An ultra-high voltage direct current converter transformer implicit fault diagnosis system

Abstract

This invention discloses a latent fault diagnosis system for ultra-high voltage direct current converter transformers, relating to the field of transformer technology. The system includes a multi-dimensional acquisition terminal, a feature decoupling terminal, and a cross-domain collaborative terminal. In this invention, the deformation unit constructs a model using DC bias current and magnetic flux density; the heat dissipation unit uses Pearson correlation coefficient analysis to effectively identify pipeline faults; and early warning is achieved through mean calculation and change monitoring to ensure stable oil temperature. The insulation unit extracts features through wavelet packet multi-scale decomposition and diagnoses faults based on energy mutations, accurately capturing potential insulation problems. The mechanical unit constructs a delay coordinate matrix from winding vibration signals and uses the maximum Lyapunov exponent to quantize the trajectory separation rate to identify latent mechanical structure faults. The collaborative multi-dimensional approach of these units ensures equipment stability and improves fault diagnosis and maintenance efficiency.

Description

Technical Field

[0001] This invention relates to the field of transformer technology, and in particular to a latent fault diagnosis system for ultra-high voltage DC converter transformers. Background Technology

[0002] As the core equipment of the UHVDC transmission system, the UHVDC converter transformer undertakes key functions such as AC / DC power conversion and voltage level transformation. Its safe and stable operation is crucial to ensuring the power grid's transmission capacity and maintaining the reliability of the power system. With the widespread application of UHVDC transmission technology, converter transformers face complex operating conditions (such as DC bias, harmonic interference, and environmental magnetic field coupling), and latent faults (such as core magnetic characteristic degradation, minor winding deformation, and early insulation degradation) are gradually becoming hidden dangers threatening equipment safety.

[0003] Currently, in existing technologies, traditional monitoring methods mainly rely on single parameter threshold alarms, which are difficult to effectively identify early hidden faults. Such faults have progressive and coupled characteristics. For example, vibration abnormalities caused by iron core bias and insulation deterioration coexist, and are affected by strong electromagnetic interference and multi-physical field coupling. Conventional diagnostic methods are prone to missed reports or misjudgments.

[0004] Therefore, a latent fault diagnosis system for ultra-high voltage DC converter transformers is proposed to solve the above problems. Summary of the Invention

[0005] The main objective of this invention is to provide a latent fault diagnosis system for ultra-high voltage DC converter transformers to solve the problems mentioned in the background above.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: a latent fault diagnosis system for ultra-high voltage DC converter transformers, the system comprising a multi-dimensional acquisition terminal, a feature decoupling terminal, and a cross-domain collaborative terminal;

[0007] The multi-dimensional acquisition terminal is used to collect multi-modal data of the transformer body and the outside in real time through IoT sensors, and to perform preprocessing.

[0008] The characteristic decoupling terminal calculates the core power loss main frequency offset based on DC bias current and magnetic flux density parameters, calculates the Pearson correlation coefficient between DC bias current and oil temperature, performs wavelet packet decomposition on the converter valve voltage waveform and extracts frequency band energy, detects insulation defects based on energy mutation, and finally calculates the maximum Lyapunov exponent of the winding vibration signal to perform comprehensive diagnosis and early warning of multiple types of latent faults.

[0009] The cross-domain collaborative terminal constructs a characteristic correlation matrix between vibration and temperature. If the largest singular value is greater than the threshold of historical normal operating conditions, it is judged as a collaborative fault. Based on the partial discharge UHF signal and the environmental noise signal, it calculates the characteristic frequency energy ratio of discharge and noise, and combines the signal singular value decomposition to separate the fault component. When the energy ratio and the main energy of the fault component both exceed the set threshold, it is determined that there is an insulation degradation fault.

[0010] Preferably, the multi-dimensional acquisition terminal includes a main body unit, an external unit, and a preprocessing unit;

[0011] The main body unit is used to collect DC bias current data of the transformer through a current sensor, magnetic flux density parameters of the transformer through a magnetic flux density sensor, temperature data of transformer oil through an oil temperature sensor, voltage waveform data of the converter valve through a voltage sensor, vibration signal data of the winding through a vibration sensor, temperature data of the transformer hot spot location through a temperature sensor, and ultra-high frequency signal data of partial discharge inside the transformer through a partial discharge ultra-high frequency sensor.

[0012] The external unit is used by the noise sensor to collect noise signal data of the transformer's operating environment;

[0013] The preprocessing unit is used to perform noise reduction, outlier replacement, and normalization on the collected multimodal data.

[0014] Preferably, the feature decoupling end includes a deformation unit, a heat dissipation unit, an insulation unit, and a mechanical unit.

[0015] Preferably, the deformation unit constructs a correlation model between DC bias current data and magnetic flux density parameters, and then uses the finite element algorithm to fit and determine the proportional coefficient in the model. The model is then used to decouple and analyze the deviation and associated vibration characteristics of the iron core caused by bias, and the hidden faults of the iron core are identified based on the decoupling results.

[0016] Preferably, the heat dissipation unit uses the Pearson correlation coefficient analysis method to calculate the linear correlation between the preprocessed DC bias current data and oil temperature sampling data, and identifies transformer pipeline faults by analyzing the value and changes of the correlation coefficient.

[0017] Preferably, the insulation unit uses wavelet packet decomposition to decompose the voltage waveform of the converter valve into multiple scales to obtain signal characteristics at different levels and frequency bands. It calculates the energy value of the corresponding frequency band at each decomposition level, calculates the energy difference between adjacent moments by monitoring the changes in frequency band energy, analyzes the energy difference, and outputs corresponding fault diagnosis information.

[0018] Preferably, the mechanical unit constructs a delay coordinate matrix of the vibration signal based on the collected winding vibration signal and according to the delay time and embedding dimension, converting the one-dimensional vibration signal into a high-dimensional spatial sequence, and then analyzes the constructed delay coordinate matrix to calculate the maximum Lyapunov exponent, thereby quantifying the stability of the winding mechanical structure.

[0019] Preferably, the cross-domain collaborative terminal includes a differentiation unit, a first collaborative unit, a second collaborative unit, and an early warning unit. The differentiation unit is used to receive the fault judgment results from the deformation unit, the heat dissipation unit, the insulation unit, and the mechanical unit, and output the corresponding operation.

[0020] Preferably, the first collaborative unit constructs a feature correlation matrix based on the amplitude and frequency of the winding vibration signal and the temperature rise rate data of the transformer hot spot area, and incorporates the covariance of vibration amplitude and temperature rise rate, the covariance of vibration frequency and temperature rise rate, the covariance of vibration amplitude and its own frequency, and the variance information of temperature rise rate. Finally, it monitors the singular values ​​of the matrix, compares the singular values ​​with a set threshold, and outputs the judgment result.

[0021] Preferably, the second collaborative unit extracts the energy of the corresponding characteristic frequency based on the partial discharge UHF signal and the environmental noise signal through time-frequency transformation, calculates the energy ratio of the two, performs singular value decomposition on the signal, separates the discharge fault-related components, and calculates the energy ratio and the energy proportion of the fault component after decomposition. When the energy ratio and the main energy of the fault component both exceed the set threshold, it is determined that there is an insulation degradation fault.

[0022] The early warning unit provides corresponding early warning information based on the results of the first and second collaborative units.

[0023] The present invention has the following beneficial effects:

[0024] 1. In this invention, the deformation unit constructs a model using DC bias current and magnetic flux density, fits coefficients using the finite element algorithm, decouples the core bias deviation and vibration characteristics, and accurately identifies hidden faults. The heat dissipation unit, based on DC bias current and oil temperature data, uses Pearson correlation coefficient analysis to effectively identify pipeline faults. Early warning is achieved through mean calculation and change monitoring to ensure stable oil temperature. The insulation unit extracts features from the converter valve voltage waveform through wavelet packet multi-scale decomposition, diagnoses faults based on energy mutations, and accurately captures potential insulation hazards. The mechanical unit constructs a delay coordinate matrix from the winding vibration signal, quantifies the trajectory separation rate using the maximum Lyapunov exponent, and identifies hidden mechanical structure faults. All units work together in a multi-dimensional manner to ensure equipment stability and improve fault diagnosis and maintenance efficiency.

[0025] 2. In this invention, the intelligent scheduling of the distinguishing units is achieved. The first collaborative unit integrates multiple physical quantities such as vibration and temperature, and uses feature correlation matrices and singular value decomposition to discover coupled anomalies, thus solving the problem of missed detection under multi-physical field interference. The second collaborative unit integrates ultra-high frequency partial discharge and noise signals, and accurately captures the latent characteristics of insulation degradation through time-frequency transformation, energy ratio analysis, and fault component separation. The early warning unit responds in stages, triggers emergency shutdown for dual faults, and initiates special retesting for single-dimensional anomalies. Through multi-source data collaboration and cross-physical quantity coupling analysis, it effectively identifies progressive and coupled faults such as core bias and insulation degradation, solves the problem of missed detection and misjudgment under strong interference, and improves the coverage and accuracy of latent fault diagnosis. Attached Figure Description

[0026] Figure 1 This is a framework diagram of a latent fault diagnosis system for an ultra-high voltage DC converter transformer according to the present invention.

[0027] Figure 2 This is a flowchart of a latent fault diagnosis system for an ultra-high voltage DC converter transformer according to the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] Please see Figure 1 and Figure 2 The present invention provides a technical solution: a latent fault diagnosis system for ultra-high voltage DC converter transformers, the system comprising a multi-dimensional acquisition terminal, a feature decoupling terminal, and a cross-domain collaborative terminal;

[0030] The multi-dimensional acquisition terminal is used to collect multi-modal data of the transformer body and its external environment in real time through IoT sensors, and to perform preprocessing.

[0031] The characteristic decoupling end calculates the core power loss main frequency offset based on DC bias current and magnetic flux density parameters, calculates the Pearson correlation coefficient between DC bias current and oil temperature, performs wavelet packet decomposition on the converter valve voltage waveform and extracts frequency band energy, detects insulation defects based on energy mutation, and finally calculates the maximum Lyapunov exponent of the winding vibration signal to perform comprehensive diagnosis and early warning of multiple types of latent faults.

[0032] The cross-domain collaborative terminal constructs a characteristic correlation matrix between vibration and temperature. If the largest singular value is greater than the threshold of historical normal operating conditions, it is judged as a collaborative fault. Based on the partial discharge UHF signal and the environmental noise signal, the characteristic frequency energy ratio of discharge and noise is calculated. Combined with the signal singular value decomposition, the fault component is separated. When the energy ratio and the main energy of the fault component both exceed the set threshold, it is judged that there is an insulation degradation fault.

[0033] Specifically, the system first acquires core parameters such as DC bias current, magnetic flux density, oil temperature, and vibration signals, as well as external parameters such as environmental noise, through a multi-dimensional acquisition terminal. After denoising and normalization, standardized data is formed. The feature decoupling terminal performs main frequency offset analysis on the core parameters to detect magnetic characteristic degradation. It identifies heat dissipation anomalies by using the correlation coefficient between oil temperature and bias current, detects frequency band energy abrupt changes in insulation materials using wavelet packet decomposition, and evaluates the winding mechanical stability by using the maximum Lyapunov exponent. The cross-domain collaborative terminal constructs a correlation matrix between vibration signals and temperature data, verifies fault correlation through singular value analysis, and performs secondary confirmation of insulation status by combining the energy ratio of partial discharge signals and noise.

[0034] The multi-dimensional acquisition terminal includes a main body unit, an external unit, and a preprocessing unit;

[0035] The main unit is used to collect DC bias current data of the transformer through current sensor, magnetic flux density parameter of the transformer through magnetic flux density sensor, temperature data of transformer oil through oil temperature sensor, voltage waveform data of converter valve through voltage sensor, vibration signal data of winding through vibration sensor, temperature data of transformer hot spot location through temperature sensor, and UHF signal data of partial discharge inside transformer through UHF sensor.

[0036] Specifically, during the data acquisition process, the sampling times are strictly aligned.

[0037] The external unit is used by the noise sensor to collect noise signal data of the transformer's operating environment;

[0038] The preprocessing unit is used to perform noise reduction, outlier replacement, and normalization on the acquired multimodal data.

[0039] The decoupling features include deformation units, heat dissipation units, insulation units, and mechanical units.

[0040] Based on DC bias current data and magnetic flux density parameters, the deformation unit constructs a correlation model between the two. Then, with the help of the finite element algorithm, the proportional coefficient in the model is determined by fitting. The model is used to decouple the deviation and related vibration characteristics of the iron core caused by bias. Based on the decoupling results, the hidden faults of the iron core are identified.

[0041] Specifically, the association model is as follows: ;

[0042] in, This represents the core power frequency offset. Represents the proportionality coefficient. Represents DC bias magnetic flux density. Represents alternating magnetic flux density, Represents DC bias current;

[0043] The transformer core was modeled using the finite element method, and the core vibration and dominant frequency shift under different DC bias currents, DC bias flux densities, and AC flux densities were calculated. The results were then fitted to obtain... The value;

[0044] The real-time collected DC bias current, DC bias magnetic flux density, and AC magnetic flux density are fitted with the obtained data. Substitute the values ​​into the correlation model to calculate the core power frequency offset.

[0045] Based on the actual vibration monitoring of the core power frequency offset and the calculated core power frequency offset, the core magnetic bias deviation is decoupled. The calculation formula is as follows:

[0046] ;

[0047] in, This represents the actual frequency offset of the iron core power frequency during vibration monitoring. This represents the offset of the core power frequency. This represents the core magnetic bias deviation obtained from decoupling;

[0048] when If the deviation exceeds the threshold, it is determined to be a latent fault in the iron core, where the deviation threshold is determined based on historical fault data; otherwise, it is determined to be a normal state.

[0049] Based on the preprocessed DC bias current data and oil temperature sampling data, the heat dissipation unit uses the Pearson correlation coefficient analysis method to calculate the degree of linear correlation between the two, and identifies transformer pipeline faults by analyzing the value and changes of the correlation coefficient.

[0050] Specifically, after averaging the DC bias current data and oil temperature sampling data, the linear correlation between the two is calculated using the following formula:

[0051] ;

[0052] Here, Pearson correlation coefficient represents the relationship between current and oil temperature. Representing the DC bias current at time , Representing the Oil temperature at any time, This represents the mean value of the DC bias current data. The average value of the oil temperature sampling data;

[0053] when Less than the association threshold or Decrease in one hour ( If the threshold value is greater than or equal to the descent threshold, it is judged as a hidden pipeline fault; where the correlation threshold and the descent threshold are both calibrated based on historical fault data.

[0054] The insulation unit uses wavelet packet decomposition to decompose the voltage waveform of the converter valve into multiple scales, obtaining signal characteristics at different levels and frequency bands. It calculates the energy value of the corresponding frequency band at each decomposition level, calculates the energy difference between adjacent moments by monitoring the changes in frequency band energy, analyzes the energy difference, and outputs corresponding fault diagnosis information.

[0055] Specifically, the voltage waveform signal undergoes multi-scale decomposition processing, and the calculation formula is as follows:

[0056] ;

[0057] in, Represents voltage waveform signal, Represents the number of decomposition levels. Representing the Number of frequency bands in the layer Representing the Layer Signal characteristics of each frequency band;

[0058] For the first Layer The signal characteristics of each frequency band are analyzed, and the energy value of that frequency band is calculated using the following formula:

[0059] ;

[0060] in, Representing the Layer Energy value of each frequency band Represents the number of decomposition levels. Representing the Number of frequency bands in the layer Represents the total number of sampling points for a discrete signal. Represents the sampling point number;

[0061] The energy difference between adjacent time points is calculated using the following formula:

[0062] ;

[0063] in, Represents the energy difference between adjacent moments. Represents a time interval, when If the energy difference is greater than twice the average energy difference under normal operating conditions, it indicates an energy mutation and a fault in the corresponding frequency band.

[0064] Based on the collected winding vibration signal, the mechanical unit constructs a delay coordinate matrix of the vibration signal according to the delay time and embedding dimension, converting the one-dimensional vibration signal into a high-dimensional spatial sequence. Then, the constructed delay coordinate matrix is ​​analyzed to calculate the maximum Lyapunov exponent, which is used to quantify the stability of the winding mechanical structure.

[0065] Specifically, the delay coordinate matrix is:

[0066] ;

[0067] in, The time-domain sequence representing the vibration signal. Represents a time variable. Represents the delay time. Represents the embedding dimension. Represents the total number of sampling points for a discrete signal;

[0068] Analyze adjacent trajectories in the delay coordinate matrix and calculate the maximum Lyapunov exponent using the following formula:

[0069] ;

[0070] in, Representing the maximum Lyapunov exponent, quantifying the separation rate of adjacent trajectories. Represents evolution time, Represents the number of trajectory pairs. and represent Adjacent trajectories at different times and Represents adjacent trajectories at the initial moment;

[0071] The separation threshold is determined by calibrating historical data. If the value is greater than the separation threshold, it indicates a hidden fault in the mechanical structure; otherwise, it indicates normal operation.

[0072] The cross-domain collaborative terminal includes a differentiation unit, a first collaborative unit, a second collaborative unit, and an early warning unit. The differentiation unit is used to receive the fault results from the deformation unit, the heat dissipation unit, the insulation unit, and the mechanical unit, and output the corresponding operation.

[0073] If any one of the deformation unit, heat dissipation unit, and mechanical unit is determined to be faulty, the system enters the first coordination unit; if the insulation unit is determined to be faulty, the system enters the second coordination unit.

[0074] The first collaborative unit constructs a feature correlation matrix based on the amplitude and frequency of the winding vibration signal and the temperature rise rate data of the transformer hot spot area. It incorporates the covariance of vibration amplitude and temperature rise rate, the covariance of vibration frequency and temperature rise rate, the covariance of vibration amplitude and its own frequency, and the variance of temperature rise rate. Finally, it monitors the singular values ​​of the matrix, compares the singular values ​​with the set threshold, and outputs the judgment result.

[0075] Specifically, the constructed feature correlation matrix is ​​as follows:

[0076] ;

[0077] in, Represents the sequence of winding vibration amplitudes. Represents the winding vibration frequency sequence. Represents the oil temperature rise rate sequence. Represents the covariance of X and Y. The covariance representing the rate of oil temperature rise;

[0078] The feature correlation matrix is ​​decomposed using Python NumPy. All singular values ​​are iterated through, and the largest singular value is selected and compared with the threshold of the historical normal operating condition. If the largest singular value is greater than the threshold of the historical normal operating condition, it is judged as a cooperative fault.

[0079] The second collaborative unit extracts the energy of the corresponding characteristic frequency based on the partial discharge ultra-high frequency signal and the environmental noise signal through time-frequency transformation, calculates the energy ratio between the two, and then performs singular value decomposition on the signal to separate the discharge fault-related components. It also calculates the energy ratio and the energy proportion of the fault component after decomposition. When both the energy ratio and the main energy of the fault component exceed the set threshold, it is determined that there is an insulation degradation fault.

[0080] Specifically, for the preprocessed partial discharge UHF signal and environmental noise signal, the time-frequency domain matrices of the partial discharge signal and environmental noise signal are obtained through short-time Fourier transform. Then, the energy of the local environmental noise signal and the partial discharge signal at the corresponding characteristic frequencies is extracted, and the energy ratio between the two is calculated. The calculation formula is as follows:

[0081] ;

[0082] in, Represents the energy ratio, This represents the energy of the extracted local environmental noise signal at the corresponding characteristic frequency. This represents the energy of the partial discharge signal at the corresponding characteristic frequency.

[0083] Simultaneously, a signal matrix is ​​constructed using time as rows, frequency as columns, and corresponding energy values ​​as matrix values. Singular value decomposition (SVD) is then performed, retaining the two largest singular values ​​to reconstruct the fault components and calculate their proportions.

[0084] ;

[0085] Through energy ratio and the proportion of fault components Compare with the corresponding energy threshold and component threshold; if the energy ratio is... and the proportion of fault components If both the energy threshold and component threshold are greater than the corresponding energy threshold and component threshold, it is judged as an insulation degradation fault.

[0086] The early warning unit provides corresponding early warning information based on the results of the first and second collaborative units;

[0087] When the deformation unit, heat dissipation unit, and mechanical unit are determined to be faulty, and the first coordinating unit is also determined to be faulty, or when both the insulation unit and the second coordinating unit are determined to be faulty, an emergency shutdown is triggered and a remote warning is issued.

[0088] When the deformation unit, heat dissipation unit, and mechanical unit are identified as faulty, and the first coordinating unit is identified as normal, or when the insulation unit is identified as faulty but the second coordinating unit is identified as normal, the unit-level fault diagnosis process is triggered, and high-precision sensors are automatically scheduled to perform fixed-point retesting on the abnormal unit.

[0089] In this invention, a latent fault diagnosis system for ultra-high voltage DC converter transformers is disclosed. The main unit comprehensively collects multi-modal data, including electrical, magnetic, thermal, mechanical, and insulation characteristics, through various sensors. The external unit collects environmental noise to supplement interference information. The preprocessing unit performs noise reduction, outlier replacement, and normalization to ensure data quality. The sampling time is strictly aligned to ensure time sequence consistency, providing comprehensive, accurate, and synchronous data support for subsequent diagnosis and laying a solid data foundation for fault diagnosis.

[0090] The deformation unit constructs a correlation model using DC bias current and magnetic flux density parameters, and then combines the finite element algorithm to fit the scaling factor to decouple and analyze the deviation and associated vibration characteristics of the iron core caused by bias. On the one hand, it can accurately identify hidden faults in the iron core. Through the constructed correlation model and decoupling algorithm, it can detect subtle anomalies caused by bias in the iron core in advance, avoid the expansion of faults, and ensure the stable operation of the transformer. On the other hand, based on the comparison of the iron core power frequency offset calculated by actual vibration monitoring, the iron core bias deviation is decoupled, making fault judgment more scientific and accurate, and providing clear fault identification basis for operation and maintenance personnel.

[0091] The heat dissipation unit uses Pearson correlation coefficient analysis based on preprocessed DC bias current and oil temperature sampling data. By calculating the linear correlation between the two, it can effectively identify transformer pipeline faults, promptly detect heat dissipation abnormalities caused by pipeline blockage or leakage, ensure stable transformer oil temperature, and avoid serious faults such as insulation damage caused by overheating. At the same time, by calculating the mean and analyzing the changes in correlation coefficients of the data, it can realize early warning of hidden pipeline faults. Based on historical fault data, the threshold is calibrated to make fault identification more in line with actual operating scenarios, improve fault prediction capability, and ensure that the transformer operates in a reasonable temperature environment.

[0092] The insulation unit employs wavelet packet decomposition for multi-scale processing of the converter valve voltage waveform. This multi-scale decomposition accurately extracts signal features from different levels and frequency bands, comprehensively capturing subtle changes in the voltage waveform. This provides rich evidence for insulation defect detection and can effectively identify latent faults such as insulation aging and partial discharge. Simultaneously, it calculates the energy of each frequency band and the energy difference between adjacent moments, judging faults based on energy mutations. This makes insulation fault diagnosis more targeted and sensitive, enabling timely detection in the early stages of faults and avoiding serious consequences such as large-scale power outages caused by insulation faults. Furthermore, by quantifying energy changes, it outputs clear fault diagnosis information, helping maintenance personnel quickly locate insulation problems.

[0093] The mechanical unit constructs a delay coordinate matrix based on the winding vibration signal and calculates the maximum Lyapunov exponent, converting the one-dimensional vibration signal into a high-dimensional spatial sequence. This allows for a more comprehensive mining of the mechanical structure information contained in the signal, capturing the complex characteristics of winding vibration, and accurately reflecting the stability of the mechanical structure. At the same time, the maximum Lyapunov exponent can quantify the separation rate of adjacent trajectories, effectively identifying hidden faults in the mechanical structure, such as winding loosening and deformation, and providing early warning of mechanical fault risks. By calibrating the separation threshold through historical data, the fault judgment criteria are made clear, avoiding more serious problems such as winding damage caused by mechanical faults.

[0094] The system employs a differentiated intelligent scheduling mechanism. The first collaborative unit integrates multiple physical quantities, including vibration and temperature, and leverages feature correlation matrices and singular value decomposition to uncover coupled anomalies, thus solving the problem of missed detections under multi-physical field interference. The second collaborative unit integrates ultra-high frequency partial discharge and noise signals, and through time-frequency transformation, energy ratio analysis, and fault component separation, accurately captures latent characteristics of insulation degradation. The early warning unit provides tiered responses, with dual faults triggering emergency shutdowns and single-dimensional anomalies initiating specialized retests. Through multi-source data collaboration and cross-physical quantity coupling analysis, it effectively identifies progressive and coupled faults such as core bias and insulation degradation, resolving the problem of missed detections and misjudgments under strong interference, and improving the coverage and accuracy of latent fault diagnosis.

[0095] 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 a process, method, article, or apparatus.

[0096] 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 variations 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 latent fault diagnosis system for ultra-high voltage DC converter transformers, characterized in that, The system includes a multi-dimensional acquisition terminal, a feature decoupling terminal, and a cross-domain collaboration terminal; The multi-dimensional acquisition terminal is used to collect multi-modal data of the transformer body and the outside in real time through IoT sensors, and to perform preprocessing. The feature decoupling terminal calculates the core power loss main frequency offset based on DC bias current and magnetic flux density parameters, analyzes and identifies core latent faults, calculates the Pearson correlation coefficient between DC bias current and oil temperature, analyzes and identifies transformer pipeline faults, performs wavelet packet decomposition on converter valve voltage waveform and extracts frequency band energy, detects insulation defects based on energy mutation, analyzes and identifies frequency band faults, and finally calculates the maximum Lyapunov exponent of winding vibration signal to analyze and identify mechanical structure latent faults, and performs comprehensive diagnosis and early warning of multiple types of latent faults. The cross-domain collaborative terminal constructs a characteristic correlation matrix between vibration and temperature. If the largest singular value is greater than the threshold of historical normal operating conditions, it is judged as a collaborative fault. Based on the partial discharge UHF signal and the environmental noise signal, it calculates the characteristic frequency energy ratio of discharge and noise, and combines the signal singular value decomposition to separate the fault component. When the energy ratio and the main energy of the fault component both exceed the set threshold, it is determined that there is an insulation degradation fault.

2. The latent fault diagnosis system for ultra-high voltage DC converter transformers according to claim 1, characterized in that, The multi-dimensional acquisition terminal includes a main body unit, an external unit, and a preprocessing unit; The main body unit is used to collect DC bias current data of the transformer through a current sensor, magnetic flux density parameters of the transformer through a magnetic flux density sensor, temperature data of transformer oil through an oil temperature sensor, voltage waveform data of the converter valve through a voltage sensor, vibration signal data of the winding through a vibration sensor, temperature data of the transformer hot spot location through a temperature sensor, and ultra-high frequency signal data of partial discharge inside the transformer through a partial discharge ultra-high frequency sensor. The external unit is used by the noise sensor to collect noise signal data of the transformer's operating environment; The preprocessing unit is used to perform noise reduction, outlier replacement, and normalization on the collected multimodal data.

3. The latent fault diagnosis system for ultra-high voltage DC converter transformers according to claim 1, characterized in that, The decoupling end features include a deformation unit, a heat dissipation unit, an insulation unit, and a mechanical unit.

4. The latent fault diagnosis system for ultra-high voltage DC converter transformers according to claim 3, characterized in that, The deformation unit constructs a correlation model based on DC bias current data and magnetic flux density parameters. Then, using the finite element algorithm, it fits and determines the scaling factor in the model. This model is used to decouple and analyze the deviation and associated vibration characteristics of the iron core caused by bias. Based on the decoupling results, the hidden faults of the iron core are identified.

5. The latent fault diagnosis system for ultra-high voltage DC converter transformers according to claim 3, characterized in that, The heat dissipation unit uses the Pearson correlation coefficient analysis method to calculate the linear correlation between the preprocessed DC bias current data and oil temperature sampling data, and identifies transformer pipeline faults by analyzing the value and changes of the correlation coefficient.

6. The latent fault diagnosis system for ultra-high voltage DC converter transformers according to claim 3, characterized in that, The insulation unit uses wavelet packet decomposition to process the voltage waveform of the converter valve in a multi-scale manner, obtaining signal characteristics at different levels and frequency bands. It calculates the energy value of the corresponding frequency band at each decomposition level, calculates the energy difference between adjacent moments by monitoring the changes in frequency band energy, analyzes the energy difference, and outputs corresponding fault diagnosis information.

7. The latent fault diagnosis system for ultra-high voltage DC converter transformers according to claim 3, characterized in that, The mechanical unit, based on the collected winding vibration signal, constructs a delay coordinate matrix of the vibration signal according to the delay time and embedding dimension, converts the one-dimensional vibration signal into a high-dimensional spatial sequence, and then analyzes the constructed delay coordinate matrix to calculate the maximum Lyapunov exponent, thereby quantifying the stability of the winding mechanical structure.

8. The latent fault diagnosis system for ultra-high voltage DC converter transformers according to claim 1, characterized in that, The cross-domain collaborative terminal includes a differentiation unit, a first collaborative unit, a second collaborative unit, and an early warning unit; The differentiation unit is used to receive the fault judgment results from the deformation unit, heat dissipation unit, insulation unit and mechanical unit, and output the corresponding operation.

9. A latent fault diagnosis system for an ultra-high voltage DC converter transformer according to claim 8, characterized in that, The first collaborative unit constructs a feature correlation matrix based on the amplitude and frequency of the winding vibration signal and the temperature rise rate data of the transformer hot spot area. It also incorporates the covariance of vibration amplitude and temperature rise rate, the covariance of vibration frequency and temperature rise rate, the covariance of vibration amplitude and its own frequency, and the variance of temperature rise rate. Finally, it monitors the singular values ​​of the matrix, compares the singular values ​​with a set threshold, and outputs the judgment result.

10. A latent fault diagnosis system for an ultra-high voltage DC converter transformer according to claim 8, characterized in that, The second collaborative unit extracts the energy of the corresponding characteristic frequency based on the partial discharge ultra-high frequency signal and the environmental noise signal through time-frequency transformation, calculates the energy ratio of the two, performs singular value decomposition on the signal, separates the discharge fault related components, and calculates the energy ratio and the energy proportion of the fault component after decomposition. When the energy ratio and the main energy of the fault component both exceed the set threshold, it is determined that there is an insulation degradation fault. The early warning unit provides corresponding early warning information based on the results of the first and second collaborative units.

Images