Diesel generating set fault detection method and system based on deep learning
By using deep learning methods to separate vibration signals from diesel generator sets, the problem of insufficient accuracy in fault detection in existing technologies has been solved, achieving higher detection accuracy and reliability.
Patent Information
- Application Number
- CN202511494793.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-20
- Publication Date
- 2025-12-12
- Estimated Expiration
- 2045-10-20
AI Technical Summary
Existing fault detection methods for diesel generator sets are ill-suited to the detection of vibration signals in diesel generator sets, making it difficult to effectively detect faults, especially under dynamic operating conditions. Existing technologies struggle to accurately separate and classify fault frequency components, resulting in insufficient detection accuracy.
A deep learning-based approach is employed, using time-frequency decomposition, adaptive filtering, k-means clustering, and random forest validation to separate abnormal frequency components and classify faults.
It improves the accuracy and reliability of fault detection in diesel generator sets, enabling precise detection of early, minor faults and reducing misjudgments.
Smart Images

Figure CN121117754A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of fault detection, and in particular to a diesel generator set fault detection method and system based on deep learning. BACKGROUND
[0002] As a key equipment in industrial production and emergency power supply and infrastructure operation, the running stability and reliability of the diesel generator set are directly related to the continuity of energy supply and production safety. Under the background of continuous growth of power demand, timely and effective fault detection of the generator set is an important means to reduce downtime risk. Vibration signals can directly reflect the running state of internal mechanical components of the equipment and are the core analysis object in the field of fault detection.
[0003] In the prior art, the fault detection method for the vibration signal of the diesel generator set relies on fixed threshold judgment or simple statistical analysis. This kind of method establishes a detection model based on a single frequency or time domain feature, and judges whether the equipment is abnormal by setting a fixed threshold of vibration amplitude. However, the running environment of the diesel generator set is complex, and the vibration signal is easily affected by factors such as load fluctuation, environmental noise and equipment aging, showing significant nonlinearity and randomness. The fixed threshold method is difficult to adapt to the dynamic change of working conditions, and often misjudges normal random fluctuations as faults, or misses early weak fault signals due to too high threshold setting. At the same time, simple statistical analysis has limited ability to extract high-dimensional time-frequency features, and cannot fully capture the dynamic evolution law of fault signals in time and frequency domain, resulting in insufficient adaptability and accuracy of the detection system under complex working conditions.
[0004] In summary, under dynamic working conditions, the fault frequency components of the diesel generator set are easily submerged by random fluctuations, making it difficult to accurately separate and classify, resulting in insufficient accuracy of diesel generator set fault detection. SUMMARY
[0005] The present application provides a diesel generator set fault detection method and system based on deep learning to realize closed-loop diagnosis from signal acquisition to fault classification and improve the accuracy of fault detection.
[0006] In the first aspect, to solve the above technical problems, the present application provides a diesel generator set fault detection method based on deep learning, comprising: obtaining first vibration signal data during running of the diesel generator set, performing time-frequency decomposition on the first vibration signal data to obtain dynamic change characteristics; performing adaptive filtering on the dynamic change characteristics to obtain a denoised time-frequency feature sequence; extracting a peak energy distribution dataset from the time-frequency feature sequence, calculating the entropy value of each frequency band according to the peak energy distribution dataset to obtain a frequency band entropy value sequence; classify the frequency band entropy value sequence by a k-means clustering algorithm, determine a random fluctuation reference mode, separate abnormal frequency components from the peak energy distribution data set based on the random fluctuation reference mode; According to the abnormal frequency component, the line spacing and amplitude ratio are calculated, and the line feature data set is obtained, and the line feature data set is classified based on the preset fault mode library to determine the fault classification result; Obtain the current second vibration signal data, calculate the similarity between the second vibration signal data and the pre-established normal mode library, and if the similarity is lower than the preset similarity threshold, output the corresponding fault feature vector according to the fault classification result; According to the normal mode library, the random forest algorithm is used to verify the fault feature vector, and the final fault detection result is obtained.
[0007] Preferably, the time-frequency decomposition of the first vibration signal data is performed to obtain dynamic change characteristics, including: Analog-to-digital conversion processing is performed on the first vibration signal data to obtain a digital vibration signal; The digital vibration signal is decomposed in time and frequency by wavelet transform to generate a time-frequency spectrum containing time and frequency information; The time-frequency spectrum is segmented according to a preset time window, and the frequency components of each time window are calculated by Fourier transform to obtain dynamic change characteristics.
[0008] Preferably, the adaptive filtering is performed on the dynamic change characteristics to obtain a denoised time-frequency feature sequence, including: A wavelet denoising method is used for preliminary denoising processing on the dynamic change characteristics to obtain a denoised first feature data set; Discrete cosine transform is used for data compression on the first feature data set to obtain a denoised time-frequency feature sequence.
[0009] Preferably, the peak energy distribution data set is extracted from the time-frequency feature sequence, the entropy value of each frequency band is calculated according to the peak energy distribution data set, and the frequency band entropy value sequence is obtained, including: According to the time-frequency feature sequence, the time domain signal is converted into a frequency domain signal by using fast Fourier transform, the peak energy distribution characteristics of the signal at different frequencies are extracted, and the peak energy distribution data set is obtained; According to the peak energy distribution data set, the energy values of each frequency band are normalized to probability distribution, the Shannon entropy of each frequency band is calculated according to the probability distribution, the Shannon entropy is arranged in frequency band order, and the frequency band entropy value sequence is obtained.
[0010] Preferably, the classification of the frequency band entropy value sequence by the k-means clustering algorithm, determining a random fluctuation reference mode, separating abnormal frequency components from the peak energy distribution data set based on the random fluctuation reference mode, comprises: When the fluctuation amplitude of the frequency band entropy value sequence is higher than the preset fluctuation threshold, the k-means clustering algorithm is used to classify the frequency band entropy value sequence to obtain the random fluctuation reference mode. According to the random fluctuation reference mode, a normal energy threshold is determined, and a frequency band with peak energy exceeding the preset energy threshold is filtered out from the peak energy distribution data set to determine the abnormal frequency component.
[0011] Preferably, the spectrum line spacing and amplitude ratio are calculated according to the abnormal frequency component to obtain a spectrum line feature data set, and the spectrum line feature data set is classified based on a preset fault mode library to determine a fault classification result, comprising: An abnormal feature sequence is obtained from the abnormal frequency component, and the spectrum line spacing and amplitude ratio of the abnormal feature sequence are calculated by fast Fourier transform to obtain the spectrum line feature data set. The spectrum line feature data set is normalized to obtain a normalized feature vector set. When the dimension of the normalized feature vector set meets the preset dimension threshold, the normalized feature vector set is divided by a support vector machine to obtain a fault type label set. The fault type label set is matched with the fault mode library to obtain the fault classification result.
[0012] Preferably, the similarity between the second vibration signal data and the pre-established normal mode library is calculated, and if the similarity is lower than a preset similarity threshold, the corresponding fault feature vector is output according to the fault classification result, comprising: For the second vibration signal data, high-frequency noise higher than a preset frequency cutoff point is filtered out by Fourier transform, and low-frequency effective signals are retained, and the low-frequency effective signals are standardized to obtain a real-time signal feature set. The distance between the real-time signal feature set and each mode in the normal mode library is calculated and normalized to obtain a normalized distance set, and each distance value in the normalized distance set is subtracted by the value 1 to form a similarity set. If the proportion of the number of similarities lower than the preset similarity threshold in the similarity set exceeds the preset proportion threshold, the abnormal mode of the real-time signal feature set is judged according to the fault classification result, and the corresponding fault feature vector is extracted.
[0013] In a second aspect, the present application provides a diesel generator set fault detection system based on deep learning, comprising: The signal time-frequency decomposition module is configured to acquire first vibration signal data of the diesel generator set during operation, and perform time-frequency decomposition on the first vibration signal data to obtain dynamic change characteristics. The dynamic denoising processing module is configured to perform adaptive filtering on the dynamic change characteristics to obtain a denoised time-frequency characteristic sequence. The frequency band entropy value calculation module is configured to extract a peak energy distribution data set from the time-frequency characteristic sequence, calculate the entropy values of each frequency band according to the peak energy distribution data set, and obtain a frequency band entropy value sequence. The abnormal frequency separation module is configured to classify the frequency band entropy value sequence by using a k-means clustering algorithm, determine a random fluctuation reference mode, and separate abnormal frequency components from the peak energy distribution data set based on the random fluctuation reference mode. The fault classification determination module is configured to calculate the spectral line spacing and amplitude ratio according to the abnormal frequency components to obtain a spectral line characteristic data set, classify the spectral line characteristic data set based on a pre-established fault mode library, and determine a fault classification result. The fault feature output module is configured to acquire current second vibration signal data, calculate the similarity of the second vibration signal data and a pre-established normal mode library, and output a corresponding fault feature vector according to the fault classification result if the similarity is lower than a pre-established similarity threshold. The random forest verification module is configured to verify the fault feature vector by using a random forest algorithm according to the normal mode library to obtain a final fault detection result.
[0014] Compared with the prior art, the present application has the following beneficial effects: (1) The present application acquires first vibration signal data of the diesel generator set during operation and performs time-frequency decomposition to obtain dynamic change characteristics, and performs adaptive filtering on the dynamic change characteristics to obtain a denoised time-frequency characteristic sequence. This process can first capture the dynamic information of the vibration signal in the time and frequency dimensions, and then dynamically suppress environmental noise and load fluctuations and other interference factors through adaptive filtering, avoiding noise from masking key fault characteristics, so that the extracted vibration data is more consistent with the real operating state of the equipment, thereby improving the reliability of fault detection.
[0015] (2) The application extracts the peak energy distribution data set from the de-noised time-frequency characteristic sequence, calculates the entropy value of each frequency band according to the peak energy distribution data set to obtain a frequency band entropy value sequence, classifies the frequency band entropy value sequence through a k-means clustering algorithm to determine a random fluctuation reference mode, and separates abnormal frequency components based on the mode. This series of operations can quantify the energy distribution and randomness of each frequency band from the de-noised characteristics, establish a random fluctuation reference that fits the dynamic working condition through a clustering algorithm, accurately distinguish normal fluctuations and abnormal frequencies, and effectively capture the frequency components corresponding to early weak faults.
[0016] (3) The application calculates the spectral line spacing and amplitude ratio according to the abnormal frequency components to obtain a spectral line feature data set, classifies the spectral line feature data set based on a pre-set fault mode library to determine a fault classification result. The classification process can convert abstract abnormal frequency components into spectral line features with clear fault direction, accurately correspond to the fault type through matching with the fault mode library, avoid misjudgment caused by unclear feature and fault association, and make the fault classification result more targeted.
[0017] (4) The application obtains the current second vibration signal data, calculates the similarity with the pre-established normal mode library, and outputs a fault feature vector if the similarity is lower than a pre-set similarity threshold, and then verifies the fault feature vector using a random forest algorithm in combination with the normal mode library to obtain a final fault detection result. The reliability of the abnormal signal is further confirmed through similarity comparison, and the multi-dimensional verification capability of the random forest algorithm is used to comprehensively evaluate the fault feature vector, reducing the limitations of single feature judgment and ensuring that the final output fault detection result has both accuracy and reliability. BRIEF DESCRIPTION OF DRAWINGS
[0018] Figure 1 is a flow diagram of one embodiment of the diesel generator set fault detection method based on deep learning provided by the application; Figure 2 is a structural diagram of one embodiment of the diesel generator set fault detection system based on deep learning provided by the application. DETAILED DESCRIPTION
[0019] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.
[0020] REFERENCE Figure 1The first embodiment of the present application provides a diesel generator set fault detection method based on deep learning, including steps S11 to S17, specifically as follows: S11, acquiring first vibration signal data of the diesel generator set during operation, performing time-frequency decomposition on the first vibration signal data to obtain dynamic change characteristics; S12, performing adaptive filtering on the dynamic change characteristics to obtain a denoised time-frequency feature sequence; S13, extracting a peak energy distribution data set from the time-frequency feature sequence, calculating the entropy value of each frequency band according to the peak energy distribution data set to obtain a frequency band entropy value sequence; S14, classifying the frequency band entropy value sequence by a k-means clustering algorithm to determine a random fluctuation reference mode, and separating abnormal frequency components from the peak energy distribution data set based on the random fluctuation reference mode; S15, calculating the spectral line spacing and amplitude ratio according to the abnormal frequency components to obtain a spectral line feature data set, and classifying the spectral line feature data set based on a pre-set fault mode library to determine a fault classification result; S16, acquiring current second vibration signal data, calculating the similarity of the second vibration signal data and a pre-established normal mode library, and if the similarity is lower than a pre-set similarity threshold, outputting a corresponding fault feature vector according to the fault classification result; S17, verifying the fault feature vector by a random forest algorithm based on the normal mode library to obtain a final fault detection result.
[0021] In step S11, the first vibration signal data of the diesel generator set during operation is acquired, and the first vibration signal data is time-frequency decomposed to obtain dynamic change characteristics, including: Analog-digital conversion processing is performed on the first vibration signal data to obtain a digital vibration signal; Wavelet transform is used to perform time-frequency decomposition on the digital vibration signal to generate a time-frequency spectrum containing time and frequency information; The time-frequency spectrum is segmented according to a pre-set time window, the frequency components of each time window are calculated by Fourier transform, and the dynamic change characteristics are obtained.
[0022] It should be noted that the first vibration signal data is collected by an acceleration sensor installed at a key position of the diesel generating set (such as a bearing, a machine body, etc.), which can directly capture the vibration generated by the movement of mechanical parts during the operation of the diesel generating set. Under the conventional signal output standard of an industrial-grade sensor, the amplitude of the output analog vibration signal is generally in the range of -5V to 5V, which can cover the vibration signal intensity of the diesel generating set from normal operation to a slight fault state. An analog-to-digital converter (ADC) is a device that converts a continuously varying analog signal into a discrete digital signal, and its working process mainly includes three steps of sampling, quantization and coding. First, the analog vibration signal is sampled according to the sampling frequency to obtain a series of signal amplitudes at discrete time points; then these amplitudes are mapped to a limited number of discrete numerical levels; finally, the quantized numerical values are converted into binary codes to form a digital vibration signal. To ensure that the subtle features in the vibration signal are not lost, a 16-bit ADC with a resolution of 0.0001526V can be selected, which can divide the signal into 65536 levels and accurately depict the slight fluctuations of the vibration signal; it should be noted that this voltage value is linearly related to the vibration acceleration value measured by the sensor, and the signal amplitude mentioned in the subsequent processing refers to this equivalent voltage value for relative comparison and analysis.
[0023] In this embodiment, wavelet transform is a mathematical method that can analyze signals in both time and frequency domains. By convolving the signal with a series of wavelet basis functions of different scales and positions, multi-resolution analysis of the signal is achieved. In this invention, Daubechies wavelet basis is selected to decompose the digital vibration signal into wavelet components of different scales, and then generate a time-frequency spectrum containing time and frequency axes. The gray level or color depth at different positions in the time-frequency spectrum represents the signal energy intensity at the corresponding time point and corresponding frequency. For example, when the digital vibration signal of a 500kW diesel generating set is wavelet transformed, the generated time-frequency spectrum can cover a frequency range of 0-500Hz. If there is a significant high-energy concentration area near 100Hz in the time-frequency spectrum at a certain time period, it can be initially determined that the unit may have a rotor imbalance problem at that time period.
[0024] It is worth noting that the length of the preset time window needs to be determined in combination with the sensor sampling frequency and the time scale of the unit vibration characteristics. For example, with a sampling frequency of 1000 Hz, a 1-second time window can be set, containing 1000 sampling points, which can meet the data quantity requirements of Fourier transform, accurately reflect the short-term frequency changes of the vibration signal, and avoid the details of the frequency changes being covered up due to a too long time window. The basic principle of Fourier transform is to calculate the correlation of the signal with different frequencies of sine or cosine functions, and to decompose the complex time-domain signal into a sum of a series of simple harmonic components with different frequencies. For the signal segment divided by the time window, the Fast Fourier Transform (FFT) can efficiently calculate the amplitude or energy of each frequency component in the segment, i.e. the frequency component, thereby quantifying the main frequency composition of the vibration signal in the corresponding time window and the intensity of each frequency. By summarizing the frequency components of all time windows, the dynamic change characteristics are obtained, reflecting the frequency distribution changes of the vibration signal at different time points.
[0025] In step S12, adaptive filtering is performed on the dynamic change characteristics to obtain a denoised time-frequency feature sequence, including: A wavelet denoising method is used to perform preliminary denoising processing on the dynamic change characteristics to obtain a first feature data set after denoising. Discrete cosine transform is used to compress the data of the first feature data set to obtain a denoised time-frequency feature sequence.
[0026] It should be noted that the core of the wavelet denoising method is to separate the noise and effective signal components in the dynamic change characteristics by using the time-frequency localization ability of wavelet transform. First, a wavelet basis suitable for the vibration signal characteristics of the diesel generator set is selected, and the dynamic change characteristics are decomposed into a plurality of wavelet components with different scales. The high-frequency wavelet components correspond to environmental noise (such as airflow noise during equipment operation, external mechanical interference, etc.), and the low-frequency wavelet components correspond to effective signals generated by the vibration of mechanical parts of the unit. Then, the standard deviation of the signal is calculated σ to obtain a robust threshold , where N is the number of sampling points of the signal segment. The high-frequency wavelet components with a signal amplitude higher than the robust threshold are filtered out, and only the low-frequency components containing effective information are retained to obtain the first feature data set after denoising. Taking a 250 kW diesel generator set as an example, the vibration signal sampling frequency is 2000 Hz, and the time-frequency spectrum of the dynamic change characteristics covers the frequency range of 0-1000 Hz. In order to retain more details of non-stationary mechanical vibration signals, the Symlet wavelet basis is used to decompose the characteristics into 5 layers of wavelet components, and the 4th and 5th high-frequency components are filtered out to obtain the first feature data set, which can significantly reduce noise interference.
[0027] In one feasible approach, the least mean square algorithm from adaptive filtering is used for further denoising. First, the first feature dataset obtained through wavelet denoising is used as the input signal. x ( n Then, starting from the signal's initial position, take all sampling points within the first window, calculate the arithmetic mean of these sampling points, and use this average as the output value corresponding to the center position of the window; then move the window one sampling point along the time axis, repeating the operation of calculating the average, until the entire input signal has been traversed. x ( n This approximates the pure effective vibration signal, which is then used as the desired signal. d ( n Then initialize the filter weight vector. w ( n Each vector element is randomly selected within a small range of -0.01 to 0.01 to avoid initial output distortion, and a convergence factor μ is set with a value of 0.001 to 0.01 to balance convergence speed and stability. Furthermore, the filter is based on the current weight vector w ( n For input signals x ( n Perform weighted calculations to obtain real-time output. y ( n )= w ( n )∙ x ( n (∙ denotes the vector dot product), calculate the error signal e ( n )= d ( n )- y ( n ), representing the deviation between the desired effective signal and the filter output; finally, through iterative formulas... Iteratively adjust the weight vector until the error... e ( n An amplitude less than 0.02V indicates that residual noise has been effectively suppressed, and the output second feature dataset mainly retains the effective vibration characteristics of the unit itself. n =0, 1,…,N-1, where N is the number of sampling points. For example, when the input signal... x ( n The 50Hz noise component in the signal significantly leads to e ( n When the value increases, the weights will be corrected in the direction of enhancing the attenuation of the 50Hz frequency; if the effective signal is excessively suppressed, the error feedback will cause the weights to be adjusted in the opposite direction to preserve the features.
[0028] It is worth mentioning that the second feature data set after denoising contains a large number of data points, and directly used for subsequent analysis will increase the computing load and data storage, while the discrete cosine transform (DCT) can greatly compress the data amount on the premise of preserving key features. Specifically, the discrete cosine transform can convert the time domain signal of the second feature data set into the cosine coefficient in the frequency domain, and the calculation formula of the frequency domain coefficient sequence y ( n ) obtained after DCT transformation of the adaptive filtering output Y ( k ) is as follows: k , wherein, is a normalization coefficient, when k =0, ; when k ≥1, , is a transform kernel function, representing a cosine basis function of different frequencies, k The smaller the value, the lower the corresponding frequency. The vibration signal energy has a concentration characteristic, and the key features of the effective vibration of the unit, such as the 80Hz main frequency of the bearing rotation, the 160Hz multiple frequency of the rotor imbalance, etc. mostly correspond to low frequency coefficients, and the energy proportion is high; while the high frequency coefficients are mostly noise residues or redundant information, and the energy is very low. By retaining the first N low frequency coefficients whose energy proportion exceeds 95%, and discarding the redundant high frequency coefficients, data compression can be realized.
[0029] Exemplarily, for the second feature data set of 10000 data points obtained after adaptive filtering of a 250kW diesel generator set, 10000 cosine coefficients can be generated by discrete cosine transform, wherein the first 200 low frequency coefficients have contained more than 95% of the energy of the original signal, which is sufficient to represent the core vibration characteristics of the unit. At this time, only by retaining these 200 coefficients, the data amount can be compressed to 2% of the original.
[0030] In step S13, a peak energy distribution data set is extracted from the time-frequency feature sequence, and the entropy values of each frequency band are calculated according to the peak energy distribution data set to obtain a frequency band entropy value sequence, including: According to the time-frequency feature sequence, the time domain signal is converted into a frequency domain signal by using fast Fourier transform, the peak energy distribution characteristics of the signal at different frequencies are extracted, and the peak energy distribution data set is obtained; According to the peak energy distribution data set, the energy values of each frequency band are normalized into a probability distribution, the Shannon entropy of each frequency band is calculated according to the probability distribution, the Shannon entropy is arranged in the order of frequency band, and the frequency band entropy value sequence is obtained.
[0031] It should be noted that for the time-frequency characteristic sequence, the sequence is first padded with zeros to the nearest power of 2, denoted as Nn; then, it is split into two subsequences of length Nn / 2 (even-indexed part, low-frequency trend part) and odd-indexed subsequence (high-frequency fluctuation part); the discrete Fourier transform results are calculated for each subsequence, and then combined using a butterfly operation to obtain the final frequency domain signal. This frequency domain signal contains Nn / 2 effective frequency points, and the square of the amplitude of each frequency point is the energy value of the corresponding frequency. The average and standard deviation of the energy values of all frequency points in the frequency domain signal are calculated, and the average value is summed with three times the standard deviation to obtain the peak threshold. Points with energy values greater than the peak threshold are identified as peak points, and the corresponding frequencies and energy values are extracted as peak features. The resulting peak energy distribution dataset can intuitively reflect the concentration of unit vibration energy at different frequencies.
[0032] Furthermore, the frequency range is divided into several continuous, equally wide frequency bands. The bandwidth is determined based on the structural parameters of the diesel generator set (such as bearing type, gear teeth number, and rotor speed) to calculate the characteristic frequencies of its key components (such as fundamental frequency, harmonics, and bearing passing frequency). Using these as the center, non-uniform frequency bands covering these characteristic frequencies are set for energy statistics. Selectable bandwidths include 10Hz, 20Hz, 50Hz, and 100Hz. For small generator sets below 50kW, the vibration characteristic frequency is usually concentrated between 0-500Hz, so the 20Hz and 50Hz bandwidths can be emphasized for testing. For large generator sets above 500kW, the vibration characteristic frequency can extend to 0-2000Hz, requiring the inclusion of wider candidate bandwidths such as 100Hz. Next, the sum of all peak energies within each frequency band is calculated to obtain the energy value of each band. Dividing the energy value of each band by the total energy value of all bands yields the proportion of that band's energy to the total energy, which is the probability distribution. These proportions together form a probability distribution sequence, whose sum is 1.
[0033] In this embodiment, Shannon entropy H The calculation formula is ,in m For the number of frequency bands, For the first i The probability distribution of each frequency band. Shannon entropy is used to quantify the randomness and complexity of energy distribution in each frequency band. The higher the entropy value, the more uniform the energy distribution in each frequency band, and the stronger the signal randomness. Arranging the Shannon entropy values of each frequency band in order of frequency frequency yields a sequence of frequency band entropy values, which reflects the complexity of vibration energy distribution in different frequency ranges.
[0034] Exemplarily, for the time-frequency feature sequence of a 250kW diesel generator set, an FFT process is performed on a sampling frequency of 2000Hz and a length of 2048 data points to obtain a frequency domain signal of 0-1000Hz, peak energy at frequencies of 80Hz, 160Hz, 320Hz, etc. is extracted to form a peak energy distribution data set; 0-1000Hz is divided into 10 frequency bands of 100Hz wide, and the energy proportion of the first frequency band (0-100Hz) is 0.4, the energy proportion of the second frequency band (100-200Hz) is 0.3, and the energy proportions of the remaining frequency bands are 0.1, 0.05, 0.05, 0.03, 0.02, 0.02, 0.02 and 0.01 in turn. After calculating the spectral entropy of each frequency band, the frequency band entropy value sequence is obtained by arranging the frequency bands in order. The frequency bands with low entropy values in the sequence correspond to the energy concentration area, which reflects the main vibration characteristics of the unit in normal operation.
[0035] In step S14, the frequency band entropy value sequence is classified by a k-means clustering algorithm to determine a random fluctuation reference mode, and based on the random fluctuation reference mode, an abnormal frequency component is separated from the peak energy distribution data set, including: When the fluctuation amplitude of the frequency band entropy value sequence is higher than a preset fluctuation threshold, the frequency band entropy value sequence is classified by a k-means clustering algorithm to obtain the random fluctuation reference mode; According to the random fluctuation reference mode, a normal energy threshold is determined, and a frequency band with peak energy exceeding a preset energy threshold is filtered out from the peak energy distribution data set to determine the abnormal frequency component.
[0036] It should be noted that the fluctuation amplitude is quantified by the standard deviation of the frequency band entropy value sequence, and the preset fluctuation threshold is determined according to historical normal operation data. By analyzing a large number of standard deviation sets of frequency band entropy values under normal working conditions, the average value and standard deviation of the set are calculated, and the average value plus 2 times the standard deviation is set as the preset fluctuation threshold. The preset fluctuation threshold of a conventional unit can be set to 0.5. When the standard deviation of the entropy value sequence calculated in real time exceeds the threshold, it reflects that the dynamic characteristics of the vibration system have changed abnormally, triggering the k-means clustering analysis.
[0037] It is worth mentioning that the k-means clustering algorithm belongs to an unsupervised learning model, which is used to cluster the frequency band entropy value sequence samples with similar characteristics into a class. The training data is the frequency band entropy value sequence when clustering is triggered and the historical entropy value sequence in the past 10 minutes, each sequence contains the entropy values of m frequency bands, forming an m-dimensional feature vector. The determination of k value needs to combine the common operating state division of the unit (such as stable operation, load switching transition, start-up stage, etc.), and the value range of k is preliminarily set to 3-5. Then, the clustering error (the sum of the squared distances of all samples to their cluster centers) under different k values is calculated, the curve of the error changing with k value is drawn, and the k value corresponding to the inflection point of the curve where the slope changes from steep to slow is taken as the optimal clustering number.
[0038] During training, first, k feature vectors are randomly selected from the training data as initial cluster centers; then the Euclidean distance between each sample and each cluster center is calculated, and the sample is assigned to the nearest cluster, and the mean of each cluster is recalculated as the new cluster center; finally, the assignment and update steps are repeated until the change in cluster center is less than the preset convergence threshold. Through this process, the frequency band entropy value sequence is divided into k classes, and the class with the highest sample number ratio and the smallest entropy value fluctuation is determined as the random fluctuation reference mode, which represents the typical energy distribution fluctuation characteristics of the unit under the current working condition, providing a reference benchmark for judging abnormalities.
[0039] Further, the peak energy distribution data corresponding to the random fluctuation reference mode is extracted, the historical maximum value or a certain high quantile (such as the 99th quantile) of the energy at each frequency point is calculated, and this value is set as the normal energy threshold. The frequency band that exceeds the corresponding frequency normal energy threshold is considered to deviate from the typical range of the reference mode, and is determined as an abnormal frequency component, corresponding to the early failure (such as the characteristic frequency generated by bearing wear) or sudden interference (such as impact vibration caused by component loosening) of the diesel generator set.
[0040] Exemplarily, for the frequency band entropy value sequence of a certain diesel generator set, the standard deviation is calculated to be 0.6, which is higher than the preset fluctuation threshold 0.5, triggering k-means clustering analysis. Select 200 entropy value sequences of the last 10 minutes as training data, each sequence containing the entropy values of 10 frequency bands, set k=3, and after 15 iterations, 3 clusters are obtained: cluster 1 accounts for 60%, the entropy value fluctuation is small (standard deviation 0.2), corresponding to the stable operation mode; cluster 2 accounts for 30%, the entropy value fluctuation is moderate; cluster 3 accounts for 10%, the entropy value fluctuation is large. Cluster 1 is determined as the random fluctuation reference mode, and in the corresponding peak energy distribution, the average energy of 80Hz frequency is 0.02V², and the standard deviation is 0.003V², and the normal energy threshold is calculated to be 0.029V². From the real-time peak energy distribution data set, it is found that the peak energy of 80Hz frequency is 0.035V², which exceeds the normal energy threshold, so 80Hz is identified as an abnormal frequency component, indicating that there may be rotor imbalance or bearing early wear problems.
[0041] In step S15, according to the abnormal frequency component, the spectral line spacing and amplitude ratio are calculated to obtain a spectral line feature data set, and the spectral line feature data set is classified based on a preset fault mode library to determine a fault classification result, including: An abnormal feature sequence is obtained from the abnormal frequency component, and the spectral line spacing and amplitude ratio of the abnormal feature sequence are calculated by fast Fourier transform to obtain the spectral line feature data set; The spectral line feature data set is normalized to obtain a normalized feature vector set; When the dimension of the normalized feature vector set meets a preset dimension threshold, the normalized feature vector set is divided by a support vector machine to obtain a fault type label set; The fault type label set is matched with the fault mode library to obtain a fault classification result.
[0042] It should be noted that the abnormal feature sequence refers to a time domain vibration signal segment of the frequency band where the abnormal frequency is located, which contains the vibration details that cause energy abnormalities. The abnormal feature sequence is processed by fast Fourier transform to obtain the corresponding frequency domain spectrum line. The frequency domain spectrum line refers to a spectrum line set composed of discrete frequency points and their amplitudes corresponding to the energy peak value presented in the frequency domain, and each spectrum line represents the vibration energy characteristics of a specific frequency, where the abscissa is the frequency and the ordinate is the energy amplitude corresponding to the frequency. The spectral line spacing refers to the frequency difference between adjacent abnormal frequency peaks, for example, the spectral line spacing between 100Hz and 200Hz is 100Hz, which is related to the parameters such as the rotating speed and the number of teeth of the mechanical part. The amplitude ratio refers to the ratio of the amplitudes of the larger peak value to the smaller peak value between two abnormal frequency peaks, which is used to represent the relative intensity of each abnormal component. The spectral line feature data set is integrated in the order of "spectral line spacing + amplitude ratio".
[0043] Further, in order to eliminate the dimensional differences between different features, the spectral feature dataset is normalized to map the data uniformly to the 0-1 interval. The normalization method adopts min-max standardization, and the formula is wherein, x is the original feature value, is the normalized feature value, and are the maximum and minimum values of the feature in the historical data, respectively. The normalized feature vector set obtained after standardization is composed of multiple normalized spectral line spacings and amplitude ratios. The preset dimension threshold can be determined through verification experiments. In the historical data set containing multiple fault types, the classification accuracy of the support vector machine under different dimension thresholds is tested, and the dimension interval that makes the accuracy reach the peak and is stable is selected as the threshold standard. At the same time, it should also be avoided that the model complexity rises due to too high dimension, which can be set to 8-10 in the diesel generator set fault diagnosis scene. The higher the dimension of the normalized feature vector set, the more abnormal feature information it contains.
[0044] It is worth noting that the support vector machine (SVM) belongs to a supervised learning model, which maps low-dimensional features to high-dimensional space through a kernel function, constructs an optimal hyperplane to realize linear separation of different categories, and the commonly used kernel function is the radial basis function (RBF), which can effectively handle nonlinear feature relationships. The training data are normalized spectral feature vectors in historical fault cases, and each vector corresponds to a known fault type label (such as rotor imbalance, bearing wear, etc.). The training data are input into the SVM, the search range of the preset adjustment penalty parameter C is [0.1, 1, 10, 100], and the order of magnitude covers from low penalty to high penalty; the search range of the kernel function parameter γ is [0.01, 0.1, 1, 10], corresponding to the kernel function width from narrow to wide; then for each (C, γ) parameter combination, the 5-fold cross-validation calculation method is used, the training data are divided into 5 parts, and 4 parts are used for training and 1 part is used for validation, the cross-validation average accuracy is obtained, the parameter combination with the highest cross-validation accuracy is finally selected, the classification accuracy of the model on the training set is the highest, and the optimal hyperplane is finally determined.
[0045] Finally, the input normalized feature vector set is divided by the hyperplane, and the fault type label corresponding to each vector is output to form a fault type label set. Collect the fault data verified by disassembly in historical fault cases, and execute the processes of S11 to S14 for each fault type, such as bearing outer ring wear, rotor imbalance, and blade crack, to extract the abnormal frequency components in the vibration signal, and record the spectral line spacing, amplitude ratio, and amplitude change trend of these frequencies. Similarly, these features are min-max normalized to form a structured typical spectral line feature template, each template corresponds to a feature vector, and finally stored according to fault type classification to form a fault mode library. Match the fault type label set with the pre-set fault mode library, calculate the cosine value of the angle between the two vectors to get the cosine similarity, and take the fault type corresponding to the highest cosine similarity as the fault classification result.
[0046] Exemplarily, a certain diesel generator set is detected in operation with abnormal frequency components of 60Hz, 120Hz and 240Hz, and the corresponding abnormal feature sequence is extracted. The sampling frequency is set to 1600Hz, and the time domain sampling point number is 2048. After fast Fourier transform processing, three main frequency domain spectral lines are obtained: 60Hz (amplitude 0.4V), 120Hz (amplitude 0.25V), and 240Hz (amplitude 0.1V). The spectral line spacing of 120Hz and 60Hz is 60Hz, and the amplitude ratio is 1.6; the spectral line spacing of 240Hz and 120Hz is 120Hz, and the amplitude ratio is 2.5; the amplitude ratio of 60Hz and 240Hz is 4. According to the order of "spectral line spacing + amplitude ratio", the spectral line feature data set is obtained as [60Hz, 120Hz, 1.6, 2.5, 4]. The maximum value of the spectral line spacing in the known historical data is 200Hz, and the minimum value is 50Hz. The maximum value of the amplitude ratio is 5, and the minimum value is 1. Therefore, the normalized feature vector set is [0.07, 0.47, 0.15, 0.38, 0.75], the dimension is 5, and the preset dimension threshold is 8. Therefore, the weak peak value near 300Hz and other frequencies is included to supplement more adjacent spectral line features, and the extended vector dimension reaches 9, meeting the threshold requirement.
[0047] Next, the extended normalized feature vector is input into the trained SVM model, and the fault type label "1" is output, corresponding to "rotor imbalance". The feature vector corresponding to the label is matched with the fault mode library, and the typical template of rotor imbalance in the library is "the spectral line spacing is an integer multiple of the fundamental frequency (such as 60Hz, 120Hz, 240Hz, etc.), and the amplitude ratio increases with the frequency". The calculated cosine similarity is 0.93, which is the highest value obtained by matching with all templates, so the fault classification result is determined to be rotor imbalance.
[0048] In step S16, the current second vibration signal data is acquired, the similarity of the second vibration signal data and the pre-established normal mode library is calculated, and if the similarity is lower than a preset similarity threshold, a corresponding fault feature vector is output according to the fault classification result, including: For the second vibration signal data, high-frequency noise higher than a preset frequency cutoff point is filtered out by Fourier transform, low-frequency effective signals are retained, and the low-frequency effective signals are normalized to obtain a real-time signal feature set; The distance of the real-time signal feature set and each mode in the normal mode library is calculated and normalized to obtain a normalized distance set, each distance value in the normalized distance set is subtracted by the value 1 to form a similarity set; If the number of similarities lower than the preset similarity threshold accounts for more than a preset proportion threshold in the similarity set, the abnormal mode of the real-time signal feature set is judged according to the fault classification result, and a corresponding fault feature vector is extracted.
[0049] It should be noted that the second vibration data set is a real-time collected unit vibration signal, which is used for dynamic monitoring of the current operation state of the unit. The preset frequency cutoff point needs to be set in combination with the characteristic frequency range of the key components of the unit. The effective vibration signal of the diesel generator set is mainly concentrated in 0-1000Hz, and the high-frequency components above 1000Hz are mainly environmental interference (such as electromagnetic noise, airflow noise, etc.). Therefore, the frequency cutoff point can be set to 1000Hz, and only the low-frequency effective signal of 0-1000Hz is retained. The z-score standardization method is adopted to subtract the mean value of the historical normal signal amplitude from the amplitude of the low-frequency effective signal, and then divide by the standard deviation of the historical normal signal amplitude to obtain the normalized low-frequency effective signal. The real-time signal feature set is obtained by combining the time sequence.
[0050] In this embodiment, the normal mode library is constructed based on the multi-condition historical data of the unit in the healthy state. First, the vibration signals of the unit under different normal conditions (such as 20% load, 50% load, 80% load, etc.) are collected, each condition is collected for at least 24 hours to ensure complete coverage of the running period; then the original signal is sliced according to a fixed time window (the same as the time window in step S11) to form an initial signal segment set; the frequency band entropy sequence of each data is calculated, and then the z-score standardization is used to eliminate the amplitude magnitude difference of the initial signal segment set. Subsequently, the core features of each segment are extracted, including the main frequency components (such as 50Hz fundamental frequency, 100Hz harmonic, etc.), the amplitude variation trend of the characteristic frequency with time, and the statistical characteristics of the signal energy distribution (such as variance, peak factor, etc.). These features are combined into a feature vector in time sequence, and all feature vectors are clustered into a plurality of cluster centers by K-means clustering algorithm. These cluster centers are the modes in the normal mode library, forming the normal mode library.
[0051] It should be noted that when calculating the distance between the real-time signal feature set and each mode in the normal mode library, the same feature vectors (e.g., root mean square, kurtosis, spectral peak, etc.) as the modes in the normal mode library are extracted from the real-time signal feature set, and the Euclidean distance between the real-time feature vectors and the feature vectors in the mode library is calculated. The smaller the distance value, the more similar the two sequences. By dividing by the maximum possible distance, the global distance is normalized to the 0-1 interval to obtain a normalized distance set. Subtracting each normalized distance value from 1 gives a similarity set ranging from 0 to 1, with 1 indicating complete similarity and 0 indicating complete dissimilarity.
[0052] It should be noted that the preset similarity threshold is used to determine whether the real-time signal deviates from the normal mode, and its value needs to be determined according to the similarity distribution of historical normal data. By statistically analyzing the similarity between a large number of real-time signals during normal operation and the normal mode library, it is found that more than 95% of the similarity is greater than 0.8, so the preset similarity threshold can be set to 0.8. The preset proportion threshold is derived from statistical analysis of the historical normal operation data of the diesel generator set. The vibration signal data of the unit under different loads and different speeds in typical normal operating conditions is selected, the similarity of each condition signal with all modes in the normal mode library is calculated, and the number of similarity less than the similarity threshold under normal state is counted. The proportion distribution shows that in more than 95% of the normal conditions, the proportion of the number of similarity less than the similarity threshold is less than 20%, so the preset proportion threshold can be set to 20%.
[0053] Finally, the fault classification result obtained in step S15 is combined to determine the abnormal mode related to the fault (e.g., a specific frequency amplitude anomaly related to rotor imbalance) in the real-time signal feature set, and the abnormal frequency, amplitude, duration, and other characteristic parameters corresponding to the abnormal mode are extracted to form a fault feature vector.
[0054] Exemplarily, the second vibration signal data of a certain diesel generator set is subjected to Fourier transform, high-frequency noise above 1000 Hz is filtered out, low-frequency signals of 0-1000 Hz are retained, and a real-time signal feature set is obtained through z-score standardization. The normal mode library contains three typical modes: low load mode (load 20%), medium load mode (load 50%) and high load mode (load 80%). By calculating the global distance of the real-time feature set and each mode, the maximum possible distance is 1, and the normalized distance set is [0.3, 0.4, 0.5], and the similarity set is [0.7, 0.6, 0.5]. Since the preset similarity threshold is 0.8, and all similarities are lower than the threshold, combined with the fault classification result “rotor imbalance”, it is judged that the amplitude of 60Hz and 120Hz frequencies in the real-time signal abnormally increases by 30% compared with the normal mode, and the amplitude, duration and other parameters of these frequencies are extracted to form a fault feature vector [60Hz, 0.35V, 120Hz, 0.22V, duration 10min], and the vector is output to guide further maintenance; it should be noted that the fault feature vector can be combined according to actual needs Select one or more of the abnormal frequency, amplitude, duration and other characteristics, or further normalized and then input into the random forest model.
[0055] In step S17, according to the normal mode library, the random forest algorithm is used to verify the fault feature vector, and the final fault detection result is obtained.
[0056] It should be noted that the random forest belongs to a supervised learning model, which is composed of multiple independent decision trees, each tree is constructed based on a random subset of training data and a random subset of features, and the final output result is determined through majority voting, which can improve the accuracy and robustness of fault judgment and avoid misjudgment of a single model. The normal mode library serves as a reference benchmark in this process, providing the distribution range of abnormal frequencies, corresponding amplitudes, duration and other characteristics under normal working conditions, wherein the abnormal frequency points to the possible fault component (such as bearing fault which corresponds to a specific high frequency), the amplitude reflects the severity of the fault (the larger the amplitude, the more serious the fault), and the duration reflects the stability of the fault, which can be used to assist the decision tree to determine whether the feature is abnormal.
[0057] In this embodiment, the training data of the random forest is the feature vector corresponding to the historical fault case and normal operation data, each vector is labeled with "fault" or "normal" label, and the specific fault type. First, randomly sample with replacement from the training data to generate multiple sample sets with the same amount of original data; for each sample set, a fixed number of partial features are randomly selected, such as 5 from 10 features, to build a decision tree; each node of the decision tree is split by the feature with the maximum information gain, which measures the reduction of sample set classification uncertainty, the calculation formula is the information entropy before splitting minus the information entropy after splitting, the smaller the information entropy, the more explicit the sample classification, such as using "60Hz amplitude whether more than 0.3V" as the splitting condition; repeat the above process to generate 100 decision trees to form a random forest.
[0058] It should be noted that when the number of decision trees increases to a certain extent, the precision improvement rate will slow down significantly, at this time, continuing to increase the number of trees will increase the calculation cost (including memory occupation and training time) without obvious benefits. In the diesel generator set fault diagnosis scene, 100 decision trees are the reasonable value obtained by testing the validation accuracy of the model under different tree numbers, which can offset the overfitting risk of a single tree through the independent judgment of multiple trees, and can control the calculation time consumption of real-time detection. After training, each tree makes an independent judgment on the input fault feature vector and outputs the classification result, and finally the judgment result of the majority of trees is taken as the verification result.
[0059] Exemplarily, a certain fault feature vector is [60Hz, 0.35V, 120Hz, 0.22V, 10min], which is input into the random forest composed of 100 decision trees. Among them, 85 trees determine "rotor imbalance", 10 trees determine "normal fluctuation", and 5 trees determine "bearing wear". Among them, the probability of "rotor imbalance" is the highest, which is 85%, and the result is consistent with the typical characteristics of rotor imbalance in the normal mode library (60Hz and its harmonic amplitude rise, long duration), so the final fault detection result is confirmed as "rotor imbalance", which is consistent with the actual disassembly inspection result.
[0060] In summary, the present application discloses a diesel generator set fault detection method based on deep learning, comprising obtaining first vibration signal data during diesel generator set operation, time-frequency decomposition of the first vibration signal data to obtain dynamic change characteristics, adaptive filtering of the dynamic change characteristics to obtain denoised time-frequency characteristic sequence, extraction of peak energy distribution dataset from the time-frequency characteristic sequence, calculation of entropy values of each frequency band based on the peak energy distribution dataset to obtain a frequency band entropy value sequence, classification of the frequency band entropy value sequence by k-means clustering algorithm to determine a random fluctuation reference mode, separation of abnormal frequency components from the peak energy distribution dataset based on the random fluctuation reference mode, calculation of spectral line spacing and amplitude ratio based on the abnormal frequency components to obtain a spectral line feature dataset, classification of the spectral line feature dataset based on a pre-set fault mode library to determine a fault classification result, obtaining current second vibration signal data, calculating the similarity of the second vibration signal data and a pre-established normal mode library, and if the similarity is lower than a pre-set similarity threshold, outputting a corresponding fault feature vector based on the fault classification result, and verifying the fault feature vector by a random forest algorithm based on the normal mode library to obtain a final fault detection result. The present application fuses time-frequency analysis technology and pattern matching method to construct a multi-level fault detection system, realizing efficient capture of abnormal features in diesel generator set vibration signals and accurate identification of fault types.
[0061] Referring to Figure 2 The second embodiment of the present application provides a diesel generator set fault detection system based on deep learning, comprising: A signal time-frequency decomposition module is configured to obtain first vibration signal data during diesel generator set operation, and perform time-frequency decomposition on the first vibration signal data to obtain dynamic change characteristics. A dynamic denoising processing module is configured to perform adaptive filtering on the dynamic change characteristics to obtain a denoised time-frequency characteristic sequence. A frequency band entropy value calculation module is configured to extract a peak energy distribution dataset from the time-frequency characteristic sequence, calculate entropy values of each frequency band based on the peak energy distribution dataset, and obtain a frequency band entropy value sequence. An abnormal frequency separation module is configured to classify the frequency band entropy value sequence by a k-means clustering algorithm, determine a random fluctuation reference mode, and separate abnormal frequency components from the peak energy distribution dataset based on the random fluctuation reference mode. A fault classification determination module is configured to calculate spectral line spacing and amplitude ratio based on the abnormal frequency components to obtain a spectral line feature dataset, classify the spectral line feature dataset based on a pre-set fault mode library, and determine a fault classification result. The fault feature output module is configured to acquire current second vibration signal data, calculate a similarity between the second vibration signal data and a pre-established normal mode library, and output a corresponding fault feature vector according to the fault classification result if the similarity is lower than a preset similarity threshold. The random forest verification module is configured to verify the fault feature vector by using a random forest algorithm according to the normal mode library to obtain a final fault detection result.
[0062] It should be noted that the diesel generator set fault detection system based on deep learning provided by the embodiment is used to execute all process steps of the diesel generator set fault detection method based on deep learning provided by the embodiment, and the working principles and beneficial effects of the two are one-to-one correspondence, thus not being repeated.
[0063] The embodiment of the present application further provides an electronic device. The electronic device comprises a processor, a memory and a computer program, such as an adaptive filtering program, stored in the memory and executable on the processor. The processor implements the steps in the above various diesel generator set fault detection methods based on deep learning when executing the computer program, such as Figure 1 The steps S11 are shown. Alternatively, the processor implements the functions of each module / unit in the above various device embodiments when executing the computer program, such as the signal time-frequency decomposition module.
[0064] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the electronic device.
[0065] The electronic device can be a desktop computer, a notebook, a palm computer and a smart tablet, etc. The electronic device can include, but is not limited to, a processor, a memory. Those skilled in the art can understand that the above components are only examples of the electronic device and do not constitute a limitation on the electronic device, and can include more or fewer components than the above, or combine certain components, or different components, for example, the electronic device can also include an input / output device, a network access device, a bus, etc.
[0066] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The processor is a control center of the electronic device, and connects various parts of the electronic device through various interfaces and lines.
[0067] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the electronic device by running or executing the computer program and / or modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, an image playing function, etc.), etc.; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a nonvolatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory device.
[0068] The modules / units integrated in the electronic device, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can implement the steps of each method embodiment when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium can include any entity or device, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. that can carry the computer program code. It should be noted that the contents included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.
[0069] It should be noted that the above-described device embodiments are only schematic, and the units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. In addition, the connection relationship between the modules in the device embodiment provided by the present application indicates that there is a communication connection between them, which can be realized as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.
[0070] The above-described specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above-described specific embodiments are only for the specific embodiments of the present application and are not used to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application should be included in the protection scope of the present application.
Claims
1. A fault detection method for diesel generator sets based on deep learning, characterized in that, include: The first vibration signal data of the diesel generator set during operation is acquired, and the first vibration signal data is decomposed into time and frequency to obtain dynamic change characteristics. Adaptive filtering is performed on the aforementioned dynamic change characteristics to obtain a denoised time-frequency feature sequence; The peak energy distribution dataset is extracted from the time-frequency feature sequence. Based on the peak energy distribution dataset, the entropy value of each frequency band is calculated to obtain the frequency band entropy value sequence. The frequency band entropy sequence is classified by k-means clustering algorithm to determine the random fluctuation baseline pattern. Based on the random fluctuation baseline pattern, the abnormal frequency components are separated from the peak energy distribution dataset. Based on the abnormal frequency components, the spectral line spacing and amplitude ratio are calculated to obtain a spectral line feature dataset. The spectral line feature dataset is then classified based on a preset fault mode library to determine the fault classification result. Acquire the current second vibration signal data, calculate the similarity between the second vibration signal data and the pre-established normal pattern library, and if the similarity is lower than the preset similarity threshold, output the corresponding fault feature vector according to the fault classification result. Based on the normal pattern library, the fault feature vector is verified using the random forest algorithm to obtain the final fault detection result.
2. The deep learning-based fault detection method for diesel generator sets according to claim 1, characterized in that, The step of performing time-frequency decomposition on the first vibration signal data to obtain dynamic change characteristics includes: The first vibration signal data is subjected to analog-to-digital conversion to obtain a digital vibration signal; Wavelet transform is used to perform time-frequency decomposition on the digital vibration signal to generate a time-spectrum diagram containing time and frequency information. The time-spectrum diagram is divided into preset time windows, and the frequency components of each time window are calculated by Fourier transform to obtain the dynamic change characteristics.
3. The deep learning-based diesel generator set fault detection method according to claim 1, characterized in that, The adaptive filtering performed on the dynamically changing features to obtain the denoised time-frequency feature sequence includes: To address the aforementioned dynamic change characteristics, a wavelet denoising method is used for preliminary denoising to obtain the first feature dataset after denoising. For the first feature dataset, discrete cosine transform is used to compress the data to obtain a denoised time-frequency feature sequence.
4. The deep learning-based diesel generator set fault detection method according to claim 1, characterized in that, The step of extracting the peak energy distribution dataset from the time-frequency feature sequence, and calculating the entropy value of each frequency band based on the peak energy distribution dataset to obtain the frequency band entropy value sequence includes: Based on the time-frequency feature sequence, the time-domain signal is converted into a frequency-domain signal using a fast Fourier transform, and the peak energy distribution features of the signal at different frequencies are extracted to obtain the peak energy distribution dataset. Based on the peak energy distribution dataset, the energy values of each frequency band are normalized into a probability distribution. The Shannon entropy of each frequency band is calculated based on the probability distribution. The Shannon entropy is then arranged in order of frequency band to obtain a sequence of frequency band entropy values.
5. The deep learning-based diesel generator set fault detection method according to claim 4, characterized in that, The step of classifying the frequency band entropy sequence using the k-means clustering algorithm to determine a random fluctuation baseline pattern, and then separating abnormal frequency components from the peak energy distribution dataset based on the random fluctuation baseline pattern, includes: When the fluctuation amplitude of the frequency band entropy value sequence is higher than the preset fluctuation threshold, the frequency band entropy value sequence is classified by the k-means clustering algorithm to obtain the random fluctuation benchmark pattern; Based on the random fluctuation baseline pattern, a normal energy threshold is determined, and frequency bands whose peak energy exceeds the preset energy threshold are selected from the peak energy distribution dataset and identified as abnormal frequency components.
6. The deep learning-based diesel generator set fault detection method according to claim 1, characterized in that, The step involves calculating the spectral line spacing and amplitude ratio based on the abnormal frequency components to obtain a spectral line feature dataset. This dataset is then classified based on a pre-defined fault mode library to determine the fault classification result, including: An abnormal feature sequence is obtained from the abnormal frequency components, and the spectral line spacing and amplitude ratio of the abnormal feature sequence are calculated by fast Fourier transform to obtain the spectral line feature dataset. The spectral line feature dataset is normalized to obtain a normalized feature vector set; When the dimension of the normalized feature vector set meets the preset dimension threshold, the normalized feature vector set is divided by a support vector machine to obtain a fault type label set. The fault type label set is matched with the fault pattern library to obtain the fault classification result.
7. The deep learning-based diesel generator set fault detection method according to claim 1, characterized in that, The similarity between the second vibration signal data and the pre-established normal pattern library is calculated. If the similarity is lower than a preset similarity threshold, the corresponding fault feature vector is output based on the fault classification result, including: For the second vibration signal data, high-frequency noise above the preset frequency cutoff point is filtered out by Fourier transform, and low-frequency effective signals are retained. The low-frequency effective signals are then standardized to obtain a real-time signal feature set. Calculate and normalize the distance between the real-time signal feature set and each pattern in the normal pattern library to obtain a normalized distance set. Subtract each distance value in the normalized distance set from the value 1 to form a similarity set. If the proportion of similarity values below a preset similarity threshold in the similarity set exceeds a preset ratio threshold, then based on the fault classification result, the abnormal mode of the real-time signal feature set is determined, and the corresponding fault feature vector is extracted.
8. A fault detection system for diesel generator sets based on deep learning, characterized in that, include: The signal time-frequency decomposition module is used to acquire the first vibration signal data during the operation of the diesel generator set, and to perform time-frequency decomposition on the first vibration signal data to obtain dynamic change characteristics. The dynamic denoising module is used to perform adaptive filtering on the dynamically changing features to obtain a denoised time-frequency feature sequence. The frequency band entropy calculation module is used to extract the peak energy distribution dataset from the time-frequency feature sequence, and calculate the entropy value of each frequency band based on the peak energy distribution dataset to obtain the frequency band entropy value sequence. An abnormal frequency separation module is used to classify the frequency band entropy value sequence using a k-means clustering algorithm, determine a random fluctuation baseline pattern, and separate abnormal frequency components from the peak energy distribution dataset based on the random fluctuation baseline pattern. The fault classification determination module is used to calculate the spectral line spacing and amplitude ratio based on the abnormal frequency components to obtain a spectral line feature dataset, classify the spectral line feature dataset based on a preset fault mode library, and determine the fault classification result. The fault feature output module is used to acquire the current second vibration signal data, calculate the similarity between the second vibration signal data and the pre-established normal pattern library, and if the similarity is lower than the preset similarity threshold, output the corresponding fault feature vector according to the fault classification result. The random forest verification module is used to verify the fault feature vectors based on the normal pattern library using the random forest algorithm to obtain the final fault detection result.
Citation Information
Patent Citations
Diesel engine fault identification method and device
CN116006322A
Belt conveyor inspection robot audio key frame extraction method based on frequency band entropy and pixel significance
CN117727325A
Fault classification detection method and device based on transformer vibration characteristics and time-frequency diagram
CN119202980A
A switch cabinet discharge fault detection method, system, device and medium
CN119780640A
Equipment supervision data real-time analysis method and system
CN120180249A
Cited By
Method and system for controlling cutting speed of aluminum plate cutting device
CN121500874A
Fault detection method and system for generator iron core lamination equipment
CN121580260A
A method and system for detecting faults in a generator core lamination press
CN121580260B
Device and method for monitoring running state of synchronous motor
CN121656835A
GIS circuit breaker spring fault monitoring method based on MEMS sensor
CN121678174A
