Fan fault prediction method and system based on optimized vibration monitoring data

Abstract

The invention belongs to the technical field of fan fault monitoring, and provides a fan fault prediction method and system based on optimized vibration monitoring data, and the method comprises the steps: calculating an effective value of data collected in a certain period of time to replace original data in the period of time, improving the data quality and processing efficiency, and maintaining the representativeness of the data; since a certain distance exists between an accelerometer used for collecting data and a vibration part, the collected data can contain certain noise influence, and it is found through statistics that certain abnormal values exist in the collected data, the maximum correlation kurtosis deconvolution method is used for carrying out noise reduction processing, and the noise reduction accuracy is improved. Pulse components representing faults in the useful signals are highlighted, whether abnormal values exist in a group of data is judged by calculating the abnormal rate of the group of data, and the abnormal values are eliminated; through noise reduction and abnormal value removal processing, the quality of original data is improved, and the data processing precision and the fault prediction precision are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of fan fault monitoring, and particularly relates to a fan fault prediction method and system based on optimized vibration monitoring data. Background Technique

[0002] Among the existing fan fault monitoring methods, vibration monitoring has the advantages of accurate fault location, simple testing, high testing efficiency, good real-time performance, etc., and is one of the effective technologies for early fault detection and prediction of fans at present.

[0003] Currently, during the fan fault monitoring process, when collecting raw data, the vibration accelerometer has a relatively fast sampling frequency and a high acquisition frequency, resulting in messy acquired data. Since there is a certain distance between the accelerometer used to collect data and the vibration part where the fan fails, the data collected by it is affected by certain noise, and there are certain outliers in the collected data; the messy data, noise influence and outliers affect the final prediction accuracy. Summary of the Invention

[0004] In order to solve the above problems, the present invention proposes a fan fault prediction method and system based on optimized vibration monitoring data. The present invention calculates the effective value of the data collected within a certain period of time to replace the raw data within this period of time, improves the data quality and processing efficiency, and does not lose the representativeness of the data; since there is still a certain distance between the accelerometer used to collect data and the vibration part, the data collected by it will contain certain noise influence, and through statistics, it is found that there are certain outliers in the collected data. In the present invention, the maximum correlation kurtosis deconvolution method is used for noise reduction processing to highlight the pulse components representing faults in the useful signals, and the abnormality rate of a set of data is calculated to determine whether there are outliers in this set of data and eliminate them; through noise reduction and outlier removal processing, the quality of the raw data is improved, and the data processing accuracy and fault prediction accuracy are improved.

[0005] In order to achieve the above object, the present invention is implemented by the following technical solutions:

[0006] In the first aspect, the present invention provides a fan fault prediction method based on optimized vibration monitoring data, including:

[0007] Obtain the vibration signal of the fan;

[0008] Take the vibration signals collected within each first preset time period as a group, and determine the effective value of each group of vibration signals;

[0009] Within the second preset time period, determine the abnormality rate by the ratio of the number of abnormal points to the number of monitoring points; delete the vibration signals corresponding to when the abnormality rate is higher than the preset value;

[0010] Denoise the vibration signal using maximum correlation kurtosis deconvolution;

[0011] Convert the vibration signal from the time domain signal to the frequency domain signal, and perform fan fault prediction according to the frequency domain.

[0012] Furthermore, the effective value X rms is:

[0013]

[0014] where x i (t) is the data collected by the accelerometer within a preset time period, i is an integer; N is the number of data collected within the preset time period.

[0015] Furthermore, taking the correlation kurtosis as the measurement scale, by reducing the proportion of noise in the useful signal, highlighting the pulse components representing faults in the useful signal, and using a filter to maximize the correlation kurtosis of the impact component.

[0016] Furthermore, the correlation kurtosis CK(T) is:

[0017]

[0018] where T is the signal period; M is the displacement order; N is the number of data collected within the preset time period; w is the impact component; n and m are integers.

[0019] Furthermore, use a window function to window and intercept the vibration signal to limit the time range of the transformation, so that the vibration signal falls within a frequency interval range, then perform Fourier transform on the windowed vibration signal, and finally superimpose the signals after windowed transformation.

[0020] Furthermore, use a hanning window to window the vibration signal in segments:

[0021]

[0022] where ω(n) is the impact component; N is the number of data collected within the preset time period; t is time; R N (n) is the value of the original signal in the time domain, representing the amplitude of the signal at the nth time point; W(ω) is the Fourier transform of the window function, ω represents the frequency variable of the signal; W R (ω) is the spectrum of the original signal without window function processing.

[0023] In a second aspect, the present invention also provides a fan fault prediction system based on optimized vibration monitoring data, including:

[0024] A data acquisition module, configured to: obtain the vibration signal of the fan;

[0025] The effective value determination module is configured to: take the vibration signals collected within each first preset time period as a group, and determine the effective value of each group of vibration signals;

[0026] The outlier removal module is configured to: within the second preset time period, determine the outlier rate through the ratio of the number of outlier points that appear to the number of monitoring points; delete the vibration signals corresponding to when the outlier rate is higher than the preset value;

[0027] The denoising module is configured to: perform denoising on the vibration signals by using maximum correlation kurtosis deconvolution;

[0028] The fault prediction module is configured to: convert the vibration signals from time domain signals to frequency domain signals, and perform fan fault prediction according to the frequency domain.

[0029] In a third aspect, the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the fan fault prediction method based on optimized vibration monitoring data described in the first aspect are implemented.

[0030] In a fourth aspect, the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and capable of running on the processor. When the processor executes the program, the steps of the fan fault prediction method based on optimized vibration monitoring data described in the first aspect are implemented.

[0031] In a fifth aspect, the present invention further provides a computer program product, where the computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the fan fault prediction method based on optimized vibration monitoring data described in the first aspect are implemented.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0033] 1. In the present invention, the effective value of the data collected within a certain time is calculated to replace the original data within this time period, solving the problem that the original data is messy due to the high acquisition frequency. On the basis of ensuring the representativeness of the data, the data quality and processing efficiency are improved; at the same time, due to the certain distance between the accelerometer for collecting data and the vibration part, the collected data will contain certain noise interference, and there will also be certain outliers in the data. In the present invention, the maximum correlation kurtosis deconvolution method is used for noise reduction processing to highlight the pulse components representing faults in the useful signals, and the outlier rate of a group of data is calculated to determine whether there are outliers in this group of data and eliminate them. The method is simple and the processing speed is fast; through noise reduction and outlier removal processing, the quality of the original data is ensured, and the data processing accuracy and fault prediction accuracy are improved.

[0034] 2. In the present invention, Fourier transform with windowing is used to effectively reduce the spectral leakage phenomenon during the vibration signal processing, improving the accuracy and precision of vibration signal processing. Specifically, in practical applications, there is a problem of spectral leakage in Fourier transform. Spectral leakage refers to the situation where when the vibration signal period is not an integer multiple, the Fourier transform distributes the spectrum of the vibration signal to other frequencies, resulting in spectral distortion. Since digital signals exist in a discrete form, it is difficult to meet the condition that the signal period is an integer multiple, and spectral leakage is likely to occur. In the present invention, the problem of spectral leakage is effectively solved through the windowing function. By weighting the signal in the time domain through the windowing function, the edge part of the signal can be attenuated, thereby reducing the spectral leakage phenomenon and ensuring the accuracy of fan fault prediction.

[0035] 3. In the present invention, after performing Fourier transform on each section of the windowed data, the spectral information of each section of data is obtained. The spectral information of each section is superimposed to obtain the time-frequency diagram of the monitoring data, and whether a fault occurs in each part of the fan is judged according to the frequency characteristics of the data, ensuring the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings forming a part of this embodiment are used to provide a further understanding of this embodiment. The schematic embodiments and descriptions thereof of this embodiment are used to explain this embodiment and do not constitute an improper limitation of this embodiment.

[0037] Figure 1 It is the data analysis flow chart of Embodiment 1 of the present invention;

[0038] Figure 2 It is the data analysis structure block diagram of Embodiment 1 of the present invention;

[0039] Figure 3 It is the schematic diagram of outlier identification of Embodiment 1 of the present invention;

[0040] Figure 4 It is the schematic diagram of the time domain characteristics of the Hanning window and other windows of Embodiment 1 of the present invention;

[0041] Figure 5 It is the schematic diagram of the frequency domain characteristics of the Hanning window and other windows of Embodiment 1 of the present invention;

[0042] Figure 6 It is the time domain diagram of the data of Embodiment 1 of the present invention;

[0043] Figure 7 It is the frequency spectrum diagram of the data of Embodiment 1 of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0044] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0045] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present application belongs.

[0046] Embodiment 1:

[0047] With the development of society, most big cities with high electricity consumption loads are in coastal areas. Wind power generation does not require long-distance transportation and is suitable for cities with large electricity consumption. The research and development of offshore wind power are not only an important part of the expansion of the wind power industry scale but also an effective way to promote the development of the marine energy industry, with good development prospects and important strategic significance.

[0048] Wind farms are characterized by large installed capacities and a large number of wind turbines. While offshore wind power is developing rapidly, it still faces many problems. At present, the technical skills of maintenance personnel in wind farms are not yet mature. There are a large number of wind turbines, which are distributed in areas with relatively harsh geographical environments and are often affected by bad weather. As a result, the wind turbines are often in a state of high-intensity variable operation, and all components are extremely prone to damage. The downtime after a failure is relatively long, which greatly reduces the availability of wind turbine units. Reducing maintenance costs and improving operation efficiency have become one of the most important issues in the maintenance strategy of wind turbine units.

[0049] Among the existing methods for monitoring wind turbine faults, vibration monitoring has the advantages of accurate fault location, simple testing, high testing efficiency, and good real-time performance. It is one of the effective technologies for early fault detection and prediction of wind turbines at present. The main function of frequency-domain analysis is to use mathematical processing transformation to convert the description of time-domain signals into a function represented by a frequency coordinate axis, so as to solve the characteristic frequencies of each component of the wind turbine and, based on the characteristic frequencies, distinguish the vibrations of each component of the wind turbine, and then carry out location analysis and diagnosis of the faults of the wind turbine. Frequency-domain analysis mainly focuses on spectrum analysis and uses Fourier transform to decompose time-domain signals to obtain information such as their frequency structure, amplitude, and phase. Frequency-domain analysis can diagnose and monitor faults such as misalignment faults, imbalance faults, and mechanical looseness in wind turbines.

[0050] As recorded in the background art, during the traditional wind turbine fault monitoring process, when collecting original data, the vibration accelerometer has a relatively fast sampling frequency and a relatively high acquisition frequency, resulting in messy acquired data. Since there is a certain distance between the accelerometer used to collect data and the vibrating part where the wind turbine fails, the data collected by it is affected by certain noise, and there are certain outliers in the collected data; the messy data, noise influence, and outliers affect the final prediction accuracy.

[0051] To solve at least one of the above problems, this embodiment provides a method for predicting wind turbine faults based on optimized vibration monitoring data, which involves the maximum correlation kurtosis deconvolution noise reduction method and the fast discrete Fourier transform (FFT) algorithm. Specifically, after removing outliers and noise from the original data collected by the accelerometer, the windowed Fourier transform (WFT) algorithm is used to convert the vibration monitoring data of the offshore wind turbine into a frequency-domain signal, and then spectrum analysis is performed to predict wind turbine faults. Specifically:

[0052] S1. Since the vibration accelerometer has a fast sampling frequency and a large amount of data is obtained in a short time, the data is messy and the analysis efficiency is not high. Therefore, in this embodiment, the vibration signals collected within each preset time period are used as a group for effective value calculation; optionally, the data obtained every 60 s is divided into a group to calculate its effective value, and the calculated effective value is used as the original data for analysis. The formula for calculating the effective value is:

[0053]

[0054] where x i (t) is the data collected by the accelerometer within the preset time period, i is an integer; N is the number of data collected within the preset time period; X rms is the effective value of the data collected within the preset time period.

[0055] S2. Identification and elimination of outliers. To exclude abnormal monitoring data caused by accidental factors, this embodiment proposes a method for excluding outliers by using the outlier rate of the data to further reduce the false alarm phenomenon of the monitoring system. Outlier rate monitoring refers to the probability of being detected as a fault within a period of time. The outlier rate is equal to the number of outliers within a specific time and the number of monitoring points within that time length. The operation formula for the outlier rate η is:

[0056] η = N a / N (2)

[0057] where N is the number of monitoring points within a specific time length; N a is the number of outliers within that time period. Optionally, the specific time length is taken as 5 minutes, and the outlier rate within every five minutes is calculated. An outlier rate above 0.2 indicates that there are outliers within these 5 minutes and they need to be removed.

[0058] S3. For the acquired raw data, the feature signals need to be denoised first. In this embodiment, Maximum Correlated Kurtosis Deconvolution (MCKD) is used for effective denoising. MCKD takes the correlated kurtosis as the measurement scale and highlights the pulse components representing faults in the useful signals by reducing the proportion of noise in the useful signals. Let w(n) be the periodic impulse component of the input signal y(n). Without considering the background noise, MCKD selects a suitable filter ψ(k) to maximize the correlated kurtosis of the impulse component w(n), and at the same time achieves the effect of denoising:

[0059]

[0060] where ψ = [ψ1, ψ2, ψ3....ψ L T ; L is the filter length.

[0061] To make the periodic pulses more prominent, the correlated kurtosis (CK) of the signal needs to be maximized by MCKD so that the result of signal processing will be optimal. The expression of the correlated kurtosis is:

[0062]

[0063] where T is the signal period; M is the displacement order.

[0064] To highlight the impact of the periodic components, the objective function of the MCKD algorithm is:

[0065]

[0066] S4. When the sampling length is fixed, the higher the sampling frequency, the denser the sampling points, the larger the amount of collected data, and the closer the obtained digital signal is to the original signal. To ensure that the signal does not lose or distort the original signal, this embodiment proposes a sampling theorem, stipulating that the minimum sampling frequency for a band-limited signal not to lose information is Fs ≥ 2Fm. Therefore, the following issues should be noted when determining the sampling frequency: First, correctly judge the frequency of the highest frequency component in the original signal; second, the same amount of data can improve the fundamental frequency resolution by changing the number of sampling points per cycle.

[0067] In the field of digital signal processing, the Fourier Transform (FT) is one of the commonly used signal processing methods at present and plays an important role in the field of time-frequency analysis. The continuous signal f(t) can be expanded in the complete orthogonal signal space as:

[0068]

[0069] ​Equation (6) is the definition formula of the Fourier transform, which is used to transform a signal from the time domain to the frequency domain.

[0070] The sufficient condition for its establishment is that f(t) is absolutely integrable in the infinite interval, that is:

[0071]

[0072] The inverse transform of the continuous Fourier transform is:

[0073]

[0074] The vibration signals of each part of the wind turbine are not stable. Since the waveform characteristics of non-stationary signals are irregular and there is no concept of frequency, in this embodiment, a window function is used to window and intercept the signal to limit the time range of the transform, so that the sampled signal is within a frequency interval range, and then the Fourier transform is performed on the windowed signal respectively. Finally, the windowed and transformed signals are superimposed, that is, the short-time Fourier transform (STFT) is performed on the signal. STFT is defined as:

[0075]

[0076] Among them, f(t) is the time-domain signal; g(t - τ) is the window function; τ is the center of the window function. When the window function g(t) = 1, STFT is simplified to the traditional Fourier transform. When the given signal is a discrete signal such as wind turbine vibration monitoring data, the signal is sampled, and the sampling time is Δt, then Equation (9) can be expressed as:

[0077]

[0078] The calculation process can be divided into the following two steps:

[0079] S4.1. Window the signal to be processed in segments.

[0080] S4.2. Perform STFT on each signal segment.

[0081] Window-Fourier Transform (WFT). The so-called windowing means multiplying the original signal by a window function. The truncation operation on the signal can be understood as adding a rectangular window to the signal. This windowing operation is essentially a weighting operation on the original signal. The weight value of each point of the unit rectangular window is 1, and these weight functions are collectively called window functions. After windowing, both the beginning and the end of the signal are 0, which weakens the influence of phase mutation to a certain extent, and the phenomenon of spectral leakage is significantly reduced. Using different window functions will bring different effects on the spectrum of the signal. Analyze three commonly used window functions: Hanning window, Hamming window, and Blackman window. The main lobe widths of the Hanning window and the Hamming window are quite similar, both smaller than that of the Blackman window. The wider the main lobe width, the more concentrated the energy is in the main lobe position, but the frequency resolution will deteriorate accordingly. The side lobe attenuation rates of the Hanning window and the Blackman window are relatively fast. In comparison, the side lobe attenuation rate of the Hamming window is slower, but the attenuation of the side lobe closest to the main lobe of the Hamming window is larger.

[0082] According to the above characteristics, for broadband signals, the Hanning window or the Blackman window with faster side lobe attenuation can be used, and the specific selection can be determined according to the frequency resolution requirements. Narrowband signals can be processed using the Hamming window, which has a high resolution for the main lobe and a good suppression effect on the adjacent side lobes. The Hanning window is a general choice. If only the harmonic components of the signal are qualitatively analyzed and the amplitude accuracy requirement is not high, the Hanning window is a good choice.

[0083] For actual signals, how to select the window function? Selecting a window function with large attenuation of the first side lobe and fast attenuation of the side lobe peak is beneficial to alleviating the spectral leakage problem generated during the truncation process. For the processing of random signals such as the vibration monitoring data of offshore wind turbines, since the Hanning window can significantly reduce the height of the side lobe without much widening the main lobe, thus effectively reducing the power leakage, the Hanning window is selected in this embodiment:

[0084]

[0085] where t is time, related to the signal sampling frequency; R N (n) is the value of the original signal in the time domain, representing the amplitude of the signal at the nth time point; W(ω) is the Fourier transform of the window function, and ω represents the frequency variable of the signal; W R (ω) is the spectrum of the original signal without being processed by the window function.

[0086] Perform DFT operation on the data. Since the time length is taken as a finite value, that is, the signal is truncated, the bandwidth of the signal is expanded. This phenomenon is called leakage. Windowing is used to reduce spectral leakage. Selecting a window function with large attenuation of the first sidelobe and fast attenuation of the sidelobe peak is beneficial to alleviating the spectral leakage problem generated during the stage process. To correct the amplitude attenuation caused by windowing, the spectrum after windowed FFT can be corrected by multiplying by the recovery coefficient of the window function. The amplitude equal recovery coefficient of the hanning window is 2, and the power equal recovery coefficient is 1.633.

[0087] S5. Obtain the spectral information of each segment of data after windowing through Fourier transform, superimpose the spectral information of each segment to obtain the time-frequency diagram of the monitoring data, and judge whether there is a fault in each part of the wind turbine according to the frequency characteristics of the data. Judging whether there is a fault in each part of the wind turbine according to the frequency characteristics of the data can be achieved by comparing the frequency of the signal in different time periods with the preset frequency, or by means of a trained fault prediction network model, etc.

[0088] Embodiment 2:

[0089] Offshore wind power has developed rapidly, but there are also many problems. At present, the technology of wind farm maintenance personnel is not yet mature. The wind turbines are distributed offshore and are often affected by bad weather, resulting in the wind turbines often being in a high-intensity operation state. Each component is extremely prone to damage, and the downtime due to faults is relatively long, greatly reducing the availability of the wind turbine units. Since all offshore wind turbines are distributed offshore, it is not conducive to people to inspect them, and faults cannot be detected in time. Therefore, when a fault occurs in the wind turbine, it cannot be detected and repaired in time.

[0090] In order to better monitor the operating state of offshore wind turbines and detect fault problems in time, sensors are installed in each part of the wind turbine to obtain the operating data of the wind turbine in real time. Since the data is complex and the obtained data is not easy to highlight whether a fault has occurred and the fault location, therefore, on the basis of Embodiment 1, this embodiment provides a wind turbine fault prediction method based on optimizing vibration monitoring data, including:

[0091] S1. Obtain the vibration monitoring data of the offshore wind turbine, which is obtained and recorded in the acquisition module in real time by the vibration monitoring accelerometers at each part of the wind turbine, and then sent to the user through the transmission submarine cable. Since the sampling frequency of the accelerometer is relatively high and a large amount of data is collected in a short time, we calculate the effective value of the data collected every 60s as the original data for analysis. Obtain the signal data of the ATY-X1 accelerometer on the No. A18 wind turbine in an offshore wind farm, and calculate the effective value. The effective value calculation formula:

[0092]

[0093] where, x i(t) is the data collected by the accelerometer during this time period; N is the number of data collected during this time period; X rms is the effective value of the data collected during this time period. Each obtained effective value is used as the original data.

[0094] S2. Perform filtering and denoising processing on the original data, identify and remove outliers. As Figure 3 shown, calculate the outlier rate within every five minutes. The outlier rate is equal to the ratio of the number of outliers within a specific time to the number of monitoring points within that time length. The operation formula is:

[0095] η = N a / N

[0096] where, N is the number of monitoring points within a specific time length, N a is the number of outliers within this time period, and the abnormal data is removed. The maximum correlated kurtosis deconvolution is one of the techniques for effective signal denoising. By reducing the proportion of noise in the useful signal, the pulse components representing faults in the useful signal are highlighted. Let w(n) be the periodic impulse component of the input signal y(n). Without considering the background of noise, MCKD selects a suitable filter ψ(k) to maximize the correlated kurtosis of the impulse component w(n) and at the same time achieve the effect of noise reduction:

[0097]

[0098] where, ψ = [ψ1, ψ2, ψ3....ψ L T ; L is the filter length.

[0099] In order to make the periodic pulse more prominent, the correlated kurtosis (CK) of the signal needs to be maximized by MCKD, so that the result of signal processing will reach the optimal. The expression of the correlated kurtosis is:

[0100]

[0101] In the formula: T is the signal period; M is the displacement order.

[0102] In order to highlight the impact of the periodic component, the objective function of the MCKD algorithm is:

[0103]

[0104] Higher-quality signal data is obtained through the above calculations.

[0105] S3. Select a window function (Hanning window) with a non-zero time. As Figure 4 、 5 ​As shown, there are Hanning window, Hamming window and Blackman window. The main lobe widths of the Hanning window and the Hamming window are comparable and both are smaller than that of the Blackman window. The wider the main lobe width is, the more concentrated the energy is at the main lobe position, but the frequency resolution will deteriorate accordingly. The sidelobe attenuation rates of the Hanning window and the Blackman window are relatively fast. In comparison, the sidelobe attenuation rate of the Hamming window is slower, but the sidelobe closest to the main lobe of the Hamming window has a larger attenuation. For the processing of random signals, since the Hanning window can significantly suppress the height of the sidelobes without much widening of the main lobe, thus effectively reducing the power leakage, the Hanning window is selected in this design:

[0106]

[0107] Equation (11) is the defining equation of the Hanning window, and W(ω) is the Fourier transform of the window function. The preprocessed data is windowed using the Hanning window to reduce spectral leakage.

[0108] S4. Intercept the signal with the window function for FFT transformation to obtain spectral information, and reduce spectral leakage by windowing. The defining equation of the Fourier transform is:

[0109]

[0110] It is used to transform the signal from the time domain to the frequency domain. The sufficient condition for it to hold is that f(t) is absolutely integrable in the infinite interval, that is:

[0111]

[0112] The Fourier transforms are respectively performed on the windowed signals, and finally the windowed and transformed signals are superimposed, that is, the short-time Fourier transform (STFT) of the signal is performed. When the given signal is a discrete signal, the continuous signal is sampled with the sampling time of Δt, then the defining equation of the STFT can be expressed as:

[0113]

[0114] The results of the Fourier transforms of each segment of data are obtained.

[0115] S5. Slide the window function along the time axis, and repeat step S4 to continue the fast Fourier transform to obtain the spectral information of each segment.

[0116] S6. Superimpose the obtained spectral information of each segment to obtain a time-frequency diagram, which presents the frequencies of the signals in different time periods. The prediction of fan faults can be carried out according to the comparison between the frequencies of the signals in different time periods and a preset frequency; or by means of a trained fault prediction network model, etc.

[0117] Embodiment 3:

[0118] This embodiment provides a fan fault prediction system based on optimized vibration monitoring data, including:

[0119] A data acquisition module, configured to: acquire the vibration signals of the fan;

[0120] An effective value determination module, configured to: take the vibration signals collected within each first preset time period as a group and determine the effective value of each group of vibration signals;

[0121] An outlier removal module, configured to: within a second preset time period, determine the outlier rate through the ratio of the number of outliers to the number of monitoring points; delete the vibration signals corresponding to when the outlier rate is higher than a preset value;

[0122] A denoising module, configured to: denoise the vibration signals by using maximum correlation kurtosis deconvolution;

[0123] A fault prediction module, configured to: convert the vibration signals from time-domain signals to frequency-domain signals and perform fan fault prediction according to the frequency domain.

[0124] The working method of the system is the same as that of the fan fault prediction method based on optimized vibration monitoring data in Embodiment 1, and will not be elaborated here.

[0125] Embodiment 4:

[0126] This embodiment provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the fan fault prediction method based on optimized vibration monitoring data described in Embodiment 1 are implemented.

[0127] Embodiment 5:

[0128] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored on the memory and capable of running on the processor. When the processor executes the program, the steps of the fan fault prediction method based on optimized vibration monitoring data described in Embodiment 1 are implemented.

[0129] Embodiment 6:

[0130] This embodiment provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps of the fan fault prediction method based on optimized vibration monitoring data described in Embodiment 1 are implemented.

[0131] The foregoing is only the preferred embodiment of this embodiment and is not intended to limit this embodiment. For those skilled in the art, this embodiment can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this embodiment shall be included within the protection scope of this embodiment.

Claims

1. A method for predicting fan failure based on optimized vibration monitoring data, characterized in that: include: Obtain the vibration signal of the fan; Taking the vibration signals collected in each first preset time period as a group, determining the effective value of each group of vibration signals; In the second preset time period, the abnormality rate is determined by the ratio of the number of abnormal points to the number of monitoring points; the vibration signal corresponding to the abnormality rate being higher than the preset value is deleted; The maximum correlation kurtosis deconvolution is used to denoise the vibration signal; The vibration signal is converted from time domain signal to frequency domain signal, and the fan fault prediction is performed based on the frequency domain.

2. A method for predicting fan failure based on optimized vibration monitoring data according to claim 1, characterized in that: Effective value X rms for: Among them, x i (t) is the data collected by the accelerometer within the preset time period, i is an integer; N is the number of data collected within the preset time period.

3. A method for predicting fan failure based on optimized vibration monitoring data according to claim 1, characterized in that: The correlation kurtosis is used as a measurement scale. By reducing the proportion of noise in the useful signal, the pulse component representing the fault in the useful signal is highlighted, and the filter is used to maximize the correlation kurtosis of the impulse component.

4. A method for predicting fan failure based on optimized vibration monitoring data as claimed in claim 3, characterized in that: The relevant kurtosis CK(T) is: Wherein, T is the signal period; M is the displacement order; N is the number of data collected within a preset time period; w is the impact component; n and m are integers.

5. A method for predicting fan failure based on optimized vibration monitoring data according to claim 1, characterized in that: The vibration signal is windowed using a window function to intercept and limit the time range of the transformation, so that the vibration signal falls within a frequency range, and then the windowed vibration signal is Fourier transformed, and finally the windowed transformed signal is superimposed.

6. A method for predicting fan failure based on optimized vibration monitoring data according to claim 4, characterized in that: Use the Hanning window to perform segmented windowing on the vibration signal: Where ω(n) is the impact component; N is the number of data collected within the preset time period; t is time; R N (n) is the value of the original signal in the time domain, which represents the amplitude of the signal at the nth time point; W(ω) is the Fourier transform of the window function, ω represents the frequency variable of the signal; W R (ω) is the spectrum of the original signal without window processing.

7. A fan fault prediction system based on optimized vibration monitoring data, characterized in that: include: The data acquisition module is configured to: obtain a vibration signal of the fan; The effective value determination module is configured to: take the vibration signals collected in each first preset time period as a group, and determine the effective value of each group of vibration signals; The abnormal value removal module is configured to: determine the abnormality rate by the ratio of the number of abnormal points to the number of monitoring points in the second preset time period; delete the vibration signal corresponding to the abnormality rate higher than the preset value; The denoising module is configured to: denoise the vibration signal using maximum correlation kurtosis deconvolution; The fault prediction module is configured to: convert the vibration signal from a time domain signal to a frequency domain signal, and perform fan fault prediction based on the frequency domain.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method for predicting fan faults based on optimizing vibration monitoring data as described in any one of claims 1 to 6 are implemented.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: When the processor executes the program, the steps of the fan fault prediction method based on optimized vibration monitoring data as described in any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the wind turbine fault prediction method based on optimized vibration monitoring data as described in any one of claims 1 to 6 are implemented.

Images