A Method, System, Storage Medium and Device for Fault Diagnosis of Machine Tool Bearings

Abstract

The present disclosure provides a machine tool bearing fault diagnosis method, system, storage medium and device, relating to the technical field of rotating machinery fault diagnosis, including: acquiring the vibration signal of the machine tool bearing; inputting the vibration signal into a fault diagnosis model, first entering a sparse wavelet convolution module, the sparse wavelet convolution module is restricted by introducing spectral kurtosis constraint and sparse constraint, and the preliminary features are output; inputting the preliminary features into a gated pulse module, in the gated pulse module, first, the pulse neuron characteristic is used to output by adjusting the gating factor to release a pulse, and then the pulse output at the previous moment and the preliminary features at the current moment are used as inputs for transformation, and a spike sequence is output, inputting the spike sequence into a deep neural network module, outputting a fault information feature vector, inputting the fault information feature vector into a classifier, and finally outputting a fault classification result to realize fault diagnosis.

Description

Technical Field

[0001] The present disclosure relates to the technical field of rotating machinery fault diagnosis, and particularly to a machine tool bearing fault diagnosis method, system, storage medium and device. Background Art

[0002] The statements in this section merely provide background technical information related to the present disclosure and do not necessarily constitute prior art.

[0003] Machine tool bearing faults are a key issue in modern industrial manufacturing. Their influence is not limited to the performance of the machine tool itself, but also runs through the entire production chain, including machining accuracy, tool life, production efficiency, and product quality. Bearings are basic components of machine tools and are widely used in industries such as automotive, power, construction, agriculture, healthcare, and aerospace. Bearing faults can not only cause equipment downtime and economic losses, but may also trigger serious safety accidents, resulting in significant losses. Therefore, machine tool bearing fault diagnosis is crucial.

[0004] Mechanical vibration signals often contain rich fault feature information. Fault features can be extracted through neural networks for identifying and diagnosing the fault types of mechanical equipment. However, the traditional neural network fault diagnosis method has the following problems:

[0005] (1) The convolution kernels of the convolutional layers of neural networks mostly adopt the random initialization method, which affects the convergence speed of the network and the quality of feature extraction;

[0006] (2) There is a large amount of noise in the vibration signals collected by sensors, making feature extraction difficult. Summary of the Invention

[0007] To solve the above problems, the present disclosure proposes a machine tool bearing fault diagnosis method, system, storage medium and device, constructs a sparse wavelet convolution module and introduces spectral kurtosis constraint and sparse constraint, which can better select the frequency band for fault feature extraction and remove redundant features, constructs a gated pulse module, and uses the characteristics of gated pulse neurons to reduce the interference of noise, improve the signal-to-noise ratio of the signal, and achieve the function of eliminating noise as much as possible while retaining fault features.

[0008] According to some embodiments, the present disclosure adopts the following technical solutions:

[0009] A machine tool bearing fault diagnosis method includes:

[0010] Obtain the vibration signal of the machine tool bearing and preprocess it;

[0011] Input the preprocessed vibration signal into the fault diagnosis model and output the fault classification result;

[0012] Among them, after the preprocessed vibration signal is input into the fault diagnosis model, it first enters the sparse wavelet convolution module. The sparse wavelet convolution module is restricted by introducing spectral kurtosis constraint and sparse constraint, and the preliminary features are output. The preliminary features are then input into the gated pulse module. In the gated pulse module, first, the characteristics of the pulse neuron are used to release pulses by adjusting the gating factor for output, and then the pulse output of the previous moment and the preliminary features of the current moment are used as inputs for transformation, and a spike train is output. The spike train is then input into the deep neural network module, and a fault information feature vector is output. The fault information feature vector is input into the classifier, and finally, the fault classification result is output to achieve fault diagnosis.

[0013] According to some embodiments, the present disclosure adopts the following technical solutions:

[0014] A machine tool bearing fault diagnosis system includes:

[0015] A signal acquisition module for acquiring the vibration signal of the machine tool bearing and performing preprocessing;

[0016] A fault classification module for inputting the preprocessed vibration signal into the fault diagnosis model and outputting the fault classification result;

[0017] Among them, after the preprocessed vibration signal is input into the fault diagnosis model, it first enters the sparse wavelet convolution module. The sparse wavelet convolution module is restricted by introducing spectral kurtosis constraint and sparse constraint, and the preliminary features are output. The preliminary features are then input into the gated pulse module. In the gated pulse module, first, the characteristics of the pulse neuron are used to release pulses by adjusting the gating factor for output, and then the pulse output of the previous moment and the preliminary features of the current moment are used as inputs for transformation, and a spike train is output. The spike train is then input into the deep neural network module, and a fault information feature vector is output. The fault information feature vector is input into the classifier, and finally, the fault classification result is output to achieve fault diagnosis.

[0018] According to some embodiments, the present disclosure adopts the following technical solutions:

[0019] A computer program product includes a computer program, and when the computer program is executed by a processor, it implements the described method for diagnosing faults in machine tool bearings.

[0020] According to some embodiments, the present disclosure adopts the following technical solutions:

[0021] A non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the described method for diagnosing faults in machine tool bearings is implemented.

[0022] According to some embodiments, the present disclosure adopts the following technical solutions:

[0023] An electronic device comprises: a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes and implements the machine tool bearing fault diagnosis method.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] A machine tool bearing fault diagnosis method disclosed in the present invention constructs a fault diagnosis model, which is initialized by a wavelet convolution kernel in a sparse wavelet convolution module. The kernel of the first convolution layer is a wavelet convolution kernel with trainable translation and scale parameters, and allows the most appropriate wavelet kernel to be selected for fault diagnosis, thereby improving the model's sensitivity to transient shock extraction, enhancing interpretability, and performing preliminary denoising on the signal.

[0026] A machine tool bearing fault diagnosis method disclosed in the present invention introduces spectral kurtosis constraint and sparsity constraint in a sparse wavelet convolution module. Spectral kurtosis, as a typical sparsity measure, is introduced into the module to better select the frequency band for fault feature extraction. Although initialization using a wavelet convolution kernel can produce more interpretable and representative features, the diversification of the selection of translation parameters, scale parameters and wavelet basis functions may hinder the improvement of fault diagnosis accuracy. Therefore, a sparse constraint module is introduced to screen the extracted features, which can better select the frequency band for fault feature extraction and remove redundant features. The sparse wavelet convolution kernel and spectral kurtosis constraint terms are used to enrich the physical meaning of the model and improve the interpretability of the model. At the same time, the module effectively extracts features containing fault features, thereby improving the performance of fault diagnosis.

[0027] A machine tool bearing fault diagnosis method disclosed in the present invention constructs a gated pulse module. The core unit of the module is a pulse neuron. When it receives a signal, it will change its membrane potential. When the membrane potential exceeds a threshold, the neuron will release a pulse for output. In order to improve the noise reduction performance and interpretability, the gated pulse module integrates multiple types of pulse neurons through the gating unit, and uses the characteristics of the gated pulse neuron to further reduce the interference of noise, improve the signal-to-noise ratio of the signal, and achieve the function of eliminating noise as much as possible while retaining the fault characteristics.

[0028] A machine tool bearing fault diagnosis method according to the present disclosure introduces a variety of gating factors into the gating pulse module, enabling the model to have good stability while maintaining a high response level. For fault diagnosis tasks under various working conditions, by adjusting the gating factors, the performance of the gating pulse module can be maximized, making the model have good stability and being able to fully utilize fault feature information. Brief Description of the Drawings

[0029] The accompanying drawings forming a part of the present disclosure are used to provide a further understanding of the present disclosure. The illustrative embodiments and descriptions thereof of the present disclosure are used to explain the present disclosure and do not constitute an improper limitation of the present disclosure.

[0030] Figure 1 Schematic flow diagram of a machine tool bearing fault diagnosis method according to an embodiment of the present disclosure;

[0031] Figure 2 Original sample signal according to an embodiment of the present disclosure;

[0032] Figure 3 Output signal of the sparse wavelet convolution module according to an embodiment of the present disclosure;

[0033] Figure 4 Output signal of the gating pulse module according to an embodiment of the present disclosure;

[0034] Figure 5 Characteristic visualization qualitative analysis diagram according to an embodiment of the present disclosure. Detailed Description of the Preferred Embodiments

[0035] The present disclosure will be further described below in conjunction with the accompanying drawings and embodiments.

[0036] It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present disclosure belongs.

[0037] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0038] Embodiment 1

[0039] An embodiment of the present disclosure provides a machine tool bearing fault diagnosis method, and the steps include:

[0040] Step 1: Obtain the vibration signal of the machine tool bearing and preprocess it;

[0041] Step 2: Input the preprocessed vibration signal into the fault diagnosis model and output the fault classification result;

[0042] Among them, after the preprocessed vibration signal is input into the fault diagnosis model, it first enters the sparse wavelet convolution module. The sparse wavelet convolution module is restricted by introducing spectral kurtosis constraint and sparse constraint, and outputs the preliminary features. The preliminary features are then input into the gated pulse module. In the gated pulse module, first, the characteristics of pulse neurons are used to release pulses by adjusting the gating factor for output. Then, the pulse output of the previous moment and the preliminary features of the current moment are used as inputs for transformation, and a spike sequence is output. The spike sequence is then input into the deep neural network module, and a fault information feature vector is output. The fault information feature vector is input into the classifier, and finally, the fault classification result is output to achieve fault diagnosis.

[0043] As an embodiment, a machine tool bearing fault diagnosis method of the present disclosure constructs a fault diagnosis model. In the fault diagnosis model, the wavelet convolution kernel of the sparse wavelet convolution module is used to initialize the deep neural network, and the gated pulse module is used to suppress noise, so as to effectively extract the fault impact components and improve the accuracy of fault diagnosis. The specific implementation process is as follows:

[0044] Step 1: Obtain the vibration signal of the machine tool bearing and preprocess it;

[0045] Specifically, the vibration signal of the machine tool bearing can be collected by a sensor, or the vibration signal of the machine tool bearing uploaded by the collection terminal can be obtained.

[0046] The preprocessing process is to segment the collected vibration signal to form samples for subsequent training of the deep neural network.

[0047] Step 2: Input the preprocessed vibration signal into the fault diagnosis model and output the fault classification result;

[0048] Specifically, the fault diagnosis module includes a sparse wavelet convolution module, a gated pulse module, a deep neural network module, and a classifier. Among them, the sparse wavelet convolution module consists of two parts, including a convolutional layer containing a wavelet convolution kernel and a sparse vector generation module composed of two fully connected layers and two fully connected layers.

[0049] The gated pulse module is composed of a series of gated pulse neurons with shared parameters in series. Each neuron takes the output of the previous neuron and the features extracted by the sparse wavelet convolution module as inputs.

[0050] The deep neural network module is composed of four residual blocks, an average pooling layer, and a fully connected layer connected in series. In addition, this module can also be freely composed of multiple convolutional layers, pooling layers, and fully connected layers.

[0051] As an embodiment, the steps of inputting the preprocessed vibration signal into the fault diagnosis model and outputting the fault classification result include:

[0052] Step 21: After the preprocessed vibration signal is input into the fault diagnosis model, it first enters the sparse wavelet convolution module. The wavelet convolution kernel with trainable translation and scale parameters is used to initialize the network. The spectral kurtosis constraint and sparse constraint are introduced to better select the frequency band for fault feature extraction and remove redundant features, and the preliminary features are output.

[0053] Specifically, the design of the sparse wavelet convolution module improves the sensitivity of the model to transient impact extraction and allows the selection of the most suitable wavelet kernel for fault diagnosis. It improves the performance of the model and enhances the interpretability. The kernel of the first convolutional layer in the module is initialized by a wavelet convolution kernel with trainable translation and scale parameters. The initialization formula of the wavelet convolution kernel is as follows:

[0054] (1)

[0055] Where, u , s respectively represent the translation and scale parameters, t represents the index of the specific value of the convolution kernel, ψ represents the selected wavelet basis function. Thus, the first layer operation of TDResNet can be expressed as:

[0056] (2)

[0057] Where, x raw is the input time-domain signal sample, h is the output after convolution, and * is the convolution operation. Here, three different wavelet convolution kernels of Laplace, Mexh, and Morlet are used, and the scale parameters on each channel of the convolution kernel are uniformly distributed in the ranges of (0.1, 2), (0.1, 3), and (0.1, 4.5) respectively.

[0058] As a typical sparsity measure, spectral kurtosis is introduced into this module to better select the frequency band for fault feature extraction. The spectral kurtosis constraint term is expressed as:

[0059] (3)

[0060] Where, I represents the number of features in, i denote the i th feature in L denote i length of i ( l ) denote i the l th element of denote h envelope of , and the specific expression is:

[0061] (4)

[0062] Initializing the wavelet convolution kernel can generate more interpretable and representative features. However, the diversity in the selection of translation parameters, scale parameters, and wavelet basis functions may hinder the improvement of the fault diagnosis accuracy. Therefore, the present disclosure further introduces a sparsity constraint to screen the extracted features.

[0063] In the sparsity constraint process, a random vector is first transformed to half of the original length through a fully connected layer, then transformed by the ReLU activation function, and subsequently restored to the original length through another fully connected layer. Finally, these values are normalized to the range (0, 1) using the sigmoid function to obtain the final sparse vector V , and then the input feature vector is multiplied by the sparse vector V to obtain the output vector, and the expression is:

[0064] (5)

[0065] where x sparse is the output of the preliminary features, h is the output after calculating the convolution, V is the sparse vector.

[0066] To remove the channels that have little impact on the results and retain the relatively critical channels containing fault features, the present disclosure further introduces a sparsity constraint, which is introduced to screen the extracted features. The sparsity constraint process introduces an L1-norm sparsity constraint term, and its expression is:

[0067] (6)

[0068] The sparse wavelet convolution module enriches the physical meaning of the model by using the sparse wavelet convolution kernel and the spectral kurtosis constraint term, improves the interpretability of the model, and at the same time, this module effectively extracts the features containing fault features, thereby improving the performance of fault diagnosis.

[0069] Step 22: Use the gated pulse module to perform deep denoising and further feature extraction to obtain the final fault information feature vector for subsequent fault diagnosis; wherein, the core unit of the gated pulse module is the pulse neuron, and the gated pulse module is composed of a plurality of pulse neurons with shared parameters connected in series. Each pulse neuron takes the output of the previous pulse neuron and the preliminary features of the output of the sparse wavelet convolution module as input. When receiving the input, it will change its membrane potential. When the membrane potential exceeds the threshold, the pulse neuron will release a pulse for output.

[0070] Specifically, after the vibration signal passes through the sparse wavelet convolution module, the amplitude of the extracted preliminary features in the fault impact section is relatively high, while for noise, even if its frequency, amplitude and other characteristics coincide with the characteristics of the wavelet convolution kernel used, its amplitude is relatively small. In order to further reduce the interference of noise and improve the signal-to-noise ratio of the signal, the present disclosure proposes a gated pulse module, the core unit of which is a pulse neuron, which changes its membrane potential when it receives a signal, and releases a pulse for output when the membrane potential exceeds the threshold. For features that conform to the wavelet characteristics, the membrane potential will continue to increase until it discharges, and for noise mixed in the features, due to the presence of leakage units in the pulse neurons, it is difficult to generate pulses relying solely on the intensity of the noise.

[0071] The gated pulse module can achieve the function of preserving fault characteristics while eliminating noise as much as possible. The gated pulse module is composed of a series of pulse neurons with shared parameters, each of which takes the output of the previous neuron and the features extracted by the sparse wavelet convolution module as input. Therefore, the preliminary features of the output of the sparse wavelet module can be converted into a spike sequence of pulses with a high signal-to-noise ratio. x G The specific formula is as follows:

[0072] (7)

[0073] (8)

[0074] Among them, the superscript ( t ) is the identifier of the time step, H(·) represents the Heaviside step function, U is the membrane potential vector, U th represents the potential threshold, L represents the membrane potential vector after the leak, I represents the incremental potential vector, F The vector representing the potential decrease caused by the release pulse, represents the Hadamard product. The specific formula of each part in equation (8) is as follows:

[0075] (9)

[0076] (10)

[0077] (11)

[0078] Among them, α G , β G , and γ G are gating factors, and their value ranges are between (0, 1). τ lin is the linear decay exponent, τ exp is the exponential decay exponent. g (t) represents the conductance exponent. U re is the soft reset value, L exp is obtained from α G τ exp U (t-1) + (1 - α G ) U (t -1) , indicating the value reduced when the neuron model performs a hard reset only on the part of the membrane potential related to the exponential decay.

[0079] To improve the noise reduction performance and interpretability, the gated pulse module fuses multiple types of spiking neurons through a gating unit. Specifically, in Equation (9), τ exp U (t-1) is the exponential decay term, ([[]] U (t-1) - τ lin ) is the linear decay term. The exponential decay mechanism helps to stabilize the neuron dynamics. However, this decay may cause the membrane potential to not reach the threshold required to release a spike. The linear decay reduces the membrane potential by a fixed value at each time step, thus maintaining a high response level. However, in the absence of input, the linear decay will cause the potential to decrease infinitely, leading to a loss of stability. Therefore, the present disclosure introduces multiple gating factors into the gated pulse module, and the gating factor α GFor fusing these two methods, enabling the model to have good stability while maintaining a high response level. In Equation (10), g (t) represents the time-varying integral weight, which helps to better process time features and provides a flexible integral method for the model. The second term is a fixed value, representing a stable and uniform integral method. The gating factor β G is used to fuse these two integral methods to achieve the flexibility of the integral process. In Equation (11), L exp represents the potential reduction in the hard reset mode, which can reduce the membrane potential to a fixed value, providing better stability for the module. However, this simple and direct reset method may lead to the loss of feature information. U re represents the potential reduction in the soft reset mode, which better retains feature information by reducing the fixed value of the membrane potential. The gating factor γ G is used to fuse these two reset modes, enabling the model to have good stability and fully utilize feature information at the same time. When the gating factor is set to 0 or 1, the gating pulse module degrades to a single mode. When the gating parameter takes a value between 0 and 1, the gating pulse module transforms into a mixture of two pulse neurons. For the fault diagnosis task under various working conditions, the performance of the gating pulse module can be maximized by adjusting the gating factor.

[0080] In the gating pulse module, t the pulse output at time - 1 and t the output features of the sparse wavelet module at time are used as the input of the gating pulse module at time t , and the output spike train effectively retains the feature information. Specifically, this module stores the feature information in the membrane potential vector U and expresses it through a series of pulses. In addition, due to the existence of the unique leakage unit of the pulse neuron, it can effectively remove the noise mixed in the features, which enables the entire model to more efficiently utilize the samples, thereby improving the accuracy of fault diagnosis.

[0081] This disclosure uses a sparse wavelet convolution module for preliminary feature extraction. The module initializes the network using wavelet convolution kernels with trainable translation and scale parameters to improve the network's sensitivity to transient shock feature extraction. The spectral kurtosis constraint is introduced to constrain the features extracted by the wavelet convolution kernel to better select the frequency band for fault feature extraction. The sparse constraint is introduced to screen the extracted features and remove redundant features. The gating pulse module uses gating pulse neurons to further extract the features, reduce the noise in the signal, improve the signal - to - noise ratio, and finally use the extracted features for fault diagnosis.

[0082] Step 23: Input the spike train into the deep neural network module, output the fault information feature vector, input the fault information feature vector into the classifier, and finally output the fault classification result to achieve fault diagnosis.

[0083] Specifically, after obtaining the spike train output by the gating pulse module, input it into the deep neural network to obtain the fault information feature vector containing fault information. z Input the fault information feature vector z into the classifier to obtain the classification result. p i,c , and then use the cross-entropy loss l c to measure the classification error. The specific formula is as follows:

[0084] (12)

[0085] where, N represents the number of samples, n c represents the number of fault status categories. y i,c is an indicator function. If the i -th sample belongs to c class, then y i,c = 1; otherwise, y i,c = 0. p i,c represents the predicted probability that the i -th sample belongs to category c . The overall objective function of the fault diagnosis model is as follows:

[0086] (13)

[0087] where, θ are the network parameters, λ 1, λ 2 are hyperparameters for balancing each loss term, l c , l SK , l sparse are respectively the classification error loss term, spectral kurtosis loss term and sparsity loss term mentioned above. By continuously training the model until the network parameters converge, and then inputting the samples to be diagnosed into the network, the fault diagnosis result can be obtained.

[0088] Simulation experiment

[0089] The method of the present disclosure was experimentally tested on the KAIST rotating machinery fault diagnosis dataset. This dataset was provided by the Department of Mechanical Engineering at the Korea Advanced Institute of Science and Technology. Its rotating machine test bench consists of an electric motor, a torque sensor, a gearbox, a bearing test module, and a rotor module. This dataset includes three different load conditions, each of which contains common bearing and rotor faults. The dataset contains four types of faults: normal, inner race bearing fault, outer race bearing fault, and shaft misalignment.

[0090] The sparse wavelet convolution module and gated impulse module framework proposed in the present disclosure were applied to the fault feature extraction of this vibration signal. Here, three different wavelet convolution kernels, namely Laplace, Mexh, and Morlet, were used. The scale parameters on each channel of the convolution kernel were uniformly distributed in the ranges of (0.1, 2), (0.1, 3), and (0.1, 4.5), respectively. The gated factor parameter and the hyperparameters for balancing each loss term were set as follows: α G = 0.4, β G = 0.4, γ G = 0.5, λ 1 = 0.2, λ 2 = 0.3.

[0091] As Figures 2 - 5 shown, the experimental results show that the proposed sparse wavelet convolution gated impulse framework can effectively extract fault features and reduce the interference of noise. Under the qualitative analysis of feature visualization, it can be seen that the features extracted from samples of the same fault type are clustered, while there are obvious boundaries between the features extracted from samples of different fault types, which proves that the model has good fault diagnosis ability and provides an effective solution for the fault diagnosis of machine tool bearings.

[0092] As an embodiment, in other embodiments, the selection of the wavelet convolution kernel and the setting of the gated factor value can be changed according to specific situations and requirements, and are not limited to the above values.

[0093] Embodiment 2

[0094] An embodiment of the present disclosure provides a machine tool bearing fault diagnosis system, including:

[0095] A signal acquisition module, configured to acquire the vibration signal of the machine tool bearing and perform preprocessing;

[0096] A fault classification module, configured to input the preprocessed vibration signal into the fault diagnosis model and output a fault classification result;

[0097] Among them, after the preprocessed vibration signal is input into the fault diagnosis model, it first enters the sparse wavelet convolution module. The sparse wavelet convolution module is restricted by introducing spectral kurtosis constraint and sparse constraint, and outputs preliminary features. The preliminary features are then input into the gated pulse module. In the gated pulse module, first, the characteristics of pulse neurons are used to output pulses by adjusting the gating factor, and then the pulse output at the previous moment and the preliminary features at the current moment are used as inputs for transformation, and a spike train is output. The spike train is then input into the deep neural network module, and a fault information feature vector is output. The fault information feature vector is input into a classifier, and finally, a fault classification result is output to achieve fault diagnosis.

[0098] Embodiment 3

[0099] In an embodiment of the present disclosure, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, it implements the described machine tool bearing fault diagnosis method.

[0100] Embodiment 4

[0101] In an embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided. The non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the described machine tool bearing fault diagnosis method is implemented.

[0102] Embodiment 5

[0103] In an embodiment of the present disclosure, an electronic device is provided, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory so that the electronic device executes and implements the described machine tool bearing fault diagnosis method.

[0104] The present disclosure is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the specified functions in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0105] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable apparatus to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one process or a plurality of processes and / or blocks Figure 1 one process or a plurality of processes and / or blocks Figure 1 steps of the functions specified in one block or a plurality of blocks.

[0106] Although the specific embodiments of the present disclosure have been described above in conjunction with the accompanying drawings, they are not intended to limit the scope of protection of the present disclosure. Those skilled in the art should understand that, based on the technical solutions of the present disclosure, various modifications or deformations that can be made by those skilled in the art without creative efforts are still within the scope of protection of the present disclosure.

Claims

1. A method for diagnosing faults in machine tool bearings, characterized in that, Including: Obtain the vibration signal of the machine tool bearing and preprocess it; Input the preprocessed vibration signal into the fault diagnosis model and output the fault classification result; Among them, after the preprocessed vibration signal is input into the fault diagnosis model, it first enters the sparse wavelet convolution module. The sparse wavelet convolution module is restricted by introducing spectral kurtosis constraint and sparse constraint, and outputs the preliminary features. Then input the preliminary features into the gated pulse module. In the gated pulse module, first use the characteristics of the pulse neuron to release pulses by adjusting the gating factor for output, and then use the pulse output of the previous moment and the preliminary features of the current moment as inputs for transformation, and output the spike train. Then input the spike train into the deep neural network module, output the fault information feature vector, input the fault information feature vector into the classifier, and finally output the fault classification result to realize fault diagnosis.

2. The method for diagnosing faults of a machine tool bearing according to claim 1, characterized in that, The fault diagnosis module includes a sparse wavelet convolution module, a gated pulse module, a deep neural network module and a classifier. The sparse wavelet convolution module first initializes the convolutional layer with a wavelet convolution kernel with trainable translation and scale parameters to improve the sensitivity of transient shock feature extraction.

3. The method for diagnosing faults of a machine tool bearing according to claim 1, wherein, The sparse wavelet convolution module is restricted by introducing spectral kurtosis constraint and sparse constraint, including: spectral kurtosis, as a typical sparsity measure, is introduced to select the frequency band for fault feature extraction, and sparse constraint is introduced to screen the extracted features, effectively extracting the preliminary features containing fault features.

4. The method for diagnosing faults of a machine tool bearing according to claim 1, wherein, The core unit of the gated pulse module is the pulse neuron. The gated pulse module is composed of multiple pulse neurons with shared parameters connected in series. Each pulse neuron uses the output of the previous pulse neuron and the preliminary features output by the sparse wavelet convolution module as inputs. When receiving the input, it will change its membrane potential. When the membrane potential exceeds the threshold, the pulse neuron will release pulses for output.

5. The machine tool bearing fault diagnosis method according to claim 4, characterized in that, A variety of gating factors are introduced in the gated pulse module to maximize the performance of the gated pulse module by adjusting the gating factors. When the gating factor is set to 0 or 1, the gated pulse module degenerates into a single mode; when the gating parameter takes a value between 0 and 1, the gated pulse module transforms into a mixture of two pulse neurons.

6. The method for diagnosing faults of a machine tool bearing according to claim 1, wherein, After the gated pulse module outputs the spike train, input it into the deep neural network module to obtain the fault information feature vector containing fault information. Input the fault information feature vector into the classifier to obtain the fault classification result. Then use the cross-entropy loss to measure the classification error, and construct the overall objective function of the fault diagnosis model, including the classification error loss term, the spectral kurtosis loss term and the sparse loss term respectively.

7. A machine tool bearing fault diagnosis system, characterized in that, Including: A signal acquisition module for obtaining the vibration signal of the machine tool bearing and preprocessing it; A fault classification module for inputting the preprocessed vibration signal into the fault diagnosis model and outputting the fault classification result; Among them, after the preprocessed vibration signal is input into the fault diagnosis model, it first enters the sparse wavelet convolution module. The sparse wavelet convolution module is restricted by introducing spectral kurtosis constraint and sparse constraint, and the preliminary features are output. The preliminary features are then input into the gated pulse module. In the gated pulse module, first, the characteristics of pulse neurons are used to output pulses by adjusting the gating factor, and then the pulse output of the previous moment and the preliminary features of the current moment are used as inputs for transformation, and the spike sequence is output. The spike sequence is then input into the deep neural network module, and the fault information feature vector is output. The fault information feature vector is input into the classifier, and finally the fault classification result is output to realize fault diagnosis.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the machine tool bearing fault diagnosis method according to any one of claims 1-6.

9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, it implements the machine tool bearing fault diagnosis method according to any one of claims 1-6.

10. An electronic device, characterized in that, Including: A processor, a memory, and a computer program; wherein, the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device runs, the processor executes the computer program stored in the memory so that the electronic device implements the machine tool bearing fault diagnosis method according to any one of claims 1-6.

Images