A fault detection method for high-voltage frequency converter

Abstract

This invention discloses a fault detection method for high-voltage frequency converters. The method utilizes a data acquisition module installed inside the high-voltage frequency converter to collect its operating parameters in real time. This invention relates to the field of data processing technology. The fault detection method establishes a fault diagnosis model by extracting data parameters from a database of similar high-voltage frequency converters. Historical fault data of the high-voltage frequency converters is then incorporated into the fault diagnosis model for training and optimization. The operating parameters of the high-voltage frequency converter requiring fault detection are then processed within the fault diagnosis model, outputting the fault diagnosis result. This allows for real-time monitoring of the possibility of equipment failure during data acquisition, rather than relying on human experience to determine faults during inspection. This improves both the efficiency and accuracy of fault detection.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a fault detection method for high-voltage frequency converters. Background Technology

[0002] The reference patent title is: A Fault Detection Method and System for High Voltage Frequency Converters (Patent Publication No.: CN117076955A, Patent Publication Date: 2023-11-17). The method includes: establishing a sample data set, which is a fault signal set; performing wavelet packet decomposition to extract voltage signal features; constructing an anomaly feature set; establishing an anomaly identification network, which is a general identification network; using the fault signal frequency as a general anomaly identification constraint to establish a general anomaly identification constraint sub-network; extracting equipment maintenance features; constructing a dedicated anomaly identification constraint and establishing a dedicated anomaly identification constraint sub-network; reconstructing the anomaly identification network; acquiring real-time signal data; preprocessing the data; and inputting it into the reconstructed anomaly identification network to obtain fault detection results. This method solves the technical problems of traditional methods, which suffer from poor fault identification accuracy due to complex circuit structures and signal characteristics, and which typically only address specific types of faults or specific equipment, lacking versatility and adaptability.

[0003] Based on the description in the above documents, the existing fault detection process for high-voltage frequency converters requires personnel to detect abnormal data and determine the specific fault. This method fails to achieve real-time monitoring and detection of faults simultaneously with the acquisition of real-time data, and to locate the time and location of the problem based on the detected fault. Therefore, this invention provides a fault detection method for high-voltage frequency converters. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a fault detection method for high-voltage frequency converters. This method solves the problem that existing fault detection processes for high-voltage frequency converters require personnel to detect abnormal data and determine the specific fault, failing to achieve real-time monitoring and detection of faults simultaneously with real-time data acquisition, and thus failing to pinpoint the time and location of the fault based on the detected fault.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a fault detection method for a high-voltage frequency converter, specifically comprising the following steps:

[0006] A1. Real-time acquisition of high-voltage frequency converter operating parameters is achieved through the acquisition module installed inside the high-voltage frequency converter.

[0007] A2. Extract data parameters of high-voltage frequency converters of the same type from the database to establish a fault diagnosis model, and introduce historical fault data of high-voltage frequency converters into the fault diagnosis model for training and optimization. At this time, the operating parameters of the high-voltage frequency converters that need to be detected are introduced into the fault diagnosis model for processing, and the fault diagnosis results are output.

[0008] A3. The fault diagnosis results are fed back to the display terminal in the form of a combination of text and charts for operators or maintenance personnel to view. At the same time, personnel can carry out fault diagnosis and repair operations based on the fault diagnosis results.

[0009] Preferably, the operation for establishing the fault diagnosis model in A2 is as follows:

[0010] a21. First, establish an initial product model by inputting the structure, functional characteristics, and historical parameters of the high-voltage frequency converter;

[0011] a22. Extract historical fault data from the database to form a training set, and then introduce the training set into the initial product model for training. The initial product model uses a convolutional neural network to extract and train the feature values ​​in the training set to obtain a fault diagnosis model.

[0012] a23. Furthermore, the fault diagnosis model can identify different types of faults by introducing the parameter data of the detection.

[0013] Preferably, the operation of introducing the high-voltage frequency converter operating parameters that need to be fault detected in A2 into the fault diagnosis model for processing is as follows:

[0014] B1. The operating parameters of the high-voltage frequency converter that needs to be fault detected are collected in real time and stored in the database. Then, the collected operating parameter data is preprocessed to form an initial dataset, labeled M.

[0015] B2. Based on the data acquisition module and the fault diagnosis model, establish a data interaction connection, input the dataset M into the fault diagnosis model, form a feature difference curve through numerical feature matching and extraction, and judge the fault problem based on the change of the feature difference curve.

[0016] B3. Transmit the fault diagnosis results to the display terminal, and trace the collection time and collection location data when the problem data was generated based on the fault diagnosis results.

[0017] Preferably, the preprocessing operation of the operating parameter data collected in B1 is as follows:

[0018] b11. Clean the collected operating parameter data to remove erroneous, incomplete, and duplicate data.

[0019] b12. Based on the cleaned data, perform classification operations, classify according to data type, and convert data of the same type to maintain consistent units;

[0020] b13. Collect the processed data to form a dataset M classified according to data type.

[0021] Preferably, the data connection between the acquisition module and the fault diagnosis model in B2 is established in the following way:

[0022] b21. The acquisition module sends a data reception signal to the signal receiving end of the fault diagnosis model, and the fault diagnosis model sends a feedback signal to the signal receiving end of the acquisition module after confirming that the signal has been received.

[0023] b22. At this point, after the acquisition module confirms that it has received the feedback signal, the data transmission connection between the acquisition module and the fault diagnosis model is established, and the data is transmitted in real time.

[0024] b23. During the data transmission process, the acquisition module continuously sends a sustaining signal to the fault diagnosis model. If the acquisition module fails to receive a feedback signal from the fault diagnosis model within the time threshold T after sending the signal, it indicates that the data transmission connection between the acquisition module and the fault diagnosis model has failed.

[0025] Preferably, the operation in B2 where the fault diagnosis model processes the dataset M to form a feature difference curve is as follows:

[0026] C1. Import the parameter data located at a certain time node from the data type of dataset M into the fault diagnosis model, and extract all the normal operating parameter data at that time node through the fault diagnosis model, and form the normal parameter dataset in the fault diagnosis model as N.

[0027] C2. Then, other parameter data in dataset M located at this time node are formed into a feature parameter dataset labeled as L, and the difference between the normal parameter dataset N and the corresponding parameter values ​​in the feature parameter dataset L is calculated.

[0028] The expression for calculating the difference is: P = p(L) n )-p(N n );

[0029] Where P is the difference between the values ​​of a subset of data in the feature parameter dataset L and the values ​​of a subset of data in the normal parameter dataset N, and p(Ln) represents the value corresponding to the subset of data in the feature parameter dataset, while Ln represents the nth feature parameter dataset in the sequential arrangement, p(Nn) represents the value corresponding to the subset of data in the normal parameter dataset, while Nn represents the nth normal parameter dataset in the sequential arrangement.

[0030] C3. According to the time node sequence during the work, establish coordinate axis points to form a feature difference curve for the parameter value difference between the normal parameter dataset N and the feature parameter dataset L.

[0031] Preferably, the operation in C3 to establish the coordinate axis based on the difference in parameter values ​​is as follows:

[0032] c31. Establish an X-axis based on the time nodes of data collection during the product's operation, in chronological order, and mark the axis nodes.

[0033] c32. The corresponding data types of the normal parameter dataset N and the feature parameter dataset L located under the coordinate axis nodes are taken as the Y-axis, and the range of parameter value differences between the normal parameter dataset N and the feature parameter dataset L are taken as the coordinate axis nodes.

[0034] c33. Mark the difference between the values ​​of the subset of data in the feature parameter dataset L and the values ​​of the subset of data in the normal parameter dataset N under the corresponding data type, and connect the marked points under the same data type in the order of time nodes to form a feature difference curve.

[0035] Preferably, the fault judgment operation based on the change of the feature difference curve in B2 is as follows:

[0036] D1. By inserting the fluctuation thresholds of the corresponding data types in the database and the normal parameter dataset N into the feature difference curve for comparison, and marking the fluctuation thresholds as ±Qm, where m represents different types of parameters;

[0037] D2. Based on the changes of the characteristic difference curve above and below the X-axis, insert a straight line y = ±Qm that intersects the characteristic difference curve. The characteristic difference curve area above y = Qm is the positive curve segment, while the characteristic difference curve area below y = -Qm is the negative curve segment.

[0038] D3. Statistically identify the types of parameters that are in the positive and negative curve segments at the same time point. Then, match the data in the fault diagnosis model with the types of parameters at the same time point to determine the type of fault.

[0039] Preferably, the starting point of the positive and negative curve segments in D2 is the abnormal parameter value, and the time node between the abnormal parameter values ​​is taken as the starting time node interval.

[0040] Preferably, the operation in D3 for the fault diagnosis model to determine the type of fault problem is as follows:

[0041] d31. First, select the data types that are within the normal parameter value range at the same time point. Then, match the fault problem datasets with the same data type values ​​that are within the normal parameter value range from the same time point of the fault diagnosis model and label them as R.

[0042] d32. Next, compare the parameter values ​​located in the positive and negative overshoot curve segments with the fault problem dataset R, and determine the corresponding fault problem based on the comparison results.

[0043] And the expression for determining the fault is:

[0044]

[0045] Where U represents the determined fault problem type, and (P→R) indicates that the difference between the parameters of the same type located in the positive and negative overshoot curve segments is compared with the difference in the fault problem dataset R. This means that two values, distinct from the results of the comparison, are generated based on the comparison results. If the R-values ​​of the compared fault problem datasets both fall within the range of positive and negative curve segments, then this type of parameter is also an abnormal parameter and is marked as 1. Conversely, if the R-values ​​of the compared fault problem datasets fall within the range of normal parameters, then this type of parameter is marked as 0. This means extracting and summarizing the parameter types marked as 1. S is the fault diagnosis result dataset formed by feedback of historical fault data in the fault diagnosis model. The fault diagnosis result is confirmed by matching the extracted and summarized parameter types with the parameter types in the fault diagnosis result dataset.

[0046] This invention provides a fault detection method for high-voltage frequency converters. Compared with existing technologies, it has the following advantages:

[0047] (1) The fault detection method of the high voltage frequency converter is to establish a fault diagnosis model by extracting data parameters of the same type of high voltage frequency converter from the database, and to introduce historical fault data of the high voltage frequency converter into the fault diagnosis model for training and optimization. At this time, the operating parameters of the high voltage frequency converter that needs to be fault detected are introduced into the fault diagnosis model for processing, and the fault diagnosis result is output. In this way, the possibility of equipment failure can be monitored in real time during the real-time data acquisition process, instead of relying on personnel experience to make judgments on fault problems during the inspection process. This improves the efficiency of fault detection and also improves the accuracy of detection.

[0048] (2) The fault detection method of the high voltage frequency converter compares the fluctuation threshold of the corresponding data type in the database and the normal parameter dataset with the feature difference curve, and obtains the parameter types of the positive and negative excess curve segments. Then, the parameter types are compared with the parameter types in the fault problem dataset at the same time node to obtain the fault diagnosis result. This reduces the impact of normal data fluctuation on the detection result and obtains the diagnosis result more accurately.

[0049] (3) The fault detection method of the high voltage frequency converter transmits the fault diagnosis results to the display terminal and traces the collection time and collection location data when the problem data is generated based on the fault diagnosis results. In this way, after the fault problem is discovered, the information and location of the fault problem are traced by the time and collection location data initially recorded by the abnormal parameter data generated by the fault problem, so as to facilitate the subsequent personnel to confirm the problem and perform corresponding maintenance. Attached Figure Description

[0050] Figure 1 This is a flowchart of the fault detection method of the present invention;

[0051] Figure 2 This is a flowchart illustrating the establishment of the fault diagnosis model of the present invention. Detailed Implementation

[0052] 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.

[0053] Please see Figure 1-Figure 2 This invention provides a technical solution: a fault detection method for a high-voltage frequency converter, specifically including the following steps:

[0054] A1. Real-time acquisition of high-voltage frequency converter operating parameters is achieved through the acquisition module installed inside the high-voltage frequency converter.

[0055] A2. Extract data parameters of high-voltage frequency converters of the same type from the database to establish a fault diagnosis model, and introduce historical fault data of high-voltage frequency converters into the fault diagnosis model for training and optimization. At this time, the operating parameters of the high-voltage frequency converters that need to be detected are introduced into the fault diagnosis model for processing, and the fault diagnosis results are output.

[0056] A3. The fault diagnosis results are fed back to the display terminal in the form of a combination of text and charts for operators or maintenance personnel to view. At the same time, personnel can carry out fault diagnosis and repair operations based on the fault diagnosis results.

[0057] This involves establishing a fault diagnosis model by extracting data parameters of similar high-voltage frequency converters from a database, and then training and optimizing the model by incorporating historical fault data of high-voltage frequency converters. The operating parameters of the high-voltage frequency converter requiring fault detection are then processed within the model, outputting the fault diagnosis results. This allows for real-time monitoring of the possibility of equipment failure during data acquisition, rather than relying on personnel experience to determine faults during inspections. This improves both the efficiency and accuracy of fault detection.

[0058] In this embodiment, the operation for establishing the fault diagnosis model in A2 is as follows:

[0059] a21. First, establish an initial product model by inputting the structure, functional characteristics, and historical parameters of the high-voltage frequency converter;

[0060] a22. Extract historical fault data from the database to form a training set, and then introduce the training set into the initial product model for training. The initial product model uses a convolutional neural network to extract and train the feature values ​​in the training set to obtain a fault diagnosis model.

[0061] a23. Furthermore, the fault diagnosis model can identify different types of faults by introducing the parameter data of the detection.

[0062] Convolutional neural networks (CNNs) are existing learning algorithms that include an input layer, hidden layers, and an output layer. They are used to learn from historical fault data of a product model, thereby forming a model that can be applied to actual data processing. After the data is input into the fault diagnosis model, the corresponding fault diagnosis results are obtained based on the extraction and matching of data features.

[0063] In this embodiment, the operation of introducing the high-voltage frequency converter operating parameters that need to be fault detected in A2 into the fault diagnosis model for processing is as follows:

[0064] B1. The operating parameters of the high-voltage frequency converter that needs to be fault detected are collected in real time and stored in the database. Then, the collected operating parameter data is preprocessed to form an initial dataset, labeled M.

[0065] B2. Based on the data acquisition module and the fault diagnosis model, establish a data interaction connection, input the dataset M into the fault diagnosis model, form a feature difference curve through numerical feature matching and extraction, and judge the fault problem based on the change of the feature difference curve.

[0066] B3. Transmit the fault diagnosis results to the display terminal, and trace the collection time and collection location data when the problem data was generated based on the fault diagnosis results.

[0067] When a fault diagnosis result is generated, the abnormal parameter data used in the fault diagnosis result is traced back to find the time when the abnormal parameter data was generated, and where the sensor was located to collect the data at that time, and where the data was collected. At the same time, based on the traced information, the approximate location of the fault is determined and confirmed and investigated by personnel.

[0068] The fault diagnosis results are transmitted to the display terminal, and the collection time and location data when the problem data was generated are traced based on the fault diagnosis results. In this way, after a fault is discovered, the information and location of the fault are traced by the time and location data initially recorded for the abnormal parameter data generated by the fault, which facilitates the subsequent personnel to confirm the problem and carry out corresponding maintenance.

[0069] In this embodiment, the preprocessing operation of the operating parameter data collected in B1 is as follows:

[0070] b11. Clean the collected operating parameter data to remove erroneous, incomplete, and duplicate data.

[0071] b12. Based on the cleaned data, perform classification operations, classify according to data type, and convert data of the same type to maintain consistent units;

[0072] b13. Collect the processed data to form a dataset M classified according to data type.

[0073] In this embodiment, the data connection between the acquisition module in B2 and the fault diagnosis model is established as follows:

[0074] b21. The acquisition module sends a data reception signal to the signal receiving end of the fault diagnosis model, and the fault diagnosis model sends a feedback signal to the signal receiving end of the acquisition module after confirming that the signal has been received.

[0075] b22. At this point, after the acquisition module confirms that it has received the feedback signal, the data transmission connection between the acquisition module and the fault diagnosis model is established, and the data is transmitted in real time.

[0076] b23. During the data transmission process, the acquisition module continuously sends a sustaining signal to the fault diagnosis model. If the acquisition module fails to receive a feedback signal from the fault diagnosis model within the time threshold T after sending the signal, it indicates that the data transmission connection between the acquisition module and the fault diagnosis model has failed.

[0077] In this embodiment, the operation of the fault diagnosis model in B2 to process the dataset M and form the feature difference curve is as follows:

[0078] C1. Import the parameter data located at a certain time node from the data type of dataset M into the fault diagnosis model, and extract all the normal operating parameter data at that time node through the fault diagnosis model, and form the normal parameter dataset in the fault diagnosis model as N.

[0079] C2. Then, other parameter data in dataset M located at this time node are formed into a feature parameter dataset labeled as L, and the difference between the normal parameter dataset N and the corresponding parameter values ​​in the feature parameter dataset L is calculated.

[0080] The expression for calculating the difference is: P = p(L) n )-p(N n );

[0081] Where P is the difference between the values ​​of a subset of data in the feature parameter dataset L and the values ​​of a subset of data in the normal parameter dataset N, and p(Ln) represents the value corresponding to the subset of data in the feature parameter dataset, while Ln represents the nth feature parameter dataset in the sequential arrangement, p(Nn) represents the value corresponding to the subset of data in the normal parameter dataset, while Nn represents the nth normal parameter dataset in the sequential arrangement.

[0082] C3. According to the time node sequence during the work, establish coordinate axis points to form a feature difference curve for the parameter value difference between the normal parameter dataset N and the feature parameter dataset L.

[0083] In this embodiment, the operation of establishing coordinate axes based on parameter numerical differences in C3 is as follows:

[0084] c31. Establish an X-axis based on the time nodes of data collection during the product's operation, in chronological order, and mark the axis nodes.

[0085] c32. The corresponding data types of the normal parameter dataset N and the feature parameter dataset L located under the coordinate axis nodes are taken as the Y-axis, and the range of parameter value differences between the normal parameter dataset N and the feature parameter dataset L are taken as the coordinate axis nodes.

[0086] c33. Mark the difference between the values ​​of the subset of data in the feature parameter dataset L and the values ​​of the subset of data in the normal parameter dataset N under the corresponding data type, and connect the marked points under the same data type in the order of time nodes to form a feature difference curve.

[0087] In this embodiment, the fault judgment operation based on the change of the feature difference curve in B2 is as follows:

[0088] D1. By inserting the fluctuation thresholds of the corresponding data types in the database and the normal parameter dataset N into the feature difference curve for comparison, and marking the fluctuation thresholds as ±Qm, where m represents different types of parameters;

[0089] D2. Based on the changes of the characteristic difference curve above and below the X-axis, insert a straight line y = ±Qm that intersects the characteristic difference curve. The characteristic difference curve area above y = Qm is the positive curve segment, while the characteristic difference curve area below y = -Qm is the negative curve segment.

[0090] D3. Statistically identify the types of parameters that are in the positive and negative curve segments at the same time point. Then, match the data in the fault diagnosis model with the types of parameters at the same time point to determine the type of fault.

[0091] In this embodiment, the starting point of D2 located between the positive and negative curve segments is the abnormal parameter value, and the time node located between the abnormal parameter values ​​is taken as the starting time node interval.

[0092] In this embodiment, the operation of the fault diagnosis model in D3 to determine the type of fault problem is as follows:

[0093] d31. First, select the data types that are within the normal parameter value range at the same time point. Then, match the fault problem datasets with the same data type values ​​that are within the normal parameter value range from the same time point of the fault diagnosis model and label them as R.

[0094] d32. Next, compare the parameter values ​​located in the positive and negative overshoot curve segments with the fault problem dataset R, and determine the corresponding fault problem based on the comparison results.

[0095] And the expression for determining the fault is:

[0096]

[0097] Where U represents the determined fault problem type, and (P→R) indicates that the difference between the parameters of the same type located in the positive and negative overshoot curve segments is compared with the difference in the fault problem dataset R. This means that two values, distinct from the results of the comparison, are generated based on the comparison results. If the R-values ​​of the compared fault problem datasets both fall within the range of positive and negative curve segments, then this type of parameter is also an abnormal parameter and is marked as 1. Conversely, if the R-values ​​of the compared fault problem datasets fall within the range of normal parameters, then this type of parameter is marked as 0. This means extracting and summarizing the parameter types marked as 1. S is the fault diagnosis result dataset formed by feedback of historical fault data in the fault diagnosis model. The fault diagnosis result is confirmed by matching the extracted and summarized parameter types with the parameter types in the fault diagnosis result dataset.

[0098] Specifically, by inserting the fluctuation thresholds of corresponding data types in the database and the normal parameter dataset into the feature difference curve for comparison, the types of parameters exceeding the curve segment positive and negative are obtained. Then, by comparing the types of parameters with the types of parameters in the fault problem dataset at the same time point, the fault diagnosis results are obtained. This reduces the impact of normal data fluctuations on the detection results and obtains more accurate diagnostic results.

[0099] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0100] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0101] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A fault detection method for a high-voltage frequency converter, characterized in that: Specifically, the following steps are included: A1. Real-time acquisition of high-voltage frequency converter operating parameters is achieved through the acquisition module installed inside the high-voltage frequency converter. A2. Extract data parameters of high-voltage frequency converters of the same type from the database to establish a fault diagnosis model, and introduce historical fault data of high-voltage frequency converters into the fault diagnosis model for training and optimization. At this time, the operating parameters of the high-voltage frequency converters that need to be detected are introduced into the fault diagnosis model for processing, and the fault diagnosis results are output. A3. The fault diagnosis results are fed back to the display terminal in the form of a combination of text and charts for operators or maintenance personnel to view. At the same time, personnel can carry out fault troubleshooting and repair operations based on the fault diagnosis results. The operation of introducing the high-voltage frequency converter operating parameters that require fault detection in A2 into the fault diagnosis model for processing is as follows: B1. The operating parameters of the high-voltage frequency converter that needs to be fault detected are collected in real time and stored in the database. Then, the collected operating parameter data is preprocessed to form an initial dataset, labeled M. B2. Based on the data acquisition module and the fault diagnosis model, establish a data interaction connection, input the dataset M into the fault diagnosis model, form a feature difference curve through numerical feature matching and extraction, and judge the fault problem based on the change of the feature difference curve. B3. Transmit the fault diagnosis results to the display terminal, and trace the collection time and collection location data when the problem data was generated based on the fault diagnosis results; The preprocessing operation for the operating parameter data collected in B1 is as follows: b11. Clean the collected operating parameter data to remove erroneous, incomplete, and duplicate data. b12. Based on the cleaned data, perform classification operations, classify according to data type, and convert data of the same type to maintain consistent units; b13. Collect the processed data to form a dataset M classified according to data type; The operation in B2 where the fault diagnosis model processes dataset M to form a feature difference curve is as follows: C1. Extract parameter data at a certain time point from the dataset M after it is classified according to data type and introduce it into the fault diagnosis model. Extract all normal operating parameter data at that time point through the fault diagnosis model and form a normal parameter dataset N from the normal operating parameter data in the fault diagnosis model. C2. Then, other parameter data in dataset M located at this time node are formed into a feature parameter dataset labeled as L, and the difference between the normal parameter dataset N and the corresponding parameter values ​​in the feature parameter dataset L is calculated. The expression for calculating the difference is: ; Where P is the difference between the values ​​of a subset of data in the feature parameter dataset L and the values ​​of a subset of data in the normal parameter dataset N, and p(Ln) represents the value corresponding to the subset of data in the feature parameter dataset, while Ln represents the nth feature parameter dataset in the sequential arrangement, p(Nn) represents the value corresponding to the subset of data in the normal parameter dataset, while Nn represents the nth normal parameter dataset in the sequential arrangement. C3. According to the time node sequence during the work, establish coordinate axis points to form a feature difference curve for the parameter value difference between the normal parameter dataset N and the feature parameter dataset L.

2. The fault detection method for a high-voltage frequency converter according to claim 1, characterized in that: The steps for establishing the fault diagnosis model in A2 are as follows: a21. First, establish an initial product model by inputting the structure, functional characteristics, and historical parameters of the high-voltage frequency converter; a22. Extract historical fault data from the database to form a training set, and then introduce the training set into the initial product model for training. The initial product model uses a convolutional neural network to extract and train the feature values ​​in the training set to obtain a fault diagnosis model. a23. Furthermore, the fault diagnosis model can identify different types of faults by introducing the parameter data of the detection.

3. The fault detection method for a high-voltage frequency converter according to claim 1, characterized in that: The data connection between the acquisition module and the fault diagnosis model in B2 is established in the following way: b21. The acquisition module sends a data reception signal to the signal receiving end of the fault diagnosis model, and the fault diagnosis model sends a feedback signal to the signal receiving end of the acquisition module after confirming that the signal has been received. b22. At this point, after the acquisition module confirms that it has received the feedback signal, the data transmission connection between the acquisition module and the fault diagnosis model is established, and the data is transmitted in real time. b23. During the data transmission process, the acquisition module continuously sends a sustaining signal to the fault diagnosis model. If the acquisition module fails to receive a feedback signal from the fault diagnosis model within the time threshold T after sending the signal, it indicates that the data transmission connection between the acquisition module and the fault diagnosis model has failed.

4. The fault detection method for a high-voltage frequency converter according to claim 3, characterized in that: The operation in C3 that establishes coordinate axes based on parameter numerical differences is as follows: c31. Establish an X-axis based on the time nodes of data collection during the product's operation, in chronological order, and mark the axis nodes. c32. The corresponding data types of the normal parameter dataset N and the feature parameter dataset L located under the coordinate axis nodes are taken as the Y-axis, and the range of parameter value differences between the normal parameter dataset N and the feature parameter dataset L are taken as the coordinate axis nodes. c33. Mark the difference between the values ​​of the subset of data in the feature parameter dataset L and the values ​​of the subset of data in the normal parameter dataset N under the corresponding data type, and connect the marked points under the same data type in the order of time nodes to form a feature difference curve.

5. The fault detection method for a high-voltage frequency converter according to claim 1, characterized in that: The fault determination operation based on the change of the feature difference curve in B2 is as follows: D1. By inserting the fluctuation thresholds of the corresponding data types in the database and the normal parameter dataset N into the feature difference curve for comparison, and marking the fluctuation thresholds as ±Qm, where m represents different types of parameters; D2. Based on the changes of the feature difference curve above and below the X-axis, insert a straight line y=±Qm that intersects the feature difference curve. The feature difference curve area above y=Qm ​​is the positive curve segment, while the feature difference curve area below y=-Qm is the negative curve segment. D3. Statistically identify the types of parameters that are in the positive and negative curve segments at the same time point. Then, match the data in the fault diagnosis model with the types of parameters at the same time point to determine the type of fault.

6. The fault detection method for a high-voltage frequency converter according to claim 5, characterized in that: The starting points of D2 located in the positive and negative curve segments are the abnormal parameter values, and the time nodes located between the abnormal parameter values ​​are taken as the starting time node intervals.

7. A fault detection method for a high-voltage frequency converter according to claim 6, characterized in that: The operation in D3 for determining the type of fault problem by the fault diagnosis model is as follows: d31. First, select the data types that are within the normal parameter value range at the same time point. Then, match the fault problem datasets with the same data type values ​​that are within the normal parameter value range from the same time point of the fault diagnosis model and label them as R. d32. Next, compare the parameter values ​​located in the positive and negative overshoot curve segments with the fault problem dataset R, and determine the corresponding fault problem based on the comparison results. And the expression for determining the fault is: ; Where U represents the determined fault problem type, and (P→R) indicates that the difference between the parameters of the same type located in the positive and negative overshoot curve segments is compared with the difference in the fault problem dataset R. This means that two values, distinct from the results of the comparison, are generated based on the comparison results. If the R-values ​​of the compared fault problem datasets both fall within the range of positive and negative curve segments, then this type of parameter is also an abnormal parameter and is marked as 1. Conversely, if the R-values ​​of the compared fault problem datasets fall within the range of normal parameters, then this type of parameter is marked as 0. This means extracting and summarizing the parameter types marked as 1. S is the fault diagnosis result dataset formed by feedback of historical fault data in the fault diagnosis model. The fault diagnosis result is confirmed by matching the extracted and summarized parameter types with the parameter types in the fault diagnosis result dataset.

Images