Motor fault diagnosis method and system based on multidimensional data under small sample condition

Abstract

The invention discloses a motor fault diagnosis method and system based on multi-dimensional data under a small sample condition, is applied to intelligent diagnosis of motor faults, and aims to solve the problem of small samples. The fault diagnosis model fusing the ball convolution operator and supervised branch and self-supervised branch cooperative training is constructed, and efficient and accurate fault diagnosis of the motor system can be realized under the condition that only a small number of labeled samples are used and multiple complex working conditions are used. In the model training stage, a label is generated through a pseudo label generation mechanism and participates in training together with original data, and a weighted loss function structure and an adaptive weight adjustment strategy are designed in a matched manner, so that the learning efficiency and generalization ability of the model on different fault categories are optimized. In conclusion, high-precision motor fault diagnosis can be realized only by depending on a small number of labeled samples, and the cost of the electric driving system in the two aspects of sample labeling and fault diagnosis is effectively reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to motor state monitoring and intelligent fault diagnosis technology, and in particular to a motor fault diagnosis method and system based on multi-dimensional data under a small sample condition. BACKGROUND

[0002] With the development of industrial automation and intelligent manufacturing, motors are widely used in new energy vehicles, aerospace, robots and national defense equipment as key execution components. The motor operation process is affected by multiple factors such as speed, load, environment, etc., and is prone to faults such as stator winding short circuit, rotor broken bar, bearing damage, imbalance and shaft eccentricity. Once a fault occurs, it not only causes equipment damage and economic loss, but also may cause safety problems. Traditional motor fault diagnosis methods mainly rely on signal processing techniques such as spectrum analysis and wavelet packet decomposition, combined with expert experience for feature extraction and discrimination, which has the disadvantages of relying on prior knowledge and manual features being difficult to be comprehensive and robust. In recent years, with the rise of deep learning technology, fault diagnosis methods based on convolutional neural networks (CNN) have made significant progress in feature learning and classification accuracy. However, such methods usually rely on a large amount of labeled data for supervised training, while in actual industrial scenarios, fault samples are difficult to obtain and the labeling cost is high, causing the model to be prone to overfitting and poor generalization performance. To solve the problem of insufficient labeled samples, self-supervised learning uses unlabeled data to learn useful representations through proxy tasks, which has been widely used in image and natural language fields. However, directly applying it to motor vibration signals faces challenges such as high data dimensionality, insufficient utilization of geometric characteristics, and difficulty in ensuring three-dimensional rotational invariance. SUMMARY

[0003] The technical problem to be solved by the present application is that most existing fault diagnosis methods use data-driven techniques, which require a large number of labeled samples to train the model. However, the labeling of samples requires the knowledge of domain experts, and the cost of labeling samples is high. To overcome the technical obstacles / challenges of effectively applying self-supervised learning to the field of motor vibration signal fault detection, the present application proposes a motor fault diagnosis method and system based on multi-dimensional data under a small sample condition, which is a fault diagnosis method based on the Self-SL framework. It combines the concept of multi-task learning and cooperatively trains two task branches of supervised and self-supervised, can adaptively learn feature representations from both supervised and self-supervised task branches, and optimally combines a dynamically adjusted loss function and introduces a spherical convolution model. The diagnosis method proposed in the present application can achieve high diagnostic accuracy with very few labeled samples.

[0004] Therefore, the present application provides the following technical solutions:

[0005] The application provides a motor fault diagnosis method based on multi-dimensional data under a small sample condition, comprising the following steps:

[0006] S101, data acquisition, acquiring three-axis vibration time series signals in various motor states through a three-axis vibration sensor, and constructing three-dimensional point cloud data of the motor by using the three-axis vibration time series signals, wherein the motor state is a real fault label;

[0007] S102, data enhancement and spherical mapping, performing a three-dimensional rotation operation on the three-dimensional point cloud data of step S101 to obtain rotated point cloud data and a real rotation label, and performing spherical mapping on the rotated point cloud data to obtain rotated spherical data; performing spherical mapping on the three-dimensional point cloud data of step S101 to obtain original spherical data;

[0008] S103, cooperative training of a supervised branch and a self-supervised branch, the supervised branch and the self-supervised branch adopt the same feature extraction module and direct the output to different classifiers, in the training stage, the rotated spherical data and the real rotation label are input into the self-supervised branch, and the original spherical data and the real fault label are input into the supervised branch;

[0009] S104, inputting the original spherical data corresponding to the three-axis vibration time series signals of the motor to be diagnosed into the trained supervised branch to obtain a fault diagnosis classification result.

[0010] Optionally, the total loss value in the network training stage is weighted by the supervised task and the self-supervised task, and a dynamic weighting mechanism is introduced:

[0011] ;

[0012] In the formula, is the total loss value, is the loss value of task i, i is 1 and 2, is the loss value of task i, that is, corresponding is the loss value of the self-supervised task, is the loss value of the supervised task; is the uncertainty parameter of task i,

[0013] the network training process, the following update exists:

[0014] ;

[0015] In the formula, denotes the updated uncertainty parameter , and η is a proportional constant.

[0016] Optionally, the self-supervised branch is provided with a feature extraction module and a linear layer 1, and the supervised branch is provided with a feature extraction module and a linear layer 2, and the linear layer 1 and the linear layer 2 are respectively classifiers for rotation type prediction and fault category prediction;

[0017] The feature extraction module is a spherical convolution module, and the spherical convolution module is composed of a plurality of convolution layers connected in sequence, and each convolution layer is stacked by a spherical convolution layer, a nonlinear layer and a pooling layer.

[0018] Optionally, the processing process of the spherical convolution layer is as follows:

[0019] Firstly, the current spherical signal and the filter are mapped to the frequency domain by spherical Fourier transform;

[0020] ;

[0021] ;

[0022] Wherein, is the signal corresponding to the spherical Fourier transform coefficient of the order , is the spherical signal on the two-dimensional unit sphere , x is a point on the sphere, is the spherical harmonic function of the order , the filter is the neural network convolution kernel of the spherical convolution layer, which is a two-dimensional matrix of 3*3, is the filter corresponding to the spherical Fourier transform coefficient of the order ; Then, the dot product operation is completed in the frequency domain to obtain ;

[0023]

[0024] ;

[0025] Finally, the spherical inverse Fourier transform is used to map back to the original space;

[0026] ;

[0027] Wherein, is the result of the current spherical convolution layer, and b is the bandwidth.

[0028] Briefly, the spherical convolution layer is inputted with the spherical data obtained by spherical mapping and the original spherical data for feature extraction. ​​

[0029] Optionally, the process of spherical mapping of the rotated point cloud data or the process of spherical mapping of the three-dimensional point cloud data in step S102 is:

[0030] mapping the rotated point cloud data or the three-dimensional point cloud data to a two-dimensional unit sphere , that is, obtaining a geometric representation in a spherical coordinate system with a radius of = 1;

[0031] wherein a center point of the rotated point cloud data or the three-dimensional point cloud data is calculated first;

[0032] then a ray is drawn from the center point according to an azimuth angle and a polar angle , and the spherical mapping can be regarded as a function, and the formula is as follows:

[0033] ;

[0034] wherein an output is a mapping value, that is, a spherical data value on , a geometric representation in a spherical coordinate system, is a distance between the center point and the farthest intersection point ; and the value ranges of and are (0, 2π] and (0, π] respectively, and the obtained is uniformly distributed through equidistant sampling.

[0035] Optionally, the three-dimensional rotation operation on the three-dimensional point cloud data in step S102 is a random three-dimensional rotation operation, that is, a randomly generated rotation direction and angle, and then a real rotation label and rotated point cloud data are obtained.

[0036] The rotation direction is rotation around the x-axis, the y-axis and the z-axis.

[0037] Optionally, the process of constructing the three-dimensional point cloud data in step S101 is: first, the collected three-axis vibration time series signal is normalized to the [0, 1] interval by using the maximum and minimum value normalization method, and then converted into a three-dimensional point cloud; then, a random rotation transformation for the purpose of adding interference and adding random Gaussian noise are performed.

[0038] The calculation formula of the random rotation is:

[0039] ​

[0040] ;

[0041] wherein, is the data after random rotation transformation, R is a rotation matrix, and p is point cloud data of a sample, such as , T is a matrix transpose, , , are three-dimensional rotation matrices around the x-axis, the y-axis, and the z-axis, respectively; a, b, and c are rotation angles of the rotation matrix , , are sampled from a uniform distribution , and the preset maximum rotation angle is π / 18.

[0042] The application also provides a system based on the above motor fault diagnosis method, comprising:

[0043] a data acquisition module for acquiring three-axis vibration time series signals in various motor states through a three-axis vibration sensor and constructing three-dimensional point cloud data of the motor using the three-axis vibration time series signals, wherein the motor state is a real fault label;

[0044] a data enhancement and spherical mapping module for performing a three-dimensional rotation operation on the three-dimensional point cloud data to obtain rotated point cloud data and a real rotation label, and performing spherical mapping on the rotated point cloud data to obtain rotated spherical data; and performing spherical mapping on the three-dimensional point cloud data of step S101 to obtain original spherical data;

[0045] a collaborative training module for collaborative training of a supervised branch and a self-supervised branch, wherein the supervised branch and the self-supervised branch use the same feature extraction module and direct the output to different classifiers, in the training stage, the rotated spherical data and the real rotation label are input into the self-supervised branch, and the original spherical data and the real fault label are input into the supervised branch;

[0046] a diagnosis module for inputting original spherical data corresponding to three-axis vibration time series signals of a motor to be diagnosed into the trained supervised branch to obtain a fault diagnosis classification result.

[0047] The application also provides a computer device, comprising one or more processors and a memory storing a computer program;

[0048] wherein the processor invokes the computer program to implement the steps of the above motor fault diagnosis method.

[0049] The application further provides a computer readable storage medium storing a computer program, wherein the computer program is called by a processor to implement steps of the motor fault diagnosis method.

[0050] The following progress is achieved compared with the prior art:

[0051] (1) The technical scheme of the application can significantly reduce the dependence on labeled data and improve the diagnostic accuracy under small sample conditions. Traditional deep learning methods usually require thousands or even tens of thousands of labeled samples to achieve high diagnostic accuracy. However, the application can achieve a diagnostic accuracy of more than 90% by using self-supervised learning with only a small number of labeled samples. Specifically, the self-supervised auxiliary task automatically generates pseudo labels (true rotation labels) using unlabeled data, enabling the model to learn more generalizable feature representations and reducing the dependence on human labeling. In the preferred technical scheme, the dynamic weight loss function adaptively adjusts the weights of the supervised classification task and the self-supervised feature learning task, ensuring that the model can still be optimized stably under limited labeled data.

[0052] (2) The three-dimensional feature representation capability is strong. The technical scheme of the application uses three-dimensional point cloud data and spherical convolution to provide richer spatial features than traditional one-dimensional / two-dimensional data, preserve the spatial distribution characteristics of the motor operating state, and improve the distinguishability of different fault samples.

[0053] (3) The generalization ability is improved. The rotation equivariance of spherical convolution enhances the robustness of the model to three-dimensional data, reducing the number of samples required for training. Experiments show that the diagnostic accuracy reaches 92.83% with 48 labeled samples. The use of spherical convolution (SO(3) group equivariance) makes the model robust to data rotation and avoids performance degradation caused by differences in sensor installation angles.

[0054] (4) Multi-task collaborative optimization. The dynamic weight mechanism in the preferred technical scheme of the application balances self-supervised and supervised tasks, accelerates model convergence, and improves diagnostic accuracy. It is suitable for various motor fault diagnoses such as stator winding short circuit, rotor broken bar, and bearing fault. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 is the technical roadmap of the motor fault diagnosis method provided by the embodiment of the application;

[0056] Figure 2 is the flowchart of the motor fault diagnosis method provided by embodiment 1 of the application. DETAILED DESCRIPTION

[0057] In order to make the purposes, technical solutions and advantages of the present application clearer, further detailed description will be given below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application. The technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0058] It should be noted that although the functional modules are divided in the device schematic diagram, and the logical sequence is shown in the flowchart, in some cases, the steps shown or described can be performed in a manner different from the module division in the device or the sequence in the flowchart. The terms "first", "second", etc. in the specification and claims and the above-described drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence.

[0059] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0060] The technical solution of the present application provides a motor fault diagnosis method and system based on multi-dimensional data under small sample conditions, which realizes motor fault diagnosis based on multi-dimensional data processing and self-supervised deep learning of three-axis vibration data under the condition of few labeled samples. In order to make the personnel in the technical field better understand the technical solution of the present application, the present application will be further described in detail below in combination with the drawings.

[0061] Embodiment 1

[0062] As shown in Figure 1 and 2 The motor fault diagnosis method based on multi-dimensional data under small sample conditions provided by the embodiment of the present application comprises the following steps:

[0063] Step 1: Data acquisition, three-axis vibration time series signals under various motor states are collected by a three-axis vibration sensor, and three-dimensional point cloud data of the motor is constructed by using the three-axis vibration time series signals, wherein the motor state is a real fault label.

[0064] Among them, through the triaxial vibration sensor along X, Y, Z three orthogonal directions, collect three-axis vibration time series signal of motor under normal and various fault conditions. If the sampling frequency is not less than 10kHz, each sample contains at least 10000 data points of single axis. After normalizing the obtained three-axis vibration time series signal, map to three-dimensional coordinate values of spatial coordinate system, form three-dimensional point cloud data of motor vibration, that is, data points from three channels are respectively taken as values in three orthogonal dimensions in three-dimensional space, x, y, z axis data of three-dimensional point cloud data correspond to three dimensions X, Y, Z of spatial rectangular coordinate system.

[0065] In this embodiment, step 1 is preferably performed according to the following process:

[0066] S1-1: Collect motor data under three-channel three load conditions and eight fault states using sensors: eight motor states are normal state NM, rotor broken bar fault BBM, stator winding fault WFM, bearing fault FBM, rotor misalignment fault RMM, single-phase open circuit fault SPM, rotor imbalance fault RUM, and rotor shaft bending fault BRM. The maximum and minimum value normalization method is used to normalize the data to the [0, 1] interval. The normalization calculation formula is as follows:

[0067]

[0068] Among them, is all data points in the th channel, and are the maximum data point value and the minimum data point value in the th channel, is the maximum value of all sampling data points. After normalization, data points from three channels are respectively taken as values in three orthogonal dimensions in three-dimensional space, and data fusion is completed to construct three-dimensional point cloud of the motor operating state.

[0069] S1-2: In order to improve the generalization ability of the model, a data enhancement technology for three-dimensional point cloud technology is introduced. In the training stage of the model, one or more minor transformations (or disturbances) are introduced to a training sample to be input to the model, but the label of the fault sample is not changed. First, the sample is randomly rotated. Assuming that the original sample is , T is the matrix transpose symbol, R represents the rotation matrix, and the calculation formula of the random rotation transformation is as follows:

[0070]

[0071] Among them, is the sample after random rotation transformation, , , are three-dimensional rotation matrices for the sample p around the x-axis, the y-axis and the z-axis respectively, a, b, c are the rotation angles of the rotation matrix , , are sampled from a uniform distribution In the embodiment of the present application π / 18 is taken. Then add random Gaussian noise to each point in the rotated sample, as shown in the following formula:

[0072]

[0073] In the formula, denotes the sample after adding Gaussian noise, denotes Gaussian noise, denotes a normal distribution.

[0074] It should be understood that through the above normalization, rotation and addition of Gaussian noise in the embodiment, the training data set is expanded, and the robustness of the model is improved by introducing interference. In other feasible embodiments, the above preprocessing methods can be selectively executed or not executed, and the present application only requires step 1 to be three-axis vibration time series signal converted into three-dimensional point cloud data, and the present application does not make specific limitations on the pre-processing and selection of motor fault types.

[0075] Step 2: Data augmentation and spherical mapping, performing a three-dimensional rotation operation on the three-dimensional point cloud data of step 1 to obtain rotated point cloud data and true rotation labels, and then performing spherical mapping on the rotated point cloud data to obtain rotated spherical data; performing spherical mapping on the three-dimensional point cloud data of step 1 to obtain original spherical data.

[0076] The technical scheme of the present application has a supervised branch and a self-supervised branch, and the data required by the two branches is different. Therefore, for the self-supervised branch, the three-dimensional point cloud data of step 1 is randomly rotated in three dimensions, and the corresponding pseudo-labels (true rotation labels) are generated according to the rotation direction and angle. The rotated data and its pseudo-labels will be used for the self-supervised task, and the data with true labels will be used for the supervised task. In detail, first, the original sample is randomly rotated in three dimensions. The three-dimensional rotation operation consists of two steps, namely direction and angle setting. The direction options include x-axis, y-axis and z-axis, indicating the selection of the rotation axis. The size of the angle can be set to multiple options, and the number of options directly determines the number of predicted labels in the self-supervised task and the difficulty of the rotation recognition task. For example, if the rotation angle is set to π / 2, there are four types of true rotation labels, which are: rotating π / 2 along the x-axis, rotating π / 2 along the y-axis, rotating π / 2 along the z-axis, and not rotating; the more the number of true rotation label types, the more difficult it is for the model to achieve high accuracy in the self-supervised task.

[0077] Based on this, the preferred execution process of the three-dimensional rotation operation in step 2 in this embodiment is as follows:

[0078] S2-1: First, perform a random three-dimensional rotation operation on the original sample.

[0079] S2-2: Based on the randomly generated rotation direction and angle, pseudo-labels are generated and formed with the rotated data to form sample-label pairs, which are then fed into the self-supervised task branch of the model. In this embodiment, three angle options, π / 3, π / 4, and π / 6, are selected. As the number of targets to be predicted increases, the difficulty of the rotation recognition task increases, and the test and diagnostic accuracy of the model also increases.

[0080] The process of spherical mapping for either rotated point cloud data or 3D point cloud data is as follows: mapping the rotated point cloud data or 3D point cloud data onto a two-dimensional unit sphere. That is, to obtain the radius The geometric representation of a spherical sphere in spherical coordinates. In this context, a two-dimensional unit sphere is the set of all points in three-dimensional Euclidean space that are equidistant from a fixed point; it can be viewed as the surface of a sphere with a radius of one. The two dimensions correspond to the intrinsic dimension of the sphere itself as a curved surface, rather than two Cartesian coordinate dimensions in three-dimensional space. Two-dimensional unit sphere It can be parameterized as:

[0081] ;

[0082] in, Polar angle, Azimuth angle, spherical signal It is a continuous function that represents the two-dimensional unit sphere. Points on the map are mapped to complex or real numbers: R1 represents the set of real numbers, and C represents the set of complex numbers.

[0083] The technical solution of this invention can reduce the dimensionality of the original 3D data through spherical mapping, avoiding the computational cost of directly processing 3D data, while retaining its original 3D characteristics. The specific process is as follows:

[0084] First, calculate the center point of the rotated point cloud data or the 3D point cloud data.

[0085] Then according to the azimuth angle and polar angle From the center point A ray is drawn from the given point, and the farthest intersection point between this ray and the original data (including rotated point cloud data and 3D point cloud data) is calculated. distance Distance value of the farthest intersection point For the mapping value, the spherical mapping can be regarded as a function, and the formula is as follows:

[0086]

[0087] where the output is the spherical data value on the surface of the sphere , the center point is the distance between the center point and the farthest intersection point ; and the ranges of and are (0, 2pi] and (0, pi] respectively, and the sampling interval value is a uniform distribution obtained by equidistant sampling. The sampling interval value is a self-defined hyperparameter, and the larger the sampling interval value, the higher the resolution and fineness, but the data storage cost and the calculation cost are also higher. The specific interval value is not limited in the present application.

[0088] Step 3: The supervised branch and the self-supervised branch are trained cooperatively. That is, the spherical data is input into a double-branch task cooperative training fault diagnosis model, one branch of the model uses limited labeled data to complete the main task of fault diagnosis, and the other branch generates rotation samples and real rotation label pairs through three-dimensional rotation to complete the designed rotation recognition task, and learns the feature representation beneficial to the main task from the task.

[0089] Wherein, the supervised branch and the self-supervised branch use the same feature extraction module, and the outputs are guided to different classifiers. The self-supervised branch of the embodiment is provided with a feature extraction module and a linear layer 1, and the supervised branch is provided with a feature extraction module and a linear layer 2, and the linear layer 1 and the linear layer 2 are respectively a classifier for rotation type prediction and fault category prediction. The feature extraction module introduces a spherical convolution module, and the spherical convolution module is composed of a plurality of convolution layers connected in series, each convolution layer is stacked by a spherical convolution layer, a nonlinear layer and a pooling layer, and the main function of the spherical convolution layer is to extract features from the input spherical data.

[0090] The spherical convolution module proposed in the embodiment has four layers, each layer includes spherical convolution, nonlinearity and pooling. The feature maps output by each convolution layer in the embodiment are 16, 32, 64 and 128 respectively. The feature vector output by the last convolution layer will be used as the input data of the classifier to complete different tasks afterwards.

[0091] The spherical convolution layer is defined as a convolution on SO(3), which is the set of all rotations in three-dimensional Euclidean space, represented as a 3x3 orthogonal matrix with determinant 1. The spherical convolution takes into account the algebraic properties of SO(3). In the application, the spherical convolution is also equivariant to three-dimensional rotations, i.e., the group elements and group operations of SO(3). In addition, the spherical convolution also satisfies the Fourier theorem. This means that the spherical convolution can be implemented in the frequency domain. The Spherical Fourier Transform (SFT) and its inverse transform (ISFT) can be obtained, as shown in the following formula:

[0092]

[0093] (6)

[0094] where b is the bandwidth of f, and is the spherical harmonic function of order l and degree m The point product calculation formula of the spherical signal in the frequency domain is as follows:

[0095]

[0096] where, is the spherical Fourier transform coefficient of order l and degree m is the original signal defined on the two-dimensional spherical surface x is a point on the spherical surface, and the filter is the convolution kernel of the neural network, which is a 3x3 two-dimensional matrix, is the filter corresponding to the spherical Fourier transform coefficient of order l and degree m . In short, the implementation of spherical convolution first maps the original spherical signal and the filter to the frequency domain through the spherical Fourier transform, then performs the point product operation in the frequency domain, and finally maps back to the original space through the spherical Fourier inverse transform.

[0097] The spherical mapping obtained by the spherical convolution layer is input into the spherical convolution layer, and the feature extraction is performed. In addition, the technical scheme of the present application introduces a dynamic weighting mechanism based on uncertainty, combines the supervised task loss and the self-supervised task loss, balances the training weights of the double branches through the regularization term, and optimizes the model parameters.

[0098] In addition, the technical scheme of the present application introduces a dynamic weighting mechanism based on uncertainty, combines the supervised task loss and the self-supervised task loss, balances the training weights of the double branches through the regularization term, and optimizes the model parameters.

[0099] In addition, the technical scheme of the present application introduces a dynamic weighting mechanism based on uncertainty, combines the supervised task loss and the self-supervised task loss, balances the training weights of the double branches through the regularization term, and optimizes the model parameters.

[0100] ​​Wherein, the original spherical data and the corresponding self-supervised label, through the self-supervised task, make the model learn the feature representation beneficial to the main task. The learning process is completed by itself. Through the designed architecture, the self-supervised task can assist the model to learn the useful data representation autonomously. In the embodiment, it is assumed that the rotation spherical data X and its corresponding pseudo label Y, the two tasks are jointly trained using the same feature extraction module, and the output results are guided to different classifiers. The goal of the self-supervised task is to gradually reduce the value of the loss function L and train the model . It can be represented by the formula:

[0101] (8)

[0102] In the formula, N is the number of samples on the self-supervised branch, is the set of trainable parameters of the model, ) is the loss function of a single sample.

[0103] By using the pseudo label generated autonomously and continuously training, the model can learn the useful feature representation without relying on external artificial data labels.

[0104] The present application is to use the supervised branch and the self-supervised branch to train cooperatively, which is essentially a multi-task learning problem. The supervised branch completes the main task, and the goal is to complete the motor fault diagnosis task in the limited labeled sample set; and the self-supervised branch completes the auxiliary main task, i.e. the rotation recognition task, and the goal is to learn the feature representation that is helpful to improve the accuracy of the main task through the rotation recognition task. How to balance the two tasks in the training phase and avoid focusing on one task and ignoring the other task will be described in detail below. Assuming that the loss value of the self-supervised task is , the loss value of the supervised task is , and the weighted sum of each loss is taken as the total loss. The total loss value is , and are the weight values of the two tasks, respectively. The size of the weight value determines the proportion of the loss value of each task to the total loss value, and further determines the importance of each task.

[0105] In some embodiments, the total loss value can be represented as:

[0106] (9)

[0107] In other embodiments, considering that manually setting the weight combination and continuously experimenting to verify the best weight combination is inefficient, therefore, the present application preferably introduces a dynamic weighting mechanism based on uncertainty in the setting of the loss function, to more perfectly balance the weights of the two tasks in the training process.

[0108] i.e. introducing a regularization term in the total loss , The expression of the total loss is represented as follows:

[0109] (10)

[0110] The uncertainty parameter δ is introduced again, which is inversely proportional to the-2 power of the weight W, i.e.:

[0111] (11)

[0112] wherein the uncertainty parameter δ is a temperature parameter in the Gibbs distribution, which determines the flatness of the final probability distribution. The fluctuation of δ will affect the learning difficulty of each loss and reflect the uncertainty of a specific task. Then, through simplification, the regularization term can also be expressed by calculation as follows:

[0113] (12)

[0114] Then, the total loss value with the dynamic weighting mechanism can be represented as follows:

[0115] (13)

[0116] The uncertainty parameter δ is also a learnable parameter. In the present embodiment, the model is updated at a lower learning rate during training, which is represented by the following formula:

[0117] (14)

[0118] In summary, the feature extraction modules of the self-supervised branch and the supervised branch in the technical scheme of the present application have the same structure, and are also difficult to avoid, and the model parameters are adjusted synchronously under the guidance of the total loss function.

[0119] After comparing the experimental results of the model with the self-supervised branch and the model without the self-supervised branch, it can be obviously found that under the condition of two sample quantities, the test accuracy of the model without the self-supervised branch is obviously lower than that of the model with the self-supervised branch, which shows that the self-supervised branch is of great significance in the motor fault diagnosis method based on multi-dimensional data processing and self-supervised learning.

[0120] Step 4: inputting the original spherical data corresponding to the three-axis vibration time sequence signal of the motor to be diagnosed into the trained supervised branch to obtain the fault diagnosis classification result.

[0121] It should be understood that, according to the foregoing steps, the three-axis vibration time sequence signals of the motor to be diagnosed are first converted into three-dimensional point cloud data, then spherical mapping is performed on the three-dimensional point cloud data to obtain spherical data, and then the trained supervised branch is input, so that the fault diagnosis classification result can be obtained.

[0122] Embodiment 2

[0123] The embodiment of the present application also provides a system based on the above motor fault diagnosis method, comprising a data acquisition module, a data enhancement module, a spherical mapping module, a cooperative training module and a diagnosis module connected in sequence or with each other.

[0124] The data acquisition module is configured to acquire three-axis vibration time sequence signals in various motor states through a three-axis vibration sensor, and to construct three-dimensional point cloud data of the motor by using the three-axis vibration time sequence signals, wherein the motor state is a real fault label.

[0125] The data enhancement and spherical mapping module is configured to perform a three-dimensional rotation operation on the three-dimensional point cloud data to obtain rotated point cloud data and a real rotation label, and to perform spherical mapping on the rotated point cloud data to obtain rotated spherical data; and to perform spherical mapping on the three-dimensional point cloud data of step S101 to obtain original spherical data.

[0126] The cooperative training module is configured to cooperatively train a supervised branch and a self-supervised branch, wherein the supervised branch and the self-supervised branch adopt the same feature extraction module, and the outputs thereof are directed to different classifiers, in the training stage, the rotated spherical data and the real rotation label are input into the self-supervised branch, and the original spherical data and the real fault label are input into the supervised branch.

[0127] The diagnosis module is configured to input the original spherical data corresponding to the three-axis vibration time sequence signals of the motor to be diagnosed into the trained supervised branch to obtain a fault diagnosis classification result.

[0128] It should also be understood that the specific implementation process of each module can refer to the above method content, and the present application will not be repeated here, and the division of the above functional modules is only for example, in some embodiments, part of the functional modules can be combined, part of the functional modules can be split, and each functional module can be implemented in a software manner or hardware or a combination of software and hardware, wherein the software and hardware devices include but are not limited to general-purpose computer devices, programmable gate arrays, digital signal processors, microprocessors and their corresponding programming or burning software.

[0129] Embodiment 3

[0130] The embodiment of the present application also provides a computer device, comprising one or more processors and a memory storing a computer program; wherein the processor invokes the computer program to implement the steps of the above motor fault diagnosis method.

[0131] Specific implementation:

[0132] S101, data acquisition, acquiring three-axis vibration time series signals in various motor states through a three-axis vibration sensor, and constructing three-dimensional point cloud data of the motor by using the three-axis vibration time series signals, wherein the motor state is a real fault label;

[0133] S102, data enhancement and spherical mapping, performing a three-dimensional rotation operation on the three-dimensional point cloud data of step S101 to obtain rotated point cloud data and a real rotation label, and then performing spherical mapping on the rotated point cloud data to obtain rotated spherical data; performing spherical mapping on the three-dimensional point cloud data of step S101 to obtain original spherical data;

[0134] S103, cooperative training of a supervised branch and a self-supervised branch, the supervised branch and the self-supervised branch use the same feature extraction module, and the outputs are directed to different classifiers, in the training phase, the rotated spherical data and the real rotation label are input into the self-supervised branch, and the original spherical data and the real fault label are input into the supervised branch;

[0135] S104, inputting the original spherical data corresponding to the three-axis vibration time series signals of the motor to be diagnosed into the trained supervised branch to obtain a fault diagnosis classification result.

[0136] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The processor is used to execute related programs to implement the technical solutions provided by the embodiments of the application.

[0137] The memory can be implemented in the form of read only memory (ROM), static storage device, dynamic storage device or random access memory (RAM), etc. The memory can store an operating system and other application programs, and when the technical solutions provided by the embodiments of the present application are implemented by software or firmware, the related program codes are saved in the memory and called and executed by the processor to execute the algorithm program of the motor fault diagnosis method of the embodiments of the application.

[0138] Embodiment 4

[0139] The embodiment of the present application also provides a computer readable storage medium storing a computer program, wherein the computer program is called by a processor to implement the steps of the motor fault diagnosis method.

[0140] Specific implementation:

[0141] S101, data acquisition, three-axis vibration time series signals in various motor states are acquired through a three-axis vibration sensor, and three-dimensional point cloud data of the motor is constructed by using the three-axis vibration time series signals, wherein the motor state is a real fault label;

[0142] S102, data enhancement and spherical mapping, three-dimensional rotation operation is performed on the three-dimensional point cloud data of step S101 to obtain rotated point cloud data and a real rotation label, and spherical mapping is performed on the rotated point cloud data to obtain rotated spherical data; spherical mapping is performed on the three-dimensional point cloud data of step S101 to obtain original spherical data;

[0143] S103, cooperative training of the supervised branch and the self-supervised branch, the supervised branch and the self-supervised branch use the same feature extraction module, and the outputs are guided to different classifiers, in the network training stage, the rotated spherical data and the real rotation label are input into the self-supervised branch, and the original spherical data and the real fault label are input into the supervised branch;

[0144] S104, the original spherical data corresponding to the three-axis vibration time series signals of the motor to be diagnosed is input into the trained supervised branch to obtain a fault diagnosis classification result.

[0145] The specific implementation process of each step is described in the foregoing diagnosis method embodiment.

[0146] The readable storage medium is a computer readable storage medium, which can be an internal storage unit of the software and hardware device in any of the foregoing embodiments, for example, a hard disk or a memory of the controller. The readable storage medium can also be an external storage device of the controller, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card and the like equipped on the controller. Further, the readable storage medium can include both the internal storage unit and the external storage device of the controller. The readable storage medium is used to store the computer program and other programs and data required by the controller. The readable storage medium can also be used to temporarily store data that has been output or will be output.

[0147] Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art, or the whole or part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in various embodiments of the present application. The aforementioned readable storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0148] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer usable program code. The present application is produced by referring to the flowcharts and the instructions executed by the processor of the method, device (system), and computer program product according to the embodiments of the present application to realize the functions specified in one or more flows of the flowcharts and / or one or more blocks of the block diagrams. These computer program instructions can also be stored in a computer readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer readable memory produce a manufactured product including instruction devices that realize the functions specified in one or more flows of the flowcharts and / or one or more blocks of the block diagrams. These computer program instructions can also be loaded into a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer implemented process, so that the instructions executed on the computer or other programmable device provide steps for realizing the functions specified in one or more flows of the flowcharts and / or one or more blocks of the block diagrams.

[0149] It should be emphasized that the examples described in the present application are illustrative rather than restrictive, and therefore the present application is not limited to the examples described in the specific embodiments, and any other embodiments derived by those skilled in the art from the technical solutions of the present application without departing from the purpose and scope of the present application, whether modified or replaced, also belong to the protection scope of the present application.

Claims

1. A motor fault diagnosis method based on multi-dimensional data under small sample condition, characterized in that: The method comprises the following steps: S101, data acquisition, acquiring three-axis vibration time series signals of various motor states through a three-axis vibration sensor, and constructing three-dimensional point cloud data of the motor by using the three-axis vibration time series signals, wherein the motor state is a real fault label; S102, data enhancement and spherical mapping, performing three-dimensional rotation operation on the three-dimensional point cloud data of step S101 to obtain rotated point cloud data and a real rotation label, and then performing spherical mapping on the rotated point cloud data to obtain rotated spherical data; performing spherical mapping on the three-dimensional point cloud data of step S101 to obtain original spherical data; S103, cooperative training of a supervised branch and a self-supervised branch, wherein the supervised branch and the self-supervised branch adopt the same feature extraction module, and the outputs are guided to different classifiers, in the training stage, the rotated spherical data and the real rotation label are input into the self-supervised branch, and the original spherical data and the real fault label are input into the supervised branch; S104, inputting the original spherical data corresponding to the three-axis vibration time series signals of the motor to be diagnosed into the trained supervised branch to obtain a fault diagnosis classification result.

2. The electric machine fault diagnostic method according to claim 1, characterized in that: The total loss value in the network training stage is the weighted sum of the supervised task and the self-supervised task, and a dynamic weighting mechanism is introduced: ； wherein, is the total loss value, is the loss value of task i, i takes values 1 and 2, i.e. corresponding is the loss value of the self-supervised task, is the loss value of the supervised task; is the uncertainty parameter of task i, network training process, the following update is made: ； wherein denotes the updated uncertainty parameter η is a proportionality constant.

3. The motor fault diagnostic method according to claim 1, characterized by: The self-supervised branch is provided with a feature extraction module and a linear layer 1, and the supervised branch is provided with a feature extraction module and a linear layer 2, and the linear layer 1 and the linear layer 2 are respectively classifiers for rotation type prediction and fault category prediction; The feature extraction module is a spherical convolution module, and the spherical convolution module is composed of a plurality of convolution layers connected in series, and each convolution layer is stacked by a spherical convolution layer, a nonlinear layer and a pooling layer.

4. The motor fault diagnostic method according to claim 3, characterized by: The processing process of the spherical convolution layer is as follows: First, the current spherical signal and the filter are mapped to the spectral domain by a spherical Fourier transform ; ； ； wherein, is a signal corresponding to an order of spherical Fourier transform coefficients, is a spherical signal on a two-dimensional unit sphere x is a point on the sphere, is an order of spherical harmonic functions; a filter is a neural network convolution kernel for a spherical convolution layer, corresponding to an order of spherical Fourier transform coefficients; Then, the dot product operation is completed in the spectral domain to obtain ; ； Finally, the original space is mapped back through inverse spherical Fourier transform; ； wherein, is the result of the current spherical convolution layer, and b is the bandwidth.

5. The motor fault diagnostic method according to claim 1, characterized by: The process of spherical mapping on the rotated point cloud data in step S102 or the process of spherical mapping on the three-dimensional point cloud data is as follows: Mapping rotational point cloud data or three-dimensional point cloud data to a two-dimensional unit sphere i.e. to get a geometric representation in a spherical coordinate system with radius = 1; wherein the center point of the rotated point cloud data or the three-dimensional point cloud data is calculated first ; According to the azimuth angle And the polar angle From the center point Draw a ray, by calculating the distance of the farthest intersection point of the ray and the rotation point cloud data or three-dimensional point cloud data The distance The distance value of the farthest intersection point As a mapping value, spherical mapping can be regarded as a function, as shown in the following formula: ； where the output is the spherical data value on the map value, is the center point and the distance between the center point and the farthest intersection point The value ranges of and are (0, 2π] and (0, π] respectively, and are uniformly distributed by equidistant sampling.​​ 6. The motor fault diagnostic method according to claim 1, characterized by: The three-dimensional rotation operation on the three-dimensional point cloud data in step S102 is a random three-dimensional rotation operation, that is, a randomly generated rotation direction and angle are obtained, and then the real rotation label and the rotated point cloud data are obtained; The rotation direction is rotation around the x-axis, the y-axis and the z-axis.

7. The motor fault diagnostic method according to claim 1, characterized by: The process of constructing the three-dimensional point cloud data in step S101 is as follows: first, the three-axis vibration time series signals collected are normalized to the [0, 1] interval by using the maximum and minimum value normalization method, and then converted into three-dimensional point cloud; then, random rotation transformation and adding random Gaussian noise are performed for the purpose of adding interference; The calculation formula of the random rotation is as follows: ； ； wherein, is the data after random rotation transformation, R is a rotation matrix, p is the point cloud data of the sample, , , are three-dimensional rotation matrices around the x-axis, the y-axis and the z-axis respectively; a, b, c are rotation angles of the rotation matrix , , is sampled from a uniform distribution , is a preset maximum rotation angle.

8. A system for diagnosing a fault of an electric machine based on the method according to any one of claims 1 to 7, characterized in that: It comprises: A data acquisition module is configured to acquire three-axis vibration time series signals of various motor states through a three-axis vibration sensor, and construct three-dimensional point cloud data of the motor by using the three-axis vibration time series signals, wherein the motor state is a real fault label; A data enhancement and spherical mapping module is configured to perform three-dimensional rotation operation on the three-dimensional point cloud data to obtain rotated point cloud data and a real rotation label, and then perform spherical mapping on the rotated point cloud data to obtain rotated spherical data; perform spherical mapping on the three-dimensional point cloud data of step S101 to obtain original spherical data; The cooperative training module is used for cooperative training of the supervised branch and the self-supervised branch, the supervised branch and the self-supervised branch adopt the same feature extraction module and guide the outputs to different classifiers, in the training stage, the rotation spherical surface data and the real rotation label are input into the self-supervised branch, the original spherical surface data and the real fault label are input into the supervised branch; The diagnosis module is used for inputting the original spherical surface data corresponding to the three-axis vibration time sequence signal of the motor to be diagnosed into the trained supervised branch to obtain a fault diagnosis classification result.

9. A computer device, characterized by: Comprise: One or more processors; A memory storing a computer program; Wherein, the processor calls the computer program to realize: The steps of the motor fault diagnosis method in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: A memory storing a computer program, the computer program is called by the processor to realize: The steps of the motor fault diagnosis method in any one of claims 1-7.

Images