Motor bearing cross-condition fault diagnosis method based on riemann characteristic sequence and meta learning
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HEFEI UNIV OF TECH
- Filing Date
- 2026-02-12
- Publication Date
- 2026-06-02
Smart Images
Figure CN122132957A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of motor fault diagnosis and intelligent operation and maintenance technology, specifically a method for cross-condition fault diagnosis of motor bearings based on Riemann feature sequences and meta-learning, as well as a cross-condition fault diagnosis system for motor bearings using this method and a computer-readable storage medium. Background Technology
[0002] Permanent magnet synchronous motors (PMSMs) are key components in new energy vehicles, high-end manufacturing equipment, and other fields, and their reliability directly affects the safety and performance of the system. Because motors operate under variable and harsh conditions for extended periods, their internal components are prone to fatigue damage, leading to potential failures in critical components such as bearings and stators. Failure to diagnose faults promptly and accurately can cause a chain reaction, resulting in equipment downtime or even serious safety accidents. Therefore, accurate and efficient diagnosis of motor faults has significant industrial importance and application value.
[0003] Most existing methods for diagnosing motor bearing faults are based on the feature analysis of vibration or current signals. However, these methods face two major challenges in practical applications: First, fault signals are often strongly modulated and interfered with by changes in operating conditions such as motor speed and load, resulting in the extracted fault features (such as harmonic components and energy indices) lacking robustness across operating conditions and having low feature accuracy. Second, in the initial stage of a fault, the feature signals are weak and easily obscured by operating noise and background harmonics, which seriously affects the accuracy of fault diagnosis and the generalization ability of the model.
[0004] Current signal processing methods, such as Fourier transform and wavelet transform, struggle to effectively separate target fault information from operating condition-related interference during feature extraction, resulting in extracted features that are sensitive to changes in operating conditions. Traditional deep learning models, such as convolutional neural networks (CNNs), while possessing powerful feature learning capabilities, are essentially trained using empirical risk minimization (ERM). This leads to models performing well on the original operating conditions but experiencing a significant drop in diagnostic performance when faced with unseen target operating conditions. Furthermore, adapting the model to new operating conditions often requires substantial computational resources and time for retraining or fine-tuning, resulting in low adaptation efficiency.
[0005] In summary, current technologies for motor bearing fault diagnosis across various operating conditions urgently require an innovative technique to overcome the shortcomings of insufficient feature robustness and low model generalization efficiency. Summary of the Invention
[0006] To address the technical problems existing in the prior art, this invention provides a cross-condition fault diagnosis method for motor bearings based on Riemann feature sequences and meta-learning. It combines the feature invariance of Riemann geometry with the rapid adaptability of the model-independent meta-learning (MAML) framework, thereby ensuring diagnostic accuracy while enabling the rapid and efficient deployment of the base learner on any new operating condition.
[0007] To achieve the above objectives, the present invention provides the following technical solution: This invention discloses a method for cross-condition fault diagnosis of motor bearings based on Riemann feature sequences and meta-learning, comprising: S1. Collect multi-channel vibration signals of motor bearings under different working conditions and segment them using the sliding window technique. Calculate the symmetric positive definite covariance matrix of the multi-channel vibration signals within each window. Project the covariance matrix into Riemann tangent space eigenvectors through the logarithmic mapping from the Riemann manifold to the tangent space. Concatenate the Riemann tangent space eigenvectors of the continuous time windows to construct a Riemann feature sequence for the meta-learning network. S2. Define each source domain condition in the open-source dataset as a meta-training task, and construct an attention residual network as the base learner of the meta-learning framework; S3. Perform two-layer meta-learning training on the base learner, including rapid adaptation of working conditions in the inner loop and meta-parameter update in the outer loop. After training is terminated, the optimal meta-initialization parameters are obtained. S4. Collect the real-time vibration signal of the motor bearing, and extract its Riemann feature sequence as the target working condition according to step S1; use the sample of the target working condition to perform rapid adaptation of the working condition of the inner cycle based on the optimal element initialization parameters to obtain the final diagnostic parameters, and apply the final diagnostic parameters to the real-time fault diagnosis of the target working condition data.
[0008] As a further improvement to the above scheme, step S1 specifically includes: S11. The motor bearing monitoring system is equipped with... The multidimensional discrete-time vibration signal sequence acquired by the vibration sensor channels is represented as a matrix. ,in This represents the total number of sampling points; It is the set of real numbers; S12. Let the length of the sliding window be... Step size is Then each sample, i.e. the first... The signal segments within a time window are represented as a matrix. ;in Indicates the first The first time window The multidimensional vibration vector of each sampling point ; S13. Window signals After centering, the covariance matrix of the samples is calculated, expressed by the following formula:
[0009] In the formula, for The result after centralized processing; For the first The covariance matrix of each time window; the superscript T is the transpose sign; S14. Regarding the covariance matrix Introducing a regularization perturbation term, the formula is as follows:
[0010] In the formula, This is the regularization perturbation term; for Identity matrix; introducing regularization perturbation term It satisfies the following two properties: symmetry, that is Positive definiteness, that is, for any non-zero vector ,satisfy ; Belongs to the C-order symmetric matrix, denote all The set space formed by symmetric positive definite matrices is At this point, the data for each time window is modeled as a manifold. A point on; S15. Using the matrix logarithmic mapping to transform the manifold Points on Projected onto the identity matrix In the tangent space at the location; where, according to the spectral decomposition theorem, Decomposed into:
[0011] In the formula, Let be an orthogonal eigenvector matrix, satisfying The superscript T is the transpose symbol; It is an eigenvalue diagonal matrix. This is the standard function used to create diagonal matrices. For matrix The 1 eigenvalue, The formula for the matrix-logarithmic mapping is:
[0012] In the formula, The matrix obtained after the matrix logarithmic mapping lies in the tangent space. In this context, the tangent space is isomorphic to... Euclidean space; S16. Using semi-vectorization operators Extracting the matrix The upper triangular independent elements are used to generate the Riemann eigenvector of the k-th time window. Thus, for a segment containing continuous Monitoring data from several time windows are used to concatenate the extracted Riemann feature vectors to construct a Riemann feature sequence tensor. , .
[0013] As a further improvement to the above scheme, The calculation formula is:
[0014] In the formula, For window signals The mean vector within, ; for A column vector of all 1s.
[0015] As a further improvement to the above scheme, in step S2, the attention residual network includes a local feature extractor, a cross-temporal feature enhancer, a fixed-dimensional transformer, and a classification decision layer connected in sequence; the local feature extractor consists of stacked one-dimensional convolutional residual blocks, used to extract local temporal features from the Riemann feature sequence. The cross-temporal feature enhancer employs a self-attention mechanism to... Perform weighted aggregation and output the attention-enhanced feature sequence. The fixed-dimensional converter is... By averaging over the time dimension, a fixed-length feature vector is obtained. The classification decision layer uses a fully connected layer and a Softmax function to... The probability distribution mapped to fault categories.
[0016] As a further improvement to the above scheme, in step S2, each source domain condition in the open-source dataset is defined as an independent meta-training task. Riemann feature data for each meta-training task Randomly partitioned into support sets for inner loop training and query set used for outer loop evaluation Base learner The parameter set is denoted as .
[0017] As a further improvement to the above scheme, the bi-layer meta-learning training in step S3 specifically includes: S31. In the inner loop, in the support set Calculate the initial parameters Adaptation loss and utilize internal learning rate conduct Gradient descent is used to simulate the rapid adaptation of the base learner to the current operating conditions, resulting in the adapted parameters. ; S32. In the outer loop, use the adapted parameters. In query set Calculate the element loss Evaluate generalization ability; calculate meta-loss with respect to initial parameters. meta gradient And combined with external learning rate Update initial parameters ; Repeat steps S31 and S32 until the preset maximum number of meta-iterations is reached or Convergence; after training is complete, the final parameters obtained are the optimal meta-initialization parameters. .
[0018] As a further improvement to the above scheme, in step S31, The calculation formula is:
[0019] In the formula, This represents the input Riemann feature sequence tensor. Indicate the fault category; This is the loss function, used to measure the difference between the predicted probability distribution and the actual fault category. The parameter is At that time, the base learner learns the input The fault prediction probability distribution output; The calculation formula is: ,at this time Is to make Minimize the local optimum. Represents the task index of the meta-task, and ; For gradient operators, it means that with respect to initial parameters Find the partial derivative.
[0020] As a further improvement to the above scheme, in step S32, The calculation formula is:
[0021] In the formula, For the batch of meta-tasks participating in the current meta-training iteration; To use the adapted parameters In query set The evaluation loss function calculated above, The adaptation parameters are obtained by the base learner through rapid adaptation in the inner loop. The following categories are output; Among them, the meta-gradient The formula for calculating the adaptive steps flowing through the inner loop using the chain rule is as follows:
[0022] In the formula, For the first Individual tasks adapt parameters The generalization loss value is as follows; Meta-loss for adapted parameters The gradient vector; To adapt to post-parameters For initial parameters The Jacobian matrix; initial parameters The update formula is: .
[0023] The present invention also discloses a fault diagnosis system for motor bearings under various operating conditions, including a data acquisition module, a processor, and a memory for storing a computer program; the data acquisition module is used to acquire vibration signals of the motor bearing; when the computer program is executed by the processor, it implements the steps of the fault diagnosis method for motor bearings under various operating conditions based on Riemann feature sequences and meta-learning as described above.
[0024] The present invention also discloses a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the motor bearing cross-operating condition fault diagnosis method based on Riemann feature sequences and meta-learning as described above.
[0025] Compared with the prior art, the beneficial effects of the present invention are: This invention, by setting up a feature sequence extraction mechanism based on Riemannian manifold projection, introducing a residual convolutional base learner with a self-attention mechanism, and a model-independent two-layer meta-learning training architecture, can decouple the operating condition modulation interference caused by changes in motor speed and load by utilizing the affine invariance of Riemann space. Based on the learned optimal meta-initialization parameters, the model has extremely strong operating condition transfer and knowledge generalization capabilities. Thus, while ensuring the accuracy of initial weak fault diagnosis of motor bearings, it can achieve rapid and efficient deployment of the base learner and real-time fault monitoring under any unknown new operating conditions. Attached Figure Description
[0026] Figure 1This is a flowchart of the motor bearing cross-condition fault diagnosis method based on Riemann feature sequences and meta-learning in Embodiment 1 of the present invention.
[0027] Figure 2 This is an architecture diagram of the motor bearing cross-operating condition fault diagnosis system in Embodiment 2 of the present invention. Detailed Implementation
[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0029] Example 1
[0030] Please see Figure 1 This embodiment provides a method for fault diagnosis of motor bearings across operating conditions based on Riemann feature sequences and meta-learning, including steps S1 to S4.
[0031] S1. Establish a robust Riemannian feature sequence for the operating conditions: Multi-channel vibration signals of motor bearings under different operating conditions are collected and segmented using the sliding window technique. The symmetric positive definite covariance matrix of the multi-channel vibration signal is calculated within each window. The covariance matrix is projected into Riemann tangent space eigenvectors through the logarithmic mapping from the Riemann manifold to the tangent space. The Riemann tangent space eigenvectors of the continuous time windows are concatenated to construct the Riemann feature sequence for the meta-learning network.
[0032] Step S1 specifically includes S11 to S16.
[0033] S11. The motor bearing monitoring system is equipped with... The multidimensional discrete-time vibration signal sequence acquired by the vibration sensor channels is represented as a matrix. ,in This represents the total number of sampling points; It is the set of real numbers.
[0034] In this embodiment, These correspond to the acceleration signals of the X-axis, Y-axis, and Z-axis, respectively.
[0035] S12. To capture the local time-varying characteristics of the signal and meet the sequence input requirements of the deep learning model, a sliding window technique is used. Perform segmentation. Let the length of the sliding window be... Step size is The sliding window extracts a subsequence of fixed length W from a long vibration signal sequence. Then, this "window" moves forward by S points (step size) to extract the next subsequence, and so on, until the entire signal has been traversed.
[0036] Each sample is the first The signal segments within a time window are represented as a matrix. ;in Indicates the first The first time window The multidimensional vibration vector of each sampling point .
[0037] S13. Window signals After centering, the covariance matrix of the samples is calculated, expressed by the following formula:
[0038]
[0039] In the formula, for The result after centralized processing; For window signals The mean vector within, ; for A column vector of all 1s; For the first The covariance matrix of each time window; the superscript T is the transpose symbol.
[0040] S14. Regarding the covariance matrix Introducing a regularization perturbation term, the formula is as follows:
[0041] In the formula, This is the regularization perturbation term; for Identity matrix; introducing regularization perturbation term It satisfies the following two properties: symmetry, that is Positive definiteness, that is, for any non-zero vector ,satisfy ; Belongs to the C-order symmetric matrix, denote all The set space formed by symmetric positive definite matrices is At this point, the data for each time window is modeled as a manifold. A point on the surface.
[0042] It should be noted that in actual engineering, in order to prevent the covariance matrix from becoming singular (i.e., the eigenvalues appearing zero or negative) due to extremely low sensor noise or strong linear correlation between signal channels, and thus to ensure that the matrix is strictly positive definite, a small regularization perturbation term is introduced.
[0043] S15. Using the matrix logarithmic mapping to transform the manifold Points on Projected onto the identity matrix In the tangent space at the location; where, according to the spectral decomposition theorem, Decomposed into:
[0044] In the formula, Let be an orthogonal eigenvector matrix, satisfying The superscript T is the transpose symbol; It is an eigenvalue diagonal matrix. This is the standard function used to create diagonal matrices. For matrix The 1 eigenvalue, The formula for the matrix-logarithmic mapping is:
[0045] In the formula, The matrix obtained after the matrix logarithmic mapping lies in the tangent space. In this context, the tangent space is isomorphic to... The Euclidean space is used. In this space, the Euclidean distance between matrices is equivalent to the affine invariant Riemannian metric on the manifold. This transformation effectively decouples the overall signal amplitude variation caused by speed fluctuations, endowing it with affine invariance, i.e., adaptability to operating conditions.
[0046] Riemannian manifold Since Euclidean space is a nonlinear curved space, direct applications of Euclidean space operations (such as Euclidean distance and direct input to neural networks) are not suitable. In order to utilize the linear properties of Euclidean space, this invention uses a matrix logarithmic map to project points on the manifold into the tangent space at the identity matrix.
[0047] Although It lies in a linear space, but it is still a linear space. It is in matrix form. To adapt to the input requirements of deep learning models, it needs to be converted into a vector.
[0048] S16. Due to Inherited Due to its symmetry, its independent elements are located only in the upper triangular region (including the diagonal), thus allowing the use of a semi-vectorization operator. Extracting the matrix The upper triangular independent elements are used to generate the Riemann eigenvector of the k-th time window. .
[0049] In this embodiment, for a triaxial signal with C=3, The form is:
[0050] The extracted Riemann eigenvector of the k-th time window for:
[0051] Note: This reflects the logarithm of the energy of each axis vibration. It reflects the logarithm of the correlation between the axes.
[0052] Finally, for a segment containing continuous Monitoring data from several time windows are used to concatenate the extracted Riemann feature vectors to construct a Riemann feature sequence tensor. , .
[0053] S2. Define the meta-learning task and base learner: Each source domain condition in the open-source dataset is defined as a meta-training task, and a residual convolutional network with attention mechanism (R-MetaNet) is used as the base learner. This network consists of a one-dimensional convolutional residual module, a self-attention mechanism, and a fully connected classification layer. It is all weight in the network and all biases The set of:
[0054] Final feature vector The formula for calculating (feature extraction) is:
[0055] Local Feature Extractor: Composed of stacked one-dimensional convolutional residual blocks, it is responsible for extracting multi-level, local temporal features from Riemann sequences. Its output is a deep feature map. .
[0056] Cross-temporal feature enhancer. For The sequences are weighted and aggregated, enabling the network to adaptively focus on the most critical time steps for fault detection within the sequence, thus enhancing the model's robustness to changes in operating conditions. Its output is the attention-enhanced feature sequence. .
[0057] Fixed-dimensional converter. By averaging over the time dimension, the variable-length sequence features are converted into a fixed-length feature vector.
[0058] Calculation (categorized output)
[0059] FC: Classification Decision Layer. Mapping to the fault category space yields the unnormalized log odds. .
[0060]
[0061] : Probability normalization function. It normalizes a vector. Transform the probability distribution into a vector P such that all output values are between [0,1] and sum to 1. The formula for calculating each element in the probability distribution vector P is:
[0062] In the formula, Indicates the total number of faults; In the log-odds vector output by the classification decision layer, the value corresponding to the first... The original score for each fault category; e The natural constant is used; the final fault prediction probability distribution vector P is the network output. Each element in the vector Represents input signal Belongs to the The probability of each type of failure.
[0063] Each source domain condition (e.g., 20Hz, 30Hz, ...) in the open-source dataset is defined as an independent meta-training task. Riemann feature data for each meta-training task The support set was randomly assigned to be used for inner loop training (simulating fast adaptation). and the query set used for outer loop evaluation (measuring generalization performance) Base learner The parameter set is denoted as .
[0064] S3.MAML inner and outer loops: The base learner is trained using a two-layer meta-learning method, which includes rapid adaptation of operating conditions in the inner loop and meta-parameter updates in the outer loop. The optimal meta-initialization parameters are obtained after training terminates.
[0065] The two-layer meta-learning training specifically includes steps S31 to S32.
[0066] S31. In the inner loop, in the support set Calculate the initial parameters Adaptation loss and utilize internal learning rate (Set a smaller value) to perform Gradient descent is used to simulate the rapid adaptation of the base learner to the current operating conditions, resulting in the adapted parameters. .
[0067] This step simulates R-MetaNet under a single operating condition. The purpose of this rapid fine-tuning process is to calculate the adapted parameters. .
[0068] Cross-entropy loss is used, and the calculation formula is as follows:
[0069] In the formula, This represents the input Riemann feature sequence tensor. Indicate the fault category; This is the loss function, used to measure the difference between the predicted probability distribution and the actual fault category. The parameter is At that time, the base learner learns the input The fault prediction probability distribution is output.
[0070] The calculation formula is: ,at this time Is to make Minimize the local optimum. Represents the task index of the meta-task, and ; For gradient operators, it means that with respect to initial parameters Find the partial derivative.
[0071] S32. In the outer loop, use the adapted parameters. In query set Calculate the element loss Evaluate generalization ability; calculate meta-loss with respect to initial parameters. meta gradient And combined with external learning rate Update initial parameters .
[0072] This step evaluates the initial parameters. The quality of, and based on its generalization ability Update the performance on the screen.
[0073] Yuan loss It measures the initial parameters The generalization performance metric is calculated using the following formula:
[0074] In the formula, The meta-task batch (a set of diagnostic tasks representing different motor operating conditions, randomly selected in the current global parameter update step) is the batch of meta-tasks participating in the current meta-training iteration. To use the adapted parameters In query set The evaluation loss function calculated above, The adaptation parameters are obtained by the base learner through rapid adaptation in the inner loop. The following categories are output; Among them, the meta-gradient The formula for calculating the adaptive steps flowing through the inner loop using the chain rule is as follows:
[0075] In the formula, For the first Individual tasks adapt parameters The generalization loss value is as follows; Meta-loss for adapted parameters The gradient vector; To adapt to post-parameters For initial parameters The Jacobian matrix; initial parameters The update formula is: .
[0076] Repeat steps S31 and S32 until the preset maximum number of meta-iterations is reached or Convergence; after training is complete, the final parameters obtained are the optimal meta-initialization parameters. .
[0077] After training is complete, the final parameters obtained are the optimal meta-initialization parameters. .
[0078] S4. Rapid adaptation and real-time diagnosis of target operating conditions: Real-time vibration signals of motor bearings are collected, and their Riemann feature sequences are extracted as target operating conditions according to step S1. Using samples of the target operating conditions, the operating conditions of the inner cycle are rapidly adapted based on the optimal meta-initialization parameters to obtain final diagnostic parameters. The final diagnostic parameters are then applied to real-time fault diagnosis of the target operating condition data.
[0079] When encountering new target working conditions At that time, only a small number of Riemann feature samples were used under this operating condition. ,based on A single rapid adaptation is all that's needed to obtain the final high-performance diagnostic model. .
[0080] Example 2
[0081] This embodiment provides a fault diagnosis system for motor bearings under various operating conditions, which applies the fault diagnosis method for motor bearings under various operating conditions based on Riemann feature sequences and meta-learning described in Embodiment 1.
[0082] like Figure 2 As shown, the fault diagnosis system provided in this embodiment includes: a data acquisition module, at least one processor 101, and a memory 102 connected to at least one processor 101. This embodiment does not limit the specific connection medium between the processor 101 and the memory 102. Figure 2 The example shown is the connection between processor 101 and memory 102 via bus 100. Bus 100 is... Figure 2 The connections between other components are shown in bold lines and are for illustrative purposes only, not as limiting information. Bus 100 can be divided into address bus, data bus, control bus, etc., for ease of representation. Figure 2 The bus is represented by a single thick line, but this does not indicate that there is only one bus or one type of bus. Alternatively, the processor 101 may also be called a controller; there is no restriction on the name.
[0083] In this embodiment, the memory 102 stores instructions that can be executed by at least one processor 101. The at least one processor 101 can execute the aforementioned method by executing the instructions stored in the memory 102.
[0084] The processor 101 is the control center of the device. It can connect to various parts of the control device through various interfaces and lines. By running or executing instructions stored in memory 102 and calling data stored in memory 102, the processor can perform various functions and process data, thereby monitoring the device as a whole.
[0085] In one possible design, processor 101 may include one or more processing units. Processor 101 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may also not be integrated into processor 101. In some embodiments, processor 101 and memory 102 may be implemented on the same chip; in some embodiments, they may also be implemented on separate chips.
[0086] Processor 101 can be a general-purpose processor, such as a central processing unit (CPU), digital signal processor, application-specific integrated circuit, field-programmable gate array or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the motor bearing cross-operating condition fault diagnosis method based on Riemann feature sequences and meta-learning disclosed in Embodiment 1 can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules in processor 101.
[0087] Memory 102, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. Memory 102 may include at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type memory, random access memory (RAM), static random access memory (SRAM), programmable read-only memory (PROM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), magnetic storage, magnetic disk, optical disk, etc. Memory 102 can be any other medium capable of carrying or storing desired program code in the form of instructions or data structures that can be accessed by a computer, but is not limited thereto. In this embodiment, memory 102 can also be a circuit or any other device capable of implementing storage functions for storing program instructions and / or data.
[0088] By designing and programming the processor 101, the code corresponding to the motor bearing cross-operating condition fault diagnosis method based on Riemann feature sequences and meta-learning described in the aforementioned embodiments can be embedded into the chip, thereby enabling the chip to execute the code during runtime. Figure 1 The steps of the cross-condition fault diagnosis method for motor bearings based on Riemann feature sequences and meta-learning are shown. How to design and program the processor 101 is a technique well-known to those skilled in the art and will not be elaborated upon here.
[0089] Example 3
[0090] This embodiment provides a computer-readable storage medium storing a computer program thereon. When the program is executed by a processor, it implements the steps of the motor bearing cross-condition fault diagnosis method based on Riemann feature sequences and meta-learning as described in Embodiment 1.
[0091] The computer-readable storage medium may include flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the storage medium may be an internal storage unit of a computer device, such as the hard disk or memory of the computer device. In other embodiments, the storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, smart memory card, secure digital card, flash memory card, etc., provided on the computer device. Of course, the storage medium may include both internal storage units and external storage devices of the computer device. In this embodiment, the memory is typically used to store the operating system and various application software installed on the computer device. In addition, the memory can also be used to temporarily store various types of data that have been output or will be output.
[0092] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A method for cross-condition fault diagnosis of motor bearings based on Riemann feature sequences and meta-learning, characterized in that, include: S1. Collect multi-channel vibration signals of motor bearings under different working conditions and segment them using the sliding window technique. Calculate the symmetric positive definite covariance matrix of the multi-channel vibration signals within each window. Project the covariance matrix into Riemann tangent space eigenvectors through the logarithmic mapping from the Riemann manifold to the tangent space. Concatenate the Riemann tangent space eigenvectors of the continuous time windows to construct a Riemann feature sequence for the meta-learning network. S2. Define each source domain condition in the open-source dataset as a meta-training task, and construct an attention residual network as the base learner of the meta-learning framework; S3. Perform two-layer meta-learning training on the base learner, including rapid adaptation of working conditions in the inner loop and meta-parameter update in the outer loop. After training is terminated, the optimal meta-initialization parameters are obtained. S4. Collect the real-time vibration signal of the motor bearing, and extract its Riemann feature sequence as the target working condition according to step S1; use the sample of the target working condition to perform rapid adaptation of the working condition of the inner cycle based on the optimal element initialization parameters to obtain the final diagnostic parameters, and apply the final diagnostic parameters to the real-time fault diagnosis of the target working condition data.
2. The method for cross-condition fault diagnosis of motor bearings based on Riemann feature sequences and meta-learning according to claim 1, characterized in that, Step S1 specifically includes: S11. The motor bearing monitoring system is equipped with... The multidimensional discrete-time vibration signal sequence acquired by the vibration sensor channels is represented as a matrix. ,in This represents the total number of sampling points; It is the set of real numbers; S12. Let the length of the sliding window be... Step size is Then each sample, i.e. the first... The signal segments within a time window are represented as a matrix. ;in Indicates the first The first time window The multidimensional vibration vector of each sampling point ; S13. Window signals After centering, the covariance matrix of the samples is calculated, expressed by the following formula: In the formula, for The result after centralized processing; For the first The covariance matrix of each time window; the superscript T is the transpose sign; S14. Regarding the covariance matrix Introducing a regularization perturbation term, the formula is as follows: In the formula, This is the regularization perturbation term; for Identity matrix; introducing regularization perturbation term It satisfies the following two properties: symmetry, that is Positive definiteness, that is, for any non-zero vector ,satisfy ; Belongs to the C-order symmetric matrix, denote all The set space formed by symmetric positive definite matrices is At this point, the data for each time window is modeled as a manifold. A point on; S15. Using the matrix logarithmic mapping to transform the manifold Points on Projected onto the identity matrix In the tangent space at the location; where, according to the spectral decomposition theorem, Decomposed into: In the formula, Let be an orthogonal eigenvector matrix, satisfying The superscript T is the transpose symbol; It is an eigenvalue diagonal matrix. This is the standard function used to create diagonal matrices. For matrix The 1 eigenvalue, The formula for the matrix-logarithmic mapping is: In the formula, The matrix obtained after the matrix logarithmic mapping lies in the tangent space. In this context, the tangent space is isomorphic to... Euclidean space; S16. Using semi-vectorization operators Extracting the matrix The upper triangular independent elements are used to generate the Riemann eigenvector of the k-th time window. Thus, for a segment containing continuous Monitoring data from several time windows are used to concatenate the extracted Riemann feature vectors to construct a Riemann feature sequence tensor. , .
3. The method for cross-condition fault diagnosis of motor bearings based on Riemann feature sequences and meta-learning according to claim 2, characterized in that, The calculation formula is: In the formula, For window signals The mean vector within, ; for A column vector of all 1s.
4. The method for cross-condition fault diagnosis of motor bearings based on Riemann feature sequences and meta-learning according to claim 1, characterized in that, In step S2, the attention residual network includes a local feature extractor, a cross-temporal feature enhancer, a fixed-dimensional transformer, and a classification decision layer connected in sequence; the local feature extractor consists of stacked one-dimensional convolutional residual blocks, used to extract local temporal features from the Riemann feature sequence. The cross-temporal feature enhancer employs a self-attention mechanism to... Perform weighted aggregation and output the attention-enhanced feature sequence. The fixed-dimensional converter is... By averaging over the time dimension, a fixed-length feature vector is obtained. The classification decision layer uses a fully connected layer and a Softmax function to... The probability distribution is mapped to the fault category.
5. The method for cross-condition fault diagnosis of motor bearings based on Riemann feature sequences and meta-learning according to claim 1, characterized in that, In step S2, each source domain condition in the open-source dataset is defined as an independent meta-training task. Riemann feature data for each meta-training task Randomly partitioned into support sets for inner loop training and query set used for outer loop evaluation Base learner The parameter set is denoted as .
6. The method for cross-condition fault diagnosis of motor bearings based on Riemann feature sequences and meta-learning according to claim 5, characterized in that, Step S3, the two-layer meta-learning training, specifically includes: S31. In the inner loop, in the support set Calculate the initial parameters Adaptation loss and utilize internal learning rate conduct Gradient descent is used to simulate the rapid adaptation of the base learner to the current operating conditions, resulting in the adapted parameters. ; S32. In the outer loop, use the adapted parameters. In query set Calculate the element loss Evaluate generalization ability; calculate meta-loss with respect to initial parameters. meta gradient And combined with external learning rate Update initial parameters ; Repeat steps S31 and S32 until the preset maximum number of meta-iterations is reached or Convergence; after training is complete, the final parameters obtained are the optimal meta-initialization parameters. .
7. The method for cross-condition fault diagnosis of motor bearings based on Riemann feature sequences and meta-learning according to claim 6, characterized in that, In step S31, The calculation formula is: In the formula, This represents the input Riemann feature sequence tensor. Indicate the fault category; This is the loss function, used to measure the difference between the predicted probability distribution and the actual fault category. The parameter is At that time, the base learner learns the input The fault prediction probability distribution output; The calculation formula is: ,at this time Is to make Minimize the local optimum. Represents the task index of the meta-task, and ; For gradient operators, it means that with respect to initial parameters Find the partial derivative.
8. The method for cross-condition fault diagnosis of motor bearings based on Riemann feature sequences and meta-learning according to claim 7, characterized in that, In step S32, The calculation formula is: In the formula, For the batch of meta-tasks participating in the current meta-training iteration; To use the adapted parameters In query set The evaluation loss function calculated above, The adaptation parameters are obtained by the base learner through rapid adaptation in the inner loop. The following categories are output; Among them, the meta-gradient The formula for calculating the adaptive steps flowing through the inner loop using the chain rule is as follows: In the formula, For the first Individual tasks adapt parameters The generalization loss value is as follows; Meta-loss for adapted parameters The gradient vector; To adapt to post-parameters For initial parameters The Jacobian matrix; initial parameters The update formula is: .
9. A fault diagnosis system for motor bearings under various operating conditions, characterized in that, The system includes a data acquisition module, a processor, and a memory for storing a computer program; the data acquisition module is used to acquire vibration signals of the motor bearing; when the computer program is executed by the processor, it implements the steps of the motor bearing cross-condition fault diagnosis method based on Riemann feature sequences and meta-learning as described in any one of claims 1 to 8.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the motor bearing cross-condition fault diagnosis method based on Riemann feature sequences and meta-learning as described in any one of claims 1 to 8.