Bolt fault diagnosis method, system and equipment based on sound signals and medium

Abstract

The invention relates to a bolt fault diagnosis method, system and device based on sound signals and a medium. The method comprises the steps that sound signals are decomposed into intrinsic mode functions through time-varying filtering empirical mode decomposition, hyper-parameters of time-varying filtering empirical mode decomposition are optimized with information entropy as a fitness function through a tuna swarm optimization algorithm, and the noise interference problem under the complex working condition is effectively solved. And setting a threshold value based on the energy ratio of the modal component, performing singular value decomposition on the modal component meeting the condition, determining a truncation order through a singular value contribution rate difference spectrum, completing signal denoising reconstruction, and further optimizing the signal quality. Time-frequency domain features of the denoised signals are extracted, dimension reduction processing is carried out through kernel principal component analysis, and a feature vector data set is constructed; and finally, fault classification is carried out on the feature vectors through an extreme random forest. The method has a good classification diagnosis effect on sound signals under different working conditions, and the bolt fault diagnosis accuracy is obviously improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent monitoring of industrial equipment, and in particular to a bolt fault diagnosis method, system, equipment and medium based on acoustic signals. Background Art

[0002] In actual industrial scenarios, bolt fault diagnosis technology based on acoustic signals has attracted much attention due to its non-destructive, fast, and efficient features. However, existing technologies still face many problems when using acoustic signals in combination with neural networks and deep learning methods for bolt fault detection. On the one hand, when collecting acoustic signals under complex working conditions, the noise interference is strong, and traditional noise reduction technologies often fail to achieve the desired noise reduction effect. On the other hand, traditional deep learning methods have many limitations, such as complex parameter adjustment and prone to overfitting when training with small samples. This not only increases the difficulty of model training but also reduces computational efficiency. Therefore, a new bolt fault diagnosis method, system, equipment, and medium based on acoustic signals are needed. Summary of the Invention

[0003] In view of the shortcomings of the prior art mentioned above, the purpose of the present invention is to provide a bolt fault diagnosis method, system, equipment and medium based on acoustic signals, which is used to solve the problems in the prior art that traditional bolt fault diagnosis methods are easily interfered by background noise in complex industrial environments, resulting in decreased detection accuracy, and it is difficult to accurately extract transient features when processing low signal-to-noise ratio, non-stationary signals and low sampling rate scenarios; common deep learning methods have problems such as many hyperparameters, complex parameter adjustment, and easy overfitting of small sample training, while traditional machine learning methods have weak generalization ability and are difficult to adapt to complex working conditions.

[0004] To achieve the above-mentioned objectives and other related objectives, the present invention provides a bolt fault diagnosis method, system, device and medium based on acoustic signals, which are applied to the field of intelligent monitoring of industrial equipment. The method includes: obtaining an acoustic signal sequence of knocking bolts under working conditions; decomposing the acoustic signal sequence according to Time-Variant Filtered Empirical Mode Decomposition (TVFEMD) to obtain corresponding modal components; performing singular value decomposition (SVD) calculation and reconstruction on the modal components according to energy proportion to obtain a denoised signal; inputting the denoised signal into a bolt fault diagnosis model to obtain a bolt fault diagnosis result; wherein the bolt fault diagnosis model is an extreme random forest model.

[0005] In one embodiment of the present invention, the step of decomposing the acoustic signal sequence according to the time-varying filter empirical mode decomposition to obtain the corresponding modal components includes: performing Hilbert transform on the acoustic signal sequence to calculate the cutoff frequency of the time-varying low-pass filter and the instantaneous mean of the acoustic signal sequence; based on the acoustic signal sequence, using the Tuna Swarm Optimization Algorithm (Tuna Swarm Optimization Algorithm) The invention relates to a method for obtaining a modal component of the acoustic signal sequence by using a time-varying low-pass filter (TSO) and a time-varying low-pass filter (TSO); inputting the acoustic signal sequence into the time-varying low-pass filter for filtering to obtain a plurality of component signals; based on the B-spline order, approximating each of the component signals by using B-spline interpolation to obtain a corresponding filtered component signal; calculating the ratio of the Loughlin instantaneous bandwidth to the instantaneous frequency of the filtered component signal as a judgment ratio; when the judgment ratio is greater than a preset threshold, determining the component signal corresponding to the judgment ratio as a modal component; when the ratio is less than or equal to the preset threshold, subtracting the instantaneous mean from the filtered component signal and re-performing modal decomposition until all the modal components of the acoustic signal sequence are determined.

[0006] In one embodiment of the present invention, the step of calculating the bandwidth threshold and B-spline order of the time-varying low-pass filter according to the tuna school optimization algorithm includes: decomposing the acoustic signal sequence according to preset initialization values ​​of the time-varying filter empirical mode decomposition method to obtain a plurality of corresponding initial modal components; calculating the information entropy of each of the initial modal components; constructing a fitness function of the optimization algorithm based on the information entropy; and using the fitness function to optimize the bandwidth threshold and B-spline order of the time-varying filter empirical mode decomposition using the tuna school optimization algorithm to obtain the optimal bandwidth threshold and B-spline order.

[0007] In one embodiment of the present invention, the step of performing singular value decomposition calculation and reconstruction on the modal components according to the energy proportion to obtain a denoised signal includes: calculating the energy proportion of each modal component, dividing the modal component into a primary modal component and a secondary modal component according to each energy proportion; performing singular value decomposition and reconstruction on the secondary modal component, and accumulating the secondary modal component with the primary modal component to obtain a denoised signal.

[0008] In one embodiment of the present invention, the step of decomposing and reconstructing the secondary modal component through singular value decomposition and accumulating it with the primary modal component to obtain a denoised signal includes: decomposing the secondary modal component through singular value decomposition to obtain a target singular value; inverse transforming the target singular value according to singular value decomposition to obtain a reconstructed modal component; and accumulating the primary modal component and the reconstructed modal component to obtain a denoised signal.

[0009] In one embodiment of the present invention, the step of obtaining a target singular value from the secondary modal component through singular value decomposition includes: performing singular value decomposition on all the secondary modal components to obtain multiple singular values; calculating a contribution rate difference spectrum based on all the singular values ​​obtained by the decomposition; determining a singular value boundary based on the contribution rate difference spectrum, and screening out a target singular value from all the singular values ​​according to the singular value boundary.

[0010] In one embodiment of the present invention, the step of inputting the denoised signal into a bolt fault diagnosis model to obtain a bolt fault diagnosis result includes: extracting the time domain features and frequency domain features of the denoised signal, and integrating the time domain features and the frequency domain features to obtain the original features; performing kernel principal component analysis on the original features to obtain reduced-dimensional acoustic signal features; and inputting the acoustic signal features into a bolt fault diagnosis model to obtain a bolt fault diagnosis result.

[0011] In one embodiment of the present invention, a bolt fault diagnosis system is also provided, which includes: a sampling module for obtaining a sound signal sequence of knocking bolts under working conditions; a modal decomposition module for decomposing the sound signal sequence according to time-varying filtering empirical mode decomposition to obtain corresponding modal components; a denoising module for performing singular value decomposition calculation and reconstruction on the modal components according to energy proportion to obtain a denoised signal; a prediction module for inputting the denoised signal into a bolt fault diagnosis model to obtain a bolt fault diagnosis result; wherein the bolt fault diagnosis model is an extreme random forest model.

[0012] In one embodiment of the present invention, an electronic device is also provided, 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 any of the above-mentioned bolt fault diagnosis methods based on acoustic signals.

[0013] In one embodiment of the present invention, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor of a computer, the computer executes the above-mentioned bolt fault diagnosis method based on acoustic signals.

[0014] As described above, the present invention provides a method, system, device, and medium for bolt fault diagnosis based on acoustic signals, which has the following beneficial effects: by combining time-varying filter empirical mode decomposition with singular value decomposition, efficient noise reduction of acoustic signals is achieved, effectively solving the problem of characteristic distortion caused by traditional methods when processing non-stationary signals. The tuna swarm optimization algorithm is used to adaptively determine the hyperparameters of the time-varying filter empirical mode decomposition using information entropy as the fitness function, thus avoiding the complexity of manual parameter adjustment and improving the degree of automation of signal processing. The use of the extreme random forest algorithm for fault classification not only has stronger generalization capabilities than traditional machine learning algorithms, but is also more efficient than deep learning methods, significantly improving detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic flow chart of a bolt fault diagnosis method based on acoustic signals provided by an embodiment of the present invention;

[0016] Figure 2 Time domain and frequency domain diagrams of noisy signals provided by embodiments of the present invention;

[0017] Figure 3 The time domain and frequency domain diagrams of the noisy signal after denoising provided by the embodiment of the present invention;

[0018] Figure 4 The time domain diagrams and frequency domain diagrams of the noisy signal before and after noise reduction provided by the embodiment of the present invention;

[0019] Figure 5 The confusion matrix of the test machine for the extreme random forest provided by the embodiment of the present invention;

[0020] Figure 6 Confusion matrix diagram of the random forest and support vector machine test sets provided by the embodiment of the present invention;

[0021] Figure 7 Shown is a structural block diagram of a bolt fault diagnosis system provided by an embodiment of the present invention;

[0022] Figure 8 Shown is a structural schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] 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 other 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.

[0024] It should be noted that the illustrations provided in the following embodiments are merely schematic illustrations of the basic concept of the present invention. Therefore, 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.

[0025] 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.

[0026] The present invention provides a bolt fault diagnosis method based on acoustic signals. The method involves obtaining an acoustic signal sequence within a preset sampling frequency; decomposing the acoustic signal sequence using time-varying filtered empirical mode decomposition (EMD) to obtain corresponding modal components. The optimal bandwidth threshold and B-spline order for the EMD are determined using a tuna swarm optimization algorithm. Singular value decomposition is then performed on the modal components based on their energy proportions, resulting in a denoised signal. The denoised signal is then input into a bolt fault diagnosis model to obtain a bolt fault diagnosis result. The bolt fault diagnosis model is an extreme random forest.

[0027] See Figure 1 , shows a flowchart of a bolt fault diagnosis method based on acoustic signals in an exemplary embodiment of the present application, comprising the following steps:

[0028] S100: Acquire a sound signal sequence of striking a bolt under working conditions.

[0029] Under actual working conditions, the sound signals generated by striking the bolts are collected at a fixed frequency of a preset sampling frequency, and the sound signal sequence is integrated in chronological order. In a preferred embodiment of the present invention, an acoustic sensor is selected to pick up the sound signals under working conditions at a fixed sampling frequency. The preferred value of the sampling frequency is 25000 Hz, and the sample data length is 8192 sampling points.

[0030] S200 , decomposing the acoustic signal sequence according to time-varying filtering empirical mode decomposition to obtain corresponding modal components.

[0031] The purpose of using time-varying filtered empirical mode decomposition is to decompose complex non-stationary signals, namely acoustic signals, into a series of simpler and easier-to-analyze intrinsic mode functions, namely modal components, thereby achieving signal simplification and feature extraction, effectively separating noise components from effective signal components, improving signal analyzability, adapting to the time-varying characteristics of non-stationary signals, and providing a data basis for subsequent signal processing, feature extraction, and fault diagnosis tasks. Compared with traditional empirical mode decomposition (EMD), time-varying filtered empirical mode decomposition significantly improves the decomposition robustness in low signal-to-noise ratio, non-stationary signal and low sampling rate scenarios. It not only effectively solves the problem of modal aliasing, but also ensures the accurate extraction of signal transient features through time-varying filtering technology, and ultimately outputs more physically meaningful intrinsic mode functions.

[0032] Specifically, in one embodiment of the present invention, the step of decomposing the acoustic signal sequence according to time-varying filtered empirical mode decomposition to obtain corresponding modal components includes:

[0033] S210 , performing Hilbert transform on the acoustic signal sequence to calculate the cutoff frequency of the time-varying low-pass filter and the instantaneous mean value of the acoustic signal sequence.

[0034] S211, calculating the instantaneous amplitude and instantaneous frequency of the acoustic signal sequence using Hilbert transform.

[0035] The acoustic signal sequence is transformed according to the following Hilbert transform formula to obtain the instantaneous amplitude and instantaneous frequency:

[0036]

[0037] Among them, A(t) is the instantaneous amplitude, t is the time sequence value, X(t) is the sound signal sequence, is the instantaneous frequency, is the Hilbert transform of the acoustic signal sequence X(t).

[0038] S212, determining a local maximum sequence and a local minimum sequence of the instantaneous amplitude.

[0039] According to the neighborhood comparison principle, the local maximum sequence and local minimum sequence of the instantaneous amplitude are determined.

[0040] S213: Interpolate the local maximum sequence to obtain a first intermediate parameter, interpolate the local minimum sequence in the same way to obtain a second intermediate parameter, and calculate the instantaneous mean and instantaneous envelope based on the first intermediate parameter and the second intermediate parameter.

[0041] The calculation formulas for the instantaneous mean and instantaneous envelope are as follows:

[0042] Where a1(t) is the instantaneous mean of the acoustic signal sequence, a2(t) is the instantaneous envelope of the acoustic signal sequence, t is the time series value, β1(t) is the first intermediate parameter, and β2(t) is the second intermediate parameter.

[0043] S214: Calculate the maximum parameter representation and the minimum parameter representation based on the local maximum sequence and the local minimum sequence.

[0044] Calculate the instantaneous amplitude and instantaneous frequency of the local maximum sequence and obtain the following maximum parameter representation:

[0045] Where A({t max}) is the maximum instantaneous amplitude, is the maximum instantaneous frequency, t max is the maximum time series value.

[0046] Calculate the instantaneous amplitude and instantaneous frequency of the local minimum sequence and obtain the following minimum parameter representation:

[0047] Where A({t min}) is the minimum instantaneous amplitude, is the minimum instantaneous frequency, t min is the minimum time series value.

[0048] According to the maximum parameter representation and the minimum parameter representation, the maximum frequency component and the minimum frequency component are calculated:

[0049] First, the maximum parameter representation is interpolated to obtain the maximum interpolation parameter representation, and the minimum parameter representation is interpolated to obtain the minimum interpolation parameter representation: η2(t).

[0050] Secondly, the maximum frequency component and the minimum frequency component are calculated according to the following formula:

[0051]

[0052]

[0053] in is the maximum instantaneous frequency component, η1(t) is the maximum interpolation parameter representation, η2(t) is the maximum interpolation parameter representation, is the instantaneous frequency component of the maximum value, a1(t) is the instantaneous mean of the acoustic signal sequence, and a2(t) is the instantaneous envelope of the acoustic signal sequence.

[0054] S215, calculating the cutoff frequency of the time-varying low-pass filter based on the maximum instantaneous frequency component and the minimum instantaneous frequency component, using the following calculation formula:

[0055] in, is the maximum instantaneous frequency component, is the maximum instantaneous frequency component, is the cutoff frequency of the time-varying low-pass filter.

[0056] S220, based on the acoustic signal sequence, using a tuna school optimization algorithm to calculate the bandwidth threshold and B-spline order of the time-varying low-pass filter;

[0057] Furthermore, in one embodiment of the present invention, the step of calculating the bandwidth threshold and B-spline order of the time-varying low-pass filter according to the tuna school optimization algorithm includes:

[0058] S221 , decomposing the acoustic signal sequence according to the preset initialization values ​​of the time-varying filtering empirical mode decomposition method to obtain a plurality of corresponding initial modal components.

[0059] S222, calculating the corresponding information entropy according to the initial modal components using the information entropy formula;

[0060] For any initial modal component, the expression of its information entropy is as follows:

[0061] Among them, x i (t) is the i-th initial modal component, p i is the ratio of the energy of the i-th modal component to the total energy, n is the number of data for each initial modal component, K is the number of initial modal components, and t is the time series value.

[0062] S223, constructing a fitness function of the optimization algorithm according to the information entropy.

[0063] The present invention uses information entropy to construct the fitness function of the optimization algorithm, that is, the sum of the information entropy of all initial modal components is used as the optimization target, and the calculation formula is as follows:

[0064] Among them, K is the number of initial modal components, and Fitness is the optimal parameter combination that can effectively separate signal features by minimizing the entropy value, that is, the optimal bandwidth threshold and B-spline order.

[0065] S224 , using the fitness function and a tuna school optimization algorithm, optimizing the bandwidth threshold and B-spline order of the time-varying filter empirical mode decomposition to obtain the optimal bandwidth threshold and B-spline order.

[0066] The Tuna School Optimization Algorithm (Tunny School Optimization Algorithm) is a metaheuristic global optimization algorithm based on swarm behavior. It simulates two foraging strategies of tuna schools: spiral foraging and parabolic foraging. It achieves global optimization by dynamically switching between these strategies. This algorithm features a small number of parameters, fast convergence, and resistance to local extremes, making it widely used in intelligent optimization problems.

[0067] Spiral foraging: When a school of tuna spots prey, they swim in a spiral pattern to approach it. The formula is as follows:

[0068]

[0069]

[0070]

[0071] β=e bl cos(2πb),

[0072]

[0073] in, is the i-th individual in the r+1-th iteration, is the optimal individual in the current population in the r+1th iteration, that is, the food position. α1 is the weight coefficient that controls the current individual to move toward the optimal individual. α2 is the weight coefficient that controls the current individual to move toward the previous individual. α is a constant used to determine the tendency of the individual to follow the optimal individual and the previous individual. b is a random number between 0 and 1. r is the number of current iterations. max is the maximum number of iterations, and l is a parameter.

[0074] When the optimal individual cannot find food (global optimal), blindly following the optimal individual is not conducive to group foraging. Therefore, generating random coordinates in the search space as reference points for spiral foraging is conducive to jumping out of the local optimal state and giving the algorithm global optimization capabilities.

[0075]

[0076] i=1,2,3…,m; where is a randomly generated reference point in the search space in the rth iteration, r max is the maximum number of iterations, and m is the total number of individuals in this iteration.

[0077] Parabolic foraging: When prey is difficult to lock onto, tuna schools surround the prey in a parabolic pattern, using the following formula:

[0078] Among them, TF is a random number, taking the value of 1 or -1, and p is the preset probability.

[0079] In summary, the mathematical expression of the tuna foraging mathematical model can be summarized as follows:

[0080]

[0081] Tuna schools approach prey using two foraging methods, each with a 50% probability. Through an iterative process, they continuously update their optimal individual positions, approaching their prey and ultimately capturing it. The tuna school optimization algorithm of the present invention terminates when the maximum number of iterations is reached, outputting the optimal individual position and terminating the iteration. If the minimum information entropy of all initial modal components is zero, the iteration is terminated early and the optimal individual position is output. If the minimum information entropy of all initial modal components is not zero and the maximum number of iterations has not been reached, iterations continue until the minimum information entropy of the initial modal components is zero or the maximum number of iterations is reached.

[0082] The B-spline order and bandwidth threshold have a significant impact on decomposition performance. Comparing the decomposition performance of acoustic signal sequences under various conditions, the optimal B-spline order range is [5, 15], which provides excellent decomposition results. When the B-spline order is greater than 15, the envelope fitting curve becomes overly smooth, which can lead to the loss of transient impulse components. When the B-spline order is less than 2, the envelope fitting curve becomes overly sensitive, which can easily introduce spurious modes. A bandwidth threshold dynamic range of [0.05-0.2] provides the best results. A bandwidth threshold less than 0.05 suppresses high-frequency noise and prolongs decomposition time. A bandwidth threshold greater than 0.2 poses the risk of residual noise.

[0083] S230: Input the acoustic signal sequence into a time-varying low-pass filter for filtering to obtain a plurality of component signals.

[0084] S240 , based on the B-spline order, approximate each component signal by using B-spline interpolation to obtain a corresponding filtered component signal.

[0085] S250 , calculating a ratio of the Loughlin instantaneous bandwidth to the instantaneous frequency of the filtered component signal as a judgment ratio.

[0086] The judgment ratio of each filtered component signal is calculated according to the following formula:

[0087] Among them, θ(t) is the judgment ratio, is the instantaneous frequency of the filtered component signal, t is the time series value, B Loughlin (t) is the Loughlin instantaneous bandwidth calculated based on the filtered component signal, and the calculation formula is as follows:

[0088] Where A(t) is the instantaneous amplitude of the filtered component signal.

[0089] S251, when the judgment ratio is greater than a preset threshold, determining the component signal corresponding to the judgment ratio as a modal component;

[0090] S252: When the ratio is less than or equal to a preset threshold, subtracting the instantaneous mean from the filtered component signal and performing modal decomposition again until all the modal components of the acoustic signal sequence are determined.

[0091] S300: Perform singular value decomposition calculation and reconstruction on the modal components according to the energy proportion to obtain a denoised signal.

[0092] Specifically, in one embodiment of the present invention, the step of performing singular value decomposition calculation and reconstruction on the modal components according to energy proportions to obtain a denoised signal includes:

[0093] S310 , calculating the energy proportion of each modal component, and dividing the modal component into a primary modal component and a secondary modal component according to each energy proportion.

[0094] S320 , decomposing and reconstructing the secondary modal component through singular value decomposition, and adding the secondary modal component to the primary modal component to obtain a denoised signal.

[0095] S321: Decompose the secondary modal components by singular value decomposition to obtain target singular values.

[0096] First, singular value decomposition is performed on all secondary modal components to obtain multiple singular values.

[0097] Singular value decomposition (SVD) is a matrix decomposition algorithm for time series signals. The Hankel matrix is ​​often used as the construction method for SVD. The acoustic signal sequence generated by the impact is a one-dimensional signal. SVD first converts it into a two-dimensional signal. The Hankel matrix for the noisy signal can then be constructed as follows:

[0098] Where M is the number of signal data, γ + ε = M, and x(·) is a single signal value of the one-dimensional signal. In a preferred embodiment of the present invention, γ = [M / 2] and ε = M + 1 - γ. This value achieves the best SVD noise reduction effect.

[0099] The SVD decomposition of the matrix H can be expressed as follows:

[0100] Where U and V are both orthogonal matrices, Σ is a diagonal matrix whose diagonal elements are composed of singular values, namely: σ1, σ2…σ k , and σ1>σ2>…>σ k >0, indicating that the k singular values ​​obtained after singular value decomposition are all greater than 0, ui and v i are the i-th column vectors of matrices U and V respectively; R is the rank of matrix H.

[0101] Secondly, the contribution rate difference spectrum is calculated based on all the singular values ​​obtained by decomposition.

[0102] Finally, the singular value boundary is determined according to the contribution rate difference spectrum, and the target singular value is screened out from all singular values ​​according to the singular value boundary.

[0103] Using Hankel matrix decomposition to decompose singular values ​​can effectively reflect the characteristics of the corresponding components according to the size of the singular values. Usually, the singular values ​​corresponding to the signal components are larger, while the singular values ​​corresponding to the noise components are smaller. Therefore, the singular value boundaries are determined based on the contribution rate difference spectrum.

[0104] Arrange the singular values ​​in descending order and calculate the singular value contribution rate according to the following singular value contribution rate formula:

[0105] δ i =σ i / σ sum -σ i+1 / σ sum , where δ i is the contribution rate of the i-th singular value, σ i is the i-th singular value, σ sum is the sum of all singular values.

[0106] The contribution rate difference spectrum is obtained according to the obtained singular value contribution rate, and the position of the maximum mutation point is found. The singular value before the mutation position is retained, and the singular value corresponding to the remaining noise signal is set to 0. The retained singular value is the target singular value.

[0107] S322 , inversely transform the target singular value according to singular value decomposition to obtain a reconstructed modal component.

[0108] S323: Accumulate the main modal component and the reconstructed modal component to obtain a denoised signal.

[0109] Furthermore, in one embodiment of the present invention, the noise reduction effect is evaluated by introducing the signal-to-noise ratio and mean square error of the signal.

[0110] Generate a simulated signal as the original signal, add Gaussian white noise to the original signal as the noisy signal. Decompose the noisy signal using the time-varying filtering empirical mode decomposition of the present invention to obtain the corresponding modal components, perform singular value decomposition calculation and reconstruction on the modal components according to the energy proportion, and denoise the original signal to obtain the noise-reduced signal for verification:

[0111] like Figure 2As shown, the left figure is the time domain diagram of the noisy signal, and the right figure is the frequency domain diagram of the noisy signal. Figure 3 As shown, the left picture is the time domain picture of the noisy signal after denoising, and the right picture is the frequency domain picture of the noisy signal after denoising. It can be clearly seen from the pictures that the time domain waveform of the noisy signal is more chaotic, and there are more noises in the frequency domain picture. The noise of the noisy signal processed by the denoising method of this application is significantly filtered out. The signal-to-noise ratio and mean square error of the noisy signal before and after denoising can be used to obtain the following index comparison table before and after denoising. Among them, the signal-to-noise ratio of the denoised signal is more than three times that of the noisy signal, and the mean square error is much smaller than that of the noisy signal. This shows that the denoising method using time-varying filtering empirical mode decomposition and singular value decomposition reconstruction can achieve significant noise reduction effect.

[0112] Table 1 Comparison of indicators before and after noise reduction

[0113]

[0114] In a preferred embodiment of the present invention, taking the acoustic signal collected under the condition of the bolt pre-tightening force of 55 MPa as an example, the time domain diagram and spectrum diagram before and after noise reduction are as follows: Figure 4 As shown:

[0115] The left image in the time domain diagram before and after noise reduction shows the acoustic signal before noise reduction, while the right image shows the acoustic signal after noise reduction. A comparison shows that the spectrum in the time domain diagram is smoother and has less noise after noise reduction. The left image shows the amplitude spectrum of the acoustic signal before noise reduction, while the right image shows the amplitude spectrum of the acoustic signal after noise reduction. The vertical axis is amplitude, and the horizontal axis is frequency (Hz). The frequency domain diagrams before and after noise reduction show that the frequency domain diagram after noise reduction is clearer and has fewer noise waves, indicating a significant noise reduction effect.

[0116] From the above two methods, it can be observed that the signal after noise reduction is purer than the original signal, and the signal after noise reduction is smoother. The corresponding features extracted from this will be more prominent, which can have a positive impact on the subsequent model establishment.

[0117] S400: Input the denoised signal into a bolt fault diagnosis model to obtain a bolt fault diagnosis result; wherein the bolt fault diagnosis model is an extreme random forest model.

[0118] The diagnostic process of the bolt fault diagnosis model is as follows:

[0119] S410: Extract the denoised signal and perform time series and frequency domain features to obtain a feature vector.

[0120] S420, performing feature fusion and data dimensionality reduction on the feature vectors using kernel principal component analysis, and constructing a feature data set.

[0121] Kernel principal component analysis is used to perform feature fusion and data dimensionality reduction on the feature vector to obtain a low-dimensional vector.

[0122] When bolt signals are collected, the same fault information may be present within a short period of time under the same operating conditions, resulting in a large amount of information overlap. Considering that actual collected signals often contain complex nonlinear relationships, kernel principal component analysis is used to compress and reduce the dimensionality of the feature vectors.

[0123] Kernel principal component analysis (KPCA) is a nonlinear generalization of principal component analysis based on the kernelization concept. Its principle is to map linearly inseparable original data points in low-dimensional space to kernel space, a higher-dimensional space, through a nonlinear mapping function, thereby making the data linearly separable. KPCA is used to extract nonlinear features from high-dimensional feature vectors using principal component analysis, perform feature fusion, and reduce data dimensionality. The number of principal components is determined by the cumulative contribution rate, and the reduced feature vector is obtained to construct a feature vector dataset. In a preferred embodiment of the present invention, the cumulative contribution rate of the features is set to be greater than or equal to 95%, and the number of principal components of the feature vector is determined to be 10.

[0124] Firstly, kernel principal component analysis is used to reduce the dimension of the feature vector, eliminate redundant information, and extract effective feature indicators.

[0125] The principle of kernel principal component analysis is as follows:

[0126] The covariance matrix of the high-dimensional feature space can be expressed as:

[0127] Where C is the covariance matrix, m is the number of samples, n is the sample dimension, φ(·) is the mapping function, and x i For the sample set X={x1,x1,…,x i}(x i ∈R n The eigenvalues ​​and eigenvectors of the covariance matrix need to satisfy the following condition: Cv = λv, where λ is the eigenvalue of the covariance matrix and v is the eigenvector of the covariance matrix.

[0128] Multiplying both sides of the conditional expression by the mapping function, we can get:

[0129] φ(x k )Cv=λφ(x k )v, where φ(x k ) is the kth sample value x k The mapping function of . The eigenvector is linearly represented by the mapping function, and the eigenvector can be expressed as: Among them, α i is the i-th coefficient, and from the above we can get:

[0130] Where j is the count value.

[0131] Secondly, the denoised signal is mapped to the kernel space to obtain the kernel space data. Based on the kernel space data and the Gaussian kernel function, principal component analysis is used to obtain the kernel matrix.

[0132] Select the kernel function. Common kernel functions include polynomial kernel, Gaussian kernel, Laplace kernel, etc. Among them, the Gaussian kernel function has few parameters and is widely used. Therefore, the Gaussian kernel is selected as the kernel function of kernel principal component analysis. Its expression is:

[0133] where σ' 2 is the bandwidth parameter of the Gaussian kernel function, x, x i are two input vectors respectively.

[0134] A symmetric positive definite kernel matrix is ​​obtained by calculating the kernel function, and the symmetric positive definite kernel matrix is ​​decentralized to obtain: Among them, K is the kernel matrix, L n is an n×n matrix, and the elements in the matrix are

[0135] Then, according to the matrix eigenvalues, eigenvectors and main elements of the kernel matrix.

[0136] Then the kernel matrix expression can be expressed as: K'α=nλα, and the eigenvalues ​​and eigenvectors of the kernel matrix K are calculated according to the formula K'. The eigenvalues ​​are: λ1≥λ2≥…≥λ m , the eigenvector is normalized, and the kth linear principal element of the sample set X is obtained as: Among them, h k refers to the kth linear principal element of the sample, v k is the normalized eigenvector of the covariance matrix C, k = 1, 2, ..., m, is the i-th coefficient corresponding to the k-th sample.

[0137] Finally, the contribution rate of each principal element is calculated based on the matrix eigenvalues ​​and eigenvectors, the principal elements whose contribution rates exceed a preset contribution rate threshold are retained, and the corresponding acoustic signal features after dimensionality reduction are determined based on the retained principal elements.

[0138] The present invention selects and calculates the cumulative contribution rate of characteristic indicators:

[0139] p is the number of principal elements with high contribution rate, and the contribution rate ranges from 80% to 98%. After calculation, it is determined that the number of principal elements of the eigenvector p is 10.

[0140] S430: Input the acoustic signal characteristics into a bolt fault diagnosis model to obtain a bolt fault diagnosis result.

[0141] The acoustic signal features are input into the bolt fault diagnosis model, and the acoustic signal features are input into the extreme random forest to obtain the average value of the classification probability regression values ​​output by all decision trees. The fault type of the bolt is determined based on the average value. In the present invention, the faulty bolt types are divided into 4 categories, namely: pre-tightening pressure of 40-45MPa is severe looseness, pre-tightening pressure of 45-50MPa is moderate looseness, pre-tightening pressure of 50-55MPa is mild looseness, and pre-tightening pressure of 55MPa or above is normal. The model output is the fault diagnosis result of the bolt, namely: severe looseness, moderate looseness, mild looseness or normal.

[0142] Furthermore, in one embodiment of the present invention, the steps of the extreme random forest construction process include:

[0143] The obtained feature vector is used as the sample set.

[0144] Construct a specified number of decision trees based on the sample set to obtain an extreme random forest. The decision tree generation process is as follows:

[0145] For a tree: Based on the i-th feature of the sample set, a split value is randomly selected between its maximum and minimum values. Samples less than the split value are classified as the left leaf node, and all samples greater than the split value are classified as the right leaf node. Each leaf node continues to randomly select split values ​​for the remaining features and grows downward until the split condition is reached. The splitting stop condition is: the decision tree reaches the maximum tree depth or the number of samples at the node is less than the minimum number of samples.

[0146] The final output of the trained extreme random forest is the average of the classification probabilities output by all decision trees. That is, the final classification probability of each sample is the average probability of all decision trees, which is expressed as:

[0147] Among them, f i is the i-th sample, L is the total number of decision trees, P I (c|f i ) is the first tree, in the sample f i The probability of belonging to category c, P(c|f i ) is the value in sample f i The average of the total probabilities of belonging to class c.

[0148] Furthermore, the present invention verifies the effectiveness of the bolt fault diagnosis method based on acoustic signals, and the verification process is as follows:

[0149] First, obtain the test data.

[0150] Data acquisition was performed using a bolt test bench, yielding the bolt test data shown in Table 2 below. The test bench primarily consists of a test flange tank, LMS acquisition equipment, a laptop computer, an acoustic sensor, and an excitation hammer. 8.8-grade M30 studs and nuts were used. To simulate loosening failures, the bolts were tightened to four different levels: 40 MPa, 45 MPa, 50 MPa, and 55 MPa, with preloads of 141.3 kN, 159 kN, 176.6 kN, and 195 kN, respectively. The excitation bolts generated acoustic signals, which were picked up by the acoustic sensor to generate a set of acoustic signals. The sampling frequency for the signals was set to 25,000 Hz.

[0151] Table 2 Bolt test data

[0152]

[0153]

[0154] Then, the obtained data is used to verify the effect of the bolt fault diagnosis method of the present invention.

[0155] The data was decomposed through time-varying filter empirical mode decomposition, singular value decomposition based on energy proportion, and reconstructed to obtain a denoised signal. This was then input into the bolt fault diagnosis model to extract the feature vector. 80% of each feature in the feature vector was accumulated to form the training set, and the remaining 20% ​​formed the test set. After multiple rounds of testing, the relevant parameters for the extreme random forest were set as follows: the number of decision trees was 50, the number of attributes randomly selected for each node was 4; the minimum sample size for splitting a node was 10; and the confusion matrix of the test set for this model is shown in the figure below. Figure 5 As shown in the figure, in the model, the different working condition data obtained by tightening the bolts to four different degrees of 40Mpa, 45Mpa, 50Mpa, and 55Mpa using a bolt tensioner are labeled 1, 2, 3, and 4 respectively. In the diagnosis results of the entire bolt fault diagnosis model, only label 1 was mistakenly diagnosed as label 2 once, and label 3 was mistakenly diagnosed as label 1 once. The total accuracy rate is 98.75%, which is a good effect.

[0156] In order to verify the advantages of extreme random forest diagnosis, the present invention chooses to compare with random forest and support vector machine based on the same test data, and the confusion matrix diagram obtained is as follows Figure 6As shown, the left figure is the confusion matrix diagram of the random forest test set, with an accuracy of 93.125%, and the right figure is the confusion matrix diagram of the support vector machine test set. The test set confusion matrix of the comparison model is as follows, with an accuracy of 91.125%, both of which are lower than the diagnostic accuracy of the extreme random forest. The classification accuracy of the extreme random forest is about 5% higher than that of the random forest and about 7% higher than that of the support vector machine, which shows that it has certain superiority.

[0157] See Figure 7 The bolt fault diagnosis system 200 includes: a sampling module 210, a modal decomposition module 220, a denoising module 230, and a prediction module 240, wherein the sampling module 210 is used to obtain an acoustic signal sequence within a preset sampling frequency; the modal decomposition module 220 is used to decompose the acoustic signal sequence according to time-varying filtering empirical mode decomposition to obtain corresponding modal components; the denoising module 230 is used to perform singular value decomposition calculation and reconstruction on the modal components according to energy proportion to obtain a denoised signal; the prediction module 240 is used to input the denoised signal into a bolt fault diagnosis model to obtain a bolt fault diagnosis result; wherein the bolt fault diagnosis model is an extreme random forest model.

[0158] The specific definition of the bolt fault diagnosis system can be found in the definition of the bolt fault diagnosis method based on acoustic signals above, which will not be repeated here. Each module in the above bolt fault diagnosis system can be implemented in whole or in part through software, hardware or a combination thereof.

[0159] The above modules may be embedded in or independent of the processor in the computer device in hardware format, or may be stored in the memory of the computer device in software format, so that the processor can call the corresponding operations of the above modules.

[0160] It should be noted that, in order to highlight the innovative part of the present invention, this embodiment does not introduce modules that are not closely related to solving the technical problem proposed by the present invention, but this does not mean that there are no other modules in this embodiment.

[0161] like Figure 8 As shown, the electronic device 3 may include a memory 32, a processor 31 and a bus, and may also include a computer program stored in the memory 32 and executable on the processor 31, such as a time-varying filtering empirical mode decomposition calculation program.

[0162] The memory 32 includes at least one type of readable storage medium, including a flash memory, a mobile hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the memory 32 may be an internal storage unit of the electronic device 3, such as a mobile hard disk of the electronic device 3. In other embodiments, the memory 32 may also be an external storage device of the electronic device 3, such as a plug-in mobile hard disk, a smart memory card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 3. Furthermore, the memory 32 may include both an internal storage unit and an external storage device of the electronic device 3. The memory 32 may be used not only to store application software installed in the electronic device 3 and various types of data, such as the code for calculating and reconstructing the singular value decomposition of modal components, but also to temporarily store data that has been output or is about to be output.

[0163] In some embodiments, the processor 31 may be composed of an integrated circuit, such as a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and a combination of various control chips. The processor 31 is the control core (Control Unit) of the electronic device 3. It utilizes various interfaces and circuits to connect the various components of the entire electronic device 3. It executes or runs programs or modules stored in the memory 32 (such as a correction program for singular value decomposition calculation and reconstruction of modal components) and calls data stored in the memory 32 to perform various functions of the electronic device 3 and process data.

[0164] The processor 31 executes the operating system and various installed application programs of the electronic device 3. The processor 31 executes the application programs to implement the steps in the above-mentioned bolt fault diagnosis method based on acoustic signals.

[0165] Exemplarily, the computer program may be divided into one or more modules, which are stored in the memory 32 and executed by the processor 31 to complete the present application. The one or more modules may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the electronic device 3.

[0166] The above-mentioned integrated unit implemented in the form of a software functional module can be stored in a computer-readable storage medium, which can be non-volatile or volatile. The above-mentioned software functional module is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, computer device, or network device, etc.) or a processor to perform some of the functions of the bolt fault diagnosis method based on acoustic signals described in various embodiments of the present application.

[0167] In summary, the present invention discloses a bolt fault diagnosis method, system, device, and medium based on acoustic signals, which can acquire the required signals without loss. During fault analysis, the system avoids the problem of fault feature distortion and deterioration of information quality after processing caused by traditional noise reduction methods when processing non-stationary signals, significantly improving the noise reduction effect of acoustic signals under working conditions. At the same time, compared with traditional deep learning models, extreme random forests have fewer hyperparameters and are easy to adjust. At the same time, they can effectively improve the efficiency and accuracy of bolt working condition diagnosis. Therefore, the present invention effectively overcomes the various shortcomings of the existing technology and has high industrial utilization value.

[0168] 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, all 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 bolt fault diagnosis method based on acoustic signals, characterized in that: The method comprises: Obtain the acoustic signal sequence of the bolt being struck under working conditions; Decomposing the acoustic signal sequence according to time-varying filtering empirical mode decomposition to obtain corresponding modal components; According to the energy proportion, singular value decomposition calculation and reconstruction are performed on the modal components to obtain a denoised signal; The denoised signal is input into a bolt fault diagnosis model to obtain a bolt fault diagnosis result; wherein the bolt fault diagnosis model is an extreme random forest model.

2. The bolt fault diagnosis method based on acoustic signals according to claim 1, characterized in that: The step of decomposing the acoustic signal sequence according to time-varying filtering empirical mode decomposition to obtain corresponding modal components includes: Performing a Hilbert transform on the acoustic signal sequence to calculate a cutoff frequency of a time-varying low-pass filter and an instantaneous mean value of the acoustic signal sequence; Based on the acoustic signal sequence, a tuna school optimization algorithm is used to calculate the bandwidth threshold and B-spline order of the time-varying low-pass filter; Inputting the acoustic signal sequence into the time-varying low-pass filter for filtering to obtain multiple component signals; Based on the B-spline order, approximating each of the component signals using B-spline interpolation to obtain a corresponding filtered component signal; The ratio of the Loughlin instantaneous bandwidth and the instantaneous frequency of the filtered component signal is calculated as the judgment ratio: When the judgment ratio is greater than a preset threshold, the component signal corresponding to the judgment ratio is determined as a modal component; When the ratio is less than or equal to a preset threshold, the instantaneous mean value is subtracted from the filtered component signal, and modal decomposition is performed again until all the modal components of the acoustic signal sequence are determined.

3. The bolt fault diagnosis method based on acoustic signals according to claim 2, characterized in that: The step of calculating the bandwidth threshold and B-spline order of the time-varying low-pass filter according to the tuna school optimization algorithm includes: Decomposing the acoustic signal sequence according to a preset initialization value of a time-varying filtering empirical mode decomposition method to obtain a plurality of corresponding initial modal components; Calculating the information entropy of each of the initial modal components; Constructing a fitness function of an optimization algorithm according to the information entropy; The bandwidth threshold and B-spline order of time-varying filter empirical mode decomposition are optimized by using the fitness function and the tuna school optimization algorithm, and the optimal bandwidth threshold and B-spline order are obtained.

4. The bolt fault diagnosis method based on acoustic signals according to claim 1, characterized in that: The step of performing singular value decomposition calculation and reconstruction on the modal components according to the energy proportion to obtain a denoised signal includes: Calculating the energy proportion of each modal component, and dividing the modal component into a primary modal component and a secondary modal component according to the energy proportion; The secondary modal component is decomposed and reconstructed through singular value decomposition, and is accumulated with the primary modal component to obtain a denoised signal.

5. The bolt fault diagnosis method based on acoustic signals according to claim 4 is characterized in that: The step of decomposing and reconstructing the secondary modal component through singular value decomposition and accumulating the secondary modal component with the primary modal component to obtain a denoised signal includes: Decomposing the secondary modal component by singular value decomposition to obtain target singular values; Inversely transforming the target singular value according to singular value decomposition to obtain a reconstructed modal component; The main modal component and the reconstructed modal component are accumulated to obtain a denoised signal.

6. The bolt fault diagnosis method based on acoustic signals according to claim 5, characterized in that: The step of obtaining a target singular value by performing singular value decomposition on the secondary modal component comprises: Performing singular value decomposition on all of the secondary modal components to obtain multiple singular values; Calculating a contribution rate difference spectrum based on all the singular values ​​obtained by decomposition; A singular value boundary is determined according to the contribution rate difference spectrum, and a target singular value is screened out from all the singular values ​​according to the singular value boundary.

7. The bolt fault diagnosis method based on acoustic signals according to claim 1, characterized in that: The step of inputting the denoised signal into a bolt fault diagnosis model to obtain a bolt fault diagnosis result comprises: Extracting time domain features and frequency domain features of the denoised signal, and integrating the time domain features and the frequency domain features to obtain original features; The original features are analyzed by kernel principal component analysis to obtain the acoustic signal features after dimensionality reduction; The acoustic signal characteristics are input into a bolt fault diagnosis model to obtain a bolt fault diagnosis result.

8. A bolt fault diagnosis system, characterized in that: The system comprises: A sampling module is used to obtain the acoustic signal sequence of the bolt being struck under working conditions; A modal decomposition module, configured to decompose the acoustic signal sequence according to time-varying filtering empirical mode decomposition to obtain corresponding modal components; A denoising module is used to perform singular value decomposition calculation and reconstruction on the modal components according to energy proportion to obtain a denoised signal; A prediction module is used to input the denoised signal into a bolt fault diagnosis model to obtain a bolt fault diagnosis result; wherein the bolt fault diagnosis model is an extreme random forest model.

9. An electronic device, characterized in that: The electronic device comprises: one or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, enables the electronic device to implement the bolt fault diagnosis method based on acoustic signals as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by a processor of a computer, the computer is caused to execute the bolt fault diagnosis method based on acoustic signals according to any one of claims 1 to 7.

Images