Gear reduction motor fault detection method and device, computer equipment and storage medium

Abstract

The application discloses a kind of gear reduction motor fault detection method, device, computer equipment and storage medium, the method includes: obtaining the vibration signal of gear reduction motor and processing;Utilize discrete short-time Fourier transform and convert the vibration signal after processing into RGB power spectrum image, use RGB power spectrum image as motor dataset;Based on attention mechanism and ConvNeXt network model, improved ConvNeXt network model is constructed, and the improved ConvNeXt network model is trained using motor dataset, and motor quality detection classification model is obtained;Obtain the vibration signal of the gear reduction motor to be measured, obtain RGB power spectrum image according to vibration signal and input RGB power spectrum image into motor quality detection classification model, obtain the fault type of the motor to be measured.The application improves the precision and accuracy of detection classification by constructing improved ConvNeXt network model.

Description

Technical Field

[0001] This invention relates to the field of motor fault detection technology, and in particular to a method, apparatus, computer equipment, and storage medium for detecting faults in geared motors. Background Technology

[0002] With the development of modern science and technology and the widespread application of scientific and technological achievements in daily life, electric motors are now widely used in various aspects of industry, agriculture, and daily life. For example, induction motors are widely used in machine tools, cranes, and conveyor belts in industrial production; diesel generators are widely used in tractors and grain processing machinery in agricultural production; and geared DC motors are widely used in smart homes, intelligent robots, and automobiles. The normal and stable operation of an electric motor determines the overall system's performance and affects all aspects of production and daily life. If an electric motor malfunctions due to substandard quality or various factors during long-term operation, it will not only damage the motor itself but also affect the normal operation of the system. This can range from disrupting people's normal work and life to causing industrial production stoppages, impacting enterprise efficiency, and even leading to serious safety accidents, endangering personal safety, and causing huge economic losses. Therefore, it is necessary to conduct quality inspections and fault diagnosis on electric motors.

[0003] Currently, motor quality inspection and fault diagnosis are mainly divided into two categories. One category involves quality inspection of motors during mass production, before they enter the market, to prevent defective motors from entering and causing accidents. The other category involves real-time status monitoring and fault diagnosis of motors during production and operation, preventing potential malfunctions due to various factors during long-term operation. Gear-reduced DC motors use gearboxes to reduce high speeds to low speeds while increasing torque. Quality inspection is generally conducted during mass production, but real-time fault monitoring can also be performed during operation.

[0004] Motor fault diagnosis methods generally fall into three categories: model-based, signal-based, and knowledge-based. Model-based fault diagnosis methods suffer from complex modeling and difficulties in achieving accurate modeling. Signal-based fault diagnosis methods may experience performance degradation when operating under unknown or unbalanced conditions. Knowledge-based motor fault diagnosis methods are currently popular in research, with deep learning-based methods being a particularly promising area.

[0005] However, current deep learning-based motor fault diagnosis methods still have some problems. Most current motor diagnosis methods using convolutional neural networks employ shallow one- or two-dimensional input convolutional neural networks with relatively few layers, mostly no more than 10 layers. There is limited research on the performance of deeper network structures in motor fault diagnosis, and the classification accuracy of these network models has room for further improvement. On the other hand, the motor datasets used in deep learning-based motor diagnosis methods are generally small, making them prone to overfitting when training deep learning models. Summary of the Invention

[0006] To address the shortcomings of the prior art, this invention provides a method, apparatus, computer device, and storage medium for detecting faults in geared motors. This method can effectively improve classification accuracy through an improved ConvNeXt model. At the same time, data augmentation techniques can be used to process the dataset, effectively expanding the dataset, achieving model regularization, preventing overfitting, and giving the improved ConvNeXt model better generalization ability.

[0007] The first objective of this invention is to provide a method for detecting faults in a geared motor.

[0008] The second objective of this invention is to provide a device for detecting faults in a geared motor.

[0009] A third objective of this invention is to provide a computer device.

[0010] A fourth objective of this invention is to provide a storage medium.

[0011] The first objective of this invention can be achieved by adopting the following technical solution:

[0012] A method for detecting faults in a geared motor, the method comprising:

[0013] The vibration signal of the geared motor is acquired and processed; wherein the vibration signal includes vibration signals of the motor during normal operation and vibration signals of various types of faults.

[0014] The processed vibration signal is converted into an RGB power spectrum image using discrete short-time Fourier transform, and the RGB power spectrum image is used as the motor dataset.

[0015] An improved ConvNeXt network model is constructed based on the original ConvNeXt network model. The improved ConvNeXt network model uses a multi-head attention mechanism to build multiple attention-based structural blocks in the fifth part of the original ConvNeXt network model.

[0016] The improved ConvNeXt network model is trained using the motor dataset to obtain a motor quality detection classification model;

[0017] The vibration signal of the gear reducer motor under test is acquired, and an RGB power spectrum image is obtained based on the vibration signal. The RGB power spectrum image is then input into the motor quality detection classification model to obtain the fault type of the motor under test.

[0018] Furthermore, the attention-based structural block comprises three layers: the first layer is a convolutional layer with a kernel size of 1×1, the second layer is a multi-head attention mechanism layer, and the third layer is a convolutional layer with a kernel size of 1×1. Each layer includes a GULE activation function and layer normalization processing.

[0019] Furthermore, training the improved ConvNeXt network model using the motor dataset includes:

[0020] Before training the improved ConvNeXt network model, data augmentation processing is performed on the RGB power spectrum images in the motor dataset to make the motor dataset include more RGB power spectrum images.

[0021] The improved ConvNeXt network model is trained using a decaying learning rate method with the processed motor dataset.

[0022] Furthermore, the data augmentation processes include Trivial Augmentation, Mixup, and Random erasing.

[0023] Furthermore, cosine annealing is used to adjust the learning rate and optimize the loss value convergence curve.

[0024] Furthermore, the processing of the vibration signal includes:

[0025] The vibration signal is input into a signal amplifier to amplify the vibration signal; then the amplified signal is input into a digital-to-analog converter to convert the analog signal into a digital signal, and then the digital signal is saved in the form of one-dimensional data.

[0026] Furthermore, the motor can be any one of a DC motor, an induction motor, and a permanent magnet synchronous motor.

[0027] The second objective of this invention can be achieved by adopting the following technical solution:

[0028] A fault detection device for a geared motor, the device comprising:

[0029] The signal acquisition module is used to acquire the vibration signal of the gear reduction motor and process the vibration signal; wherein, the vibration signal includes the vibration signal of the motor during normal operation and the vibration signal of various types of faults.

[0030] The dataset acquisition module is used to convert the processed vibration signal into an RGB power spectrum image using discrete short-time Fourier transform, and then use the RGB power spectrum image as the motor dataset.

[0031] The model building module is used to build an improved ConvNeXt network model based on the original ConvNeXt network model. The improved ConvNeXt network model builds multiple attention mechanism-based structural blocks by using a multi-head attention mechanism on the fifth part of the original ConvNeXt network model.

[0032] The model training module is used to train the improved ConvNeXt network model using the motor dataset to obtain a motor quality detection classification model.

[0033] The fault type detection module is used to acquire the vibration signal of the gear reducer motor under test, obtain an RGB power spectrum image based on the vibration signal, and input the RGB power spectrum image into the motor quality detection classification model to obtain the fault type of the motor under test.

[0034] The third objective of this invention can be achieved by adopting the following technical solution:

[0035] A computer device includes a processor and a memory for storing a processor-executable program, wherein the processor implements the detection method described above when executing the program stored in the memory.

[0036] The fourth objective of this invention can be achieved by adopting the following technical solution:

[0037] A storage medium storing a program that, when executed by a processor, implements the above-described detection method.

[0038] The present invention has the following advantages over the prior art:

[0039] 1. The method provided by this invention applies an improved ConvNeXt deep learning classification network to gear reducer motor fault detection. This improved ConvNeXt network is based on a multi-head attention mechanism, combining the GULE activation function, layer normalization, convolutional layers, and residual structures to construct attention-based structural blocks, replacing some structural blocks of the original ConvNeXt network, thus introducing the attention mechanism into the pure convolutional ConvNeXt network. Compared to the original ConvNeXt network, the improved ConvNeXt network exhibits higher classification accuracy and precision in gear reducer motor fault detection tasks under the same number of training epochs.

[0040] 2. The method provided by this invention utilizes data augmentation techniques (Trivial Augment, Mixup, and Randomizerasing) to process the training set, thereby expanding the training set. This effectively prevents overfitting, achieves model regularization, and improves the model's classification accuracy and generalization. Simultaneously, a cosine annealing method is introduced during model training to adjust the learning rate. The learning rate of the model is dynamically adjusted according to the training rounds, optimizing the convergence curve of the loss value during model training, so that the model loss value can converge to the global optimal solution. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0042] Figure 1 This is a flowchart of the fault detection method for a gear reduction motor according to Embodiment 1 of the present invention.

[0043] Figure 2 This is a schematic diagram of the fault detection method for a gear reduction motor according to Embodiment 1 of the present invention.

[0044] Figure 3 This is a structural block diagram of the basic structural block in the ConvNeXt network of Embodiment 1 of the present invention.

[0045] Figure 4 This is a structural block diagram of the attention-based structural block in the improved ConvNeXt network of Embodiment 1 of the present invention.

[0046] Figure 5 This is a schematic diagram of the cosine annealing learning rate adjustment curve in Embodiment 1 of the present invention.

[0047] Figure 6This is a structural block diagram of the gear reduction motor fault detection device according to Embodiment 3 of the present invention.

[0048] Figure 7 This is a structural block diagram of the computer device according to Embodiment 4 of the present invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be understood that the specific embodiments described are merely used to explain this application and are not intended to limit this application.

[0050] Example 1:

[0051] Motors come in various types, such as DC motors, induction motors, and permanent magnet synchronous motors. Generally, fault diagnosis for most motors involves using vibration sensors to measure the motor's vibration signals and then determining the fault based on these signals. The geared motor fault detection method provided by this invention is applicable to fault detection for all of the aforementioned motor types. This embodiment uses a geared DC motor as an example to illustrate the method provided by this invention.

[0052] like Figure 1 , 2 As shown in the figure, this embodiment provides a method for detecting faults in a geared motor, including the following steps:

[0053] S101. Collect vibration signals from the gear-reduced DC motor and process the vibration signals.

[0054] Vibration sensors are used to collect vibration signals of a geared DC motor running unloaded at rated power. The vibration signals include vibration signals of the geared motor during normal operation as well as vibration signals of various types of faults.

[0055] The vibration signal is input to a signal amplifier and a digital-to-analog converter. After amplification, the vibration signal is converted from an analog signal to a digital signal and then input into a computer to be stored as one-dimensional data.

[0056] S102. The processed vibration signal is converted into an RGB power spectrum image using discrete short-time Fourier transform, and the RGB power spectrum image is used as the motor dataset.

[0057] The power spectral density of the motor vibration signal is obtained by using the discrete short-time Fourier transform on the one-dimensional data of the vibration signal obtained in step S101.

[0058]

[0059] Where P is the power spectral density defined on the sample (time) n and frequency w of signal x, STFT is an abbreviation for Short-Time Fourier Transform, π is pi, j is the imaginary unit, x(m) is the discrete sequence form of one-dimensional data of motor vibration signal, m is the sample, and y(nm) is the analysis window function, which is time-reversed and has an offset of n samples. The window function slides along the discrete sequence x(m) to form windows and calculates the discrete Fourier transform of each segment of the window data. The window function used is the Hamming window.

[0060]

[0061] Where w(n) is the Hamming window function, n is the number of samples, and M is the total number of samples.

[0062] After calculating the power spectrum matrix P, the values ​​in the power spectrum matrix will be mapped to the corresponding colors through normalization and scalar color mapping rules. Finally, the power spectrum matrix will be transformed into a three-channel RGB image.

[0063] S103. An improved ConvNeXt model is constructed based on the attention mechanism and the ConvNeXt network model.

[0064] Furthermore, step S103 includes:

[0065] (1) Original ConvNeXt network model.

[0066] The overall structure of the original ConvNeXt network model is shown in Table 1.

[0067] Table 1 ConvNeXt Network Structure

[0068]

[0069] The ConvNeXt network first uses a convolutional layer with a kernel size of 4×4 and a stride of 4 as the first part of the network to perform downsampling on the input image. The second, third, fourth, and fifth parts are built using the basic pure convolutional structure blocks of the ConvNeXt network, with the number of blocks in the second, third, fourth, and fifth parts in a ratio of 3:3:9:3. The basic structure blocks of the ConvNeXt network are shown below. Figure 3As shown, the basic structure of the ConvNeXt network consists of a first layer that is a depthwise separable convolutional layer with a kernel size of 7×7, and second and third layers that are convolutional layers with kernels of 1×1. Layer normalization is added between the first and second layers, and the GELU activation function is added between the second and third layers. The input and the output after passing through the three convolutional layers are added to form a residual structure, which serves as the output of the entire block structure. The ConvNeXt network adds separate downsampling layers between each of the second to fifth parts. These separate downsampling layers consist of layer normalization and convolutional layers with a kernel size of 2×2 and a stride of 2, achieving both normalization and downsampling functions.

[0070] (2) Improved ConvNeXt network model.

[0071] The improved ConvNeXt network has the same structure for the first four parts as the original ConvNeXt network. The improvement lies in the fifth part of the ConvNeXt network, which uses a multi-head attention mechanism to build an attention-based structural block, as shown in the figure. Figure 4 As shown, the fifth part includes multiple structure blocks based on the attention mechanism. In this embodiment, the number of structure blocks is 3.

[0072] The attention-based block structure consists of three layers: a first layer with a 1×1 kernel, a second layer with a multi-head attention mechanism, and a third layer with a 1×1 kernel. Each layer is separated by a Gule activation function and layer normalization. A residual structure is used to sum the input and the output after passing through the network layers, resulting in the output of the entire block structure. The formula for the attention mechanism is as follows:

[0073]

[0074] Where Q represents the query vector obtained after linear transformation of the feature map input and the q-linear layer of the attention mechanism layer, K represents the key vector obtained after linear transformation of the feature map input and the k-linear layer of the attention mechanism layer, and V represents the value vector obtained after linear operation of the feature map input and the v-linear layer of the attention mechanism layer. This is the scaling factor.

[0075] The formula for the multi-head attention mechanism is as follows:

[0076]

[0077] Wherein, parameter matrix

[0078] The improved ConvNeXt network is the same as the original ConvNeXt network: separate downsampling layers are added between each part from two to five. The separate downsampling layers consist of layer normalization processing and convolutional layers with a kernel size of 2×2 and a stride of 2, so as to realize the normalization and downsampling functions.

[0079] The output of the improved ConvNeXt network is the same as the original ConvNeXt network, consisting of pooling layers, layer normalization processing, flattening processing, and linear layers. The overall structure of the improved ConvNeXt network is shown in Table 2.

[0080] Table 2 Improved ConvNeXt Network Structure

[0081]

[0082]

[0083] S104. Based on the decaying learning rate method, the improved ConvNeXt model is trained using the motor dataset to obtain the motor quality detection classification model.

[0084] The power spectrum images of motors of the same type with different qualities or faults obtained in step S102 are divided into training set and test set. Each set has the same number of power spectrum images of the same quality or fault type, but the number of images of each specific type can be different.

[0085] Data augmentation is performed on the training set to include more power spectrum images. The model is trained using a decaying learning rate method, and the model weight parameters are updated during training. After training, a motor quality detection classification model is obtained.

[0086] During training, data augmentation techniques such as Trivial Augmentation, Mixup, and Random Erasing are introduced to process the training set, which is equivalent to expanding the training set. This can effectively prevent overfitting, achieve model regularization, and improve the model's classification accuracy. At the same time, cosine annealing is introduced to adjust the decay of the learning rate to optimize the convergence curve of the loss value.

[0087] Trivial Augment includes various operations such as scaling, cropping, rotation, and contrast adjustment, and the algorithms are shown in Table 3.

[0088] Table 3 Trivial Augment

[0089]

[0090] Where 'a' represents a data augmentation operation, A represents a set of data augmentation operations such as scaling, cropping, rotation, and contrast adjustment, and m represents the intensity value.

[0091] Random Erasing randomly erases a rectangular region from the power spectrum image using random values ​​without changing the image label. This reduces the risk of overfitting and makes the model robust to occlusion. The Random Erasing algorithm is shown in Table 4.

[0092] Table 4 Random Erasing

[0093]

[0094]

[0095] Mixup performs a random linear combination of two types of images, generating labels for the linear combination to expand the dataset. This allows the model to learn a linear transition between different categories, enhancing its generalization ability. The Mixup formula is as follows:

[0096]

[0097] Where x i x j For the image, y i y j The label corresponding to the image. For the generated images, For mixed labels, λ∈[0,1] is the probability value.

[0098] A cosine annealing learning rate decay method is introduced, continuously adjusting the learning rate over multiple epochs of training to optimize the loss convergence curve during model training. This allows the model to learn with a larger learning rate in the early stages of training, accelerating convergence, and then adjust with a smaller learning rate in the later stages, allowing the model to converge near its optimal value. The formula for cosine annealing learning rate adjustment is as follows:

[0099]

[0100] Where η t Let η be the current learning rate. min η is the minimum learning rate. max T represents the maximum learning rate. cur For the current epoch, T max For the maximum epoch. The cosine annealing adjusts the learning rate curve as follows: Figure 5 As shown.

[0101] Training ends when the improved ConvNeXt network model achieves the required high classification accuracy on the test set, resulting in the motor quality detection classification model.

[0102] S105. Obtain the vibration signal of the DC motor under test, convert the vibration signal into an RGB power spectrum image, input the RGB power spectrum image into the motor quality detection classification model, and obtain the fault type of the motor under test.

[0103] The three-channel RGB power spectrum of the motor under test is obtained by a feature extraction method based on short-time Fourier transform. The three-channel RGB power spectrum is then input into the motor quality detection classification model. The motor quality detection classification model is used to judge the specific quality or fault of the motor under test, thus completing the quality or fault detection of the gear reduction DC motor under test.

[0104] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware, and the corresponding program can be stored in a computer-readable storage medium.

[0105] It should be noted that although the method operations of the above embodiments are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the order of execution of the described steps may be changed. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0106] Example 2:

[0107] For the JL-12FN20-37 geared DC motor, a MOTECH LPS-305 digitally controlled linear DC regulated power supply is used to provide a stable voltage to the motor, enabling it to operate normally under its rated operating voltage. The CT1050LC vibration sensor uses a sampling frequency of 12kHz to collect the vibration signal of the normally operating motor. This vibration signal is input to the CT5201 single-channel constant current adapter and the MCC1608G multi-function data acquisition unit. After amplification and conversion from analog to digital signals, the vibration signal is input into the computer and stored as one-dimensional discrete data.

[0108] Data was collected from 100 normal motors and 100 faulty motors. For each micro geared DC motor, 20 sets of 5-second vibration time-domain signals were continuously recorded. These 5-second signals were then evenly divided into 5 sets of 1-second signals and labeled: label 0 for normal motors and label 1 for motors with abnormal noise. The one-dimensional discrete data of the 1-second motor vibration signals were then processed using a discrete short-time Fourier transform (DSFT) method to obtain a 92×129 power spectrum matrix. Color mapping of this matrix yielded a 92×129 three-channel RGB image. This resulted in a dataset of 20,000 power spectrum images of geared DC motor samples with roughly equal numbers of normal and abnormal noise. This dataset was then divided into training, validation, and test sets in a 7:2:1 ratio.

[0109] An improved ConvNeXt network was built using PyTorch in a Python environment, incorporating convolutional layers, attention mechanisms, activation functions, and linear layers. The training set was read and augmented with Trivial Augmentation, Random Erasing, and Mixup. The improved ConvNeXt network was trained for 300 epochs on a server using four 2080ti GPUs. Cosine annealing was introduced during training to adjust the learning rate, causing it to decay. A validation set was used to verify the model's classification accuracy. The final improved ConvNeXt model achieved a classification accuracy of 99.501% on the test set.

[0110] The vibration signal of the JL-12FN20-37 geared DC motor under test, measured and acquired under normal operating voltage, was obtained using a MOTECH LPS-305 CNC linear DC regulated power supply, a CT5201 single-channel constant current adapter, an MCC1608G multi-function data acquisition unit, and a CT1050LC vibration sensor. This signal was then transmitted to the JL-12FN20-37 geared DC motor quality inspection program on a computer, where a feature extraction method based on short-time Fourier transform was used to obtain the power spectrum. This power spectrum was then input into an improved ConvNeXt model trained and deployed in the computer program. The model outputs whether the motor under test is a normal motor or a motor producing abnormal noise. This completes the quality inspection task of the JL-12FN20-37 geared DC motor in the factory production environment, preventing defective motors from entering the market and causing accidents.

[0111] Example 3:

[0112] like Figure 6As shown, this embodiment provides a fault detection device for a geared motor. The device includes a signal acquisition module 601, a dataset acquisition module 602, a model building module 603, a model training module 604, and a fault type detection module 605, wherein:

[0113] The signal acquisition module 601 is used to acquire the vibration signal of the gear reduction motor and process the vibration signal; wherein, the vibration signal includes the vibration signal of the motor operating normally and the vibration signal of various types of faults.

[0114] The dataset acquisition module 602 is used to convert the processed vibration signal into an RGB power spectrum image using discrete short-time Fourier transform, and use the RGB power spectrum image as the motor dataset.

[0115] The model building module 603 is used to build an improved ConvNeXt network model based on the ConvNeXt network model. The improved ConvNeXt network model builds multiple attention mechanism-based structural blocks by using a multi-head attention mechanism on the fifth part of the original ConvNeXt network model.

[0116] The model training module 604 is used to train the improved ConvNeXt network model using the motor dataset to obtain a motor quality detection classification model.

[0117] The fault type detection module 605 is used to acquire the vibration signal of the gear reducer motor under test, obtain an RGB power spectrum image based on the vibration signal, and input the RGB power spectrum image into the motor quality detection classification model to obtain the fault type of the motor under test.

[0118] The specific implementation of each module in this embodiment can be found in Embodiment 1 above, and will not be repeated here. It should be noted that the device provided in this embodiment is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure can be divided into different functional modules to complete all or part of the functions described above.

[0119] Example 4:

[0120] This embodiment provides a computer device, which can be a computer, such as... Figure 7As shown, the processor 702, memory, input device 703, display 704, and network interface 705 are connected via system bus 701. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium 706 and internal memory 707. The non-volatile storage medium 706 stores the operating system, computer programs, and database. The internal memory 707 provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. When the processor 702 executes the computer program stored in the memory, it implements the detection method of Embodiment 1 described above, as follows:

[0121] The vibration signal of the geared motor is acquired and processed; wherein the vibration signal includes vibration signals of the motor during normal operation and vibration signals of various types of faults.

[0122] The processed vibration signal is converted into an RGB power spectrum image using discrete short-time Fourier transform, and the RGB power spectrum image is used as the motor dataset.

[0123] An improved ConvNeXt network model is constructed based on the original ConvNeXt network model. The improved ConvNeXt network model uses a multi-head attention mechanism to build multiple attention-based structural blocks in the fifth part of the original ConvNeXt network model.

[0124] The improved ConvNeXt network model is trained using the motor dataset to obtain a motor quality detection classification model;

[0125] The vibration signal of the gear reducer motor under test is acquired, and an RGB power spectrum image is obtained based on the vibration signal. The RGB power spectrum image is then input into the motor quality detection classification model to obtain the fault type of the motor under test.

[0126] Example 5:

[0127] This embodiment provides a storage medium, which is a computer-readable storage medium, storing a computer program. When the computer program is executed by a processor, it implements the detection method of Embodiment 1 above, as follows:

[0128] The vibration signal of the geared motor is acquired and processed; wherein the vibration signal includes vibration signals of the motor during normal operation and vibration signals of various types of faults.

[0129] The processed vibration signal is converted into an RGB power spectrum image using discrete short-time Fourier transform, and the RGB power spectrum image is used as the motor dataset.

[0130] An improved ConvNeXt network model is constructed based on the original ConvNeXt network model. The improved ConvNeXt network model uses a multi-head attention mechanism to build multiple attention-based structural blocks in the fifth part of the original ConvNeXt network model.

[0131] The improved ConvNeXt network model is trained using the motor dataset to obtain a motor quality detection classification model;

[0132] The vibration signal of the gear reducer motor under test is acquired, and an RGB power spectrum image is obtained based on the vibration signal. The RGB power spectrum image is then input into the motor quality detection classification model to obtain the fault type of the motor under test.

[0133] It should be noted that the computer-readable storage medium in this embodiment can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0134] In summary, the gear reducer motor fault detection method provided by this invention includes: collecting vibration signals of the gear reducer motor under no-load operation at rated power using a vibration sensor. The collected vibration signal data of various types of faults in the gear reducer motor are used for feature extraction using short-time Fourier transform, converting the one-dimensional vibration signal into a three-channel RGB image, and dividing it into training, validation, and test sets for model training of an improved ConvNeXt network. A deep learning network model block is proposed based on multi-head attention mechanism, convolutional layers, GULE activation function, layer normalization, and residual structure. This model block replaces part of the original ConvNeXt model block, which has excellent classification capabilities, resulting in an improved ConvNeXt model. During model training, data augmentation techniques, including Trivial Augmentation, Mixup, and Random Erasing, are incorporated to expand the dataset and act as a regularization mechanism. Simultaneously, a decaying learning rate method is used to optimize the loss value convergence curve during model training. The vibration feature map of the geared motor under test, extracted using a short-time Fourier transform-based feature extraction method, is input into the trained improved ConvNeXt model. The model can then output the fault type of the geared motor and whether its quality is up to standard. This method uses data augmentation techniques to regularize the model, effectively preventing overfitting. Furthermore, compared to traditional motor fault detection methods based on modeling, time-frequency analysis, shallow convolutional neural networks, or ordinary deep networks, this method achieves better recognition accuracy and classification precision due to the use of the improved ConvNeXt deep network.

[0135] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, shall fall within the scope of protection of the present invention.

Claims

1. A method for detecting faults in a geared motor, characterized in that, The method includes: The vibration signal of the geared motor is acquired and processed; wherein the vibration signal includes vibration signals of the motor during normal operation and vibration signals of various types of faults. The processed vibration signal is converted into an RGB power spectrum image using discrete short-time Fourier transform, and the RGB power spectrum image is used as the motor dataset. An improved ConvNeXt network model is constructed based on the ConvNeXt network model. This improved model replaces the fifth part of the ConvNeXt network model with multiple attention-based structural blocks built using a multi-head attention mechanism. Each attention-based structural block consists of three layers: a first layer is a 1×1 convolutional layer, a second layer is a multi-head attention mechanism layer, and a third layer is also a 1×1 convolutional layer. Each layer includes a GULE activation function and layer normalization. The output of the third layer is added to the input of the attention-based structural block to obtain its output. The improved ConvNeXt network model is trained using the motor dataset to obtain a motor quality detection classification model; The vibration signal of the gear reducer motor under test is acquired, and an RGB power spectrum image is obtained based on the vibration signal. The RGB power spectrum image is then input into the motor quality detection classification model to obtain the fault type of the motor under test.

2. The detection method according to claim 1, characterized in that, The step of training the improved ConvNeXt network model using the motor dataset includes: Before training the improved ConvNeXt network model, data augmentation processing is performed on the RGB power spectrum images in the motor dataset to make the motor dataset include more RGB power spectrum images. The improved ConvNeXt network model is trained using a decaying learning rate method with the processed motor dataset.

3. The detection method according to claim 2, characterized in that, The data augmentation processes include Trivial Augment, Mixup, and Random erasing.

4. The detection method according to claim 2, characterized in that, Cosine annealing is used to adjust the learning rate and optimize the loss value convergence curve.

5. The detection method according to any one of claims 1 to 4, characterized in that, The processing of the vibration signal includes: The vibration signal is input into a signal amplifier to amplify the vibration signal; then the amplified signal is input into a digital-to-analog converter to convert the analog signal into a digital signal, and then the digital signal is saved in the form of one-dimensional data.

6. The detection method according to any one of claims 1 to 4, characterized in that, The motor can be any one of a DC motor, an induction motor, or a permanent magnet synchronous motor.

7. A fault detection device for a geared motor, characterized in that, The device includes: The signal acquisition module is used to acquire the vibration signal of the gear reduction motor and process the vibration signal; wherein, the vibration signal includes the vibration signal of the motor during normal operation and the vibration signal of various types of faults; The dataset acquisition module is used to convert the processed vibration signal into an RGB power spectrum image using discrete short-time Fourier transform, and then use the RGB power spectrum image as the motor dataset. The model building module is used to construct an improved ConvNeXt network model based on the existing ConvNeXt network model. This improved ConvNeXt network model replaces the fifth part of the existing ConvNeXt network model with multiple attention-based structural blocks built using a multi-head attention mechanism. Each attention-based structural block consists of three layers: a first layer with a 1×1 kernel, a second layer with a multi-head attention mechanism, and a third layer with a 1×1 kernel. Each layer includes a GULE activation function and layer normalization. The output of the third layer is added to the input of the attention-based structural block to obtain its output. The model training module is used to train the improved ConvNeXt network model using the motor dataset to obtain a motor quality detection classification model. The fault type detection module is used to acquire the vibration signal of the gear reducer motor under test, obtain an RGB power spectrum image based on the vibration signal, and input the RGB power spectrum image into the motor quality detection classification model to obtain the fault type of the motor under test.

8. A computer device comprising a processor and a memory for storing a processor-executable program, characterized in that, When the processor executes the program stored in the memory, it implements the detection method according to any one of claims 1-6.

9. A storage medium storing a program, characterized in that, When the program is executed by the processor, it implements the detection method according to any one of claims 1-6.

Images