A motor bearing fault diagnosis method, device, equipment and storage medium

Abstract

The application relates to a motor bearing fault diagnosis method, device, equipment and storage medium, comprising the following steps: acquiring a one-dimensional vibration signal of a motor, and converting the one-dimensional vibration signal into a two-dimensional time-frequency graph; inputting the two-dimensional time-frequency graph into a pre-trained complete hybrid neural network model to obtain a diagnosis result, wherein the hybrid neural network model comprises a CNN layer, an LSTM layer, an Attention layer and a full connection layer; the CNN layer and the LSTM layer are used for parallel processing of the two-dimensional time-frequency graph, and local texture features and time dynamic evolution features of the two-dimensional time-frequency graph are extracted respectively; the Attention layer is used for weighted fusion of the local texture features and the time dynamic evolution features to obtain fusion features; the full connection layer is used for outputting the diagnosis result according to the fusion features; through the synergistic effect of the parallel structure and the Attention layer, in the positive multiple type fault diagnosis, the accuracy is improved compared with a traditional CNN-LSTM serial model, and the model training convergence speed is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of motor fault diagnosis technology, and in particular to a method, apparatus, equipment and storage medium for diagnosing motor bearing faults. Background Technology

[0002] As a core power equipment in industrial production, the condition of the bearings in electric motors directly affects their operational stability and service life. Traditional methods for diagnosing motor bearing faults often rely on manual experience to analyze the time or frequency domain characteristics of vibration signals, resulting in low diagnostic efficiency, high subjectivity, and insufficient accuracy in identifying complex faults.

[0003] With the development of deep learning technology, Convolutional Neural Networks (CNNs) are used for fault diagnosis due to their ability to extract local spatial features, but they struggle to capture the temporal dynamics of signals. Long Short-Term Memory Networks (LSTMs), while capable of modeling temporal dependencies, have limited ability to extract local texture features. The existing mainstream CNN-LSTM serial structure suffers from compression loss of original information during feature transfer and fails to fully utilize the complementarity of the two types of features, resulting in room for improvement in the accuracy and training efficiency of multi-type fault diagnosis. Furthermore, the large amount of random noise contained in vibration signals can interfere with the recognition of effective features, further affecting the reliability of diagnostic results. Therefore, there is an urgent need for a motor bearing fault diagnosis scheme that can simultaneously extract both spatial and temporal features and effectively suppress noise interference. Summary of the Invention

[0004] In view of this, it is necessary to provide a method, device, equipment and storage medium for diagnosing motor bearing faults, so as to solve the problems of incomplete feature extraction, weak anti-interference ability, insufficient focus on key features and low accuracy of multi-type fault diagnosis in the existing technology.

[0005] In a first aspect, embodiments of the present invention provide a method, apparatus, device, and storage medium for diagnosing motor bearing faults, comprising the following steps: The one-dimensional vibration signal of the motor is acquired and converted into a two-dimensional time-frequency diagram. A two-dimensional time-frequency image is input into a pre-trained hybrid neural network model to obtain a diagnostic result. The hybrid neural network model includes a CNN layer, an LSTM layer, an Attention layer, and a fully connected layer. The CNN layer and the LSTM layer are used to process the two-dimensional time-frequency image in parallel and extract the local texture features and the temporal dynamic evolution features of the two-dimensional time-frequency image, respectively. The Attention layer is used to perform weighted fusion of the local texture features and the temporal dynamic evolution features to obtain fused features. The fully connected layer is used to output the diagnostic result based on the fused features.

[0006] Further, the step of acquiring the one-dimensional vibration signal of the motor and converting the one-dimensional vibration signal into a two-dimensional time-frequency diagram includes: A one-dimensional vibration signal of the motor is acquired, and the one-dimensional vibration signal is subjected to interference removal processing. The one-dimensional vibration signal after interference removal is subjected to continuous wavelet transform to balance the time resolution and frequency resolution of the one-dimensional vibration signal, and a two-dimensional time-frequency diagram corresponding to the one-dimensional vibration signal is generated.

[0007] Furthermore, the CNN layer comprises three convolutional layers, two max-pooling layers, and one flattening layer, wherein the convolutional layers employ the ReLU activation function to extract the local texture features of the two-dimensional time-frequency map.

[0008] Furthermore, the Attention layer employs a multi-head attention mechanism, and the method for obtaining the fused features is as follows: The attention weights of the local texture features and the temporal dynamic evolution features are calculated using a parallel attention head. The attention weights of the local texture features and the temporal dynamic evolution features are normalized. Based on the attention weights after normalization, the local texture features and the temporal dynamic evolution features are weighted and fused to obtain the fused features.

[0009] Furthermore, it also includes: The diagnostic results are optimized using a sliding window hard voting method to obtain the final fault diagnosis result.

[0010] Furthermore, the optimization of the diagnostic results based on the sliding window hard voting method to obtain the final fault diagnosis result includes: Set the sliding window length to 20, meaning that 20 consecutive preliminary fault diagnosis results are counted each time; The 20 preliminary fault diagnosis results within the sliding window are counted, and the occurrence frequency of each type of fault is statistically analyzed. The most frequently occurring fault type is taken as the final diagnostic result corresponding to the sliding window. If there are two or more most frequently occurring types, the preliminary fault diagnosis result at the middle position of the sliding window is taken as the final diagnostic result to ensure the uniqueness of the diagnostic result.

[0011] In a second aspect, embodiments of the present invention provide a motor bearing fault diagnosis device, comprising: The signal acquisition module is used to acquire the one-dimensional vibration signal of the motor and convert the one-dimensional vibration signal into a two-dimensional time-frequency diagram. A diagnostic module is used to input a two-dimensional time-frequency image into a pre-trained hybrid neural network model to obtain a diagnostic result. The hybrid neural network model includes a CNN layer, an LSTM layer, an Attention layer, and a fully connected layer. The CNN layer and the LSTM layer process the two-dimensional time-frequency image in parallel and extract local texture features and temporal dynamic evolution features from the image, respectively. The Attention layer performs a weighted fusion of the local texture features and the temporal dynamic evolution features to obtain a fused feature. The fully connected layer outputs the diagnostic result based on the fused feature.

[0012] In a third aspect, embodiments of the present invention provide an electronic device, comprising: processor; The memory stores computer instructions that cause the processor to perform the methods described above.

[0013] In a fourth aspect, embodiments of the present invention provide a computer-readable storage medium storing computer instructions for implementing the method described above.

[0014] In a fifth aspect, embodiments of the present invention provide a data acquisition system, including a vibration acquisition board and the electronic device described above, wherein the vibration acquisition board includes: Multiple vibration sensors are installed in the drive bearing housing and the non-drive bearing housing of the motor, respectively; The signal conditioning circuit includes a differential amplifier circuit, a low-pass filter circuit, and a voltage follower circuit connected to the vibration sensor, used to amplify and filter the weak vibration signal output by the vibration sensor. The data acquisition chip has its input terminal connected to the output terminal of the signal conditioning circuit, and is used to convert the conditioned analog vibration signal into a digital signal. The output terminal of the data acquisition chip is electrically connected to the electronic device.

[0015] Compared with existing technologies, 1. Parallel processing of the same time-frequency image using CNN and LSTM layers avoids feature compression and loss in sequential structures, fully preserving the complementarity of local texture features and global dynamic features, thus improving feature representation capabilities. An Attention layer is introduced to focus on key fault features such as periodic impact stripes through multi-dimensional weight calculations, reducing random noise interference and making the fused features more discriminative, especially improving the diagnostic accuracy of non-mechanical faults such as demagnetization.

[0016] 2. Through the synergistic effect of parallel structure and attention layer, the accuracy is improved compared with the traditional CNN-LSTM serial model in the diagnosis of multiple types of positive faults, and the model training convergence speed is greatly improved. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a motor bearing fault diagnosis method according to an embodiment of the present disclosure is shown schematically. Figure 2 This table compares the loss values ​​of the CNN-LSTM-Attention framework in this application with those of the traditional CNN-LSTM framework. Figure 3 This table compares the accuracy of the CNN-LSTM-Attention framework in this application with that of the traditional CNN-LSTM framework. Figure 4 A flowchart illustrating a method for diagnosing motor bearing faults according to yet another embodiment of the present disclosure is shown. Figure 5 A schematic diagram of a motor bearing fault diagnosis device according to an embodiment of the present disclosure is shown. Figure 6 A schematic diagram of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0018] The technical solutions of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, not all of them. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this specification. Furthermore, the technical features involved in the different embodiments of this specification described below can be combined with each other as long as they do not conflict with each other.

[0019] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0020] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0021] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0022] This disclosure provides a method 100 for diagnosing motor bearing faults, which is described below in conjunction with... Figure 1 The motor bearing fault diagnosis method 100 according to an exemplary embodiment of the present disclosure may include steps S101-S102: Step S101: Obtain the one-dimensional vibration signal of the motor and convert the one-dimensional vibration signal into a two-dimensional time-frequency diagram; Step S102: Input the two-dimensional time-frequency image into a pre-trained hybrid neural network model to obtain the diagnostic result. The hybrid neural network model includes a CNN layer, an LSTM layer, an Attention layer, and a fully connected layer. The CNN layer and LSTM layer are used to process the two-dimensional time-frequency image in parallel and extract the local texture features and the time dynamic evolution features of the two-dimensional time-frequency image, respectively. The Attention layer is used to perform weighted fusion of the local texture features and the time dynamic evolution features to obtain the fused features. The fully connected layer is used to output the diagnostic result based on the fused features.

[0023] The implementation process of each step is illustrated below as an example.

[0024] Step S101: Obtain the one-dimensional vibration signal of the motor and convert the one-dimensional vibration signal into a two-dimensional time-frequency diagram.

[0025] In this embodiment, acquiring a one-dimensional vibration signal of the motor and converting the one-dimensional vibration signal into a two-dimensional time-frequency diagram includes: The one-dimensional vibration signal of the motor is acquired, and the one-dimensional vibration signal is processed to remove interference. The one-dimensional vibration signal after interference removal is subjected to continuous wavelet transform to balance the time resolution and frequency resolution of the one-dimensional vibration signal, and a two-dimensional time-frequency diagram corresponding to the one-dimensional vibration signal is generated.

[0026] In some embodiments, a one-dimensional vibration signal can be acquired using a vibration sensor mounted on the motor, wherein the vibration sensor can be mounted on both the drive end and the non-drive end of the motor. During signal acquisition, the signal is filtered and amplified by a signal conditioning circuit to remove environmental interference noise.

[0027] In some embodiments, the scale parameter and time translation parameter are adjusted by using the continuous wavelet transform (CWT) with Morse wavelet as the basis function to convert the one-dimensional vibration signal into a two-dimensional time-frequency diagram, balancing the time resolution and frequency resolution, and preserving the complete time-frequency characteristics.

[0028] For example, the specific parameters of the continuous wavelet transform (CWT) include: using the Morse wavelet as the basis wavelet, a sampling frequency of 20kHz, transforming a single-channel data with a length of 2048 points, generating a time-frequency graph through image processing after the transformation, and then cropping and resizing the time-frequency graph to fit the input size of the convolutional neural network.

[0029] Step S102: Input the two-dimensional time-frequency image into a pre-trained hybrid neural network model to obtain the diagnostic result. The hybrid neural network model includes a CNN layer, an LSTM layer, an Attention layer, and a fully connected layer. The CNN layer and LSTM layer are used to process the two-dimensional time-frequency image in parallel and extract the local texture features and the time dynamic evolution features of the two-dimensional time-frequency image, respectively. The Attention layer is used to perform weighted fusion of the local texture features and the time dynamic evolution features to obtain the fused features. The fully connected layer is used to output the diagnostic result based on the fused features.

[0030] In this embodiment, the hybrid neural network includes a parallel CNN layer and LSTM layer, an Attention layer, and a fully connected layer connected in sequence. The two-dimensional time-frequency image is input into the CNN layer and the LSTM layer. The CNN layer extracts local texture features from the two-dimensional time-frequency image, the LSTM layer extracts temporal dynamic evolution features, and the Attention layer performs weighted fusion of the local texture features and the temporal dynamic evolution features to obtain fused features. The fully connected layer outputs diagnostic results based on the fused features.

[0031] In some embodiments, the CNN layer is a convolutional neural network and the LSTM layer is a long short-term memory network. The two processes the two-dimensional time-frequency map synchronously, avoiding feature compression and loss caused by the serial structure (i.e., the CNN layer and the LSTM layer are connected in parallel), fully preserving the complementarity between local texture features and global dynamic features, and improving feature representation capabilities.

[0032] For example: A convolutional neural network (CNN) consists of an input layer, a feature extraction module, a classification module, and an output layer connected in sequence; the input layer is normalized using the "zscore" method, and the input size is consistent with the size of the time-frequency image after wavelet transform processing; the feature extraction module contains 3 convolutional units, each of which consists of a 3×3 convolutional layer, a batch normalization layer, a ReLU activation layer, and a 2×2 max pooling layer; each max pooling layer of the network is used to reduce the dimensionality and enhance the local texture features of the time-frequency image.

[0033] The specific parameters of the CNN layer are as follows: it contains 3 convolutional layers (3×3 kernel size, with 32, 64, and 128 kernels respectively), 2 max pooling layers (2×2 kernel size), and 1 flattening layer. The convolutional layers use the ReLU activation function to extract local texture features from the time-frequency map. The parameters of each unit are as follows: First convolutional unit: 3×3 convolutional layer (number of kernels: 32, stride: 1, padding: same) → batch normalization layer → ReLU activation layer → 2×2 max pooling layer (stride: 2, padding: same), the output feature map size is scaled down proportionally to the input image size; Second convolutional unit: 3×3 convolutional layer (number of kernels: 64, stride: 1, padding: same) → batch normalization layer → ReLU activation layer → 2×2 max pooling layer, the output feature map size is further reduced proportionally; The third convolutional unit: 3×3 convolutional layer (number of kernels: 128, stride: 1, padding: same) → batch normalization layer → ReLU activation layer → 2×2 max pooling layer, the output feature map size continues to be reduced proportionally; Feature flattening and mapping layer: The feature map output by the third convolutional unit is flattened to convert the two-dimensional feature map into a one-dimensional feature vector, which is then output through the ReLU activation layer.

[0034] For example, the LSTM structure contains two hidden layers, employs the Adam optimizer with a learning rate of 0.001, and extracts dynamic evolution features of the time-frequency graph along the time axis. Specifically, the LSTM unit layer contains two LSTM units and one Dropout layer. A Dropout layer is added after the first LSTM layer, with the Dropout layer ratio set to 0.2 to prevent overfitting through regularization. The output is then processed through a ReLU activation layer to ensure consistency in feature dimensions with the CNN branch outputs.

[0035] In some embodiments, the Attention layer is a multi-head attention mechanism, and the method for obtaining fused features is as follows: The attention weights for local texture features and temporal dynamic evolution features are calculated separately using a parallel attention head; The attention weights for local texture features and temporal dynamic evolution features are normalized. Based on the attention weights after normalization, local texture features and temporal dynamic evolution features are weighted and fused to obtain fused features.

[0036] For example, the implementation process of the multi-head attention mechanism is as follows: the local texture features output by the CNN layer are converted into feature vectors through global average pooling and used together with the hidden state vectors output by the LSTM as attention inputs; the attention weights of the two are calculated separately through parallel attention heads, and the feature vectors are weighted, fused and concatenated to obtain fused features that include local time-frequency patterns and global temporal evolution.

[0037] The specific parameters of the Attention layer are as follows: the number of attention heads is set to 8, and the key, query and value of each attention head are calculated in parallel. Each attention head sums the value vector according to its weight, and the sums are concatenated to form the final fused feature.

[0038] In some embodiments, the fully connected layer outputs diagnostic results based on fusion features. These diagnostic results represent the operating status of the motor bearings; specifically, they include normal operation, inner race fault, outer race fault, rolling element fault, and demagnetization fault.

[0039] For example, the activation function of the first fully connected layer is ReLU; the activation function of the second fully connected layer is softmax, and the output is the probability value of each type of fault. The category with the highest probability is the preliminary fault diagnosis result.

[0040] In this implementation scheme, the diagnostic structure output by the fully connected layer is a preliminary result, and further optimization of the diagnostic structure is required.

[0041] For example, the diagnostic results are optimized based on the sliding window hard voting method to obtain the final fault diagnosis result. The specific steps are as follows: Sliding window settings: Set the sliding window length to 20, which means that 20 consecutive preliminary fault diagnosis results will be counted each time (each result corresponds to a set of vibration signal diagnosis outputs). Voting statistics: Count the 20 preliminary results in the window and count the number of occurrences of various faults (including normal status); Final result determination: The most frequent fault type is taken as the final diagnosis result for that window. If there are two or more most frequent fault types (a tie), the preliminary result in the middle of the window is taken as the final result to ensure the uniqueness of the result.

[0042] The final result is compared with that of traditional algorithms. Figure 2 and Figure 3 As shown, the hybrid neural network architecture (i.e., CNN-LSTM-Attention) in this implementation scheme outperforms the traditional CNN-LSTM architecture in both convergence speed and accuracy.

[0043] Figure 4This schematically illustrates a flowchart 400 of a motor bearing fault diagnosis method according to yet another embodiment of the present disclosure. Figure 4 As shown, a schematic diagram of the flowchart 400 of the motor bearing fault diagnosis method of this disclosure embodiment may include steps S401 to S409.

[0044] Step S401: Obtain the one-dimensional vibration signal of the motor.

[0045] S402. After the conditioning signal removes interference, the one-dimensional vibration signal is preprocessed by continuous wavelet transform and time-frequency features are extracted. The one-dimensional vibration signal is then converted into a two-dimensional time-frequency image containing complete time-frequency information, which is used as the input of the hybrid neural network.

[0046] S403. Input the two-dimensional time-frequency graph into a pre-trained hybrid neural network, wherein the hybrid neural network includes CNN layers, LSTM layers, attention layers and fully connected layers.

[0047] The S404, CNN, and LSTM layers are used to process the two-dimensional time-frequency map in parallel, and extract the local texture features and temporal dynamic evolution features of the two-dimensional time-frequency map, respectively.

[0048] S405, the multi-head attention mechanism weightedly fuses local texture features with temporal dynamic evolution features to obtain fused features.

[0049] S406, the fully connected layer outputs preliminary diagnostic results based on the fusion characteristics.

[0050] S407. By using the sliding window hard voting method, the preliminary result with the highest frequency within the window is used as the final fault diagnosis result to reduce random errors.

[0051] Figure 5 A schematic diagram of a motor bearing fault diagnosis device according to another embodiment of the present disclosure is shown. Figure 5 As shown, the motor bearing fault diagnosis device 500 includes: The signal acquisition module 501 is used to acquire the one-dimensional vibration signal of the motor and convert the one-dimensional vibration signal into a two-dimensional time-frequency diagram. The diagnostic module 502 is used to input the two-dimensional time-frequency image into a pre-trained hybrid neural network model to obtain diagnostic results. The hybrid neural network model includes CNN layers, LSTM layers, attention layers, and fully connected layers. The CNN and LSTM layers are used to process the two-dimensional time-frequency image in parallel and extract the local texture features and temporal dynamic evolution features of the two-dimensional time-frequency image, respectively. The attention layer is used to perform weighted fusion of the local texture features and temporal dynamic evolution features to obtain fused features. The fully connected layer is used to output diagnostic results based on the fused features.

[0052] In some embodiments, acquiring a one-dimensional vibration signal of the motor and converting the one-dimensional vibration signal into a two-dimensional time-frequency diagram includes: The one-dimensional vibration signal of the motor is acquired, and the one-dimensional vibration signal is processed to remove interference. The one-dimensional vibration signal after interference removal is subjected to continuous wavelet transform to balance the time resolution and frequency resolution of the one-dimensional vibration signal, and a two-dimensional time-frequency diagram corresponding to the one-dimensional vibration signal is generated.

[0053] In some embodiments, the CNN layer comprises three convolutional layers, two max-pooling layers, and one flattening layer, wherein the convolutional layers employ the ReLU activation function to extract local texture features from the two-dimensional time-frequency map.

[0054] In some embodiments, the Attention layer is a multi-head attention mechanism, and the method for obtaining fused features is as follows: The attention weights for local texture features and temporal dynamic evolution features are calculated separately using a parallel attention head; The attention weights for local texture features and temporal dynamic evolution features are normalized. Based on the attention weights after normalization, local texture features and temporal dynamic evolution features are weighted and fused to obtain fused features.

[0055] In some embodiments, it also includes: The diagnostic results are optimized based on the sliding window hard voting method to obtain the final fault diagnosis result.

[0056] In some embodiments, the diagnostic results are optimized based on a sliding window hard voting method to obtain the final fault diagnosis result, including: Set the sliding window length to 20, meaning that 20 consecutive preliminary fault diagnosis results are counted each time; Count the 20 preliminary fault diagnosis results in the sliding window and count the number of occurrences of each type of fault; The most frequently occurring fault type is taken as the final diagnostic result corresponding to the sliding window. If there are two or more most frequently occurring fault types, the preliminary fault diagnosis result at the middle position of the sliding window is taken as the final diagnostic result to ensure the uniqueness of the diagnostic result.

[0057] Figure 6 A schematic diagram of an electronic device according to an embodiment of the present disclosure is shown below, in conjunction with... Figure 6 This specification describes some embodiments of electronic devices.

[0058] Reference Figure 6The electronic device 600 may include one or more of the following components: processing component 602, memory 604, power supply component 606, multimedia component 608, audio component 610, input / output (I / O) interface 612, sensor component 616, and communication component 618.

[0059] Processing component 602 typically controls the overall operation of electronic device 600, such as operations associated with display, telephone calls, data communication, camera operation, and recording operations. Processing component 602 may include one or more processors 620 to execute instructions. Furthermore, processing component 602 may include one or more modules to facilitate interaction between processing component 602 and other components. For example, processing component 602 may include a multimedia module to facilitate interaction between multimedia component 608 and processing component 602. As another example, processing component 602 may read executable instructions from memory to implement relevant functions of the electronic device.

[0060] Memory 604 is configured to store various types of data to support the operation of electronic device 400. Examples of this data include instructions for any application or method operating on electronic device 600, contact data, phonebook data, messages, pictures, videos, etc. Memory 604 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0061] Power supply component 606 provides power to various components of electronic device 600. Power supply component 606 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 600.

[0062] Multimedia component 608 includes a display screen that provides an output interface between electronic device 600 and user. In some embodiments, multimedia component 608 includes a front-facing camera and / or a rear-facing camera. When electronic device 600 is in an operating mode, such as shooting mode or video mode, the front-facing camera and / or rear-facing camera can receive external multimedia data. Each front-facing camera and rear-facing camera can be a fixed optical lens system or have focal length and optical zoom capabilities.

[0063] Audio component 610 is configured to output and / or input audio signals. For example, audio component 610 includes a microphone (MIC) configured to receive external audio signals when electronic device 600 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 606 or transmitted via communication component 618. In some embodiments, audio component 610 also includes a speaker for outputting audio signals.

[0064] I / O interface 612 provides an interface between processing component 602 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, power buttons, and lock buttons.

[0065] Sensor assembly 616 includes one or more sensors for providing state assessments of various aspects of electronic device 600. For example, sensor assembly 616 can detect the on / off state of electronic device 600, the relative positioning of components such as the display and keypad of electronic device 600, changes in position of electronic device 600 or a component of electronic device 600, the presence or absence of user contact with electronic device 600, orientation or acceleration / deceleration of electronic device 600, and temperature changes of electronic device 600. Sensor assembly 616 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 616 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 616 may also include an accelerometer, gyroscope, magnetometer, pressure sensor, or temperature sensor.

[0066] Communication component 618 is configured to facilitate wired or wireless communication between electronic device 600 and other devices. Electronic device 600 can access wireless networks based on communication standards, such as Wi-Fi, 2G, 3G, 6G, 5G, or combinations thereof. In one exemplary embodiment, communication component 618 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 618 also includes a short-range communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0067] In an exemplary embodiment, the electronic device 600 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components.

[0068] In an exemplary embodiment, a non-transitory computer-readable storage medium, such as a memory for instructions, is also provided, which, when executed by a processor of an electronic device, enables the electronic device to perform the motor bearing fault diagnosis method of any embodiment of the present disclosure.

[0069] Non-transitory computer-readable storage media can be ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage devices, etc.

[0070] In a fifth aspect, embodiments of the present invention also provide a data acquisition system, including a vibration acquisition board and the above-mentioned electronic equipment, wherein the vibration acquisition board includes: Multiple vibration sensors are installed in the drive bearing housing and the non-drive bearing housing of the motor, respectively; The signal conditioning circuit includes a differential amplifier circuit, a low-pass filter circuit, and a voltage follower circuit connected to the vibration sensor, which are used to amplify and filter the weak vibration signal output by the vibration sensor. The data acquisition chip has its input terminal connected to the output terminal of the signal conditioning circuit. It is used to convert the conditioned analog vibration signal into a digital signal. The output terminal of the data acquisition chip is electrically connected to the electronic device.

[0071] Among them, the differential amplifier circuit is used to amplify the difference and suppress common mode; the low-pass filter circuit is used to allow low-frequency signals to pass through and suppress or attenuate high-frequency signals; the voltage follower circuit has high input impedance, low output impedance, and a voltage gain of 1, and is used to isolate the preceding and following stages to achieve impedance matching.

[0072] In one embodiment, the input terminal of the data acquisition chip is connected to the output terminal of the signal conditioning circuit to convert the conditioned analog vibration signal into a digital signal. The data acquisition chip can be a 16-bit precision ADC chip, with a sampling frequency of 20kHz and a single-channel data acquisition length of 2048 points. It is used to convert the conditioned analog vibration signal into a digital signal and transmit the digital signal to the top-level computer through an Ethernet interface. The storage format is a CSV file for easy subsequent data processing.

[0073] The electronic device reads the CSV format digital vibration signal acquired by the vibration acquisition board. By removing null and outlier values ​​from the data, data integrity is ensured. The Morse wavelet is selected as the basis wavelet to perform CWT transformation on the cleaned one-dimensional vibration signal, generating a two-dimensional time-frequency matrix. The amplitude of the time-frequency matrix represents the signal energy at the corresponding time-frequency point. By converting the time-frequency matrix into a grayscale image and cropping out invalid edge areas, the image size is ensured to fit the input requirements.

[0074] Compared with existing technologies: 1. Parallel processing of the same time-frequency image using CNN and LSTM layers avoids feature compression and loss in sequential structures, fully preserving the complementarity of local texture features and global dynamic features, thus improving feature representation capabilities. An Attention layer is introduced to focus on key fault features such as periodic impact stripes through multi-dimensional weight calculations, reducing random noise interference and making the fused features more discriminative, especially improving the diagnostic accuracy of non-mechanical faults such as demagnetization.

[0075] 2. Through the synergistic effect of parallel structure and attention layer, the accuracy is improved compared with the traditional CNN-LSTM serial model in the diagnosis of multiple types of positive faults, and the model training convergence speed is greatly improved.

[0076] 3. The hardware acquisition module is adapted to various motor application scenarios, the signal conditioning circuit can remove environmental noise, and the sliding window hard voting method reduces real-time diagnostic errors, making it suitable for complex working conditions in industrial sites.

[0077] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for diagnosing motor bearing faults, characterized in that, Includes the following steps: The one-dimensional vibration signal of the motor is acquired and converted into a two-dimensional time-frequency diagram. A two-dimensional time-frequency image is input into a pre-trained hybrid neural network model to obtain diagnostic results. The hybrid neural network model includes a CNN layer, an LSTM layer, an Attention layer, and a fully connected layer. The CNN layer and the LSTM layer process the two-dimensional time-frequency image in parallel, extracting local texture features and temporal dynamic evolution features from the image, respectively. The Attention layer weightedly fuses the local texture features and the temporal dynamic evolution features to obtain a fused feature. The fully connected layer is used to output the diagnostic results based on the fusion features.

2. The method for diagnosing motor bearing faults according to claim 1, characterized in that, The step of acquiring the one-dimensional vibration signal of the motor and converting the one-dimensional vibration signal into a two-dimensional time-frequency diagram includes: A one-dimensional vibration signal of the motor is acquired, and the one-dimensional vibration signal is subjected to interference removal processing. The one-dimensional vibration signal after interference removal is subjected to continuous wavelet transform to balance the time resolution and frequency resolution of the one-dimensional vibration signal, and a two-dimensional time-frequency diagram corresponding to the one-dimensional vibration signal is generated.

3. The method for diagnosing motor bearing faults according to claim 1, characterized in that, The CNN layer comprises three convolutional layers, two max pooling layers, and one flattening layer. The convolutional layers employ the ReLU activation function to extract the local texture features of the two-dimensional time-frequency map.

4. The method for diagnosing motor bearing faults according to claim 1, characterized in that, The Attention layer employs a multi-head attention mechanism, and the method for obtaining the fused features is as follows: The attention weights of the local texture features and the temporal dynamic evolution features are calculated using a parallel attention head. The attention weights of the local texture features and the temporal dynamic evolution features are normalized. Based on the attention weights after normalization, the local texture features and the temporal dynamic evolution features are weighted and fused to obtain the fused features.

5. The method for diagnosing motor bearing faults according to claim 1, characterized in that, Also includes: The diagnostic results are optimized using a sliding window hard voting method to obtain the final fault diagnosis result.

6. The method for diagnosing motor bearing faults according to claim 5, characterized in that, The optimization of the diagnostic results based on the sliding window hard voting method to obtain the final fault diagnosis result includes: Set the sliding window length to 20, meaning that 20 consecutive preliminary fault diagnosis results are counted each time; The 20 preliminary fault diagnosis results within the sliding window are counted, and the occurrence frequency of each type of fault is statistically analyzed. The most frequently occurring fault type is taken as the final diagnostic result corresponding to the sliding window. If there are two or more most frequently occurring types, the preliminary fault diagnosis result at the middle position of the sliding window is taken as the final diagnostic result to ensure the uniqueness of the diagnostic result.

7. A motor bearing fault diagnosis device, characterized in that, include: The signal acquisition module is used to acquire the one-dimensional vibration signal of the motor and convert the one-dimensional vibration signal into a two-dimensional time-frequency diagram. A diagnostic module is used to input a two-dimensional time-frequency image into a pre-trained hybrid neural network model to obtain diagnostic results. The hybrid neural network model includes CNN layers, LSTM layers, attention layers, and fully connected layers. The CNN and LSTM layers process the two-dimensional time-frequency image in parallel and extract local texture features and temporal dynamic evolution features from the image, respectively. The attention layer performs a weighted fusion of the local texture features and the temporal dynamic evolution features to obtain fused features. The fully connected layer is used to output the diagnostic results based on the fusion features.

8. An electronic device, characterized in that, include: processor; A memory storing computer instructions for causing a processor to perform the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The device stores computer instructions for implementing the method according to any one of claims 1 to 6.

10. A data acquisition system, characterized in that, Includes a vibration acquisition board and the electronic device as described in claim 8, wherein the vibration acquisition board includes: Multiple vibration sensors are installed in the drive bearing housing and the non-drive bearing housing of the motor, respectively; The signal conditioning circuit includes a differential amplifier circuit, a low-pass filter circuit, and a voltage follower circuit connected to the vibration sensor, used to amplify and filter the weak vibration signal output by the vibration sensor. The data acquisition chip has its input terminal connected to the output terminal of the signal conditioning circuit, and is used to convert the conditioned analog vibration signal into a digital signal. The output terminal of the data acquisition chip is electrically connected to the electronic device.

Images