Water-cooling leakage signal detection system for wind turbine generator

Abstract

The invention belongs to the technical field of water-cooling leakage signal detection, and discloses a wind turbine generator water-cooling leakage signal detection system which comprises a detection unit, a signal processing unit, a spectrum analysis unit and a self-learning unit. The detection unit can convert detected liquid motion into an analog signal containing Doppler frequency shift information; the signal processing unit can be used for conditioning and converting the electric signals and converting the analog signals subjected to lifting processing into digital signals; the spectrum analysis unit can generate a corresponding spectrogram and calculate the total energy and frequency variance of the Doppler frequency shift region based on the spectrogram, so as to quantify the leakage scale and dynamic characteristics of the leakage. According to the application, the frequency signal generated at the moment of leakage of the water-cooled liquid can be sensed in real time, and even a small amount of liquid drops can cause the change of the frequency signal, so that early warning is realized, and precious response time is won for operation and maintenance personnel.

Description

Technical Field

[0001] This application belongs to the field of water-cooled leakage signal detection technology, specifically relating to a water-cooled leakage signal detection system for wind turbine generators. Background Technology

[0002] With the increasing global demand for renewable energy, wind power has been widely adopted as a clean and renewable energy source. In large wind turbines, water cooling is typically used to ensure heat dissipation for core components such as the generator under high loads. However, due to the complex and harsh operating environment and the aging of the equipment itself, water cooling systems inevitably experience leaks. For pressurized water cooling systems, leaks, especially small initial leaks, often manifest as a high-speed, fine liquid jet. This jet produces specific acoustic characteristics: firstly, it generates broadband noise; secondly, the numerous high-speed liquid droplets in the jet can act as reflectors of sound waves.

[0003] Leaks can not only affect cooling efficiency and cause overheating damage to equipment, but also lead to serious safety accidents such as electrical short circuits, resulting in huge economic losses for wind farms. Therefore, developing an efficient and reliable water-cooled leak detection system for wind turbines is of significant practical importance.

[0004] However, most current wind turbine water-cooled leak detection systems currently use one or more sensors installed in low-lying areas at the bottom of the tower platform. When leaking liquid comes into contact with the two electrodes of the sensor, the loop resistance changes, triggering an alarm. However, this approach suffers from a very high false alarm rate in humid environments inside the tower, and it struggles to accurately detect small leaks in the early stages, thus compromising the accuracy and reliability of the detection. Summary of the Invention

[0005] This application provides a water-cooled leakage signal detection system for wind turbines, aiming to solve the problems of existing technologies having an extremely high false alarm rate when the internal environment of the tower is humid, and being unable to accurately detect small-flow leaks in the early stages of leakage, thus failing to guarantee the accuracy and reliability of the detection.

[0006] A water-cooled leakage signal detection system for wind turbine generators includes a detection unit, a signal processing unit, a spectrum analysis unit, and a self-learning unit;

[0007] The detection unit is constructed based on a sound wave transmitter and a sound wave receiver, and can convert the detected liquid motion into an analog signal containing Doppler frequency shift information;

[0008] The signal processing unit can condition and convert electrical signals to extract weak signals from noise and boost them to the volt level, and convert the boosted analog signals into digital signals.

[0009] The spectrum analysis unit can decompose a digital signal into a superposition of different frequency components through Fourier transform, generate a corresponding spectrum, and calculate the total energy and frequency variance of the Doppler frequency shift region based on the spectrum to quantify the leakage scale and dynamic characteristics; the spectrum analysis unit includes a generation subunit, a feature parameter extraction subunit, and a threshold subunit.

[0010] The self-learning unit is constructed based on the spectral characteristic curve under the leak-free state. It can determine whether the frequency shift spectral characteristic curve parameters exceed the preset threshold by recording the spectral characteristic curve and canceling steady-state noise interference in real time.

[0011] Furthermore, the detection unit also includes a transducer and a temperature compensation circuit. The transducer adopts a multi-layer stacked structure and is equipped with a metal backing mass block to form a resonant cavity to enhance the vibration amplitude.

[0012] The temperature compensation circuit monitors changes in the working environment in real time and automatically adjusts the drive parameters to offset the resonant frequency shift caused by temperature drift.

[0013] Furthermore, the specific contents of the signal processing unit include signal amplification and filtering, and signal conversion;

[0014] The signal amplification and filtering: The front end of the receiving channel of the signal processing unit is equipped with a preamplifier composed of low noise field effect, which amplifies the microvolt-level echo signal to the millivolt level, allowing only the frequency components related to the leakage signal to pass through.

[0015] Furthermore, the generating subunit can receive digital signals and decompose the continuous time series signal into a series of orthogonal basis functions linear combinations through Fourier transform to determine the complex amplitude value X at the corresponding frequency point k. k By projecting the time-domain waveform onto different frequency axes, the complex amplitude distribution of each frequency component is obtained, and the complex amplitude value X is obtained. k The calculation formula is as follows:

[0016]

[0017] Where x[n] is the time-domain sample, N is the data length, n is the time, k is the frequency coefficient, and j is the imaginary unit.

[0018] Furthermore, the generation subunit can also calculate the echo signal frequency change Δf caused by droplet motion, which is used to detect the echo signal frequency change of low-speed leaking liquid. The calculation formula is as follows:

[0019]

[0020] Where v is the radial velocity of the droplet, θ is the angle between the direction of ultrasonic wave propagation and the direction of droplet motion, and λ is the wavelength.

[0021] Furthermore, the feature parameter extraction subunit can perform intelligent feature analysis on the extracted Doppler spectrum: calculating the total energy E of the frequency shift component. doppler and droplet velocity value The total energy E doppler The calculation formula is as follows:

[0022]

[0023] Wherein, Ω represents the set of frequency ranges for the Doppler frequency shift components;

[0024] The droplet velocity value The calculation formula is as follows:

[0025]

[0026] Where μ is the average frequency, f i Here, M represents the frequency value of a single frequency point, and M represents the number of effective frequency points.

[0027] Furthermore, the threshold subunit can determine the total energy E. doppler and frequency variance Does it exceed the preset dynamic threshold?

[0028] Furthermore, after the self-learning unit enters the training mode, it will continuously collect environmental background noise data for a period of time, including signals generated by steady-state interference sources such as fan operation and mechanical vibration. After smoothing the original data through a moving average filter, a standardized spectral characteristic curve is generated.

[0029] Furthermore, the self-learning unit captures the monitoring spectrum feature curve of the current frame in real time and performs a point-by-point subtraction operation with the baseline spectrum in memory to determine whether the frequency shift spectrum feature curve parameters are potentially abnormal.

[0030] Compared with the prior art, this application has at least the following beneficial effects:

[0031] Based on further analysis and research of existing technical problems, this application utilizes the principles of acoustic wave transmitters, acoustic wave receivers, and spectral characteristic curves to detect in real time the frequency signal generated by water-cooled liquid at the moment of leakage. Even a small amount of droplets can cause changes in the frequency signal, enabling early warning and saving valuable response time for maintenance personnel.

[0032] Furthermore, analyzing the total energy and droplet velocity values ​​of the Doppler signal can provide a preliminary assessment of the severity of the leak. The total energy indicates the magnitude of the leak; a higher total energy suggests a more severe leak, providing richer information for operational and maintenance decisions. Attached Figure Description

[0033] Figure 1 This is a schematic diagram illustrating the application environment of a wind turbine water-cooled leakage signal detection system provided in one embodiment of this application. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments.

[0035] As shown in the figure, the wind turbine water cooling leakage signal detection system provided in this application includes a detection unit, a signal processing unit, a spectrum analysis unit, and a self-learning unit.

[0036] The detection unit is built based on a sound wave transmitter and a sound wave receiver. It can convert the detected liquid motion into an electrical signal containing Doppler frequency shift information, providing the most basic data foundation for the entire detection process.

[0037] Under the action of an external driving circuit, the acoustic transmitter generates high-frequency mechanical vibration, thereby continuously emitting a stable and constant-intensity continuous ultrasonic beam towards the preset monitoring area. The ultrasonic frequency is selected in the range of 30kHz to 220kHz to avoid interference from the human hearing range and the low-to-mid-frequency noise generated by most mechanical equipment inside the tower, ensuring the environmental specificity of the detection signal.

[0038] The emitted sound waves propagate through the air and interact with objects in the monitored space. Sound waves are reflected from any surface they encounter. In an ideal, leak-free environment, the reflectors within the monitored area are primarily stationary elements such as the walls of water-cooled pipes, supports, and tower foundations. According to the Doppler effect, since these reflectors are stationary relative to the detector, the frequency of their reflected echoes does not change significantly compared to the transmission frequency.

[0039] The acoustic receiver is responsible for capturing all these echo signals after complex reflections and frequency shifts. It precisely converts the received sound pressure changes into corresponding weak electrical signals. This synthesized electrical signal contains both a large number of strong, unchanging reflected signals from a stationary background and weak signals from a group of moving droplets with positive and negative frequency shifts.

[0040] The detection unit also includes a transducer and temperature compensation circuit. The transducer adopts a multi-layer stacked structure and is equipped with a metal backing mass to form a resonant cavity to enhance the vibration amplitude. An acoustic matching layer and a focusing lens are added to the front end to focus the divergent vibration energy into a narrow beam of ultrasonic pulses.

[0041] The temperature compensation circuit monitors changes in the working environment in real time and automatically adjusts the drive parameters to offset the resonant frequency shift caused by temperature drift.

[0042] The signal processing unit can condition and convert electrical signals, extracting weak signals from noise and boosting them to the volt level, then converting the boosted analog signals into digital signals for subsequent circuit processing. Details are as follows:

[0043] 1) Signal amplification and filtering

[0044] The front end of the receiving channel of the signal processing unit is equipped with a preamplifier composed of low-noise field-effect transistors (FETs) to boost the microvolt-level echo signal to the millivolt level. A differential input is implemented using an instrumentation amplifier architecture, allowing only frequency components related to the leakage signal to pass through, while strongly attenuating out-of-band interference and effectively suppressing common-mode interference.

[0045] 2) Signal conversion

[0046] The conditioned analog signal is fed into a high-precision Σ-Δ ADC for oversampling digitization. Undersampling technology is employed to optimize the sampling rate setting, reducing data throughput while meeting the Nyquist criterion. A built-in anti-aliasing low-pass filter strictly limits the signal bandwidth and eliminates high-frequency image components. A digital calibration algorithm is implemented during the conversion process to correct gain errors and phase shifts between channels. A FIFO buffer enables seamless data transfer, ensuring no data loss during continuous acquisition. This design not only fully preserves the signal's time-domain characteristics but also creates ideal conditions for subsequent digital down-conversion, FFT operations, and other processing.

[0047] The signal processing unit also embeds a hardware monitoring loop to continuously track key node parameters: it measures the actual output power through a detection circuit and compares it with the set value, triggering a protection interrupt when an anomaly occurs.

[0048] The spectrum analysis unit can decompose a digital signal into a superposition of different frequency components through Fourier transform (FFT), generate a corresponding spectrum, and calculate the total energy and frequency variance of the Doppler shift region based on the spectrum to quantify the leakage scale and dynamic characteristics.

[0049] The spectrum analysis unit includes a generation subunit, a feature parameter extraction subunit, and a threshold subunit.

[0050] The generator subunit can receive digital signals and decompose the time-domain waveform into a superposition of different frequency components using Fourier Transform (FFT). Fourier Transform (FFT) is a discretized Fourier series expansion tool that can decompose a continuous time-series signal into a series of orthogonal basis functions linear combinations to determine the complex amplitude X at the corresponding frequency point k. k By projecting the time-domain waveform onto different frequency axes, the complex amplitude distribution of each frequency component is obtained. The complex amplitude value X... k The calculation formula is as follows:

[0051]

[0052] Where x[n] represents the time-domain sample, N represents the data length, n represents time, k represents the frequency coefficient, and j represents the imaginary unit.

[0053] When an ultrasonic wave encounters a moving droplet, the frequency change Δf of the echo signal caused by the droplet's motion is calculated. This frequency shift is represented in the spectrum as a symmetrically broadened band centered on the transmission frequency. The high resolution of the FFT allows even minute velocity changes (e.g., on the order of 0.1 m / s) to produce observable frequency shifts, thus enabling sensitive detection of frequency changes in the echo signal of low-velocity leaking liquid. The calculation formula is as follows:

[0054]

[0055] Where v is the radial velocity of the droplet, θ is the angle between the direction of ultrasonic wave propagation and the direction of droplet motion, and λ is the wavelength.

[0056] The feature parameter extraction subunit can perform intelligent feature analysis on the extracted Doppler spectrum: calculating the total energy E of the frequency shift component. doppler and droplet velocity value Droplet velocity values ​​reflect the dispersion of droplet velocity and are a characteristic indicator of water-cooled leaks. By determining whether these characteristic values ​​simultaneously exceed preset thresholds, background noise and actual leaks can be ultimately distinguished. Total energy E doppler The calculation formula is as follows:

[0057]

[0058] Wherein, Ω represents the set of frequency ranges for the Doppler shift components.

[0059] When a leak occurs, the droplet swarm interacts with the ultrasonic waves, increasing the signal energy within the frequency shift range. A higher total energy indicates a stronger interaction between the droplet swarm and the ultrasonic waves, suggesting a potentially larger leak. Therefore, calculating the total energy allows for a preliminary assessment of the leak's intensity.

[0060] Droplet velocity value The calculation formula is as follows:

[0061]

[0062] Where μ is the average frequency, f i Here, M represents the frequency value of a single frequency point, and M represents the number of effective frequency points.

[0063] Frequency variance reflects the dispersion of droplet velocities. During a leak, the individual droplets in a swarm often exhibit different velocities due to various factors such as gravity and air resistance. A larger frequency variance indicates greater velocities and more unstable droplet motion, reflecting the dynamic characteristics of the leaking droplet swarm. By calculating the frequency variance, the system can gain a deeper understanding of the dynamic changes in the leak, providing richer information for accurate leak assessment.

[0064] in,

[0065] The threshold subunit can preset a dynamic threshold. During actual detection, the threshold is only applied when the total energy E measured is within the threshold value. doppler Exceeding the total energy threshold, and the frequency variance Exceeding the frequency variance threshold Only when the condition is met will the system determine it as a valid leak. This dual-criteria design effectively eliminates false triggers caused by a single interfering factor.

[0066] The self-learning unit is constructed based on the spectral characteristic curve under a leak-free state. It can record the spectral characteristic curve and differentially cancel steady-state noise interference in real time, determining whether the frequency-shifted spectral characteristic curve parameters exceed a preset threshold. This enables it to adapt to the environment and maintain high detection accuracy. Details are as follows:

[0067] Upon initial power-on or manual calibration, the self-learning unit enters training mode. During this time, it continuously collects environmental background noise data for a period of time, including signals from steady-state interference sources such as fan operation and mechanical vibration. After smoothing the raw data using a moving average filter, a standardized spectral characteristic curve is generated. This curve, with frequency on the horizontal axis and power density on the vertical axis, is stored in the EEPROM as a benchmark for subsequent comparisons.

[0068] During normal operation, the self-learning unit captures the monitoring spectral characteristic curve of the current frame in real time and performs point-by-point subtraction with the baseline spectrum in memory. Due to the high repeatability of fixed noise sources, they appear as similar waveform profiles in the two graphs; while sudden leakage signals will form significant energy bulges in the difference results.

[0069] For example, if the baseline spectrum has a value of A at a certain frequency point and the corresponding real-time spectrum has a value of B, then the difference value is Δ = BA. Only when Δ exceeds a preset sensitivity threshold is it considered a potential anomaly.

[0070] To address slow drift caused by equipment aging or gradual environmental changes, the self-learning unit uses an exponentially weighted moving average algorithm to periodically fine-tune baseline parameters. Each update assigns a weight of α (e.g., 0.95) to new data and a weight of 1-α to historical data, ensuring a smooth transition rather than abrupt replacement. This avoids false alarms caused by temperature changes and promptly captures long-term trend changes.

[0071] The self-learning unit also includes an early warning subunit. When the extracted frequency shift spectrum characteristic curve parameters exceed a preset threshold, the early warning subunit will determine that a leakage has occurred and trigger multiple alarm methods to notify relevant personnel. Alarm methods include audible and visual alarms, SMS alarms, and email alarms. Audible and visual alarms can attract the attention of on-site personnel; SMS and email alarms can send detailed alarm information to remote maintenance personnel's mobile phones or email addresses, enabling them to promptly understand the abnormal status of the equipment and take appropriate measures.

[0072] The aforementioned wind turbine water-cooled leakage signal detection system, based on the principles of acoustic wave transmitter, acoustic wave receiver, and spectral characteristic curve, can sense the frequency signal generated by the water-cooled liquid at the moment of leakage in real time. Even a small amount of droplets can cause changes in the frequency signal, achieving early warning and winning valuable response time for maintenance personnel.

[0073] Furthermore, analyzing the total energy and droplet velocity values ​​of the Doppler signal can provide a preliminary assessment of the severity of the leak. The total energy indicates the magnitude of the leak; a higher total energy suggests a more severe leak, providing richer information to support operational decisions (whether to immediately shut down the system).

[0074] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A water-cooled leakage signal detection system for wind turbine generators, characterized in that, It includes a detection unit, a signal processing unit, a spectrum analysis unit, and a self-learning unit; The detection unit is constructed based on a sound wave transmitter and a sound wave receiver, and can convert the detected liquid motion into an analog signal containing Doppler frequency shift information; The signal processing unit can condition and convert electrical signals to extract weak signals from noise and boost them to the volt level, and convert the boosted analog signals into digital signals. The spectrum analysis unit can decompose a digital signal into a superposition of different frequency components through Fourier transform, generate a corresponding spectrum, and calculate the total energy and frequency variance of the Doppler frequency shift region based on the spectrum to quantify the leakage scale and dynamic characteristics; the spectrum analysis unit includes a generation subunit, a feature parameter extraction subunit, and a threshold subunit. The self-learning unit is constructed based on the spectral characteristic curve under the leak-free state. It can determine whether the frequency shift spectral characteristic curve parameters exceed the preset threshold by recording the spectral characteristic curve and canceling steady-state noise interference in real time.

2. The wind turbine water-cooled leakage signal detection system according to claim 1, characterized in that, The detection unit also includes a transducer and a temperature compensation circuit. The transducer adopts a multi-layer stacked structure and is equipped with a metal backing mass block to form a resonant cavity to enhance the vibration amplitude. The temperature compensation circuit monitors changes in the working environment in real time and automatically adjusts the drive parameters to offset the resonant frequency shift caused by temperature drift.

3. The wind turbine water-cooled leakage signal detection system according to claim 1, characterized in that, The specific contents of the signal processing unit include signal amplification and filtering, and signal conversion; The signal amplification and filtering: The front end of the receiving channel of the signal processing unit is equipped with a preamplifier composed of low noise field effect, which amplifies the microvolt-level echo signal to the millivolt level, allowing only the frequency components related to the leakage signal to pass through.

4. The wind turbine water-cooled leakage signal detection system according to claim 1, characterized in that, The generating subunit can receive digital signals and decompose the continuous time series signal into a series of orthogonal basis functions linear combinations through Fourier transform to determine the complex amplitude value X at the corresponding frequency point k. k By projecting the time-domain waveform onto different frequency axes, the complex amplitude distribution of each frequency component is obtained, and the complex amplitude value X is obtained. k The calculation formula is as follows: Where x[n] is the time-domain sample, N is the data length, n is the time, k is the frequency coefficient, and j is the imaginary unit.

5. The wind turbine water-cooled leakage signal detection system according to claim 1, characterized in that, The generation subunit can also calculate the echo signal frequency change Δf caused by droplet motion, which is used to detect the echo signal frequency change of low-speed leaking liquid. The calculation formula is as follows: Where v is the radial velocity of the droplet, θ is the angle between the direction of ultrasonic wave propagation and the direction of droplet motion, and λ is the wavelength.

6. The wind turbine water-cooled leakage signal detection system according to claim 1, characterized in that, The feature parameter extraction subunit can perform intelligent feature analysis on the extracted Doppler spectrum: calculating the total energy E of the frequency shift component. doppler and droplet velocity value The total energy E doppler The calculation formula is as follows: Wherein, Ω represents the set of frequency ranges for the Doppler frequency shift components; The droplet velocity value The calculation formula is as follows: Where μ is the average frequency, f i Here, M represents the frequency value of a single frequency point, and M represents the number of effective frequency points.

7. The wind turbine water-cooled leakage signal detection system according to claim 1, characterized in that, The threshold subunit can determine the total energy E. doppler and frequency variance Does it exceed the preset dynamic threshold? 8. The wind turbine water-cooled leakage signal detection system according to claim 1, characterized in that, After entering the training mode, the self-learning unit will continuously collect environmental background noise data for a period of time, including signals generated by steady-state interference sources such as fan operation and mechanical vibration. After smoothing the original data through a moving average filter, a standardized spectral characteristic curve is generated.

9. A wind turbine water-cooled leakage signal detection system according to claim 8, characterized in that, The self-learning unit captures the monitoring spectrum feature curve of the current frame in real time and performs point-by-point subtraction with the baseline spectrum in memory to determine whether the frequency shift spectrum feature curve parameters are potentially abnormal.

Images