A fault diagnosis method for mechanical rotating parts based on stacked convolutional networks

Abstract

The present invention provides a method for diagnosing mechanical rotating component faults based on a stacked convolutional network. The method includes receiving vibration acceleration signals when a rated number of mechanical rotating components fail, wherein different fault types corresponding to the vibration acceleration signals are identified; random vibration acceleration signal samples are randomly selected from a rated number of vibration acceleration signals; pulse enhancement and noise reduction are performed on the random vibration acceleration signal samples through a stacked convolution module, and fault feature samples are extracted; and a fault diagnosis model is obtained based on the fault feature samples for fault diagnosis. Pulse enhancement and noise reduction are performed through a stacked convolution module, and fault feature samples are extracted to train a fault diagnosis model, which helps to solve the technical problem of the lack of a mechanical component fault diagnosis method in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of mechanical component fault diagnosis, and in particular to a mechanical component fault diagnosis method, system, device, and storage medium. Background Art

[0002] With the improvement of information technology and the rapid development of modern industry, mechanical equipment is gradually moving towards integration, electrification and automation, and its equipment structure is becoming more and more sophisticated and complex. Mechanical rotating parts are key components in mechanical equipment. Once a failure occurs, the entire equipment will not be able to operate, which may increase downtime or cause a lot of economic losses or even casualties. Therefore, conducting state detection and fault diagnosis technology research can maintain the industrial production safety of mechanical equipment, which has important application value and potential economic benefits.

[0003] Data samples that reflect the operating status of mechanical rotating parts are mostly collected by vibration acceleration sensors. However, the data samples collected in actual work are also mixed with various vibration source signals from other equipment, components, etc., as well as inevitable environmental background noise. This makes the signals that can reflect the effective fault information of mechanical rotating parts seriously interfered or even completely submerged in the noise, greatly increasing the difficulty of fault type identification.

[0004] Therefore, there is an urgent need for a mechanical component fault diagnosis method, system, device, and storage medium, which can help solve the technical problem of the lack of a mechanical component fault diagnosis method in the prior art. Summary of the invention

[0005] In one embodiment, the present invention provides a mechanical component fault diagnosis method, which performs pulse enhancement and noise reduction through a stacked morphological convolution module and extracts fault feature samples to train a fault diagnosis model, thereby helping to solve the technical problem of the lack of a mechanical component fault diagnosis method in the prior art.

[0006] The method includes:

[0007] receiving vibration acceleration signals when a rated number of mechanical rotating components fail, wherein different fault types corresponding to the vibration acceleration signals are identified;

[0008] Randomly selecting a rated number of the vibration acceleration signals to obtain random vibration acceleration signal samples;

[0009] Performing pulse enhancement and noise reduction on the random vibration acceleration signal samples through a stacked morphological convolution module and extracting fault feature samples;

[0010] A fault diagnosis model is obtained by training the fault feature samples for fault diagnosis.

[0011] In one embodiment, the stacked morphological convolution module is a parallel branch with different morphological operators, and the different morphological operators include an opening operation and a closing operation;

[0012] The opening operation is used to perform erosion and expansion operations in sequence, and the closing operation is used to perform expansion and erosion operations in sequence.

[0013] In one embodiment, the performing pulse enhancement and noise reduction on the random vibration acceleration signal sample through a stacked morphological convolution module and extracting the fault feature sample comprises:

[0014] retaining negative pulses using an open operation according to the random vibration acceleration signal, and extracting positive pulses using a closed operation according to the random vibration acceleration signal;

[0015] Extracting components related to fault characteristics from the negative pulse and the positive pulse to obtain corresponding negative pulse components and positive pulse components;

[0016] The fault feature sample is obtained by fusing the negative pulse component and the positive pulse component.

[0017] In one embodiment, the step of obtaining the fault feature sample by fusing the negative pulse component and the positive pulse component comprises:

[0018] Obtaining a fusion feature map by fusing the negative pulse component and the positive pulse component;

[0019] The weight parameters are obtained according to the influence of different fault categories on different characteristic channels;

[0020] Obtaining feature sizes of different channels according to the weight parameters and the different feature channels;

[0021] The fault feature template is obtained by applying preset attention parameters according to the feature sizes of different channels.

[0022] In one embodiment, after the step of obtaining a fault diagnosis model by training the fault feature samples for fault diagnosis, the method further comprises:

[0023] The fault diagnosis model is optimized using a back propagation algorithm through a Softmax cross entropy loss function.

[0024] In one embodiment, the present invention further provides a mechanical component fault diagnosis system, the mechanical component fault diagnosis system comprising a background system;

[0025] The background system is used to receive vibration acceleration signals when a rated number of mechanical rotating parts fail, wherein different fault types corresponding to the vibration acceleration signals are identified; a rated number of vibration acceleration signals are randomly selected to obtain random vibration acceleration signal samples; the random vibration acceleration signal samples are pulse enhanced and denoised through a stacked morphological convolution module and fault feature samples are extracted; and a fault diagnosis model is obtained by training according to the fault feature samples for fault diagnosis.

[0026] In one embodiment, the mechanical component fault diagnosis system includes a vibration acceleration sensor;

[0027] The vibration acceleration sensor is used to collect vibration acceleration signals.

[0028] In one embodiment, the present invention further provides a mechanical component fault diagnosis device, the device comprising:

[0029] A receiving module, used for receiving vibration acceleration signals when a rated number of mechanical rotating parts fail, wherein different fault types corresponding to the vibration acceleration signals are identified;

[0030] A random module, used for randomly selecting a rated number of the vibration acceleration signals to obtain random vibration acceleration signal samples;

[0031] An enhancement and noise reduction module, used for performing pulse enhancement and noise reduction on the random vibration acceleration signal sample through a stacked morphological convolution module and extracting fault feature samples;

[0032] The training module is used to obtain a fault diagnosis model for fault diagnosis based on the fault feature samples.

[0033] In one embodiment, the present invention further provides a mechanical component fault diagnosis device, the device comprising: a processor and a memory;

[0034] The memory stores an application program executable by the processor, which is used to enable the processor to execute the steps of the mechanical component fault diagnosis method.

[0035] In one embodiment, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the mechanical component fault diagnosis method when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A schematic diagram of a process 100 for diagnosing a mechanical component fault in one embodiment of the present invention;

[0037] Figure 2A schematic diagram of a process 200 for diagnosing a mechanical component fault in another embodiment of the present invention;

[0038] Figure 3 A schematic diagram of the data architecture of the SMCN structural block diagram in another embodiment of the present invention;

[0039] Figure 4 A schematic diagram of a process 300 for diagnosing a mechanical component fault in another embodiment of the present invention;

[0040] Figure 5 A schematic diagram of a process 400 for diagnosing a mechanical component fault in another embodiment of the present invention;

[0041] Figure 6 FIG. 4 is a schematic diagram of an analog circuit fault detection device in another embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0043] Various aspects and features of the present application are described herein with reference to the accompanying drawings.

[0044] These and other characteristics of the present application will become apparent from the following description of a preferred form of embodiment given as a non-limiting example with reference to the accompanying drawings.

[0045] It should also be understood that, although the present application has been described with reference to some specific examples, those skilled in the art will be able to realize many other equivalent forms of the present application that have the features described in the claims and are therefore within the scope of protection defined thereby.

[0046] The above and other aspects, features and advantages of the present application will become more apparent in view of the following detailed description when taken in conjunction with the accompanying drawings.

[0047] Specific embodiments of the present application are described hereinafter with reference to the accompanying drawings; however, it should be understood that the embodiments applied for are merely examples of the present application, which may be implemented in a variety of ways. Well-known and / or repeated functions and structures are not described in detail to determine the true intent based on the user's historical operations and to avoid unnecessary or redundant details that make the present application unclear. Therefore, the specific structural and functional details applied for herein are not intended to be limiting, but are merely used as the basis and representative basis for the claims to teach those skilled in the art to use the present application in a variety of ways with substantially any suitable detailed structure.

[0048] This specification may use the phrases "in one embodiment," "in another embodiment," "in yet another embodiment," or "in other embodiments," all of which may refer to one or more of the same or different embodiments according to the present application.

[0049] The inventors have discovered that for fault diagnosis of mechanical parts, such as bearing parts when they are rotating, relevant parameters of their vibration can be collected, and a fault diagnosis model can be obtained through analysis based on the collected parameters through training, thereby realizing fault diagnosis of mechanical parts.

[0050] Figure 1 FIG. 1 is a schematic diagram of a process 100 for diagnosing a mechanical component fault in an embodiment of the present invention. Figure 1 As shown, in one embodiment, the present invention also provides a mechanical component fault diagnosis method, the method comprising:

[0051] S101, receiving vibration acceleration signals when a rated number of mechanical rotating components fail, wherein different fault types corresponding to the vibration acceleration signals are identified.

[0052] In this step, a specific step of receiving vibration acceleration signals when a rated number of mechanical rotating components fail is provided, wherein different fault types corresponding to the vibration acceleration signals are identified.

[0053] Specifically, when a mechanical rotating part fails, the output vibration signal will change accordingly. In this step, the vibration acceleration signal of the mechanical rotating part is monitored and collected to identify its different fault types. The vibration acceleration sensor is used to collect multimodal data of different positions and fault degrees of the mechanical rotating part, and an original data set X = {x1, x2, ..., x N}.

[0054] In addition, the original vibration acceleration signal can be converted into a sample set X = {x1, x2, ..., x N} Randomly divided into training sample sets for training SMCN model And a test sample set to verify the performance of the trained model in represents the training data, Its corresponding category label, n tr is the number of training samples, which is the same as is the test data, n te is the number of test samples.

[0055] S102: randomly select a rated number of the vibration acceleration signals to obtain random vibration acceleration signal samples.

[0056] In this step, a specific step of randomly selecting a rated number of the vibration acceleration signals to obtain random vibration acceleration signal samples is provided.

[0057] Specifically, for example, a robust and adaptable bearing fault diagnosis method should be able to obtain stable feature representations from the input, and this feature representation should have similar structure and content for the complete input and the partially damaged input. A vibration signal containing noise is a partially damaged signal, so a robust and adaptable bearing fault diagnosis method should be able to resist certain noise interference. The present invention simulates noise interference through a random sampling strategy, forcing the network to learn features that better represent the nature of the fault from the vibration signal containing interference.

[0058] The specific process of random sampling is as follows: Figure 1 As shown in the figure, each point in the figure represents a data point in the sequence. The sampling of each data point follows a Bernoulli distribution with probability p. In the present invention, probability p is 0.5. By randomly shutting down the neurons in the input layer, the random sampling process of the input signal can be simulated to obtain

[0059]

[0060] After randomly sampling the data at the input layer, the network will use the sampled samples for subsequent feature extraction and fault diagnosis. The specific method will be described in detail later, so I will not go into details here.

[0061] S103, performing pulse enhancement and noise reduction on the random vibration acceleration signal samples through a stacked morphological convolution module and extracting fault feature samples.

[0062] In this step, a specific step is provided for performing pulse enhancement and noise reduction on the random vibration acceleration signal sample through a stacked morphological convolution module and extracting a fault feature sample.

[0063] Specifically, in order to avoid the influence of noise interference on the performance of the fault diagnosis model, the noise interference should be eliminated as much as possible through certain technical means. The present invention achieves the goal of removing the interference components of the noisy signal by introducing mathematical morphological filtering technology.

[0064] S104: training a fault diagnosis model based on the fault feature samples for fault diagnosis.

[0065] In this step, a specific step is provided for obtaining a fault diagnosis model for fault diagnosis based on the fault feature sample training.

[0066] In this embodiment, a specific implementation method of a mechanical component fault diagnosis method is provided. First, a rated number of vibration acceleration signals when a mechanical rotating component fails are received, wherein different fault types corresponding to the vibration acceleration signals are identified; a rated number of the vibration acceleration signals are randomly selected to obtain random vibration acceleration signal samples; the random vibration acceleration signal samples are pulse enhanced and denoised through a stacked morphological convolution module and fault feature samples are extracted; and a fault diagnosis model is obtained by training the fault feature samples for fault diagnosis.

[0067] Figure 2 FIG. 2 is a schematic diagram of a process 200 for diagnosing a mechanical component fault in another embodiment of the present invention. Figure 3 This is a data architecture diagram of the SMCN structure block diagram in another embodiment of the present invention. Figure 2 and Figure 3 In one embodiment, the stacked morphological convolution module is a parallel branch with different morphological operators, and the different morphological operators include an opening operation and a closing operation;

[0068] The opening operation is used to perform erosion and expansion operations in sequence, and the closing operation is used to perform expansion and erosion operations in sequence.

[0069] In this embodiment, a specific data architecture of the stacked convolution module is provided.

[0070] from Figure 3 As can be seen from the figure, the input signal is filtered through two parallel branches with opening and closing operators, which is the morphological layer. The morphological layer uses the opening and closing operators as the core of the morphological layer to extract pulse features and reduce noise. The opening operation performs an expansion operation on the eroded signal, while the closing operation performs an erosion operation on an expanded signal. The erosion is achieved by subtracting the elements between the structural element (SE), that is, the core, and the vibration signal and finding the partial minimum value, while the expansion is achieved by adding the elements and finding the partial maximum value.

[0071] In one embodiment, the performing pulse enhancement and noise reduction on the random vibration acceleration signal sample through a stacked morphological convolution module and extracting the fault feature sample comprises:

[0072] S201, retaining negative pulses according to the random vibration acceleration signal by using an open operation, and extracting positive pulses according to the random vibration acceleration signal by using a closed operation.

[0073] In this step, a specific step is provided for retaining negative pulses according to the random vibration acceleration signal using an open operation, and extracting positive pulses according to the random vibration acceleration signal using a closed operation. The negative pulses of the vibration signal are retained and the positive pulses are suppressed by the open operation; the positive pulses of the vibration signal are extracted by the closed operation.

[0074] Specifically, for the input signal x∈R N and SEss d and e , the operation process of the four morphological operators is as follows:

[0075]

[0076]

[0077]

[0078]

[0079] Among them, x i is the feature map in the i-th input channel, y j is the feature map in the jth output channel. SEs (i.e., s d and e ) is a kernel of length l. In order to improve computational efficiency, the height of SE in SMCN is set to zero.

[0080] S202, extracting components related to fault characteristics from the negative pulse and the positive pulse to obtain corresponding negative pulse components and positive pulse components.

[0081] In this step, a specific step is provided for extracting components related to fault characteristics from the negative pulse and the positive pulse to obtain corresponding negative pulse components and positive pulse components.

[0082] S203, obtaining the fault feature sample by fusing the negative pulse component and the positive pulse component.

[0083] In this step, a specific step is provided for obtaining the fault feature sample by fusing the negative pulse component and the positive pulse component. In order to extract the positive pulse component and the negative pulse component related to the fault feature from the vibration signal, two parallel branches with opening and closing operations are used simultaneously in the deep network. Since the vibration signal after morphological filtering contains different pulse features, which is an understanding of the vibration signal at different levels, it is necessary to fuse them.

[0084] It is also worth pointing out that in SMCN, a nonlinear process of fusing different features using the self-attention mechanism is introduced, which is called feature enhancement fusion. This module achieves dynamic adjustment of the receptive field through two operations: fusion and selection. Figure 3 As shown in Figure 2, the fusion operation combines the information from each branch stream to generate a global feature description. The selection operation uses the above description to recalibrate the feature maps and then aggregate them.

[0085] In this embodiment, a specific implementation method is provided for performing pulse enhancement and noise reduction on the random vibration acceleration signal sample through a stacked morphological convolution module and extracting a fault feature sample. First, the negative pulse is retained according to the random vibration acceleration signal using an open operation, and the positive pulse is extracted according to the random vibration acceleration signal using a closed operation. The components related to the fault feature are extracted from the negative pulse and the positive pulse to obtain the corresponding negative pulse component and positive pulse component. The fault feature sample is obtained by fusing the negative pulse component and the positive pulse component.

[0086] Figure 4 FIG. 3 is a schematic diagram of a process 300 for diagnosing a mechanical component fault in another embodiment of the present invention. Figure 4 As shown, in one embodiment, the step of obtaining the fault feature sample by fusing the negative pulse component and the positive pulse component includes:

[0087] S301, obtaining a fusion feature map by fusing the negative pulse component and the positive pulse component.

[0088] In this step, a specific step is provided for obtaining a fusion feature map by fusing the negative pulse component and the positive pulse component.

[0089] Specifically, in the fusion operation part, the results of the two branches are first fused through the element-wise summation operation to obtain a new feature map Y:

[0090] Y=y d +y e (5)

[0091] S302: weight parameters are obtained according to the influence of different fault categories on different characteristic channels.

[0092] In this step, a specific step is provided for configuring and obtaining weight parameters according to the influence of different fault categories on different characteristic channels.

[0093] Considering that different feature channels often contribute differently to extracting weakly identifiable fault features, a small sub-network is used to obtain a set of weights.

[0094] S303: Obtain feature sizes of different channels according to the weight parameters and the different feature channels.

[0095] In this step, a specific step is provided for obtaining the feature sizes of different channels according to the weight parameter and the different feature channels.

[0096] The above set of weights is multiplied by the features of each channel respectively to adjust the size of each channel feature.

[0097] S304, applying preset attention parameters according to feature sizes of different channels to obtain the fault feature template.

[0098] In this step, a specific step is provided for applying preset attention parameters according to the feature sizes of different channels to obtain the fault feature template, thereby applying different sizes of attention on each feature channel, enhancing the fault features and suppressing redundant features.

[0099] In this embodiment, a specific implementation method is provided for obtaining the fault feature sample by fusing the negative pulse component and the positive pulse component. A fused feature map is obtained by fusing the negative pulse component and the positive pulse component. A weight parameter is obtained according to the configuration of the influence of different fault categories on different feature channels. The feature sizes of different channels are obtained according to the weight parameter and different feature channels. The fault feature template is obtained by applying a preset attention parameter according to the feature sizes of different channels.

[0100] The specific algorithm is as follows:

[0101] In order to model the channel relationship, we first generate global information through a global average pooling operation.

[0102] Where C is the feature dimension of S in the model graph. Then the relationship between channels is modeled through a simple and effective fully connected layer to guide the adaptive selection:

[0103]

[0104] Where δ represents the ReLU activation function, For BN, where d is controlled by the reduction ratio r:

[0105] d=max(C / r,L) (8)

[0106] Among them, L=32, which is the classic value in experiments.

[0107] In the selection operation part, different attention sizes are applied to different channels through the Softmax function, and different spatial scales are adaptively selected:

[0108]

[0109] in, And a, b are pairs of y d and e The soft attention applied has the following relationship: a c +b c =1.c ∈R 1×d represents the cth row of A, a c is the cth element of a. The final feature map is obtained by applying different weights to different feature maps:

[0110]

[0111] Where V=[V1,V2,…,V C ],

[0112] The morphological layer and the feature enhancement fusion layer constitute a morphological convolution module (MCM). The MCM first uses two parallel branches with opening and closing operations to reduce noise and extract pulse features from the vibration signal. Then, the feature enhancement fusion layer based on the attention mechanism is used to adaptively fuse the denoised signals of different branches to enhance the fault-related features. Stacking three MCMs in the SMCN gradually reduces the noise-related features and retains the fault-related features.

[0113] In one embodiment, after the step of obtaining a fault diagnosis model by training the fault feature samples for fault diagnosis, the method further comprises:

[0114] The fault diagnosis model is optimized using a back propagation algorithm through a Softmax cross entropy loss function.

[0115] In this embodiment, a specific implementation method of optimizing the fault diagnosis model using a back propagation algorithm through a Softmax cross entropy loss function is provided.

[0116] The enhanced fault features are input into the classification layer to obtain the preliminary classification results. Then, the back propagation algorithm is used to optimize the SMCN model training with the Softmax cross entropy loss function (Formula (11)) as the objective function until the number of iterations required is met and the model training is completed.

[0117]

[0118] Among them, Z l (j) is the input of the lth neuron in the lth fully connected layer.

[0119] Finally, the trained SMCN fault diagnosis model is used to diagnose the test set data. Perform diagnostic classification and obtain the fault diagnosis accuracy of mechanical rotating parts.

[0120] In one embodiment, the present invention further provides a mechanical component fault diagnosis system, the mechanical component fault diagnosis system comprising a background system;

[0121] The background system is used to receive vibration acceleration signals when a rated number of mechanical rotating parts fail, wherein different fault types corresponding to the vibration acceleration signals are identified; a rated number of vibration acceleration signals are randomly selected to obtain random vibration acceleration signal samples; the random vibration acceleration signal samples are pulse enhanced and denoised through a stacked morphological convolution module and fault feature samples are extracted; and a fault diagnosis model is obtained by training according to the fault feature samples for fault diagnosis.

[0122] In one embodiment, the mechanical component fault diagnosis system includes a vibration acceleration sensor;

[0123] The vibration acceleration sensor is used to collect vibration acceleration signals.

[0124] Figure 5 FIG. 4 is a schematic diagram of a process 400 for diagnosing a mechanical component fault in another embodiment of the present invention. Figure 6 FIG. 4 is a schematic diagram of an analog circuit fault detection device in another embodiment of the present invention. Figure 5 and Figure 6 As shown, in one embodiment, the present invention further provides a mechanical component fault diagnosis device, the device comprising:

[0125] The receiving module 101 is used to receive vibration acceleration signals when a rated number of mechanical rotating parts fail, wherein different fault types corresponding to the vibration acceleration signals are identified;

[0126] A random module 102, configured to randomly select a rated number of the vibration acceleration signals to obtain random vibration acceleration signal samples;

[0127] The enhancement and noise reduction module 103 is used to perform pulse enhancement and noise reduction on the random vibration acceleration signal sample through a stacked morphological convolution module and extract fault feature samples;

[0128] The training module 104 is used to train the fault feature samples to obtain a fault diagnosis model for fault diagnosis.

[0129] In one embodiment, the present invention further provides a mechanical component fault diagnosis device, the device comprising: a processor and a memory;

[0130] The memory stores an application program executable by the processor, which is used to enable the processor to execute the steps of the mechanical component fault diagnosis method.

[0131] In one embodiment, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the mechanical component fault diagnosis method when executed by a processor.

[0132] Advantages of the present invention:

[0133] In the end-to-end mechanical rotating component fault diagnosis algorithm proposed in the present invention, the random sampling strategy can improve the adaptability of the model; the stacked morphology convolution module can extract deep fault features, and the two parallel branches with opening and closing operators in the module have a noise reduction effect. The adaptive fusion strategy based on the attention mechanism can recalibrate the feature map, suppress redundant features, and realize the enhanced fusion of fault features; the classifier processes the enhanced fault features to obtain the fault diagnosis result. Therefore, the fault algorithm proposed in the present invention can significantly improve the accuracy of fault diagnosis of mechanical rotating components.

[0134] The above embodiments are only exemplary embodiments of the present invention and are not intended to limit the present invention. The protection scope of the present invention is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present invention within the essence and protection scope of the present invention, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present invention.

Claims

1. A mechanical component fault diagnosis method, characterized in that: The method includes: receiving vibration acceleration signals when a rated number of mechanical rotating components fail, wherein different fault types corresponding to the vibration acceleration signals are identified; Randomly selecting a rated number of the vibration acceleration signals to obtain random vibration acceleration signal samples; Performing pulse enhancement and noise reduction on the random vibration acceleration signal samples through a stacked morphological convolution module and extracting fault feature samples; A fault diagnosis model is obtained by training the fault feature samples for fault diagnosis; The morphological layer and the feature enhancement fusion layer constitute a stacked morphological convolution module, which first uses two parallel branches with opening and closing operations to reduce noise and extract pulse features from the vibration signal, and then uses a feature enhancement fusion layer based on the attention mechanism to adaptively fuse the denoised signals of different branches to enhance fault-related features. The stacked morphological convolution module is a parallel branch with different morphological operators, and the different morphological operators include opening operations and closing operations; The opening operation is used to perform erosion and expansion operations in sequence, and the closing operation is used to perform expansion and erosion operations in sequence.

2. The mechanical component fault diagnosis method according to claim 1, characterized in that: The step of performing pulse enhancement and noise reduction on the random vibration acceleration signal sample through a stacked morphological convolution module and extracting a fault feature sample comprises: retaining negative pulses using an open operation according to the random vibration acceleration signal, and extracting positive pulses using a closed operation according to the random vibration acceleration signal; Extracting components related to fault characteristics from the negative pulse and the positive pulse to obtain corresponding negative pulse components and positive pulse components; The fault feature sample is obtained by fusing the negative pulse component and the positive pulse component.

3. The mechanical component fault diagnosis method according to claim 2, characterized in that: The fusing the negative pulse component and the positive pulse component to obtain the fault feature sample comprises: Obtaining a fusion feature map by fusing the negative pulse component and the positive pulse component; The weight parameters are obtained according to the influence of different fault categories on different characteristic channels; Obtaining feature sizes of different channels according to the weight parameters and the different feature channels; The fault feature template is obtained by applying preset attention parameters according to the feature sizes of different channels.

4. The mechanical component fault diagnosis method according to claim 3, characterized in that: After the step of obtaining a fault diagnosis model for fault diagnosis by training the fault feature samples, the method further comprises: The fault diagnosis model is optimized using a back propagation algorithm through a Softmax cross entropy loss function.

5. A mechanical component fault diagnosis system, characterized in that: The mechanical component fault diagnosis system includes a background system; The background system is used to receive vibration acceleration signals when a rated number of mechanical rotating parts fail, wherein different fault types corresponding to the vibration acceleration signals are identified; random vibration acceleration signal samples are obtained by randomly selecting a rated number of the vibration acceleration signals; pulse enhancement and noise reduction are performed on the random vibration acceleration signal samples through a stacked morphological convolution module and fault feature samples are extracted; and a fault diagnosis model is obtained by training the fault feature samples for fault diagnosis; The morphological layer and the feature enhancement fusion layer constitute a stacked morphological convolution module, which first uses two parallel branches with opening and closing operations to reduce noise and extract pulse features from the vibration signal, and then uses a feature enhancement fusion layer based on the attention mechanism to adaptively fuse the denoised signals of different branches to enhance fault-related features. The stacked morphological convolution module is a parallel branch with different morphological operators, and the different morphological operators include opening operations and closing operations; The opening operation is used to perform erosion and expansion operations in sequence, and the closing operation is used to perform expansion and erosion operations in sequence.

6. The mechanical component fault diagnosis system according to claim 5, characterized in that: The mechanical component fault diagnosis system includes a vibration acceleration sensor; The vibration acceleration sensor is used to collect vibration acceleration signals.

7. A mechanical component fault diagnosis device, characterized in that: The device comprises: A receiving module, used for receiving vibration acceleration signals when a rated number of mechanical rotating parts fail, wherein different fault types corresponding to the vibration acceleration signals are identified; A random module, used for randomly selecting a rated number of the vibration acceleration signals to obtain random vibration acceleration signal samples; An enhancement and noise reduction module, used for performing pulse enhancement and noise reduction on the random vibration acceleration signal sample through a stacked morphological convolution module and extracting fault feature samples; A training module, used for training a fault diagnosis model based on the fault feature samples for fault diagnosis; The morphological layer and the feature enhancement fusion layer constitute a stacked morphological convolution module, which first uses two parallel branches with opening and closing operations to reduce noise and extract pulse features from the vibration signal, and then uses a feature enhancement fusion layer based on the attention mechanism to adaptively fuse the denoised signals of different branches to enhance fault-related features. The stacked morphological convolution module is a parallel branch with different morphological operators, and the different morphological operators include opening operations and closing operations; The opening operation is used to perform erosion and expansion operations in sequence, and the closing operation is used to perform expansion and erosion operations in sequence.

8. A mechanical component fault diagnosis device, characterized in that: The device comprises: a processor and a memory; The memory stores an application program executable by the processor, which is used to enable the processor to execute the steps of the mechanical component fault diagnosis method according to any one of claims 1 to 4.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the mechanical component fault diagnosis method according to any one of claims 1 to 4 are implemented.

Images