Mechanical fault diagnosis method and system based on deep wavelet kernel variational autoencoder

Abstract

The invention discloses a mechanical fault diagnosis method and system based on a deep wavelet kernel variational autoencoder. The diagnosis method comprises the following steps: obtaining two-channel nonlinear vibration data for different faults of an electromechanical fluid system and randomly dividing the data into a training set and a test set at a ratio of 1:1; sequentially connecting a kernel function mapping module, a deep kernel feature learning module and a classification module in series to form a deep wavelet kernel variational autoencoder; wherein the kernel function mapping module projects the collected data set into a high-dimensional feature space using a kernel function mapping method, the deep kernel feature learning module uses a wavelet variational autoencoder for stacking learning and obtains a classification vector, and the classification module performs fault classification on the classification vector; the deep wavelet kernel variational autoencoder is trained using a training set to complete the deep extraction of nonlinear fault information, and the test set is input into the trained deep wavelet kernel variational autoencoder to automatically realize fault diagnosis of the electromechanical fluid system.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical fault diagnosis, and in particular to a mechanical fault diagnosis method and system based on a deep wavelet kernel variational autoencoder. Background Art

[0002] Due to the harsh operating environment and complex working conditions of high-end equipment (such as wind turbines, aircraft engines, and high-speed rail rotating parts), it is difficult to obtain vibration monitoring data of high-end equipment under high-speed, variable load and other large fluctuation conditions, and the sample size is difficult to meet actual needs. The essential fault information contained in it is weak, coupled and uncertain, that is, the useful information contained in small sample data under fluctuating conditions is difficult to reveal, and traditional diagnostic algorithms are powerless to do so. Therefore, how to use effective analysis methods to reveal the discriminative fault information contained in small sample data is a difficult problem in high-end equipment fault diagnosis.

[0003] At present, deep learning theory provides an effective way to solve the difficult problem that high-end equipment failures are difficult to diagnose due to mutual coupling, weak features, and small amount of useful information. However, traditional deep learning models (such as deep belief networks, convolutional neural networks, sparse autoencoders, denoising autoencoders, variational autoencoders, etc.) have high data requirements during model training. When faced with severe data shortage, their feature learning performance and fault diagnosis capabilities drop sharply. Summary of the invention

[0004] In response to the problems and needs raised above, this solution proposes a mechanical fault diagnosis method and system based on a deep wavelet kernel variational autoencoder. Due to the adoption of the following technical features, it can achieve the above technical objectives and bring about many other technical effects.

[0005] An object of the present invention is to propose a mechanical fault diagnosis method based on a deep wavelet kernel variational autoencoder, comprising the following steps:

[0006] S10: Obtain two-channel nonlinear vibration data for different faults in the electromechanical and hydraulic system x k and The number of sampling points is n, and they are randomly divided into training set and test set in a 1:1 ratio;

[0007] S20: Connecting the kernel function mapping module, the deep kernel feature learning module and the classification module in series in sequence to form a deep wavelet kernel variational autoencoder; wherein the kernel function mapping module is configured to project the collected data set into a high-dimensional feature space using a kernel function mapping method, the deep kernel feature learning module is configured to use a wavelet variational autoencoder for stacking learning and obtain a classification vector, and the classification module is configured to perform fault classification on the classification vector;

[0008] S30: The deep wavelet kernel variational autoencoder is trained using the training set to complete the deep extraction of nonlinear fault information, and the test set is input into the trained deep wavelet kernel variational autoencoder to automatically realize the fault diagnosis of the electromechanical hydraulic system.

[0009] In one example of the present invention, in step S20,

[0010] The kernel function mapping module is composed of an input layer and a mapping layer connected in series, and is configured to output the data of the input layer after being processed by the kernel function mapping of the mapping layer;

[0011] The deep kernel feature learning module is composed of at least three wavelet variational autoencoders connected in series, and is configured to use the feature vector output by the last wavelet variational autoencoder as a classification vector;

[0012] The classification module is configured to use an unsupervised clustering algorithm to achieve automatic identification of fault types.

[0013] In one example of the present invention, the expression of the classification module is:

[0014]

[0015] In the formula, is the output of the third wavelet variational autoencoder, c n is the number of clusters, and clustering(·) is a function in the MATLAB clustering toolkit.

[0016] In one example of the present invention, the wavelet variational autoencoder comprises a wavelet encoding module, a wavelet sampling module and a wavelet decoding module which are cascaded in sequence; wherein,

[0017] The wavelet encoding module is composed of an input layer and a wavelet hidden layer connected in series, and is configured to learn the hidden layer feature information;

[0018] The wavelet sampling module is configured to perform feature sampling from hidden layer features that satisfy Gaussian distribution;

[0019] The wavelet decoding module is composed of a wavelet hidden layer and an output layer connected in series, and is configured to reconstruct signals according to the characteristics of the samples to generate interpretable feature data.

[0020] In one example of the present invention, the wavelet hidden layer in the wavelet encoding module uses an asymmetric real Laplace wavelet ψ(t) as an activation function, and its output is q j ,j=1,2,…,m, the specific calculation formula is:

[0021]

[0022] In the formula, f b and f c are the bandwidth and center frequency of the wavelet, respectively, q j is the output of the jth wavelet hidden layer node in the wavelet coding module, d j and c j The scale factor and translation factor are respectively, jk is the weight between the jth wavelet hidden layer node and the kth input layer node.

[0023] In one example of the present invention, the activation functions of the wavelet hidden layer and the output layer in the wavelet decoding module respectively use an asymmetric real Laplace wavelet ψ(t) and a hyperbolic tangent function to convert the hidden variable feature z output by the wavelet sampling module into j As the input of the wavelet decoding module, the decoding operation process is carried out, and the output results are p j ,j=1,2,…,m and y k ,k=1,2,…,n, the specific calculation formula is:

[0024]

[0025] In the formula, f b and f c are the bandwidth and center frequency of the wavelet, respectively, j is the output of the jth wavelet hidden layer node in the wavelet decoding module, d j and c j The scale factor and translation factor are respectively, j is the weight between the jth wavelet hidden layer node and the jth sampling node, W kj is the weight between the kth output layer node and the jth wavelet hidden layer node, and tanh(·) is the hyperbolic tangent function.

[0026] Another object of the present invention is to propose a mechanical fault diagnosis system based on a deep wavelet kernel variational autoencoder, comprising:

[0027] Signal acquisition and division device for obtaining two-channel nonlinear vibration data for different faults in electromechanical and hydraulic systems k and The number of sampling points is n, and they are randomly divided into training set and test set in a 1:1 ratio;

[0028] A model building device, used for sequentially connecting a kernel function mapping module, a deep kernel feature learning module and a classification module to form a deep wavelet kernel variational autoencoder; wherein the kernel function mapping module is configured to project the collected data set into a high-dimensional feature space using a kernel function mapping method, the deep kernel feature learning module is configured to use a wavelet variational autoencoder to perform stacking learning and obtain a classification vector, and the classification module is configured to perform fault classification on the classification vector;

[0029] The training and fault identification device is used to train the deep wavelet kernel variational autoencoder with a training set to complete the deep extraction of nonlinear fault information, and input the test set into the trained deep wavelet kernel variational autoencoder to automatically realize the fault diagnosis of the electromechanical hydraulic system.

[0030] In one example of the present invention,

[0031] The kernel function mapping module is composed of an input layer and a mapping layer connected in series, and is configured to output the data of the input layer after being processed by the kernel function mapping of the mapping layer;

[0032] The deep kernel feature learning module is composed of at least three wavelet variational autoencoders connected in series, and is configured to use the feature vector output by the last wavelet variational autoencoder as a classification vector;

[0033] The classification module is configured to use an unsupervised clustering algorithm to achieve automatic identification of fault types.

[0034] In one example of the present invention, the wavelet variational autoencoder comprises a wavelet encoding module, a wavelet sampling module and a wavelet decoding module which are cascaded in sequence; wherein,

[0035] The wavelet encoding module is composed of an input layer and a wavelet hidden layer connected in series, and is configured to learn the hidden layer feature information;

[0036] The wavelet sampling module is configured to perform feature sampling from hidden layer features that satisfy Gaussian distribution;

[0037] The wavelet decoding module is composed of a wavelet hidden layer and an output layer connected in series, and is configured to reconstruct signals according to the characteristics of the samples to generate interpretable feature data.

[0038] In one example of the present invention, the wavelet hidden layer in the wavelet encoding module uses an asymmetric real Laplace wavelet ψ(t) as an activation function, and its output is q j ,j=1,2,…,m, the specific calculation formula is:

[0039]

[0040] In the formula, f b and fc are the bandwidth and center frequency of the wavelet, respectively, q j is the output of the jth wavelet hidden layer node in the wavelet coding module, d j and c j The scale factor and translation factor are respectively, jk is the weight between the jth wavelet hidden layer node and the kth input layer node.

[0041] The best embodiment for carrying out the present invention will be described in more detail below with reference to the accompanying drawings so that the features and advantages of the present invention can be easily understood. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings of the embodiments of the present invention, wherein the drawings are only used to illustrate some embodiments of the present invention, but not to limit all embodiments of the present invention thereto.

[0043] Figure 1 is an overall implementation flow chart of a fault diagnosis method according to an embodiment of the present invention;

[0044] Figure 2 2 is a schematic diagram of the structure of a deep wavelet kernel variational autoencoder according to an embodiment of the present invention;

[0045] Figure 3 A schematic diagram of the structure of a wavelet variational autoencoder according to an embodiment of the present invention;

[0046] Figure 4 A rotating machinery fault simulation test bench according to an embodiment of the present invention;

[0047] Figure 5 A time domain waveform diagram of vibration data of a rolling bearing under different health conditions according to an embodiment of the present invention;

[0048] Figure 6 This is a comparison chart of recognition accuracy achieved by different deep learning methods according to an embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the technical solution of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described in conjunction with the drawings of specific embodiments of the present invention. The same figure marks in the drawings represent the same parts. It should be noted that the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] Since it is difficult to obtain a large sample of vibration monitoring data for high-end equipment under fluctuating working conditions such as high-speed and variable load, and the fault information contained therein is weak, coupled and uncertain, traditional diagnostic models are unable to solve such problems. To solve this problem, the present invention discloses a mechanical fault diagnosis method based on a deep wavelet kernel variational autoencoder. This method integrates kernel function mapping processing, wavelet activation function and unsupervised clustering algorithm into the input layer, hidden layer and output layer of the traditional variational autoencoder respectively, and constructs a deep wavelet kernel variational autoencoder through layer-by-layer cascade learning. It can deeply mine the weak and multi-coupled fault information contained in small sample monitoring data, and can overcome the problem that traditional diagnostic models are heavily dependent on data category labeling, saving the cost of manually labeling data categories and ensuring efficient diagnostic accuracy.

[0051] It should be noted that the unsupervised clustering algorithm involved in the present invention uses functions in the MATLAB clustering toolkit, including K-means clustering, fuzzy C-means clustering, hierarchical clustering, density peak clustering and other algorithms, but is not limited to these clustering algorithm toolkits.

[0052] According to the first aspect of the present invention, a mechanical fault diagnosis method based on a deep wavelet kernel variational autoencoder is provided. Figures 1 to 3 As shown, the following steps are included:

[0053] S10: Obtain two-channel nonlinear vibration data for different faults in the electromechanical and hydraulic system x k and The number of sampling points is n, and they are randomly divided into training set and test set in a 1:1 ratio;

[0054] S20: Connecting the kernel function mapping module, the deep kernel feature learning module and the classification module in series in sequence to form a deep wavelet kernel variational autoencoder; wherein the kernel function mapping module is configured to project the collected data set into a high-dimensional feature space using a kernel function mapping method, the deep kernel feature learning module is configured to use a wavelet variational autoencoder for stacking learning and obtain a classification vector, and the classification module is configured to perform fault classification on the classification vector;

[0055] S30: The deep wavelet kernel variational autoencoder is trained using the training set to complete the deep extraction of nonlinear fault information, and the test set is input into the trained deep wavelet kernel variational autoencoder to automatically realize the fault diagnosis of the electromechanical hydraulic system.

[0056] This diagnostic method projects the collected data set into a high-dimensional feature space by using a kernel function mapping method, performs serial learning by multiple wavelet variational autoencoders, and uses an unsupervised clustering algorithm for fault identification, thus forming a deep wavelet kernel variational autoencoder. The above diagnostic method can effectively reveal the discriminative weak fault information hidden in small sample data, and complete the accurate diagnosis of the equipment failure mode and degree under fluctuating working conditions. At the same time, it has the excellent characteristics of variational autoencoders such as strong robustness, generalization and anti-interference, and it does not need to rely on data category labels, which can save the cost of manual data category labeling.

[0057] It should be noted that the present invention constructs a deep wavelet kernel variational autoencoder model by integrating kernel function mapping processing, wavelet activation function and unsupervised clustering algorithm into the input layer, hidden layer and output layer of the traditional variational autoencoder respectively. It can not only get rid of the dependence of traditional deep learning models on data category labels, but also reduce learning errors in the model training process, realize accurate identification of fault types, and has good robustness and generalization ability.

[0058] In one example of the present invention, in step S20,

[0059] The kernel function mapping module is composed of an input layer and a mapping layer connected in series, and is configured to output the data of the input layer after being processed by the kernel function mapping of the mapping layer;

[0060] The deep kernel feature learning module is composed of at least three wavelet variational autoencoders connected in series, and is configured to use the feature vector output by the last wavelet variational autoencoder as a classification vector;

[0061] The classification module is equivalent to an output layer and is configured to use an unsupervised clustering algorithm to achieve automatic identification of fault types.

[0062] It should be noted that the number of wavelet variational autoencoders of the deep kernel feature learning module is selected according to the size of the computing task. Preferably, three wavelet variational autoencoders are selected; the three wavelet variational autoencoders are connected in series in sequence, and the output of the first wavelet variational autoencoder is As the input of the second wavelet variational autoencoder, the output of the second wavelet variational autoencoder As the input of the third wavelet variational autoencoder.

[0063] In one example of the present invention, the specific expression of the kernel function mapping module is:

[0064]

[0065] In the formula, k = 1, 2, ..., n, n is the number of sampling points, x k and is the collected data, K(·) is the Gaussian kernel function, and γ is the kernel parameter.

[0066] In one example of the present invention, the expression of the classification module is:

[0067]

[0068] In the formula, is the output of the third wavelet variational autoencoder, c n is the number of clusters, and clustering(·) is a function in the MATLAB clustering toolkit.

[0069] In one example of the present invention, the wavelet variational autoencoder comprises a wavelet encoding module, a wavelet sampling module and a wavelet decoding module which are cascaded in sequence; wherein,

[0070] The wavelet encoding module is composed of an input layer and a wavelet hidden layer connected in series, and is configured to learn the hidden layer feature information;

[0071] The wavelet sampling module is configured to perform feature sampling from hidden layer features that satisfy Gaussian distribution;

[0072] The wavelet decoding module is composed of a wavelet hidden layer and an output layer connected in series, and is configured to reconstruct signals according to the characteristics of the samples to generate interpretable feature data.

[0073] In one example of the present invention, the wavelet hidden layer in the wavelet encoding module uses an asymmetric real Laplace wavelet ψ(t) as an activation function, and its output is q j ,j=1,2,…,m, the specific calculation formula is:

[0074]

[0075] In the formula, f b and f c are the bandwidth and center frequency of the wavelet, respectively, q j is the output of the jth wavelet hidden layer node in the wavelet coding module, d j and c j The scale factor and translation factor are respectively, jk is the weight between the jth wavelet hidden layer node and the kth input layer node.

[0076] In an example of the present invention, the wavelet sampling module is equivalent to a resampling process. Specifically, firstly, the output result q of the wavelet encoding module is calculated according to the encoding process. j, j=1,2,…,m corresponding to μ mean and variance σ, then randomly sample ε from the normal distribution N(1,0), and output the latent variable feature z according to the formula z=μ+εσ j ,j=1,2,L,m.

[0077] In one example of the present invention, the activation functions of the wavelet hidden layer and the output layer in the wavelet decoding module respectively use an asymmetric real Laplace wavelet ψ(t) and a hyperbolic tangent function to convert the hidden variable feature z output by the wavelet sampling module into j As the input of the wavelet decoding module, the decoding operation process is carried out, and the output results are p j ,j=1,2,…,m and y k ,k=1,2,…,n, the specific calculation formula is:

[0078]

[0079] In the formula, f b and f c are the bandwidth and center frequency of the wavelet, respectively, j is the output of the jth wavelet hidden layer node in the wavelet decoding module, d j and c j The scale factor and translation factor are respectively, j is the weight between the jth wavelet hidden layer node and the jth sampling node, W kj is the weight between the kth output layer node and the jth wavelet hidden layer node, and tanh(·) is the hyperbolic tangent function.

[0080] According to the second aspect of the present invention, a mechanical fault diagnosis system based on a deep wavelet kernel variational autoencoder comprises:

[0081] Signal acquisition and division device for obtaining two-channel nonlinear vibration data for different faults in electromechanical and hydraulic systems k and The number of sampling points is n, and they are randomly divided into training set and test set in a 1:1 ratio;

[0082] A model building device, used for sequentially connecting a kernel function mapping module, a deep kernel feature learning module and a classification module to form a deep wavelet kernel variational autoencoder; wherein the kernel function mapping module is configured to project the collected data set into a high-dimensional feature space using a kernel function mapping method, the deep kernel feature learning module is configured to use a wavelet variational autoencoder for stacking learning and obtain a classification vector, and the classification module is configured to perform fault classification on the classification vector;

[0083] The training and fault identification device is used to train the deep wavelet kernel variational autoencoder with a training set to complete the deep extraction of nonlinear fault information, and input the test set into the trained deep wavelet kernel variational autoencoder to automatically realize the fault diagnosis of the electromechanical hydraulic system.

[0084] The diagnostic system projects the collected data set into a high-dimensional feature space by adopting a kernel function mapping method, performs serial learning by multiple wavelet variational autoencoders, and uses an unsupervised clustering algorithm for fault identification, thus forming a deep wavelet kernel variational autoencoder. The above diagnostic system can effectively reveal the discriminative weak fault information hidden in small sample data, and complete the accurate diagnosis of the equipment failure mode and degree under fluctuating working conditions. At the same time, it has the excellent characteristics of variational autoencoders such as strong robustness, generalization and anti-interference, and it does not need to rely on data category labels, which can save the cost of manual data category labeling.

[0085] It should be noted that the present invention constructs a deep wavelet kernel variational autoencoder model by integrating kernel function mapping processing, wavelet activation function and unsupervised clustering algorithm into the input layer, hidden layer and output layer of the traditional variational autoencoder respectively. It can not only get rid of the dependence of traditional deep learning models on data category labels, but also reduce learning errors in the model training process, realize accurate identification of fault types, and has good robustness and generalization ability.

[0086] In one example of the present invention,

[0087] The kernel function mapping module is composed of an input layer and a mapping layer connected in series, and is configured to output the data of the input layer after being processed by the kernel function mapping of the mapping layer;

[0088] The deep kernel feature learning module is composed of at least three wavelet variational autoencoders connected in series, and is configured to use the feature vector output by the last wavelet variational autoencoder as a classification vector;

[0089] The classification module is configured to use an unsupervised clustering algorithm to achieve automatic identification of fault types.

[0090] In one example of the present invention, the wavelet variational autoencoder comprises a wavelet encoding module, a wavelet sampling module and a wavelet decoding module which are cascaded in sequence; wherein,

[0091] The wavelet encoding module is composed of an input layer and a wavelet hidden layer connected in series, and is configured to learn the hidden layer feature information;

[0092] The wavelet sampling module is configured to perform feature sampling from hidden layer features that satisfy Gaussian distribution;

[0093] The wavelet decoding module is composed of a wavelet hidden layer and an output layer connected in series, and is configured to reconstruct signals according to the characteristics of the samples to generate interpretable feature data.

[0094] In one example of the present invention, the wavelet hidden layer in the wavelet encoding module uses an asymmetric real Laplace wavelet ψ(t) as an activation function, and its output is q j ,j=1,2,…,m, the specific calculation formula is:

[0095]

[0096] In the formula, f b and f c are the bandwidth and center frequency of the wavelet, respectively, q j is the output of the jth wavelet hidden layer node in the wavelet coding module, d j and c j The scale factor and translation factor are respectively, jk is the weight between the jth wavelet hidden layer node and the kth input layer node.

[0097] Specific implementation cases:

[0098] This case uses a rotating machinery fault simulation test bench as the object to carry out artificial implant bearing fault tests. Figure 4 The figure is a structural diagram of the rotating machinery fault simulation test bench, which is mainly composed of a motor, a coupling, a bearing, a data acquisition system and an electrical control system. During the test, four different faults were machined on bearing 1 using electric spark technology, namely outer ring fault, inner ring fault, outer ring-rolling element composite fault and inner and outer ring composite fault. In addition, when the motor speed is about 1050r / min, the sampling frequency of the signal is set to 10240Hz and the number of sampling points is 10240 points. By installing two acceleration sensors in the vertical direction of bearing 1 and bearing 4 respectively, the bearing vibration data under different fault conditions are collected. There are 50 groups of data samples collected under each bearing health state, of which 25 groups of samples are randomly divided into training sets, and the remaining 25 groups of samples are used as test sets. Figure 5 The time domain waveform of the vibration data of the rolling bearing under different health conditions. Figure 5 It can be seen that the bearing vibration data under different health conditions have certain similarities. It is not easy to accurately identify the failure mode of rolling bearings by directly observing the signal waveform with the naked eye. Therefore, it is urgent to explore an effective intelligent diagnosis method to achieve accurate identification of bearing failure modes.

[0099] The method of the present invention is used to analyze the collected bearing vibration data, and compared with five representative deep learning methods (i.e., variational autoencoder VAE, stacked denoising autoencoder SDAE, convolutional neural network CNN, deep belief network DBN, long short-term memory network LSTM) to verify the effectiveness and superiority of the method of the present invention. First, a training set with a size of 250×10240 is used to train different deep learning models; then, the test set of the same size is input into each trained deep learning model to automatically realize the recognition of different bearing fault modes and obtain equipment health status monitoring and fault diagnosis results. In addition, in order to avoid the randomness of the diagnosis results, each deep learning model is run 10 times to compare their fault recognition performance. Figure 6 Shows the recognition accuracy achieved by different deep learning methods. Figure 6 It can be seen that the method of the present invention can obtain an average recognition accuracy of 99.92%, while the other five representative deep learning methods (i.e., VAE, SDAE, CNN, DBN and LSTM) obtained average recognition accuracies of 96.48%, 97.92%, 94.48%, 86% and 81.22%, respectively. Obviously, the average recognition accuracy of the method of the present invention is higher than that of other deep learning methods. In addition, the volatility of the recognition accuracy of the method of the present invention is relatively small. Therefore, through the above comparative analysis, it can be concluded that the method of the present invention can effectively identify different bearing health states. Moreover, compared with traditional deep learning-based diagnostic methods, the method of the present invention has better fault identification capabilities, is very suitable for fault information learning of small sample monitoring data under fluctuating conditions, and has good engineering application prospects.

[0100] The exemplary implementation scheme of the mechanical fault diagnosis method and system based on the deep wavelet kernel variational autoencoder proposed in the present invention is described in detail above with reference to the preferred embodiments. However, those skilled in the art will appreciate that, without departing from the concept of the present invention, various modifications and variations can be made to the above-mentioned specific embodiments, and various technical features and structures proposed in the present invention can be combined in various ways without exceeding the protection scope of the present invention, which is determined by the attached claims.

Claims

1. A mechanical fault diagnosis method based on deep wavelet kernel variational autoencoder, characterized in that: The steps include: S10: Obtain two-channel nonlinear vibration data for different faults in the electromechanical and hydraulic system x k and The number of sampling points is n, and they are randomly divided into training set and test set in a 1:1 ratio; S20: Connecting the kernel function mapping module, the deep kernel feature learning module and the classification module in series in sequence to form a deep wavelet kernel variational autoencoder; wherein the kernel function mapping module is configured to project the collected data set into a high-dimensional feature space using a kernel function mapping method, the deep kernel feature learning module is configured to use a wavelet variational autoencoder for stacking learning and obtain a classification vector, and the classification module is configured to perform fault classification on the classification vector; S30: The deep wavelet kernel variational autoencoder is trained using the training set to complete the deep extraction of nonlinear fault information, and the test set is input into the trained deep wavelet kernel variational autoencoder to automatically realize the fault diagnosis of the electromechanical hydraulic system.

2. The mechanical fault diagnosis method based on deep wavelet kernel variational autoencoder according to claim 1 is characterized in that: In step S20, The kernel function mapping module is composed of an input layer and a mapping layer connected in series, and is configured to output the data of the input layer after being processed by the kernel function mapping of the mapping layer; The deep kernel feature learning module is composed of at least three wavelet variational autoencoders connected in series, and is configured to use the feature vector output by the last wavelet variational autoencoder as a classification vector; The classification module is configured to use an unsupervised clustering algorithm to achieve automatic identification of fault types.

3. The mechanical fault diagnosis method based on deep wavelet kernel variational autoencoder according to claim 2 is characterized in that: The expression of the classification module is: In the formula, is the output of the third wavelet variational autoencoder, c n is the number of clusters, and clustering(·) is a function in the MATLAB clustering toolkit.

4. The mechanical fault diagnosis method based on deep wavelet kernel variational autoencoder according to claim 2 is characterized in that: The wavelet variational autoencoder comprises a wavelet encoding module, a wavelet sampling module and a wavelet decoding module which are cascaded in sequence; wherein, The wavelet encoding module is composed of an input layer and a wavelet hidden layer connected in series, and is configured to learn the hidden layer feature information; The wavelet sampling module is configured to perform feature sampling from hidden layer features that satisfy Gaussian distribution; The wavelet decoding module is composed of a wavelet hidden layer and an output layer connected in series, and is configured to reconstruct signals according to the characteristics of the samples to generate interpretable feature data.

5. The mechanical fault diagnosis method based on deep wavelet kernel variational autoencoder according to claim 4 is characterized in that: The wavelet hidden layer in the wavelet coding module uses an asymmetric real Laplace wavelet ψ(t) as the activation function, and its output is q j ,j=1,2,…,m, the specific calculation formula is: In the formula, f b and f c are the bandwidth and center frequency of the wavelet, respectively, q j is the output of the jth wavelet hidden layer node in the wavelet coding module, d j and c j The scale factor and translation factor are respectively, jk is the weight between the jth wavelet hidden layer node and the kth input layer node.

6. The mechanical fault diagnosis method based on deep wavelet kernel variational autoencoder according to claim 4 is characterized in that: The activation functions of the wavelet hidden layer and output layer in the wavelet decoding module respectively use asymmetric real Laplace wavelet ψ(t) and hyperbolic tangent function to transform the hidden variable feature z output by the wavelet sampling module into j As the input of the wavelet decoding module, the decoding operation process is carried out, and the output results are p j ,j=1,2,…,m and y k ,k=1,2,…,n, the specific calculation formula is: In the formula, f b and f c are the bandwidth and center frequency of the wavelet, respectively, j is the output of the jth wavelet hidden layer node in the wavelet decoding module, d j and c j They are scale factor and translation factor, W j is the weight between the jth wavelet hidden layer node and the jth sampling node, W kj is the weight between the kth output layer node and the jth wavelet hidden layer node, and tanh(·) is the hyperbolic tangent function.

7. A mechanical fault diagnosis system based on deep wavelet kernel variational autoencoder, characterized in that: include: Signal acquisition and division device for obtaining two-channel nonlinear vibration data for different faults in electromechanical and hydraulic systems k and The number of sampling points is n, and they are randomly divided into training set and test set in a 1:1 ratio; A model building device, used for sequentially connecting a kernel function mapping module, a deep kernel feature learning module and a classification module to form a deep wavelet kernel variational autoencoder; wherein the kernel function mapping module is configured to project the collected data set into a high-dimensional feature space using a kernel function mapping method, the deep kernel feature learning module is configured to use a wavelet variational autoencoder for stacking learning and obtain a classification vector, and the classification module is configured to perform fault classification on the classification vector; The training and fault identification device is used to train the deep wavelet kernel variational autoencoder with a training set to complete the deep extraction of nonlinear fault information, and input the test set into the trained deep wavelet kernel variational autoencoder to automatically realize the fault diagnosis of the electromechanical hydraulic system.

8. The mechanical fault diagnosis system based on deep wavelet kernel variational autoencoder according to claim 7 is characterized in that: The kernel function mapping module is composed of an input layer and a mapping layer connected in series, and is configured to output the data of the input layer after being processed by the kernel function mapping of the mapping layer; The deep kernel feature learning module is composed of at least three wavelet variational autoencoders connected in series, and is configured to use the feature vector output by the last wavelet variational autoencoder as a classification vector; The classification module is configured to use an unsupervised clustering algorithm to achieve automatic identification of fault types.

9. The mechanical fault diagnosis system based on deep wavelet kernel variational autoencoder according to claim 7, characterized in that: The wavelet variational autoencoder comprises a wavelet encoding module, a wavelet sampling module and a wavelet decoding module which are cascaded in sequence; wherein, The wavelet encoding module is composed of an input layer and a wavelet hidden layer connected in series, and is configured to learn the hidden layer feature information; The wavelet sampling module is configured to perform feature sampling from hidden layer features that satisfy Gaussian distribution; The wavelet decoding module is composed of a wavelet hidden layer and an output layer connected in series, and is configured to reconstruct signals according to the characteristics of the samples to generate interpretable feature data.

10. The mechanical fault diagnosis system based on deep wavelet kernel variational autoencoder according to claim 7, characterized in that: The wavelet hidden layer in the wavelet coding module uses an asymmetric real Laplace wavelet ψ(t) as the activation function, and its output is q j ,j=1,2,…,m, the specific calculation formula is: In the formula, f b and f c are the bandwidth and center frequency of the wavelet, respectively, q j is the output of the jth wavelet hidden layer node in the wavelet coding module, d j and c j The scale factor and translation factor are respectively, jk is the weight between the jth wavelet hidden layer node and the kth input layer node.

Images