Fault detection model training method and device, fault detection method and device and electronic equipment

Abstract

The invention relates to the technical field of motors, and discloses a fault detection model training method, a fault detection method, a fault detection device and electronic equipment. The fault detection model training method comprises the steps of obtaining a training data set; the training data set is a data set comprising current data when the motor is in a normal state, different eccentric types and different eccentric degrees; performing variational mode decomposition on each current data in the training data set to obtain a plurality of mode components corresponding to each current data; determining a target modal component corresponding to each piece of current data from the plurality of modal components; and converting the target modal component into an image, and inputting the image into a pre-constructed fault detection model for model training to obtain a trained fault detection model. According to the invention, the accuracy of motor eccentric fault detection can be further improved.

Description

Technical Field

[0001] The present application relates to the technical field of motors, and for example, relates to a method for training a fault detection model, a fault detection method, an apparatus, and an electronic device. Background Art

[0002] Motors are widely used in all aspects of production and life, such as in various equipment and machines like machine tools, printing machines, etc., and in household appliances such as electric fans, air conditioners, etc. During the installation process of motors, due to problems such as inaccurate positioning or loose mechanical structures, the rotor (or rotating) center of the motor may be misaligned with the stator center, that is, an eccentricity fault. The eccentricity fault will cause uneven torque during the operation of the motor, resulting in increased vibration and noise, and will also generate additional mechanical losses, reducing the efficiency of the motor and even posing a safety hazard. Therefore, diagnosing the eccentricity fault of the motor has important industrial value.

[0003] The related art discloses a method for inspecting the eccentricity fault diagnosis of an asynchronous motor, including the following steps: obtaining historical fault data of the eccentricity fault of the asynchronous motor, and extracting the eccentricity fault characteristics from the historical fault data; diagnosing whether the asynchronous motor has an eccentricity fault according to the eccentricity fault characteristics, and if an eccentricity fault occurs, outputting primary fault data; obtaining the current data of the asynchronous motor and the corresponding detection environment, and performing quality screening on the current data according to the detection environment to form current spectrum data; according to the current spectrum data, extracting the frequency and amplitude in the current spectrum data, and inspecting whether the asynchronous motor has an eccentricity fault according to the frequency and amplitude and outputting an inspection result; if the inspection result is that the asynchronous motor has an eccentricity fault, obtaining the monitoring image during the operation of the asynchronous motor, and extracting the vibration data of the asynchronous motor by combining the attention mechanism; obtaining the vibration modes of different components of the asynchronous motor, and combining the vibration data to confirm the eccentricity position and the severity of the eccentricity to obtain a fault result.

[0004] Although the related art improves the accuracy of the inspection of the eccentricity fault diagnosis of the motor, it only relies on the frequency and amplitude information in the current spectrum data for the eccentricity fault inspection, and the degree of accuracy improvement is limited.

[0005] It should be noted that the information disclosed in the above background art section is only used to enhance the understanding of the background of the present application, and therefore may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] To have a basic understanding of some aspects of the disclosed embodiments, a simple summary is given below. This summary is not a general review, nor is it intended to identify key / important constituent elements or delineate the protection scope of these embodiments, but rather serves as a preface to the subsequent detailed description.

[0007] Embodiments of the present disclosure provide a training method, a fault detection method, an apparatus, and an electronic device for a fault detection model, which can further improve the accuracy of motor eccentricity fault detection.

[0008] In some embodiments, a training method for a fault detection model is provided, including: obtaining a training data set; the training data set is a data set including current data of a motor in a normal state, different eccentricity types, and different eccentricity degrees; performing variational mode decomposition on each current data in the training data set to obtain a plurality of modal components corresponding to each current data; determining a target modal component corresponding to each current data from the plurality of modal components; converting the target modal component into an image and inputting it into a pre-constructed fault detection model for model training to obtain a trained fault detection model.

[0009] Optionally, determining a target modal component corresponding to each current data from the plurality of modal components includes: calculating the correlation coefficient between the current data and each modal component corresponding to the current data respectively; determining the target modal component corresponding to the current data from the plurality of modal components corresponding to each current data according to the correlation coefficient.

[0010] Optionally, determining the target modal component corresponding to the current data from the plurality of modal components corresponding to each current data according to the correlation coefficient includes: using the modal component with the largest correlation coefficient among the plurality of modal components corresponding to each current data as the target modal component corresponding to the current data; or using the modal components with a correlation coefficient greater than or equal to a correlation coefficient threshold among the plurality of modal components corresponding to each current data as the target modal component corresponding to the current data.

[0011] Optionally, the pre-constructed fault detection model includes a two-dimensional convolutional neural network model; converting the target modal component into an image and inputting it into the pre-constructed fault detection model for model training to obtain a trained fault detection model includes: converting the target modal component into a grayscale image to obtain a training image; inputting the training image into the two-dimensional convolutional neural network model for model training to obtain a trained fault detection model.

[0012] Optionally, input the training images into a two-dimensional convolutional neural network model for model training to obtain a trained fault detection model, including: dividing the training images to obtain a training image set and a test image set; wherein, both the training image set and the test image set include training images corresponding to current data in the normal state, different eccentricity types, and different eccentricity degrees; input the training images in the training image set into the two-dimensional convolutional neural network model for model training; input the training images in the test image set into the trained two-dimensional convolutional neural network model for testing, and obtain the test accuracy; when the test accuracy meets the preset conditions, obtain the trained fault detection model.

[0013] Optionally, inputting the training images into a two-dimensional convolutional neural network model for model training includes: initializing the model parameters of the two-dimensional convolutional neural network model; inputting the training images into the two-dimensional convolutional neural network model for forward propagation to obtain a prediction result; comparing the prediction result with the true result corresponding to the training images, and calculating the loss function; calculating the gradient value of the loss function with respect to the model parameters through backpropagation, and updating the model parameters of the two-dimensional convolutional neural network model according to the gradient value.

[0014] In some embodiments, a fault detection method is provided, including: acquiring real-time current data of a motor to be detected; performing variational mode decomposition on the real-time current data to obtain a plurality of modal components; determining a target modal component corresponding to the real-time current data from the plurality of modal components; converting the target modal component into an image and inputting it into a fault detection model to obtain a detection result; wherein, the fault detection model is trained by using the training method of the fault detection model described in the above embodiments.

[0015] Optionally, the fault detection method further includes: acquiring real-time operating parameters of the motor to be detected; inputting the real-time operating parameters into a trained reconstruction model for encoding and decoding, and obtaining the real-time reconstruction error of the real-time operating parameters; the reconstruction model is trained by using the operating parameters of the motor in the normal state; verifying the detection result according to the real-time reconstruction error; when the verification is passed, sending the detection result to a target terminal device for display; when the verification fails, updating the fault detection model according to the real-time reconstruction error.

[0016] In some embodiments, a training device for a fault detection model is provided, including: a first acquisition module configured to obtain a training data set; the training data set is a data set including current data of the motor in a normal state, different eccentricity types, and different eccentricity degrees; a first decomposition module configured to perform variational mode decomposition on each current data in the training data set to obtain a plurality of modal components corresponding to each current data; a first determination module configured to determine a target modal component corresponding to each current data from the plurality of modal components; a model training module configured to convert the target modal component into an image and input it into a pre-constructed fault detection model for model training to obtain a trained fault detection model.

[0017] In some embodiments, a fault detection device is provided, including: a second acquisition module configured to obtain real-time current data of a motor to be detected; a second decomposition module configured to perform variational mode decomposition on the real-time current data to obtain a plurality of modal components; a second determination module configured to determine a target modal component corresponding to the real-time current data from the plurality of modal components; a fault detection module configured to convert the target modal component into an image and input it into the fault detection model to obtain a detection result; wherein, the fault detection model is trained by using the training method of the fault detection model described in the above embodiments.

[0018] In some embodiments, an electronic device is provided, including: a processor; and a memory storing program instructions, the processor is configured to execute the training method of the fault detection model described in the above embodiments, or the fault detection method described in the above embodiments.

[0019] The training method, fault detection method, device, and electronic device for the fault detection model provided by the embodiments of the present disclosure can achieve the following technical effects:

[0020] In the embodiments of the present disclosure, the current data can be decomposed into a plurality of modal components through variational mode decomposition, and each modal component represents different features in the current data, realizing a more refined capture of the feature information in the current data, especially the subtle changes related to the eccentricity fault. Although feature extraction is also performed in the related art, the detailed information in the current signal cannot be deeply analyzed. Compared with the related art, the embodiments of the present disclosure can more deeply and effectively extract the fault features in the current data of the motor through variational mode decomposition and image processing, thereby further improving the accuracy of fault detection.

[0021] The above general description and the following description are only exemplary and explanatory, and are not used to limit this application. Description of the Drawings

[0022] One or more embodiments are exemplarily illustrated by corresponding drawings. These exemplary illustrations and the drawings do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are shown as similar elements. The drawings do not constitute a scale limitation, and wherein:

[0023] Figure 1 is a schematic diagram of a method for training a fault detection model provided by an embodiment of the present disclosure;

[0024] Figure 2 is a schematic diagram of a method for training a fault detection model provided by another embodiment of the present disclosure;

[0025] Figure 3 is a schematic diagram of a fault detection method provided by an embodiment of the present disclosure;

[0026] Figure 4 is a schematic diagram of a device for training a fault detection model provided by an embodiment of the present disclosure;

[0027] Figure 5 is a schematic diagram of a fault detection device provided by an embodiment of the present disclosure;

[0028] Figure 6 is a schematic diagram of an electronic device provided by an embodiment of the present disclosure. Detailed Description of the Embodiment

[0029] In order to be able to understand the features and technical content of the embodiments of the present disclosure in more detail, the implementation of the embodiments of the present disclosure will be described in detail below with reference to the drawings. The attached drawings are for reference and illustration only and are not used to limit the embodiments of the present disclosure. In the following technical description, for the sake of explanation, numerous details are provided to give a thorough understanding of the disclosed embodiments. However, one or more embodiments may still be implemented without these details. In other instances, well-known structures and devices may be shown in a simplified manner to simplify the drawings.

[0030] In the embodiments of the present disclosure, terms such as "first" and "second" in the specification, claims, and the above-mentioned drawings are used to distinguish similar objects and do not necessarily have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so as to implement the embodiments of the present disclosure described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion.

[0031] Unless otherwise specified, the term "plurality" means two or more.

[0032] In the embodiments of the present disclosure, the character " / " indicates that the objects before and after are in an "or" relationship. For example, A / B means: A or B.

[0033] The term "and / or" describes the relationship between objects and indicates that there can be three relationships. For example, A and / or B means: A, B, and A and B, these three relationships.

[0034] The term "corresponding" can refer to a relationship of association or a binding relationship. That A corresponds to B means that there is a relationship of association or a binding relationship between A and B.

[0035] It should be noted that, without conflict, the embodiments in the present disclosure and the features in the embodiments can be combined with each other.

[0036] Combined with Figure 6 As shown, an electronic device 60 is provided in an embodiment of the present disclosure. The electronic device 60 includes a processor 600 and a memory 601 storing program instructions.

[0037] Among them, the processor 600 can obtain a training data set; the training data set is a data set including current data of the motor in normal states, different eccentric types, and different eccentric degrees; it can perform variational mode decomposition on each current data in the training data set to obtain multiple modal components corresponding to each current data; it can determine the target modal component corresponding to each current data from the multiple modal components; it can convert the target modal component into an image and input it into a pre-constructed fault detection model for model training to obtain a trained fault detection model.

[0038] Combined with Figure 6 the electronic device 60 shown, an embodiment of the present disclosure provides a method for training a fault detection model, as Figure 1 shown, the training method includes:

[0039] S101, the processor obtains a training data set.

[0040] The training data set is a data set including current data of the motor in normal states, different eccentric types, and different eccentric degrees.

[0041] In some embodiments, the training data set is obtained in the following manner: the current data of the motor in normal states, different eccentric types, and different eccentric degrees at a preset rotational speed of the motor are collected through a motor driver according to a preset sampling frequency; the collected current data is marked to generate a training data set.

[0042] A motor driver is an electronic device used to control the speed, torque, and direction of a motor, also known as a motor controller. In this embodiment, the motor driver collects the current data of the motor to generate a training data set, without the need for additional modification of the motor or installation of sensors, improving the convenience and accuracy of current data collection, reducing the data collection cost, and further improving the convenience, cost of training the fault detection model, and the accuracy of the fault detection model.

[0043] The preset speed refers to the operating speed preset for the motor. In this embodiment, the number of preset speeds can be one or more. When the number of preset speeds is multiple, the speed values of the multiple preset speeds are all different. The specific values of the preset speeds are not limited in the embodiments of the present disclosure.

[0044] The preset sampling frequency refers to the frequency of collecting data preset in advance. By collecting the current data of the motor at the preset speed, in the normal state, different eccentricity types, and different eccentricity degrees respectively according to the preset sampling frequency, it is ensured that the training data set can comprehensively cover various possible states of the motor, improving the diversity of the training data set. After the training data set is used for training the fault detection model, the generalization ability of the fault detection model is improved, the recognition ability of the fault detection model for different types and degrees of faults is enhanced, and the detection accuracy of the fault detection model is improved.

[0045] At the same time as or after collecting the current data, each current data is marked to indicate the operating speed and the represented state (such as the normal state, a certain eccentricity type, or eccentricity degree) of the current data. The marked current data will form a training data set to realize the production of the training data set for subsequent training of the fault detection model.

[0046] In a practical application, the training data set is obtained in the following manner: Set the preset sampling frequency to 4000 Hz (Hertz), the preset rotational speed to 1000 r / min (revolutions per minute), and the eccentricity types include parallel eccentricity and angular eccentricity. Among them, the parallel eccentricity amounts (i.e., the eccentricity degrees) of parallel eccentricity are 0.5 mm (millimeters), 1 mm, and 1.5 mm respectively, and the eccentricity angle values (i.e., the eccentricity degrees) of angular eccentricity are 0.8°, 1.2°, and 1.5° respectively; When the motor speed is 1000 r / min, the current data in the normal state (i.e., no eccentricity), parallel eccentricity amount of 0.5 mm, parallel eccentricity amount of 1 mm, parallel eccentricity amount of 1.5 mm, eccentricity angle of 0.8°, eccentricity angle of 1.2°, and eccentricity angle of 1.5° are collected through the motor driver, obtaining a total of 7 groups of current data; Among them, the sampling duration is 10 s (seconds), and the sampling frequency is 4000 Hz, so each group of current data has a total of 40000 data points; Each group of collected current data is marked, including but not limited to marking the operating speed (1000 r / min), status (normal state, parallel eccentricity or angular eccentricity), eccentricity degree (no eccentricity, 0.5 mm, 1 mm, 1.5 mm, 0.8°, 1.2°, or 1.5°), etc. The marked current data is formed into a training data set to achieve the acquisition of the training data set.

[0047] S102. The processor performs variational mode decomposition on each current data in the training data set to obtain multiple modal components corresponding to each current data.

[0048] Variational Mode Decomposition (VMD) is an adaptive and non-recursive method for signal processing, aiming to decompose complex multi-component signals into a series of intrinsic mode functions (IMFs) with sparse characteristics. These IMF components are compact and bandwidth-limited in the frequency domain, and the central frequency and bandwidth of each IMF component will adaptively change during the decomposition process. The core idea of VMD is to decompose the signal into multiple IMF components such that the sum of the estimated bandwidths of each IMF component is minimized, that is, to find a solution through iterative optimization to minimize the bandwidth of each IMF component.

[0049] S103. The processor determines the target modal component corresponding to each current data from the multiple modal components.

[0050] In this step, the target modal component refers to the modal component among the multiple modal components that is related to the characteristics of the motor operating state. The target modal component can accurately reflect the characteristics of the motor operating state.

[0051] S104. The processor converts the target modal component into an image and inputs it into a pre-constructed fault detection model for model training to obtain a trained fault detection model.

[0052] By using the fault detection model training method provided in the embodiments of the present disclosure, the current data can be decomposed into multiple modal components through variational mode decomposition. Each modal component represents different features in the current data, achieving a more refined capture of the feature information in the current data, especially the subtle changes related to the eccentricity fault. Although feature extraction is also carried out in the related art, the detailed information in the current signal cannot be deeply analyzed. Compared with the related art, the embodiments of the present disclosure can more deeply and effectively extract the fault features in the current data of the motor through variational mode decomposition and image processing, thereby further improving the accuracy of fault detection.

[0053] In a specific application, the variational mode decomposition of the current data includes:

[0054] Initialization: Set the number K of modal components (i.e., IMF components). The number K of modal components represents the expected number of modal components to be decomposed; initialize the central frequency ω of each modal component i n , that is, set the central frequency to ω i 1 , n represents the number of iterations, ω i n represents the central frequency corresponding to the i-th modal component at the n-th iteration. The value of i is 1, 2, 3, ……, K; set the convergence conditions, such as the upper limit of the number of iterations, the bandwidth change threshold, etc.

[0055] For each iteration n (starting from 1): For the current estimate u i n-1 (t) of each modal component, use the Hilbert transform to construct its analytic signal; u i n-1 (t) represents the current estimate of the i-th modal component at the n-th iteration, which is 0 or random noise at the first iteration (n = 1); estimate the bandwidth of each modal component and calculate the total bandwidth of all modal components by demodulation (i.e., multiplying the analytic signal by ) and smoothing in a preset form (such as Gaussian smoothing or Wiener filtering); where, in each iteration process, update the current estimate of each modal component and the corresponding central frequency ω i n so as to minimize the total bandwidth of all modal components.

[0056] One can transform the solution of a variational problem into the minimization problem of an augmented Lagrangian function: to update the current estimate \(u\) i n-1 (t) and the corresponding central frequency \(\omega\) i n of each modal component, minimizing the total bandwidth of all modal components. Here, \(u(t)\) refers to the original signal (current data in this embodiment), and \(u\) i (t) refers to the current estimate of the \(i\)-th modal component obtained by decomposition, and \(\omega\) i refers to the central frequency corresponding to the \(i\)-th modal component. \(\alpha\) is a regularization parameter that balances the reconstruction accuracy and bandwidth, and \(\delta(t)\) is the Dirac function. By optimizing the first term of the formula, that is, to minimize the bandwidth of each modal component \(u\) i (t); by optimizing the second term of the formula, that is, to ensure that the sum of all modal components can reconstruct the original signal \(u(t)\).

[0057] Convergence check and result output: Check whether the convergence condition is satisfied, such as the change in the central frequency of all modal components is less than the bandwidth change threshold, or the upper limit of the number of iterations is reached; if the convergence condition is satisfied, output all modal components \(u\) i (t) (\(i = 1, 2, 3, \cdots, K\)) and the central frequency \(\omega\) i corresponding to each modal component \(u\) i ; if the convergence condition is not satisfied, continue the iteration process.

[0058] Optionally, from multiple modal components, determine the target modal component corresponding to each current data, including: calculating the correlation coefficient between the current data and each modal component corresponding to the current data respectively; according to the correlation coefficient, determine the target modal component corresponding to the current data from the multiple modal components corresponding to each current data.

[0059] The correlation coefficient method is a statistical method used to measure the strength and direction of the linear relationship between two variables. The correlation coefficient can be represented by \(\rho\), and the value range of the correlation coefficient is \([-1, 1]\). When \(\rho = 1\), it means that the two variables are completely positively correlated; when \(\rho = -1\), it means that the two variables are completely negatively correlated; when \(\rho = 0\), it means that there is no linear relationship between the two variables.

[0060] In this embodiment, by calculating the correlation coefficient between the current data and the modal components obtained by decomposition, the similarity or correlation between the current data and each modal component is measured, so as to select the modal components related to the current data for subsequent analysis or processing.

[0061] In this embodiment, the correlation coefficient between the current data and each modal component corresponding to the current data is calculated as follows: Among them, ρ(i) represents the correlation coefficient between the i-th modal component and the original signal u(t) (i.e., the current data); L represents the signal length of the original signal u(t), that is, the number of data points obtained after sampling the current data. For example, if the sampling frequency is 4000 Hz and the sampling duration is 10 s, then L = 4000×10 = 40000, a total of 40000 data points, and the signal length L is 40000; x(j) represents the instantaneous value of the data point obtained by sampling the original signal u(t) at the j-th sampling point; F MI,i (j) represents the corresponding value of the data point obtained by sampling at the j-th sampling point in the i-th modal component (obtained by variational mode decomposition), which is used to represent the local feature components with different frequencies and bandwidths in the original signal u(t); denotes the sum of the data points obtained by sampling all sampling points (from j = 1 to j = L); the value of j is 1, 2, 3... L.

[0062] In this embodiment, is the sum of the point-by-point products of the instantaneous value of the data point corresponding to each sampling point of the original signal u(t) and the value in the time domain corresponding to this data point in the i-th modal component, which reflects their covariance. is the product of the square root of the sum of the squares of the instantaneous value of the data point corresponding to each sampling point of the original signal u(t) and the value in the time domain corresponding to this data point in the i-th modal component, which is used to standardize the output result to ensure that the output correlation coefficient is within the range [-1, 1]. The formula in this embodiment is used to screen the modal component most relevant to the original signal u(t) (i.e., the target modal component) to improve the accuracy of the subsequent fault detection model.

[0063] In this embodiment, by calculating the correlation coefficient between the current data and the decomposed modal components, the target modal component corresponding to each current data can be determined more accurately. Even if the current data is affected by noise or other interference factors, the modal components related to the operating state characteristics of the motor can still be selected to enhance the robustness and anti-interference ability of the fault detection model. In addition, by accurately determining the target modal component, more useful information can be input into the fault detection model, thereby improving the training efficiency and accuracy of the model and reducing the computational amount and time cost in the model training process.

[0064] Optionally, according to the correlation coefficient, the target modal component corresponding to each current data is determined from multiple modal components corresponding to the current data, including: taking the modal component with the largest correlation coefficient among the multiple modal components corresponding to each current data as the target modal component corresponding to the current data.

[0065] The modal component with the largest correlation coefficient has the strongest correlation with the original current data and can best reflect the operating state characteristics of the motor in the current data. In this embodiment, by taking the modal component with the largest correlation coefficient as the target modal component corresponding to the current data, it is ensured that the modal component most matching the current data is selected, thereby further improving the accuracy and efficiency of detection.

[0066] Optionally, according to the correlation coefficient, the target modal component corresponding to each current data is determined from multiple modal components corresponding to the current data, including: taking the modal components with a correlation coefficient greater than or equal to the correlation coefficient threshold among the multiple modal components corresponding to each current data as the target modal components corresponding to the current data.

[0067] The correlation coefficient is used to measure the degree of linear correlation between the current data and the modal component corresponding to the current data. In this embodiment, a correlation coefficient threshold is preset in advance to screen out the modal components with a sufficiently high correlation with the current data. By taking the modal components with a correlation coefficient greater than or equal to the correlation coefficient threshold as the target modal components corresponding to the current data, not only can the modal component most matching the current data (i.e., the modal component with the largest correlation coefficient) be captured, but also other modal components with a significant correlation with the current data can be captured, so as to more comprehensively understand the characteristics of the current data, thereby further improving the accuracy and reliability of fault detection.

[0068] In some embodiments, the pre-constructed fault detection model includes a two-dimensional convolutional neural network model.

[0069] A Convolutional Neural Network (CNN) is a model in the field of deep learning. By introducing convolutional operations, the convolutional neural network can gradually extract hierarchical features of the input data, learning from the initial edge features to more advanced abstract features, and achieving a more effective representation of the input data. Through the training of large-scale data, the convolutional neural network can automatically learn high-level abstract representations of the input signal, has strong generalization ability, and can be applied to fault diagnosis problems under different types and working conditions. In this embodiment, a fault detection model is constructed based on a two-dimensional convolutional neural network model, utilizing the powerful feature extraction ability, generalization ability, and classification ability of the two-dimensional convolutional neural network model to automatically learn the deep features and differences between the fault signals generated by the motor under different working conditions, more accurately identify the features in the current data, and thus improve the accuracy of detection.

[0070] In the embodiments of the present disclosure, the pre-constructed two-dimensional convolutional neural network model includes an input layer, a convolutional layer, an activation function layer, a pooling layer, a fully connected layer, and an output layer. The input layer is used to receive input data (in this embodiment, the input data is an image converted from the target modal component). For image data, the input layer is a three-dimensional tensor, respectively representing the height, width, and color channel number of the image (such as the RGB three channels).

[0071] The convolutional layer includes multiple learnable convolutional kernels (or filters) for performing convolutional operations on the input data to extract local features in the image. The convolutional kernel slides on the input data and performs a dot product operation with the data within the sliding window to obtain a new feature map. For image data, since each pixel in the image is associated with the surrounding pixels, a two-dimensional convolutional kernel is used to better capture the spatial structure and patterns in the image. Specifically, the two-dimensional convolutional kernel can be represented as a matrix of learnable weight parameters:

[0072]

[0073] The two-dimensional convolutional operation is a process of performing point-by-point multiplication and summation on the local regions of the input image and the convolutional kernel. The convolutional kernel slides (or convolves) on the input image. Each time it slides to a new position, the input image region covered by the convolutional kernel is dot-producted with the convolutional kernel to generate a pixel value of the output image. Specifically, in the embodiments of the present disclosure, if f represents the input image data, then for a certain position (i, j) of the image data f, the calculation formula for its convolutional operation is as follows:

[0074]

[0075] Among them, M and N are the height and width of the convolution kernel respectively; f(i - m, j - n) is the pixel value of the input image data at the position (i - m, j - n); ω(m, n) is the weight parameter of the convolution kernel at the position (m, n).

[0076] In practical applications, the convolutional layer contains multiple convolution kernels. Each convolution kernel learns different features of the input data and performs different feature extractions on the input data, such as edges, textures, colors, etc. Each convolution kernel will perform a convolution operation on the input data to generate a new feature map. In addition, in this embodiment, the concept of weight sharing is introduced in the convolution operation, so that the convolution kernel uses the same weight in the local receptive field, which is of great significance for reducing the number of parameters and improving the operation efficiency. Due to weight sharing, the convolution operation has translational invariance, making the features learned by the model invariant to position changes in the image and making the model more generalizable.

[0077] The activation function layer is used to perform a non - linear mapping on the feature map output by the convolutional layer through an activation function (such as ReLU, sigmoid or tanh) to introduce non - linear factors and increase the expression ability and robustness of the model. The activation function converts the pixel value of each feature map into a new value, introducing non - linear characteristics, thereby allowing the model to learn more complex features.

[0078] The pooling layer is used to reduce the spatial dimensions (height and width) of the feature map, reduce the number of parameters and the amount of computation, reduce the computational complexity while retaining important information, that is, the main features of the feature map. In the embodiment of the present disclosure, the pooling operation adopts max - pooling, that is, the maximum value of the features in each pooling window is selected as the pooled feature value, the window size is 2 * 2, and the stride is 2.

[0079] After the pooling layer, the Flatten operation is adopted to flatten the multi - dimensional feature map into a one - dimensional vector and input it into the fully - connected layer. The role of the Flatten operation is to convert the multi - dimensional feature map into a one - dimensional vector. Specifically, the Flatten operation will arrange all the elements of the feature map in a certain order (such as row - major or column - major) into a one - dimensional array. In this way, the elements that originally had a spatial structure in the feature map are converted into a linear sequence for convenient input into the fully - connected layer for further processing.

[0080] The fully - connected layer is also called the dense - connected layer. Each neuron in the fully - connected layer is connected to all neurons in the previous layer, and each connection has a weight, forming a fully - connected structure to learn the non - linear combination between features and map the learned features to the label space of the training data.

[0081] In the embodiments of the present disclosure, the ReLU function is used as the activation function of the fully connected layer, which plays a role in non-linear transformation and increases the expression ability of the network: ReLU(x) = max(0, x). Compared with other activation functions, such as sigmoid or tanh, the ReLU function is more computationally efficient because it only needs to compare whether the input is greater than zero. When the input is negative, the output of the ReLU function is zero, which inactivates some neurons and produces a sparse activation effect, thereby preventing overfitting and improving the generalization ability of the model.

[0082] The number of neurons in the output layer is equal to the number of categories in the classification task. Exemplarily, in a specific application, there are a total of 7 categories including the normal state (i.e., no eccentricity), parallel eccentricity of 0.5 mm, parallel eccentricity of 1 mm, parallel eccentricity of 1.5 mm, eccentricity angle of 0.8°, eccentricity angle of 1.2°, and eccentricity angle of 1.5°. Then the number of neurons in the output layer is 7. In this embodiment, the Softmax function is used as the activation function of the output layer. The Softmax function can convert a given input vector into a probability distribution. Specifically, given an input vector z = (z 1 , z 2 , …, z k ), where k is the dimension of the input vector, the Softmax function will output a new vector σ(z) = (σ(z) 1 , σ(z) 2 , …, σ(z) k ). Among them, the calculation formula for the i-th element σ(z) i is:

[0083]

[0084] The calculation formula for the i-th element σ(z) i ensures that each element in the output vector is within the range of (0, 1), and the sum of all elements is 1, thus forming a valid probability distribution. The Softmax function assigns a higher probability to the larger values in the input vector and relatively smaller probabilities to other values, making the output of the model closer to the true probability distribution.

[0085] As shown in combination Figure 2 , the embodiments of the present disclosure provide another method for training a fault detection model, including:

[0086] S201, the processor obtains a training data set.

[0087] The training data set is a data set including current data of the motor at a preset rotational speed in different normal states, different eccentricity types, and different degrees of eccentricity.

[0088] S202, the processor performs variational mode decomposition on each current data in the training dataset to obtain multiple modal components corresponding to each current data.

[0089] S203, the processor determines the target modal component corresponding to each current data from the multiple modal components.

[0090] S204, the processor converts the target modal component into a grayscale image to obtain the training image.

[0091] In this step, converting the target modal component into a grayscale image can be achieved by mapping the values of the target modal component to the grayscale value range (such as 0 - 255), thereby generating the image data for training, that is, the training image.

[0092] S205, the processor inputs the training image into the two-dimensional convolutional neural network model for model training to obtain the trained fault detection model.

[0093] The training method of the fault detection model provided by the embodiments of the present disclosure can convert the target modal component into a grayscale image before model training. A grayscale image is an image that only contains luminance information, and the luminance change is closely related to the fluctuation and change pattern of the current data. Based on the grayscale image, the characteristic changes of the current data can be analyzed more clearly, and the relationship between the current data and the fault can be better understood, thereby improving the accuracy of fault detection. In addition, compared with color images, the data processing volume of grayscale images is greatly reduced, reducing the computational complexity and storage requirements, and improving the speed and efficiency of data processing.

[0094] In some embodiments, a single current data corresponds to multiple target modal components. In the case where a single current data corresponds to multiple target modal components, inputting the training image into the two-dimensional convolutional neural network model for model training includes: inputting the grayscale images converted from the multiple target modal components corresponding to the same current data into the two-dimensional convolutional neural network model for feature extraction to obtain multiple feature maps; fusing the multiple feature maps to obtain the target feature map; analyzing the target feature map to perform model training on the two-dimensional convolutional neural network model.

[0095] In this embodiment, when multiple target modal components correspond to a single current data, the grayscale images converted from the multiple target modal components are first input into a two-dimensional convolutional neural network model for feature extraction. Through the convolutional layer, activation function layer, and pooling layer, the two-dimensional convolutional neural network model can automatically learn and extract low-level to high-level features in the grayscale images. Then, the feature maps extracted from the multiple grayscale images are fused to obtain a target feature map. Feature map fusion can be achieved by concatenation or weighted summation to make full use of the complementary information in different target modal components, improve the prediction performance of the model, so as to more accurately identify and analyze the feature changes of the current data, and improve the accuracy and reliability of fault detection. The fused target feature map is analyzed to continue training the two-dimensional convolutional neural network model, optimize the model parameters, and improve the prediction performance of the model.

[0096] Optionally, inputting the training images into a two-dimensional convolutional neural network model for model training to obtain a trained fault detection model includes: dividing the training images to obtain a training image set and a test image set; wherein, both the training image set and the test image set include the training images corresponding to the current data in the normal state, different eccentricity types, and different eccentricity degrees; inputting the training images in the training image set into the two-dimensional convolutional neural network model for model training; inputting the training images in the test image set into the trained two-dimensional convolutional neural network model for testing, and obtaining the test accuracy; when the test accuracy meets the preset conditions, obtaining the trained fault detection model.

[0097] In this embodiment, by pre-dividing the training images, it is ensured that both the training image set and the test image set contain the training images corresponding to the current data in the normal state, different eccentricity types, and different eccentricity degrees, so that the model can fully learn the features in various states during the training process and verify its generalization ability in the test stage. In the model training stage, the training images in the training image set are input into the two-dimensional convolutional neural network model for model training. The two-dimensional convolutional neural network model continuously iterates and optimizes its parameters to learn how to extract useful features from the input image data and classify or regress these features to identify different states of the current data. In the model test stage, the training images in the test image set are input into the trained two-dimensional convolutional neural network model for testing, and the test accuracy is calculated. The test accuracy is an important indicator to measure the performance of the model, which reflects the generalization ability of the model on unseen data. When the test accuracy meets the preset conditions, the model is considered to be trained and saved and used as a fault detection model. If the test accuracy does not meet the preset conditions, it is necessary to adjust the model structure, optimize the algorithm, or increase the training data to improve the performance of the model.

[0098] In this embodiment, the training images generated from the grayscale images converted from multiple target modal components corresponding to the same current data should be assigned to the same set.

[0099] Optionally, the preset conditions include that the test accuracy is greater than or equal to the accuracy threshold and the number of training epochs is greater than or equal to the epoch threshold. The number of training epochs (Epoch) refers to the number of times the entire training dataset is completely traversed once.

[0100] In this embodiment, by setting two preset conditions of test accuracy and the number of training epochs, double guarantees are provided for the training of the model, ensuring that the model not only needs to have a certain generalization ability on unseen data (i.e., the test accuracy is greater than or equal to the accuracy threshold), but also needs to go through enough training epochs to optimize its parameters (i.e., the number of training epochs is greater than or equal to the epoch threshold), thereby improving the final performance of the model. In addition, by setting two preset conditions of test accuracy and the number of training epochs, it is also possible to avoid premature stopping of training resulting in the model not fully learning, or too late stopping of training resulting in overfitting of the model.

[0101] Exemplarily, the values of the accuracy threshold are 0.9, 0.92, 0.94, or 0.96, and the values of the epoch threshold are 35, 40, 45, or 50. It should be noted that the specific values of the accuracy threshold and the epoch threshold need to be preset by technicians according to the nature of the specific task and the characteristics of the data, and are not specifically limited here.

[0102] Optionally, inputting the training images in the training image set into a two-dimensional convolutional neural network model for model training includes: initializing the model parameters of the two-dimensional convolutional neural network model; inputting the training images in the training image set into the two-dimensional convolutional neural network model for forward propagation to obtain a prediction result; comparing the prediction result with the true result corresponding to the training image to calculate the loss function; calculating the gradient value of the loss function with respect to the model parameters through backpropagation, and updating the model parameters of the two-dimensional convolutional neural network model according to the gradient value.

[0103] In this embodiment, before starting the training, the model parameters of the two-dimensional convolutional neural network model are first initialized, such as the weights and biases of the convolutional kernels, the weights and biases of the fully connected layers, etc. The model parameters can be initialized with random values or pre-trained weights. The training images in the training image set are sequentially input into the two-dimensional convolutional neural network model for forward propagation. During the forward propagation process, the image data passes through structures such as convolutional layers, pooling layers, and fully connected layers, and finally outputs a prediction result. The prediction result is compared with the true result corresponding to the training image (i.e., the label of the training image), and the loss function is calculated. The loss function is used to measure the difference between the prediction result of the model and the actual result, and the cross-entropy loss function or the mean squared error loss function can be used as the loss function. The gradient value of the loss function with respect to the model parameters is calculated through the backpropagation algorithm. The backpropagation algorithm uses the chain rule to gradually transmit the gradient of the loss function from the output layer back to the input layer, so as to obtain the gradient value of each parameter. The model parameters of the two-dimensional convolutional neural network model are updated according to the gradient value. In this embodiment, optimization algorithms (such as SGD, Adam, etc.) can be used to update the model parameters to minimize the loss function. After each training round, the performance of the model is evaluated. If the test accuracy of the model is greater than or equal to the preset accuracy threshold, and the current training round is greater than or equal to the round threshold, the training is stopped and the model is saved to obtain a trained fault detection model.

[0104] Combined with Figure 3 As shown, an embodiment of the present disclosure provides a fault detection method, including:

[0105] S301, the processor acquires the real-time current data of the motor to be detected.

[0106] In this step, the real-time current data of the motor is acquired by collecting the real-time current data through the motor driver.

[0107] S302, the processor performs variational mode decomposition on the real-time current data to obtain multiple modal components.

[0108] S303, the processor determines the target modal component corresponding to the real-time current data from the multiple modal components.

[0109] S304, the processor converts the target modal component into an image and inputs it into the fault detection model to obtain a detection result.

[0110] Among them, the fault detection model is trained by using the training method of the fault detection model as described in the above embodiment.

[0111] The fault detection method provided by the embodiments of the present disclosure can decompose real-time current data into multiple modal components through variational mode decomposition. Each modal component represents different features in the real-time current data, achieving a more refined capture of the feature information in the real-time current data, especially the subtle changes related to the eccentricity fault. Although feature extraction is also performed in the related art, the detailed information in the current signal cannot be deeply analyzed. Compared with the related art, the embodiments of the present disclosure can more deeply and effectively extract the fault features in the real-time current data of the motor through variational mode decomposition and image processing, thereby further improving the accuracy of fault detection.

[0112] In addition, the fault detection model in the embodiments of the present disclosure is trained by using the training method of the fault detection model described in the above embodiments. Therefore, the technical effects possessed by the fault detection model described in the above embodiments are all possessed by the embodiments of the present disclosure, and will not be elaborated herein.

[0113] Optionally, the fault detection method further includes: obtaining the real-time operating parameters of the motor to be detected; inputting the real-time operating parameters into the trained reconstruction model for encoding and decoding, and obtaining the real-time reconstruction error of the real-time operating parameters; the reconstruction model is trained by using the operating parameters of the motor in the normal state; verifying the detection result according to the real-time reconstruction error; in the case of successful verification, sending the detection result to the target terminal device and displaying it; in the case of failed verification, updating the fault detection model according to the real-time reconstruction error.

[0114] In this embodiment, the real-time operating parameters of the motor to be detected include but are not limited to the vibration signal, current, voltage, temperature, etc. of the motor to be detected. The target terminal device refers to the terminal device corresponding to the staff such as motor maintenance personnel or technicians, and may include but are not limited to the console of the monitoring center, the mobile devices (such as mobile phones, tablets) of the maintenance personnel, etc.

[0115] In this embodiment, the reconstruction model is trained using the operating parameters of the motor in the normal state, learning the normal behavior pattern of the motor. By inputting the acquired real-time operating parameters into the pre-trained reconstruction model, through the encoding and decoding processes, the reconstructed operating parameters are generated, and the error between the real-time operating parameters and the reconstructed operating parameters is obtained, that is, the real-time reconstruction error. The real-time reconstruction error reflects the difference between the real-time operating parameters and the normal behavior pattern predicted by the model. Based on the real-time reconstruction error, the detection result is verified. If the verification passes, the detection result is sent to the target terminal device and displayed to ensure the accuracy of the detection result, and at the same time so that relevant personnel can quickly understand the fault situation and take corresponding countermeasures, improving the timeliness and accuracy of fault handling. If the verification fails, the fault detection model is updated according to the real-time reconstruction error. By updating, the fault detection model can more accurately reflect the actual fault situation, improving the accuracy and reliability of fault detection.

[0116] Compared with the related technology, the reconstruction model in this embodiment is trained using the operating parameters of the motor in the normal state, without the need to obtain a large amount of historical fault data with labeled motor eccentricity faults. In addition, since the reconstruction model can adaptively adapt to the normal operation mode of the motor. Through the real-time reconstruction error, the detection result of the fault detection model can be verified, improving the accuracy of the fault diagnosis of the fault detection model, thereby improving the accuracy and reliability of the overall fault detection.

[0117] In some embodiments, the reconstruction model includes an encoder and a decoder. The encoder is responsible for mapping the input data to a low-dimensional representation, and the decoder is responsible for restoring this low-dimensional representation to an output as close as possible to the original input. The encoding and decoding processes force the reconstruction model to learn to extract the key features in the input data, thus achieving an effective representation of the data.

[0118] Optionally, the reconstruction model is trained in the following manner: Obtain the operating parameters of the motor in the normal state as training data; Add noise to the operating parameters in the training data; Input the operating parameters with added noise into the encoder of the reconstruction model for encoding to obtain encoded data; Input the encoded data into the decoder of the reconstruction model for decoding to reconstruct the operating parameters without added noise.

[0119] In this embodiment, the reconstruction model can add noise to the input data and attempt to recover the original data from it, thereby learning the internal structure and features of the input data. During the encoding process, the encoder is responsible for converting the input data into an internal representation. This process is achieved by reducing the dimension of the data, that is, mapping the input data to a low-dimensional representation, thereby learning the compressed form of the input data. The structure of the decoder is a mirror image of the encoder. During the decoding process, the decoder is responsible for receiving the encoded data output by the encoder and attempting to reconstruct the original form of the input data. This process is achieved by gradually increasing the dimension of the data until it reaches the size of the original input data.

[0120] Optionally, training the reconstruction model further includes: using the operating parameters before noise addition as the original data; and using the reconstructed operating parameters without added noise as the reconstructed data; using the mean squared error loss function to evaluate the training reconstruction error between the original data and the reconstructed data; and obtaining the trained reconstruction model when the training reconstruction error is less than or equal to the second error threshold.

[0121] In this embodiment, the normal operating parameters of the motor before noise addition are used as the original data, that is, the target data that the reconstruction model needs to recover. The operating parameters obtained after the reconstruction model reconstructs the input data with added noise are used as the reconstructed data, that is, the output data of the reconstruction model, for comparison with the original data. The mean squared error (MSE) loss function is used to evaluate the training reconstruction error between the original data and the reconstructed data, which is used to measure the degree of difference between the two data sets. By clearly defining the original data and the reconstructed data and using the mean squared error loss function to evaluate the training reconstruction error, the reconstruction performance of the reconstruction model can be measured more accurately, so as to adjust the model parameters in a timely manner during the training process and improve the reconstruction ability and fault detection accuracy of the model.

[0122] In this embodiment, a second error threshold is preset to determine whether the reconstruction model is trained. During the training process, the training reconstruction error is continuously calculated and compared with the second error threshold. If the training reconstruction error is less than or equal to the second error threshold, it is considered that the reconstruction model can basically reconstruct the original data with a small error, and the reconstruction model has been trained, and the training process can be stopped. By setting the second error threshold as the training stop condition, overfitting or underfitting of the model is avoided.

[0123] In some embodiments, the reconstruction model includes multiple encoders and multiple decoders. The multiple encoders and multiple decoders are arranged at intervals in sequence. An adjacent encoder and decoder form a group, and multiple groups of encoders and decoders are stacked together to form the reconstruction model.

[0124] In this embodiment, the reconstruction model is composed of multiple groups of encoders and decoders stacked together, and each layer gradually learns a higher-level abstract representation of the input data. The output data of each layer can be used as the input data to train the next layer, thus forming a deep neural network. By learning the hierarchical structure of the data layer by layer, the reconstruction model can capture the complex structures and features in the data. By learning the hierarchical structure of the data layer by layer, the reconstruction model can capture the complex structures and features in the data, thereby improving the reconstruction ability of the model and the accuracy of fault detection. Since the reconstruction model can learn the complex structures and features of the data, when there are significant differences between the real-time operating parameters of the motor and the normal state (i.e., the fault condition), the reconstruction error will increase significantly, thereby improving the sensitivity of fault detection. Since noise is added and denoised in each layer of the reconstruction model, the model has stronger robustness to the noise of the input data. Even if the operating parameters of the motor are disturbed by noise during actual operation, the reconstruction model can still accurately reconstruct the normal operating state of the motor.

[0125] In some embodiments, the reconstruction model is trained with a denoising autoencoder or a stacked denoising autoencoder as the framework.

[0126] Optionally, the detection result includes the operating state of the motor to be detected, and the operating state includes a normal state and a fault state. In the case where the operating state is a fault state, the detection result further includes the eccentricity type and the degree of eccentricity. Verifying the detection result according to the real-time reconstruction error includes: when the real-time reconstruction error is less than or equal to the first error threshold and the operating state of the motor to be detected is a normal state, determining that the verification passes; or, when the real-time reconstruction error is greater than the first error threshold, the operating state of the motor to be detected is a fault state, and the fault level output by the reconstruction model is the same as the degree of eccentricity, determining that the verification passes; otherwise, determining that the verification fails. Wherein, the second error threshold is less than or equal to the first error threshold.

[0127] In this embodiment, a first error threshold is preset in advance. Since the reconstruction model can adaptively adapt to the normal operating mode of the motor, by comparing the real-time reconstruction error with the first error threshold, the operating state of the motor to be detected can be judged. Specifically, if the real-time reconstruction error is less than or equal to the first error threshold, it can be considered that the motor to be detected is operating normally; if the real-time reconstruction error is greater than the first error threshold, it can be considered that the motor to be detected has a fault. Then when the real-time reconstruction error is less than or equal to the first error threshold and the operating state of the motor to be detected output by the fault detection model is a normal state, the judgment results of the two are the same, and the verification passes.

[0128] When the real-time reconstruction error is greater than the first error threshold and the operating state of the motor to be detected output by the fault detection model is a fault state, the judgment results of the two on the operating state of the motor to be detected are the same. Further, the reconstruction model in this embodiment can output a fault level to characterize the severity of the fault. By comparing the eccentricity degree output by the fault detection model with the fault level output by the reconstruction model, it is determined whether the classifications corresponding to the eccentricity degree and the fault level are the same. When the classifications corresponding to the eccentricity degree and the fault level are the same, the judgment results of the two on the fault degree of the motor to be detected are the same, and the verification passes.

[0129] In this embodiment, the output of the fault level of the reconstruction model can be achieved by presetting multiple error thresholds. Exemplarily, a first error threshold, a third error threshold, and a fourth error threshold are preset. Among them, the first error threshold < the third error threshold < the fourth error threshold. When the real-time reconstruction error is greater than the first error threshold and less than the third error threshold, the fault level is determined to be mild; when the real-time reconstruction error is greater than the third error threshold and less than the fourth error threshold, the fault level is determined to be moderate; when the real-time reconstruction error is greater than the fourth error threshold, the fault level is determined to be severe. The eccentricity degree output by the fault detection model can mark the eccentricity degree for different eccentricity amounts during the model training process. The eccentricity degree markings include mild, moderate, and severe. Exemplarily, a parallel eccentricity amount of 0.5 mm or an eccentricity angle value of 0.8° corresponds to a mild eccentricity degree, a parallel eccentricity amount of 1 mm or an eccentricity angle value of 1.2° corresponds to a moderate eccentricity degree, and a parallel eccentricity amount of 1.5 mm or an eccentricity angle value of 1.5° corresponds to a severe eccentricity degree, so as to output the eccentricity degree marking (mild, moderate, or severe) corresponding to the eccentricity amount to characterize the fault degree after the fault detection.

[0130] Optionally, updating the fault detection model according to the real-time reconstruction error includes: when the real-time reconstruction error is less than or equal to the first error threshold, marking the real-time operating parameters as the normal state and retraining the fault detection model with the new training samples; when the real-time reconstruction error is greater than the first error threshold, marking the real-time operating parameters according to the eccentricity type output by the fault detection model and the fault level output by the reconstruction model, and retraining the fault detection model with the marked real-time operating parameters as the new training samples. Through the update, the fault detection model can more accurately reflect the actual fault situation and improve the accuracy and reliability of the fault detection.

[0131] It should be noted that the specific values of the first error threshold, the second error threshold, the third error threshold, and the fourth error threshold involved in the embodiments of the present disclosure need to be preset by those skilled in the art according to the nature of the actual specific task and the characteristics of the input data, and are not specifically limited here.

[0132] Combined Figure 4 As shown, an embodiment of the present disclosure provides a training device 40 for a fault detection model, including a first acquisition module 410, a first decomposition module 420, a first determination module 430, and a model training module 440. The first acquisition module 410 is configured to obtain a training data set; the training data set is a data set including current data of a motor at a preset rotational speed under normal conditions, different eccentricity types, and different eccentricity degrees; the first decomposition module 420 is configured to perform variational mode decomposition on each current data in the training data set to obtain multiple modal components corresponding to each current data; the first determination module 430 is configured to determine a target modal component corresponding to each current data from the multiple modal components; the model training module 440 is configured to convert the target modal component into an image and input it into a pre-constructed fault detection model for model training to obtain a trained fault detection model.

[0133] The training device 40 for the fault detection model provided by the embodiment of the present disclosure can implement the training method of the fault detection model described in the above embodiment. Therefore, the technical effects possessed by the training method of the fault detection model described in the above embodiment are all possessed by the embodiment of the present disclosure, and will not be elaborated here.

[0134] Optionally, the first determination module 430 is further configured to calculate the correlation coefficient between the current data and each modal component corresponding to the current data respectively; and determine the target modal component corresponding to the current data from the multiple modal components corresponding to each current data according to the correlation coefficient.

[0135] Optionally, the first determination module 430 is further configured to use the modal component with the largest correlation coefficient among the multiple modal components corresponding to each current data as the target modal component corresponding to the current data; or use the modal components with a correlation coefficient greater than or equal to a correlation coefficient threshold among the multiple modal components corresponding to each current data as the target modal component corresponding to the current data.

[0136] Optionally, the pre-constructed fault detection model includes a two-dimensional convolutional neural network model. The model training module 440 is further configured to convert the target modal component into a grayscale image to obtain a training image; and input the training image into the two-dimensional convolutional neural network model for model training to obtain a trained fault detection model.

[0137] Optionally, the model training module 440 is further configured to divide the training images to obtain a training image set and a test image set; wherein, both the training image set and the test image set include training images corresponding to current data in a normal state, different eccentricity types, and different eccentricity degrees; input the training images in the training image set into the two-dimensional convolutional neural network model for model training; input the training images in the test image set into the trained two-dimensional convolutional neural network model for testing, and obtain the test accuracy; when the test accuracy meets the preset conditions, obtain the trained fault detection model.

[0138] Optionally, the model training module 440 is further configured to initialize the model parameters of the two-dimensional convolutional neural network model; input the training images in the training image set into the two-dimensional convolutional neural network model for forward propagation to obtain a prediction result; compare the prediction result with the true result corresponding to the training image, and calculate the loss function; calculate the gradient value of the loss function with respect to the model parameters through backpropagation, and update the model parameters of the two-dimensional convolutional neural network model according to the gradient value.

[0139] Combined with Figure 5 As shown, an embodiment of the present disclosure provides a fault detection device 50, including a second acquisition module 510, a second decomposition module 520, a second determination module 530, and a fault detection module 540. The second acquisition module 510 is configured to acquire real-time current data of the motor to be detected; the second decomposition module 520 is configured to perform variational mode decomposition on the real-time current data to obtain a plurality of modal components; the second determination module 530 is configured to determine the target modal component corresponding to the real-time current data from the plurality of modal components; the fault detection module 540 is configured to convert the target modal component into an image and input it into the fault detection model to obtain a detection result; wherein, the fault detection model is trained by using the training method of the fault detection model described in the above embodiment.

[0140] The fault detection device 50 provided by the embodiment of the present disclosure can implement the fault detection method described in the above embodiment. Therefore, the technical effects possessed by the fault detection method described in the above embodiment are all possessed by the embodiment of the present disclosure, and will not be elaborated here.

[0141] Optionally, the fault detection module 540 is further configured to acquire real-time operating parameters of the motor to be detected; input the real-time operating parameters into the trained reconstruction model for encoding and decoding, and obtain the real-time reconstruction error of the real-time operating parameters; the reconstruction model is trained by using the operating parameters of the motor in a normal state; verify the detection result according to the real-time reconstruction error; when the verification is passed, send the detection result to the target terminal device for display; when the verification fails, update the fault detection model according to the real-time reconstruction error.

[0142] Combined Figure 6 As shown, an embodiment of the present disclosure provides an electronic device 60, including a processor 600 and a memory 601. Optionally, the electronic device 60 may further include a communication interface 602 and a bus 603. Among them, the processor 600, the communication interface 602, and the memory 601 can complete mutual communication through the bus 603. The communication interface 602 can be used for information transmission. The processor 600 can call the logical instructions in the memory 601 to execute the training method of the fault detection model described in the above embodiments, or the fault detection method described in the above embodiments.

[0143] In addition, when the logical instructions in the above-mentioned memory 601 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0144] The memory 601, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as the program instructions / modules corresponding to the methods in the embodiments of the present disclosure. The processor 600 executes functional applications and data processing by running the program instructions / modules stored in the memory 601, that is, implements the training method of the fault detection model described in the above embodiments, or the fault detection method described in the above embodiments.

[0145] The memory 601 may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the terminal device, etc. In addition, the memory 601 may include a high-speed random access memory and may also include a non-volatile memory.

[0146] An embodiment of the present disclosure provides a computer-readable storage medium storing computer-executable instructions, and the computer-executable instructions are set to execute the training method of the fault detection model described in the above embodiments, or the fault detection method described in the above embodiments.

[0147] The technical solution of the embodiments of the present disclosure can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes one or more instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the embodiments of the present disclosure. The foregoing storage medium may be a non-transitory storage medium, such as: a USB flash drive, a portable hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc, etc., which are various media that can store program codes.

[0148] The above description and drawings fully illustrate the embodiments of the present disclosure so that those skilled in the art can practice them. Other embodiments may include structural, logical, electrical, process, and other changes. Embodiments merely represent possible variations. Unless explicitly required, separate components and functions are optional, and the order of operations may vary. Parts and features of some embodiments may be included in or substituted for parts and features of other embodiments. Moreover, the terms used in this application are only for describing embodiments and are not used to limit the claims. As used in the description of the embodiments and the claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to also include the plural forms. Similarly, as used in this application, the term "and / or" refers to any and all possible combinations including one or more of the associated listed items. Additionally, when used in this application, the term "comprise" and its variants "comprises" and / or "comprising", etc. mean the presence of the stated features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or groups of these. Without further limitation, an element defined by the statement "comprising one..." does not exclude the presence of another identical element in the process, method, or device comprising the element. In this document, each embodiment may focus on the differences from other embodiments, and the same or similar parts among the embodiments may be referred to each other. For the methods, products, etc. disclosed in the embodiments, if they correspond to the method part disclosed in the embodiments, the relevant parts may refer to the description of the method part.

[0149] Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner can depend on the specific application and design constraints of the technical solution. The skilled person can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of the embodiments of the present disclosure. The skilled person can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0150] In the embodiments disclosed herein, the disclosed methods, products (including but not limited to devices, equipment, etc.) can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units can be merely a logical function division, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms. The units described as separate components can be or can not be physically separated, and the components displayed as units can be or can not be physical units, that is, they can be located in one place or can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to implement this embodiment. In addition, in the embodiments of the present disclosure, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0151] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a portion of code, which contains one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions noted in the blocks may occur in a different order than noted in the accompanying drawings. For example, two consecutive blocks may in fact be executed substantially in parallel, or they may sometimes be executed in the reverse order, depending on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks may also occur in a different order than disclosed in the description, and sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may in fact be executed substantially in parallel, or they may sometimes be executed in the reverse order, depending on the functions involved. Each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

Claims

1. A method for training a fault detection model, characterized in that: include: Get the training dataset; The training data set is a data set including current data of the motor in a normal state, different eccentricity types and different eccentricity degrees; Performing variational modal decomposition on each current data in the training data set to obtain multiple modal components corresponding to each current data; Determine a target modal component corresponding to each current data from the multiple modal components; The target modal components are converted into images and input into a pre-built fault detection model for model training to obtain a trained fault detection model.

2. The training method according to claim 1, characterized in that: From multiple modal components, determine the target modal component corresponding to each current data, including: respectively calculating the correlation coefficient between the current data and each modal component corresponding to the current data; According to the correlation coefficient, a target modal component corresponding to each current data is determined from a plurality of modal components corresponding to the current data.

3. The training method according to claim 2, characterized in that: According to the correlation coefficient, a target modal component corresponding to each current data is determined from multiple modal components corresponding to the current data, including: Among the multiple modal components corresponding to each current data, the modal component with the largest correlation coefficient is used as the target modal component corresponding to the current data; or, Among the multiple modal components corresponding to each current data, the modal component with a correlation coefficient greater than or equal to the correlation coefficient threshold is used as the target modal component corresponding to the current data.

4. The training method according to any one of claims 1 to 3, characterized in that: Pre-built fault detection models include a 2D convolutional neural network model; Convert the target modal component into an image and input it into a pre-built fault detection model for model training to obtain a trained fault detection model, including: Convert the target modal component into a grayscale image to obtain a training image; The training images are input into the two-dimensional convolutional neural network model for model training to obtain a trained fault detection model.

5. The training method according to claim 4, characterized in that: Inputting the training images into a two-dimensional convolutional neural network model for model training to obtain a trained fault detection model, including: dividing the training images to obtain a training image set and a test image set; wherein the training image set and the test image set both include training images corresponding to current data in a normal state, different eccentricity types, and different eccentricity degrees; inputting the training images in the training image set into the two-dimensional convolutional neural network model for model training; inputting the training images in the test image set into the trained two-dimensional convolutional neural network model for testing, and obtaining a test accuracy; when the test accuracy meets a preset condition, obtaining a trained fault detection model; and / or, The training image is input into the two-dimensional convolutional neural network model for model training, including: initializing the model parameters of the two-dimensional convolutional neural network model; inputting the training image into the two-dimensional convolutional neural network model for forward propagation to obtain a prediction result; comparing the prediction result with the actual result corresponding to the training image, and calculating the loss function; calculating the gradient value of the loss function to the model parameters through back propagation, and updating the model parameters of the two-dimensional convolutional neural network model according to the gradient value.

6. A fault detection method, characterized in that: include: Obtain real-time current data of the motor to be tested; Perform variational modal decomposition on real-time current data to obtain multiple modal components; Determine a target modal component corresponding to the real-time current data from among the multiple modal components; The target modal component is converted into an image and input into a fault detection model to obtain a detection result; wherein the fault detection model is trained using the training method for the fault detection model as described in any one of claims 1 to 5.

7. The fault detection method according to claim 6, characterized in that: Also includes: Obtain the real-time operating parameters of the motor to be tested; The real-time operation parameters are input into the trained reconstruction model for encoding and decoding, and the real-time reconstruction error of the real-time operation parameters is obtained; The reconstructed model is obtained by training with the operating parameters of the motor under normal conditions; Verify the detection results based on the real-time reconstruction error; If the verification is successful, the test results will be sent to the target terminal device and displayed; In case the verification fails, the fault detection model is updated according to the real-time reconstruction error.

8. A training device for a fault detection model, characterized in that: include: A first acquisition module is configured to acquire a training data set; The training data set is a data set including current data of the motor in a normal state, different eccentricity types and different eccentricity degrees; A first decomposition module is configured to perform variational modal decomposition on each current data in the training data set to obtain multiple modal components corresponding to each current data; A first determination module is configured to determine a target modal component corresponding to each current data from a plurality of modal components; The model training module is configured to convert the target modal component into an image and input it into a pre-built fault detection model for model training to obtain a trained fault detection model.

9. A fault detection device, characterized in that: include: A second acquisition module is configured to obtain real-time current data of the motor to be detected; A second decomposition module is configured to perform variational modal decomposition on the real-time current data to obtain a plurality of modal components; A second determination module is configured to determine a target modal component corresponding to the real-time current data from a plurality of modal components; The fault detection module is configured to convert the target modal component into an image and input it into a fault detection model to obtain a detection result; wherein the fault detection model is trained using the training method of the fault detection model as described in any one of claims 1 to 6.

10. An electronic device, characterized in that: include: processor; and A memory storing program instructions, wherein the processor is configured to execute the training method of the fault detection model according to any one of claims 1 to 6, or the fault detection method according to claim 7.

Images