Mining ventilator bearing fault diagnosis method and device and electronic equipment

Abstract

The invention discloses a mining ventilator bearing fault diagnosis method and device and electronic equipment, and belongs to the technical field of ventilator fault diagnosis, and the mining ventilator bearing fault diagnosis method comprises the steps: obtaining a ventilator bearing vibration signal; the ventilator bearing vibration signal is input into a pre-trained fault diagnosis model, a fault classification result is obtained, and the fault diagnosis model adopts a TimeBlock-MCNN structure. According to the model, a Times Block module is used for learning periodic and cross-period characteristics of a vibration signal, and an MCNN module is used for learning multi-scale characteristics, so that complex information in the vibration signal is effectively captured. Experimental results show that the model has high accuracy on a test set, and various bearing faults can be effectively distinguished. The invention provides a new method for mining ventilator bearing fault diagnosis, and the method has high accuracy and practicability.

Description

Technical Field

[0001] The present invention relates to the technical field of ventilator fault diagnosis, and particularly to a fault diagnosis method, device and electronic equipment for a mine ventilator bearing. Background Art

[0002] At present, the demand for coal mining is increasing day by day, which puts forward higher requirements for the degree of automation underground. On the other hand, with the demand for the construction of modern coal mines, the level of the underground ventilation system is also constantly improving. According to the current situation of coal mine production, gas poisoning and gas explosion accidents caused by underground ventilator failures occur from time to time, seriously affecting the safe production and economic benefits of coal mines. The problems of ventilators mainly stem from backward monitoring means, low monitoring accuracy and slow response speed, which are difficult to meet the needs of the high-speed development of modern coal mines.

[0003] The ventilators used in underground mines are mainly used to discharge underground harmful gases and dust that may cause pneumoconiosis, ensuring good air quality underground. There are two main reasons for poor mine ventilation. One is that the ventilation equipment is aging and not repaired and updated in time. On the other hand, there is a lack of effective countermeasures for ventilator failures in mines, resulting in problems not being discovered in time. The coal mine ventilator is a non-linear system with a poor operating environment and many influencing factors, so it is very difficult to conduct online monitoring and bearing fault diagnosis on it. In order to keep the ventilator bearing working normally, it is necessary to combine other technologies such as computers and modern detection to conduct fault diagnosis on it and improve the accuracy of fault diagnosis.

[0004] In related research at home and abroad, the most commonly used method is the time series algorithm, followed by the genetic optimization algorithm, and then the particle swarm optimization algorithm. It can be seen from this kind of literature that there are two problems in the results of related research: one is that the algorithm is easy to fall into local optimum, and the other is that there is a certain deviation between the accuracy and the calculated value. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a fault diagnosis method, device and electronic equipment for a mine ventilator bearing with high accuracy.

[0006] To solve the above technical problem, the present invention provides the following technical solutions: On the one hand, an embodiment of the present invention provides a fault diagnosis method for a mine ventilator bearing, including: Obtaining the vibration signal of the ventilator bearing; Inputting the vibration signal of the ventilator bearing into a pre-trained fault diagnosis model to obtain a fault classification result, wherein the fault diagnosis model adopts a TimesBlock-MCNN structure.

[0007] On the other hand, an embodiment of the present invention provides a fault diagnosis device for a mine ventilator bearing, including: An acquisition module, configured to acquire the vibration signal of the ventilator bearing; A prediction module, configured to input the vibration signal of the ventilator bearing into a pre-trained fault diagnosis model to obtain a fault classification result, wherein the fault diagnosis model adopts a TimesBlock-MCNN structure.

[0008] On yet another aspect, an embodiment of the present invention provides an electronic device, which includes: a housing, a processor, a memory, a circuit board, and a power supply circuit. Among them, the circuit board is arranged inside the space enclosed by the housing, and the processor and the memory are arranged on the circuit board; the power supply circuit is used to supply power to each circuit or device of the above-mentioned electronic device; the memory is used to store executable program codes; the processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, and is used to execute any one of the foregoing methods.

[0009] On still another aspect, an embodiment of the present invention further provides a computer-readable storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement any one of the foregoing methods.

[0010] The present invention has the following beneficial effects: The fault diagnosis method, device and electronic device for a mine ventilator bearing provided by the embodiment of the present invention first acquire the vibration signal of the ventilator bearing, and then input the vibration signal of the ventilator bearing into a pre-trained fault diagnosis model to obtain a fault classification result, wherein the fault diagnosis model adopts a TimesBlock-MCNN structure. This model uses the TimesBlock module to learn the periodic and cross-period characteristics of the vibration signal, and uses the MCNN module to learn multi-scale characteristics, so as to effectively capture the complex information in the vibration signal. Experimental results show that this model has achieved high accuracy on the test set and can effectively distinguish various types of bearing faults. This application provides a new method for the fault diagnosis of mine ventilator bearings, which has high accuracy and practicality. Description of the Drawings

[0011] Figure 1 It is a schematic flowchart of an embodiment of the fault diagnosis method for a mine ventilator bearing of the present invention; Figure 2 It is a schematic structural diagram of the fault diagnosis model (TimesBlock-MCNN structure) in the present invention; Figure 3 It is Figure 2 a schematic structural diagram of a single TimesBlock in Figure 4 For Figure 2 The structural schematic diagram of the MCNN in a single TimesBlock in Figure 5 The classification confusion matrix in the example verification of the present invention; Figure 6 The structural schematic diagram of the embodiment of the bearing fault diagnosis device for mine ventilators of the present invention; Figure 7 The structural schematic diagram of the embodiment of the electronic device of the present invention. Detailed implementation manners

[0012] To make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0013] Aiming at the problems in the research related to the faults of mine ventilators, this application constructs a multi-scale convolutional neural network and uses TimesBlock as the basic network. MCNN (Multi-scale Convolutional Neural Networks) can learn features at different levels and scales, and TimesBlock can learn intra-period changes and cross-period changes. By combining these two network structures, a high-precision and low-model-volume bearing fault diagnosis model for mine ventilators is realized, and this module can effectively capture short-term and long-term dependencies in vibration signals. The proposed model achieves high precision while maintaining a small model size and fast inference speed. This application demonstrates the applicability of TimesBlock in the field of bearing fault diagnosis of mine ventilators, and verifies the practicability of small models in the field of fault diagnosis by using a simple MCNN as the backbone network.

[0014] Discrete Fourier Transform The Discrete Fourier Transform (DFT) is a technique that converts discrete-time domain signals into discrete-frequency domain signals. It has been widely used in vibration signal processing, image processing, and machine vision. The DFT converts a time series into a frequency-domain series. By analyzing the peaks in the spectrogram of the signal in the frequency domain, the periodic components of the signal can be obtained. The computational complexity of the traditional DFT is , so the computational efficiency is not high. The Fast Fourier Transform (FFT) is an efficient algorithm based on the DFT. It uses the symmetry and segmentation strategy of the signal to reduce the complexity of the DFT to . In this application, the calculation of the DFT will use the FFT to achieve the time-frequency domain conversion of the signal. The calculation formula of the DFT is shown as follows: (1) (2) In the formula, is the nth point in the time domain sequence, is the th point in the frequency domain sequence, is the amplitude of the th frequency in the frequency domain sequence, represents the real part, represents the imaginary part. Based on the DFT calculation results, the formula for extracting the first cycles is as follows: (3) (4) In the formula, represents the amplitude sequence of the frequency, represents the length of the time sequence, represents the period length.

[0015] Convolutional neural network Convolutional neural networks have been widely used in image processing because they can effectively learn two-dimensional data through convolutional calculations and deep structures. A convolutional neural network usually consists of two main parts: the convolutional layer and the pooling layer. The convolutional layer uses a convolutional kernel with shared weights to extract local features from the input signal. Compared with the fully connected network, this method provides better computational efficiency. The mathematical expression of the convolutional layer is shown in the following formula: (5) In the formula, is the convolutional output, is the convolutional kernel, and are the coordinates of the output feature, is the number of layers of the convolutional network, and are the coordinates of the convolutional kernel, and are the width and height of the convolutional kernel.

[0016] The pooling layer is used to select and filter the extracted features. Commonly used pooling layers include the average pooling layer and the max pooling layer. The max pooling layer is widely used because of its high computational efficiency and good effect. The mathematical expression of the pooling layer is shown in the following formula: (6) In the formula, is the max pooling output, is the max pooling output, and are the coordinates of the maximum pooling output, and are the width and height of the window of the maximum pooling.

[0017] On the one hand, an embodiment of the present invention provides a method for diagnosing faults in a mine ventilation fan bearing, as Figure 1 shown, including: Step S101: Obtain the vibration signal of the ventilation fan bearing; Step S102: Input the vibration signal of the ventilation fan bearing into a pre-trained fault diagnosis model to obtain a fault classification result, where the fault diagnosis model adopts a TimesBlock-MCNN structure.

[0018] This application proposes an end-to-end fault diagnosis model for a mine ventilation fan bearing. The vibration signal of the ventilation fan bearing is directly input into the model without manual feature extraction. The TimesBlock-MCNN structure is as Figure 2 shown.

[0019] Preferably, the fault diagnosis model, namely TimesBlock-MCNN, is composed of 5 - 30 (the specific number can be flexibly selected according to needs, such as 10, 15, 20, etc.) cascaded TimesBlocks and a fully connected layer. Each TimesBlock contains an FFT module for extracting periodic information, and two MCNN modules for learning the cross-period and intra-period changes of the time series. The fully connected layer is located after the output of the last TimesBlock, maps the output to a vector of a preset length (specifically determined according to the number of states, that is, the number of fault classifications, and can be 5 - 15, such as 6, 8, 10, 12, etc.), and transforms it into a probability distribution through the Softmax function, that is, the fault classification result is obtained. The Softmax calculation formula is shown as the following formula: (7) In the formula, represents the th element in the vector.

[0020] Regarding TimesBlock TimesBlock is a framework module that can effectively learn the cross-period and intra-period information of the time series. TimesBlock utilizes the periodicity of the time series to convert a one-dimensional sequence into a two-dimensional tensor, and then inputs these two-dimensional tensors into the visual backbone network, namely MCNN, to learn intra-period and cross-period features.

[0021] The overall structure of a single TimesBlock module is as Figure 3As shown, the input one-dimensional signal is processed by the FFT module to determine the period length and period intensity. Then, according to the determined period length, the one-dimensional signal is stacked to form a two-dimensional tensor. After that, the MCNN is used to further extract the features of these two-dimensional tensors. Finally, the two-dimensional tensor is converted into a one-dimensional signal. The period intensity output by the FFT module is used to perform weighted fusion on different one-dimensional signals, so as to obtain the final output.

[0022] Regarding the MCNN Each TimesBlock module contains two parts: the FFT module and the MCNN module. The FFT module stacks the one-dimensional vibration signal along the period length into a two-dimensional tensor. The structure of the MCNN is as Figure 4 shown.

[0023] In each TimesBlock, the two MCNN modules are the first MCNN module (i.e., the first layer of MCNN, see the left side of the activation function in Figure 4 ), and the second MCNN module (i.e., the second layer of MCNN, see the right side of the activation function in Figure 4 ). Each MCNN module consists of several parallel convolutional kernels, and the number of convolutional kernels can be flexibly set according to needs. In the example shown in the figure, each MCNN module consists of 5 parallel convolutional kernels with sizes of 1*1, 3*3, 5*5, 7*7, and 9*9. The convolutional kernels in each MCNN module perform padding operations on the input so that the outputs of different-sized kernels have the same shape.

[0024] The output of the first layer of MCNN passes through the Gaussian Error Linear Units (GELU) activation function and then outputs to the second layer of MCNN. In practical applications, the GELU function has been proven to achieve good results in deep learning and is widely used in various types of neural network models. The expression of the GELU function is as shown in the following formula: (8) In the formula, is the cumulative distribution function of the Gaussian distribution.

[0025] The second layer of MCNN compresses the output channels of the first layer of MCNN to the number of input signal channels of the first layer of MCNN, and then reshapes the output of each convolutional kernel from a two-dimensional tensor to a one-dimensional tensor. The one-dimensional tensor is pooled by combining the period intensity signal output by the FFT, and the output after pooling is added and fused with the input of this TimesBlock to generate an output with the same shape as the input signal of this TimesBlock. Through the above operations, TimesBlocks can be directly stacked without paying attention to the shape of the transmitted signal.

[0026] Example verification 1. Data introduction In this experiment, the vibration dataset of rolling bearings publicly available from Case Western Reserve University (CWRU) in the United States was used. This dataset has been widely applied in various research on bearing fault diagnosis and life prediction. The experimental data was collected under four working conditions: 0HP, 1HP, 2HP, and 3HP, with corresponding rotational speeds of 1797 rmp, 1772 rmp, 1750 rmp, and 1730 rmp. Vibration sensors were placed at the drive end and the fan end. The frequency at the fan end was 12 kHz, and the frequency at the drive end was 48 kHz. Artificial electrical discharge machining (EDM) was used to create defects with sizes of 0.007 inches, 0.014 inches, and 0.021 inches on the inner ring, outer ring, and rolling elements of the bearing, generating 9 types of fault data and 1 type of normal data.

[0027] Table 1 Bearing dataset of Case Western Reserve University Fault type Fault size Number of samples Sample length Sample label Normal - 240 2000 0 Rolling element fault 0.07 240 2000 1 Rolling element fault 0.14 240 2000 2 Rolling element fault 0.21 240 2000 3 Inner race fault 0.07 240 2000 4 Inner race fault 0.14 240 2000 5 Inner race fault 0.21 240 2000 6 Outer race fault 0.07 240 2000 7 Outer race fault 0.14 240 2000 8 Outer race fault 0.21 240 2000 9 For this experiment, the data collected at the drive end was selected, with a sampling frequency of 48 kHz, a working load of 3HP, and a rotational speed of 1730 rmp. To improve the efficiency of model training, the original vibration signal was subjected to Wiener filtering to remove high-order noise and downsampled with a step size of 8. To verify the effectiveness of the model, the data was divided into a training set and a test set in a ratio of 8:2, as shown in Table 1 for details.

[0028] 2. Model parameters and metrics The input and output dimensions of the TimesBlock layer are shown in Table 2, and the corresponding parameters of the MCNN are shown in Table 3.

[0029] Table 2 Input and output of the TimesBlock-MCNN model Input shape Output shape Number of parameters TimesBlock-0 [250, 1] [250, 1] 1010 TimesBlock-1 [250, 1] [250, 1] 1010 TimesBlock-2 [250, 1] [250, 1] 1010 TimesBlock-3 [250, 1] [250, 1] 1010 TimesBlock-4 [250, 1] [250, 1] 1010 TimesBlock-5 [250, 1] [250, 1] 1010 … … … … TimesBlock-19 [250, 1] [250, 1] 1010 Linear

[10]

[10] 2510 Table 3 MCNN parameters Input shape Output shape Number of parameters Conv2d-1 [1, 36, 7] [3, 36, 7] 6 Conv2d-2 [1, 36, 7] [3, 36, 7] 30 Conv2d-3 [1, 36, 7] [3, 36, 7] 78 Conv2d-4 [1, 36, 7] [3, 36, 7] 150 Conv2d-5 [1, 36, 7] [3, 36, 7] 246 MCNN-6 [5, 1, 36, 7] [3, 36, 7] 0 GELU-7 [3, 36, 7] [3, 36, 7] 0 Conv2d-8 [3, 36, 7] [1, 36, 7] 4 Conv2d-9 [3, 36, 7] [1, 36, 7] 28 Conv2d-10 [3, 36, 7] [1, 36, 7] 76 Conv2d-11 [3, 36, 7] [1, 36, 7] 148 Conv2d-12 [3, 36, 7] [1, 36, 7] 244 MCNN-13 [5, 1, 36, 7] [1, 36, 7] 0 TimesBlock-14 [1, 36, 7] [250, 1] 0 During the model training process, a dropout layer with a dropout rate of 0.5 was used to prevent the model from overfitting. The loss function adopted the cross-entropy loss function, the optimizer adopted the Adam optimizer, the learning rate was set to 0.001, the batch size was set to 256, and the model was trained for 200 training epochs in total. The model evaluation metrics adopted accuracy, precision, and F1-score, and the calculation formulas are shown as follows: (9) (10) (11) In the formula, is the number of true positive samples, is the number of true negative samples, is the number of false positive samples, is the number of false negative samples. In formula (11), is the recall rate, .

[0030] 3. Experimental Results In this experiment, 10 independent random experiments were conducted to exclude the influence of randomness caused by model training. The evaluation metrics of the method proposed in this application on the test set are shown in Table 4 below.

[0031] Table 4 Evaluation Metrics of the Test Set Label Accuracy F1 score 0 1.000 0.994 1 1.000 1.000 2 1.000 1.000 3 0.921 0.951 4 1.000 1.000 5 0.924 0.970 6 1.000 1.000 7 1.000 0.979 8 1.000 1.000 9 1.000 1.000 Precision 0.990 As can be seen from Table 4, the accuracy of the model on the test set reaches 99.0%, which means that the model can correctly identify most of the bearing fault samples. In addition, the F1 scores of all labels are greater than 0.95, among which the F1 scores of labels 0, 1, 2, 4, 6, 8, 9 are greater than 0.98, and the F1 scores of labels 1, 2, 4, 6, 8, 9 are as high as 1.00, indicating that the model's recognition ability for various types of faults is at a relatively high level.

[0032] Figure 5 is the confusion matrix of the model prediction values and actual values in one of the experimental results. The model can well distinguish various fault modes. Except for label 3, the model has strong recognition ability for each state. It can be seen that the model has a phenomenon of confusing recognition between label 3 and label 5, which may be due to the relatively similar fault characteristics of label 3 (rolling element fault) and label 5 (inner race fault).

[0033] Table 5 Comparison of Model Sizes and Inference Speeds Model LSTM CNN CNN-LSTM TimesBlock-MCNN Number of parameters (pcs) 158218 337954 382498 121320 Model size (MB) 0.8 1.5 2.2 0.09 Inference speed (s) 0.94 1.16 1.51 0.26 Note: The instance of this application corresponding to this table is the result obtained when TimesBlock is stacked 30 times. The fewer the stacks, the fewer the corresponding parameters and the faster the speed.

[0034] As can be seen from Table 5 above, compared with the existing network models such as LSTM (Long Short-Term Memory), CNN (Convolutional Neural Networks), and CNN-LSTM, the TimesBlock-MCNN model of this application has a small model size and a fast inference speed.

[0035] In summary, this application proposes a fault diagnosis model for mine ventilator bearings based on TimesBlock-MCNN, and its effectiveness is verified through experiments. This model can effectively learn the periodic and multi-scale features in the vibration signals of ventilator bearings, thereby improving the accuracy of fault diagnosis. The main advantages are as follows: (1) The TimesBlock module can effectively learn the cross-cycle and intra-cycle information of time series and convert it into a two-dimensional tensor for input into the MCNN module, thus better capturing the time-frequency features of vibration signals.

[0036] (2) MCNN can learn features at different levels and scales, thus better capturing the complex information in vibration signals.

[0037] (3) The model does not require manual feature extraction and can directly input vibration signals into the network for fault diagnosis, simplifying the fault diagnosis process.

[0038] This model provides a new deep learning-based method for fault diagnosis of mine ventilator bearings, which can improve the accuracy and efficiency of fault diagnosis, thus ensuring the safe production of mines and having certain theoretical and application values. In the future, more effective deep learning models for fault diagnosis of mine ventilator bearings will be explored, and further research will be carried out on the lightweight method of the model so that it can run on resource-constrained devices.

[0039] On the other hand, an embodiment of the present invention provides a fault diagnosis device for mine ventilator bearings, as Figure 6 shown, including: An acquisition module 11 for acquiring the vibration signal of the ventilator bearing; A prediction module 12 for inputting the vibration signal of the ventilator bearing into a pre-trained fault diagnosis model to obtain a fault classification result, where the fault diagnosis model adopts a TimesBlock-MCNN structure.

[0040] The device of this embodiment can be used to execute Figure 1 the technical solutions of the method embodiments shown, and its implementation principle and technical effects are similar, so they will not be elaborated here.

[0041] Preferably, the fault diagnosis model consists of several serially connected TimesBlocks and a fully connected layer. Each TimesBlock contains an FFT module for extracting periodic information and two MCNN modules for learning the cross-cycle and intra-cycle changes of time series; the fully connected layer is located after the output of the last TimesBlock, maps the output to a vector of a preset length, and transforms it into a probability distribution through the Softmax function to obtain the fault classification result.

[0042] Preferably, the number of TimesBlocks is 5 - 30; and / or, the preset length is 5 - 15.

[0043] Preferably, in each TimesBlock, the input one-dimensional signal is processed by an FFT module to determine the period length and period intensity, and then according to the determined period length, the one-dimensional signals are stacked to form a two-dimensional tensor for inputting into the MCNN module to further extract the features of these two-dimensional tensors.

[0044] Preferably, in each TimesBlock, the two MCNN modules are respectively a first MCNN module and a second MCNN module, and both the first MCNN module and the second MCNN module are composed of several parallel convolutional kernels; The output of the first MCNN module, after passing through a GELU activation function, is output to the second MCNN module; the second MCNN module compresses the output channels of the first MCNN module to the number of input signal channels of the first MCNN module, and then reshapes the output of each convolutional kernel from a two-dimensional tensor into a one-dimensional tensor. The one-dimensional tensor is pooled by combining the period intensity signal output by the FFT, and the output after pooling is added and fused with the input of this TimesBlock to generate an output with the same shape as the input signal of this TimesBlock.

[0045] Preferably, each MCNN module is composed of 5 parallel convolutional kernels with sizes of 1*1, 3*3, 5*5, 7*7, and 9*9.

[0046] Preferably, the convolutional kernels in each MCNN module perform a padding operation on the input so that the outputs of different-sized kernels have the same shape.

[0047] The embodiment of the present invention also provides an electronic device, Figure 7 which is a schematic structural diagram of an embodiment of the electronic device of the present invention and can implement the present invention Figure 1 shown in the embodiment, as Figure 7 shown, the above-mentioned electronic device may include: a housing 41, a processor 42, a memory 43, a circuit board 44, and a power supply circuit 45. Among them, the circuit board 44 is arranged inside the space surrounded by the housing 41, and the processor 42 and the memory 43 are arranged on the circuit board 44; the power supply circuit 45 is used to supply power to each circuit or device of the above-mentioned electronic device; the memory 43 is used to store executable program codes; the processor 42 runs a program corresponding to the executable program code by reading the executable program code stored in the memory 43, and is used to execute the method described in any one of the foregoing embodiments.

[0048] The specific execution process of the above steps by the processor 42 and the steps further executed by the processor 42 by running the executable program code can be referred to the description of the embodiments shown in the present invention Figure 1 and will not be elaborated here.

[0049] The electronic device exists in various forms, including but not limited to: (1) Mobile communication device: Such devices are characterized by having mobile communication functions and mainly aim to provide voice and data communication. Such terminals include: smart phones (such as iPhone), multimedia phones, functional phones, and low-end phones, etc.

[0050] (2) Ultra-mobile personal computer device: Such devices belong to the category of personal computers, have computing and processing functions, and generally also have the characteristic of mobile Internet access. Such terminals include: PDA, MID, and UMPC devices, etc., such as iPad.

[0051] (3) Portable entertainment device: Such devices can display and play multimedia content. Such devices include: audio and video players (such as iPod), handheld game consoles, e-books, and intelligent toys and portable vehicle navigation devices.

[0052] (4) Server: A device that provides computing services. The composition of the server includes a processor, hard disk, memory, system bus, etc. The server is similar to a general computer architecture, but due to the need to provide highly reliable services, it has higher requirements in terms of processing power, stability, reliability, security, scalability, and manageability.

[0053] (5) Other electronic devices with data interaction functions.

[0054] The embodiment of the present invention also provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the method steps described in any of the above method embodiments are implemented.

[0055] The embodiment of the present invention also provides an application program, which is executed to implement the method provided in any method embodiment of the present invention.

[0056] It should be noted that in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or device comprising said element.

[0057] Each embodiment in this specification is described in a related manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description of the method embodiments. For the convenience of description, the above device is described by dividing its functions into various units / modules. Of course, when implementing the present invention, the functions of each unit / module can be realized in one or more software and / or hardware.

[0058] Those of ordinary skill in the art can understand that all or part of the processes of implementing the above method embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above method embodiments. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0059] As mentioned above, the above are only the specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for diagnosing bearing faults of a mine ventilator, characterized in that, Including: Obtain the vibration signal of the ventilator bearing; Input the vibration signal of the ventilator bearing into a pre-trained fault diagnosis model to obtain a fault classification result, where the fault diagnosis model adopts a TimesBlock-MCNN structure.

2. The method according to claim 1, wherein The fault diagnosis model consists of several serially connected TimesBlocks and a fully connected layer. Each TimesBlock contains an FFT module for extracting periodic information and two MCNN modules for learning the cross-period and intra-period variations of the time series; the fully connected layer is located after the output of the last TimesBlock, maps the output to a vector of a preset length, and transforms it into a probability distribution through the Softmax function to obtain the fault classification result.

3. The method according to claim 2, wherein The number of the TimesBlocks is 5 - 30; And / or, the preset length is 5 - 15.

4. The method according to claim 2, wherein In each TimesBlock, the input one-dimensional signal is processed by the FFT module to determine the period length and period intensity, and then according to the determined period length, the one-dimensional signals are stacked to form a two-dimensional tensor for inputting into the MCNN module to further extract the features of these two-dimensional tensors.

5. The method according to claim 4, wherein In each TimesBlock, the two MCNN modules are respectively a first MCNN module and a second MCNN module, and both the first MCNN module and the second MCNN module consist of several parallel convolutional kernels; The output of the first MCNN module passes through the GELU activation function and then outputs to the second MCNN module; The second MCNN module compresses the output channels of the first MCNN module to the number of input signal channels of the first MCNN module, then reshapes the output of each convolutional kernel from a two-dimensional tensor to a one-dimensional tensor, and the one-dimensional tensor is pooled by combining the period intensity signal output by the FFT, and the pooled output is added and fused with the input of this TimesBlock to generate an output with the same shape as the input signal of this TimesBlock.

6. The method according to claim 5, wherein Each MCNN module consists of 5 parallel convolutional kernels with sizes of 1*1, 3*3, 5*5, 7*7, and 9*9 respectively.

7. The method according to claim 6, wherein The convolutional kernels in each MCNN module perform padding operations on the input so that the outputs of different-sized kernels have the same shape.

8. A bearing fault diagnosis device for a mine ventilator, characterized in that, Including: An acquisition module for acquiring the vibration signal of the ventilator bearing; A prediction module for inputting the vibration signal of the ventilator bearing into a pre-trained fault diagnosis model to obtain a fault classification result, where the fault diagnosis model adopts a TimesBlock-MCNN structure.

9. An electronic device, characterized in that, The electronic device includes: a housing, a processor, a memory, a circuit board, and a power supply circuit. The circuit board is arranged inside the space enclosed by the housing, and the processor and the memory are arranged on the circuit board; the power supply circuit is used to supply power to each circuit or device of the above electronic device; the memory is used to store executable program codes; the processor runs the program corresponding to the executable program codes by reading the executable program codes stored in the memory, and is used to execute the method according to any one of claims 1 - 7 above.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the method according to any one of claims 1-7 above.

Images