Methods, devices, equipment and media for detecting motor bearing faults in port equipment

Abstract

This application provides a method, apparatus, equipment, and medium for detecting motor bearing faults in port equipment. In this method, the acquired vibration time-domain signal of the motor bearing is processed by signal conversion to obtain a vibration frequency-domain signal and a vibration time-frequency-domain signal; then, the vibration time-domain signal, vibration frequency-domain signal, and vibration time-frequency-domain signal are image-processed to obtain a combined vibration signal image; this combined vibration signal image is then input into a feature extraction model to obtain a vibration feature vector; finally, the vibration feature vector is input into a target random forest model to obtain the fault detection result. This solution improves detection accuracy by detecting based on the vibration time-domain signal, vibration frequency-domain signal, and vibration time-frequency-domain signal, and by using a random forest model for classification.

Description

Technical Field

[0001] This application relates to the field of port technology, and in particular to a method, apparatus, equipment and medium for detecting motor bearing faults in port equipment. Background Technology

[0002] In the port sector, port equipment needs to operate as continuously as possible to ensure the normal functioning of the port. Motor bearings are crucial components affecting the safety and reliability of port equipment. During long-term operation, motor bearings are subject to load variations and environmental influences, making them prone to failure and necessitating fault detection.

[0003] In the existing technology, the fault detection of motor bearings usually involves installing a vibration sensor at the motor bearing, inputting the collected vibration signal into a trained neural network model for fault detection, and obtaining the detection result.

[0004] In summary, existing motor fault detection methods rely solely on vibration signals, resulting in low detection accuracy. Summary of the Invention

[0005] This application provides a method, apparatus, equipment, and medium for detecting motor bearing faults in port equipment, which solves the problem of low detection accuracy caused by using only vibration signals for detection in the prior art.

[0006] In a first aspect, embodiments of this application provide a method for detecting motor bearing faults in port equipment, including:

[0007] The acquired vibration time-domain signal of the motor bearing is processed by signal conversion to obtain the vibration frequency-domain signal and the vibration time-frequency-domain signal;

[0008] The vibration time-domain signal, the vibration frequency-domain signal, and the vibration time-frequency-domain signal are processed into an image to obtain a combined vibration signal image.

[0009] The combined vibration signal image is input into a feature extraction model to obtain a vibration feature vector; the feature extraction model is a target classification model with the classification head removed, and the target classification model is a neural network model that has been pre-trained through transfer learning to obtain fault detection results based on the combined vibration signal image;

[0010] The vibration feature vector is input into the target random forest model to obtain the fault detection result. The fault detection result is an indicator that the motor bearing is fault-free or a fault type indicator. The target random forest model is a pre-trained random forest model that obtains the fault detection result based on the vibration feature vector.

[0011] In one possible implementation, the step of performing signal conversion processing on the acquired vibration time-domain signal of the motor bearing to obtain a vibration frequency-domain signal and a vibration time-frequency-domain signal includes:

[0012] The vibration time-domain signal is processed by fast Fourier transform and amplitude spectrum extraction to obtain the vibration frequency-domain signal;

[0013] The vibration time-domain signal is subjected to Hilbert-Huang transform to obtain the vibration time-frequency domain signal.

[0014] In one possible implementation, the step of performing image processing on the vibration time-domain signal, the vibration frequency-domain signal, and the vibration time-frequency-domain signal to obtain a combined vibration signal image includes:

[0015] The vibration time-domain signal, the vibration frequency-domain signal, and the vibration time-frequency-domain signal are truncated and scaled to obtain a time-domain scaled signal, a frequency-domain scaled signal, and a time-frequency-domain scaled signal.

[0016] The time-domain scaled signal is arranged according to a preset number of rows or a preset number of columns to obtain the channel value of the first channel of each pixel in the vibration signal composite image; the preset number of rows is the number of rows of pixels in the vibration signal composite image, and the preset number of columns is the number of columns of pixels in the vibration signal composite image.

[0017] The frequency domain scaling signal is arranged according to the preset number of rows or the preset number of columns to obtain the channel value of the second channel of each pixel in the vibration signal combined image;

[0018] For each pixel in the combined vibration signal image, the channel value of the third channel of the pixel is determined based on the channel values ​​of the first and second channels of the pixel, as well as the time-frequency domain scaling signal.

[0019] In one possible implementation, the step of truncating and scaling the vibration time-domain signal, the vibration frequency-domain signal, and the vibration time-frequency-domain signal to obtain a time-domain scaled signal, a frequency-domain scaled signal, and a time-frequency-domain scaled signal includes:

[0020] According to a preset number of segments, the vibration time-domain signal and the vibration frequency-domain signal are segmented to obtain a time-domain segmented signal and a frequency-domain segmented signal. The preset number of segments is the product of the preset number of rows and the preset number of columns.

[0021] Based on the time of the time-domain intercepted signal and the frequency of the frequency-domain intercepted signal, the vibration time-frequency domain signal is intercepted to obtain the time-frequency domain intercepted signal.

[0022] Based on a preset channel value threshold, the time-domain truncated signal, the frequency-domain truncated signal, and the time-frequency-domain truncated signal are scaled to obtain a time-domain scaled signal, a frequency-domain scaled signal, and a time-frequency-domain scaled signal.

[0023] In one possible implementation, determining the channel value of the third channel of the pixel based on the channel values ​​of the first and second channels of the pixel, and the time-frequency domain scaling signal, includes:

[0024] The time corresponding to the channel value of the first channel of the pixel in the time-domain scaled signal is used as the time index;

[0025] The frequency corresponding to the channel value of the second channel of the pixel in the frequency domain scaling signal is used as the frequency index;

[0026] In the time-frequency domain scaled signal, the signal value corresponding to the time index and the frequency index is used as the channel value of the third channel of the pixel.

[0027] In one possible implementation, before performing signal conversion processing on the acquired time-domain vibration signal of the motor bearing to obtain the frequency-domain vibration signal and the time-frequency-domain vibration signal, the method further includes:

[0028] Obtain a combined image dataset of the ImageNet dataset and vibration signals;

[0029] The initial neural network model is trained based on the ImageNet dataset to obtain a first neural network model, which includes a backbone network, a first-stage layer, a first downsampling layer, a second-stage layer, a second downsampling layer, a third-stage layer, a third downsampling layer, a fourth-stage layer, a global average pooling layer, and a classification head.

[0030] Freeze the backbone network, first stage layer, first downsampling layer and second stage layer in the first neural network model to obtain the second neural network model;

[0031] The second neural network model is trained based on the vibration signal combined image dataset to obtain the target classification model.

[0032] In one possible implementation, the method further includes:

[0033] The feature extraction model is obtained by removing the classifier head from the target random forest model.

[0034] The images in the vibration signal combined image dataset are sequentially input into the feature extraction model to obtain a training feature vector dataset.

[0035] The random forest model is trained based on the training feature vector dataset to obtain the target random forest model.

[0036] Secondly, embodiments of this application provide a motor bearing fault detection device for port equipment, comprising:

[0037] Processing module, used for:

[0038] The acquired vibration time-domain signal of the motor bearing is processed by signal conversion to obtain the vibration frequency-domain signal and the vibration time-frequency-domain signal;

[0039] The vibration time-domain signal, the vibration frequency-domain signal, and the vibration time-frequency-domain signal are processed into an image to obtain a combined vibration signal image.

[0040] The feature extraction module is used to input the combined vibration signal image into the feature extraction model to obtain a vibration feature vector; the feature extraction model is a target classification model with the classification head removed, and the target classification model is a neural network model that has been pre-trained through transfer learning to obtain fault detection results based on the combined vibration signal image;

[0041] The detection module is used to input the vibration feature vector into the target random forest model to obtain the fault detection result. The fault detection result is an indicator indicating that the motor bearing is fault-free, or a fault type indicator. The target random forest model is a pre-trained random forest model that obtains the fault detection result based on the vibration feature vector.

[0042] Thirdly, embodiments of this application provide an electronic device, including:

[0043] Processor, memory, communication interface;

[0044] The memory is used to store the executable instructions of the processor;

[0045] The processor is configured to execute the motor bearing fault detection method for port equipment according to any one of the first aspects by executing the executable instructions.

[0046] Fourthly, embodiments of this application provide a readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method for detecting motor bearing faults in port equipment as described in any of the first aspects.

[0047] The port equipment motor bearing fault detection method, apparatus, equipment, and medium provided in this application embodiment perform signal conversion processing on the acquired motor bearing vibration time-domain signal to obtain vibration frequency-domain signal and vibration time-frequency domain signal; then, the vibration time-domain signal, vibration frequency-domain signal, and vibration time-frequency domain signal are image-processed to obtain a vibration signal combination image; the vibration signal combination image is then input into a feature extraction model to obtain a vibration feature vector; the vibration feature vector is then input into a target random forest model to obtain the fault detection result. This scheme improves detection accuracy by detecting based on vibration time-domain signal, vibration frequency-domain signal, and vibration time-frequency domain signal, and by using a random forest model for classification. Attached Figure Description

[0048] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0049] Figure 1 A flowchart illustrating an embodiment of the port equipment motor bearing fault detection method provided in this application;

[0050] Figure 2 A schematic diagram of the vibration time-domain signal provided in this application;

[0051] Figure 3 A schematic diagram of the vibration frequency domain signal provided in this application;

[0052] Figure 4 A schematic diagram of the vibration time-frequency domain signal provided in this application;

[0053] Figure 5 The layout flowchart provided for this application;

[0054] Figure 6 A schematic diagram of the motor bearing fault detection process provided in this application;

[0055] Figure 7 A flowchart illustrating Embodiment 2 of the port equipment motor bearing fault detection method provided in this application;

[0056] Figure 8 A flowchart illustrating Embodiment 3 of the port equipment motor bearing fault detection method provided in this application;

[0057] Figure 9 A schematic diagram of an embodiment of the motor bearing fault detection device for port equipment provided in this application;

[0058] Figure 10 This is a schematic diagram of the structure of an electronic device provided in this application.

[0059] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0060] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0061] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0062] In the port sector, port equipment needs to operate as continuously as possible to ensure the normal functioning of the port. Motor bearings are crucial components affecting the safety and reliability of port equipment. During long-term operation, motor bearings are subject to load variations and environmental influences, making them prone to failure and necessitating fault detection.

[0063] In existing technologies, fault detection of motor bearings typically involves installing vibration sensors at the bearing location and inputting the collected vibration signals into a trained neural network model for fault detection, thus obtaining the detection result. However, relying solely on vibration signals for detection leads to low accuracy.

[0064] To address the problems existing in the prior art, the inventors, during their research on motor bearing fault detection methods for equipment, discovered that the collected vibration signals are time-domain signals. While time-domain signals can retain the characteristics of instantaneous bearing impact, they cannot reproduce periodic and non-stationary features. Therefore, to improve detection accuracy, time-domain, frequency-domain, and time-frequency-domain signals can be used for detection. Frequency-domain signals can retain periodic features, while time-frequency-domain signals can retain non-stationary features. These three signals are converted into images, and then feature vectors are extracted using a feature extraction model. A random forest model is then used for classification to obtain the fault detection results, further improving detection accuracy. Based on the above inventive concept, the motor bearing fault detection scheme for equipment described in this application was designed.

[0065] The execution subject of the port equipment motor bearing fault detection method in this application can be a computer, or a server, terminal equipment, etc. This application does not limit it. The following description uses a computer as an example.

[0066] The following provides an example illustrating the application scenarios of the motor bearing fault detection method for port equipment provided in this application.

[0067] For example, in this application scenario, in a port, port equipment is constantly running. In order to detect the failure of the motor bearing of the port equipment in a timely manner and carry out maintenance, it is necessary to install a vibration sensor at the motor bearing.

[0068] The vibration sensor is connected to the computer. After the computer collects the vibration time-domain signal of the motor bearing through the vibration sensor, it performs signal conversion processing to obtain the vibration frequency-domain signal and the vibration time-frequency-domain signal.

[0069] Then, the vibration time-domain signal, vibration frequency-domain signal, and vibration time-frequency-domain signal are processed into images to obtain a combined vibration signal image.

[0070] The combined vibration signal image is input into the feature extraction model to obtain the vibration feature vector.

[0071] The vibration feature vector is then input into the target random forest model to obtain the fault detection results.

[0072] The computer can display the fault detection results so that staff can determine whether the motor bearing is faulty. If the motor bearing is faulty, the fault detection results can also determine the type of fault so that the motor bearing can be repaired in a timely manner.

[0073] It should be noted that the above scenario is only an example of an application scenario provided by the embodiments of this application. The embodiments of this application do not limit the actual form of the various devices included in the scenario, nor do they limit the interaction method between devices. In the specific application of the solution, it can be set according to actual needs.

[0074] The technical solution of this application will now be described in detail through specific embodiments. It should be noted that the following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0075] Figure 1 This is a flowchart illustrating an embodiment of the port equipment motor bearing fault detection method provided in this application. This embodiment describes how a computer performs motor bearing fault detection based on vibration time-domain signals, vibration frequency-domain signals, and vibration time-frequency-domain signals. The method in this embodiment can be implemented through software, hardware, or a combination of both. Figure 1 As shown, the method for detecting motor bearing failures in this port equipment specifically includes the following steps:

[0076] S101: Perform signal conversion processing on the acquired vibration time-domain signal of the motor bearing to obtain the vibration frequency-domain signal and the vibration time-frequency-domain signal.

[0077] In this step, after the computer acquires the vibration time-domain signal of the motor bearing, it performs signal conversion processing to improve the accuracy of fault detection, thereby obtaining the vibration frequency-domain signal and the vibration time-frequency-domain signal.

[0078] Specifically, the vibration time-domain signal is processed by fast Fourier transform and amplitude spectrum extraction to obtain the vibration frequency-domain signal.

[0079] Vibration time-domain signals can be used Let t represent time. N represents the vibration quantity, where N is the number of vibration quantities in the vibration time-domain signal, and also the number of moments.

[0080] Performing a Fast Fourier Transform on the vibration time-domain signal, that is, using the formula To obtain the complex spectrum signal Where k represents frequency, Represents the spectrum of a complex number.

[0081] Next, amplitude spectrum extraction is performed on the complex spectrum signal, which is done using the formula... The vibration frequency domain signal is obtained. Where k represents frequency, This represents the amplitude spectrum.

[0082] It should be noted that the vibration amount can be acceleration, velocity or displacement. The embodiments of this application do not limit the vibration amount, which can be determined according to the actual situation.

[0083] The vibration time-domain signal is transformed using the Hilbert-Huang transform to obtain the vibration time-frequency domain signal. The vibration time-frequency domain signal represents the energy values ​​corresponding to different times and frequencies.

[0084] For example, Figure 2 A schematic diagram of the vibration time-domain signal provided in this application is shown below. Figure 2 As shown, the horizontal axis represents time, and the vertical axis represents the vibration amount.

[0085] Figure 3 A schematic diagram of the vibration frequency domain signal provided in this application is shown below. Figure 3 As shown, the horizontal axis represents time, and the vertical axis represents the amplitude spectrum.

[0086] Figure 4 A schematic diagram of the vibration time-frequency domain signal provided in this application is shown below. Figure 4 As shown, the figure includes a time axis, a frequency axis, and an energy value axis. The raised part in the figure is the instantaneous frequency trajectory.

[0087] S102: Perform image processing on the vibration time-domain signal, vibration frequency-domain signal, and vibration time-frequency-domain signal to obtain a combined vibration signal image.

[0088] In this step, after the computer obtains the vibration time-domain signal, vibration frequency-domain signal, and vibration time-frequency-domain signal, it performs image processing on the vibration time-domain signal, vibration frequency-domain signal, and vibration time-frequency-domain signal to obtain a combined vibration signal image so that the subsequent model can process it.

[0089] Specifically, the vibration time-domain signal, vibration frequency-domain signal, and vibration time-frequency-domain signal are first truncated and scaled to obtain the time-domain scaled signal, frequency-domain scaled signal, and time-frequency-domain scaled signal.

[0090] Since there is a large amount of data in the three signals, the number of vibration magnitudes, amplitude spectra, and energy values ​​should be the same as the number of pixels in the combined vibration signal image. Therefore, truncating is necessary. The truncating method is as follows:

[0091] Based on a preset number of segments, the vibration time-domain signal and vibration frequency-domain signal are truncated to obtain a time-domain truncated signal and a frequency-domain truncated signal. The number of vibration quantities in the time-domain truncated signal and the number of amplitude spectra in the frequency-domain truncated signal are both preset number of segments.

[0092] Based on the time of the time-domain truncation signal and the frequency of the frequency-domain truncation signal, the vibration time-frequency domain signal is truncated to obtain the time-frequency domain truncation signal. That is, for each vibration quantity in the time-domain truncation signal, and for each amplitude spectrum in the frequency-domain truncation signal, the energy value corresponding to that time and frequency in the vibration time-frequency domain signal is extracted.

[0093] It should be noted that the preset number of segments is the product of the preset number of rows and the preset number of columns. The preset number of rows is the number of rows of pixels in the combined vibration signal image, and the preset number of columns is the number of columns of pixels in the combined vibration signal image. The preset number of rows and preset number of columns can be 32, 64, 128, etc. This application embodiment does not limit the preset number of rows, preset number of columns, and preset number of segments, and can be determined according to the actual situation.

[0094] Since each pixel in the vibration signal composite image has three channels—R (red), G (green), and B (blue)—and each channel has a threshold value, scaling of the time-domain, frequency-domain, and time-frequency-domain truncated signals is required to form an image. The scaling method is as follows:

[0095] Based on the preset channel value threshold, the time-domain truncation signal, frequency-domain truncation signal, and time-frequency-domain truncation signal are scaled to obtain the time-domain scaled signal, frequency-domain scaled signal, and time-frequency-domain scaled signal.

[0096] The preset channel value thresholds include an upper limit and a lower limit. The upper limit is 255 and the lower limit is 0. Therefore, the vibration amount in the time domain truncated signal is scaled to between 0 and 255, the amplitude spectrum in the frequency domain truncated signal is scaled to between 0 and 255, and the energy value in the time and frequency domain truncated signal is scaled to between 0 and 255.

[0097] After cropping and scaling the three signals, an image can be generated. First, the time-domain scaled signals are arranged according to a preset number of rows or columns to obtain the channel value of the first channel of each pixel in the vibration signal composite image.

[0098] The arrangement method can be as follows: the time-domain scaling signal is divided into segments according to a preset number of rows, and the vibration amount of each segment is used as the channel value of the first channel of a row of pixels in the vibration signal combined image.

[0099] For example, Figure 5 The layout flowchart provided for this application is as follows: Figure 5 As shown, the preset number of rows is 4 and the preset number of columns is 3. The time-domain scaling signal is divided into 4 segments on average, and the vibration amount of each segment is used as the channel value of the first channel of a row of pixels in the vibration signal composite image.

[0100] Another arrangement method is to divide the time-domain scaled signal into segments according to a preset number of columns, and use the vibration amount of each segment as the channel value of the first channel of a column of pixels in the vibration signal combined image.

[0101] Based on a preset number of rows or columns, the frequency domain scaled signal is arranged to obtain the channel value of the second channel of each pixel in the vibration signal composite image.

[0102] It should be noted that the arrangement of frequency domain scaled signals is similar to that of time domain scaled signals, and will not be elaborated here.

[0103] For each pixel in the combined vibration signal image, the channel value of the third channel of the pixel is determined based on the channel values ​​of the first and second channels of the pixel, as well as the time-frequency domain scaling signal.

[0104] In other words, the time corresponding to the channel value of the first channel of the pixel in the time-domain scaled signal is used as the time index. The frequency corresponding to the channel value of the second channel of the pixel in the frequency-domain scaled signal is used as the frequency index. The signal values ​​corresponding to the time index and frequency index in the time-frequency domain scaled signal are used as the channel value of the third channel of the pixel.

[0105] It should be noted that the correspondence between the first, second, and third channels and the R, G, and B channels can be as follows: the first channel is the R channel, the second channel is the G channel, and the third channel is the B channel. Alternatively, the first channel can be the G channel, the second channel is the R channel, and the third channel is the B channel. Another possible correspondence is: the first channel is the B channel, the second channel is the R channel, and the third channel is the G channel. This application does not limit the channel correspondence method; it can be determined according to the actual situation.

[0106] S103: Input the combined vibration signal image into the feature extraction model to obtain the vibration feature vector.

[0107] In this step, after the computer obtains the combined image of vibration signals, it inputs the combined image of vibration signals into the feature extraction model to obtain the vibration feature vector.

[0108] Among them, the feature extraction model is the target classification model with the classification head removed, and the target classification model is a neural network model that has been pre-trained through transfer learning and is used to obtain fault detection results based on the combined images of vibration signals.

[0109] The target classification model is the ConvNeXt model. Features are extracted using the ConvNeXt model, whose large kernel depthwise separable convolutions can cover longer vibration periods, identifying periodic fault texture information from a global perspective, thus ignoring spatial transient noise interference and improving detection accuracy.

[0110] S104: Input the vibration feature vector into the target random forest model to obtain the fault detection results.

[0111] In this step, after the computer obtains the vibration feature vector, it inputs the vibration feature vector into the target random forest model to obtain the fault detection result.

[0112] Among them, the fault detection result is either an indicator that the motor bearing is fault-free, or a fault type indicator.

[0113] The target random forest model is a pre-trained random forest model that obtains fault detection results based on vibration feature vectors.

[0114] It should be noted that the fault type identifier is an identifier for the fault type, which can be a fault in the inner ring of the bearing, a fault in the outer ring of the bearing, a fault in the bearing balls, etc. This application embodiment does not limit the fault type, and it can be determined according to the actual situation.

[0115] For example, Figure 6 This is a schematic diagram of the motor bearing fault detection process provided in this application, such as... Figure 6 As shown, the vibration time-domain signal is processed by signal conversion to obtain the vibration frequency-domain signal and the vibration time-frequency-domain signal; the vibration time-domain signal, the vibration frequency-domain signal, and the vibration time-frequency-domain signal are processed by image processing to obtain a combined vibration signal image; the combined vibration signal image is input into the feature extraction model to obtain the vibration feature vector; the vibration feature vector is input into the target random forest model to obtain the fault detection result.

[0116] The port equipment motor bearing fault detection method provided in this embodiment involves signal conversion processing of the acquired motor bearing vibration time-domain signal to obtain vibration frequency-domain and vibration time-frequency-domain signals; then, image processing of the vibration time-domain, vibration frequency-domain, and vibration time-frequency-domain signals to obtain a combined vibration signal image; finally, inputting the combined vibration signal image into a feature extraction model to obtain a vibration feature vector; and finally, inputting the vibration feature vector into a target random forest model to obtain the fault detection result. This scheme detects faults based on vibration time-domain, vibration frequency-domain, and vibration time-frequency-domain signals. The vibration time-domain signal can retain the characteristics of instantaneous impact of the bearing, the frequency-domain signal can retain periodic characteristics, and the time-frequency-domain signal can retain non-stationary characteristics, thus improving detection accuracy. In addition, the target random forest model adopts a soft voting mechanism, and the model has high anti-overfitting ability and cross-operating condition stability. Using the target random forest model for classification can reduce noise interference and improve detection accuracy.

[0117] Figure 7 This is a flowchart illustrating a second embodiment of the port equipment motor bearing fault detection method provided in this application. Based on the above embodiments, this application describes the process of transferring and training an initial neural network model using a computer to obtain a target classification model. For example... Figure 7 As shown, the method for detecting motor bearing failures in this port equipment specifically includes the following steps:

[0118] S701: Obtain the combined image dataset of the ImageNet dataset and vibration signal.

[0119] In this step, since there are few vibration time-domain signals of the motor bearing, transfer training is required in order to train the target classification model. This requires obtaining the ImageNet dataset and the vibration signal combined image dataset.

[0120] It should be noted that the ImageNet dataset is a large image dataset with over 14 million images, organized according to the WordNet hierarchy, with over 20,000 image categories and corresponding annotations for each image.

[0121] The images in the vibration signal combined image dataset are obtained using steps S101-S102 in Example 1, and each image has a corresponding fault detection result.

[0122] S702: Train the initial neural network model based on the ImageNet dataset to obtain the first neural network model.

[0123] In this step, after the computer obtains the ImageNet dataset, it can train the initial neural network model based on the ImageNet dataset to obtain the first neural network model.

[0124] It should be noted that the model training process in this step is consistent with the model training process in existing technologies, and will not be described again here.

[0125] The initial neural network model is the ConvNeXt model. The initial neural network model and the first neural network model consist of the backbone network (Stem layer), the first stage layer, the first downsampling layer, the second stage layer, the second downsampling layer, the third stage layer, the third downsampling layer, the fourth stage layer, the global average pooling layer, and the classification head, respectively.

[0126] The backbone network performs an initial downsampling of the input image, mapping it to an initial feature representation. The first-stage layer extracts shallow features such as edges, textures, and local details. The first downsampling layer reduces the spatial resolution of the feature map while increasing the number of channels, enabling subsequent networks to extract richer features within a larger receptive field. The second-stage layer extracts combined local features. The second downsampling layer performs downsampling and dimensionality upsampling. The third-stage layer integrates the previously extracted features to form stronger discriminative representations. The third downsampling layer reduces resolution and increases semantic expression density. The fourth-stage layer extracts abstract, high-level semantic features that approximate the final classification decision. The global average pooling layer compresses the two-dimensional feature map into a one-dimensional feature vector, reducing the number of parameters, mitigating overfitting, and preserving information from the entire feature map globally for use by the subsequent classification head. The classification head is used for classification.

[0127] S703: Freeze the backbone network, first stage layer, first downsampling layer and second stage layer in the first neural network model to obtain the second neural network model.

[0128] In this step, after the computer obtains the first neural network model, it freezes the backbone network, the first stage layer, the first downsampling layer, and the second stage layer in the first neural network model to obtain the second neural network model.

[0129] It should be noted that freezing means that these layers will not be changed during subsequent training.

[0130] S704: Based on the vibration signal combined image dataset, train the second neural network model to obtain the target classification model.

[0131] In this step, after the computer obtains the second neural network model, it trains the second neural network model based on the vibration signal combined image dataset to obtain the target classification model.

[0132] It should be noted that the model training process in this step is consistent with the model training process in existing technologies, and will not be described again here.

[0133] The port equipment motor bearing fault detection method provided in this embodiment trains the initial neural network model using the ImageNet dataset, giving the model a strong ability to recognize low-level textures. By freezing the backbone network, the first stage layer, the first downsampling layer, and the second stage layer, the general feature extraction capability can be preserved, effectively preventing the model from overfitting on noisy small sample target domains. Using a vibration signal combined image dataset to train the second neural network model, training can be completed with less data, resulting in a model with better feature extraction capabilities.

[0134] Figure 8 This is a flowchart illustrating Embodiment 3 of the port equipment motor bearing fault detection method provided in this application. Based on the above embodiments, this application describes the process of training a random forest model to obtain a target random forest model after the computer obtains the target classification model. Figure 8 As shown, the method for detecting motor bearing failures in this port equipment specifically includes the following steps:

[0135] S801: Remove the classifier head from the target random forest model to obtain the feature extraction model.

[0136] In this step, after the computer obtains the target random forest model, in order to obtain the data needed to train the random forest model, the classifier head in the target random forest model needs to be removed to obtain the feature extraction model.

[0137] S802: The images in the vibration signal combined image dataset are sequentially input into the feature extraction model to obtain the training feature vector dataset.

[0138] In this step, after the computer obtains the feature extraction model, it inputs the images from the vibration signal combined image dataset into the feature extraction model in sequence to obtain the training feature vector dataset.

[0139] S803: Train the random forest model based on the training feature vector dataset to obtain the target random forest model.

[0140] In this step, after the computer obtains the training feature vector dataset, it trains the random forest model based on the training feature vector dataset to obtain the target random forest model.

[0141] It should be noted that the model training process in this step is consistent with the model training process in existing technologies, and will not be described again here.

[0142] The port equipment motor bearing fault detection method provided in this embodiment improves the classification accuracy of the target random forest model by training the training feature vector dataset obtained by the feature extraction model.

[0143] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0144] Figure 9 This is a schematic diagram of an embodiment of the motor bearing fault detection device for port equipment provided in this application. Figure 9 As shown, the motor bearing fault detection device 90 of the port equipment includes:

[0145] Processing module 91 is used for:

[0146] The acquired vibration time-domain signal of the motor bearing is processed by signal conversion to obtain the vibration frequency-domain signal and the vibration time-frequency-domain signal;

[0147] The vibration time-domain signal, vibration frequency-domain signal, and vibration time-frequency-domain signal are processed into an image to obtain a combined vibration signal image.

[0148] The feature extraction module 92 is used to input the combined vibration signal image into the feature extraction model to obtain the vibration feature vector; the feature extraction model is the target classification model with the classification head removed, and the target classification model is a neural network model that has been pre-trained for transfer learning and is used to obtain fault detection results based on the combined vibration signal image;

[0149] The detection module 93 is used to input the vibration feature vector into the target random forest model to obtain the fault detection result. The fault detection result is an indicator indicating that the motor bearing is fault-free, or a fault type indicator. The target random forest model is a pre-trained random forest model that obtains the fault detection result based on the vibration feature vector.

[0150] Furthermore, processing module 91 is specifically used for:

[0151] The vibration time-domain signal is processed by fast Fourier transform and amplitude spectrum extraction to obtain the vibration frequency-domain signal;

[0152] The vibration time-domain signal is transformed by Hilbert-Huang to obtain the vibration time-frequency domain signal.

[0153] Furthermore, processing module 91 is specifically used for:

[0154] The vibration time-domain signal, vibration frequency-domain signal, and vibration time-frequency-domain signal are truncated and scaled to obtain time-domain scaled signal, frequency-domain scaled signal, and time-frequency-domain scaled signal;

[0155] The time-domain scaled signal is arranged according to a preset number of rows or columns to obtain the channel value of the first channel of each pixel in the vibration signal composite image; the preset number of rows is the number of rows of pixels in the vibration signal composite image, and the preset number of columns is the number of columns of pixels in the vibration signal composite image.

[0156] Based on a preset number of rows or columns, the frequency domain scaled signal is arranged to obtain the channel value of the second channel of each pixel in the vibration signal composite image;

[0157] For each pixel in the combined vibration signal image, the channel value of the third channel of the pixel is determined based on the channel values ​​of the first and second channels of the pixel, as well as the time-frequency domain scaling signal.

[0158] Furthermore, processing module 91 is specifically used for:

[0159] According to the preset number of segments, the vibration time domain signal and vibration frequency domain signal are segmented to obtain the time domain segmented signal and the frequency domain segmented signal. The preset number of segments is the product of the preset number of rows and the preset number of columns.

[0160] Based on the time of the time-domain intercepted signal and the frequency of the frequency-domain intercepted signal, the vibration time-frequency domain signal is intercepted to obtain the time-frequency domain intercepted signal;

[0161] Based on the preset channel value threshold, the time-domain truncation signal, frequency-domain truncation signal, and time-frequency-domain truncation signal are scaled to obtain the time-domain scaled signal, frequency-domain scaled signal, and time-frequency-domain scaled signal.

[0162] Furthermore, processing module 91 is specifically used for:

[0163] Use the time corresponding to the channel value of the first channel of a pixel in the time-domain scaled signal as the time index;

[0164] Use the frequency corresponding to the channel value of the second channel of the pixel in the frequency domain scaled signal as the frequency index;

[0165] In the time-frequency domain scaled signal, the signal values ​​corresponding to the time index and frequency index are used as the channel values ​​of the third channel of the pixel.

[0166] Furthermore, before performing signal conversion processing on the acquired vibration time-domain signal of the motor bearing to obtain the vibration frequency-domain signal and the vibration time-frequency-domain signal, the processing module 91 is also used for:

[0167] Obtain a combined image dataset of the ImageNet dataset and vibration signals;

[0168] The initial neural network model is trained based on the ImageNet dataset to obtain the first neural network model, which includes a backbone network, a first-stage layer, a first downsampling layer, a second-stage layer, a second downsampling layer, a third-stage layer, a third downsampling layer, a fourth-stage layer, a global average pooling layer, and a classification head.

[0169] Freeze the backbone network, first stage layer, first downsampling layer and second stage layer in the first neural network model to obtain the second neural network model;

[0170] Based on the vibration signal combined image dataset, the second neural network model is trained to obtain the target classification model.

[0171] Furthermore, processing module 91 is also used for:

[0172] The classifier head is removed from the target random forest model to obtain the feature extraction model;

[0173] Images from the vibration signal combined image dataset are sequentially input into the feature extraction model to obtain a training feature vector dataset.

[0174] The random forest model is trained based on the training feature vector dataset to obtain the target random forest model.

[0175] The port equipment motor bearing fault detection device provided in this embodiment is used to execute the technical solution in any of the aforementioned method embodiments. Its implementation principle and technical effect are similar, and will not be described again here.

[0176] Figure 10 This is a schematic diagram of the structure of an electronic device provided in this application. Figure 10 As shown, the electronic device 1000 includes:

[0177] Processor 1001, memory 1002, and communication interface 1003;

[0178] The memory 1002 is used to store the executable instructions of the processor 1001;

[0179] The processor 1001 is configured to execute the technical solutions in any of the foregoing method embodiments by executing executable instructions.

[0180] Optionally, the memory 1002 can be either standalone or integrated with the processor 1001.

[0181] Optionally, when the memory 1002 is a device independent of the processor 1001, the electronic device 1000 may further include:

[0182] Bus 1004, memory 1002 and communication interface 1003 are connected to processor 1001 through bus 1004 and complete communication with each other. Communication interface 1003 is used to communicate with other devices.

[0183] Optionally, the communication interface 1003 can be implemented using a transceiver. The communication interface is used to enable communication between the database access device and other devices (e.g., clients, read-write databases, and read-only databases). The memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk drive.

[0184] Bus 1004 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus.

[0185] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0186] The electronic device is used to execute the technical solutions in any of the foregoing method embodiments. Its implementation principle and technical effect are similar, and will not be described again here.

[0187] This application also provides a readable storage medium storing a computer program thereon, which, when executed by a processor, implements the technical solutions provided in any of the foregoing method embodiments.

[0188] This application also provides a computer program product, including a computer program, which, when executed by a processor, is used to implement the technical solutions provided in any of the foregoing method embodiments.

[0189] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0190] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for detecting motor bearing faults in port equipment, characterized in that, include: The acquired vibration time-domain signal of the motor bearing is processed by signal conversion to obtain the vibration frequency-domain signal and the vibration time-frequency-domain signal; The vibration time-domain signal, the vibration frequency-domain signal, and the vibration time-frequency-domain signal are processed into an image to obtain a combined vibration signal image. The combined vibration signal image is input into a feature extraction model to obtain a vibration feature vector; the feature extraction model is a target classification model with the classification head removed, and the target classification model is a neural network model that has been pre-trained through transfer learning to obtain fault detection results based on the combined vibration signal image; The vibration feature vector is input into the target random forest model to obtain the fault detection result. The fault detection result is an indicator that the motor bearing is fault-free or a fault type indicator. The target random forest model is a pre-trained random forest model that obtains the fault detection result based on the vibration feature vector.

2. The method according to claim 1, characterized in that, The process of converting the acquired time-domain vibration signal of the motor bearing into a frequency-domain vibration signal and a time-frequency-domain vibration signal includes: The vibration time-domain signal is processed by fast Fourier transform and amplitude spectrum extraction to obtain the vibration frequency-domain signal; The vibration time-domain signal is subjected to Hilbert-Huang transform to obtain the vibration time-frequency domain signal.

3. The method according to claim 1, characterized in that, The step of performing image processing on the vibration time-domain signal, the vibration frequency-domain signal, and the vibration time-frequency-domain signal to obtain a combined vibration signal image includes: The vibration time-domain signal, the vibration frequency-domain signal, and the vibration time-frequency-domain signal are truncated and scaled to obtain a time-domain scaled signal, a frequency-domain scaled signal, and a time-frequency-domain scaled signal. The time-domain scaled signal is arranged according to a preset number of rows or a preset number of columns to obtain the channel value of the first channel of each pixel in the vibration signal composite image; the preset number of rows is the number of rows of pixels in the vibration signal composite image, and the preset number of columns is the number of columns of pixels in the vibration signal composite image. The frequency domain scaling signal is arranged according to the preset number of rows or the preset number of columns to obtain the channel value of the second channel of each pixel in the vibration signal combined image; For each pixel in the combined vibration signal image, the channel value of the third channel of the pixel is determined based on the channel values ​​of the first and second channels of the pixel, as well as the time-frequency domain scaling signal.

4. The method according to claim 3, characterized in that, The step of truncating and scaling the vibration time-domain signal, the vibration frequency-domain signal, and the vibration time-frequency-domain signal to obtain a time-domain scaled signal, a frequency-domain scaled signal, and a time-frequency-domain scaled signal includes: According to a preset number of segments, the vibration time-domain signal and the vibration frequency-domain signal are segmented to obtain a time-domain segmented signal and a frequency-domain segmented signal. The preset number of segments is the product of the preset number of rows and the preset number of columns. Based on the time of the time-domain intercepted signal and the frequency of the frequency-domain intercepted signal, the vibration time-frequency domain signal is intercepted to obtain the time-frequency domain intercepted signal. Based on a preset channel value threshold, the time-domain truncated signal, the frequency-domain truncated signal, and the time-frequency-domain truncated signal are scaled to obtain a time-domain scaled signal, a frequency-domain scaled signal, and a time-frequency-domain scaled signal.

5. The method according to claim 3, characterized in that, Determining the channel value of the third channel of the pixel based on the channel values ​​of the first and second channels of the pixel, and the time-frequency domain scaling signal, includes: The time corresponding to the channel value of the first channel of the pixel in the time-domain scaled signal is used as the time index; The frequency corresponding to the channel value of the second channel of the pixel in the frequency domain scaling signal is used as the frequency index; In the time-frequency domain scaled signal, the signal value corresponding to the time index and the frequency index is used as the channel value of the third channel of the pixel.

6. The method according to any one of claims 1 to 5, characterized in that, Before performing signal conversion processing on the acquired vibration time-domain signal of the motor bearing to obtain the vibration frequency-domain signal and the vibration time-frequency-domain signal, the method further includes: Obtain a combined image dataset of the ImageNet dataset and vibration signals; The initial neural network model is trained based on the ImageNet dataset to obtain a first neural network model, which includes a backbone network, a first-stage layer, a first downsampling layer, a second-stage layer, a second downsampling layer, a third-stage layer, a third downsampling layer, a fourth-stage layer, a global average pooling layer, and a classification head. Freeze the backbone network, first stage layer, first downsampling layer and second stage layer in the first neural network model to obtain the second neural network model; The second neural network model is trained based on the vibration signal combined image dataset to obtain the target classification model.

7. The method according to claim 6, characterized in that, The method further includes: The feature extraction model is obtained by removing the classifier head from the target random forest model. The images in the vibration signal combined image dataset are sequentially input into the feature extraction model to obtain a training feature vector dataset. The random forest model is trained based on the training feature vector dataset to obtain the target random forest model.

8. A fault detection device for motor bearings in port equipment, characterized in that, include: Processing module, used for: The acquired vibration time-domain signal of the motor bearing is processed by signal conversion to obtain the vibration frequency-domain signal and the vibration time-frequency-domain signal; The vibration time-domain signal, the vibration frequency-domain signal, and the vibration time-frequency-domain signal are processed into an image to obtain a combined vibration signal image. The feature extraction module is used to input the combined vibration signal image into the feature extraction model to obtain a vibration feature vector; the feature extraction model is a target classification model with the classification head removed, and the target classification model is a neural network model that has been pre-trained through transfer learning to obtain fault detection results based on the combined vibration signal image; The detection module is used to input the vibration feature vector into the target random forest model to obtain the fault detection result. The fault detection result is an indicator indicating that the motor bearing is fault-free, or a fault type indicator. The target random forest model is a pre-trained random forest model that obtains the fault detection result based on the vibration feature vector.

9. An electronic device, characterized in that, include: Processor, memory, communication interface; The memory is used to store the executable instructions of the processor; The processor is configured to execute the motor bearing fault detection method for port equipment according to any one of claims 1 to 7 by executing the executable instructions.

10. A readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for detecting motor bearing faults in port equipment as described in any one of claims 1 to 7.

Images