Tunnel fan composite fault adaptive identification method and device and storage medium

Abstract

This invention relates to the field of tunnel ventilation fan fault identification technology, specifically disclosing an adaptive identification method, device, and storage medium for composite faults in tunnel ventilation fans. The method includes: acquiring multi-dimensional sensor data of the tunnel ventilation fan; performing environmental noise processing and speed interference processing on the multi-dimensional sensor data to obtain processed data; determining the general characteristics and operating condition characteristics of the tunnel ventilation fan based on the processed data; performing feature decoupling operations based on the general characteristics and operating condition characteristics to obtain feature vectors; and performing an adaptive composite fault discrimination operation based on the feature vectors to generate a composite fault identification result. By improving traditional ventilation fan fault identification methods, and adopting targeted identification schemes based on the actual operating conditions, environmental conditions, and fault characteristics of the tunnel ventilation fan, the accuracy and reliability of fault identification are improved.

Description

Technical Field

[0001] This invention relates to the field of tunnel ventilation fan fault identification technology, specifically to an adaptive identification method, device, and storage medium for complex faults in tunnel ventilation fans. Background Technology

[0002] As the core equipment of a tunnel ventilation system, tunnel ventilation fans play a crucial role in replacing exhaust gases, maintaining air quality, and ensuring visibility. Due to the enclosed nature of tunnels, limited pollutant dispersion, and the need for continuous long-term operation of the fans, their operational status directly affects tunnel traffic safety and operational efficiency. Failures in tunnel ventilation fans can not only reduce ventilation capacity and create traffic safety hazards, but also lead to serious consequences such as tunnel closure and economic losses.

[0003] In the existing technology, the vibration signal analysis and fault diagnosis methods for industrial motors are relatively mature. The existing technology usually uses fixed bandpass filtering or simple mean filtering for signal preprocessing, without fully considering the influence of complex factors such as strong airflow interference, power frequency noise, and dust erosion in the tunnel environment on the acquired signal. The preprocessing algorithm lacks adaptive noise reduction capability for tunnel interference characteristics, making it difficult to effectively extract weak fault features, and the signal-to-noise ratio cannot meet the requirements of high-precision diagnosis.

[0004] Meanwhile, existing fault diagnosis algorithms are mostly based on fixed threshold discrimination strategies, which fail to consider the dynamic changes in operating conditions caused by frequent start-stop and variable load operation of tunnel fans according to ventilation needs. When the speed and load fluctuate within a large range, the fault characteristics exhibit significant dynamic features, and the fixed threshold method is prone to misjudgment or missed judgment during the switching of operating conditions.

[0005] In addition, the fault types of tunnel ventilation fans differ from those of ordinary industrial motors. Fault types such as blade imbalance and blade cracks account for a higher proportion of the fault spectrum of tunnel ventilation fans. Furthermore, the characteristics of blade faults and bearing faults exhibit certain aliasing in the frequency domain. Existing algorithms lack an effective feature decoupling mechanism, making it difficult to accurately distinguish between composite faults. Summary of the Invention

[0006] To overcome the aforementioned technical problems in the prior art, embodiments of the present invention provide an adaptive identification method, device, and storage medium for composite faults of tunnel ventilation fans. By improving the traditional ventilation fan fault identification method, a targeted identification scheme is adopted based on the actual operating conditions, environmental conditions, and fault characteristics of the tunnel ventilation fan, thereby improving the accuracy and reliability of fault identification.

[0007] To achieve the above objectives, embodiments of the present invention provide an adaptive identification method for composite faults in tunnel ventilation fans. The method includes: acquiring multi-dimensional sensor data of the tunnel ventilation fan; performing environmental noise processing and speed interference processing on the multi-dimensional sensor data to obtain processed data; determining the general characteristics and operating condition characteristics of the tunnel ventilation fan based on the processed data; performing feature decoupling operations based on the general characteristics and the operating condition characteristics to obtain feature vectors; and performing an adaptive composite fault discrimination operation based on the feature vectors to generate a composite fault identification result.

[0008] Preferably, the step of performing environmental noise processing and rotational speed interference processing on the multi-dimensional sensing data to obtain processed data includes: determining a preset energy threshold; performing adaptive analysis of the number of decomposition levels of wavelet decomposition on the multi-dimensional sensing data based on the preset energy threshold to determine the adaptive decomposition level; performing wavelet packet decomposition on the multi-dimensional sensing data based on the Daubechies-db6 wavelet function and the adaptive decomposition level to obtain decomposed data; performing adaptive bandpass filtering on the decomposed data to obtain filtered data; performing resonance demodulation processing on the filtered data based on Hilbert transform to obtain the bearing outer ring fault characteristic frequency, bearing inner ring fault characteristic frequency, and rolling element fault characteristic frequency; and generating processed data based on the bearing outer ring fault characteristic frequency, bearing inner ring fault characteristic frequency, and rolling element fault characteristic frequency.

[0009] Preferably, the step of performing adaptive bandpass filtering on the decomposed data to obtain filtered data includes: acquiring the rotational speed frequency of the tunnel fan; determining the center frequency of the tunnel fan based on the rotational speed frequency; determining the noise power of the tunnel fan; dynamically determining the quality factor based on the signal power of the tunnel fan and the noise power; and performing adaptive bandpass filtering on the decomposed data based on the center frequency and the quality factor to obtain filtered data.

[0010] Preferably, determining the operating characteristics of the tunnel ventilation fan based on the processed data includes: determining a fault characteristic frequency band; determining a Zoom-FFT analysis center frequency based on the fault characteristic frequency band; performing complex modulation processing on the processed data based on the Zoom-FFT analysis center frequency to obtain a complex-modulated signal; performing low-pass filtering processing on the complex-modulated signal to obtain a low-pass filtered signal; performing FFT transform on the low-pass filtered signal to obtain refined spectral features; determining the speed change rate and start / stop status based on the real-time speed of the tunnel ventilation fan; generating operating characteristics based on the speed change rate and the start / stop status; and generating ventilation fan operating characteristics based on the refined spectral features and the operating characteristics.

[0011] Preferably, the step of performing feature decoupling operation based on the general features and the wind turbine operating condition features to obtain feature vectors includes: generating fused features based on the general features and the wind turbine operating condition features; determining the covariance matrix of the fused features; determining the variance contribution rate of each principal component based on the covariance matrix; generating a decoupled feature matrix based on the variance contribution rate and a preset contribution rate threshold; determining the inter-class variance and intra-class variance based on the decoupled feature matrix; and filtering the decoupled feature matrix based on the Fisher criterion, the inter-class variance, and the intra-class variance to obtain feature vectors.

[0012] Preferably, the step of performing adaptive composite fault discrimination based on the feature vector to generate a composite fault identification result includes: processing the feature vector based on a fuzzy C-means clustering algorithm to obtain multiple operating condition categories; obtaining a preset fault feature library; matching the preset fault feature library based on the feature vector and the multiple operating condition categories to obtain a matching result; performing fault mechanism verification based on the matching result to generate a mechanism verification result; determining the prior probability corresponding to each fault type based on historical fault data; generating a posterior probability for tunnel fan faults based on the matching result, the mechanism verification result, and the prior probability; and generating a composite fault identification result based on the posterior probability.

[0013] Preferably, the step of performing fault mechanism verification based on the matching result and generating a mechanism verification result includes: when the matching result indicates the existence of a fault, obtaining the fault type; when the fault type is a bearing fault, performing a bearing mechanism verification operation on the feature vector based on a preset bearing mechanism verification rule to generate a bearing fault verification result; when the fault type is a fan blade fault, performing a fan blade mechanism verification operation on the feature vector based on a preset fan blade mechanism verification rule to generate a fan blade fault verification result; when the fault type is an abnormal vibration, performing a vibration mechanism verification operation on the feature vector based on a preset temperature-vibration verification rule to generate a vibration fault verification result; and generating a mechanism verification result based on the bearing fault verification result, the fan blade fault verification result, and the vibration fault verification result.

[0014] Preferably, the step of performing fault mechanism verification based on the matching result and generating mechanism verification results further includes: when the fault type is a sealing fault, obtaining the spectrum and envelope spectrum based on the general features; determining the high-frequency friction component based on the spectrum, and determining the rotational speed frequency and harmonics based on the envelope spectrum; determining the harmonic amplitude ratio based on the harmonics; performing a sealing mechanism verification operation on the feature vector based on the high-frequency friction component and the harmonic amplitude ratio to generate a sealing fault verification result; obtaining the vertical / horizontal vibration phase difference; and performing an alignment mechanism verification operation on the feature vector based on the harmonic amplitude ratio and the vertical / horizontal vibration phase difference to generate an alignment fault verification result.

[0015] Accordingly, the present invention also provides an adaptive identification device for composite faults of tunnel ventilation fans. The device includes: a data acquisition unit for acquiring multi-dimensional sensor data of the tunnel ventilation fan; a processing unit for performing environmental noise processing and speed interference processing on the multi-dimensional sensor data to obtain processed data; a feature extraction unit for determining the general features and operating condition features of the tunnel ventilation fan based on the processed data; a decoupling unit for performing feature decoupling operations based on the general features and the operating condition features to obtain feature vectors; and a fault identification unit for performing an adaptive composite fault discrimination operation based on the feature vectors to generate composite fault identification results.

[0016] On the other hand, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method provided in the embodiments of the present invention.

[0017] The present invention has at least the following technical effects through the technical solution provided by the present invention:

[0018] By improving existing conventional wind turbine fault identification methods, a multi-scale analysis method combining adaptive wavelet packet denoising and resonance demodulation can be used to effectively remove strong noise in the tunnel environment, extract weak fault feature signals, significantly improve the signal-to-noise ratio, and enhance the identification rate and accuracy of tunnel wind turbine faults.

[0019] By using a dynamic threshold adjustment strategy that adapts to operating conditions, combined with automatic operating condition identification achieved through FCM clustering, it can adapt to the dynamic operating conditions of frequent start-stop and variable load operation of tunnel ventilation fans, significantly improving the recognition rate of all operating conditions.

[0020] Furthermore, feature overlap is eliminated by decoupling PCA features, and dual verification of the mechanism features of bearing faults and fan blade faults is achieved by combining multi-level rule discrimination, which significantly improves the accuracy of distinguishing composite faults and greatly reduces the misjudgment rate.

[0021] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0022] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0023] Figure 1 This is a flowchart illustrating the specific implementation of the adaptive identification method for composite faults in tunnel ventilation fans provided in this embodiment of the invention.

[0024] Figure 2 This is a schematic diagram of the structure of the tunnel ventilation fan composite fault adaptive identification device provided in an embodiment of the present invention. Detailed Implementation

[0025] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0026] In this invention, the terms "system" and "network" are used interchangeably. "Multiple" refers to two or more; therefore, in this invention, "multiple" can also be understood as "at least two." "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / ", unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship. Furthermore, it should be understood that in the description of this invention, terms such as "first" and "second" are used only for descriptive purposes and should not be construed as indicating or implying relative importance or order.

[0027] Please see Figure 1 This invention provides an adaptive identification method for complex faults in tunnel ventilation fans, the method comprising:

[0028] S10. Acquire multi-dimensional sensor data of the tunnel ventilation fan;

[0029] S20. Perform environmental noise processing and rotational speed interference processing on the multi-dimensional sensing data to obtain processed data;

[0030] S30. Determine the general characteristics and operating conditions of the tunnel ventilation fan based on the processed data;

[0031] S40. Perform feature decoupling operation based on the general features and the wind turbine operating condition features to obtain feature vectors;

[0032] S50. Perform an adaptive composite fault discrimination operation based on the feature vector to generate a composite fault identification result.

[0033] In one possible implementation, multi-dimensional sensing data on tunnel risks are first acquired in real time. For example, vibration, temperature, and speed signals of the tunnel ventilation fan are collected using vibration sensors, temperature sensors, and speed sensors, respectively. The vibration sensor is an ICP-type accelerometer, installed near the tunnel ventilation fan bearing housing. The sampling frequency is set to 5120Hz, the sampling duration is 10 seconds, and the sampling accuracy is 16-bit A / D conversion, ensuring that the vibration signal covers the bearing fault characteristic frequency band (1kHz-10kHz) and the fan blade fault characteristic frequency band (0.1Hz-500Hz). The temperature sensor is a PT100 platinum resistance temperature sensor, installed on the surface of the tunnel ventilation fan housing, used to monitor the bearing housing temperature and motor winding temperature. The temperature measurement range is -50℃ to 200℃, and the measurement accuracy is ±0.1℃. The speed sensor is a photoelectric speed sensor, installed near the fan main shaft, used to acquire the fan speed signal in real time. The speed measurement range is 0 to 3000rpm, and the speed resolution is better than 1rpm.

[0034] After collecting the aforementioned multi-dimensional sensor data, it is preprocessed. Specifically, due to the complex working environment of tunnel ventilation fans, which are affected by strong tunnel noise, dust, temperature changes, and frequent start-stop operations, the acquisition of sensor data is greatly interfered with. Therefore, conventional data processing preprocessing methods cannot reliably eliminate the operating noise of tunnel ventilation fans. At the same time, because the fault data of tunnel ventilation fans is weaker than the interference noise during daily operation and is not easy to detect, if the raw data is used directly for fault detection and analysis, early fault signals may be missed, and the fault may only be discovered when it is obvious, which is not in line with the actual interests of the enterprise.

[0035] To address the aforementioned technical problems, in this embodiment of the invention, the step of performing environmental noise processing and rotational speed interference processing on the multi-dimensional sensing data to obtain processed data includes: determining a preset energy threshold; performing adaptive analysis of the wavelet decomposition level of the multi-dimensional sensing data based on the preset energy threshold to determine the adaptive decomposition level; performing wavelet packet decomposition on the multi-dimensional sensing data based on the Daubechies-db6 wavelet function and the adaptive decomposition level to obtain decomposed data; performing adaptive bandpass filtering on the decomposed data to obtain filtered data; performing resonance demodulation processing on the filtered data based on Hilbert transform to obtain the bearing outer ring fault characteristic frequency, bearing inner ring fault characteristic frequency, and rolling element fault characteristic frequency; and generating processed data based on the bearing outer ring fault characteristic frequency, bearing inner ring fault characteristic frequency, and rolling element fault characteristic frequency.

[0036] In one possible implementation, to address the issues of strong background noise and non-stationary vibration in the tunnel environment, basic noise reduction processing is first performed on the original multi-dimensional sensor data. Specifically, a preset energy threshold is first determined, for example, 0.01. Then, an adaptive analysis of the wavelet decomposition layer number is performed on the multi-dimensional sensor data to determine the adaptive decomposition layer number. For example, by sequentially calculating the signal energy of each decomposition node of the wavelet decomposition of the multi-dimensional sensor data, the decomposition automatically stops when the ratio of the energy of a certain layer to the total energy is less than the preset energy threshold of 0.01, thereby achieving adaptive control of the decomposition depth. At this point, the adaptive decomposition layer number can be determined.

[0037] Then, the Daubechies-db6 wavelet function is used in conjunction with the adaptive decomposition layer to perform wavelet packet decomposition on the multi-dimensional sensor data. The Daubechies-db6 wavelet function has tight support characteristics and approximate symmetry, which is suitable for processing non-stationary vibration signals caused by the start-up and shutdown of tunnel fans and load changes in tunnel environments. Preferably, the adaptive decomposition layer is 3-6 layers. By determining the adaptive decomposition depth, noise can be accurately removed while retaining weak fault characteristics, and a high signal-to-noise ratio preliminary vibration signal can be obtained, laying the foundation for subsequent demodulation and filtering.

[0038] In a preferred embodiment, the Symlet wavelet function (sym4) can also be used for wavelet packet decomposition. This function has better symmetry and can effectively reduce phase distortion. The adaptive judgment threshold for the number of decomposition layers can be adjusted from 0.01 to 0.005 to retain more weak fault features and improve the fault identification rate.

[0039] During wavelet packet decomposition, the noise reduction threshold for the wavelet packet is determined using an adaptive soft thresholding strategy. Specifically, the standard deviation estimate is calculated using the absolute value of the median. Where MAD is the median absolute value function, These are the high-frequency coefficients after wavelet decomposition. Then, the noise reduction threshold is calculated and determined based on the signal length. Where N is the signal length, during the processing, coefficients with absolute values ​​less than the threshold are set to zero by the shrinkage function, and coefficients with absolute values ​​greater than the threshold are shrunk. After the soft thresholding is completed, the signal is reconstructed by the wavelet packet reconstruction algorithm to obtain the vibration signal after preliminary noise reduction.

[0040] Because the tunnel ventilation fan itself generates variable speed and signal-to-noise ratio interference, it can significantly affect the accuracy of the data. Traditional bandpass filters have a fixed center frequency, and once the tunnel ventilation fan's speed changes, they cannot accurately capture fault information, thus failing to meet practical requirements. To solve this technical problem, an adaptive bandpass filter is further employed to process the decomposed data to obtain filtered data.

[0041] In this embodiment of the invention, performing adaptive bandpass filtering on the decomposed data to obtain filtered data includes: acquiring the rotational speed frequency of the tunnel fan; determining the center frequency of the tunnel fan based on the rotational speed frequency; determining the noise power of the tunnel fan; dynamically determining the quality factor based on the signal power of the tunnel fan and the noise power; and performing adaptive bandpass filtering on the decomposed data based on the center frequency and the quality factor to obtain filtered data.

[0042] Specifically, the operating conditions of the tunnel ventilation fan are acquired in real time, including its speed and frequency. Then determine the center frequency of the adaptive bandpass filter: 'k' is the harmonic factor, and the center frequency automatically adjusts with the rotational speed and its harmonics, effectively adapting to fan speed fluctuations. Simultaneously, by estimating the signal-to-noise ratio in real time, for example, based on the ratio of signal power to noise power (e.g., obtained through power statistics in the low-frequency band of the signal (0 to 50 Hz), the quality factor (Q) of the filter is dynamically adjusted to suppress noise while preserving effective characteristics. For example, its adjustment range is 2 to 10. When the signal-to-noise ratio is higher than 15 dB, the Q value is reduced to 2 to 4 to widen the bandwidth and retain more signal components; when the signal-to-noise ratio is lower than 10 dB, the Q value is increased to 6 to 10 to compress the bandwidth, improve filtering accuracy, and suppress noise interference.

[0043] In this embodiment of the invention, by dynamically adjusting the filtering of the tunnel ventilation fan's operating data according to its actual operating conditions, adaptive monitoring of the tunnel ventilation fan's operating conditions can be achieved. No matter how the speed or noise changes, the filter can accurately identify fault characteristics, thereby greatly improving the accuracy of subsequent fault identification of the tunnel ventilation fan.

[0044] Furthermore, based on the Hilbert transform, resonance demodulation processing is performed on the filtered data. Since bearing pitting and spalling will generate high-frequency resonance impacts, this signal is easily submerged by noise and cannot be identified or detected, resulting in the inability to identify early faults of tunnel ventilation fans in a timely and accurate manner. By using the Hilbert transform to extract the envelope, the high-frequency resonance can be converted into a low-frequency envelope spectrum, thereby making the bearing fault impacts that are submerged by noise explicit, which can effectively assist in the identification of bearing composite faults.

[0045] Specifically, in order to capture early, subtle fault information of the tunnel ventilation fan bearings, a Hilbert transform is first performed on the filtered data to extract the envelope signal, which can be represented as follows: This process converts the original signal into an analytic signal, where the real part is the original signal and the imaginary part is the result of the Hilbert transform. The envelope signal can be obtained from the magnitude of this analytic signal. A Fast Fourier Transform (FFT) is then performed on the envelope signal to obtain the envelope spectrum. Based on the bearing fault characteristic frequency band, the demodulation frequency band range can be determined, for example, from 1kHz to 10kHz. From this envelope spectrum, the bearing outer ring fault characteristic frequency (BPFO), bearing inner ring fault characteristic frequency (BPFI), and rolling element fault characteristic frequency (BSF) can be extracted, for example, represented as: , and Where Z is the number of rolling elements, n is the spindle speed (rpm), d is the diameter of the rolling elements, D is the bearing pitch circle diameter, and α is the contact angle.

[0046] At this point, processed data is generated based on the aforementioned bearing outer ring fault characteristic frequency, bearing inner ring fault characteristic frequency, and rolling element fault characteristic frequency to facilitate accurate fault identification in the future.

[0047] In this embodiment of the invention, by using three methods—adaptive wavelet packet denoising, adaptive bandpass filtering, and resonance demodulation—to perform joint preprocessing on the original signal, tunnel-environment noise can be effectively removed and weak fault features can be extracted to obtain a denoised vibration signal. This can greatly improve the reliability of subsequent feature extraction, decoupling, and discrimination, increase the recognition rate of composite faults, and improve the fault recognition rate under all working conditions.

[0048] After obtaining the processed data, multidimensional feature extraction is performed to further support the accurate identification of subsequent complex faults. Specifically, firstly, general features of the tunnel ventilation fan are extracted, including but not limited to higher-order time-domain features, frequency-domain features, refined spectral features, and time-frequency energy features. Specifically, higher-order time-domain features include but are not limited to signal peak-to-peak value, skewness, kurtosis factor, impulse factor, margin factor, and waveform factor, for example, skewness is characterized as: Where μ is the signal mean, σ is the standard deviation, and E(·) is the expected value. The skewness can determine the degree of deviation of the signal symmetry; the kurtosis factor is characterized as: The kurtosis factor can be used to determine the significance of the impact characteristics of tunnel ventilation fan bearings; a larger kurtosis factor indicates a more significant impact component in the signal. Based on the signal peak value... and signal RMS value The pulse factor can be determined to characterize the intensity of the impact pulse; the margin factor is characterized as follows: ,in, The peak value of the signal. The square root amplitude is used to characterize the degree to which the signal deviates from the mean; the waveform factor is characterized as follows: ,in, The mean value of the signal is used to characterize the degree of waveform distortion.

[0049] Since wind turbine imbalance, misalignment, blade cracks, and other anomalies are clearly reflected in the frequency harmonics during operation, frequency domain features are extracted for further fault identification. Frequency domain feature extraction is based on Fast Fourier Transform (FFT). Specifically, the sampled data is processed through a Hanning window and then subjected to an FFT to obtain the spectrum. The frequency domain features include the dominant frequency and its amplitude. For example, the frequency point with the largest amplitude in the spectrum is taken as the dominant frequency, and its corresponding amplitude is recorded. The amplitude is determined by extracting the amplitudes of the 1st to 5th harmonic components based on the first five harmonic components. On this basis, the spectral centroid and spectral variance are calculated. Specifically, the spectral centroid is represented as: , For the i-th frequency point, Let be the amplitude at the i-th frequency point; this feature reflects the central tendency of the spectral energy distribution. The spectral variance is characterized as: This feature reflects the degree of dispersion in the spectral energy distribution.

[0050] Based on this, due to the characteristics of tunnel ventilation applications, such as strong tunnel noise, variable operating conditions, and numerous complex faults, traditional fault identification or signal processing methods suffer from significant interference and noise. The fault identification rate for tunnel ventilation is often only around 70%, unable to reach over 95%. To further improve the fault identification rate, time-frequency energy features are extracted. Specifically, a four-layer wavelet packet decomposition is used to decompose the signal into 16 frequency bands. The wavelet packet decomposition employs the Daubechies-db4 wavelet function. During the decomposition process, the sub-band energy is calculated at each node, specifically represented as follows: ,in, For the wavelet packet coefficients of the i-th frequency band, the energy of each frequency band is normalized to a probability distribution: At this point, Shannon entropy is calculated based on normalized energy as a time-frequency energy feature: This feature is used to characterize the distribution characteristics of signal energy in different frequency bands. When a fault occurs, the energy distribution will change significantly.

[0051] For conventional fans or motors, their operating conditions are generally relatively stable. However, in actual applications, tunnel fans are subject to frequent start-stop, speed changes, and load changes. For example, during start-stop, the impact is significant and can easily be misjudged as a fault, causing great trouble for technicians and significantly increasing the maintenance costs or safety risks of tunnel fans.

[0052] In this embodiment of the invention, determining the operating condition characteristics of the tunnel ventilation fan based on the processed data includes: determining a fault characteristic frequency band; determining a Zoom-FFT analysis center frequency based on the fault characteristic frequency band; performing complex modulation processing on the processed data based on the Zoom-FFT analysis center frequency to obtain a complex-modulated signal; performing low-pass filtering processing on the complex-modulated signal to obtain a low-pass filtered signal; performing FFT transform on the low-pass filtered signal to obtain refined spectral features; determining the speed change rate and start / stop status based on the real-time speed of the tunnel ventilation fan; generating operating condition characteristics based on the speed change rate and the start / stop status; and generating ventilation fan operating condition characteristics based on the refined spectral features and the operating condition characteristics.

[0053] In one possible implementation, the operating characteristics of the tunnel ventilation fan include refined spectral characteristics and operating condition characteristics. While all motor fault analyses utilize high-resolution frequencies, tunnel ventilation fans exhibit characteristics such as low speed, weak fault characteristics, and strong environmental noise during operation. Therefore, it is necessary to refine their spectrum to enable the analysis and identification of weak signals, improving the adaptability of fault identification in tunnel ventilation scenarios. Specifically, by refining the fault characteristic frequency band (0.5... Up to 5 The target frequency resolution is better than 0.1Hz, determined using a complex modulation Zoom-FFT algorithm for refined analysis. Specifically, the frequency band of interest is first shifted to near zero frequency through complex modulation, and the complex-modulated signal is represented as follows: ,in For Zoom-FFT analysis center frequency, Here, n is the sampling frequency, n is the sampling point number, and j is the imaginary unit. The complex-modulated signal is then low-pass filtered to remove high-frequency components. A finite impulse response (FIR) digital filter can be used, with the passband cutoff frequency set to 0.5 times the target analysis bandwidth and the stopband attenuation greater than 60 dB. Next, the filtered signal is resampled. The resampling rate is determined based on the target resolution, and the rate reduction factor is the ratio of the target bandwidth to the original bandwidth. Finally, an FFT transform is performed on the resampled signal to obtain refined spectral features.

[0054] At this point, the real-time rotational speed of the tunnel ventilation fan is further acquired, and the rotational speed change rate is further determined, as well as the corresponding start-up and shutdown states are generated. For example, when the rotational speed rises from rest to more than 10% of the rated speed, it is marked as the start-up state; when the rotational speed drops to less than 10% of the rated speed, it is marked as the shutdown state; and the remaining states are marked as the running state. The current signal can also be further acquired and its time derivative calculated to generate the load current change rate. At this point, the ventilation fan operating condition characteristics can be generated based on the above rotational speed change rate, start-up state, and load current change rate. The load current change rate is not a mandatory item; in some embodiments, it can be used as a supplementary item to further improve the dimensionality of the operating condition characteristics and help improve the accuracy of tunnel ventilation fan fault identification.

[0055] In this embodiment of the invention, by extracting the general features of tunnel ventilation fans, multi-dimensional identification of potential faults can be achieved from the perspective of the motor. At the same time, based on the actual physical operating conditions of tunnel ventilation fans, targeted feature extraction is performed for situations such as high tunnel wind noise, structural vibration, electromagnetic interference, easy obscuring of single features, multiple compound faults, and extreme changes in operating conditions. This allows for more accurate analysis and identification of tunnel ventilation fan faults from multiple dimensions, improving the subsequent fault identification rate (from about 70% to 95%) and identification accuracy.

[0056] After determining the general features and the operating characteristics of the wind turbine, fault identification can be performed. However, in practical applications, the acquired features have many dimensions, and these features are highly correlated and overlapping. If directly input into a classifier or discriminator, it will lead to problems such as significantly reduced computation speed, model overfitting, and noise amplification, resulting in faults that cannot be efficiently and accurately identified and distinguished, failing to meet the actual needs of tunnel wind turbine fault identification. Therefore, further feature decoupling is performed to optimize features and improve the efficiency and accuracy of subsequent discrimination algorithms.

[0057] In this embodiment of the invention, the step of performing feature decoupling operation based on the general features and the wind turbine operating condition features to obtain feature vectors includes: generating fused features based on the general features and the wind turbine operating condition features; determining the covariance matrix of the fused features; determining the variance contribution rate of each principal component based on the covariance matrix; generating a decoupled feature matrix based on the variance contribution rate and a preset contribution rate threshold; determining the inter-class variance and intra-class variance based on the decoupled feature matrix; and filtering the decoupled feature matrix based on Fisher's criterion, the inter-class variance, and the intra-class variance to obtain feature vectors.

[0058] In one possible implementation, fused features are first generated based on the aforementioned general features and wind turbine operating condition features. For example, high-order time-domain features (6-dimensional), frequency-domain features (9-dimensional), refined spectrum features (5-dimensional), time-frequency energy features (16-dimensional), and operating condition features (3-dimensional) are sequentially concatenated to obtain fused features with a total feature dimension of no less than 50 dimensions. Then, a feature matrix is ​​constructed. Where n is the number of samples and m is the feature dimension, the covariance matrix is ​​calculated based on this feature matrix, for example, represented as: Then, the eigenvalues ​​λ and corresponding eigenvectors v of the covariance matrix are solved, and the variance contribution rate of each principal component is calculated, for example, represented as: ,in For the i-th eigenvalue, a decoupled feature matrix is ​​generated based on the variance contribution rate and a preset contribution rate threshold. Specifically, principal component analysis is used to retain k principal components whose cumulative variance contribution rate is greater than a preset contribution rate threshold (e.g., 95%), for example, represented as: This achieves decoupling and dimensionality reduction of the fused features, eliminating overlapping interference between features.

[0059] In this embodiment of the invention, by employing feature decoupling, fault information and noise information are separated, retaining only 95% of the effective information. This filters out noise, increases the visibility of weak noise, and improves the efficiency and accuracy of subsequent fault identification. Simultaneously, through this decoupling operation, the dimensions of the features can be reduced from 50 to approximately 10, significantly reducing computational complexity, improving the efficiency of fault calculation for tunnel ventilation fans, and meeting the online real-time monitoring requirements of tunnel ventilation fans.

[0060] Further, based on the decoupled feature matrix, the inter-class variance and intra-class variance are determined. For example, the formula for calculating the inter-class variance is: Where c is the number of categories, Let be the prior probability of the i-th class. Let be the feature mean of the i-th class, and μ be the population mean. The formula for calculating the within-class variance is: ,in, Let be the number of samples in the i-th category. Therefore, according to Fisher's criterion, a corresponding criterion function can be constructed, for example, represented as:

[0061]

[0062] In the final feature selection process, features with J values ​​greater than a preset threshold (e.g., 1.0) are selected to eliminate features with weak discriminative ability, thereby achieving feature optimization and improving the efficiency and accuracy of subsequent discrimination algorithms.

[0063] In this embodiment of the invention, by using a feature structure combined with Fisher's criteria for secondary screening, features with weaker distinguishing ability can be further filtered out, and features with the strongest distinguishing ability can be selected. This allows bearing faults, blade faults, and motor faults to each have their own independent feature channels, thereby facilitating the accurate identification of subsequent fault features and effectively overcoming the problems of feature drift and high false alarms caused by frequent start-up and shutdown of tunnel ventilation fans and changing operating conditions.

[0064] After generating the aforementioned feature vectors, fault identification begins.

[0065] In this embodiment of the invention, the step of performing adaptive composite fault discrimination based on the feature vector to generate a composite fault identification result includes: processing the feature vector based on a fuzzy C-means clustering algorithm to obtain multiple operating condition categories; obtaining a preset fault feature library; matching the preset fault feature library based on the feature vector and the multiple operating condition categories to obtain a matching result; performing fault mechanism verification based on the matching result to generate a mechanism verification result; determining the prior probability corresponding to each fault type based on historical fault data; generating a posterior probability for tunnel ventilation fan faults based on the matching result, the mechanism verification result, and the prior probability; and generating a composite fault identification result based on the posterior probability.

[0066] In one possible implementation, to achieve rapid location of tunnel ventilation fan faults, after obtaining the aforementioned decoupled clean features, the fuzzy C-means clustering algorithm is first used to cluster the feature vectors to classify them into operating conditions, for example, dividing the operating conditions into three categories: start-stop state (labeled 1), steady state (labeled 2), and transition state (labeled 3). The clustering objective function is, for example, characterized as: ,in, Let m be the membership degree of sample i to category j, and m be the fuzzy factor (e.g., a value of 2). Let be the distance from sample i to the center of category j, and let the membership function satisfy the normalization constraint. During the clustering iteration process, an alternating optimization strategy is used to update the membership degree and cluster center until convergence or the maximum number of iterations is reached (e.g., it can be set to 100). Finally, the current working condition category is determined based on the maximum membership degree, providing working condition context information for subsequent discrimination.

[0067] After determining the initial operating condition category, a preset fault feature library is obtained. This library, for example, is a fault feature library generated in advance by technicians based on historical fault samples through statistical learning. It includes categories such as normal operating conditions, bearing outer ring faults, bearing inner ring faults, rolling element faults, fan blade imbalance, and fan blade cracks. It includes the mean vector and covariance matrix of each category's features. Then, the Mahalanobis distance between the current feature vector and each category's feature library is calculated, for example, represented as: ,in, and Let be the feature mean vector and covariance matrix of the i-th category, respectively. Then, the category with the smallest distance is taken as the preliminary discrimination result, i.e. the matching result, thus realizing the coarse screening of fault categories.

[0068] However, in practical applications, identifying and judging fault categories solely based on data similarity has significant drawbacks. It does not focus on the specific physical mechanism, thus leading to misjudgments and failing to meet actual needs. Therefore, in this embodiment of the invention, the fault mechanism is further verified to improve the accuracy of fault identification.

[0069] In this embodiment of the invention, the step of performing fault mechanism verification based on the matching result and generating verification results includes: when the matching result indicates the existence of a fault, obtaining the fault type; when the fault type is a bearing fault, performing a bearing mechanism verification operation on the feature vector based on a preset bearing mechanism verification rule to generate a bearing fault verification result; when the fault type is a fan blade fault, performing a fan blade mechanism verification operation on the feature vector based on a preset fan blade mechanism verification rule to generate a fan blade fault verification result; when the fault type is an abnormal vibration, performing a vibration mechanism verification operation on the feature vector based on a preset temperature-vibration verification rule to generate a vibration fault verification result; and generating a mechanism verification result based on the bearing fault verification result, the fan blade fault verification result, and the vibration fault verification result.

[0070] Specifically, if the initial matching reveals a fault in the tunnel ventilation fan, the fault type is further determined based on the matching results. If the initially identified fault type is a bearing fault, the presence of bearing outer ring fault characteristic frequencies, bearing inner ring fault characteristic frequencies, rolling element fault characteristic frequencies, and their sideband components in the envelope spectrum is detected. When the error between the detected frequency components and the theoretical characteristic frequencies is less than 2%, it is determined that bearing fault characteristics exist in the envelope spectrum. The detection of sideband components uses a spectral peak search method, with the search range being the characteristic frequency ± the rotational frequency interval.

[0071] If the initial fault type is identified as a fan blade fault, the presence of 1x, 2x, and 3x frequency components in the frequency spectrum is detected. When the ratio of the amplitude of the frequency component to the amplitude of the fundamental frequency component is greater than 0.3, it is determined that there are fan blade fault characteristics in the frequency spectrum. By utilizing the mechanism that fan blade imbalance or crack faults will lead to a significant increase in frequency and its harmonic components, the accurate identification of specific tunnel fan faults can be achieved.

[0072] If the initial fault type is identified as abnormal vibration, the temperature characteristics are simultaneously detected when the vibration characteristics are abnormal. When the vibration characteristic value exceeds the normal range and the temperature change exceeds the threshold (set to 5℃), a temperature-vibration fault correlation is established, and the fault type is determined to be a related fault such as poor bearing lubrication or motor overheating.

[0073] In this embodiment of the invention, by accurately identifying subdivided faults based on feature vectors and specific mechanism verification rules for subdivided faults, it is possible to effectively identify misjudgments in the initial fault identification, prevent normal impacts from being mistaken for faults, and thus improve the accuracy and reliability of tunnel fan fault identification.

[0074] After generating the above matching and verification results, the fault is accurately identified. Specifically, the prior probability corresponding to each fault type is determined based on historical fault data. For example, technicians pre-determine the prior probability of different fault types based on the historical occurrence frequency of each fault according to historical fault statistics of tunnel ventilation fans. Then, based on the above discrimination results, the corresponding posterior probability is determined using the Bayesian posterior probability calculation formula, which can be represented as follows: ,in, Let be the prior probability of the i-th class. The likelihood function is calculated from the matching results. In the specific fault determination process, the category with the highest posterior probability is selected as the final judgment result, and the output confidence level is the maximum posterior probability value to ensure the confidence and reliability of the final judgment result. When the confidence level is lower than 0.7, an alarm message is output to indicate that manual review is required.

[0075] For example, in a specific application instance, real-time monitoring is performed on an axial flow tunnel fan in a ventilation system of a long tunnel. The fan has a rated power of 315kW, a rated speed of 1480rpm, and is equipped with double-row rolling bearings (model: SKF22232, number of rolling elements Z=16, pitch circle diameter D=200mm, rolling element diameter d=30mm, contact angle α=0°). The vibration sensor is installed vertically on the bearing housing at the drive end, the sampling frequency is set to 5120Hz, and a continuous acquisition mode is used. The temperature sensor is installed on the bearing housing housing to monitor the bearing temperature rise.

[0076] During real-time monitoring, when the fan is in steady-state operation, the vibration signal exhibits relatively stable random vibration characteristics. The theoretical value of the bearing outer ring fault characteristic frequency (BPFO) is calculated as follows:

[0077]

[0078] Under steady-state conditions, adaptive wavelet packet denoising was performed. The decomposition level was adaptively determined to be 5 levels. The denoising threshold was calculated using the median absolute value method. After resonant demodulation, the envelope spectrum was extracted. A significant spectral peak was detected near 177.6Hz, with the peak amplitude exceeding the normal value by more than 3 times, accompanied by a frequency shift sideband component (1480 / 60=24.67Hz). The first-level FCM clustering determined the steady-state condition. The second-level Mahalanobis distance calculation determined the minimum distance to the bearing outer ring fault category. The third-level rule verification detected the BPFO characteristic frequency and its sidebands, with a frequency error of less than 1%, satisfying the bearing characteristic frequency verification condition. The fourth-level Bayesian posterior discriminant calculation calculated the posterior probability, and the posterior probability of the bearing outer ring fault category reached 0.93. The final output judgment result was bearing outer ring fault, with a confidence level of 0.93. Based on the judgment result, the maintenance personnel inspected the bearing and found pitting and peeling faults on the bearing outer ring. The bearing was replaced in time, preventing further expansion of the fault.

[0079] In another embodiment, when the fan is in the start-up transition state, the speed gradually increases from 0 rpm to the rated speed. During the start-up process, the vibration signal contains a large impact component, and the speed change rate is high. The first-level FCM clustering determines the operating condition as a transition state. The threshold adaptive adjustment module relaxes the vibration feature threshold by 1.5 times to avoid misjudgment caused by normal start-up impact. The second-level Mahalanobis distance calculation determines that the distance from the normal operating condition is the smallest. The third-level rule verification does not detect any abnormal bearing characteristic frequency or fan blade characteristic frequency. The fourth-level Bayesian posterior discriminant calculation calculates the posterior probability. The posterior probability of the normal operating condition category reaches 0.87, and the final output judgment result is normal state. This example verifies the accurate differentiation capability of the present invention between normal start-up impact and fault characteristics under transition state operating conditions.

[0080] In this embodiment of the invention, by adopting a multi-level adaptive discrimination method, accurate faults can be identified based on the actual operating mechanism of the tunnel ventilation fan, and corresponding identification results and confidence levels can be provided. The three levels constrain and complement each other, effectively improving the identification reliability of the tunnel ventilation fan under strong noise, variable operating conditions, and complex fault scenarios, thus meeting the actual needs of enterprises.

[0081] In practical applications, the operating conditions of tunnel ventilation fans differ significantly from those of ordinary fans or conventional motors. On the one hand, tunnel ventilation fans operate at high speeds for extended periods without human intervention, making misalignment, imbalance, and seal failure common faults. On the other hand, tunnels are often damp, dusty, and contain corrosive gases, which makes the fan seals prone to aging and failure. Furthermore, tunnel ventilation fans do not operate continuously in a steady state but require frequent start-stop, acceleration, deceleration, and variable air volume adjustments based on factors such as traffic flow, smoke, and CO concentration. Therefore, traditional fan fault analysis methods are insufficient, and the three fault classifications mentioned above also fail to meet the needs of analyzing actual operating conditions.

[0082] In this embodiment of the invention, the step of performing fault mechanism verification based on the matching result and generating mechanism verification results further includes: when the fault type is a sealing fault, obtaining a spectrum and envelope spectrum based on the general features; determining a high-frequency friction component based on the spectrum, and determining the rotational speed frequency and harmonics based on the envelope spectrum; determining the harmonic amplitude ratio based on the harmonics; performing a sealing mechanism verification operation on the feature vector based on the high-frequency friction component and the harmonic amplitude ratio to generate a sealing fault verification result; obtaining the vertical / horizontal vibration phase difference; and performing an alignment mechanism verification operation on the feature vector based on the harmonic amplitude ratio and the vertical / horizontal vibration phase difference to generate an alignment fault verification result.

[0083] In one possible implementation, to further improve the precision of fault identification for tunnel ventilation fans and achieve accurate identification of specific operating conditions and fault types in tunnel application scenarios, in addition to the aforementioned bearing fault verification results, fan blade fault verification results, and vibration fault verification results, the identification of categories such as gas seal faults, coupling misalignment faults, and rotor imbalance faults is further added. Specifically, by further adding categories such as gas seal faults, coupling misalignment faults, and rotor imbalance faults to the preset fault feature library to cover more fault types, during cluster identification, since tunnel ventilation fans themselves vibrate significantly during acceleration / deceleration, using "steady-state" analysis would lead to false alarms.

[0084] Therefore, during the clustering process, the acceleration and deceleration states of the tunnel ventilation fan are further identified to refine the granularity of the operating condition classification. Based on this, matching and identification are performed with a dedicated fault feature library corresponding to acceleration / deceleration, and the corresponding mean and covariance are used when calculating the Mahalanobis distance. In subsequent verification, the thresholds of the verification rules are dynamically adjusted, such as increasing the search tolerance of bearing fault frequency BPFO / BPFI, increasing the threshold of fan blade overtone amplitude, increasing the threshold of misalignment amplitude ratio, and increasing the threshold of seal friction energy, thereby effectively avoiding the identification of normal impacts as faults under acceleration / deceleration conditions.

[0085] Furthermore, when performing mechanism verification, mechanism verification for sealing failure and counterweight failure is also added. Specifically, the spectrum and envelope spectrum are obtained according to the above general characteristics, and the rotational frequency and harmonics are extracted from them. The high-frequency friction component is determined from the spectrum (for example, the 500Hz-2000Hz frequency band is truncated). The amplitudes of 1 times the rotational frequency, 2 times the rotational frequency and 3 times the rotational frequency are determined according to the rotational frequency and harmonics respectively. The harmonic amplitude ratio is further determined, for example, 2 times amplitude / 1 times amplitude, 3 times amplitude / 1 times amplitude. At this time, the sealing mechanism verification operation is further performed. For example, when any two of the following conditions are met at the same time, it is determined that there is a sealing failure: (1) 2 times amplitude / 1 times amplitude > 0.4; (2) 3 times amplitude / 1 times amplitude > 0.25; (3) the high-frequency friction component > 1.8 times the normal average value.

[0086] Meanwhile, for centering faults, if any of the following conditions are met, a centering fault is determined to exist: (1) 2 times amplitude / 1 times amplitude > 0.6; (2) 1 times amplitude > 2 times the normal value; (3) 1 times amplitude and 2 times amplitude both exceed the threshold and 2 times amplitude / 1 times amplitude is between 0.5 and 0.8.

[0087] The above verification can effectively identify minor faults in tunnel ventilation fans in the early stages of use, rather than judging them as normal vibrations. This is beneficial for early warning and maintenance of tunnel ventilation fans, improving their service life and ensuring early identification of tunnel ventilation fan faults.

[0088] In this embodiment of the invention, by further adopting the matching identification of acceleration and deceleration states, and introducing sealing verification and centering verification, a scenario-based fault optimization identification scheme is realized based on the environmental characteristics, operating characteristics and fault characteristics of tunnel ventilation fans. This further improves the accuracy and reliability of tunnel ventilation fan fault identification, reduces enterprise maintenance costs, and meets the actual needs of enterprises.

[0089] Please see Figure 2 Based on the same inventive concept, embodiments of the present invention provide an adaptive identification device for composite faults of tunnel ventilation fans. The device includes: a data acquisition unit for acquiring multi-dimensional sensor data of the tunnel ventilation fan; a processing unit for performing environmental noise processing and speed interference processing on the multi-dimensional sensor data to obtain processed data; a feature extraction unit for determining the general features and operating condition features of the tunnel ventilation fan based on the processed data; a decoupling unit for performing feature decoupling operations based on the general features and the operating condition features to obtain feature vectors; and a fault identification unit for performing an adaptive composite fault discrimination operation based on the feature vectors to generate composite fault identification results.

[0090] Furthermore, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the methods described in the embodiments of the present invention.

[0091] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention.

[0092] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not describe the various possible combinations separately.

[0093] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0094] Furthermore, various different implementations of the present invention can be combined arbitrarily, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed in the present invention.

Claims

1. An adaptive identification method for composite faults in tunnel ventilation fans, characterized in that, The method includes: Acquire multi-dimensional sensor data of tunnel ventilation fans; Environmental noise processing and rotational speed interference processing are performed on the multi-dimensional sensor data to obtain processed data; Based on the processed data, the general characteristics and operating conditions of the tunnel ventilation fan are determined. Based on the general features and the wind turbine operating condition features, a feature decoupling operation is performed to obtain a feature vector; Based on the feature vector, an adaptive composite fault discrimination operation is performed to generate composite fault identification results.

2. The method according to claim 1, characterized in that, The step of performing environmental noise processing and rotational speed interference processing on the multi-dimensional sensing data to obtain processed data includes: Determine the preset energy threshold; Based on the preset energy threshold, the number of wavelet decomposition layers for the multidimensional sensing data is adaptively analyzed to determine the adaptive decomposition layer. Wavelet packet decomposition is performed on the multidimensional sensor data based on the Daubechies-db6 wavelet function and the adaptive decomposition level to obtain the decomposed data. An adaptive bandpass filter is performed on the decomposed data to obtain filtered data; Based on the Hilbert transform, resonance demodulation processing is performed on the filtered data to obtain the bearing outer ring fault characteristic frequency, the bearing inner ring fault characteristic frequency, and the rolling element fault characteristic frequency. Processed data is generated based on the bearing outer ring fault characteristic frequency, the bearing inner ring fault characteristic frequency, and the rolling element fault characteristic frequency.

3. The method according to claim 2, characterized in that, The step of performing adaptive bandpass filtering on the decomposed data to obtain filtered data includes: Obtain the rotational speed and frequency of the tunnel ventilation fan; The center frequency of the tunnel ventilation fan is determined based on the rotational speed frequency. The noise power of the tunnel ventilation fan is determined, and the quality factor is dynamically determined based on the signal power and noise power of the tunnel ventilation fan. Based on the center frequency and the quality factor, the decomposed data is subjected to adaptive bandpass filtering to obtain filtered data.

4. The method according to claim 1, characterized in that, The step of determining the operating characteristics of the tunnel ventilation fan based on the processed data includes: Determine the fault characteristic frequency band, and determine the Zoom-FFT analysis center frequency based on the fault characteristic frequency band; The processed data is subjected to complex modulation based on the Zoom-FFT analysis center frequency to obtain a complex modulated signal; The complexed signal is subjected to low-pass filtering to obtain a low-pass filtered signal; The refined spectral features are obtained by performing an FFT transform on the low-pass filtered signal. The rotational speed change rate and start / stop status are determined based on the real-time rotational speed of the tunnel ventilation fan. Operating condition characteristics are generated based on the speed change rate and the start / stop state; Wind turbine operating characteristics are generated based on the refined spectral features and the operating condition features.

5. The method according to claim 1, characterized in that, The step of performing feature decoupling operation based on the general features and the wind turbine operating condition features to obtain feature vectors includes: A fused feature is generated based on the general features and the wind turbine operating condition features; Determine the covariance matrix of the fusion features, and determine the variance contribution rate of each principal component based on the covariance matrix; A decoupled feature matrix is ​​generated based on the variance contribution rate and the preset contribution rate threshold. The inter-class variance and intra-class variance are determined based on the decoupled feature matrix. The decoupled feature matrix is ​​filtered based on Fisher's criterion, the inter-class variance, and the intra-class variance to obtain the feature vector.

6. The method according to claim 1, characterized in that, The step of performing adaptive composite fault discrimination based on the feature vector to generate composite fault identification results includes: The feature vectors are processed using a fuzzy C-means clustering algorithm to obtain multiple working condition categories; Obtain a preset fault feature library, and match the preset fault feature library based on the feature vector and the multiple operating condition categories to obtain matching results; Based on the matching results, a fault mechanism verification is performed, and a mechanism verification result is generated. Determine the prior probability corresponding to each fault type based on historical fault data; Based on the matching results, the mechanism verification results, and the prior probability, a posterior probability for tunnel ventilation fan failure is generated. A composite fault identification result is generated based on the posterior probability.

7. The method according to claim 6, characterized in that, The step of performing fault mechanism verification based on the matching result and generating mechanism verification results includes: If the matching result indicates the presence of a fault, the fault type is obtained; When the fault type is a bearing fault, a bearing mechanism verification operation is performed on the feature vector based on a preset bearing mechanism verification rule to generate a bearing fault verification result. When the fault type is a fan blade fault, a fan blade mechanism verification operation is performed on the feature vector based on a preset fan blade mechanism verification rule to generate a fan blade fault verification result. When the fault type is abnormal vibration, a vibration mechanism verification operation is performed on the feature vector based on a preset temperature-vibration verification rule to generate a vibration fault verification result. Mechanism verification results are generated based on the bearing fault verification results, the fan blade fault verification results, and the vibration fault verification results.

8. The method according to claim 7, characterized in that, The step of performing fault mechanism verification based on the matching result and generating mechanism verification result further includes: In the case where the fault type is a sealing fault, the spectrum and envelope spectrum are obtained based on the general characteristics; The high-frequency friction component is determined based on the spectrum, and the rotational speed frequency and harmonics are determined based on the envelope spectrum. The frequency doubling amplitude ratio is determined based on the frequency doubling. Based on the high-frequency friction component and the harmonic amplitude ratio, a sealing mechanism verification operation is performed on the feature vector to generate a sealing fault verification result. Obtain the phase difference of vibration in the vertical / horizontal directions; Based on the frequency harmonic amplitude ratio and the vertical / horizontal vibration phase difference, a centering mechanism verification operation is performed on the feature vector to generate a centering fault verification result.

9. A tunnel ventilation fan composite fault adaptive identification device, characterized in that, The device includes: The data acquisition unit is used to acquire multi-dimensional sensor data from the tunnel ventilation fan. The processing unit is used to perform environmental noise processing and rotational speed interference processing on the multi-dimensional sensing data to obtain processed data; The feature extraction unit is used to determine the general features and operating condition features of the tunnel ventilation fan based on the processed data. The decoupling unit is used to perform feature decoupling operations based on the general features and the wind turbine operating condition features to obtain feature vectors; The fault identification unit is used to perform an adaptive composite fault discrimination operation based on the feature vector to generate a composite fault identification result.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the method described in any one of claims 1-8.

Images