Fault diagnosis method and system for multi-phase permanent magnet motor in cross-working mode

Abstract

The invention provides a cross-working-mode multi-phase permanent magnet motor fault diagnosis method and system, relates to the field of motor fault diagnosis, and solves the technical problems of model redundancy, large calculation amount, long training time and large occupied storage space caused by the fact that a deep learning model needs to be independently constructed for each working mode in an existing fault diagnosis method. The method comprises the following steps: constructing a single switch tube open-circuit fault sample library of the multi-phase permanent magnet motor system in multiple working modes, wherein each sample is provided with a label; the samples are grouped according to labels and integrated into a training data set of a CNN model, the CNN model is trained, and a loss function of the CNN model is calculated; on the basis of the loss function, global parameters of the CNN model are iteratively optimized through a back propagation algorithm, and CNN model training is completed; and analyzing the fault type of the to-be-diagnosed multiphase permanent magnet motor system through the CNN model. The method is used in the fault diagnosis process of the multi-phase permanent magnet motor.

Description

Technical Field

[0001] This application relates to the field of motor fault diagnosis, and in particular to a method and system for diagnosing faults in a multiphase permanent magnet motor that operates across different modes. Background Technology

[0002] With the development of artificial intelligence, fault diagnosis methods for motor systems based on deep learning have received increasing attention. However, current deep learning-based fault diagnosis methods only apply to motor systems operating under normal working conditions. Multiphase permanent magnet motor systems, in addition to operating under normal conditions, also operate under fault-tolerant conditions. Following existing fault diagnosis approaches, separate fault identification models need to be built for each operating mode. Furthermore, deep learning-based fault diagnosis methods themselves require significant training time and computational resources, leading to redundancy in the fault identification models. The complexity and storage space occupied by these models will also be excessive. Summary of the Invention

[0003] This application provides a fault diagnosis method for multiphase permanent magnet motors that operates across different modes, solving the technical problems of existing fault diagnosis methods that require building separate deep learning models for each operating mode, resulting in model redundancy, high computational load, long training time, and high storage space consumption. This application also provides a fault diagnosis system for multiphase permanent magnet motors that operates across different modes.

[0004] To achieve the above objectives, this application adopts the following technical solution: Firstly, a fault diagnosis method for multiphase permanent magnet motors operating across different modes is provided, including: A sample library of single switch open-circuit faults of multiphase permanent magnet motor systems under various operating modes is constructed, and each sample is labeled; the labels include fault type labels and operating mode labels. The labeled samples are grouped according to their labels and integrated into the training dataset of the CNN model. During training, the loss function of the CNN model is calculated. Based on the loss function, the global parameters of the CNN model are iteratively optimized through the backpropagation algorithm to complete the training of the CNN model. The CNN model includes a shared feature extraction layer and multiple output branches that correspond one-to-one with the working mode. Each output branch is used to classify multiple fault types in each working mode. The system acquires the operating data and current operating mode of the multiphase permanent magnet motor system to be diagnosed. The operating data and operating mode labels are then input into the trained CNN model. The CNN model activates the corresponding output branch based on the operating mode label, and the output branch outputs the fault type label of an open circuit of a single switch in the current operating mode.

[0005] Based on the above technical solutions, in multiphase permanent magnet motor systems, the motor not only operates in normal mode but may also enter various fault-tolerant operating modes due to faults, such as one-phase loss or two-phase loss. The current harmonic distribution and fault characteristics differ significantly under different operating modes. If the traditional "single model unified diagnosis" strategy is adopted, the model cannot adequately account for the characteristic differences of various operating conditions, leading to a decrease in diagnostic accuracy. Conversely, if a separate deep learning diagnostic model is built for each operating mode, it will result in redundant model structure, a surge in parameters, long training time, high storage overhead, and difficulty in deployment on embedded platforms. To address this technical challenge, this solution proposes a lightweight multi-task diagnostic architecture of "shared feature extraction + multi-branch output": by constructing a dual-labeled sample library containing operating mode labels and fault type labels, the model learns the joint mapping relationship between modes and faults during the training phase; shared convolutional layers are used to extract common fault features across modes, significantly reducing parameter redundancy; simultaneously, output branches corresponding one-to-one with the operating mode are set, and the corresponding branches are dynamically activated during inference based on the input operating mode labels, achieving "on-demand diagnosis." This method, while maintaining high diagnostic accuracy across all fault-tolerant modes, significantly reduces model complexity and computational resource consumption. Furthermore, it alleviates the sample imbalance problem through a weighted loss function, effectively improving the ability to identify rare fault-tolerant modes. The overall solution achieves the goal of "one lightweight model covering all operating conditions and accurately identifying faults," solving key technical bottlenecks such as redundancy, poor generalization, and deployment difficulties in fault diagnosis models under multiple operating conditions. It provides an efficient and feasible solution for online intelligent diagnosis of high-reliability multiphase motor systems.

[0006] In conjunction with the first aspect above, in one possible implementation, the working mode includes a normal operating mode. A fault-tolerant working mode with one missing phase and A fault-tolerant operating mode for two-phase deficiency; The normal operating mode label is operating mode 0; The The fault-tolerant working mode with one missing phase is tagged as working mode. , , It is a positive integer; The The fault-tolerant operating modes for missing two phases include two categories: missing two adjacent phases and missing two non-adjacent phases. Operating mode tags. , , It is a positive integer; Each working mode corresponds to a system operating state under a single-phase open-circuit fault.

[0007] In conjunction with the first aspect above, in one possible implementation, the construction of a sample library of open-circuit faults of a single switch in a multiphase permanent magnet motor system under various operating modes includes: Several multiphase current data points for each operating mode are obtained from historical data, and the multiphase currents for each operating mode are transformed into Clark transformations using Clark transformation. α-β Subspace current components i α , i β and xy Subspace current components i x and i y ; Will i α , i β , i x and i y Perform normalization, and with i α For the horizontal axis, i β Using the vertical axis, we construct... α-β The multi-harmonic plane current vector trajectory diagram in the subspace, and, with i x For the horizontal axis, i y Using the vertical axis, we construct... xy Multiharmonic plane current vector trajectory diagram in subspace; Each multi-harmonic plane current vector trajectory diagram is used as sample data, and each sample data is labeled with an operating mode label and a fault type label; based on all labeled sample data, a sample library of single switch open circuit faults of multiphase permanent magnet motor system is constructed.

[0008] In conjunction with the first aspect above, in one possible implementation, the time length of the multi-harmonic plane current vector trajectory diagram is not less than one complete electrical cycle, and the sampling frequency satisfies the Nyquist sampling theorem.

[0009] In conjunction with the first aspect above, in one possible implementation, the shared feature extraction layer is used to extract common features from the input data. The shared feature extraction layer includes an input layer, a convolutional layer, and a pooling layer. Each output branch corresponds to a specific working mode for performing fault classification tasks under each working mode. The output branch includes at least one fully connected layer, a global average pooling layer, or an output layer.

[0010] In conjunction with the first aspect above, in one possible implementation, the number of output branches is equal to the total number of operating modes of the multiphase permanent magnet motor system to be diagnosed, and the structure of each output branch can be adjusted according to the characteristics of its corresponding operating mode.

[0011] In conjunction with the first aspect above, in one possible implementation, the calculation of the loss function of the CNN model includes: The feature vector output by the shared feature extraction layer With input working mode information The features are concatenated to obtain the fused feature vector. ; ; The output branch corresponding to working mode k outputs: ;in, For the working mode label, k=0,1, , The number of fault-tolerant operating modes that are missing one phase. The number of fault-tolerant operating modes with two phases missing. This is the weight matrix of the GAP layer corresponding to working mode k. The bias vector for working mode k; Calculate the classification loss function under working mode k: ; in, Let k be the number of samples for working mode k. Let j be the j-th sample element of the mask vector, used to identify whether sample j belongs to working mode k; if yes, then... ;otherwise , Here, r is the indicator function; r is the fault type label in operating mode k. The total number of fault types under operating mode k. The label for the true fault type of sample j; Classification loss function based on working mode k Through formula The loss function of the CNN model is calculated. .

[0012] In conjunction with the first aspect above, in one possible implementation, the working mode information is a one-dimensional one-hot encoded vector with a length of p1+p2+1; wherein, only one element in the encoded vector is 1, and the rest are 0, which is used to uniquely identify the working mode of the current sample.

[0013] In conjunction with the first aspect mentioned above, in one possible implementation, the CNN model further includes lightweight processing: Replacing fully connected layers with global average pooling layers reduces the dimensionality of the feature maps. ; in, For the first The output of the global average pooling layer for each channel The number of eigenvalues ​​for each channel, where f is the eigenvalue index. For the first The f-th feature value of each channel.

[0014] Secondly, this application provides a multiphase permanent magnet motor fault diagnosis system that operates across multiple modes, including a construction module, a model training module, and a diagnosis module. The construction module is used to build a sample library of single-switch open-circuit faults of the multiphase permanent magnet motor system under various operating modes, with each sample labeled. The model training module is used to group the labeled samples according to their labels and integrate them into a training dataset for a CNN model. During training, the loss function of the CNN model is calculated. Based on the loss function, the global parameters of the CNN model are iteratively optimized using a backpropagation algorithm to complete the CNN model training. The diagnosis module is used to acquire the operating data and current operating mode of the multiphase permanent magnet motor system to be diagnosed, input the operating data and operating mode labels into the trained CNN model, and activate the corresponding output branch according to the operating mode label. The output branch outputs the fault type label of a single-switch open circuit in the current operating mode.

[0015] This application provides a method and system for fault diagnosis of multiphase permanent magnet motors across operating modes. By constructing an image-based sample library with dual labels for fault type and operating mode, it achieves refined characterization of fault features under complex operating conditions. The designed multi-branch convolutional neural network adopts a shared feature extraction layer structure, which can automatically learn common fault features across modes from the input data, significantly reducing the number of model parameters and avoiding structural redundancy and resource waste caused by independently modeling each operating mode. Each output branch corresponds one-to-one with a specific operating mode. During the inference phase, the model dynamically activates the corresponding branch based on the input operating mode label. This branch completes the fault classification task within the mode based on the features extracted from the shared layer, balancing the universality of feature extraction and the specificity of classification decisions. This mechanism effectively improves the diagnostic accuracy and generalization ability of the model under various fault-tolerant operating conditions, especially suitable for fault identification under asymmetric conditions such as phase loss. A weighted loss function further alleviates the class imbalance problem and enhances the sensitivity to sparse patterns. The overall solution achieves high-accuracy diagnosis while reducing model complexity. It has advantages such as high computational efficiency, easy deployment, and strong robustness, providing an effective solution for intelligent and lightweight fault diagnosis of multiphase motor systems.

[0016] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0017] Figure 1 A system architecture diagram of a multiphase permanent magnet motor fault diagnosis system with cross-operating modes provided in this application embodiment; Figure 2 A flowchart illustrating a method for diagnosing faults in a multiphase permanent magnet motor across operating modes, provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a fault diagnosis device provided in an embodiment of this application; Figure 4 This is a schematic diagram of the hardware structure of a fault diagnosis device provided in an embodiment of this application. Detailed Implementation

[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0019] The cross-operating-mode multiphase permanent magnet motor fault diagnosis method provided in this application embodiment can be applied to a cross-operating-mode multiphase permanent magnet motor fault diagnosis system, such as... Figure 1 As shown, the system includes: a building module, a model training module, and a diagnostic module; The construction module is used to build a sample library of single switch open-circuit faults in a multiphase permanent magnet motor system under various operating modes, and each sample is labeled. The model training module is used to group the labeled samples according to their labels and integrate them into the training dataset of the CNN model; during the training process, the loss function of the CNN model is calculated; based on the loss function, the global parameters of the CNN model are iteratively optimized through the backpropagation algorithm to complete the training of the CNN model. The diagnostic module is used to acquire the operating data and current operating mode of the multiphase permanent magnet motor system to be diagnosed. The operating data and operating mode labels are input into the trained CNN model. The CNN model activates the corresponding output branch according to the operating mode label. The output branch outputs the fault type label of a single open circuit of a single switch in the current operating mode.

[0020] To address the technical problems of existing fault diagnosis methods requiring the separate construction of deep learning models for each operating mode, resulting in model redundancy, high computational load, long training time, and high storage space consumption, this application provides a cross-operating mode multiphase permanent magnet motor fault diagnosis method. This method includes: constructing a sample library of single switch open-circuit faults in a multiphase permanent magnet motor system under various operating modes, with each sample labeled; grouping the labeled samples according to their labels and integrating them into a training dataset for a CNN model; calculating the loss function of the CNN model during training; iteratively optimizing the global parameters of the CNN model using a backpropagation algorithm based on the loss function to complete the CNN model training; acquiring the operating data and current operating mode of the multiphase permanent magnet motor system to be diagnosed, inputting the operating data and operating mode labels into the trained CNN model, and having the CNN model activate the corresponding output branch according to the operating mode label, with the output branch outputting the fault type label for a single switch open circuit under the current operating mode. Based on this, by constructing a dual-label sample library of fault type and operating mode, accurate modeling of switch open-circuit faults under multiple operating conditions is achieved. A multi-branch CNN structure with shared feature extraction layers is adopted. Common features of each mode are extracted by sharing convolutional layers, which significantly reduces the number of parameters and avoids model redundancy. Each output branch corresponds one-to-one with the working mode. During inference, the corresponding branch is activated according to the mode label to realize fault classification within this mode, taking into account both feature universality and decision specificity.

[0021] like Figure 2 As shown in the embodiment of this application, a fault diagnosis method for a multiphase permanent magnet motor with multiple operating modes is provided, including: S201. Construct a sample library of single switch open-circuit faults for multiphase permanent magnet motor systems under various operating modes, and assign a label to each sample.

[0022] The labels include fault type labels and operating mode labels. The operating mode labels are used to distinguish different system operating states, and the fault type labels are used to identify the location of the switch where an open-circuit fault occurred.

[0023] The working modes include one normal working mode, A fault-tolerant working mode with one missing phase and A fault-tolerant operating mode for two-phase deficiency; The normal operating mode is labeled as operating mode 0; The fault-tolerant working mode with one missing phase is tagged as working mode. , , For positive integers, each working mode This corresponds to the system operating state under a single-phase open-circuit fault. The fault-tolerant operating modes for missing two phases include two categories: missing two adjacent phases and missing two non-adjacent phases. Operating mode tags. , , It is a positive integer.

[0024] For example, the operating modes of a five-phase permanent magnet motor system include one normal operating mode, five fault-tolerant operating modes with one phase missing, and ten fault-tolerant operating modes with two phases missing. The normal operating mode is labeled as operating mode 0. The five fault-tolerant operating modes with one phase missing are labeled as operating mode 1, operating mode 2, ..., operating mode 5. The ten fault-tolerant operating modes with two phases missing are divided into two categories: missing two adjacent phases and missing two non-adjacent phases, labeled as operating mode 6, operating mode 7, ..., operating mode 15.

[0025] S202. Group the labeled samples according to their labels and integrate them into the training dataset for the CNN model. During training, calculate the loss function of the CNN model. Based on the loss function, iteratively optimize the global parameters of the CNN model through the backpropagation algorithm to complete the training of the CNN model.

[0026] The CNN model includes a shared feature extraction layer and multiple output branches that correspond one-to-one with the operating modes. Each output branch is used to classify multiple fault types in each operating mode.

[0027] The shared feature extraction layer is used to extract common features from the input data. The shared feature extraction layer includes an input layer, a convolutional layer, and a pooling layer. Each output branch corresponds to a specific working mode and is used to perform the fault classification task under each working mode. The output branch includes at least one fully connected layer, a global average pooling layer, or an output layer.

[0028] In addition, to achieve lightweight CNN models, fully connected layers are replaced with global average pooling (GAP) layers to reduce the dimension of feature maps, thereby reducing computational cost and parameter count. ; in, For the first The output of the global average pooling layer for each channel The number of eigenvalues ​​for each channel, where f is the eigenvalue index. For the first The f-th feature value of each channel.

[0029] Based on the above technical solutions, in multiphase permanent magnet motor systems, the current fault characteristics under different operating modes, such as normal, one-phase missing, and two-phase missing, differ significantly. Using a general-purpose CNN model with a single output head makes it difficult to meet the discrimination requirements of various operating conditions. Training a separate model for each mode leads to structural redundancy, parameter explosion, and high training costs, making deployment on resource-constrained embedded platforms difficult. Therefore, this solution adopts a lightweight multi-task architecture of "shared feature extraction layer + dedicated output branches for multiple operating modes": the shared layer extracts common fault features across modes through convolution and pooling, avoiding redundant learning; each output branch corresponds one-to-one with the operating mode, activating the corresponding branch only during inference based on the input mode label, achieving accurate and efficient fault classification. To further reduce model complexity, a global average pooling (GAP) layer replaces the traditional fully connected layer. By compressing the spatial dimension of each feature channel, class-discriminative features are directly generated, significantly reducing the number of parameters and computational overhead, while enhancing the model's generalization ability and suppressing overfitting. This design ensures high diagnostic accuracy under various operating conditions while significantly improving the model's lightweight nature and deployment efficiency, effectively solving key technical challenges such as model redundancy, computational complexity, and poor generalization in multi-mode fault diagnosis.

[0030] S203. Obtain the operating data and current working mode of the multiphase permanent magnet motor system to be diagnosed. Input the operating data and working mode label into the trained CNN model. The CNN model activates the corresponding output branch according to the working mode label. The output branch outputs the fault type label of a single open circuit of a single switch in the current working mode.

[0031] The number of output branches is equal to the total number of operating modes of the multiphase permanent magnet motor system to be diagnosed, and the structure of each output branch can be adjusted according to the characteristics of its corresponding operating mode.

[0032] Based on the above technical solutions, in a multiphase permanent magnet motor system, the motor not only operates in a normal state but may also switch to various fault-tolerant operating modes due to faults such as open circuits in switching transistors, such as asymmetrical operating states with one, two, or even more phases missing. The current waveforms, harmonic components, and fault characteristics under these different operating modes exhibit significant differences: in the normal mode, the current is symmetrical and the harmonic content is low, while in the fault-tolerant mode, the current is severely distorted, specific harmonics (such as the 5th and 7th harmonics) are significantly enhanced, and the fault characteristics caused by different combinations of phase loss are all different. If the traditional "single model unified diagnosis" method is adopted, that is, using a single output head to classify faults under all operating conditions, the model cannot adapt to such diverse feature distributions simultaneously, which can easily lead to a significant drop in diagnostic accuracy under certain fault-tolerant modes. Conversely, if an independent deep learning model is designed for each operating mode (such as 16 types), it will lead to a surge in the number of models, a huge total number of parameters, repetitive and redundant training processes, and excessive consumption of storage and computing resources, severely restricting its practical deployment in vehicle-mounted, shipborne, or industrial embedded control systems.

[0033] To address the aforementioned technical bottlenecks, this solution innovatively proposes a lightweight, multi-task fault diagnosis architecture based on "shared feature extraction + multi-branch output." First, a sample library with dual semantic labels is constructed—each sample is simultaneously labeled with its operating mode label (e.g., "missing C phase") and fault type label (e.g., "open circuit in upper arm of C phase"), enabling the model to establish a joint "mode-fault" mapping during training. Second, the network front-end employs shared convolutional and pooling layers to automatically extract common fault features across operating modes (e.g., harmonic trajectory shape, asymmetry, etc.) from the input current vector trajectory image, avoiding repeated learning of low-level features for each mode and significantly compressing model parameters. The back-end sets up multiple output branches corresponding one-to-one with the operating mode, each branch dedicated to fine-grained fault classification within that mode. During the inference phase, the system dynamically activates the corresponding output branch based on the real-time acquired operating mode labels, achieving "on-demand diagnosis," ensuring classification accuracy while avoiding the computational overhead of irrelevant branches. Furthermore, by introducing a weighted loss function based on the number of samples, the class imbalance problem caused by the fact that there are far more normal mode samples than fault-tolerant mode samples (especially the rare mode with two missing samples) is effectively alleviated, and the sensitivity and robustness of the model to small sample conditions are significantly improved.

[0034] In summary, this solution uses a single lightweight model to uniformly cover all operating conditions. While significantly reducing model complexity, memory usage, and inference latency, it achieves high-precision and highly adaptable open-circuit fault diagnosis of switching transistors. It effectively solves core problems such as redundancy, weak generalization ability, and deployment difficulties in fault diagnosis models under multiple operating conditions, and provides a practical and feasible technical path for intelligent operation and maintenance of multiphase permanent magnet motor systems in high-reliability scenarios such as aerospace, new energy vehicles, and ship propulsion.

[0035] Furthermore, in this embodiment, a sample library with dual labels for fault type and operating mode is constructed, which can comprehensively characterize the fault characteristics of multiphase permanent magnet motors under different fault-tolerant operating states, breaking through the limitation of traditional methods that only model a single normal mode. By simultaneously labeling each sample with both "fault type" and "operating mode" labels, a refined description of complex fault scenarios such as phase loss and open circuit of switching transistors is achieved, enabling the model to learn the differences and evolution patterns of fault characteristics under different operating modes during the training phase. This dual-label mechanism not only enhances the information integrity of the sample data, but also provides key prior information for the subsequent multi-branch network structure to achieve collaborative decision-making of "pattern recognition-fault classification," supporting the model to dynamically match the optimal diagnostic path based on the current operating state during inference. Especially after the motor enters a fault-tolerant operating mode with one or two phases missing, traditional methods are prone to misjudgment due to feature shift, while this solution, guided by pattern labels, significantly improves the adaptability and accuracy of fault diagnosis, providing a reliable data foundation for intelligent diagnosis under complex multi-conditions.

[0036] In one possible implementation of this application embodiment, the above-mentioned S201 can be specifically implemented by the following S301, S302 and S303, which are described in detail below: S301. Obtain several multiphase current data points for each operating mode from historical data, and convert the multiphase current data for each operating mode into a Clark transformation. α-β Subspace current components i α , i β and xy Subspace current components i x and i y ; Will i α , i β , i x and i y Perform normalization, and with i α For the horizontal axis, i β Using the vertical axis, we construct... α-β The multi-harmonic plane current vector trajectory diagram in the subspace, and, with i x For the horizontal axis, i y Using the vertical axis, we construct... xy Multiharmonic plane current vector trajectory diagram in subspace; Normalization refers to normalizing the amplitude of each current component according to its peak or effective value under normal operating mode, so that the current amplitude of all samples is within the range of [0,1] or [-1,1].

[0037] It should be noted that the time length of the multi-harmonic plane current vector trajectory diagram is no less than one complete electrical cycle, and the sampling frequency satisfies the Nyquist sampling theorem. Furthermore, the current vector trajectory diagram is stored as a grayscale image or a pseudo-color image with a resolution of no less than 64×64 pixels, and is used as input for training subsequent convolutional neural network models.

[0038] S303. Each multi-harmonic plane current vector trajectory diagram is used as sample data, and each sample data is labeled with a working mode label and a fault type label; based on all labeled sample data, a sample library of single switch open circuit faults of multiphase permanent magnet motor system is constructed.

[0039] For example, current data of a five-phase permanent magnet motor (A, B, C, D, E) under specific operating modes and fault conditions are extracted from historical data, respectively. , , , , For example, within a complete electrical cycle, such as 20ms corresponding to a 50Hz frequency, the current time series of each phase is collected.

[0040] For a five-phase permanent magnet motor, the Clark transformation matrix is ​​typically used to convert the current in the five-phase coordinate system to a two-phase stationary coordinate system (α-β coordinate system) and two other harmonic component coordinate systems (xy coordinate system). The Clark transformation formula is as follows: ; in, This is the Clark transformation matrix, which varies depending on the specific structure of the five-phase system. Applying this transformation yields four new current components. , , ,and .

[0041] For the obtained current components , , ,and Normalization is performed so that the current amplitude of all samples falls within the range of [0,1] or [-1,1]. Typically, the maximum or effective value of the current component under normal operating conditions is used as a reference for scaling.

[0042] Using the normalized current components, construct respectively α-β subspace and xy The multi-harmonic plane current vector trajectory plot in the subspace. This means we need to plot the... For the horizontal axis, A graph with the vertical axis as the ordinate, and with... For the horizontal axis, The graph is plotted on the vertical axis. These two graphs reflect the operating characteristics of the motor in different subspaces.

[0043] Each generated trajectory image is assigned a corresponding operating mode label and fault type label. For example, "Operating Mode 3" can represent a specific fault-tolerant operating state, while "Fault Type 3" may refer to a specific situation where a switch is open-circuited.

[0044] Based on the above annotation information, all samples are integrated into a sample library of single-switch open-circuit faults in multiphase permanent magnet motor systems. This sample library can be used to train convolutional neural network models to identify different operating modes and fault types.

[0045] Based on the above technical solutions, traditional fault diagnosis methods for multiphase permanent magnet motors mostly rely on the time-domain or frequency-domain characteristics of the original current signal, making it difficult to effectively capture the complex harmonic coupling and trajectory distortion characteristics caused by faults under different operating modes, such as normal operation, one-phase loss, and two-phase loss. Especially under fault-tolerant operating conditions, the current waveform is highly asymmetrical, and traditional characteristics are easily affected by operating conditions, resulting in poor diagnostic robustness. Therefore, this solution proposes a technical approach based on Clark transform to construct a multi-harmonic plane current vector trajectory diagram: by decoupling the five-phase (or multi-phase) current into... α-β (fundamental wave dominant) and xy (Harmonic-dominated) Two orthogonal subspaces are used, and the dynamic evolution of the current vector within a complete electrical cycle is visualized in the form of a trajectory diagram. This not only fully preserves the phase, amplitude, and harmonic distribution characteristics caused by the fault, but also transforms the one-dimensional time-series signal into a two-dimensional image with spatial structure information, naturally adapting to the feature extraction mechanism of convolutional neural networks. Furthermore, normalization is used to eliminate the influence of amplitude fluctuations caused by changes in operating conditions such as motor load and speed, ensuring comparability between samples. Simultaneously, the trajectory diagram duration is set to be no less than one electrical cycle, sampling satisfies the Nyquist criterion, and the resolution is no less than 64×64 pixels, balancing signal integrity and model input standardization. Finally, by combining dual-label annotation of operating mode and fault type, a clearly structured and semantically explicit image-based fault sample library is constructed, providing high-quality, highly discriminative training data for subsequent multi-branch lightweight CNN models. This method fundamentally solves key problems in traditional diagnosis, such as weak feature representation ability, poor adaptability to operating conditions, and insufficient model generalization ability, significantly improving the identification accuracy and reliability of single switch open-circuit faults under multiple operating conditions.

[0046] In one possible implementation of this application embodiment, the above-mentioned S202 can be specifically implemented by the following S401, S402 and S403, which are described in detail below: S401, the feature vector output by the shared feature extraction layer With input working mode information The features are concatenated to obtain the fused feature vector. ; ; The output branch corresponding to working mode k outputs: ;in, For the working mode label, k=0,1, , The number of fault-tolerant operating modes that are missing one phase. The number of fault-tolerant operating modes with two phases missing. This is the weight matrix of the GAP layer corresponding to working mode k. The bias vector for working mode k; It should be noted that the working mode information is a one-dimensional one-hot encoded vector with a length of p1+p2+1; wherein, there is only one element in the encoded vector that is 1, and the rest are all 0, which is used to uniquely identify the working mode of the current sample.

[0047] S402. Calculate the classification loss function under working mode k: ; in, Let k be the number of samples for working mode k. Let j be the j-th sample element of the mask vector, used to identify whether sample j belongs to working mode k; if yes, then... ;otherwise , Here, r is the indicator function; r is the fault type label in operating mode k. The total number of fault types under operating mode k. The label for the true fault type of sample j; S403, Classification Loss Function Based on Working Mode k Through formula The loss function of the CNN model is calculated. .

[0048] For example, the motor type is a five-phase permanent magnet synchronous motor, and the total number of working modes is as follows: normal mode: 1 type, fault-tolerant mode for missing one phase: 5 types (missing A, B, C, D, E), fault-tolerant mode for missing two phases: 10 types (such as missing AB, AC, AD...DE), and the total number of working modes is 16, i.e., k=0,1,...,15; The types of faults to be diagnosed in each operating mode: for example, 5 types (corresponding to 5 open circuits of switching transistors), therefore ; Suppose that the current training batch contains samples from multiple working modes; some of them belong to the "C-phase missing" mode (let k=3).

[0049] S401: Feature concatenation and branch output; Shared feature extraction, input a single image α-β or xy The planar trajectory image (64×64 grayscale image) is passed through a CNN shared layer (convolution + GAP) to output a feature vector. Assume d = 128; Working mode information encoding: The current sample belongs to "C phase missing" → corresponding working mode label k=3, construct a one-hot vector. , ; fused feature vector ; Output branch for working mode k=3 .

[0050] S402: Calculate the loss function for working mode k=3; Suppose there are 100 samples in the current batch; of which 20 belong to k=3 (i.e., ... For each sample j, check if it belongs to pattern 3; if yes, then... ;otherwise ; Assuming sample j=50 belongs to operating mode k=3, its actual fault is "C-phase upper arm open circuit" → Then the loss function for sample j=50 with working mode k+3 is: ; S403: Calculate the total loss; The loss functions corresponding to each working mode are as follows: , , , ... Therefore, the loss function of the CNN model .

[0051] Based on the above technical solutions, in multiphase permanent magnet motor systems, the fault manifestations under different operating modes, such as normal operation, one-phase loss, and two-phase loss, differ significantly. If a unified classification head is used for fault diagnosis, the model cannot simultaneously adapt to the vastly different feature distributions under various operating conditions, leading to a decrease in diagnostic accuracy. On the other hand, training an independent model for each mode would result in parameter redundancy, high training costs, and deployment difficulties. To address this challenge, this solution designs a collaborative training mechanism of "shared features + pattern-aware branch + weighted multi-task loss". First, the common feature vector output from the shared feature extraction layer is concatenated with the working mode encoding to form a fused feature vector, enabling each output branch to utilize common features while also being aware of its current operating state. Second, a masking mechanism ensures that each sample only participates in the loss calculation of the branch corresponding to its working mode, achieving task decoupling and precise supervision. Finally, a total loss function weighted by the number of samples is used to normalize and weight the losses of different modes, effectively mitigating the class imbalance problem caused by the fact that there are far more samples in the normal mode than in the fault-tolerant mode, preventing model training from being dominated by large-sample modes, and thus significantly improving the fault identification capability for rare but critical fault-tolerant operating conditions. This technology achieves high-precision and robust diagnosis under multiple operating conditions while maintaining a lightweight model, solving the core problems faced by traditional methods in multi-mode scenarios, such as poor generalization, model redundancy, and failure of small-sample mode diagnosis. It provides an efficient, scalable, and engineering-practical solution for intelligent fault diagnosis of complex multiphase motor systems.

[0052] The above primarily describes the solutions of the embodiments of this application from the perspective of device implementation. It is understood that each device, for example, a multi-phase permanent magnet motor fault diagnosis device spanning multiple operating modes, includes at least one of the hardware structures and software modules corresponding to each function in order to achieve the above-mentioned functions. Those skilled in the art should readily recognize that, based on the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0053] This application embodiment can divide a multi-phase permanent magnet motor fault diagnosis device across operating modes into functional units based on the above method example. For example, each function can be divided into separate functional units, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents a logical functional division; other division methods may be used in actual implementation.

[0054] When using integrated units, Figure 3 A possible structural schematic diagram of a multiphase permanent magnet motor fault diagnosis device (referred to as fault diagnosis device 50) involving cross-operating modes involved in the above embodiments is shown. The fault diagnosis device 50 includes a processing unit 501 and a communication unit 502, and may also include a storage unit 503. Figure 3 The structural diagram shown can be used to illustrate the structure of a multiphase permanent magnet motor fault diagnosis device that spans multiple operating modes involved in the above embodiments.

[0055] when Figure 3 The structural diagram shown illustrates the structure of a multi-phase permanent magnet motor fault diagnosis device that operates across multiple modes as described in the above embodiments. The processing unit 501 is used to control and manage the operation of the multi-phase permanent magnet motor fault diagnosis device that operates across multiple modes. The communication unit 502 is used for the multi-phase permanent magnet motor fault diagnosis device that operates across multiple modes to communicate with other devices. The storage unit 503 is used to store the program code and data of the multi-phase permanent magnet motor fault diagnosis device that operates across multiple modes.

[0056] For example, communication unit 502 is used to build a sample library of single switch open circuit faults of multiphase permanent magnet motor system in multiple operating modes, and each sample is labeled. The processing unit 501 is used to group the labeled samples according to their labels and integrate them into the training dataset of the CNN model; during the training process, it calculates the loss function of the CNN model; based on the loss function, it iteratively optimizes the global parameters of the CNN model through the backpropagation algorithm to complete the training of the CNN model. In one possible implementation, the processing unit 501 is further configured to acquire the operating data and current operating mode of the multiphase permanent magnet motor system to be diagnosed, input the operating data and operating mode label into the trained CNN model, and activate the corresponding output branch according to the operating mode label. The output branch outputs the fault type label of a single open circuit of a single switch in the current operating mode.

[0057] The processing unit 501 can be a processor or a controller, and the communication unit 502 can be a communication interface, transceiver, transceiver circuit, transceiver device, etc. The term "communication interface" is a general term and may include one or more interfaces. The storage unit 503 can be a memory. When the fault diagnosis device 50 is a chip, the processing unit 501 can be a processor or a controller, and the communication unit 502 can be an input interface and / or an output interface, pins, or circuits, etc. The storage unit 503 can be a storage unit within the chip (e.g., a register, cache, etc.) or a storage unit located outside the chip (e.g., read-only memory (ROM), random access memory (RAM, etc.).

[0058] The communication unit can also be called a transceiver unit. The antenna and control circuit with transceiver functions in the fault diagnosis device 50 can be considered as the communication unit 502 of the fault diagnosis device 50, and the processor with processing functions can be considered as the processing unit 501 of the fault diagnosis device 50. Optionally, the device in the communication unit 502 that implements the receiving function can be considered as a communication unit, which is used to execute the receiving steps in the embodiments of this application. The communication unit can be a receiver, a receiver circuit, etc. The device in the communication unit 502 that implements the transmitting function can be considered as a transmitting unit, which is used to execute the transmitting steps in the embodiments of this application. The transmitting unit can be a transmitter, a transmitter, a transmitting circuit, etc.

[0059] Figure 3 If the integrated units in the process are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods in the various embodiments d of this application. Storage media for storing computer software products include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0060] Figure 3 The units in the process can also be called modules; for example, a processing unit can be called a processing module.

[0061] This application also provides a hardware structure diagram of a multi-phase permanent magnet motor fault diagnosis device (denoted as fault diagnosis device 60) that operates across multiple modes. See [link to diagram]. Figure 4 The fault diagnosis device 60 includes a processor 601, and optionally, a memory 602 connected to the processor 601.

[0062] In the first possible implementation, see Figure 4 The fault diagnosis device 60 also includes a transceiver 603. The processor 601, memory 602, and transceiver 603 are connected via a bus. The transceiver 603 is used to communicate with other devices or communication networks. Optionally, the transceiver 603 may include a transmitter and a receiver. The device in the transceiver 603 that implements the receiving function can be considered as a receiver, which is used to perform the receiving steps in the embodiments of this application. The device in the transceiver 603 that implements the transmitting function can be considered as a transmitter, which is used to perform the transmitting steps in the embodiments of this application.

[0063] Based on the first possible implementation method Figure 4 The structural diagram shown can be used to illustrate the structure of a multiphase permanent magnet motor fault diagnosis device that spans multiple operating modes involved in the above embodiments.

[0064] in, Figure 4 This can also be illustrated as a system chip in a multiphase permanent magnet motor fault diagnosis device that operates across multiple modes. In this case, the actions performed by the aforementioned multiphase permanent magnet motor fault diagnosis device can be implemented by this system chip. The specific actions performed are described above and will not be repeated here.

[0065] Some of the data in the above formula are calculated by removing dimensions and taking their numerical values. The formula is the closest to the real situation obtained by software simulation of a large amount of collected data. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

Claims

1. A method for fault diagnosis of a multiphase permanent magnet motor across operating modes, characterized in that, include: A sample library of single switch open-circuit faults of multiphase permanent magnet motor systems under various operating modes is constructed, and each sample is labeled; the labels include fault type labels and operating mode labels. The samples are grouped according to labels and integrated into a training dataset for the CNN model. The CNN model is then trained, and its loss function is calculated. Based on the loss function, the global parameters of the CNN model are iteratively optimized using the backpropagation algorithm to complete the CNN model training. The CNN model includes a shared feature extraction layer and multiple output branches that correspond one-to-one with the working mode. Each output branch is used to classify multiple fault types in each working mode. The system acquires the operating data and current operating mode of the multiphase permanent magnet motor system to be diagnosed. The operating data and operating mode labels are then input into the trained CNN model. The CNN model activates the corresponding output branch based on the operating mode label, and the output branch outputs the fault type label of an open circuit of a single switch in the current operating mode.

2. The method according to claim 1, wherein, The working modes include a normal working mode, a one-phase fault-tolerant working mode, and a two-phase fault-tolerant working mode. The normal operating mode label is operating mode 0; The A fault-tolerant operating mode tag is an operating mode , , is a positive integer; The A fault-tolerant operation mode of missing two phases includes two categories of missing adjacent two phases and missing non-adjacent two phases, and an operation mode label , , is a positive integer; Each working mode corresponds to a system operating state under a single-phase open-circuit fault.

3. The method according to claim 1, wherein, The constructed sample library of single switch open-circuit faults for multiphase permanent magnet motor systems under various operating modes includes: Obtain several polyphase current data in each working mode from historical data, convert the polyphase current in each working mode into α-β subspace current components i α , i β and xy subspace current components i x and i y ; Will i α , i β , i x and i y Perform normalization, and with i α For the horizontal axis, i β Using the vertical axis, we construct... α-β The multi-harmonic plane current vector trajectory diagram in the subspace, and, with i x For the horizontal axis, i y Using the vertical axis, we construct... xy Multiharmonic plane current vector trajectory diagram in subspace; Each multi-harmonic plane current vector trajectory diagram is used as sample data, and each sample data is labeled with an operating mode label and a fault type label; based on all labeled sample data, a sample library of single switch open circuit faults of multiphase permanent magnet motor system is constructed.

4. The method according to claim 3, wherein, The time length of the multi-harmonic plane current vector trajectory diagram is not less than one complete electrical cycle, and the sampling frequency satisfies the Nyquist sampling theorem.

5. The method of claim 1, wherein, The shared feature extraction layer is used to extract common features from the input data. The shared feature extraction layer includes an input layer, a convolutional layer, and a pooling layer. Each output branch corresponds to a specific working mode and is used to perform fault classification tasks under each working mode. The output branch includes at least one fully connected layer, a global average pooling layer, or an output layer.

6. The method of claim 1, wherein, The number of output branches is equal to the total number of operating modes of the multiphase permanent magnet motor system to be diagnosed, and the structure of each output branch can be adjusted according to the characteristics of its corresponding operating mode.

7. The method of claim 1, wherein, The calculation of the loss function for the CNN model includes: The feature vector output by the shared feature extraction layer With the input mode information The splicing is performed to obtain a fused feature vector ; ； The output branch corresponding to the working mode k outputs ; wherein is a working mode label, k = 0, 1, , is the number of fault-tolerant working modes with one missing phase, is the number of fault-tolerant working modes with two missing phases, is the weight matrix of the GAP layer corresponding to the working mode k, is the bias vector of the working mode k; Calculate the classification loss function under working mode k: ； wherein, is the number of samples for working mode k, is the jth sample element of the mask vector, identifying whether sample j belongs to working mode k; yes, if ; no, if , is the indicator function; r is the failure type label in working mode k, is the total number of failure types under working mode k, is the true failure type label of sample j; Classification loss function based on working mode k Through formula The loss function of the CNN model is calculated. .

8. The method according to claim 7, wherein, The working mode information is a one-dimensional one-hot encoded vector with a length of p1+p2+1; wherein, there is only one element in the encoded vector that is 1, and the rest are all 0, which is used to uniquely identify the working mode of the current sample.

9. The method of claim 1, wherein, The CNN model also includes lightweight processing: Replacing fully connected layers with global average pooling layers reduces the dimensionality of the feature maps. ； wherein, is the output of the global average pooling layer for the th channel, is the number of feature values per channel, f is the feature value index, is the fth feature value for the th channel.

10. A cross-working-mode multiphase permanent magnet motor fault diagnosis system based on the cross-working-mode multiphase permanent magnet motor fault diagnosis method of any one of claims 1-9, characterized in that, It includes a model training module, as well as connected building and diagnostic modules; The construction module is used to build a sample library of single switch open circuit faults of multiphase permanent magnet motor system under various working modes, and each sample is labeled. The model training module is used to group the labeled samples according to their labels and integrate them into a training dataset for the CNN model. In the training process, a loss function of the CNN model is calculated; based on the loss function, global parameters of the CNN model are iteratively optimized through a back propagation algorithm, and the training of the CNN model is completed; The diagnostic module is configured to acquire operation data of the multiphase permanent magnet motor system to be diagnosed and a current working mode in which the multiphase permanent magnet motor system is located, input the operation data and the working mode label into the trained CNN model, activate a corresponding output branch according to the working mode label, and output a fault type label of a single switch tube open circuit in the current working mode.

Images