High-frequency distortion phenomenon and solution of low frequency vibration sensor

This paper analyzes the difference between several acceleration over-limit faults, and finds out the cause of false acceleration fault of wind turbines, which is because the low-frequency vibration sensor used by the unit is interfered with the high-frequency signal generated by the friction of the yaw system and causes the sensor output signal distortion. The performance of different manufacturers’ sensors under different installation positions is reproduced on the spot. A cost-effective solution to the high frequency distortion problem (installing a physical low-pass filter) is developed, and the comparative test results before and after its application are demonstrated experimentally.

1. Instruction "Acceleration over-limit" is a common problem of wind turbines.Normally, it is the real acceleration of the wind turbines caused by abnormal wind resources or the control strategy instability.However, according to the data analysis, more than 80% of the "acceleration over-limit" faults are caused by abnormal conditions, which is due to the false positive phenomenon caused by the distortion of the vibration sensor signal acquisition, which seriously affects the MTBF [1] and power generation of the wind turbines.
At present, almost all wind turbine manufacturers think that the abnormal acceleration caused by the control strategy or abnormal wind resources.Therefore, the common but the wrong way to solve the problem is to optimize control strategy or reset the wind turbine.The fault can not be resolved, it still exists.
This paper analyzes the difference between several acceleration over-limit faults and find out a right way to solve the false acceleration fault.It can reduce nearly all the false "acceleration over-limit" faults.At present, the scheme has been applied in the Goldwind wind turbines (nearly 800 wind turbines within warranty period), which can increase direct or indirect revenue for more than 10 million RMB.

Terms and definitions
The following terms and definitions apply to this document.

Acceleration sensor
The sensing equipment used to detect the change of the displacement and acceleration of the object is divided into low frequency, medium frequency and high frequency according to the frequency range that it can detect, and there is no clear boundary.
2 Vibration sensors in wind turbines are generally low-frequency acceleration sensors, and their linear undamped response frequency range is generally lower than 200Hz, and the performance of different sensors is not consistent.
Same as vibration sensor below.

Frequency distortion
When the input signal of the amplifier circuit is a complex frequency signal, if the amplifier circuit has different gain amplitude for different frequency components of the signal, it will cause distortion of the output waveform, which is called amplitude distortion.If the relative phase shift changes, it is called phase distortion, and the two are collectively referred to as frequency distortion.What happens in the low frequency band is called low-frequency distortion, and what happens in the high frequency band is called high-frequency distortion.
Frequency distortion is caused by the linear reactance element of the circuit, and the frequency distortion is characterized by the distortion of the output signal compared with the input signal.The output signal produces a new frequency component that the input signal does not have, or the output signal has a large amplitude difference from the input signal [2] .

Vibration failure caused by control strategy
Through the analysis of the value of the vibration sensor, combined with the data of the generator speed, power, etc., as shown in Figure 1, it can be judged that the fault is caused by the divergence of control, and it is correct and necessary to trigger the fault shutdown [3] .

False fault caused by high frequency vibration
Data characteristics analysis: In terms of data characteristics, the vibration value floats away from zero for a long time in a single direction (that is, "zero drift", Figure 2 [4] ) or returns to zero soon after bias zero, but there is no reverse vibration value (that is, "jump", Figure 3) [5] .The normal vibration value is generally manifested as obvious oscillation divergence.Or shock vibration, will fluctuate up and down the zero position many times, similar to a simple pendulum motion.Therefore, the data is characterized as abnormal vibration data, and the possibility of wind turbines control strategy instability and shock vibration is excluded.

The basic situation of yaw
The main suspicion is that some factors in the yaw process cause the vibration sensor data to be too large and the fault is reported, so the wind turbine with frequent failure is selected for testing.
The manual yaw test is as follows: 1) There is impact vibration at the start and end of yaw, causing impact noise and slight cabin shaking (small wind manual yaw), which generally stabilizes after a few seconds.
2) In the process of continuous yaw, the nacelle is slight shaking, there are intermittent or continuous strong abnormal sound, part of the process noise has a beat which can obviously perceive the change of the beat.The sound is harsh and unbearable.
3) When yaw is abnormal, the vibration sensor value will change greatly.
The audio analysis software is used to analyze the frequency spectrum of the sound with normal yaw and the sound with harsh yaw: a) The concentration in sound with normal yaw is below 1KHz, and the maximum amplitude of frequencies above 1K is nearly 10dB smaller than the maximum amplitude below 1K, as shown in Figure 4. b) The main frequency of yaw sound is harsh above 1KHz, especially between 2K-3KHz, and the maximum frequency amplitude below 1K is nearly 25dB, as shown in Figure 5. c) The snippets of yaw sound from normal to abnormal in the screenshot were analyzed by short-time Fourier analysis.The darker the color, the larger the amplitude, the sound was normal in the first 5 seconds, the dominant frequency was concentrated below 1KHz, and the high frequency above 2kHz became the dominant frequency in the last 5 seconds, as shown in Figure 6.Due to the lack of professional high-frequency vibration detection equipment, the frequency spectrum components of the vibration generated by the yaw process are not clear at present, and can only be estimated through the sound frequency.The prominent frequency may be concentrated in 2K-3KHz, and the amplitude is large.

Test results of different manufacturers' sensors when yaw abnormal sound
Gsensor (Goldwind acceleration senor) in the nacelle cabinet: there is a zero drift phenomenon that the vibration data is not near 0 when it is not yawing.When yaw noise is abnormal, the vibration data will mainly show zero drift distortion, which can exceed ±0.2g in severe cases, and there is also oscillation distortion, as shown in Figure 7. Compared with the sensor PCH1216(Made by Denmark): when yaw abnormal sound, the vibration data is mainly manifested as oscillation distortion, but it is not easy to trigger, and the anti-high-frequency distortion ability is strong.There is no distorted data in the figure, but it can be seen that the burr phenomenon is intensified when the sound is abnormal, as shown in Figure 8.

Test conclusion
The following conclusions can be drawn by combining the fault analysis and comparative test data analysis: 1.The root cause of "Acceleration over-limit during yaw" fault in the field lies in the high frequency distortion and data anomaly caused by the high frequency vibration during yaw, and the secondary cause is the insufficient anti-high frequency performance of the sensor.
2.Different manufacturers of sensors have the problem of high frequency distortion, each anti-high frequency distortion ability is different, the phenomenon of high frequency distortion is not consistent, common zero drift distortion and concussion distortion.

Solutions
Considering the high frequency vibration sources and detection links, two solutions are proposed to solve the problem of fault false positives: From the source to weaken high-frequency vibration: the main work is brake disc grinding, cleaning, replacement of brake disc and even brake disc, etc.The solution is a mechanical program, is currently promoted in the field, but after the implementation of the program for a long time running will appear again high-frequency vibration, need periodic maintenance and costs are high; Install a physical low-pass filter: Install a physical low-pass filter, before the high-frequency vibration components are transmitted to the detection chip, first through a "low-pass filter", the signal attenuation to the sensor can work normally, to avoid high-frequency distortion.The "vibration sensor + physical low-pass filter" as a whole, the installation of physical low-pass filter is equivalent to enhancing the overall high-frequency suppression ability.The solution is easy to install and low cost.

Physical low-pass filter
The physical low-pass filter solution [6] , as shown in Figure 9, has the advantages of clear principle, low cost, easy installation, no need for additional support tooling, and strong applicability, but it is only used to ensure the reliability of sensor detection, and does not solve the root problem of high-frequency vibration.

Solution test
For low-frequency vibration sensors used for vibration detection of wind turbine, the installation of high-frequency vibration damper can effectively suppress the influence of high-frequency vibration components on the sensor, while it has no influence on the detection of low-frequency vibration.
Instruction: a) High frequency vibration is a relative concept, where it is the cut-off frequency relative to the high frequency vibration damper. b) The frequency of the wind turbine collecting vibration data is 50Hz.c) In order to avoid the phenomenon of frequency domain aliasing, the digital vibration sensor uses the digital eight-order low-pass filtering algorithm to filter the vibration value before output, and the cut-off frequency is 20Hz.The analog vibration sensor does not have this function, and there may be frequency domain aliasing problem.

Time series data analysis
Two sensors in the same (near the same) position, one has the damper and the other one does not have.High-frequency sampling data: refer to the high-frequency data collected synchronously ( not strictly synchronized).In the yaw process under maintenance mode, high-frequency vibration exists almost the whole time, accompanied by sharp sounds.
Access the main control data: When the vibration of high frequency sampling is prominent, the three sensors connected to the main control.The sensor without vibration damper has the serious jump and zero drift phenomenon, and the frequency and amplitude are large.The data waveform of the sensor installed with the damper is continuous, there is no jump and zero drift phenomenon, and there is burr phenomenon, but the amplitude is small, and the low-frequency waveform is not affected.
During the test, it was found that the data channel in Y direction where the vibration sensor with the damper was damaged, so the comparison test only analyzed the data in X direction.

Frequency domain analysis of high frequency data
According to the data analysis of comparative test, the larger the amplitude of high-frequency vibration, the greater the possibility of abnormal vibration sensor data connected to the main control.It is necessary to further analyze the high-frequency sampling data.
The X-direction data of a test with high-frequency sampling in the comparison test is selected to conduct short-time Fourier analysis.The deeper and brighter the color in the figure indicates the larger the vibration amplitude, which can be seen in Figure 10.
When the amplitude of high-frequency vibration is large, there are many peaks above 1000Hz, mainly concentrated in the vicinity of 2000Hz and 4000Hz.The prominent frequency will also change with the yaw process, and the external performance is abnormal yaw sound and the tone and intensity of the sound will change.There is a trend that the greater the amplitude of high-frequency vibration, the greater the yaw sound.
When the amplitude of high-frequency vibration is small, its frequency component is concentrated below 1000Hz, and the yaw sound is stable and low, and this part of the frequency component exists in the whole yaw process, and the intensity is small and basically unchanged.
Combined with the analysis of the data, it is found that the high frequency component and its intensity in the yaw process may cause the abnormal value of the vibration sensor.When the yaw is normal, the frequency components are concentrated between 500 and 2000Hz, and the prominent frequencies are 1295Hz, 1728Hz, etc., as shown in Figure 12, but the maximum value is much smaller than the amplitude when the yaw sound is abnormal, and the amplitude in the Z direction is the largest.Combined with modeling analysis, a low-pass filter is added before the signal is transmitted to the sensor.The vibration data collected by the high-frequency sampling device represents the original vibration data, and the digital filter is used for low-pass filtering, and the cutoff frequency of the digital filter is set to the cutoff frequency of the high-frequency vibration damping pad.Assuming that the detection of the vibration sensor can fully reflect the vibration signal of the sensor chip, the final data of the sensor can be replaced by the filtered data.

Test effect analysis
Only X direction data is selected for analysis.Although the amplitude of high-frequency sampling is greater than 3g, there are more high-frequency components.After low-pass filtering at 200Hz cut-off frequency, the high-frequency components are filtered out, while the low-frequency components are completely retained, and the amplitude of the first-order tower frequency is within the range of ±0.005g, as shown in Figure 13.The waveform of the sensor data installed with the damper and the data after digital filtering is continuous, without jump and zero drift phenomenon, as shown in Figure 14.During the period of 360 to 380 seconds, the high-frequency vibration component is large, resulting in the sensor data installed on the vibration absorption pad is still affected, and there is a burr phenomenon.

Conclusion
After the analysis of test data, the conclusion is as follows: 1) The installation of high frequency damper can effectively suppress the influence of high frequency vibration components on the low frequency vibration sensor, and has no influence on the detection of low frequency vibration components, which greatly improves the reliability of the sensor signal.
2) If the high-frequency vibration component of the installation point is too strong, there may still be a slight "zero drift" and "burr" high-frequency distortion phenomenon in the vibration sensor data, which is greatly reduced compared with the case of no vibration damper.

Figure 1 .
Figure 1.Typical vibration data for control strategy causing failure.

Figure 2 .
Figure 2. The characteristics of vibration curves of "zero drift" and "jump".

Figure 3 .
Figure 3. Vibration curve characteristics of "control strategy instability" and "shock vibration.

Figure 4 .
Figure 4. Sound frequency diagram when the yaw sound is normal.

Figure 5 .
Figure 5. Sound frequency diagram when the yaw sound is abnormal.

Figure 6 .
Figure 6.Plot of the sound frequency of yaw sound from normal to abnormal.

Figure 10 .
Figure 10.Short time Fourier analysis of high frequency vibration data.

Figure 11 .
Figure 11.Power spectrum analysis of high frequency acquisition with high amplitude.

Figure 12 .
Figure 12.Power spectrum analysis of high frequency acquisition with normal amplitude.

Figure 13 .
Figure 13.Time series curve before and after filtering.

Figure 14 .
Figure 14.Contrast and local curve magnification of sensor and filtered data.