed, and the state variable particularly comprises: - a generated power, - an acceleration power of the rotor, and / or - a generator torque detected.

5. Method according to one of the preceding claims, characterized in that, in the comparability test, with respect to at least two estimated loading variables (620, 630), a check is executed as to whether a sensor malfunction is present, wherein: - the first loading variable is a or the estimated gravitational load (630), and - the second loading variable is a or the estimated wind load (620).

6. Method according to one of the preceding claims, characterized in that a projected loading variable (320) is determined, by way of the estimated loading variable (620, 630) projected for a measurement direction of the load sensor (112, 114), and - in the comparability test, according to the projected loading variable (320), a check is executed as to whether a non-constant sensor signal (310, 610) is to be anticipated.

7. Method according to one of the preceding claims, characterized in that: - if it has been detected that the sensor signal (310, 610) is temporally constant, - in the comparability test executed according to the at least one estimated loading variable (620, 630), and particularly according to a or the at least one projected loading variable (320), it is determined whether a temporally non-constant sensor signal (310, 610) was anticipated, and - depending upon the anticipation of a temporally non-constant sensor signal (310, 610) a sensor malfunction is identified.

8. Method according to one of the preceding claims, characterized in that, in the comparability test for the appraisal of whether a temporally non-constant sensor signal (310, 610) was to be anticipated, - a check deviation (410, 510, 622, 632) is generated in the form of a difference between a or the projected loading variable (320) at a current time point, and a or the projected loading variable (320) for a time point at which the sensor signal (310, 610) was last not detected as temporally constant, and - the check deviation (410, 510, 622, 632) is compared with a specified limiting value (G, GG, GW), and - if the check deviation does not exceed the limiting value (G, GG, GW), it is expected that the sensor signal (310, 610) is temporally constant, wherein, in particular: - different limiting values (GG, GW) are specified for the different loading variables (620, 630) considered, wherein, in particular: - a different limiting value (G, GG, GW) is specified in each case for the check deviation with respect to the gravitational load (510, 632) and for the check deviation with respect to the wind load (410, 622).

9. Method according to one of the preceding claims, characterized in that - if the sensor signal (310, 610) has been detected as temporally constant, - but, by the execution of the comparability test, it is established that a temporally constant signal was not anticipated, - a counter (480, 580, 680) is advanced, wherein - a sensor malfunction is identified, if the counter (480, 580, 680) has achieved a stipulated threshold value.

10. Method according to one of the preceding claims, characterized in that - if, during the comparability test, it has been detected that the sensor signal (310, 610) has changed, and has thus not been detected as constant, a or the counter (480, 580, 680) is reset, or - if the sensor signal (310, 610) has been detected as constant, and a constant sensor signal (310, 610) is anticipated, the counter (480, 580, 680) is left at its current reading.

11. Method according to one of the preceding claims, characterized in that - in the comparability test, if the sensor signal (310, 610) is constant, - a gravitational load comparison, by way of a comparison with an estimated gravitational load (630), and - a wind load comparison, by way of a comparison with an estimated wind load (620) is executed, wherein - in each check, an identical or the same counter (480, 580, 680) are respectively advanced by one upward increment, if it is established that a non-constant sensor signal (310, 610) is anticipated, and wherein a sensor malfunction is assumed, if the counter (480, 580, 680) has achieved a or the stipulated threshold value, wherein, in particular - in the gravitational load comparison, a different upward increment is applied to that employed in the wind load comparison.

12. Method according to one of the preceding claims, characterized in that the sensor signal (310, 610) is assumed to be noisy, and - the sensor signal (310, 610) is identified as temporally constant, if the sensor signal (310, 610) fluctuates or varies about a signal value, wherein a predefinable fluctuation amplitude or variation amplitude is not quantitatively exceeded.

13. Wind power installation (100) having a load sensor (112, 114), a rotor (106) and at least one rotor blade (108), characterized in that - the wind power installation (100) is configured to execute a method for detecting a sensor malfunction, according to one of the preceding claims, and - the wind power installation (100) particularly comprises a measurement and control device, on which the method for detecting the sensor malfunction is implemented.