Fault detection method and system for high-voltage test equipment

Abstract

The invention relates to the field of electrical equipment fault detection, in particular to a fault detection method and system for high-voltage test equipment, and the method comprises the steps: preliminarily recognizing time periods in which transient faults possibly exist through frequency domain analysis, extracting voltage transient features in the time periods in combination with time domain analysis, and carrying out the fault confirmation through current features. And feature vectors at fault moments are constructed for dimensionality reduction processing and are used for classification model training. And finally, realizing real-time fault detection of the direct-current high-voltage generator based on the trained classification model. According to the method, the real-time transient fault detection of the direct-current high-voltage generator is realized through voltage and current feature extraction and SVM classification in a mode of combining frequency domain and time domain analysis.

Description

Technical Field

[0001] The present invention relates to the field of electrical equipment fault detection, and in particular to a fault detection method and system for high-voltage test equipment. Background Art

[0002] The primary functions of high-voltage testing equipment include testing the insulation strength, withstand voltage, and electric shock resistance of equipment to ensure it can withstand various voltage stresses in actual operation. A crucial component is the DC high-voltage generator, whose core components typically include a transformer, rectifier circuit, filter circuit, and control system. Compared to AC high-voltage generators, DC high-voltage generators are more effective in identifying minor defects and aging issues in insulation materials because the DC electric field more realistically simulates the voltage stress conditions experienced by the equipment in actual operation.

[0003] However, DC high-voltage generators can experience various issues during actual operation, such as unstable output voltage, overheating, and abnormal discharge. These issues not only affect the accuracy of test results but can also damage the equipment itself and even lead to safety incidents. Unstable output voltage is a common fault in DC high-voltage generators. Frequency domain analysis, a key method for detecting voltage fluctuation faults in existing technologies, typically converts time-domain voltage signals into frequency-domain signals through Fast Fourier Transform (FFT) to identify frequency components and abnormal patterns in voltage fluctuations.

[0004] For example, Chinese patent application publication number CN118534268A discloses an arc fault identification method, apparatus, device, and storage medium. The method extracts frequency domain feature information of an arc signal to be tested as input feature information; inputs the input feature information into at least two algorithm-based arc fault identification models to obtain classification results, which are then combined into a classification result set; and determines the fault detection result of the arc data to be tested based on the distribution of the classification results in the classification result set. However, this method still has some technical drawbacks. For example, because commonly used frequency domain transformations can only perform spectrum conversion on all data within the entire signal window, it cannot effectively capture and identify sudden transient changes. Summary of the Invention

[0005] In response to the above-mentioned problem of being unable to effectively capture transient changes, in the first aspect, the present invention proposes a fault detection method for high-voltage test equipment, including: obtaining a voltage sequence and a current sequence of a DC high-voltage generator; dividing the voltage sequence using windows; obtaining the high-frequency components of each window based on frequency domain analysis, and recording the windows with high-frequency components greater than a set high-frequency threshold as suspected fault windows; obtaining the time domain transient characteristics of the voltage at each sampling moment in each suspected fault window, the time domain transient characteristics are positively correlated with the voltage mean value in the corresponding suspected fault window and the difference between the voltage at the corresponding sampling moment, are positively correlated with the second-order derivative of the voltage at the corresponding sampling moment, and are positively correlated with the highest voltage in the corresponding suspected fault window. The time difference between the highest voltage peak and the lowest voltage trough is negatively correlated; the sampling moment corresponding to the time-domain transient feature greater than the set threshold is recorded as the suspected fault sampling moment; the normal current interval is obtained based on the peak mean and the trough mean of the current sequence; in response to the current at the suspected fault sampling moment being in the normal current interval, the corresponding suspected fault sampling moment is determined to be the transient fault sampling moment; the time-domain transient features of the voltage at the normal sampling moment and the transient fault sampling moment are obtained to obtain a feature matrix and add a label; the real-time voltage is obtained, the time-domain transient features of the real-time voltage are calculated and input into the trained classification model, and in response to the output result being a fault label, the staff is notified to perform maintenance.

[0006] This method uses frequency-domain analysis to initially identify possible fault time periods. It then uses time-domain transient signature analysis to pinpoint the likely time of the fault. A secondary verification based on current signatures ensures that the fault signal is distinguishable from fluctuations in normal equipment operation, accurately identifying the moment of transient fault occurrence. Finally, a classification model built using historical data enables real-time detection. Compared to existing technologies, this method more accurately identifies transient fault time points during detection, offering higher real-time performance and reliability. This significantly improves system stability, particularly in complex signal environments.

[0007] Furthermore, obtaining the high frequency components of each window based on frequency domain analysis also includes: using discrete wavelet transform DWT to decompose the voltage sequence of each window into multiple layers; obtaining the first In the window The high frequency energy ratio of the layer is recorded as , the specific method of obtaining it is:

[0008]

[0009] in Indicates the After the window wavelet transform Layer detail factor; Indicates the In the window sampling points; Indicates the The total number of decomposition layers of wavelet transform within a window; Indicates the After the window wavelet transform The high-frequency energy ratio of the layer; the high-frequency energy ratio is the high-frequency component of the window.

[0010] The discrete wavelet transform (DWT) is used to perform a multi-layer decomposition of the voltage sequence and extract the proportion of high-frequency energy. This method can effectively separate the high-frequency components in the signal, especially with good capture of sudden high-frequency fault signals. Compared with traditional frequency domain transform algorithms, the discrete wavelet transform can more accurately reflect high-frequency anomalies in local voltage signals, improving the screening accuracy of suspected fault windows.

[0011] Furthermore, the window with high frequency component greater than the set high frequency threshold is recorded as a suspected fault window, and the method further includes: obtaining the high frequency energy proportion of the highest layer in all time windows of the historical normal voltage sequence; recording the mean and standard deviation of the high frequency energy proportion of the highest layer as and ; The high frequency threshold is recorded as , the specific method of obtaining it is:

[0012] in represents the parameter adjustment coefficient; in response to the existence of a highest layer where the high-frequency energy ratio is greater than the high-frequency threshold, the corresponding window is recorded as a suspected fault window.

[0013] Furthermore, the calculation method of the time domain transient characteristics is specifically as follows:

[0014] in Indicates any suspected fault window Neidi Time domain transient characteristics of voltage at moment t; Indicates any suspected fault window Neidi The second derivative of the voltage at time t; Indicates any suspected fault window The first Voltage at the moment; It means the The suspected failure window at that time The mean voltage within Indicates any suspected fault window The time difference between the highest voltage peak and the lowest voltage trough.

[0015] By calculating the time-domain transient characteristics of the voltage, the fault signal changes are further amplified. By comprehensively judging multiple factors such as the voltage mean, second-order derivative, and the time difference between peaks and troughs, time-domain analysis effectively extracts the significant characteristics of the fault point. Combined with frequency-domain information, this further improves the accuracy of fault location.

[0016] Furthermore, the sampling moment corresponding to the time-domain transient feature greater than the set threshold is recorded as the suspected fault sampling moment, and the method also includes: using the maximum inter-class variance method to determine the adaptive threshold for all the acquired time-domain transient features; in response to the existence of a time-domain transient feature greater than the adaptive threshold, the corresponding sampling moment is recorded as the suspected fault sampling moment.

[0017] Furthermore, obtaining the time-domain transient characteristics of the voltage at all sampling moments to obtain a feature matrix and adding labels also includes: obtaining the time-domain transient characteristics of the voltage at all transient fault sampling moments to construct a first matrix, and adding a fault label after each eigenvector in the first matrix; obtaining a historical normal voltage sequence of a set length, and obtaining the time-domain transient characteristics of the voltage at all sampling moments in the historical normal voltage sequence to construct a second matrix, and adding a normal label after each eigenvector in the second matrix; merging the first matrix and the second matrix to obtain a feature matrix.

[0018] Furthermore, the feature matrix is ​​input into a classification model for training to obtain a trained classification model, which also includes: using PCA principal component analysis to reduce the dimension of the first matrix and the second matrix; inputting the reduced dimension labeled first matrix and second matrix into an SVM support vector machine for training to obtain a trained SVM classification model.

[0019] By reducing the dimensionality of the feature matrix through principal component analysis (PCA), we effectively reduced data redundancy, avoided overfitting, and increased the training speed of the classifier. Combined with support vector machines (SVMs) for model training, the model possesses stronger generalization capabilities in complex, high-dimensional data environments.

[0020] Furthermore, obtaining the voltage sequence and current sequence of the DC high-voltage generator also includes: installing a high-voltage mutual inductance sensor and a Hall current sensor at the output end of the DC high-voltage generator respectively; using an ADC module to connect the two sensors, collecting the output voltage and current of the DC high-voltage generator and storing them in a database; and obtaining the voltage sequence and current sequence of the DC high-voltage generator from the database.

[0021] Furthermore, it also includes performing data cleaning operations on the data stored in the database, and the data cleaning includes mean filtering noise reduction and Z-Score algorithm standardization.

[0022] Through data cleaning such as mean filtering noise reduction and Z-Score standardization, noise and abnormal data can be effectively removed, ensuring the reliability and consistency of the data, and providing a higher quality data foundation for subsequent analysis.

[0023] In a second aspect, the present invention provides a fault detection system for high-voltage test equipment, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the fault detection method for high-voltage test equipment of the present invention is implemented.

[0024] The technical effects of the present invention are: This invention achieves precise detection of transient faults through combined frequency and time domain analysis, combined with voltage and current signal characteristics. The method first identifies suspected fault windows through frequency domain analysis, then extracts the suspected fault time point using time domain transient characteristics, and finally confirms the fault using current characteristics. Real-time monitoring and alarming are achieved through a classification model trained on historical data. Compared to existing technologies, this invention achieves higher detection accuracy and real-time performance in complex signal environments, improving the safety and stability of equipment operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an illustrative and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein: Figure 1 is a flow chart schematically illustrating a fault detection method for high-voltage test equipment in an embodiment of the present invention; Figure 2 is a grayscale diagram schematically illustrating a DC high voltage generator according to an embodiment of the present invention; Figure 3 FIG. 1 is a block diagram schematically illustrating the structure of a fault detection system of a high-voltage test device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.

[0027] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0028] Example of a fault detection method for high voltage test equipment: like Figure 1 As shown, a fault detection method for high-voltage test equipment of the present invention includes: S1. Collect the historical output voltage and current of the DC high voltage generator.

[0029] like Figure 2 As shown, the DC high-voltage generator is mainly used to test the insulation of some power equipment. If the output voltage is unstable during operation, it may affect the test results of the power equipment. In particular, when the instantaneous voltage is too large, it may damage the power equipment being tested, resulting in unnecessary losses.

[0030] In this embodiment, a high-voltage mutual inductance sensor and a Hall current sensor are first installed at the output end of the DC high-voltage generator; then an ADC module with a sampling rate of 10kHz is used to connect the two sensors to collect the real-time output voltage and current of the DC high-voltage generator; finally, the voltage and current at each sampling moment are stored in a database for subsequent analysis and model training. In this embodiment, in addition to the sampling rate of 10kHz, the implementer can also select 8kHz, 12kHz or 14kHz according to actual conditions to ensure that transient fault signals can be captured. It should be noted that the data stored in the database need to undergo data cleaning operations, including mean filtering noise reduction and Z-Score algorithm standardization. The above two pre-processing algorithms are well-known technologies and will not be described here.

[0031] Through data cleaning such as mean filtering noise reduction and Z-Score standardization, noise and abnormal data can be effectively removed, ensuring the reliability and consistency of the data, and providing a higher quality data foundation for subsequent analysis.

[0032] S2. Perform frequency domain analysis on the voltage to obtain a suspected fault window.

[0033] Since the sampling frequency in this embodiment is high and the working time of the DC high voltage generator is usually not less than 30 minutes each time, the voltage and current data containing transient faults for 30 consecutive minutes are selected from the database to obtain the voltage sequence containing transient faults. and current sequence Then obtain the normal operation data of the DC high voltage generator without failure for 30 consecutive minutes from the database and extract the normal voltage sequence and normal current sequence .

[0034] When a DC high-voltage generator experiences a transient fault, it typically causes rapid changes and spikes in the voltage signal, thereby introducing some high-frequency components into the output voltage signal and increasing the high-frequency energy. Therefore, in this embodiment, the discrete wavelet transform (DWT) can be used to decompose the voltage sequence. The input to this algorithm is the voltage sequence, the wavelet function type, and the number of decomposition layers. The wavelet function can be selected as db4, which belongs to the Dobesi wavelet series and has good time-frequency localization characteristics, making it more suitable for transient signal analysis. The number of decomposition layers can be selected from the empirical value of 4 in this embodiment, but implementers can also select other values ​​based on actual conditions. The output of the algorithm is the approximation coefficient and detail coefficient of each layer after decomposition. The approximation coefficient captures the low-frequency trend of the signal, while the detail coefficient corresponds to the highest-frequency signal component, which best reflects the high-frequency changes of transient faults.

[0035] The discrete wavelet transform (DWT) is used to perform a multi-layer decomposition of the voltage sequence and extract the proportion of high-frequency energy. This method can effectively separate the high-frequency components in the signal, especially with good capture of sudden high-frequency fault signals. Compared with traditional frequency domain transform algorithms, the discrete wavelet transform can more accurately reflect high-frequency anomalies in local voltage signals, improving the screening accuracy of suspected fault windows.

[0036] For normal voltage sequence In this embodiment, a time window with an empirical value of 100 milliseconds can be selected for division, so each window contains The above discrete wavelet transform DWT algorithm is used to decompose the voltage time series in each time window to obtain the normal voltage sequence. The detail coefficients of each time window after wavelet transformation are calculated, and the proportion of high-frequency energy in the highest layer of the decomposition result is calculated. The following are some examples: Middle After wavelet transform of the time window The detail factor of the layer is , then After wavelet transform of the time window The high-frequency energy ratio of the layer is:

[0037]

[0038] in Indicates the After wavelet transform of the time window High-frequency energy of the layer; Indicates the In the time window sampling points; Indicates the The total number of decomposition layers of wavelet transform within a time window is 4 in this implementation; Indicates the After wavelet transform of the time window The high-frequency energy ratio of the layer.

[0039] Since this embodiment mainly analyzes the highest frequency component in the signal, the high frequency energy ratio of the highest layer after wavelet transform, that is, the fourth layer, is obtained here. Here we first calculate the normal voltage sequence The mean and standard deviation of the high-frequency energy proportion of the fourth layer after wavelet transformation in all time windows are recorded as and ; Then set the energy ratio threshold , the specific calculation method is:

[0040] in is the parameter adjustment coefficient, and in this embodiment, the empirical value 3 can be used. That is, when the voltage sequence containing transient faults After the time window is divided, if the high-frequency energy ratio of the fourth layer after wavelet transformation in a certain time window is greater than the energy ratio threshold , it means that there may be a transient fault in the time window. Example explanation: For a voltage sequence containing a transient fault , there is a time window , after wavelet transform, the high frequency energy ratio of the 4th layer is , and there are , then the time window There may be a transient fault within this time window, and this time window is recorded as a suspected fault window. It should be noted that the above voltage sequence The time window division method and the calculation method of high frequency energy ratio after time window wavelet transform are the same as those of normal voltage sequence. same.

[0041] In summary, the voltage series can be calculated The high-frequency energy proportion of the 4th layer after wavelet transformation in all time windows is compared with the energy proportion threshold Compare and get the voltage sequence including transient fault All suspected fault windows in the. In this embodiment, the use of high-frequency energy ratio as a comparison indicator for suspected fault windows has better adaptability than using high-frequency energy alone, which is mainly reflected in the following: when the load of the test object of the DC voltage generator changes, the output voltage will also change accordingly, and the voltage change at this time is normal, but if only the high-frequency energy of the current voltage change is evaluated with the high-frequency energy of the normal voltage, the voltage change time window caused by the load change may be misjudged as a suspected fault window; by comparing the high-frequency energy ratio of each time window, even if the voltage changes due to load changes, its high-frequency energy ratio will hardly change significantly, so the voltage change window caused by transient faults, that is, the suspected fault window, can be obtained more accurately.

[0042] S3. Perform voltage time domain analysis based on the suspected fault window to obtain the suspected fault sampling time.

[0043] Frequency domain analysis performs well in identifying high-frequency energy anomalies, but it can only capture the time period when the energy anomalies are concentrated, making it less accurate in locating the time of transient faults. Time domain analysis can further amplify and refine these transient characteristics, ensuring a low missed detection rate and improving overall detection sensitivity and accuracy.

[0044] In step S2, the voltage sequence containing the transient fault is obtained For each suspected fault window, the mean value of the voltage in the window can be calculated first, which is used to detect the deviation degree of the voltage in the window. Then, the time domain transient characteristics of the output voltage at each sampling moment are obtained. The specific acquisition process is as follows: For any suspected fault window The first At this moment, first obtain its transient amplitude ,have:

[0045] in Indicates a suspected fault window The first Voltage at the moment; It means the The suspected failure window at that time Then for the suspected fault window , get the maximum peak and minimum trough of the voltage in the window, respectively recorded as and , calculate the time difference between peak and trough ,have:

[0046] in Indicates the The suspected fault window at that time Maximum voltage peak The corresponding moment; It means the The suspected fault window at that time Minimum voltage valley The corresponding moment; is the normalization function. Then for the suspected fault window The first Voltage at the moment , calculate its voltage change acceleration ,have:

[0047] Finally get any suspected fault window Neidi Time domain transient characteristics of voltage at moment , specifically:

[0048] in Indicates a suspected fault window Neidi The voltage change acceleration of the voltage at the moment is greater, indicating that the DC high voltage generator is The more likely the output voltage at a moment is to change and the shorter the time it takes to change, the more likely a transient fault is to occur at that moment. The bigger; Indicates a suspected fault window The time difference between the voltage peak and valley within the voltage fluctuation period is the duration of the voltage fluctuation. The shorter the voltage fluctuation time within the suspected fault window at the moment, the more likely the DC high voltage generator is to have a transient fault. The bigger; Suspected fault window Neidi The transient amplitude of the voltage at the moment, which reflects the voltage signal based on the voltage mean When the transient amplitude is large, it may indicate the occurrence of a fault. The bigger.

[0049] At this point, the time domain transient characteristics of the voltage at each sampling moment in each suspected fault window are obtained. Here, all the obtained time domain transient characteristics are used to determine the adaptive time domain transient feature segmentation threshold using the maximum inter-class variance method, which is recorded as , and get all the values ​​greater than The sampling time corresponding to the time domain transient characteristics is constructed as the suspected transient fault time set ,Right now The maximum inter-class variance method mentioned above is a well-known technology and will not be described in detail here.

[0050] By calculating the time-domain transient characteristics of the voltage, the fault signal changes are further amplified. By comprehensively judging multiple factors such as the voltage mean, second-order derivative, and the time difference between peaks and troughs, time-domain analysis effectively extracts the significant characteristics of the fault point. Combined with frequency-domain information, this further improves the accuracy of fault location.

[0051] S4. Calculate the current characteristics of each suspected transient fault sampling point and determine the transient fault sampling time.

[0052] In step S3, the sampling times of all suspected transient faults are obtained. However, relying solely on the voltage signal can lead to false alarms because voltage fluctuations can be caused by non-fault factors, such as errors in the data acquisition equipment due to external environmental interference. Therefore, this embodiment subsequently incorporates the transient characteristics of the current signal to eliminate false faults caused by abnormal voltage signals alone and minimal current changes, thereby improving the robustness and reliability of the overall detection system.

[0053] For DC high voltage generators, the equipment they test usually includes high resistance products such as zinc oxide lightning arresters. Therefore, the current change of the DC high voltage generator is relatively small during normal operation. However, if a transient fault occurs in the DC high voltage generator, the current change is more obvious than during normal operation. In this embodiment, the derivative method is used to obtain the normal current sequence. All the peaks and troughs in the wave, and calculate the mean of all the peaks and record it as ; Calculate the mean of all troughs and record it as , get the floating range of normal current The above derivative method for obtaining peaks and troughs is a well-known technique and will not be described in detail here.

[0054] For the suspected transient fault time set obtained in step S3 , based on current sequences containing transient faults Get Collection The current corresponding to each suspected fault sampling moment in The current corresponding to the suspected fault sampling time is recorded as . Example: When Belongs to the floating range When , it indicates that the current is normal at the suspected fault sampling moment, that is, the abnormality in the time domain transient characteristics of the voltage at the suspected fault sampling moment may be caused by non-fault factors; when Not in the floating range When , it indicates that the current at the suspected fault sampling moment is abnormal, and the time domain transient characteristics of the voltage at the suspected fault sampling moment are also abnormal. It can be determined that the suspected fault sampling moment is the actual transient fault sampling moment.

[0055] In summary, based on the floating range of normal current For suspected transient fault time set The suspected fault sampling time in the is verified twice to obtain the confirmed transient fault sampling time.

[0056] S5. Obtain the feature vectors of each historical sampling moment to perform dimensionality reduction and model training to obtain a trained classification model.

[0057] For any transient fault sampling time , obtain its characteristic vector, including the voltage at the sampling moment ,in Indicates the time window where the sampling moment is located; the proportion of high-frequency energy of the wavelet transform in the time window The calculation method of the high-frequency energy ratio is shown in step S2; the transient amplitude of the voltage at the sampling moment is ; The time window where the sampling moment is located Voltage peak-valley time difference ; Voltage change acceleration at this sampling moment ; The current at this sampling moment In summary, any transient fault sampling moment There is a corresponding eigenvector , as shown below:

[0058] At this point, the characteristic vectors of all transient fault sampling moments are obtained to form the fault feature matrix and recorded as ; Then based on the normal voltage sequence and normal current sequence The same method as above is used to obtain the eigenvector of any normal sampling moment and form a normal eigenmatrix, which is recorded as .

[0059] Since high-dimensional feature data may have information redundancy, in this embodiment, PCA principal component analysis can be used to analyze the matrix and Processing to reduce feature dimensions and improve subsequent model training efficiency while retaining the matrix The input of the algorithm is the feature matrix and hyperparameters , used to control the retained principal components. This hyperparameter can be obtained by calculating the explained variance ratio and will not be described in detail in this embodiment. The output of the algorithm is the principal component matrix, principal component direction and eigenvalue after dimensionality reduction.

[0060] Fault feature matrix The principal component matrix after dimensionality reduction is recorded as , the normal characteristic matrix The principal component matrix after dimensionality reduction is recorded as ; where the matrix Contains the reduced-dimensional feature vector corresponding to any transient fault sampling moment, and adds a fault label to each reduced-dimensional feature vector in the matrix; and the matrix Contains the reduced-dimensional feature vector corresponding to any normal sampling moment, and adds a normal label to each reduced-dimensional feature vector in the matrix for subsequent model training.

[0061] In this embodiment, the SVM support vector machine is used to train the feature vectors after dimension reduction at each sampling moment. The input of the algorithm includes the matrix and , 80% of the data in each matrix is ​​used as the training set, and 20% of the data in each matrix is ​​used as the validation set. It should be noted that: Any feature vector in corresponds to a fault label, Any feature vector in the corresponds to a normal label. The algorithm input also includes key hyperparameters, where the regularization parameter can be taken as an empirical value of 10 in this embodiment, and the kernel function parameter can be taken as an empirical value of 0.1. The output of the algorithm is a trained SVM classification model.

[0062] By reducing the dimensionality of the feature matrix through principal component analysis (PCA), we effectively reduced data redundancy, avoided overfitting, and increased the training speed of the classifier. Combined with support vector machines (SVMs) for model training, the model possesses stronger generalization capabilities in complex, high-dimensional data environments.

[0063] S6. Obtain the real-time output voltage and current of the DC high-voltage generator; calculate the feature vector at the current sampling moment, and use the trained classification model to perform classification to obtain the detection result.

[0064] When the DC high-voltage generator is actually working, the voltage and current for a total of 100 milliseconds before the current sampling moment are obtained, the current time window is determined to calculate the eigenvector of the current sampling moment, the eigenvector of the current sampling moment is obtained using the method of step S5, and the principal component direction obtained by the PCA principal component analysis in step S5 is used to reduce its dimension. The reduced eigenvector is input into the trained SVM classification model obtained in step S5 to obtain the final fault detection result.

[0065] Exemplary description: There is a real-time sampling moment First, obtain the voltage and current data for 100 milliseconds before the sampling moment and determine the time window of the current sampling moment. ; Then calculate the sampling time based on the method of step S5 The eigenvector of ,have:

[0066] in Indicates the sampling time The voltage, Indicates the time window where the sampling moment is located; Indicates the sampling time Time window The proportion of high-frequency energy in the highest layer of the inner wavelet transform; Indicates the sampling time The transient amplitude of the voltage; Sampling time The time window The time difference between the voltage peak and trough; Indicates the sampling time Voltage change acceleration; Indicates the sampling time of current.

[0067] Then use the principal component direction obtained in step S5 to eigenvector Perform dimensionality reduction to obtain the eigenvector after dimensionality reduction ; Finally, the feature vector The trained SVM classification model in step S5 is input. If the output result of the classification model is normal, it indicates that no transient fault has occurred in the DC high-voltage generator at the current sampling moment. If the output result of the classification model is fault, it indicates that a transient fault may have occurred in the DC high-voltage generator at the current sampling moment. The equipment automatically stops operating and an alarm is triggered to prompt maintenance personnel to conduct further analysis of the equipment to determine the specific cause of the transient fault.

[0068] This method uses frequency-domain analysis to initially identify possible fault time periods. It then uses time-domain transient signature analysis to pinpoint the likely time of the fault. A secondary verification based on current signatures ensures that the fault signal is distinguishable from fluctuations in normal equipment operation, accurately identifying the moment of transient fault occurrence. Finally, a classification model built using historical data enables real-time detection. Compared to existing technologies, this method more accurately identifies transient fault time points during detection, offering higher real-time performance and reliability. This significantly improves system stability, particularly in complex signal environments.

[0069] Example of a fault detection system for high voltage test equipment: On the other hand, the present invention also provides a fault detection system for high voltage test equipment. Figure 3 As shown, the fault detection system of high-voltage test equipment includes a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, a fault detection method for high-voltage test equipment according to the first aspect of the present invention is implemented.

[0070] The fault detection system of the high-voltage test equipment also includes other components well known to those skilled in the art, such as a communication interface. The configuration and functions of these components are known in the art and will not be described in detail here.

[0071] In the present invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, the computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store the required information and can be accessed by an application, module, or both. Any such computer storage medium can be part of, accessible to, or connectable to a device. Any application or module described in the present invention can be implemented using computer-readable / executable instructions that can be stored or otherwise maintained by such a computer-readable medium.

[0072] In the description of this specification, "multiple" and "several" mean at least two, such as two, three or more, etc., unless otherwise clearly defined.

[0073] While several embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Numerous modifications, variations, and alternatives will occur to those skilled in the art without departing from the concept and spirit of the present invention. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention.

Claims

1. A fault detection method for high voltage test equipment, characterized in that: The method comprises: Obtaining the voltage sequence and current sequence of the DC high voltage generator; The voltage sequence is divided into windows; the high-frequency components of each window are obtained based on frequency domain analysis, and the windows with high-frequency components greater than a set high-frequency threshold are recorded as suspected fault windows; the time-domain transient characteristics of the voltage at each sampling moment in each suspected fault window are obtained, and the time-domain transient characteristics are positively correlated with the difference between the voltage mean in the corresponding suspected fault window and the voltage at the corresponding sampling moment, positively correlated with the second-order derivative of the voltage at the corresponding sampling moment, and negatively correlated with the time difference between the highest voltage peak and the lowest voltage trough in the corresponding suspected fault window; the sampling moment corresponding to the time-domain transient characteristics greater than the set threshold is recorded as the suspected fault sampling moment; A normal current interval is obtained based on the peak mean and the trough mean of the current sequence; in response to the current at the suspected fault sampling moment being in the normal current interval, the corresponding suspected fault sampling moment is determined to be a transient fault sampling moment; the time domain transient characteristics of the voltage at the normal sampling moment and the transient fault sampling moment are obtained to obtain a feature matrix and add labels; The real-time voltage is obtained, the time-domain transient characteristics of the real-time voltage are calculated and input into the trained classification model, and in response to the output result being a fault label, the staff is notified to perform maintenance.

2. A fault detection method for high voltage test equipment according to claim 1, characterized in that: Obtain the high-frequency components of each window based on frequency domain analysis, including: The voltage sequence of each window is decomposed into multiple layers using discrete wavelet transform DWT; In the window The high frequency energy ratio of the layer is recorded as , the specific method of obtaining is: ； ； in Indicates After the window wavelet transform Layer detail factor; Indicates In the window sampling points; Indicates The total number of decomposition layers of wavelet transform within a window; Indicates After the window wavelet transform The high-frequency energy proportion of the layer; the high-frequency energy proportion is the high-frequency component of the window.

3. A fault detection method for high voltage test equipment according to claim 2, characterized in that: The windows whose high-frequency components are greater than the set high-frequency threshold are recorded as suspected fault windows, including: Obtain the high-frequency energy proportion of the highest layer in all time windows of the historical normal voltage sequence; record the mean and standard deviation of the high-frequency energy proportion of the highest layer as and ; The high frequency threshold is recorded as , the specific method of obtaining is: ； in represents the parameter adjustment coefficient; In response to the presence of a highest layer where the high frequency energy ratio is greater than the high frequency threshold, the corresponding window is recorded as a suspected fault window.

4. A fault detection method for high voltage test equipment according to claim 1, characterized in that: The calculation method of the time domain transient characteristics is specifically as follows: ； in Indicates any suspected fault window Neidi Time domain transient characteristics of voltage at moment ; Indicates any suspected fault window Neidi The second derivative of the voltage at time t; Indicates any suspected fault window The first Voltage at the moment; It means the The suspected failure window at the time The mean voltage within Indicates any suspected fault window The time difference between the highest voltage peak and the lowest voltage trough.

5. A fault detection method for high voltage test equipment according to claim 1, characterized in that: The sampling time corresponding to the time domain transient characteristics greater than the set threshold is recorded as the suspected fault sampling time, including: All the acquired time-domain transient features are used to determine the adaptive threshold using the maximum inter-class variance method; In response to the presence of a time-domain transient feature greater than an adaptive threshold, the corresponding sampling moment is recorded as a suspected fault sampling moment.

6. A fault detection method for high voltage test equipment according to claim 1, characterized in that: Obtain the time-domain transient characteristics of the voltage at all sampling moments to obtain the feature matrix and add labels, including: Obtain the time-domain transient characteristics of the voltage at all transient fault sampling moments to construct a first matrix, and add a fault label after each characteristic vector in the first matrix; Obtain a historical normal voltage sequence of a set length, and obtain the time-domain transient characteristics of the voltage at all sampling moments in the historical normal voltage sequence to construct a second matrix, and add a normal label after each feature vector in the second matrix; The first matrix and the second matrix are combined to obtain the characteristic matrix.

7. A fault detection method for high voltage test equipment according to claim 6, characterized in that: Inputting the feature matrix into the classification model for training to obtain a trained classification model includes: Use PCA principal component analysis to reduce the dimension of the first matrix and the second matrix; The first matrix and the second matrix with labels after dimension reduction are input into the SVM support vector machine for training to obtain a trained SVM classification model.

8. A fault detection method for high voltage test equipment according to claim 1, characterized in that: Obtain the voltage sequence and current sequence of the DC high voltage generator, including: A high-voltage mutual inductance sensor and a Hall current sensor are respectively installed at the output end of the DC high-voltage generator; Use the ADC module to connect two sensors, collect the output voltage and current of the DC high voltage generator and store them in the database; The voltage sequence and current sequence of the DC high voltage generator are obtained from the database.

9. A fault detection method for high voltage test equipment according to claim 8, characterized in that: It also includes performing data cleaning operations on the data stored in the database, wherein the data cleaning includes mean filtering noise reduction and Z-Score algorithm standardization.

10. A system for fault detection of high voltage test equipment, characterized in that: It comprises a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the fault detection method for high-voltage test equipment according to any one of claims 1 to 9 is implemented.

Images