Reduction gearbox fault monitoring and early warning method and system based on voiceprint features

Through the monitoring method based on voiceprint features, fast independent component analysis and packet wave decomposition technology, combined with BP neural network and KNN algorithm, the problems of insufficient recognition capability of gearbox fault monitoring and low noise monitoring accuracy in the existing technology are solved. Early warning and efficient monitoring of gearbox failures are achieved.

CN118913688APending Publication Date: 2024-11-08李慧

Patent Information

Application Number
CN202410997948.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-24
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The existing gearbox fault monitoring technology has problems such as insufficient identification capabilities, inability to detect faults in time, risk of damage to the equipment, and low noise monitoring accuracy.

Method used

Using a monitoring method based on voiceprint features, the whole-domain sound signals are collected through bone conduction voiceprint sensors, and the signal separation and feature extraction are performed using fast independent component analysis and improved packet wave decomposition technology. Combined with BP neural network and KNN algorithm, the gearbox is realized. Early warning of failure.

Benefits of technology

It improves the accuracy and reliability of gearbox fault monitoring, can issue early warnings in a timely manner, avoid equipment downtime, and ensure production and operation efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118913688A_ABST
    Figure CN118913688A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of reduction gearbox fault monitoring and early warning, and provides a reduction gearbox fault monitoring and early warning method and system based on voiceprint characteristics, and the method comprises the steps: firstly collecting a global sound signal of a reduction gearbox which operates healthily under a normal working condition, taking the global sound signal as sample set signal data, and then carrying out the blind source separation algorithm based on rapid independent component analysis, s3, separating mixed sound signals of the sample set, extracting features of source sound signals of the sample set based on an improved packet wave decomposition technology, calculating sample entropy for frequency band signals obtained through decomposition in the step S3, and constructing voiceprint feature vectors of the reduction gearbox in a normal health state; according to the invention, 0-20KHz frequency response sounds of different point positions of the reduction gearbox are collected through the bone conduction voiceprint sensor, and full-frequency-domain sound signal data are obtained; through separation of mixed sound signals, abnormal sound interference is reduced, feature extraction and self-learning are carried out on the separated sound signals, and finally early warning of early fault features of the reduction gearbox is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of gearbox fault monitoring and early warning, and specifically relates to a gearbox fault monitoring and early warning method and system based on voiceprint features. Background Art

[0002] Currently, the technical methods for monitoring deceleration faults on the market include reduction gearbox oil analysis, vibration monitoring analysis, temperature monitoring analysis, and noise monitoring analysis.

[0003] Oil analysis can only identify a small number of faults such as gearbox wear and lubricant performance changes. Offline oil analysis can only be done periodically and cannot detect faults in a timely manner. Online oil monitoring and analysis requires drilling holes in the equipment for installation, which poses a certain risk of damage to the gearbox itself.

[0004] When the speed of the gearbox is low, that is, the shaft speed is basically less than 30 rpm, it is difficult to capture useful information using vibration monitoring technology, mainly because the energy of the impact signal is too low. In some industrial scenarios, the entire transmission mechanism is complex (the gearbox is only one of the devices), and there are too many interference signals, which affects the collection and use of vibration signals.

[0005] Temperature monitoring and analysis is applicable to serious faults such as broken teeth that have occurred inside the gearbox, resulting in severe wear and tear and rising equipment temperature. It belongs to late-stage fault monitoring and has low monitoring value.

[0006] Noise monitoring and analysis is applicable to late stage gearbox failures. Due to the limited audible sound frequency range and the large overlap between the gearbox noise and the fault sound band, even after using noise removal technology, it is still unable to meet the accuracy requirements for fault noise feature extraction.

[0007] The above technical methods all have more or less usage restrictions when used, which leads to poor results.

[0008] To this end, those skilled in the art have proposed a reduction gearbox fault monitoring and early warning method and system based on voiceprint features to solve the problems raised by the background technology. Summary of the invention

[0009] In order to solve the above technical problems, the present invention provides a gearbox fault monitoring and early warning method and system based on voiceprint features. Through this method, the gearbox working failure can be discovered earlier, and the technicians can effectively determine whether there are any abnormalities in the operation of the gearbox, thereby realizing the prediction and diagnosis of the gearbox failure with high accuracy and reliability.

[0010] In order to achieve the above object, the present invention provides a speed reducer fault monitoring and early warning method based on voiceprint features, comprising the following steps:

[0011] S1. First, collect the global sound signal of the gearbox (normal sample gearbox) under normal working conditions and running healthily, and use it as the sample set signal data (normal group data), and then use the bone conduction voiceprint sensor to collect the sound signal;

[0012] S2. Then, based on the blind source separation algorithm of fast independent component analysis (ICA), the mixed sound signal of the sample set is separated to extract the original independent sound signal, and then the source sound signal, i.e., the sound signal of the device itself, is solved and the noise signal is marked;

[0013] S3, then extract the characteristics of the sample set source sound signal based on the improved small packet wave decomposition technology, and then use the BP neural network model to improve it to obtain the frequency band signal;

[0014] S4, then calculating the sample entropy of the frequency band signal decomposed in step S3, constructing the voiceprint feature vector of the reducer in a normal and healthy state, and using it as the input vector of the KNN algorithm, calculating the K value in the KNN algorithm, and forming a classification label for the sample data;

[0015] S5, then using the method in step S1, collect the global sound signal of the reduction gearbox in actual production work (whether the reduction gearbox is faulty is unknown) as the test set sound signal data;

[0016] S6, then use the method in step S2 to separate and filter the test set sound signal data;

[0017] S7, then use the method of step S3 to decompose and extract features from the test set sound signal data;

[0018] S8. Use the sound signal feature data of the test set as the input instance of the KNN algorithm, select the K sample data instances closest to the input instance as the nearest neighbors, and then use the voting method to determine the category to which the input instance belongs; if the category to which the input instance belongs is within the classification of the sample data, output the result, the sound signal feature collected by the gearbox is normal, and the gearbox operates normally without fault; if the category to which the input instance belongs is not within the classification of the sample data, output the result, the sound signal feature collected by the gearbox is abnormal, and the gearbox has hidden faults.

[0019] Preferably, in step S1, the range of the collected sound is 0-20KHz.

[0020] Preferably, in step S3, the sample set source sound signal is based on the physical characteristics of the reduction gearbox, and its sound frequency is mainly distributed within 4000 Hz.

[0021] Preferably, in step S3, the feature extraction of the sample set source sound signal based on the improved small packet decomposition technology is specifically: using packet decomposition, if the decomposition level is too small, such as 3 layers, the energy is mainly distributed in the first and second frequency bands, and the sound is difficult to recognize; if the decomposition level is too large, such as 9 layers, 512 frequency bands will be formed, increasing the difficulty of feature recognition.

[0022] Preferably, in step S3, the steps of using the BP neural network model for improvement are as follows:

[0023] S31, firstly, decompose the sound into 3 layers of wavelet packets, and input the 8 frequency band feature vectors into the BP neural network model;

[0024] S32, then identifying the energy distribution and identifying the frequency band where the energy distribution is concentrated;

[0025] S33, then continue to decompose the frequency band where the energy is concentrated in the 4th layer, and input the decomposed frequency band feature vector into the BP neural network model again for secondary recognition, and cyclically decompose to the 9th layer;

[0026] S34, finally output the decomposed frequency bands.

[0027] It provides a new fault monitoring method for industrial gearbox equipment operation and maintenance technicians. Through this method, pre-processing warnings can be issued in time to avoid equipment shutdown caused by gearbox failure during the production process of the enterprise, thereby reducing production operation efficiency and ensuring production levels.

[0028] A speed reducer fault monitoring and early warning system based on voiceprint features, the fault monitoring and early warning system comprising:

[0029] Data acquisition module: responsible for collecting sound data during the operation of the reduction gearbox;

[0030] Acoustic signal processing module: pre-processes and extracts features of the collected sound data, and extracts the soundprint features of the reducer in working state;

[0031] Fault detection module: performs real-time analysis and comparison of sound data to detect whether there is abnormal sound caused by a gearbox fault;

[0032] Early warning system module: When the system detects a gearbox failure, it will sound an alarm and display detailed fault information, notifying the equipment operator to handle it in time;

[0033] Historical data recording module: records historical sound data for fault diagnosis and analysis, helping to determine the cause of the fault or prevent the occurrence of the fault in advance;

[0034] The data acquisition module, the acoustic signal processing module, the fault detection module, the early warning system module and the historical data recording module are electrically connected in sequence.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] The present invention collects 0Hz-20KHz frequency response sounds at different points of the reduction gearbox through a bone conduction voiceprint sensor to obtain full-frequency domain sound signal data; by separating the mixed sound signal, abnormal sound interference is reduced, and features are extracted from the separated sound signal, and self-learning is performed, ultimately achieving early warning of the reduction gearbox's early fault features. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flow chart of the speed reducer fault monitoring and early warning method based on voiceprint features of the present invention;

[0038] Figure 2 This is a framework diagram of the gearbox fault monitoring and early warning system based on voiceprint features of the present invention;

[0039] Figure 3 It is a blind source separation flow chart of the present invention. DETAILED DESCRIPTION

[0040] The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0041] like Figure 1 As shown:

[0042] Embodiment 1: The present invention provides a speed reducer fault monitoring and early warning method based on voiceprint features, comprising the following steps:

[0043] S1. First, collect the global sound signal of the reducer (normal sample reducer) under normal working conditions and running healthily, where the range of the collected sound is 0-20KHz, and use it as the sample set signal data (normal group data), and then use the bone conduction voiceprint sensor to collect the sound signal;

[0044] S2. Then, based on the blind source separation algorithm of fast independent component analysis (ICA), the mixed sound signal of the sample set is separated to extract the original independent sound signal, and then the source sound signal, i.e., the sound signal of the device itself, is solved and the noise signal is marked;

[0045] S3, then extracting the features of the sample set source sound signal based on the improved small packet wave decomposition technology, the sample set source sound signal is based on the physical characteristics of the reduction box, and its sound frequency is mainly distributed within 4000Hz, and then using the BP neural network model to improve it to obtain the frequency band signal;

[0046] S4, then calculating the sample entropy of the frequency band signal decomposed in step S3, constructing the voiceprint feature vector of the reducer in a normal and healthy state, and using it as the input vector of the KNN algorithm, calculating the K value in the KNN algorithm, and forming a classification label for the sample data;

[0047] S5, then using the method in step S1, collect the global sound signal of the reduction gearbox in actual production work (whether the reduction gearbox is faulty is unknown) as the test set sound signal data;

[0048] S6, then use the method in step S2 to separate and filter the test set sound signal data;

[0049] S7, then use the method of step S3 to decompose and extract features from the test set sound signal data;

[0050] S8. Use the sound signal feature data of the test set as the input instance of the KNN algorithm, select the K sample data instances closest to the input instance as the nearest neighbors, and then use the voting method to determine the category to which the input instance belongs; if the category to which the input instance belongs is within the classification of the sample data, output the result, the sound signal feature collected by the gearbox is normal, and the gearbox operates normally without fault; if the category to which the input instance belongs is not within the classification of the sample data, output the result, the sound signal feature collected by the gearbox is abnormal, and the gearbox has hidden faults.

[0051] From the above, it can be seen that the present invention collects the 0Hz-20KHz frequency response sound at different points of the gearbox through the bone conduction voiceprint sensor to obtain full-frequency domain sound signal data; reduces abnormal sound interference by separating the mixed sound signal; extracts features from the separated sound signal, performs self-learning, and finally realizes early warning of the gearbox early fault characteristics.

[0052] Furthermore, the monitored sound is mixed, including the sound of the gearbox itself, as well as the noise of the surrounding environment, such as the sound of people passing by, wind, the sound of collisions and collisions with the gearbox, the sound of on-site operation instruction horns, etc.; the core of the blind source separation algorithm is that it attempts to restore the original, independent sound signal; the central limit theorem: the distribution of the sum of two independent random variables is closer to the Gaussian distribution than any of the original variables; so the premise or assumption of using this algorithm is "if there is noise, the noise signal and the source signal are independent of each other, and the Gaussian distribution signal is the noise signal."

[0053] Specifically, based on the blind source separation algorithm of fast independent component analysis (ICA), the method steps for separating the mixed sound signal of the sample set are as follows:

[0054] (1) Steps of ICA separation

[0055] Step 1: Return the mean to zero: subtract the mean of each component of the signal to be processed;

[0056] Step 2: Data whitening: Find the autocorrelation function of the signal and perform whitening on it;

[0057] Step 3: Select an arbitrary orthogonal matrix as the initial value;

[0058] Step 4: Update each column of W using the iterative formula

[0059] Step 5: Orthogonalize the matrix W;

[0060] Step 6: If it does not converge (W changes a lot after iteration), return to step 4, such as Figure 3 As shown;

[0061] The specific algorithm is as follows:

[0062] 1) Variable meaning:

[0063] X: indicates the observed signal; W: indicates the weight vector; S: indicates the independent component;

[0064] 2) Algorithm steps

[0065] Step 1: Remove the mean: Calculate the mean μ for each observation vector x, and then center all observation vectors, that is, x′=x-μ; this step ensures that the mean of the data is zero, preparing the data for subsequent whitening and independent component extraction;

[0066] Step 2: Whitening: Calculate the covariance matrix of the centered data C = E[x′x′ T ]; Perform eigenvalue decomposition C = EDE on the covariance matrix T , where D is the eigenvalue diagonal matrix and E is the corresponding eigenvector matrix;

[0067] Transform the data using eigenvalues ​​and eigenvectors to obtain whitened data;

[0068] Step 3: Find the maximum non-Gaussian direction:

[0069] Initialization: Select a random unit vector W as the initial value of the weight vector;

[0070] Fixed-point iterative update: Apply the following update rule to the weight vector W until convergence: + =E[x”g(w T x”)]-E[g'(w T x”)]w Orthogonalization: If multiple independent components are extracted, the new weight vectors need to be orthogonalized to ensure that they are independent of each other;

[0071] Step 4: Calculate independent components:

[0072] Use the final weight vector W to project the whitened data onto the independent components, that is, s = w T x″ represents the source signal;

[0073] Step 5: Repeat the extraction of other components

[0074] Repeat steps 3 and 4 to extract more independent components, and each time ensure that the new weight vector is orthogonal (or independent) to the previously extracted independent components;

[0075] (2) Quantitative judgment of Gaussianity

[0076] Gaussian signal refers to a signal whose probability distribution follows Gaussian distribution (normal distribution). In practical applications, the sound of the gearbox itself naturally presents a non-Gaussian distribution, and noise often conforms to Gaussian distribution. Using this feature, the mixed signal can be separated into its independent source signals. The methods for measuring or judging non-Gaussian include kurtosis and negentropy. The kurtosis method is used in the present invention.

[0077] kurt(y)=E(y 4 )-3[E(y 2 )] 2

[0078]

[0079] When the value is 0, it is a noise signal, and when it is not 0, it is a sound signal from the reduction gearbox;

[0080] Specifically, in step S3, the feature extraction of the sample set source sound signal based on the improved small packet decomposition technology is as follows: using packet decomposition, if the decomposition level is too small, such as 3 layers, the energy is mainly distributed in the first and second frequency bands, and the sound is difficult to recognize; if the decomposition level is too large, such as 9 layers, 512 frequency bands will be formed, increasing the difficulty of feature recognition.

[0081] From the above, it can be seen that by adopting the improved small packet wave decomposition technology, features can be extracted from the source sound signal to train the fault detection model, thereby accurately detecting the gearbox fault.

[0082] Specifically, in step S3, the steps of using the BP neural network model for improvement are as follows:

[0083] S31, firstly, decompose the sound into 3 layers of wavelet packets, and input the 8 frequency band feature vectors into the BP neural network model;

[0084] S32, then identifying the energy distribution and identifying the frequency band where the energy distribution is concentrated;

[0085] S33, then continue to decompose the frequency band where the energy is concentrated in the 4th layer, and input the decomposed frequency band feature vector into the BP neural network model again for secondary recognition, and cyclically decompose to the 9th layer;

[0086] S34, finally output the decomposed frequency bands.

[0087] From the above, it can be seen that by adopting the BP neural network model, more refined feature recognition results can be obtained, thereby improving the accuracy and reliability of gearbox fault detection.

[0088] Specifically, wavelet packet decomposition: The function of wavelet packet decomposition is to decompose sound information into high-frequency and low-frequency signals; wavelet packet decomposition is a commonly used method in signal processing, which is used to decompose signals into approximate signals and detail signals with different frequency components; its principle is to divide the signal into different frequency bands, the signal frequency range in each frequency band is different, and then perform wavelet decomposition on each frequency band;

[0089] Wavelet packet decomposition is an extended form of wavelet analysis, which can decompose signals more finely so as to better understand and process signals. Its main steps include selecting wavelet basis functions and the number of frequency bands, calculating wavelet packet coefficients, and performing threshold processing and reconstruction on the coefficients.

[0090] Specifically, the BP neural network model is used in the process of extracting the characteristics of the sample set source sound signal using the small packet wave decomposition technology, and includes the following steps:

[0091] 1. Data preprocessing

[0092] 1) Framing: split the voiceprint signal into short frames (e.g., 20-30 milliseconds per frame);

[0093] 2) Windowing: Apply a window function (such as a Hamming window) to each frame to reduce spectral leakage;

[0094] 3) FFT (Fast Fourier Transform): Perform FFT on the windowed frame to obtain the spectrum;

[0095] 2. Feature extraction

[0096] 1) Calculate the spectrum energy: square the modulus of the FFT result to obtain the spectrum energy distribution;

[0097] 2) Select frequency bands: divide the spectrum energy distribution into multiple frequency bands (for example, using the Mel frequency scale) and calculate the sum of the energy in each frequency band;

[0098] 3) Normalization: Normalize the energy of each frequency band so that the sum of the energy of all frequency bands is 1;

[0099] 3. Construct BP neural network model

[0100] 1) Input layer: the number of nodes is equal to the number of frequency bands;

[0101] 2) Hidden layer: One or more layers can be set as needed, and the number of nodes in each layer can be determined based on experience or experiments.

[0102] 3) Output layer: The number of nodes is usually set to the number of categories (i.e., the number of frequency band categories); for the task of identifying concentrated frequency bands, one output node may be required to indicate whether the frequency band is an energy concentrated frequency band, or multiple output nodes may be used to indicate the possibility of multiple frequency bands;

[0103] 4. Training BP neural network

[0104] 1) Forward propagation: For a given input, the output is calculated through the neural network; the formula is as follows:

[0105] z (l) =W (l) a (l-1 )+b (l)

[0106] a (l) =f(z (l) )

[0107] Where l is the number of layers, W(l) and b(l) are the weights and biases of the lth layer, f is the activation function (such as Sigmoid, ReLU, etc.), and a(l) is the output of the lth layer (i.e., the input of the next layer);

[0108] 2) Calculate the loss: Use a loss function (such as mean square error, cross entropy loss, etc.) to calculate the difference between the predicted output and the actual output;

[0109] 3) Back propagation: Calculate the gradient through the chain rule and use the gradient descent algorithm to update the weights and biases; the formula is as follows:

[0110]

[0111] δ (l) =((W (l+1) ) T δ (l+1) )☉f′( z(l) )

[0112]

[0113] Where C is the loss function, δ(l) is the error term of the lth layer, and ⊙ represents the element-wise product;

[0114] 5. Evaluation and prediction

[0115] 1) Use the test set to evaluate the performance of the model;

[0116] 2) For new audio input, extract features according to the above steps and input them into the model to obtain the prediction results;

[0117] 6. Determine the frequency band where the sound energy distribution is concentrated

[0118] 1) According to the output of the model, find the frequency band predicted to be the energy concentrated frequency band;

[0119] 2) It is possible to determine which frequency bands are energy concentrated by setting a threshold (for example, if the output value is greater than a certain threshold, it is considered to be an energy concentrated frequency band).

[0120] Specifically, KNN algorithm: KNN algorithm is used to classify the sound signal and determine the classification of the sound signal under the normal operation of the gearbox; the sound signal data under the normal operation of the gearbox is used as the training set data, and the steps are as follows:

[0121] Step 1: Collect data: Collect labeled training data sets, including features and corresponding classification or regression results;

[0122] Step 2: Calculate the distance: For a new input instance, calculate the distance between it and each instance in the training data set; usually use metrics such as Euclidean distance or Manhattan distance;

[0123] Step 3. Choose the K value: Determine how many nearest neighbors need to be considered to make a decision; a smaller K value will lead to a higher model complexity and prone to overfitting; a larger K value will lead to a lower model complexity and prone to underfitting, which includes:

[0124] (1) Selecting a range of K values: Select a range of K values, for example, starting from 1 and gradually increasing, or select a range that includes several potential optimal values;

[0125] (2) Cross-validation: For each value of K, use cross-validation (such as K-fold cross-validation) to evaluate the performance of the model;

[0126] (3) Divide the training set into K subsets (in K-fold cross-validation, K is different from K in the KNN algorithm); for each value of K, perform the following iterations:

[0127] Select K-1 subsets as the current training set;

[0128] The remaining subset is used as a validation set;

[0129] Use the current K value and training set to train the KNN model;

[0130] Use the validation set to evaluate the performance of the model and record performance metrics (such as accuracy);

[0131] After all iterations are completed, calculate the average value or other statistics of the performance indicator;

[0132] Select the optimal K value: Based on the results of cross-validation, select a K value with the best performance indicator (such as the highest accuracy) as the optimal K value;

[0133] Final evaluation: Use the selected optimal K value and the entire training set to train the final KNN model, and use the test set to evaluate the performance of the model;

[0134] Step 4: Find K nearest neighbors: Select the K training data instances that are closest to the input instance as the nearest neighbors;

[0135] Step 5: Make a decision: Make a decision based on the classification or regression results of the nearest neighbors. For classification problems, a voting method is used to determine the category to which the input instance belongs. For regression problems, the average of the prediction results of the K nearest neighbors is used as the prediction result of the input instance.

[0136] Step 6: Output results: Output the category to which the input instance belongs or the prediction result.

[0137] like Figure 2 As shown:

[0138] Embodiment 2: A gearbox fault monitoring and early warning system based on voiceprint features, the fault monitoring and early warning system comprising:

[0139] Data acquisition module: responsible for collecting sound data during the operation of the reduction gearbox;

[0140] Acoustic signal processing module: pre-processes and extracts features of the collected sound data, and extracts the soundprint features of the reducer in working state;

[0141] Fault detection module: performs real-time analysis and comparison of sound data to detect whether there is abnormal sound caused by a gearbox fault;

[0142] Early warning system module: When the system detects a gearbox failure, it will sound an alarm and display detailed fault information, notifying the equipment operator to handle it in time;

[0143] Historical data recording module: records historical sound data for fault diagnosis and analysis, helping to determine the cause of the fault or prevent the occurrence of the fault in advance;

[0144] The data acquisition module, the acoustic signal processing module, the fault detection module, the early warning system module and the historical data recording module are electrically connected in sequence.

[0145] From the above, it can be seen that the gearbox fault monitoring and early warning system based on voiceprint features uses sound sensors to collect sound data when the gearbox is running, and extracts soundprint features through acoustic signal processing technology; the system will analyze the sound data in real time to monitor whether the gearbox has abnormal sounds; when the system detects that the gearbox makes abnormal sounds, it will sound an alarm and display detailed fault information so that the equipment operator can deal with it in time; in addition, the system can also record historical sound data for fault diagnosis and analysis; through this gearbox fault monitoring and early warning system based on voiceprint features, gearbox faults can be discovered in time, equipment damage and accidents can be reduced, and operating safety and the service life of the gearbox can be improved.

[0146] The embodiments of the present invention are provided for the purpose of illustration and description. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations of the present invention. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A speed reducer fault monitoring and early warning method based on voiceprint features, characterized in that: The following steps are involved: S1. First, collect the global sound signal of the reducer under normal working conditions and running healthily, and use it as the sample set signal data, and then use the bone conduction voiceprint sensor to collect the sound signal; S2. Then, based on the blind source separation algorithm of fast independent component analysis, the mixed sound signal of the sample set is separated to extract the original independent sound signal, and then the source sound signal, i.e., the sound signal of the device itself, is solved and the noise signal is marked; S3, then extract the characteristics of the sample set source sound signal based on the improved small packet wave decomposition technology, and then use the BP neural network model to improve it to obtain the frequency band signal; S4, then calculating the sample entropy of the frequency band signal decomposed in step S3, constructing the voiceprint feature vector of the reducer in a normal and healthy state, and using it as the input vector of the KNN algorithm, calculating the K value in the KNN algorithm, and forming a classification label for the sample data; S5, then using the method in step S1 to collect the global sound signal of the reduction gearbox in actual production work as the test set sound signal data; S6, then use the method of step S2 to separate and filter the test set sound signal data; S7, then use the method of step S3 to decompose and extract features from the test set sound signal data; S8, using the test set sound signal feature data as an input instance of the KNN algorithm, selecting K sample data instances closest to the input instance as the nearest neighbors, and then using a voting method to determine the category to which the input instance belongs; If the category to which the input instance belongs is within the classification of the sample data, the output result is that the characteristics of the sound signal collected by the reduction gearbox are normal, and the reduction gearbox is operating normally without fault; If the category to which the input instance belongs is not within the classification of the sample data, the output result is that the sound signal characteristics collected by the gearbox are abnormal and the gearbox has a potential fault.

2. A method for monitoring and early warning of reduction gearbox faults based on voiceprint features as claimed in claim 1, characterized in that: In step S1, the range of the collected sound is 0-20KHz.

3. A method for monitoring and early warning of reduction gearbox faults based on voiceprint features as claimed in claim 1, characterized in that: In step S3, the sample set source sound signal is based on the physical characteristics of the reduction gearbox, and its sound frequency is mainly distributed within 4000 Hz.

4. A speed reducer fault monitoring and early warning method based on voiceprint features as claimed in claim 1, characterized in that: In step S3, the feature extraction of the sample set source sound signal based on the improved small packet decomposition technology is specifically: using packet decomposition, if the decomposition level is too small, such as 3 layers, the energy is mainly distributed in the first and second frequency bands, and the sound is difficult to recognize; if the decomposition level is too large, such as 9 layers, 512 frequency bands will be formed, increasing the difficulty of feature recognition.

5. A method for monitoring and early warning of reduction gearbox faults based on voiceprint features as claimed in claim 1, characterized in that: In step S3, the steps of using the BP neural network model for improvement are as follows: S31, firstly, decompose the sound into 3 layers of wavelet packets, and input the 8 frequency band feature vectors into the BP neural network model; S32, then identifying the energy distribution and identifying the frequency band where the energy distribution is concentrated; S33, then continue to decompose the frequency band where the energy is concentrated in the 4th layer, and input the decomposed frequency band feature vector into the BP neural network model again for secondary recognition, and cyclically decompose to the 9th layer; S34, finally outputting the decomposed frequency bands.

6. A speed reducer fault monitoring and early warning system based on voiceprint features according to any one of claims 1 to 5, characterized in that: The fault monitoring and early warning system comprises: Data acquisition module: responsible for collecting sound data during the operation of the reduction gearbox; Acoustic signal processing module: pre-processes and extracts features of the collected sound data, and extracts the soundprint features of the reducer in working state; Fault detection module: performs real-time analysis and comparison of sound data to detect whether there is abnormal sound caused by a gearbox fault; Early warning system module: When the system detects a gearbox failure, it will sound an alarm and display detailed fault information, notifying the equipment operator to handle it in time; Historical data recording module: records historical sound data for fault diagnosis and analysis, helping to determine the cause of the fault or prevent the occurrence of the fault in advance; The data acquisition module, the acoustic signal processing module, the fault detection module, the early warning system module and the historical data recording module are electrically connected in sequence.

Citation Information

Patent Citations

  • Rolling bearing fault detection method based on improved wavelet neural network

    CN107122802A

  • Transformer sound anomaly detection method based on improved wavelet packet and deep learning

    CN111259921A

  • Power equipment health state detection system and method based on voiceprint features

    CN113283310A

  • Rolling bearing fault detection method

    CN114964777A

  • Transformer fault sound feature recognition method based on neural network

    CN115438685A

Cited By

  • Pump machine health monitoring method based on voiceprint characteristics

    CN119900706A

  • Photovoltaic box transformer substation abnormal operation state diagnosis and fault location method based on sound signals

    CN121415805A

  • A voiceprint-based rotating machinery fault diagnosis method

    CN122921583A