A centrifugal pump acoustic signal fault diagnosis method and system

Abstract

The present invention provides a method for diagnosing centrifugal pump acoustic signal faults, comprising: obtaining acoustic signals of a centrifugal pump under different operating conditions; extracting Mel-frequency cepstral coefficients of the acoustic signals as the initial feature matrix of the acoustic signals; obtaining a Mel-frequency cepstral scatter entropy feature matrix as a sample set using a scatter entropy algorithm; training a pre-trained support vector machine classifier using the sample set, and optimizing the parameters of the pre-trained support vector machine classifier using a bat algorithm during the training process to obtain an acoustic signal fault diagnosis model; and diagnosing centrifugal pump faults using the trained acoustic signal fault diagnosis model. The present invention incorporates scatter entropy into the Mel-frequency cepstral coefficient feature extraction process to deeply mine the signal, thereby improving the accuracy of centrifugal pump fault diagnosis based on acoustic signals. At the same time, the bat optimization algorithm is used to optimize the kernel function and penalty factor of the classifier, thereby determining various fault conditions of the centrifugal pump.

Description

Technical Field

[0001] The present invention belongs to the technical field of centrifugal pump fault diagnosis classification, and in particular relates to a centrifugal pump acoustic signal fault diagnosis method and system. Background Art

[0002] Centrifugal pumps are widely used in the petrochemical industry due to their wide range of applications and simple structure. Failures in centrifugal pumps can impact production efficiency at the very least, or even lead to serious accidents, endangering life and property. Early fault diagnosis of centrifugal pumps can identify and warn of faulty equipment in advance, allowing for repairs or replacements before a failure occurs. Compared to using centrifugal pump vibration signals for fault diagnosis, using centrifugal pump acoustic radiation signals for condition monitoring can avoid the problem of vibration acceleration sensors being unable to be installed due to abnormal environments such as high temperatures and corrosion. Fault diagnosis methods based on acoustic signals have important application value in the petrochemical industry.

[0003] Numerous studies have been conducted on centrifugal pump fault diagnosis by scholars both domestically and internationally. Zhou Yunlong et al., based on multi-point noise measurements, employed singular value decomposition (SVD) denoising and second-generation wavelet packet decomposition to extract the noise energy spectrum at each point, trained a BP neural network, and performed early-stage cavitation fault diagnosis for centrifugal pumps. However, given the interference of large amounts of distributed noise and abnormal structural vibration, SVD was unable to effectively extract hydrodynamic noise, resulting in numerous misjudgments at the boundary between normal and early cavitation conditions. Araste et al. used the electrical characteristics of centrifugal pumps as a basis for fault diagnosis of motor-driven centrifugal pumps, inferring the health of the centrifugal pump using motor voltage and current signals, thus reducing the need for acceleration sensors. However, fault diagnosis methods based on electrical signals were unable to analyze the noise and vibration generated during cavitation, resulting in poor cavitation diagnosis. Chen et al. introduced the concept of the machine learning k-nearest neighbor (KNN) algorithm into traditional Mahalanobis distance fault diagnosis and proposed an improved KNN centrifugal pump fault prediction model based on Mahalanobis distance. This method can distinguish specific centrifugal pump fault types based on vibration signals, but its vibration-based approach has its limitations.

[0004] Based on acoustic signals, Mel-frequency cepstral coefficients have been gradually applied to acoustic signal fault diagnosis of various equipment. Wang Qian et al. used Mel-frequency cepstral coefficients to extract the features of the noise signal of rolling bearings, and reduced the dimension of the extracted features using the Compensated Distance Evaluation (CDET). The features after dimensionality reduction can accurately and effectively identify the type of bearing faults. However, the kernel function and penalty factor of the support vector machine were not optimized, and the artificial setting of parameters reduced the generalization performance of the model. (Wang Qian, Wang Gang, Jiang Hanhan, et al. Research on rolling bearing fault diagnosis method based on MFCC and CDET [J]. Control Engineering, 2019 (9): 5.) Yan et al. proposed a diesel engine acoustic fault diagnosis method based on variational mode decomposition mapping Mel-frequency cepstral coefficients and long short-term memory network. However, the feature extraction method of variational mode decomposition mapping Mel-frequency cepstral coefficients still has confusion between some faults, and this method has only been verified in the field of diesel engine acoustic fault diagnosis, and the model migration capability needs to be studied. Di Xiaodong et al. improved MFCC by linearly superimposing the feature parameters of traditional MFCC with those of GFCC to obtain a hybrid parameter MGCC, which was used for transformer fault diagnosis. However, this method has poor fault signal extraction capabilities in noisy environments and cannot maintain a high recognition rate in low signal-to-noise ratio environments. Summary of the Invention

[0005] In view of the shortcomings of the above-mentioned prior art, the present invention extracts Mel-frequency cepstral coefficients from acoustic signals collected on-site as the initial features of the signal. The scatter entropy of these initial Mel-frequency cepstral features is then calculated. The penalty coefficient and kernel function parameters of a support vector machine are optimized using a bat optimization algorithm to obtain an acoustic signal fault diagnosis model. This trained acoustic signal fault diagnosis model is then used to diagnose centrifugal pump faults. This overcomes the difficulty of installing vibration sensors, while enabling on-site, non-contact measurement for centrifugal pump fault diagnosis in high-temperature, corrosive environments, thereby increasing the frequency richness of the information source for centrifugal pump fault diagnosis.

[0006] To achieve the above-mentioned purpose and other related purposes, the present invention provides a centrifugal pump acoustic signal fault diagnosis method, which includes obtaining the acoustic signal of the centrifugal pump under different working conditions; extracting the Mel-cepstral coefficients of the acoustic signal as the initial feature matrix of the acoustic signal; calculating the scatter entropy of each frame of the initial feature matrix by a scatter entropy algorithm to obtain a Mel-cepstral scatter entropy feature matrix as a sample set; using the sample set to train a pre-trained support vector machine classifier, and during the training process using a bat algorithm to automatically optimize the parameters of the pre-trained support vector machine classifier to obtain an acoustic signal fault diagnosis model; and using the trained acoustic signal fault diagnosis model to perform centrifugal pump fault diagnosis.

[0007] In an optional embodiment of the present invention, the step of extracting the Mel-cepstral coefficients of the acoustic signal as the initial feature matrix of the acoustic signal includes: preprocessing the acoustic signal to obtain a time domain signal of the preprocessed acoustic signal; performing a fast Fourier transform on each frame of the time domain signal to obtain a frequency domain signal; calculating the frequency domain signal to obtain an energy spectrum; filtering the energy spectrum through multiple Mel filters and calculating the logarithmic energies output by multiple Mel filters; and obtaining Mel-cepstral coefficients as the initial feature matrix of the acoustic signal through discrete cosine transform based on the multiple logarithmic energies.

[0008] In an optional embodiment of the present invention, the step of preprocessing the acoustic signal to obtain the time domain signal of the preprocessed acoustic signal includes: performing pre-emphasis processing on the acoustic signal; performing framing processing on the acoustic signal after the pre-emphasis processing; and performing windowing processing on the acoustic signal after the framing processing to obtain the time domain signal of the preprocessed acoustic signal.

[0009] In an optional embodiment of the present invention, the step of obtaining a frequency domain signal by performing a fast Fourier transform on each frame of the time domain signal is implemented by the following formula:

[0010]

[0011] Wherein, the value range of k is 0≤k≤Z, Z represents the number of Fourier transform points, S'(c) represents the time domain signal, X(k) represents the frequency domain signal, j represents a complex number, the value range of c is 1≤c≤C, and C is the total number of windows.

[0012] In an optional embodiment of the present invention, the step of obtaining Mel-frequency cepstral coefficients based on the logarithmic quantity by discrete cosine transform as the initial feature matrix of the acoustic signal is implemented by the following formula:

[0013]

[0014] Where M represents the number of filters, E(m) is the logarithmic energy output by the Mth filter group, C(z) represents the Mel-frequency cepstral coefficient, the value range of n is 1≤n≤N, and N represents the number of sampling points.

[0015] In an optional embodiment of the present invention, the step of calculating the scatter entropy of each frame of the initial feature matrix by the scatter entropy algorithm to obtain a Mel-cepstral scatter entropy feature matrix includes: mapping the one-dimensional time series of each frame of the initial feature matrix by the normal cumulative distribution integral to obtain a first one-dimensional time series; mapping the first one-dimensional time series by linear transformation to obtain a second one-dimensional time series; calculating an embedded subsequence based on the second one-dimensional time series; calculating a scatter pattern corresponding to the embedded subsequence based on the embedded subsequence; calculating the probability of each scatter pattern based on the corresponding scatter pattern; obtaining the scatter entropy of the one-dimensional time series of each frame of the initial feature matrix by defining the probability of the scatter pattern through Shannon entropy; normalizing the scatter entropy of the one-dimensional time series of each frame to obtain a Mel-cepstral scatter entropy feature matrix as a sample set.

[0016] In an optional embodiment of the present invention, the step of normalizing the scatter entropy of the one-dimensional time series of each frame to obtain a Mel-cephalometric scatter entropy feature matrix as a sample set includes: performing dimensionality reduction processing on the Mel-cephalometric scatter entropy feature matrix through principal component analysis; and using the Mel-cephalometric scatter entropy feature matrix after dimensionality reduction as a sample set.

[0017] In an optional embodiment of the present invention, the steps of using the sample set to train the pre-trained support vector machine classifier and optimizing the parameters of the pre-trained support vector machine classifier using the bat algorithm during the training process to obtain the acoustic signal fault diagnosis model include: building a pre-trained support vector machine classifier in the training set; optimizing the penalty coefficient and kernel function of the pre-trained support vector machine classifier by the bat algorithm to obtain the optimal penalty coefficient and kernel function; and inputting the optimal penalty coefficient and kernel function into the pre-trained support vector machine classifier to obtain the acoustic signal fault diagnosis model.

[0018] In an optional embodiment of the present invention, the step of optimizing the penalty coefficient and kernel function of the pre-trained support vector machine classifier by the bat algorithm to obtain the optimal penalty coefficient and kernel function includes: randomly generating two particles in the bat algorithm by a random number generator; calculating the local optimal solution of the two particles; iteratively updating the information of the two particles based on the local optimal solution; judging whether the two particles after the iterative update meet the output conditions, and if the output conditions are met, inputting them into the pre-trained support vector machine classifier as the optimal penalty coefficient and kernel function; if the output conditions are not met, returning to the step of randomly generating two particles in the bat algorithm by the random number generator.

[0019] To achieve the above-mentioned purpose and other related purposes, the present invention also provides a centrifugal pump acoustic signal fault diagnosis system including: an acquisition module, which acquires the acoustic signal of the centrifugal pump under different working conditions; an extraction module, which extracts the Mel-cepstral coefficients of the acoustic signal as the initial feature matrix of the acoustic signal; a calculation module, which calculates the scatter entropy of each frame of the initial feature matrix through a scatter entropy algorithm to obtain a Mel-cepstral scatter entropy feature matrix as a sample set; a training module, which uses the sample set to train a pre-trained support vector machine classifier, and uses the bat algorithm to optimize the parameters of the pre-trained support vector machine classifier during the training process to obtain an acoustic signal fault diagnosis model; and a diagnosis module, which uses the trained acoustic signal fault diagnosis model to perform centrifugal pump fault diagnosis.

[0020] The technical effect of the present invention is to provide a centrifugal pump acoustic signal fault diagnosis method. The present invention can deeply mine the signal by adding spread entropy through the Mel-frequency cepstral coefficient feature extraction part, which helps to improve the accuracy of centrifugal pump fault diagnosis based on acoustic signals. At the same time, the bat optimization algorithm is used to optimize the two particles of the classifier, namely the kernel function and the penalty factor, to find the optimal parameters in the classifier, thereby judging various fault conditions of the centrifugal pump, solving the problem of difficult installation of traditional vibration sensors and avoiding the problem of staff identifying faults in harsh sites, while improving the accuracy of centrifugal pump fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is an application scenario diagram of a centrifugal pump acoustic signal fault diagnosis method proposed by the present invention;

[0022] Figure 2 This is a flow chart of a centrifugal pump acoustic signal fault diagnosis method proposed by the present invention;

[0023] Figure 3 This is a specific flow chart of extracting the Mel-frequency cepstral coefficients of the acoustic signal proposed by the present invention;

[0024] Figure 4 This is a specific flow chart of the pre-processing proposed by the present invention;

[0025] Figure 5 This is a specific flow chart of obtaining the Mel-Cepstrum Scatter Entropy feature matrix proposed by the present invention;

[0026] Figure 6 A specific flow chart as a sample set proposed by the present invention;

[0027] Figure 7 A specific flow chart of the pre-trained support vector machine classifier proposed by the present invention;

[0028] Figure 8This is a specific flow chart of the optimization proposed by the present invention;

[0029] Figure 9 Partial time domain diagram of the noise signal proposed by the present invention;

[0030] Figure 10 This is a comparison diagram before and after the normal working condition feature extraction proposed by the present invention;

[0031] Figure 11 This is a comparison diagram before and after the extraction of cavitation working condition characteristics proposed by the present invention;

[0032] Figure 12 This is a comparison diagram before and after the bolt loosening working condition feature extraction proposed by the present invention;

[0033] Figure 13 This is a comparison diagram before and after the extraction of the misalignment working condition characteristics proposed by the present invention;

[0034] Figure 14 This is the t-SNE visualization two-dimensional graph proposed in the present invention;

[0035] Figure 15 This is the t-SNE visualization three-dimensional graph proposed by the present invention;

[0036] Figure 16 This is the confusion matrix diagram of the bat algorithm classifier after dimensionality reduction proposed by the present invention;

[0037] Figure 17 This is the confusion matrix diagram of the Mel-cepstral scatter entropy feature matrix in the classifier proposed by the present invention;

[0038] Figure 18 This is the confusion matrix diagram of the bat algorithm classifier under variational mode decomposition proposed by the present invention;

[0039] Figure 19 This is the confusion matrix diagram of the bat algorithm classifier proposed in the present invention;

[0040] Figure 20 This is a comparison chart of the accuracy iteration curve proposed by the present invention;

[0041] Figure 21 This is a functional module diagram of the model improvement device proposed in the present invention;

[0042] Figure 22 This is a structural block diagram of the electronic device proposed by the present invention. DETAILED DESCRIPTION

[0043] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0044] It should be noted that the illustrations provided in the following embodiments are merely schematic illustrations of the basic concept of the present invention. The illustrations only show components related to the present invention and are not drawn according to the number, shape, and size of components in actual implementation. In actual implementation, the type, quantity, and proportion of each component may be changed arbitrarily, and the component layout may also be more complex.

[0045] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the embodiments of the present invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring the embodiments of the present invention.

[0046] Mel-Frequency Cepstral Coefficients (MFCCs) are the coefficients that make up the Mel-Frequency Cepstral. They are derived from the cepstrum of an audio clip. The difference between the cepstrum and the Mel-Frequency Cepstral is that the Mel-Frequency Cepstral uses equally spaced frequency bands on the Mel scale, which better approximates the human auditory system than the linearly spaced bands used in the normal logarithmic cepstrum. This nonlinear representation can improve the representation of audio signals in various fields, such as audio compression.

[0047] The Bat Algorithm (BA) is a highly efficient bio-inspired algorithm developed by observing the biological characteristics of bats in searching for prey and avoiding obstacles. The algorithm generates random bat particles through initialization, each with a specific velocity and position. Each particle continuously iterates its own parameters, such as pulse frequency, velocity, position, loudness, and pulse emission rate, as the search environment changes.

[0048] Support vector machines (SVMs) are a machine learning method based on statistical theory and have been widely used in fault diagnosis. Their basic principle is to find the optimal splitting hyperplane to make the problem linearly separable. If the samples are linearly separable, the SVM can achieve linear segmentation of the dataset using the linear hyperplane. However, most samples are nonlinearly separable. For this type of data, the SVM uses a kernel function to map the data into a high-dimensional space and then finds a hyperplane within that space to achieve linear segmentation. For outliers in the dataset, the SVM model can be optimized by adding a relaxation factor, or a penalty factor. The SVM is a learning machine based on a kernel function, and its generalization ability depends significantly on the kernel function chosen. The choice of kernel function directly affects the mapping of samples into the high-dimensional space, and thus the SVM's classification performance. The kernel function value affects the precision of the sample segmentation. Too large a value can lead to poor classification results, while too small a value can lead to overfitting. The penalty factor value determines the impact of outliers on the model. Too large a value can lead to overfitting, while too small a value can lead to underfitting.

[0049] Dispersion entropy (DE) is an algorithm used to measure the degree of irregularity and uncertainty of nonlinear time series.

[0050] Figure 1 The application scenario diagram of the model method provided by the embodiment of the present invention is as follows: the present invention first pre-processes the acoustic signal and extracts the Mel-frequency cepstral coefficients, then performs scatter entropy processing on the extracted Mel-frequency cepstral coefficients, and then uses principal component analysis to reduce the dimension. A support vector machine classifier is created, and the support vector machine classifier is trained using the bat algorithm to obtain a support vector machine classifier based on the bat algorithm. Finally, the accuracy between the predicted working conditions and the actual working conditions of the support vector machine based on the bat algorithm is calculated. In other application scenarios, fault diagnosis for different working conditions of centrifugal pumps can be set according to actual conditions, and the embodiments of the present invention are not limited to this.

[0051] The electronic device may be any electronic product that can perform human-computer interaction with a user, such as a personal computer, a tablet computer, a smart phone, a personal digital assistant (PDA), an interactive network television (IPTV), a smart wearable device, etc.

[0052] The electronic device may further include a network device and / or a user device, wherein the network device includes, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of hosts or network servers.

[0053] The network where the electronic device is located includes but is not limited to the Internet, a wide area network, a metropolitan area network, a local area network, a virtual private network (VPN), etc.

[0054] Figure 2 This is a flow chart of a centrifugal pump sound signal fault diagnosis method provided by an embodiment of the present invention. It should be noted that the present invention is based on the acquisition and diagnosis of sound signals under four working conditions: normal, cavitation, bolt loosening and misalignment, and compares with multiple diagnostic methods. This method can be applied to Figure 1 The implementation environment shown is a schematic diagram. It should be understood that the method can also be applied to other exemplary implementation environments and specifically executed by devices in other implementation environments. This embodiment does not limit the implementation environment to which the method is applicable.

[0055] like Figure 2 As shown, a centrifugal pump acoustic signal fault diagnosis method of this embodiment at least includes:

[0056] Step S20: Acquire the acoustic signals of the centrifugal pump under different working conditions. It should be noted that the acoustic signals of the centrifugal pump under different working conditions, including normal, cavitation, loose bolts, and misalignment, are acquired.

[0057] Step S21: extracting the Mel-frequency cepstral coefficients of the acoustic signal as the initial feature matrix of the acoustic signal. It should be noted that extracting the Mel-frequency cepstral coefficients is specifically divided into four parts, including audio file preprocessing, fast Fourier transform, Mel-frequency filter filtering and discrete cosine transform.

[0058] Step S22: Calculate the scatter entropy of each frame of the initial feature matrix using a scatter entropy algorithm to obtain a Mel-cephalometric scatter entropy feature matrix as a sample set. It should be noted that the scatter entropy algorithm specifically includes normal cumulative partial integration, linear transformation, calculation of embedded subsequences, calculation of scatter patterns, calculation of scatter pattern probabilities, and calculation of scatter entropy.

[0059] Step S23: Use the sample set to train the pre-trained support vector machine classifier, and use the bat algorithm to optimize the parameters of the pre-trained support vector machine classifier during the training process to obtain an acoustic signal fault diagnosis model. It should be noted that the pre-trained support vector machine classifier is first built and then the training set is input. The default parameters of the bat algorithm include population size, number of iterations, frequency range and dimension, etc. The bat algorithm generates random particles and calculates the fitness of the particles in the continuous iteration process to evaluate the quality of individuals. Furthermore, the sample set includes a training set and a test set.

[0060] Step S24: Perform centrifugal pump fault diagnosis using the trained acoustic signal fault diagnosis model. It should be noted that the acoustic signals in the test set are input into the trained pre-trained support vector machine classifier. The trained pre-trained support vector machine classifier outputs a predicted operating condition. The predicted operating condition is then compared with the actual operating condition for verification and classification accuracy is calculated. The classification accuracy indicates that the closer the predicted operating condition is to the actual operating condition, the better the training effect of the pre-trained support vector machine classifier is.

[0061] like Figure 3 As shown, the specific steps of extracting the Mel-cepstral coefficients of the sound signal provided in this embodiment include:

[0062] Step S30: Preprocess the acoustic signal to obtain a time domain signal of the preprocessed acoustic signal. It should be noted that the preprocessing specifically includes pre-emphasis, framing, and windowing.

[0063] In a specific embodiment, if Figure 4 As shown, the specific steps of preprocessing provided in this embodiment include:

[0064] Step S40: Pre-emphasize the sound signal. It should be noted that the pre-emphasis function in the pre-emphasis of the sound signal is expressed as follows:

[0065] y(n)=x(n)-α·x(n-1)

[0066] Wherein, α is the pre-emphasis coefficient, y(n) is the preprocessed signal, x(n) is the original signal containing N sampling points, and the value range of n is 1≤n≤N.

[0067] Step S41: Framing the pre-emphasized acoustic signal. It should be noted that to ensure that the time domain signal is converted into a frequency domain signal while retaining timing information, the time domain signal needs to be framed. This means that the speech signal is divided into segments of equal length, with the number of segments resulting being the number of frames. The number of sampling points at which a frame of speech signal intersects with the preceding and following frames is the number of frames. For example, the frame length is generally set to 25ms, and the frame shift is set to 10ms.

[0068] Step S42: Windowing is performed on the framed acoustic signal to obtain the time domain signal of the pre-processed acoustic signal. It should be noted that a window function is required to eliminate the influence of spectral leakage on the framed acoustic signal. The window function is generally set to a Hamming window, and the Hamming window expression is as follows:

[0069]

[0070] The value range of c is 1≤c≤C, and C is the total number of windows.

[0071] Step S31: Perform a fast Fourier transform on each frame of the time domain signal to obtain a frequency domain signal. It should be noted that the framed signal S(c) is multiplied by the Hamming window W(c) to obtain the windowed signal S'(c). The frequency domain signal is obtained by the following formula:

[0072]

[0073] Wherein, the value range of k is 0≤k≤Z, Z represents the number of Fourier transform points, S'(c) represents the time domain signal, X(k) represents the frequency domain signal, j represents a complex number, the value range of c is 1≤c≤C, and C is the total number of windows.

[0074] Step S32: Calculate the frequency domain signal to obtain an energy spectrum. It should be noted that the energy spectrum is X(k) 2 energy spectrum.

[0075] Step S33: Filter the energy spectrum using multiple Mel filters, and calculate the logarithmic energy of the outputs of the multiple Mel filters. It should be noted that the calculation of the logarithmic number of the outputs of the multiple Mel filters is achieved by the following formula:

[0076]

[0077] Among them, E(m) represents the logarithmic energy of the output of the M filter group, and the value range of m is 0 <m<M,H m (k) represents the transfer function of the m-th filter, X(k) represents the frequency domain signal, the value range of k is 0≤k≤Z, and Z represents the number of Fourier transform points.

[0078] Specifically, the transfer function of the m-th filter is expressed as follows:

[0079]

[0080] Among them, H m (k) The range of the transfer function is:<f(m-1);f(m-1)≤k≤f(m);f(m)≤k≤f(m+1);k> f(m-1), f(m) are the center frequencies of the triangular filter.

[0081] Step S34: Based on the plurality of logarithmic energies, Mel-frequency cepstral coefficients are obtained by discrete cosine transform to serve as the initial feature matrix of the acoustic signal. It should be noted that the Mel-frequency cepstral coefficients obtained by discrete cosine transform are implemented as follows:

[0082]

[0083] Among them, C(z) represents the Mel cepstral coefficient, E(m) represents the logarithmic energy of the output of the M filter group, Z represents the number of points of Fourier transform, and m ranges from 0 <m<M。

[0084] like Figure 5 As shown, the specific steps of obtaining the Mel-cepstrum scatter entropy feature matrix provided in this embodiment include:

[0085] Step S50: Mapping the one-dimensional time series of each frame of the initial feature matrix by normal cumulative distribution integral to obtain a first one-dimensional time series. It should be noted that the expression of the normal cumulative distribution integral is as follows:

[0086]

[0087] Where y={y1,y2,…,y N} represents a one-dimensional time series x with a length of N n ={x1,x2,…,x N}, y i ∈(0,1), μ and σ 2 are the expectation and variance of the time series x, respectively.

[0088] Step S51: Map the first one-dimensional time series through linear transformation to obtain a second one-dimensional time series. It should be noted that the linear transformation is expressed as follows:

[0089]

[0090] Where z={z1,z2,…,z N} means y={y1,y2,…,y N}, z n ∈[1,c], where R is the rounding function and c is the number of categories Represents the nth element of the cth category.

[0091] Step S52: Calculate the embedded subsequence according to the second one-dimensional time series. It should be noted that the calculation of the embedded subsequence is achieved by the following formula:

[0092]

[0093] Where j = 1, 2, ..., N-(m-1)d, m represents the embedding dimension, and d represents the delay.

[0094] Step S53: Based on the embedded subsequence, the dispersion pattern corresponding to the embedded subsequence is calculated. It should be noted that the dispersion pattern corresponding to the embedded subsequence is calculated and implemented by the following formula:

[0095]

[0096] in, The corresponding scatter pattern is c m Indicates each The number of scattering modes, m represents the embedding dimension, and d represents the delay.

[0097] Step S54: Calculate the probability of each of the scattering patterns according to the corresponding scattering patterns. It should be noted that the probability of each of the scattering patterns is calculated by the following formula:

[0098]

[0099] in, express Mapping to each scatter pattern The number of express Mapping to each scatter pattern The number of divided by The number of elements in , m represents the embedding dimension, and d represents the delay.

[0100] Step S55: Based on the probability of the dispersion pattern, the dispersion entropy of the one-dimensional time series of each frame of the initial feature matrix is obtained by the definition of Shannon entropy. It should be noted that the dispersion entropy of the one-dimensional time series is obtained by the following formula:

[0101]

[0102] in, express Mapping to each scatter pattern The number of divided by The number of elements in , DE(x,m,c,d) represents the diffusion entropy.

[0103] Step S56: normalize the scatter entropy of the one-dimensional time series of each frame to obtain a Mel-cepstrum scatter entropy feature matrix as a sample set. It should be noted that the Mel-cepstrum scatter entropy feature matrix is obtained by the following formula:

[0104]

[0105] Among them, DE(x,m,c,d) represents the diffusion entropy, c mIndicates each The number of scatter patterns.

[0106] Specifically, such as Figure 6 As shown, the specific steps provided as a sample set in this embodiment include:

[0107] Step S60: Dimensionality reduction processing is performed on the Mel-Cepstral Scatter Entropy feature matrix using principal component analysis. It should be noted that in actual fault diagnosis, the original signal after feature extraction is still high-dimensional. If the high-dimensional feature vector is fed into the pre-trained support vector machine classifier, the large amount of redundant pre-noise data in the feature vector will occupy a large amount of computer storage space, thereby increasing the time consumed by fault diagnosis.

[0108] Specifically, principal component analysis can reduce the complexity of calculations while trying to retain the effectiveness of feature classification. It can compress features to a certain extent. When collecting signals from centrifugal pumps, the same fault information may be contained in a short period of time under the same operating conditions, which leads to a large amount of information overlap. In order to avoid data redundancy caused by information overlap, principal component analysis extracts principal components by constructing a linear combination of the original variables, and the number of principal components is less than the number of original variables.

[0109] Step S61: using the Mel-cepstrum scatter entropy feature matrix after dimension reduction as a sample set.

[0110] like Figure 7 As shown, the specific steps of obtaining the pre-trained support vector machine classifier provided in this embodiment include:

[0111] Step S70: Building a pre-trained support vector machine classifier in the training set.

[0112] Step S71: Optimize the penalty coefficient and kernel function of the pre-trained support vector machine classifier using the Bat Algorithm to obtain the optimal penalty coefficient and kernel function. It should be noted that the parameter selection is performed within a range, and two particles are randomly generated using a random number generator within this range. These two particles represent the penalty coefficient and kernel coefficient. The fitness values of the particles are calculated during the continuous iteration process to evaluate their quality. Finally, the particles are judged when outputting the individual particles.

[0113] Step S72: inputting the optimal penalty coefficient and kernel function into the pre-trained support vector machine classifier to obtain an acoustic signal fault diagnosis model.

[0114] like Figure 8 As shown, the specific steps of performing optimization through the bat algorithm provided in this embodiment include:

[0115] Step S80: randomly generating two particles in the bat algorithm through a random number generator.

[0116] Step S81: Calculate and obtain the local optimal solution of the two particles. It should be noted that the calculation to obtain the local optimal solution of the two particles is implemented by the following formula:

[0117]

[0118] Among them, f i represents the pulse frequency of particle i; β represents a random factor that obeys uniform distribution and ranges from [0,1]; represents the velocity of particle i at time t; similarly represents the position of particle i at time t; x * represents the global optimal solution.

[0119] Step S82: Iteratively update the information of the two particles based on the local optimal solution. It should be noted that the expression of the iterative update is as follows:

[0120] x new =x old +ε·A t

[0121] Among them, x new is a new solution for local search; x old is a local search for old solutions; ε is a random number in the range [-1,1]; A t is the average loudness of all bat particles at time t.

[0122] Specifically, the iterative update expression of the loudness and pulse emission rate is as follows:

[0123]

[0124] in, represents the loudness of particle i at time t+1; represents the pulse emission rate of particle i at time t+1; Indicates the maximum pulse emission rate in the particle group; α and γ both represent the coefficient of variation, α∈(0,1), γ>0. During the particle iteration process,

[0125] Step S83: Determine whether the two particles after the iterative update meet the output conditions. If the output conditions are met, the two particles are input into the pre-trained support vector machine classifier as the optimal penalty coefficient and kernel function. If the output conditions are not met, return to the random number generator to randomly generate two particles in the bat algorithm.

[0126] It should be noted that the optimal parameter after optimization is the particle individual output last. The particle output requirement is that when the fitness value of the particle is less than the set value or the number of iterations of the particle is greater than the set value, the particle is output as the optimal parameter.

[0127] like Figure 9-20 As shown, the technical solution of the present invention is described below with reference to specific embodiments:

[0128] It should be noted that the rated speed of the centrifugal pump is 2950 rpm / min, a microphone is used to collect the sound signal, and the signal sampling frequency is set to 51200 Hz.

[0129] like Figure 9 As shown, Figure 9 The figure includes the normal working condition, the cavitation working condition obtained by changing the inlet and outlet pressure of the centrifugal pump, the loose bolt working condition obtained by loosening the base bolts, and the partial time domain diagram of the noise signal under the misalignment working condition obtained by adjusting the offset between the motor and the pump body drive shaft.

[0130] like Figure 10-13 As shown in the figure, 20 seconds of noise signals were selected from each of the four operating conditions, totaling 1,024,000 sampling points. Each 4,096 sampling points constituted a frame for initial feature matrix extraction, with 250 groups for each operating condition, for a total of 1,000 groups. The collected noise signals were divided into training samples (800 training samples) and test samples (200 test samples). The initial feature dimension was set to 13, the frame length was 25ms, and the frame shift was 10ms. Feature extraction was performed on each frame based on the selected parameters, resulting in a 272×13 feature matrix, or 13 feature vectors of length 272, for each frame. To preserve the original dynamic characteristics of the noise signal, two differential operations were performed on the initial features to preserve the characteristics and trend of the acoustic signal. This resulted in an initial feature matrix of 272×39 per frame.

[0131] The embedding dimension of the scatter entropy is set to 3, the number of categories is set to 6, and the features extracted by Mel cepstrum are used as input. A scatter entropy value can be calculated for each frame column vector, and the size of the feature matrix for each frame is 1×39. The comparison between the feature sequence extracted by Mel cepstrum scatter entropy feature and the original time domain waveform is shown in the figure below. Figure 7-10 As shown in the figure, the number of sampling points after feature extraction is significantly reduced compared to the original signal, and it can roughly represent the original time domain signal under different working conditions, effectively optimizing the problem of too many sampling points. Specifically, Figure 10 This is a comparison chart before and after the extraction of normal working condition features. Figure 11 This is a comparison chart before and after cavitation working condition feature extraction. Figure 12 This is a comparison chart before and after the extraction of bolt loosening working condition features. Figure 13 This is a comparison diagram before and after the misalignment working condition feature extraction.

[0132] like Figure 14-15 As shown in the figure, the feature matrix extracted by the Mel-cephalogram scatter entropy is reduced in dimension in the row direction, that is, the Mel-cephalogram scatter entropy features in a time series are represented by one or a few values. The noise signals after feature extraction for 4 working conditions and a total of 1000 frames are all subjected to PCA (principal component analysis) dimensionality reduction, and the dimension can be reduced from the original 39 to 16, which greatly reduces the problem of long training time caused by data redundancy. The final data set size is 1000×16 (250×4×16), of which the training set size is 800×16 (200×4×16) and the validation set size is 200×16 (50×4×16). The extracted features are visualized by t-SNE to intuitively observe whether the features of the original signal are retained while reducing the dimensionality and whether it has a certain degree of discrimination and recognition. The signal visualization discrimination after feature extraction and dimensionality reduction is good, and the difference between normal working conditions and fault conditions is obvious. It can be substituted into the subsequent support vector machine classifier model based on the bat algorithm for classification. Specifically, Figure 14 Visualize the 2D graph for t-SNE, Figure 15 Visualize the 3D plot for t-SNE.

[0133] like Figure 16 As shown in the figure, after Mel-Cepstral Scatter Entropy feature extraction and dimensionality reduction, the input is input into the support vector machine classifier model based on the bat algorithm, where the basic parameters of the bat algorithm are set as follows: population size is 20, number of iterations is 50, loudness is 0.5, pulse rate is 0.5, frequency range is [0,2], and dimension is 2. The bat optimization algorithm obtains the optimal solution of the support vector machine penalty coefficient c and kernel function g through global search and imports it into the support vector machine classifier. Figure 16 is the confusion matrix of the support vector machine classifier based on the bat algorithm after Mel-cepstral scatter entropy feature extraction.

[0134] like Figure 17-20 As shown, in order to verify the diagnostic advantage of the support vector machine classifier model based on the bat algorithm after Mel-cepstral scatter entropy feature extraction, the present invention is compared with the MFCC-SVM (support vector machine classifier based on Mel-cepstral), VMD-BASVM (support vector machine classifier based on the bat algorithm after variational mode decomposition), and MFCC-BASVM (support vector machine classifier based on the bat algorithm after Mel-cepstral feature extraction) models. Confirm the optimization ability of the bat optimization algorithm, the improvement ability of DE on MFCC feature extraction, and the difference in diagnostic ability with the mainstream feature extraction method VMD. In order to eliminate the influence of singular results in model diagnosis on the final classification accuracy of the model, the present invention performs ten random diagnoses on the four models respectively, and calculates the average diagnostic accuracy of each fault category. Specifically, as Figure 17is the MFCC-SVM confusion matrix, Figure 18 is the VMD-BASVM confusion matrix, Figure 19 is the MFCC-BASVM confusion matrix, Figure 20 The comparison chart of the average accuracy iteration curves of the four models is shown in Table 1.

[0135] Table 1

[0136]

[0137] Specifically, such as Figure 20 As can be seen from Table 1, the MFCC-SVM model does not include an optimization process, and the support vector machine parameters are manually set, so its accuracy does not increase with the number of iterations. The VMD-BASVM model has slightly higher accuracy than the MFCC-BASVM model, but takes longer to converge. The MFCCDE-BASVM model has slightly higher accuracy than the VMD-BASVM model and converges faster. The addition of the bat optimization algorithm significantly improves the classification accuracy of the support vector machine. Although the feature extraction effect of MFCC is not as good as that of VMD, the optimization of MFCC through the spread entropy not only retains the original advantage of MFCC's fast convergence, but also improves the final classification accuracy.

[0138] The feature extraction and diagnosis time of MFCC is very short, but the accuracy of traditional MFCC is not as good as VMD. The MFCC improved by spread entropy slightly increases the feature extraction and diagnosis time, and the average diagnosis accuracy is improved from the original 94.4% to 96.45%, and the feature extraction time is shorter than VMD.

[0139] like Figure 21 As shown, the present invention also provides a centrifugal pump sound signal fault diagnosis system including an acquisition module 210, an extraction module 211, a calculation module 212, a training module 213, and a diagnosis module 214. The acquisition module 210 is used to acquire the sound signal of the centrifugal pump under different working conditions; the extraction module 211 is used to extract the Mel-cephalogram coefficient of the sound signal based on the sound signal as the initial feature matrix of the sound signal; the calculation module 212 is used to calculate the scatter entropy of each frame of the initial feature matrix by a scatter entropy algorithm to obtain a Mel-cephalogram scatter entropy feature matrix; the training module 213 is used to train a pre-trained support vector machine classifier using the sample set, and optimize the parameters of the pre-trained support vector machine classifier using the bat algorithm during the training process to obtain a sound signal fault diagnosis model; the diagnosis module 214 is used to perform centrifugal pump fault diagnosis using the trained sound signal fault diagnosis model.

[0140] It should be noted that the above embodiments provide Figure 21The centrifugal pump acoustic signal fault diagnosis system illustrated herein shares the same concept as the centrifugal pump acoustic signal fault diagnosis method provided in the aforementioned embodiment. The specific manner in which each module and unit performs its operations has been described in detail in the method embodiment and will not be repeated here. In actual applications, the centrifugal pump acoustic signal fault diagnosis system provided in the aforementioned embodiment may, as needed, allocate the aforementioned functions to different functional modules. This means dividing the internal structure of the device into different functional modules to perform all or part of the aforementioned functions, and this is not a limitation herein.

[0141] An embodiment of the present invention also provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the electronic device implements the centrifugal pump acoustic signal fault diagnosis method provided in the above-mentioned embodiments.

[0142] Figure 22 FIG1 shows a schematic diagram of the structure of a computer system suitable for implementing an electronic device according to an embodiment of the present invention. Figure 22 The computer system 800 of the electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0143] like Figure 22 As shown, the computer system 2200 includes a central processing unit (CPU) 2201, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 2202 or the program loaded from the storage part 2206 into the random access memory (RAM) 2203, such as the method described in the above embodiment. Various programs and data required for system operation are also stored in the RAM 2203. The CPU 2201, ROM 2202 and RAM 2203 are connected to each other via a bus 2204. An input / output (I / O) interface 2205 is also connected to the bus 2204.

[0144] The following components are connected to the I / O interface 2205: an input section 2206 including a keyboard, a mouse, and the like; an output section 2207 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and speakers; a storage section 2206 including a hard disk; and a communication section 2208 including a network interface card such as a LAN (Local Area Network) card or a modem. The communication section 2208 performs communication processing via a network such as the Internet. A drive 2210 is also connected to the I / O interface 2205 as needed. Removable media 2211, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 2210 as needed, so that computer programs read from the removable media can be installed in the storage section 2206 as needed.

[0145] In particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product that includes a computer program carried on a computer-readable medium, the computer program including a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 2208 and / or installed from the removable medium 2211. When the computer program is executed by the central processing unit (CPU) 2201, the various functions defined in the system of the present invention are performed.

[0146] It should be noted that the computer-readable medium shown in the embodiments of the present invention may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. The computer-readable storage medium may, for example, be an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries a computer-readable computer program. Such a propagated data signal may take a variety of forms, including, but not limited to, an electromagnetic signal, an optical signal, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. A computer program embodied on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0147] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. Each box in the flowchart or block diagram can represent a module, program segment, or part of the code, and the above-mentioned module, program segment, or part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two boxes shown in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.

[0148] The units involved in the embodiments of the present invention may be implemented in software or hardware, and the units described may also be provided in a processor. In some cases, the names of these units do not limit the units themselves.

[0149] Another aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon. When executed by a computer processor, the computer program causes the computer to perform the centrifugal pump acoustic signal fault diagnosis method described above. The computer-readable storage medium may be included in the electronic device described in the above embodiments, or may exist independently and not be incorporated into the electronic device.

[0150] Another aspect of the present invention provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the centrifugal pump acoustic signal fault diagnosis method provided in each of the above embodiments.

[0151] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, any equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be covered by the claims of the present invention.

Claims

1. A centrifugal pump acoustic signal fault diagnosis method, characterized in that: include: Obtain the acoustic signals of the centrifugal pump under different working conditions; extracting Mel-frequency cepstral coefficients of the acoustic signal as an initial feature matrix of the acoustic signal; Calculating the scatter entropy of each frame of the initial feature matrix by a scatter entropy algorithm to obtain a Mel-cepstrum scatter entropy feature matrix as a sample set; The sample set is used to train a pre-trained support vector machine classifier, and during the training process, the bat algorithm is used to optimize the parameters of the pre-trained support vector machine classifier to obtain an acoustic signal fault diagnosis model; The trained acoustic signal fault diagnosis model is used to diagnose centrifugal pump faults.

2. The centrifugal pump sound signal fault diagnosis method according to claim 1, characterized in that: The step of extracting the Mel-frequency cepstral coefficients of the acoustic signal as the initial feature matrix of the acoustic signal includes: Preprocessing the acoustic signal to obtain a time domain signal of the preprocessed acoustic signal; Performing a fast Fourier transform on each frame of the time domain signal to obtain a frequency domain signal; Calculating the frequency domain signal to obtain an energy spectrum; Filtering the energy spectrum using a plurality of Mel filters, and calculating the logarithmic energy of outputs of the plurality of Mel filters; Mel-frequency cepstral coefficients are obtained through discrete cosine transformation based on the plurality of logarithmic energies to serve as the initial feature matrix of the acoustic signal.

3. The centrifugal pump sound signal fault diagnosis method according to claim 2, characterized in that: The step of preprocessing the acoustic signal to obtain a time domain signal of the preprocessed acoustic signal includes: performing pre-emphasis processing on the acoustic signal; performing frame processing on the acoustic signal after the pre-emphasis processing; The frame-processed acoustic signal is subjected to windowing processing to obtain a pre-processed time domain signal of the acoustic signal.

4. The centrifugal pump acoustic signal fault diagnosis method according to claim 2, characterized in that: The step of obtaining a frequency domain signal by performing a fast Fourier transform on each frame of the time domain signal is achieved by the following formula: Wherein, the value range of k is 0≤k≤Z, Z represents the number of Fourier transform points, S'(c) represents the time domain signal, X(k) represents the frequency domain signal, j represents a complex number, the value range of c is 1≤c≤C, and C is the total number of windows.

5. The centrifugal pump sound signal fault diagnosis method according to claim 2, characterized in that: The step of obtaining Mel-frequency cepstral coefficients as the initial feature matrix of the acoustic signal by discrete cosine transform based on the logarithmic quantity is implemented by the following formula: Where M represents the number of filters, E(m) is the logarithmic energy output by the Mth filter group, C(z) represents the Mel-frequency cepstral coefficient, the value range of n is 1≤n≤N, and N represents the number of sampling points.

6. The centrifugal pump sound signal fault diagnosis method according to claim 1, characterized in that: The step of calculating the scatter entropy of each frame of the initial feature matrix by using a scatter entropy algorithm to obtain a Mel-cepstrum scatter entropy feature matrix as a sample set includes: Mapping the one-dimensional time series of each frame of the initial feature matrix by normal cumulative distribution integral to obtain a first one-dimensional time series; Mapping the first one-dimensional time series through linear transformation to obtain a second one-dimensional time series; Calculating an embedded subsequence according to the second one-dimensional time series; Obtaining a distribution pattern corresponding to the embedded subsequence by calculation based on the embedded subsequence; Calculating the probability of each scattering pattern according to the corresponding scattering pattern; Obtaining the scatter entropy of the one-dimensional time series of each frame of the initial feature matrix based on the probability of the scatter pattern defined by Shannon entropy; The scatter entropy of the one-dimensional time series of each frame is normalized to obtain a Mel-cepstral scatter entropy feature matrix as a sample set.

7. The centrifugal pump acoustic signal fault diagnosis method according to claim 6, characterized in that: After the step of normalizing the scatter entropy of the one-dimensional time series of each frame to obtain a Mel-cepstrum scatter entropy feature matrix as a sample set, the method further includes: Performing dimensionality reduction processing on the Mel-cepstral scatter entropy feature matrix by principal component analysis; The Mel-cepstral scatter entropy feature matrix after dimension reduction is used as a sample set.

8. The centrifugal pump acoustic signal fault diagnosis method according to claim 1, characterized in that: The step of training a pre-trained support vector machine classifier using the sample set and optimizing the parameters of the pre-trained support vector machine classifier using the bat algorithm during the training process to obtain an acoustic signal fault diagnosis model comprises: Building a pre-trained support vector machine classifier in the sample set; The penalty coefficient and kernel function of the pre-trained support vector machine classifier are optimized by the bat algorithm to obtain the optimal penalty coefficient and kernel function; The optimal penalty coefficient and kernel function are input into the pre-trained support vector machine classifier to obtain an acoustic signal fault diagnosis model.

9. The centrifugal pump acoustic signal fault diagnosis method according to claim 8, characterized in that: The step of optimizing the penalty coefficient and kernel function of the pre-trained support vector machine classifier by the bat algorithm to obtain the optimal penalty coefficient and kernel function includes: Two particles are randomly generated in the bat algorithm by a random number generator; Calculating a local optimal solution of the two particles; Iteratively updating the information of the two particles based on the local optimal solution; The two particles after the iterative update are judged whether the two particles meet the output condition; if the output condition is met, the two particles are input into the pre-trained support vector machine classifier as the optimal penalty coefficient and kernel function; if the output condition is not met, the two particles are returned to the random number generator in the bat algorithm to randomly generate two particles.

10. A centrifugal pump acoustic signal fault diagnosis system, characterized in that: include: An acquisition module is used to obtain the acoustic signals of the centrifugal pump under different working conditions; An extraction module extracts Mel-frequency cepstral coefficients of the acoustic signal as an initial feature matrix of the acoustic signal; A calculation module calculates the scatter entropy of each frame of the initial feature matrix by using a scatter entropy algorithm to obtain a Mel-cepstrum scatter entropy feature matrix as a sample set; A training module, which uses the sample set to train a pre-trained support vector machine classifier, and uses a bat algorithm to optimize the parameters of the pre-trained support vector machine classifier during the training process to obtain an acoustic signal fault diagnosis model; The diagnostic module uses the trained acoustic signal fault diagnosis model to perform centrifugal pump fault diagnosis.

Images