Fault diagnosis method based on acoustic signal of carrier roller bearing of belt conveyor

Abstract

The invention discloses a fault diagnosis method based on acoustic signals of a carrier roller bearing of a belt conveyor, and the method comprises the steps: carrying out the self-adaptive global optimization of key parameters in a variational mode decomposition algorithm through employing a raccoon optimization algorithm, obtaining a global optimal solution, feeding back the global optimal solution to the variational mode decomposition algorithm, and obtaining the acoustic signals of the carrier roller bearing of the belt conveyor. Carrying out adaptive signal decomposition on the acquired carrier roller bearing acoustic signal by using the algorithm to obtain an IMF; screening components in the IMF by using the comprehensive index to obtain an optimal IMF component, and calculating a time domain feature value of the optimal IMF component to form a denoised signal feature vector; and optimizing hyper-parameters of the convolutional neural network by using a raccoon optimization algorithm to obtain an optimized convolutional neural network fault diagnosis model, and inputting the denoised signal feature vector into the model for fault classification diagnosis to obtain a diagnosis result. According to the method, the acoustic signal of the carrier roller bearing can be decomposed in a self-adaptive manner, the fault features are extracted, and the problems of mode aliasing and the like are effectively solved.

Description

Technical Field

[0001] The invention relates to the technical field of bearing fault diagnosis, and in particular to a fault diagnosis method based on acoustic signals of belt conveyor roller bearings. Background Art

[0002] Belt conveyors are important continuous conveying equipment in the process of coal mining in mines. Among them, rollers are the most commonly used rotating parts in mining belt conveyors, and bearings are important components of rollers. Their operating status directly affects the performance, operating efficiency and service life of the equipment. However, most of the fault diagnosis of roller bearings nowadays uses their vibration signals, and a large number of sensors need to be arranged to obtain vibration data, which is time-consuming and labor-intensive. Therefore, using the sound signals generated during its working process for fault diagnosis can not only solve the problem of non-contact long-distance fault diagnosis, but also reduce the probability of accidents of mechanical equipment, and provide reliable decision support for the later maintenance plan of the equipment, which has high practical significance. Since the rolling bearing sound signal is non-stationary, nonlinear, and susceptible to environmental noise interference, it brings difficulties to the extraction of fault information. The VMD (variational mode decomposition) proposed by Dragomiretskiy can decompose the input signal into a series of intrinsic mode functions with limited bandwidth and certain frequency. However, the excellent performance of VMD depends on the appropriate parameter selection. Inappropriate preset values ​​of the mode number K will lead to over-decomposition or under-decomposition. Zhao et al. introduced improved complete EMD (Empirical mode decomposition) with adaptive noise and MOMOEDA (Multipoint Optimal Minimum Entropy Deconvolution Adjusted) to reduce strong noise interference and extract defect features from rolling bearing signals. However, MOMEDA also has parameter selection problems. How to determine the parameter period T and filter length L is a problem that needs to be considered. Inappropriate parameter settings have a significant impact on the results of signal decomposition and diagnostic classification. The selection of these parameter values ​​directly affects the stability and diagnostic accuracy of the diagnostic model. In order to solve this impact, a fault diagnosis method based on the acoustic signal of belt conveyor roller bearings is proposed to solve the above problems. Summary of the invention

[0003] The purpose of the present invention is to provide a fault diagnosis method based on acoustic signals of belt conveyor roller bearings, which can optimize the key parameters in VMD and the hyperparameters in convolutional neural networks, thereby improving the overall diagnostic accuracy and efficiency.

[0004] The present invention adopts the following technical solution:

[0005] A fault diagnosis method based on acoustic signals of belt conveyor roller bearings comprises the following steps:

[0006] S1. Collecting acoustic signals of roller bearings using sound sensors;

[0007] S2. Use COA (Coati Optimization Algorithm) to adaptively and globally optimize the key parameters in the VMD algorithm to obtain the global optimal solution; the key parameters include the number of modes K and the penalty factor α.

[0008] S3. Feeding back the key parameters at the end of the optimization to the VMD algorithm, using the algorithm to perform adaptive signal decomposition on the acoustic signal in step S1 to obtain the corresponding IMF (Intrinsic Mode Function).

[0009] S4. Use comprehensive indicators to screen the components in the IMF, and take the component with the largest comprehensive indicator as the optimal IMF component.

[0010] S5. Calculate the time domain eigenvalue of the optimal IMF component to form a signal eigenvector after noise reduction.

[0011] S6. Use COA to optimize the hyperparameters of the convolutional neural network to obtain the optimized CNN (Convolutional Neural Networks) fault diagnosis model.

[0012] S7. Input the signal feature vector after noise reduction into the optimized CNN fault diagnosis model to perform fault classification diagnosis and obtain the diagnosis result.

[0013] Furthermore, in step S1, the sound sensor collects acoustic signals of the roller bearings of the belt conveyor when it is unloaded in a normal state and a fault state respectively.

[0014] Furthermore, in step S2, the adaptive global optimization of key parameters includes the following contents:

[0015] S201, setting the parameters of COA, including the population size, the maximum number of iterations, and the optimization upper and lower bounds.

[0016] S202, initialize the population and calculate the fitness values ​​of key parameters in the VMD algorithm.

[0017] S203. Update the locations of all raccoons and iguanas according to the basic principles of COA.

[0018] S204, determine whether the key parameters in the VMD algorithm exceed the set boundaries; if exceeded, return to step S203 and re-operate; if not exceeded, enter step S205.

[0019] S205, modifying key parameters and fitness values ​​in the VMD algorithm.

[0020] S206. Repeat steps S202-S205 until the termination criterion is met, and the global optimal solution [K, α] is obtained.

[0021] Furthermore, in step S4, the comprehensive index includes the mutual correlation coefficient, the permutation entropy and the mutual information entropy, and the specific expressions are as follows:

[0022]

[0023]

[0024] MI(x i ,y i )=H(y i )-H(y i |x i )

[0025]

[0026]

[0027] Among them, CC, PE, MI, mi, and ci represent the mutual correlation coefficient, permutation entropy, mutual information entropy, and normalized and composite index of mutual information entropy, respectively. i represents the IMF component of the i-th sound signal, represents the average value of all IMF components, y i represents the i-th sound signal, represents the average value of all sound signals, N represents the total number of sound signals, and p j represents the probability of the jth arrangement of the kth modal component of the ith sound signal, k = 1, 2, ..., n, n represents the total number of modal components, x[n] represents the original sound signal, H(y i ) represents the signal y i The entropy, H(y i |x i ) indicates that x is known i Time i The conditional entropy, H(x i ) represents the signal x i The entropy of .

[0028] Furthermore, in step S5, there are 9 time domain characteristic values, namely, average value, peak-to-peak value, standard deviation, root mean square, kurtosis value, peak factor, pulse factor, waveform factor and margin factor.

[0029] Furthermore, in step S6, the hyperparameters of the convolutional neural network include a learning rate, hidden layer nodes, and a regularization coefficient, and optimizing the hyperparameters includes the following:

[0030] S601, setting COA parameters, including population size, maximum number of iterations, and optimization upper and lower bounds.

[0031] S602, initialize the population, and calculate the fitness values ​​of the learning rate, hidden layer nodes, and regularization coefficient in the convolutional neural network.

[0032] S603. Update the locations of all raccoons and iguanas according to the basic principles of COA.

[0033] S604, determine whether each target exceeds the boundary, if so, return to step S603 and repeat the operation; if not, proceed to step S605.

[0034] S605, modifying the learning rate, hidden layer nodes, regularization coefficients and fitness values.

[0035] S606. Repeat steps S602-S605 until the termination criterion is met, and obtain the globally optimal learning rate, hidden layer nodes and regularization coefficient.

[0036] S607. The optimized CNN fault diagnosis model includes a folding layer, two convolutional layers, two activation layers and an anti-folding layer connected in sequence.

[0037] Furthermore, in step S7, the diagnosis result obtained includes the following contents:

[0038] The signal feature vector after denoising is divided into a training set and a validation set according to a ratio of 7:3. The training set is input into the optimized CNN fault diagnosis model for training until the accuracy reaches 100%. The validation set is input into the trained model to obtain the fault classification diagnosis result, which is output as a confusion matrix diagram.

[0039] Furthermore, the present invention also proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the fault diagnosis method based on the acoustic signal of the belt conveyor roller bearing when executing the computer program.

[0040] Furthermore, the present invention also proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is run by a processor, the fault diagnosis method based on the acoustic signal of the belt conveyor roller bearing is executed.

[0041] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0042] 1. The present invention realizes long-distance non-contact fault diagnosis of belt conveyor roller bearings by noise reduction processing of sound signals, solves the disadvantage of needing to arrange a large number of vibration sensors, and provides an effective new way for long-distance non-contact diagnosis of belt conveyor rollers.

[0043] 2. The variational modal decomposition used in the present invention has good noise robustness and the ability to effectively suppress modal aliasing and endpoint effects, and can better extract fault information.

[0044] 3. The combination of variational mode decomposition parameters in the present invention has a significant impact on the decomposition effect; at the same time, the selection of optimal IMF components for different sound signals is also different.

[0045] 4. The present invention adopts the raccoon optimization algorithm to perform global optimization on the key parameters of variational modal decomposition, which can not only effectively determine the optimal parameter combination, but also adaptively decompose and extract different sound signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is an overall implementation flow chart of the present invention.

[0047] Figure 2 It is the optimization flow chart of the raccoon algorithm of the present invention.

[0048] Figure 3 It is the key parameter optimization flow chart of the present invention.

[0049] Figure 4 It is a hyperparameter optimization flow chart of the present invention.

[0050] Figure 5 4 is a spectrum diagram after adaptive noise reduction according to an embodiment of the present invention.

[0051] Figure 6 4 is a confusion matrix diagram of an embodiment of the present invention without COA optimization.

[0052] Figure 7 4 is a confusion matrix diagram of COA-VMD-CNN according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0054] To achieve the above object, the present invention proposes a fault diagnosis method based on the acoustic signal of the belt conveyor roller bearing, such as Figure 1 As shown, the specific steps are as follows:

[0055] S1. Use sound sensors to collect acoustic signals of roller bearings of belt conveyors when they are unloaded in normal and faulty states. The faulty state refers to the state in which iron filings enter the bearings, causing the inner and outer rings to wear and become damaged.

[0056] S2, such as Figure 2 As shown in the figure, COA (Coati Optimization Algorithm) is a swarm intelligence optimization algorithm that simulates the hunting behavior of raccoons for global optimization, combining the strategies of raccoons in the hunting attack and escape from predation stages. In this algorithm, the global search of raccoons is divided into two types, namely hunting attack (exploration stage) and escape from predation (exploitation stage). The algorithm finds the optimal parameters through these behaviors and abstracts them into mathematical language.

[0057] (1) Initialize the population. The specific formula is:

[0058]

[0059] Among them, x d,z It means individual; represents the lower boundary of optimization; represents the upper boundary of optimization; r represents a random number between [0,1].

[0060] (2) Exploration phase

[0061] Search the global position, the specific formula is:

[0062]

[0063] in, represents the new position of the d-th raccoon in the z-th dimension; G z represents the position of the iguana in the zth dimension, that is, the position of the best member; I represents a number randomly selected from the set {1,2}; M represents the total number of raccoons; Indicates no more than The maximum integer of ; m represents the total number of decision variables.

[0064] Based on the random position in the search space after the iguana falls to the ground, the raccoon moves, and the specific formula is:

[0065]

[0066]

[0067] in, represents the position of the iguana on the ground in the zth dimension, represents the objective function value after the iguana falls to the ground in the zth dimension, F d,z Represents the objective function value of the d-th raccoon in the z-th dimension.

[0068] If the updated individual is better, update the current individual; otherwise keep it as it is. The specific formula is:

[0069]

[0070] in, represents the objective function value of the d-th raccoon in the new position; F d represents the objective function value of the dth raccoon at its previous position, X d represents the position of the dth raccoon, represents the new position of the dth raccoon.

[0071] (3) Development stage

[0072] Generate a random position near each raccoon's position using the following formula:

[0073]

[0074]

[0075]

[0076] in, represents the local lower bound of the z-th decision variable; represents the local upper bound of the z-th decision variable, q represents the number of iterations, q = 1, 2, ..., Q, Q represents the maximum number of iterations, represents the new position of the d-th raccoon in the z-th dimension.

[0077] If the updated individual is better, update the current individual, otherwise keep it as it is. The specific formula is:

[0078]

[0079] COA is used to adaptively optimize the key parameters in the VMD (Variational Mode Decomposition) algorithm globally to obtain the global optimal solution; the key parameters include the number of modes K and the penalty factor α; Figure 3 As shown, the specific contents are:

[0080] S201, setting the parameters of COA, including the population size, the maximum number of iterations, and the optimization upper and lower bounds.

[0081] S202, initialize the population and calculate the fitness values ​​of key parameters in the VMD algorithm.

[0082] S203. Update the locations of all raccoons and iguanas according to the basic principles of COA.

[0083] S204, determine whether the key parameters in the VMD algorithm exceed the set boundaries; if exceeded, return to step S203 and re-operate; if not exceeded, enter step S205.

[0084] S205, modifying key parameters and fitness values ​​in the VMD algorithm.

[0085] S206. Repeat steps S202-S205 until the termination criterion is met, and the global optimal solution [K, α] is obtained.

[0086] In this embodiment, the population size of COA is set to 50, the maximum number of iterations is set to 500, the upper and lower bounds of K are set to [3, 12], and the upper and lower bounds of α are set to [100, 2500].

[0087] S3, feeding back the key parameters at the end of optimization to the VMD algorithm, and using the algorithm to perform adaptive signal decomposition on the acoustic signal in step S1, wherein VMD mainly adopts non-recursive technology, constructs and solves a finite variation problem, decomposes the complex initial signal into several IMF component sequences according to the frequency from high to low, and obtains the corresponding IMF (Intrinsic Mode Function); the specific content is:

[0088] S301, construct a variational problem, whose constraint condition is that the sum of the decomposed components is the original signal, and the constraint expression is:

[0089]

[0090]

[0091] Where K represents the number of decomposition modes that need to be determined in advance; {μ k}{ω k} respectively represent the kth modal component and center frequency after decomposition; δ(t) represents the Dirac function; * represents the convolution operator; f(t) represents the original signal.

[0092] S302, solving the variational problem, by introducing a quadratic penalty factor and a Lagrange multiplication operator, transforming the constrained variational problem into an unconstrained variational problem, and obtaining an augmented Lagrange expression, specifically:

[0093]

[0094] in, It is used to ensure the reconstruction accuracy of the signal, and λ(t) represents the Lagrange multiplication operator.

[0095] S303, update {μ k (ω)} to obtain the modal component, the specific formula is:

[0096]

[0097] Update the center frequency {ω k}, the specific formula is:

[0098]

[0099] Update the Lagrange multiplication operator, the specific formula is:

[0100]

[0101] Among them, γ represents the noise tolerance; Represent the n+1th modal component respectively Modal function μ i where λ(t) is the Fourier transform of f(t) and λ(t).

[0102] S304, repeat steps S301-S303 until the following conditions are met and then stop:

[0103]

[0104] Where φ represents the convergence accuracy, which is 1×10 -7 .

[0105] S4. Use comprehensive indicators to screen the components in the IMF, and take the component with the largest comprehensive indicator as the optimal IMF component; the specific contents are:

[0106] Comprehensive indicators include cross-correlation coefficient, permutation entropy and mutual information entropy; among them, CC (Cross-correlation Coefficient) criterion is a measure of the strength of the linear relationship between two variables. The larger the CC value, the higher the similarity between the IMF component and the original signal. The size of PE (Permutation Entropy) measures the degree of random change of the signal time series. The larger the PE value, the more random the signal time series and the more complex the signal; conversely, the more regular the signal sequence and the less complex it is. MI (Mutual Information) is mainly used in information theory to indicate the degree of correlation between two events. It is not easily affected by external factors. The larger the mutual information entropy value, the stronger the correlation between the two events. As for the IMF component, the richer the original signal feature information it contains, the larger the mutual information entropy value.

[0107] The specific calculation formula of the comprehensive index is:

[0108]

[0109]

[0110] MI(x i ,y i )=H(y i )-H(y i |x i )

[0111]

[0112]

[0113] Among them, mi and ci represent the normalized and composite index of mutual information entropy, respectively, and x i represents the IMF component of the i-th sound signal, represents the average value of all IMF components, y i represents the i-th sound signal, represents the average value of all sound signals, N represents the total number of sound signals, and p j represents the probability of occurrence of the jth arrangement in the kth modal component of the ith sound signal, k = 1, 2, ..., n, n represents the total number of modal components, j = 1, 2, ..., J, J represents the total number of arrangements, H(y i ) represents y i The entropy, H(y i |x i ) indicates that x is known i Time i The conditional entropy, H(x i ) represents xi The entropy of .

[0114] S5. Calculate the time domain eigenvalues ​​of the optimal IMF component to form the signal eigenvector after noise reduction. There are 9 time domain eigenvalues, namely, mean value, peak-to-peak value, standard deviation, root mean square, kurtosis value, peak factor, pulse factor, waveform factor and margin factor. The specific expression is:

[0115]

[0116] Peak-to-peak value = x max -x min .

[0117]

[0118]

[0119]

[0120]

[0121]

[0122]

[0123]

[0124] S6. CNN (Convolutional Neural Networks) is a deep learning model similar to the artificial neural network multi-layer perceptron, which has powerful feature extraction capabilities. Its network consists of input layer, hidden layer and output layer, where the hidden layer includes convolution layer, activation layer, pooling layer and fully connected layer.

[0125] The convolution layer is used to extract features. It generates output features by convolving the local area of ​​the input with the convolution kernel. This paper increases the size of the convolution kernel of the first layer to obtain a larger receptive field and extract short-term features. The remaining layers are small convolution kernels, which is helpful to prevent overfitting. Its operation expression is:

[0126]

[0127] Among them, x represents input; Represents the feature output after convolution of the oth convolution kernel of the lth layer; They represent the weight and bias of the o-th convolution kernel in the l-th layer respectively.

[0128] The function of the activation layer is to enable the model to have multi-layer nonlinear mapping learning capabilities by adding a nonlinear activation function. The Relu function is selected as the activation function. Its operation expression is:

[0129]

[0130] The pooling layer downsamples the data after the convolutional layer to extract the main features, reducing the model size and effectively avoiding overfitting.

[0131] The fully connected layer is responsible for integrating the features extracted by the model and sending them as input to the classifier to output the classification results.

[0132] Among them, the learning rate is a parameter that controls the update step size of the back-propagation error. If the learning rate is too large, the network may not converge, and if it is too small, the network will converge too slowly. The number of hidden layers directly affects the expressive power of the neural network. Increasing the number of hidden layers can increase the complexity of the network, enabling it to learn more complex patterns and features. However, the more layers there are, the more complex the training process is, the greater the amount of calculation is, and the longer the fault diagnosis time is. The choice of the regularization coefficient has an important influence on the generalization ability of the model. A smaller regularization coefficient value will result in a weaker regularization effect, which may not be enough to prevent overfitting. A larger regularization coefficient value will result in too strong regularization, which may affect the fitting ability of the model. Therefore, choosing an appropriate regularization coefficient value requires finding a balance between the generalization ability of the model and the training error.

[0133] The hyperparameters of the convolutional neural network include learning rate, hidden layer nodes and regularization coefficient. COA is used to optimize the hyperparameters of the convolutional neural network to obtain the optimized CNN fault diagnosis model; Figure 4 As shown, the specific contents are:

[0134] S601, setting COA parameters, including population size, maximum number of iterations, and optimization upper and lower bounds.

[0135] S602, initialize the population, and calculate the fitness values ​​of the learning rate, hidden layer nodes, and regularization coefficient in the convolutional neural network.

[0136] S603. Update the locations of all raccoons and iguanas according to the basic principles of COA.

[0137] S604, determine whether each target exceeds the boundary, if so, return to step S603 and repeat the operation; if not, proceed to step S605.

[0138] S605, modifying the learning rate, hidden layer nodes, regularization coefficients and fitness values.

[0139] S606. Repeat steps S602-S605 until the termination criterion is met, and obtain the globally optimal learning rate, hidden layer nodes and regularization coefficient.

[0140] S607. The optimized CNN fault diagnosis model includes a folding layer, two convolutional layers, two activation layers and an anti-folding layer connected in sequence.

[0141] In this embodiment, the lower bounds of the learning rate, hidden layer nodes and regularization coefficient are set to 1e-3, 10 and 1e-4 respectively, and the upper bounds are set to 1e-2, 30 and 1e-1 respectively.

[0142] S7. Divide the signal feature vector after denoising into a training set and a validation set according to a ratio of 7:3, input the training set into the optimized CNN fault diagnosis model for training until the accuracy reaches 100%, input the validation set into the trained model to obtain the fault classification diagnosis result, and output the result in the form of a confusion matrix diagram.

[0143] In this embodiment, the collected roller bearing data is taken from the belt conveyor roller bearing fault simulation test bench built in the laboratory. Three groups of rollers are set up, each group is 1m apart, and the roller size is 250mm. The motor frequency is 1.6kW, the belt speed is 2m / s, and the sampling frequency is 48KHz. The sampling points of the acoustic signals of the two roller bearings are 2048, and 1200 groups of samples are collected respectively, for a total of 2400 groups. The 2400 groups of data are numbered by fault type, 1 for normal and 2 for fault, and are divided into training set and verification set in a ratio of 7:3. There are 840 training sets and 360 verification sets corresponding to each acoustic signal.

[0144] Figure 5 The schematic diagram of the parameter combination obtained by adaptive optimal modal decomposition of the two types of data collected, normal signal and inner and outer ring wear fault signal, is shown. Taking the inner and outer ring wear fault as an example, the IMF component spectrum diagram of the optimal decomposition is shown. It can be seen that the IMF components after adaptive decomposition effectively avoid problems such as modal aliasing and endpoint effect. The diagnostic results output by the existing method and the method proposed in the present invention are shown as follows: Figure 6 and Figure 7 shown. Figure 6 The accuracy rate is 93.5%, mainly due to inaccurate identification of faulty rollers; Figure 7 The accuracy rate is 96.667%. The recognition accuracy rate of faulty rollers has increased, and the recognition accuracy rates of the two categories are comparable.

[0145] The embodiment of the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. It should be noted that when the processor executes the computer program, the specific steps of the method provided in the embodiment of the present invention are corresponding to the specific steps of the method provided in the embodiment of the present invention, and the processor has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided in the embodiment of the present invention.

[0146] The embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program. It should be noted that when the computer program is executed by the processor, it corresponds to the specific steps of the method provided in the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided in the embodiment of the present invention.

[0147] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A fault diagnosis method based on acoustic signals of belt conveyor roller bearings, characterized in that: include: S1. Collecting acoustic signals of roller bearings using sound sensors; S2. Using the raccoon optimization algorithm to adaptively and globally optimize the key parameters in the variational mode decomposition algorithm to obtain the global optimal solution; the key parameters include the number of modes K and the penalty factor α; S3, feeding back the key parameters at the end of the optimization to the variational mode decomposition algorithm, using the algorithm to perform adaptive signal decomposition on the acoustic signal in step S1 to obtain the corresponding IMF; S4. Use comprehensive indicators to screen the components in the IMF, and take the component with the largest comprehensive indicator as the optimal IMF component; S5, calculating the time domain eigenvalue of the optimal IMF component to form a signal eigenvector after noise reduction; S6. Use the raccoon optimization algorithm to optimize the hyperparameters of the convolutional neural network to obtain an optimized convolutional neural network fault diagnosis model; S7. Input the denoised signal feature vector into the optimized convolutional neural network fault diagnosis model to perform fault classification diagnosis and obtain a diagnosis result.

2. The fault diagnosis method based on the acoustic signal of the belt conveyor roller bearing according to claim 1 is characterized in that: In step S1, the sound sensor collects acoustic signals of the roller bearing of the belt conveyor when it is unloaded in a normal state and a fault state respectively.

3. The fault diagnosis method based on the acoustic signal of the belt conveyor roller bearing according to claim 1 is characterized in that: In step S2, the adaptive global optimization of key parameters includes the following contents: S201, setting the parameters of the raccoon optimization algorithm, including the population size, the maximum number of iterations, and the optimization upper and lower bounds; S202, initializing the population and calculating the fitness values ​​of key parameters in the variational mode decomposition algorithm; S203, updating the positions of all raccoons and iguanas according to the basic principles of the raccoon optimization algorithm; S204, determining whether key parameters in the variational mode decomposition algorithm exceed set boundaries; If it exceeds, then return to step S203 and repeat the operation; if it does not exceed, then proceed to step S205; S205, modifying key parameters and fitness values ​​in the variational mode decomposition algorithm; S206. Repeat steps S202-S205 until the termination criterion is met, and the global optimal solution [K, α] is obtained.

4. The fault diagnosis method based on the acoustic signal of the belt conveyor roller bearing according to claim 1 is characterized in that: In step S4, the comprehensive indicators include mutual correlation coefficient, permutation entropy and mutual information entropy, and the specific expressions are as follows: MI(x i ,and i )=H(and i )-H(and i |x i ) Among them, CC, PE, MI, mi, and ci represent the mutual correlation coefficient, permutation entropy, mutual information entropy, and normalized and composite index of mutual information entropy, respectively. i represents the IMF component of the i-th sound signal, represents the average value of all IMF components, y i represents the i-th sound signal, represents the average value of all sound signals, N represents the total number of sound signals, and p j represents the probability of the jth arrangement of the kth modal component of the ith sound signal, k = 1, 2, ..., n, n represents the total number of modal components, x[n] represents the original sound signal, H(y i ) represents the signal y i The entropy, H(y i |x i ) indicates that x is known i Time i The conditional entropy, H(x i ) represents the signal x i The entropy of .

5. The fault diagnosis method based on the acoustic signal of the belt conveyor roller bearing according to claim 1 is characterized in that: In step S5, there are 9 time domain characteristic values, namely, average value, peak-to-peak value, standard deviation, root mean square, kurtosis value, peak factor, pulse factor, waveform factor and margin factor.

6. The fault diagnosis method based on acoustic signals of belt conveyor roller bearings according to claim 1 is characterized in that: In step S6, the hyperparameters of the convolutional neural network include learning rate, hidden layer nodes and regularization coefficient, and the optimization of the hyperparameters includes the following: S601, setting the parameters of the raccoon optimization algorithm, including the population size, the maximum number of iterations, and the optimization upper and lower bounds; S602, initializing the population, and calculating the fitness values ​​of the learning rate, hidden layer nodes, and regularization coefficient in the convolutional neural network; S603, updating the positions of all raccoons and iguanas according to the basic principles of the raccoon optimization algorithm; S604, determine whether each target exceeds the boundary, if so, return to step S603 and repeat the operation; if not, proceed to step S605; S605, modifying the learning rate, hidden layer nodes, regularization coefficients and fitness values ​​thereof; S606. Repeat steps S602-S605 until the termination criterion is met, and obtain the globally optimal learning rate, hidden layer nodes and regularization coefficient. S607. The optimized convolutional neural network fault diagnosis model includes a folding layer, two convolutional layers, two activation layers and an anti-folding layer connected in sequence.

7. The fault diagnosis method based on acoustic signals of belt conveyor roller bearings according to claim 1 is characterized in that: In step S7, the diagnosis result obtained includes the following contents: The signal feature vector after denoising is divided into training set and validation set according to the ratio of 7:

3. The training set is input into the optimized convolutional neural network fault diagnosis model for training until the accuracy reaches 100%. The validation set is input into the trained model to obtain the fault classification diagnosis result, which is output as a confusion matrix diagram.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the fault diagnosis method based on the acoustic signal of the belt conveyor roller bearing described in any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the fault diagnosis method based on acoustic signals of belt conveyor roller bearings according to any one of claims 1 to 7 is executed.

Images