A double three-phase permanent magnet synchronous motor drive system open-circuit fault diagnosis method under limited label samples

Abstract

The application discloses a kind of open-circuit fault diagnosis methods of double three-phase permanent magnet synchronous motor drive system under limited label sample, comprising: first, the characteristics after system fault are analyzed to electrical characteristics, combine position angle and harmonic plane current trajectory as input fault characteristics, collect harmonic plane current signal under specific position angle and normalize to [‑1,1];Second, the application designs a kind of clustering-based semi-supervised learning algorithm, which can automatically calculate the number of fault categories contained in the data set.The final output of the algorithm is the optimal key feature point under each fault category and the corresponding safety radius;Finally, the application determines the final fault category by feature matching.Compared with the same data-driven method, which trains a large-scale neural network with a large amount of labeled data, the application can achieve high-precision diagnosis and detection of unknown faults with small computational effort under limited labeled samples, making it more suitable for practical engineering applications.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology for multiphase motor drive systems, specifically a method for diagnosing open-circuit faults in dual three-phase permanent magnet synchronous motor drive systems with limited labeled samples. Background Technology

[0002] Multiphase motors have many advantages such as low torque ripple, high power density, and high reliability. They are now widely used in fields with high requirements for reliability and power density, such as aerospace, electric vehicles, ships, and wind power generation.

[0003] Due to the complex application environments of real-world systems, improving the reliability of multiphase motor drive systems, especially dual three-phase permanent magnet synchronous motor drive systems, becomes particularly important. Fault diagnosis and fault-tolerant control are generally considered the main ways to improve the reliability of motor drive systems, and accurate fault diagnosis is a prerequisite for fault-tolerant control. Therefore, achieving accurate diagnosis of open-circuit faults in dual three-phase permanent magnet synchronous motor drive systems has significant practical implications.

[0004] With the continuous development of technologies such as machine learning, data-driven intelligent diagnostic methods have gradually attracted the attention of most researchers. However, existing data-driven intelligent diagnostic methods often require a large amount of labeled fault data from the actual system for training complex neural networks, which is impractical in most industrial automation applications. Manually labeling this data is very time-consuming and costly. In addition, existing data-driven intelligent diagnostic methods also suffer from problems such as high computational load and inability to detect unknown faults, making them difficult to apply in practice. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing similar methods mentioned in the background art and to provide a method for diagnosing open circuit faults in a dual three-phase permanent magnet synchronous motor drive system with limited labeled samples.

[0006] The technical solution adopted in this invention is: a method for diagnosing open-circuit faults in a dual three-phase permanent magnet synchronous motor drive system with limited tag samples, comprising the following steps:

[0007] Step 1: Collect the harmonic plane current at a specific position angle, normalize it to [-1,1], and construct the original dataset;

[0008] Step 2: Randomly divide the labeled samples in the original dataset into a training set and a validation set. The data in the training set is used to initialize the key feature points of each known fault category, and the data in the validation set is used to initialize the safety radius of each known fault category.

[0009] Step 3: Under the constraint of the safety radius, iteratively assign the unlabeled samples in the original dataset to the nearest fault category to determine the number of fault categories contained in the dataset;

[0010] Step 4: Iteratively update the optimal key feature points for each fault category by defining a clustering objective function. Update the corresponding safety radius based on the optimal key feature points for each fault category.

[0011] Step 5: Under the constraint of the safety radius, determine the final fault category based on the principle of closest feature matching distance.

[0012] Furthermore, system analysis demonstrates that the harmonic plane current described in step 1 is robust to changes in parameters and operating conditions, and can effectively distinguish all open-circuit faults. For phase A loss and switch open-circuit faults, based on fault mechanism and electrical characteristic analysis, the harmonic plane current after different faults approximately has the following expression:

[0013]

[0014] Among them, i x_OPF-A and i y_OPF-A i x_OSF-SA1 and i y_OSF-SA1 i x_OSF-SA2 and i y_OSF-SA2 These represent the harmonic plane currents after a phase loss fault in phase A, an open circuit in the upper switch, and an open circuit in the lower switch, respectively. θ represents the reference value of the motor's q-axis current. e It is the electrical angle of the rotor.

[0015] Furthermore, the normalization to [-1, 1] described in step 1 is achieved through the following formula:

[0016]

[0017] Where x and x normal These represent the original input fault features and the normalized input fault features, respectively. This is the reference value for the q-axis current of the motor.

[0018] Furthermore, the initialization of key feature points for each known fault category in step 2 is achieved by calculating the average value of key feature points of samples of the same fault category in the training set.

[0019] Furthermore, the initialization of the safety radius for each known fault category in step 2 is achieved through the following formula:

[0020]

[0021] Among them, SMi It is the safe radius for the i-th known fault category. It is the feature of the j-th sample in the verification set that belongs to the ith fault category. It is the feature of the j-th sample in the validation set that belongs to the remaining fault category, i.e., excluding fault category i, c i Key feature points representing the i-th known fault category, It is the number of samples in the validation set that belong to the i-th known fault category. It represents the number of samples in the validation set that belong to other fault categories. max() and min() represent functions that find the maximum and minimum values, respectively.

[0022] Furthermore, the determination of the number of fault categories in the dataset in step 3 is achieved in the following way:

[0023] First, the distances from unlabeled samples to key feature points of each known fault category are calculated. If the smallest distance exceeds the safety radius of the corresponding fault category, the sample is considered a new fault category. In the next iteration, a new fault category's key feature point is randomly selected from these unlabeled samples. This process is repeated until no new fault categories appear, thus determining the number of fault categories in the dataset.

[0024] Furthermore, the clustering objective function defined in step 4 is implemented through the following function:

[0025]

[0026] Where J is the defined clustering objective function, and a is the number of fault categories in the dataset. It is the number of unlabeled samples in the dataset. It is the j-th unlabeled sample in the dataset, z ji It is a binary variable used to determine unlabeled samples Whether it belongs to the i-th fault category, c i It is the key feature point of the i-th known fault category, where z and c are the elements z and c respectively. ji and c i The collective term for symbols.

[0027] Furthermore, the determination of the final fault category in step 5 is achieved through the following function:

[0028]

[0029] Where 'a' is the number of fault categories in the dataset, and 'x' is the number of fault categories in the dataset. n These are the key feature points of the normalized sample to be diagnosed, c i It is the key feature point of the i-th known fault category. The output should make ||x n -c i ||2 is the smallest i.

[0030] Compared with the prior art, the beneficial effects of the present invention are:

[0031] (1) It can be demonstrated by combining theoretical analysis that the combination of harmonic current and position angle used in this invention is robust to changes in parameters and operating conditions, and the extracted features have distinct differences among different types of faults, providing certain theoretical support for the proposed diagnostic method.

[0032] (2) The present invention can extract fault information by using both labeled and unlabeled samples in the dataset, which is more suitable for practical engineering applications.

[0033] (3) The computational load of the present invention is small, and online diagnosis can be achieved in a general controller.

[0034] (4) The present invention can detect unknown faults based on the safety radius, which can reduce the false diagnosis rate in practical applications. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of a dual three-phase permanent magnet synchronous motor drive system according to the present invention;

[0036] Figure 2 This is a flowchart of a specific implementation method of the present invention. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining this invention and not for limiting it; that is, the described embodiments represent only selected embodiments of this invention and not all embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0038] Figure 1This is a schematic diagram of a dual three-phase permanent magnet synchronous motor drive system studied in this invention. This invention first extracts robust fault features through theoretical analysis as input to the algorithm. Then, it designs a targeted comprehensive clustering algorithm to extract fault information from the unknown dataset to calculate the optimal key feature points for each fault category. The final fault category is determined through feature matching. Compared to similar methods, this invention can achieve high-precision fault diagnosis with fewer labeled samples and lower computational cost, making it very suitable for implementation in general embedded systems. More specifically, the specific implementation steps of the open-circuit fault diagnosis method for a dual three-phase permanent magnet synchronous motor drive system with limited labeled samples are as follows (see...). Figure 2 ):

[0039] Step 1: Collect the harmonic plane current at a specific position angle, normalize it to [-1,1], and construct the original dataset;

[0040] This invention demonstrates that the harmonic plane current after a fault is robust to changes in parameters and operating conditions, and can effectively distinguish all open-circuit faults. For phase A loss and open-circuit faults of the switching transistor, based on the fault mechanism and electrical characteristic analysis, the harmonic plane current after different faults approximately has the following expression:

[0041]

[0042] Among them, i x_OPF-A and i y_OPF-A i x_OSF-SA1 and i y_OSF-SA1 i x_OSF-SA2 and i y_OSF-SA2 These represent the harmonic plane currents after a phase loss fault in phase A, an open circuit in the upper switch, and an open circuit in the lower switch, respectively. θ represents the reference value of the motor's q-axis current. e This is the electrical angle of the rotor. The expressions for the harmonic plane currents corresponding to other phase faults can be obtained using the same approach.

[0043] Therefore, based on the expression for the harmonic plane current after a fault, the harmonic plane current corresponding to a suitable position angle can be selected as the input fault feature. In this embodiment, the harmonic plane currents corresponding to position angles of 0, π / 3, 2π / 3, π, 4π / 3, and 5π / 3 are selected as the input fault features.

[0044] After collecting the required harmonic plane currents, they need to be normalized before being used as samples in the original dataset. In this embodiment, normalization to [-1, 1] is achieved using the following function:

[0045]

[0046] Where x and x normalThese represent the original input fault features and the normalized input fault features, respectively. This is the reference value for the q-axis current of the motor.

[0047] In this embodiment, 60 samples were collected for each fault category, totaling 18 fault categories. Of the 60 samples from 17 fault categories, 20 samples were considered labeled, and the remaining 40 samples were unlabeled. The remaining 60 samples from the one fault category were all considered unlabeled.

[0048] Step 2: Randomly divide the labeled samples in the original dataset into a training set and a validation set. The data in the training set is used to initialize the key feature points of each known fault category, and the data in the validation set is used to initialize the safety radius of each known fault category.

[0049] First, the known fault categories can be determined from the labeled samples in the dataset. In this embodiment, 10 sets of 20 labeled samples from each known fault category are selected as the training set, and the remaining 10 sets are selected as the validation set.

[0050] Next, this invention needs to initialize the key feature points and corresponding safety radii for each known fault category. In this embodiment, the initialization of the key feature points for each known fault category is achieved by calculating the average value of the key feature points of samples of the same fault category in the training set.

[0051] In this embodiment, the safety radius corresponding to each known fault category is initialized using the following function:

[0052]

[0053] Among them, SM i It is the safe radius for the i-th known fault category. It is the feature of the j-th sample in the verification set that belongs to the ith fault category. It is the feature of the j-th sample in the validation set that belongs to the remaining fault category, i.e., excluding fault category i, c i Key feature points representing the i-th known fault category, It is the number of samples in the validation set that belong to the i-th known fault category. It represents the number of samples in the validation set that belong to other fault categories. max() and min() represent functions that find the maximum and minimum values, respectively.

[0054] Step 3: Under the constraint of the safety radius, iteratively assign the unlabeled samples in the original dataset to the nearest fault category to determine the number of fault categories contained in the dataset;

[0055] In this embodiment, the number of fault categories contained in the dataset is determined in the following way:

[0056] For each unlabeled sample in the original dataset, calculate its distance to key feature points of each known fault category. If the nearest distance is greater than the safety radius corresponding to that fault category, then the unlabeled sample is identified as a potential new fault category. After all unlabeled samples have been calculated, select the unlabeled sample with the largest nearest distance from the potential new fault categories as the key feature point of the new fault category. Repeat this process for the remaining unlabeled samples until no new fault categories appear.

[0057] Step 4: Iteratively update the optimal key feature points for each fault category by defining a clustering objective function. Update the corresponding safety radius based on the optimal key feature points for each fault category.

[0058] In this embodiment, the objective function J for clustering can be expressed as follows:

[0059]

[0060] Where J is the defined clustering objective function, and a is the number of fault categories in the dataset. It is the number of unlabeled samples in the dataset. It is the j-th unlabeled sample in the dataset, z ji It is a binary variable used to determine unlabeled samples Whether it belongs to the i-th fault category, c i It is the key feature point of the i-th known fault category, where z and c are the elements z and c respectively. ji and c i The collective term for symbols.

[0061] In this invention, the criterion for updating the optimal key feature points under each fault category is to minimize the aforementioned objective function. In this embodiment, the iterative update formula can be expressed as follows:

[0062]

[0063] Among them, c i It is the key feature point of the i-th known fault category. It is the j-th unlabeled sample in the dataset, z ji It is a binary variable used to determine unlabeled samples Does it belong to the i-th fault category? It represents the number of unlabeled samples in the dataset, and min() represents the function to find the minimum value.

[0064] Step 5: Under the constraint of the safety radius, determine the final fault category based on the principle of closest feature matching distance.

[0065] During online diagnostics, it is necessary to first determine whether a fault has occurred. In this embodiment, the fault indicator variable I is used. xy The amplitude of the harmonic plane current can be expressed as follows:

[0066]

[0067] Among them, i x and i y This represents the current in the motor's harmonic plane. A system fault is considered to have occurred when the calculated fault indicator variable exceeds a preset current threshold. In this embodiment, the current threshold is selected as 0.8A.

[0068] After a fault occurs, this invention begins to collect the necessary harmonic current sample information, and then uses the following function to determine the final fault category:

[0069]

[0070] Where 'a' is the number of fault categories in the dataset, and 'x' is the number of fault categories in the dataset. n These are the key feature points of the normalized sample to be diagnosed, c i It is the key feature point of the i-th known fault category. The output should make ||x n -c i ||2 is the smallest i.

[0071] When the calculated minimum distance exceeds the safe radius corresponding to that fault category, the system is considered to have experienced an unknown fault. In practical applications, this can significantly reduce the false diagnosis rate and improve system reliability.

[0072] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed in the specific embodiments and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be modified within the scope of the concept described in the present invention through the teachings and inspirations above or through technology or knowledge in related fields. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for open-circuit fault diagnosis of a dual three-phase permanent magnet synchronous motor drive system with limited label samples, characterized in that, The method comprises the following steps: Step 1: Collecting the harmonic plane current under a specific position angle, normalizing to [-1, 1], and constructing an original data set; Step 2: Randomly dividing the labeled samples in the original data set into a training set and a validation set, using the data in the training set to initialize the key feature points of each known fault category, and using the data in the validation set to initialize the safety radius of each known fault category; Step 3: Under the limitation of the safety radius, iteratively assigning the unlabeled samples in the original data set to the nearest fault category, thereby determining the number of fault categories contained in the data set; Step 4: Iteratively updating the optimal key feature points under each fault category by defining a clustering objective function, and updating the corresponding safety radius according to the optimal key feature points of each fault category; Step 5: Determining the final fault category based on the principle of matching the nearest feature under the limitation of the safety radius; The safety radius of each known fault category initialized in step 2 is realized by the following formula: where SM i is the safety radius of the ith known failure class, is the feature of the jth sample in the validation set belonging to the ith failure class, is the feature of the jth sample in the validation set belonging to the remaining failure classes, i.e., other than the ith failure class, i represents the key feature point of the ith known failure class, is the number of samples in the validation set belonging to the ith known failure class, is the number of samples in the validation set belonging to the remaining failure classes, max() and min() represent the functions of finding the maximum and minimum values, respectively. The clustering objective function J in step 4 can be expressed in the following form: where J is the defined clustering objective function, a is the number of fault categories contained in the data set, n ul is the number of unlabeled samples in the data set, is the jth unlabeled sample in the data set, z ji is a binary variable used to determine whether the unlabeled sample belongs to the ith fault category, c i is the key feature point of the ith known fault category, z and c are collectively referred to as the symbols of elements z ji and c i .

2. The open-circuit fault diagnosis method for a dual three-phase permanent magnet synchronous motor drive system with limited label samples according to claim 1, characterized in that, It can be proved by system analysis that the harmonic plane current in step 1 is robust to parameter and working condition changes, and can effectively distinguish all open circuit faults. For A-phase open circuit and switch tube open circuit faults, combined with fault mechanism and electrical characteristic analysis, the harmonic plane current after different faults has the following expressions: where i x_OPF-A and i y_OPF-A , i x_OSF-SA1 and i y_OSF-SA1 , i x_OSF-SA2 and i y_OSF-SA2 represent the harmonic plane currents after phase failure of phase A, open circuit of upper switch tube and open circuit of lower switch tube, respectively, represents the motor q-axis current reference value, θ e is the electrical angle of the rotor.

3. The open-circuit fault diagnosis method for a dual three-phase permanent magnet synchronous motor drive system with limited label samples according to claim 1, characterized in that, The normalization to [-1, 1] in step 1 is realized by the following formula: wherein x and x normal respectively represent the original input fault feature and the normalized input fault feature, is the motor q-axis current reference value.

4. The open-circuit fault diagnosis method for a dual three-phase permanent magnet synchronous motor drive system with limited label samples according to claim 1, characterized in that, The key feature points of each known fault category initialized in step 2 are realized by calculating the average value of the key feature points of the same fault category in the training set.

5. The open-circuit fault diagnosis method for dual three-phase permanent magnet synchronous motor drive system with limited label samples according to claim 1, characterized in that, The determination of the number of fault categories contained in the data set in step 3 is realized by the following method: First, calculate the distance from the unlabeled sample to the key feature points of each known fault category. If the smallest distance exceeds the safety radius of the corresponding fault category, consider this sample as a new fault category. In the next iteration, a key feature point of a new fault category will be randomly selected from these unlabeled samples. Iterate until no new fault category appears, thereby determining the number of fault categories contained in the data set.

6. The open-circuit fault diagnosis method of a dual three-phase permanent magnet synchronous motor drive system with limited label samples according to claim 1, characterized in that, The determination of the final fault category in step 5 is realized by the following function: wherein a is the number of fault categories contained in the data set, x n is the normalized key feature point of the sample to be diagnosed, c i is the key feature point of the i-th known fault category, indicates the output i that minimizes || x n - c i ||2.

Images