Oil-filled sleeve fault positioning method and system based on palladium film hydrogen sensor array

Abstract

The invention discloses an oil-filled sleeve fault positioning method and system based on a palladium film hydrogen sensor array, and relates to the technical field of power equipment monitoring. The method comprises the following steps: deploying a palladium film hydrogen sensor array at a plurality of monitoring points communicated with an oil cavity of an oil-filled sleeve, and collecting hydrogen content signals of a multi-point oil sample; through time synchronization, temperature compensation and feature extraction, obtaining a multi-measuring-point hydrogen response arrival time sequence and curve morphological features; establishing an oil-filled sleeve digital twinborn simulation model containing a heat-electricity-diffusion coupling model, simulating hydrogen diffusion processes under different fault types and positions, and generating a virtual data set for algorithm training; training an artificial intelligence algorithm by using the data set, and constructing a fault position prediction model; during actual operation, real-time monitoring and processing data of the sensor are input, a fault positioning result is output, and early warning is realized. According to the method, the digital twinning algorithm and the AI algorithm are fused, so that the fault positioning accuracy is improved, and reliable guarantee is provided for safe operation of the oil-filled sleeve.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring and intelligent diagnosis technology, specifically to a method and system for locating faults in oil-filled bushings based on a palladium thin-film hydrogen sensor array. Background Technology

[0002] Oil-filled bushings (such as oil-impregnated paper capacitor bushings / oil-filled capacitor bushings) are key components in power systems for enabling high-voltage conductors to pass through walls and be insulated from external sources. Their operational reliability directly affects the safety and stability of substations and main equipment. During long-term operation, oil-filled bushings may fail due to various reasons, including aging and moisture absorption of the capacitor core insulation, partial discharge, overheating caused by poor contact at the ends or conductive connections, breakdown of the oil-paper insulation, arc discharge, and leakage due to seal deterioration. These failures often lead to the decomposition of the insulating material in the oil, producing various dissolved gases. Hydrogen is generally considered one of the important indicator gases for early faults related to discharge and overheating. Therefore, online monitoring of hydrogen in the oil for early warning of oil-filled bushing faults is of practical significance.

[0003] Existing technologies for hydrogen detection in oil mainly include chromatographic gas analysis, optical sensing, and metal oxide semiconductor gas sensors. Chromatographic methods offer high detection accuracy, but typically require sampling and offline analysis, making online real-time monitoring difficult. Optical methods involve complex and costly equipment, and suffer from insufficient stability and engineering adaptability in oil-based measurement environments. While metal oxide sensors respond well at high temperatures, they are prone to problems such as large drift, poor selectivity, and excessively high operating temperatures in insulating oil environments.

[0004] In recent years, palladium thin-film resistive hydrogen sensors have gradually become a research hotspot for dissolved gas detection in oil due to their high sensitivity to hydrogen, rapid response, and ability to operate at relatively low temperatures. However, existing research mostly focuses on single-point concentration monitoring or trend analysis, lacking a systematic study of the relationship between multi-point hydrogen diffusion characteristics inside the equipment and the spatial distribution of faults, making it difficult to effectively identify and locate fault locations.

[0005] Compared to high-oil-volume equipment such as transformers, oil-filled bushings are typical "low-oil equipment." They have small oil volumes, narrow and elongated oil chambers, and limited oil circuit connectivity. Internal air gaps, interfaces, and boundary effects lead to localized enrichment after fault release, transmission delays, and concentration curve morphology significantly influenced by structural boundaries. Furthermore, low-oil equipment is more sensitive to seal integrity, oil cleanliness, and pressure balance. Traditional high-flow-rate circulating sampling or invasive multi-point deployment schemes are prone to disturbances, making it difficult to balance "multi-point information acquisition" with "engineering feasibility." Therefore, in oil-filled bushing scenarios, diagnostic methods relying solely on single-point or single-parameter threshold methods often fail to distinguish between different fault locations (upper / middle / lower) and struggle to maintain stable and reliable early warning capabilities under complex operating conditions.

[0006] Therefore, how to address the low-oil-content structure of oil-filled casings, and achieve repeatable acquisition of hydrogen information from multiple points without significantly disrupting the seal or the working conditions of the oil cavity, and further combine digital twin modeling and artificial intelligence algorithms to establish an online monitoring system and method capable of early identification of internal faults in oil-filled casings and achieving spatial (axial / radial) fuzzy positioning, has become an urgent technical problem to be solved. Summary of the Invention

[0007] To address the common problems in existing oil gas monitoring for oil-filled casings, such as offline sampling not being real-time, difficulty in locating single-point monitoring, easy disturbance of the seal and oil cavity conditions when sampling at multiple points in low-oil structures, and poor adaptability of threshold / empirical methods to changes in operating conditions, this invention proposes a fault location method and system for oil-filled casings based on a palladium thin-film hydrogen sensor array. Under the premise of minimizing damage to the casing seal and not significantly changing the oil quantity / pressure balance of low-oil equipment, it achieves online early warning and spatial (axial / radial) fuzzy location of internal faults in oil-filled casings.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] Compared with the prior art, the present invention has at least the following beneficial effects:

[0010] (1) Online real-time and early warning: The high sensitivity and fast response of the palladium thin film resistive hydrogen sensor to hydrogen gas enables early online monitoring of discharge / overheating faults in the bushing.

[0011] (2) Low oil consumption and low disturbance: Through structures such as micro-sampling chamber, flow-limiting microchannel or micro-circulation bypass, multi-point information acquisition can be achieved without significantly changing the oil volume and pressure balance and without damaging the seal as much as possible, making the project highly feasible.

[0012] (3) Upgrade from “single-point trend” to “multi-point positioning”: By utilizing the time difference of response arrival and curve shape differences of multiple measurement points, combined with digital twin data-driven and artificial intelligence learning, the axial / radial fuzzy positioning and probability output of internal casing faults are realized, significantly improving the amount of diagnostic information.

[0013] (4) Higher adaptability and robustness: Digital twins take into account the electro-thermal-diffusion coupling mechanism, and AI models incorporate low oil characteristic correction, which can reduce the impact of operating condition changes, oil circuit differences and boundary effects on diagnostic accuracy. Attached Figure Description

[0014] Figure 1 The system structure diagram of the oil-filled casing fault location system based on a palladium thin-film hydrogen sensor array is shown.

[0015] Figure 2A schematic diagram of the sensor arrangement method for an oil-filled casing fault location system based on a palladium thin-film hydrogen sensor array is shown.

[0016] Numbering on the map:

[0017] 1. Oil-filled bushing; 2. Palladium thin-film resistive hydrogen sensor array module; 2-1, 2-2, 2-3, three sets of sensor modules; 3. Low-oil adaptive sampling / coupling module; 3-1. Micro-sampling chamber; 3-2. Flow-limiting microchannel; 3-3. Micro-circulation bypass oil circuit; 4. Signal acquisition and synchronization module; 5. Data processing and storage module; 6. Digital twin simulation module; 7. Artificial intelligence algorithm module; 8. Fault location and early warning module.

[0018] Various other advantages and benefits of the present invention will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiments below. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. Detailed Implementation

[0019] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. Although the drawings and embodiments show specific embodiments of the invention, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0020] It should be noted that certain terms are used in the specification and claims to refer to specific components. Those skilled in the art will understand that different terms may be used to refer to the same component. This specification and claims do not distinguish components based on differences in terminology, but rather on differences in function. The terms "comprising" or "including" used throughout the specification and claims are open-ended and should be interpreted as "comprising but not limited to." The following descriptions are preferred embodiments for carrying out the invention; however, these descriptions are for the purpose of understanding the general principles of the specification and are not intended to limit the scope of the invention. The scope of protection of this invention is determined by the appended claims.

[0021] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," "outer," "front," and "rear," etc., indicate the orientation or positional relationship based on the orientation or positional relationship in the working state of this invention, and are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention. Furthermore, the terms "first," "second," "third," and "fourth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0022] Example 1

[0023] like Figure 1 As shown, this embodiment provides a fault location system for an oil-filled bushing based on a palladium thin-film hydrogen sensor array. This system is applied to an oil-filled bushing 1 and includes a palladium thin-film resistive hydrogen sensor array module 2, illustrated as three sensor modules: 2-1, 2-2, and 2-3; a low-oil adaptation sampling / coupling module 3; a signal acquisition and synchronization module 4; a data processing and storage module 5; a digital twin simulation module 6; an artificial intelligence algorithm module 7; and a fault location and early warning module 8. The oil-filled bushing 1 is an oil-impregnated paper capacitor type or an oil-filled capacitor type bushing, internally comprising: a conductor, a capacitor core, oil-paper insulation, an upper oil chamber, a lower oil cavity, a flange oil cavity, and a shell. The oil-filled bushing is a low-oil device, with a relatively small internal oil volume, a narrow and elongated oil cavity, and limited oil circuit connectivity.

[0024] like Figure 2 As shown, a palladium thin-film resistive hydrogen sensor array module 2 is installed at multiple preset positions communicating with the internal oil cavity of the oil-filled sleeve 1, preferably including at least two or more of the following monitoring areas:

[0025] (1) Oil chamber area at the top of the casing;

[0026] (2) The flange oil cavity near the sleeve flange or the bypass micro-sampling cavity connected to it;

[0027] (3) The lower oil cavity area near the lower end of the conductor or the lower end terminal.

[0028] Preferably, in a 220kV bushing, 3 to 4 monitoring points are set along the axial direction to form a multi-point monitoring sequence arranged from bottom to top, so as to obtain the concentration change of fault gas at different locations and the response arrival time difference.

[0029] To avoid significant alterations to the bushing structure, the sensor connects to the oil cavity via a dedicated sealing connector or a modified detection port. The sensor body is fixed to the bushing housing by a metal bracket or insulating support, and a reliable seal is achieved using oil seals, O-rings, etc., to prevent leakage.

[0030] The low-oil-adaptive sampling / coupling module 3 includes: a micro-sampling chamber 3-1; a flow-limiting microchannel 3-2; and a micro-circulation bypass oil path 3-3 (optional), which are connected to the oil cavity inside the casing.

[0031] The microsampling chamber 3-1 is a small-volume metal or insulating shell, whose interior is in direct contact with the sensitive area of ​​the palladium thin-film resistive hydrogen sensor. The microsampling chamber is connected to the corresponding oil chamber of the sleeve via a flow-limiting microchannel 3-2. The diameter and length of the flow-limiting microchannel are designed to minimize oil exchange per unit time, thereby:

[0032] (1) It does not significantly change the overall oil volume and internal pressure balance of the casing;

[0033] (2) Reduce disturbances to the sealing status and oil medium cleanliness of low-oil equipment;

[0034] (3) Form a repeatable “transmission delay characteristic” to provide calibrable parameters for subsequent digital twin modeling and positioning algorithms.

[0035] When necessary, a micro-circulation bypass oil circuit 3-3 can be added outside the sleeve flange, driven by a micro pump to achieve small-flow circulation, so that the sensor points can obtain representative oil samples, while avoiding the traditional large-flow external circulation structure.

[0036] Signal acquisition and synchronization module 4 is electrically connected to each palladium thin-film resistive hydrogen sensor and is used for:

[0037] (1) Provide sensor bias voltage or constant current source;

[0038] (2) Collect the resistance change signals at each measuring point and convert them into digital quantities;

[0039] (3) Synchronize the data of each channel with time and set a unified timestamp;

[0040] (4) Perform preliminary filtering, noise reduction and temperature compensation.

[0041] Preferably, the signal acquisition and synchronization module 4 includes a multi-channel A / D converter, a temperature acquisition unit, and a local microprocessor, and uses a unified system clock or external timing signal (such as GPS / time network) to achieve multi-channel synchronous sampling.

[0042] The data processing and storage module 5 can be located in a field control box near the bushing, or it can be connected to a remote monitoring host via a communication network (such as RS485, Ethernet, fiber optic, etc.). This module is used for:

[0043] (1) Receive multi-channel hydrogen concentration or resistance change data;

[0044] (2) Calculate short-term and long-term trend quantities, such as the upward slope, rate of change, moving average, etc.

[0045] (3) Store raw data and feature data to form a historical database;

[0046] (4) Upload the data to the digital twin simulation module 6 and the artificial intelligence algorithm module 7.

[0047] Palladium thin-film resistive hydrogen sensors include:

[0048] Substrate: such as alumina ceramic substrate, glass substrate, or other oil-resistant and heat-resistant materials;

[0049] Sensitive thin film layer: Palladium or palladium alloy thin film deposited on the substrate, used as a hydrogen sensitive resistor layer;

[0050] Electrode and lead structure: disposed at both ends of the sensitive film to extract resistance signals;

[0051] Temperature measurement / compensation unit: such as Pt100, thermistor or built-in temperature sensing chip, used for temperature compensation;

[0052] Encapsulation and sealing structure: including a metal housing, insulating seals, and a protective layer in contact with the oil medium, ensuring that the sensitive film is reliably exposed to oil or hydrogen in oil, while guaranteeing overall airtightness and oil resistance.

[0053] The working principle of the palladium thin-film resistive sensor is as follows: hydrogen molecules diffuse into the palladium thin film and form hydrides with the metal, causing a reversible change in the film's resistance; the hydrogen concentration can be reflected by measuring the change in resistance. To adapt to the environment of oil-filled bushings, the sensor's operating temperature in this embodiment is preferably set within the range of room temperature to approximately 90°C to reduce the impact on the insulating oil and extend its service life.

[0054] Preferably, each sensor can be calibrated in an environment with a known hydrogen concentration before installation to obtain the resistance-concentration relationship; during system operation, the data processing module can dynamically correct the sensor output according to the temperature and calibration curve to improve measurement accuracy.

[0055] Example 2

[0056] A method for locating faults in oil-filled bushings based on a palladium thin-film hydrogen sensor array, comprising S1 to S6:

[0057] S1. Sensor Setup and Initialization

[0058] Based on the structural dimensions of the oil-filled casing and the typical fault distribution, several palladium thin-film resistive hydrogen sensors are arranged in the upper oil chamber, flange oil chamber (or its bypass micro-sampling chamber), and lower oil chamber of the oil-filled casing. Multi-point equivalent sampling is achieved by connecting these sensors to the oil chambers via a low-oil-adaptive sampling / coupling module. Before system operation, zero-point checks and temperature compensation parameter configurations are performed on each sensor.

[0059] S2. Online Data Acquisition and Feature Extraction

[0060] Signal acquisition and synchronization module 4 synchronously acquires the outputs of each sensor and performs filtering and temperature compensation on the data. Data processing and storage module 5 extracts the following features from the sampled data:

[0061] (1) The time of the first significant increase in hydrogen concentration or resistance at each monitoring point;

[0062] (2) The upward slope and maximum rate of change of each monitoring point within the preset time window;

[0063] (3) Peak or plateau values ​​and their occurrence time;

[0064] (4) Whether there are curve morphological characteristics such as a second rise or a slow tail;

[0065] (5) The sequence of response arrival time differences and relative change magnitudes between each monitoring point.

[0066] The above features form a multi-dimensional time-series feature vector, which is provided as input to the fault diagnosis model.

[0067] S3. Establish a digital twin model of the oil-filled casing.

[0068] In digital twin simulation module 6, a thermal-electrical-gas release-diffusion coupling model of the oil-filled bushing is established based on actual equipment structural parameters (conductor radius, capacitor core layered structure, oil cavity volume, bushing dimensions, oil-paper dielectric parameters, etc.), specifically including:

[0069] (1) Electric field distribution model: used to determine the local electric field enhancement region and possible local discharge location;

[0070] (2) Thermal field distribution model: Considering factors such as load current and ambient temperature, the temperature field is calculated;

[0071] (3) Fault release model: The hydrogen generation rate and duration are set for different fault types such as partial discharge, spark discharge, and local overheating.

[0072] (4) Hydrogen diffusion and transport model: Considering the effects of the narrow structure of the oil cavity, the top air gap and the boundary interface under low oil conditions, the volume and transport delay parameters of the micro-sampling cavity and the flow-limiting microchannel are also included in the model.

[0073] Numerical simulation was used to obtain time series data of hydrogen concentration at each monitoring point under different fault locations (upper end, middle end, lower end, near the inner or outer layer of the capacitor core, etc.), different fault types, and different operating conditions, forming a virtual sample set.

[0074] S4, Artificial Intelligence Model Training

[0075] Artificial intelligence algorithm module 7 uses supervised or semi-supervised learning methods to train on the aforementioned virtual sample set. Preferably, one or more of the following models can be used:

[0076] (1) Temporal feature extraction model based on convolutional neural network (CNN);

[0077] (2) Sequence modeling networks based on Long Short-Term Memory (LSTM) networks or Gated Recurrent Units (GRUs);

[0078] (3) Multi-channel time series modeling method based on temporal convolutional network (TCN).

[0079] The model input is the characteristic sequence of sensor signals from each monitoring point or the original time series, and the output is:

[0080] (1) Fault axial segmentation region (e.g., lower region, middle region, upper region);

[0081] (2) Or the radial level of the fault (e.g., near the conductor, in the middle of the core, near the outer shell);

[0082] (3) or spatial probability distribution in the internal coordinate system (r,z) of the casing.

[0083] To improve robustness in low-oil equipment scenarios, the following constraints or corrections are introduced during training:

[0084] (1) Use the microchannel transmission delay parameter as an adjustable feature or auxiliary input to distinguish between the true diffusion delay and the sampling delay;

[0085] (2) Feature annotation is performed on phenomena such as top air gap accumulation and boundary reflection causing the curve to rise twice, to enhance the model's ability to recognize changes in curve shape.

[0086] (3) By adding noise and disturbance parameters to simulate the differences under different oil temperatures, loads and sealing conditions, the generalization ability of the model can be improved.

[0087] With available on-site operational data, transfer learning or online incremental learning can be further employed to continuously optimize the model.

[0088] S5. Online Fault Detection and Location

[0089] After the system is put into operation, the data processing and storage module 5 continuously collects hydrogen data from multiple points and extracts features according to step S2, which are then fed into the trained artificial intelligence model for inference. The fault location and early warning module 8 determines the fault location based on the spatial probability distribution and early warning level output by the model:

[0090] (1) When the hydrogen concentration and the rate of rise are lower than the preset threshold, it is determined to be a normal state or a slight abnormality, and the trend is recorded.

[0091] (2) When the hydrogen concentration or characteristic value reaches the warning threshold, a warning message and possible fault area are given, such as "the probability of overheating / discharge in the lower area is high".

[0092] (3) When the hydrogen concentration or characteristic value exceeds the alarm threshold, a clear high-risk warning is output and the fault location result is pushed to the operation and maintenance system.

[0093] S6, Operation and Maintenance Decision-Making and Feedback Optimization

[0094] Maintenance personnel can formulate maintenance strategies based on the system's warning level and location results, combined with other monitoring methods (such as partial discharge measurement and infrared thermography). After bushing maintenance or disassembly, the actual fault location and type information can be transmitted back to the system for:

[0095] (1) Correct the parameters in the digital twin model;

[0096] (2) As new samples, they are added to the training set to retrain or fine-tune the artificial intelligence model;

[0097] (3) Continuously improve the diagnostic accuracy of the system under different models and operating environments.

[0098] Example 3

[0099] A method for locating faults in oil-filled bushings based on a palladium thin-film hydrogen sensor array includes:

[0100] A thin-film hydrogen sensor array is set at multiple monitoring points connected to the oil cavity of the oil-filled casing to obtain the hydrogen content of oil samples from multiple points.

[0101] Based on the structural dimensions of the oil-filled casing and the typical fault distribution, several palladium thin-film resistive hydrogen sensors are arranged in the upper oil chamber, flange oil chamber (or its bypass micro-sampling chamber), and lower oil chamber of the oil-filled casing. Multi-point equivalent sampling is achieved by connecting these sensors to the oil chambers via a low-oil-adaptive sampling / coupling module. Before system operation, zero-point checks and temperature compensation parameter configurations are performed on each sensor.

[0102] Time synchronization, temperature compensation, and feature extraction were performed on the sensor-acquired signals to obtain multi-point hydrogen response arrival time series data and curve morphology characteristics;

[0103] The following features are extracted from the sampled signal:

[0104] (1) The time of the first significant increase in hydrogen concentration or resistance at each monitoring point;

[0105] (2) The upward slope and maximum rate of change of each monitoring point within the preset time window;

[0106] (3) Peak or plateau values ​​and their occurrence time;

[0107] (4) Whether there are curve morphological characteristics such as a second rise or a slow tail;

[0108] (5) The sequence of response arrival time differences and relative change magnitudes between each monitoring point.

[0109] The above features form a multi-dimensional time-series feature vector, which is provided as input to the fault diagnosis model.

[0110] A digital twin simulation model of an oil-filled casing was established to simulate the hydrogen diffusion process under different fault types and locations of the oil-filled casing, so as to generate a virtual dataset for algorithm training.

[0111] The digital twin simulation model includes a thermal-electrical-diffusion coupling model, which includes potential control equations, heat conduction equations, and diffusion equations; it may also include a fault gas release model.

[0112] Based on the actual equipment structural parameters (conductor radius, capacitor core layered structure, oil cavity volume, bushing dimensions, oil-paper dielectric parameters, etc.),

[0113] (1) Electric potential control equation: used to determine the local electric field enhancement region and possible local discharge location;

[0114] (2) Heat conduction equation: Considering factors such as load current and ambient temperature, calculate the temperature field;

[0115] (3) Fault release model: The hydrogen generation rate and duration are set for different fault types such as partial discharge, spark discharge, and local overheating.

[0116] (4) Diffusion equation: Considering the effects of the narrow structure of the oil cavity, the top air gap and the boundary interface under low oil conditions, the volume and transmission delay parameters of the micro-sampling cavity and the flow-limiting microchannel are also included in the model.

[0117] Numerical simulation was used to obtain time series data of hydrogen concentration at each monitoring point under different fault locations (upper end, middle end, lower end, near the inner or outer layer of the capacitor core, etc.), different fault types, and different operating conditions in the oil-filled casing, forming a virtual dataset.

[0118] Artificial intelligence algorithms are trained to form a fault location prediction model using a virtual dataset generated by a digital twin simulation model;

[0119] The virtual dataset is trained using supervised or semi-supervised learning methods. Preferably, one or more of the following models can be used:

[0120] (1) Temporal feature extraction model based on convolutional neural network (CNN);

[0121] (2) Sequence modeling networks based on Long Short-Term Memory (LSTM) networks or Gated Recurrent Units (GRUs);

[0122] (3) Multi-channel time series modeling method based on temporal convolutional network (TCN).

[0123] The fault location prediction model takes as input the characteristic sequence of sensor signals from each monitoring point or the original time series, and outputs:

[0124] (1) Fault axial segmentation region (e.g., lower region, middle region, upper region);

[0125] (2) Or the radial level of the fault (e.g., near the conductor, in the middle of the core, near the outer shell);

[0126] (3) or spatial probability distribution in the internal coordinate system (r,z) of the casing.

[0127] In actual operation, the oil-filled casing sensor is input into the fault location prediction model to monitor the time series data in real time, and the location result of the oil-filled casing fault is output to achieve early warning.

[0128] Spatial probability distribution and early warning level output by the fault location prediction model:

[0129] (1) When the hydrogen concentration and the rate of rise are lower than the preset threshold, it is determined to be a normal state or a slight abnormality, and the trend is recorded.

[0130] (2) When the hydrogen concentration or characteristic value reaches the warning threshold, a warning message and possible fault area are given, such as "the probability of overheating / discharge in the lower area is high".

[0131] (3) When the hydrogen concentration or characteristic value exceeds the alarm threshold, a clear high-risk warning is output and the fault location result is pushed to the operation and maintenance system.

[0132] The aforementioned potential control equation:

[0133]

[0134] Based on the multi-medium structure and dielectric parameters of the oil-filled casing, the equation uses a quasi-static electric field control equation (potential field equation / Laplace-Poisson type equation) to solve for the electric potential field φ(x,t), and obtains the electric field intensity E(x,t) from the potential gradient relationship; where ε(x,T) is the dielectric constant (which can vary with temperature and space); the boundary conditions include the electrode boundary conditions with the high-voltage end applied potential V(t) and the grounding end potential being 0, and the electric potential and normal electric displacement are satisfied at the multi-medium interface; the electric field sub-model outputs the electric field distribution E and related electric stress indices, and calculates the electric loss heat source term of the heating model and calls the fault gas release model;

[0135] Heat conduction equation

[0136]

[0137] Q e (x,t)=σ(T,x)||E(x,t)|| 2

[0138] The transient heat conduction control equation (heat conduction equation / heat diffusion equation) is established based on the parameters of the multi-medium material of the casing to solve for the temperature field T(x,t), where ρ(x) is the density, c(x) is the specific heat capacity, and k(x) is the thermal conductivity. The right-hand side of the control equation includes a volume heat source term Q(x,t), which at least includes the electrical loss heat source Q output by the electric field sub-model. e (x,t) and the additional heat source Q caused by the fault f (x,t), and can make Q e The electric field strength E(x,t) and electrical conductivity σ(T,x) are used for calculation; the thermal boundary conditions include convective heat transfer boundaries, where h is the convective heat transfer coefficient and T is the electrical conductivity. ∞ The ambient temperature can be further included, and may include an adiabatic boundary or a given temperature boundary; the thermal field submodel outputs the temperature distribution T(x,t), which is used to couple with the temperature-dependent diffusion coefficient and fault release source term of the diffusion submodel.

[0139] Diffusion equation:

[0140]

[0141] Establish a hydrogen concentration field based on the law of mass conservation. The migration governing equation (diffusion equation / convection-diffusion equation) is given, wherein the diffusion flux term is determined by the diffusion coefficient D(T,x) and the concentration gradient, and the diffusion coefficient D can be a temperature-dependent function to achieve coupling with the thermal field; the governing equation includes a source term. Used to characterize hydrogen production / release processes under different fault types and locations, and may further include sink / decrease terms R. sink(x,t) is used to characterize the adsorption, dissolution, recombination, or other consumption effects of hydrogen in the material; boundary conditions include sealed boundaries (zero flux boundaries) and escaping boundaries (mass transfer boundaries), where the escaping boundary is determined by the mass transfer coefficient k. m With environmental hydrogen concentration Characterizing the mass exchange between the casing and the external environment; diffusion / migration sub-model output And the boundary release flux is used to generate virtual data and fault location identification.

[0142] The following constraints or corrections can be introduced during the training of artificial intelligence algorithms:

[0143] (1) The microchannel transmission delay parameter Δt i As an adjustable feature or auxiliary input, it can be used to distinguish between the true diffusion delay and the sampling delay;

[0144] (2) Feature annotation is performed on phenomena such as top air gap accumulation and boundary reflection causing the curve to rise twice, to enhance the model's ability to recognize changes in curve shape.

[0145] (3) By adding noise and disturbance parameters to simulate the differences under different oil temperatures, loads and sealing conditions, the generalization ability of the model can be improved.

[0146] With available on-site operational data, transfer learning or online incremental learning can be further employed to continuously optimize the model.

[0147] The following constraints or corrections can be introduced during the training of artificial intelligence algorithms:

[0148] Add microchannel transmission delay Δt i With low oil content and rapid enrichment characteristics τ e Correction;

[0149] The artificial intelligence algorithm model introduces a low-oil feature parameter vector Θ={τ} for low-oil equipment. e ,κ top ,r ref ,Δt i}, where τ e For rapid enrichment time constant, κ top r is the top air gap concentration factor. ref The boundary reflection causes a second-order increase coefficient, Δt i The microchannel transmission delay compensation amount for the i-th monitoring point; and the hydrogen response signal C of each monitoring point i (t) according to the formula

[0150]

[0151] Perform parameterization correction / feature mapping to obtain the feature vector f = Ψ({Ci (t)};Θ) is used as input to the fault location model to improve the robustness of the location.

[0152] Among them, the microchannel transmission delay Δt i To address the inconsistency in transmission delay caused by differences in the length, inner diameter, and flow resistance of the sampling oil path / microchannel at multiple monitoring points in the oil-filled casing, a microchannel transmission delay compensation amount Δt is introduced for the i-th monitoring point. i Furthermore, the hydrogen response signal at this monitoring point was time-axis aligned and corrected. Specifically, the original sampling sequence C... i (t) or the equivalent input sequence C after enrichment with less oil i,en (t) according to t→t-Δt i Time shift compensation is performed to obtain the delayed-corrected sequence C. i,en (t-Δt i This ensures that the arrival times of responses from each monitoring point are comparable, preventing the network from misinterpreting differences in channel transmission as differences in fault location; where Δt i It can be obtained through oil circuit calibration tests (injecting a step / pulse hydrogen source and measuring the arrival time), online correlation alignment (cross-correlation peak / dynamic time warping), or by joint estimation of digital twin and field data.

[0153] Low-oil rapid enrichment characteristics τ e To address the rapid enrichment effect of the low-oil-volume sampling module in small-volume oil samples, which causes the hydrogen response curve observed by the sensor to exhibit "accelerated rising edge, compressed time scale, or reshaped morphology," a rapid enrichment time constant τ is introduced. e The response kinetics are parameterized. Specifically, the observed signal is represented as a superposition of the actual hydrogen concentration input and the enrichment / mass transfer kinetics, using τ... e Constructing a first-order enrichment model (such as an exponential response kernel or a first-order inertial element) for C i,en (t;τ e Perform inversion or reparameterization mapping to ensure that the obtained correction sequence has consistent rising edge characteristics and comparable morphological characteristics under different low-oil operating conditions; τ e It can be obtained through mass transfer calibration experiments under different oil volume / temperature conditions, or it can be used as a low-oil feature parameter in deep networks as a conditional variable to adaptively correct the sensitivity of feature extraction to the rising edge speed and the plateau establishment process.

[0154] Top air gap aggregation coefficient κ top To address the issue of hydrogen accumulation at the top air gap / oil-gas interface of the oil-filled casing, leading to systematic amplitude bias at monitoring points at different axial positions, a top air gap accumulation coefficient κ is introduced. topThe amplitude and channel weights of the multi-point response are normalized and corrected. Specifically, this can be achieved by constructing a position-dependent weighting function w. i (κ top The response at the i-th monitoring point is scaled (e.g., by channel gain compensation or normalized weighting) to make the amplitude differences between monitoring points more reflect the actual fault release-diffusion differences rather than the top accumulation effect; κ top It can be obtained by simulation of digital twins under different oil level / air gap volume fraction conditions, or by fitting and estimating long-term statistical data from the field, and can be used as a conditional parameter to participate in the adaptive modulation of cross-sensor fusion weights.

[0155] Boundary reflection second rise coefficient r ref This refers to the "secondary rise / echo" phenomenon caused by oil cavity boundaries, structural interfaces, or reflux channels, i.e., the phenomenon occurring after the main response and delayed by τ. ref The morphological distortion of nearby secondary peaks or shoulders introduces the boundary reflection coefficient r. ref The secondary component is modeled and corrected. Specifically, the observed response of the i-th channel is represented as the superposition of the delayed-corrected principal component and its delayed replica: This allows for explicit characterization of the second ascent intensity and suppression of its interference with arrival time and localization features during feature extraction or deep network encoding; ref With τ ref It can be obtained through statistical analysis of the second peak amplitude ratio / peak interval under pulse excitation, or by inverse calculation from the boundary mass transfer and reflection condition parameters of the digital twin.

[0156] The preferred artificial intelligence model is a multi-sensor temporal deep network. It performs temporal encoding (1D-CNN / TCN / Transformer) on the synchronous response curves of each palladium thin-film hydrogen sensor, then performs cross-sensor fusion (attention or GNN) according to the casing axial topology, and incorporates the low-oil feature parameter Θ={τ e ,κ top ,r ref ,Δt i Given the condition, output the probability distribution of the fault location;

[0157] The low-oil characteristic parameter vector Θ serves as a parameterized representation of measurement link distortion. On one hand, it is used to perform time-shift compensation, enrichment kinetic correction, top accumulation amplitude normalization, and boundary reflection secondary response modeling for hydrogen responses at each monitoring point, thereby forming a correction sequence more consistent with the fault propagation mechanism. This system corrects time errors, enrichment biases, and abnormal features caused by the structure of the measurement system or low-oil equipment. On the other hand, Θ is injected as a conditional variable into a multi-sensor temporal deep network to achieve adaptive modulation of feature extraction and cross-sensor fusion weights under different low-oil operating conditions, thereby improving the robustness and generalization ability of the fault location probability output. During training, labeled virtual data generated by digital twins is used for supervised learning, and unlabeled field data can be introduced for semi-supervised consistency training / pseudo-label training to improve generalization.

[0158] The digital twin simulation module also includes a fault gas release model, which includes a type function, a location function, and a time function.

[0159] Type function

[0160]

[0161] Fault type s∈{thermal fault, partial discharge, electric arc, ...}; location x f The fault center is δ; the influence radius / initial diffusion scale is δ.

[0162] Position function

[0163]

[0164] ω(x;x f ,δ) is used to characterize the spatial distribution weight of the fault release source term, where x is the spatial coordinate, x f The fault center location is defined by δ, which is the influence scale parameter (characterizing the fault influence radius / expansion range). The location distribution function is used to expand the fault source terms from a centralized source to a distributed source, making the source terms have a larger weight near the fault center and decay with increasing spatial distance. Different distribution function families (Gaussian, exponential decay, piecewise constant, or ellipsoidal / cylindrical distribution) can be selected according to the fault type to adapt to different fault locations and geometries. The time gating function...

[0165] u(t;t0,τ)=H(t-t0)-H(t-t0-τ)

[0166] H(·) can be a Heaviside step function. The time-gating function u(t; t0, τ) is used to characterize the start-stop control of the fault release source term in the time dimension, where t is the simulation time, t0 is the fault start time, and τ is the fault duration. The time-gating function is used to limit the fault source term to take effect within a preset time interval, so that the source term is activated during the fault occurrence period and suppressed during the non-fault period. Different gating function families can be selected according to the fault characteristics, including step gating, pulse gating, smooth gating (with rising / falling transition), and periodic gating (intermittent / repeated triggering) to realize the simulation of different fault durations, repetition frequencies, and duty cycles.

[0167] It also includes type-dependent release intensity functions.

[0168] (1) Thermal failure type Arrhenius:

[0169]

[0170] (2) Power-law type of field strength for discharge / partial discharge:

[0171]

[0172] The gas release intensity function is used to characterize the equivalent hydrogen production / release intensity per unit time and per unit volume under different fault types, and serves as the type weight term for the fault gas release source term.

[0173] Among them, the thermal failure type release intensity function g th (T)(Arrhenius type) is used to characterize hydrogen production processes caused by hot spots, overheating, or thermal decomposition. Its intensity increases with increasing temperature T and can be limited by activation parameters (e.g., activation energy-related parameters) and threshold temperature parameters, so that the gas release contribution is suppressed when the temperature is below the threshold.

[0174] Among them, the discharge / partial discharge type gas release intensity function g pd (||E||,T) (Power Law Type) is used to characterize hydrogen production processes caused by partial discharge, arcing, or surface discharge. Its intensity increases with the increase of electric field strength ||E||. The field strength exponent parameter can be set to reflect the sensitivity of different discharge modes to electric stress. Temperature-related weights can be introduced to characterize the thermal effects or material reaction rate changes associated with the discharge. T and E are output by the thermal field sub-model and the electric field sub-model, respectively, to achieve coupling.

[0175] Example 2

[0176] A fault location system for oil-filled casing based on a palladium thin-film hydrogen sensor array, comprising:

[0177] The multi-point sensing data acquisition module is used to acquire the raw hydrogen content signal of each monitoring point from the palladium thin film hydrogen sensor array set at multiple monitoring points connected to the oil cavity of the oil-filled casing.

[0178] The signal preprocessing and feature extraction module is used to perform time synchronization correction and temperature compensation on the original signal, and extract the arrival time series data and curve morphology features of hydrogen response at multiple measurement points to form a structured monitoring feature vector.

[0179] The digital twin simulation modeling module is used to construct a digital twin simulation model of an oil-filled casing. The digital twin simulation model includes a thermal-electrical-diffusion coupling model and a fault gas release model.

[0180] The thermo-electric-diffusion coupling model includes an electric potential control equation, a heat conduction equation, and a hydrogen diffusion equation, which are used to simulate the coupled evolution of the electric field, temperature field, and hydrogen concentration field inside the casing.

[0181] The fault release model includes a type function, a location function, and a time-gated function, which are used to characterize the hydrogen release behavior under different fault types, spatial locations, and temporal characteristics, and to provide initial source terms for the diffusion equation.

[0182] The virtual training data generation module is used to traverse various fault types and fault location combinations in the digital twin simulation model, simulate and generate corresponding multi-measurement point hydrogen concentration time series response data, and combine the measurement link distortion parameterization model to perform system error correction on the simulation data to form a labeled virtual training dataset.

[0183] The fault location model training module is used to train an artificial intelligence algorithm based on the virtual training dataset to generate a fault location prediction model that can predict the spatial location of the fault based on the hydrogen response characteristics of multiple measurement points.

[0184] The online fault location and early warning module is used to receive the real-time monitoring feature vector processed by the signal preprocessing and feature extraction module during the actual operation of the oil-filled casing, input it into the fault location prediction model, output the fault location result of the oil-filled casing, and trigger an early warning based on the location result.

[0185] Digital twin simulation modeling module: Based on the structural parameters of the oil-filled casing, including structural parameter vector / mapping function, geometric modeling function, material parameter field function, thermal-electrical-diffusion coupling model (including potential control equation, electric field definition, electrical loss / Joule heat source term, heat conduction equation, diffusion equation, coupling relationship function) and fault gas release model (including fault type selection function, total gas release source term function, fault location distribution function, fault time gating function, type intensity function family, fault heat source function), a digital twin model of the casing is established to simulate the hydrogen generation and diffusion process under different fault types and locations, so as to generate a virtual dataset for algorithm training;

[0186] The palladium thin-film resistive hydrogen sensor includes a palladium or palladium alloy thin-film resistive sensing element and a temperature control or temperature compensation component, and is sealed in an oil-resistant environment to adapt to long-term immersion in tubing oil.

[0187] The monitoring points include at least two or more of the following: the upper oil chamber of the casing, the oil cavity in the flange area or the micro-sampling cavity connected to it, the lower oil cavity or the oil circuit location near the lower terminal area, forming a multi-point response sequence along the casing axis.

[0188] The online fault location and early warning module outputs the following results: axial segmentation area labels, radial hierarchical labels, and / or spatial probability distribution map of the casing.

[0189] Example 4

[0190] It should be noted that the technical features in the above embodiments can be arbitrarily combined to form new implementation methods. For example:

[0191] (1) The number of sensors can be adjusted from 2 to 6 depending on the voltage level and length of the bushing;

[0192] (2) The micro-sampling chamber can be set outside the sleeve flange, or it can be modified using the original oil test valve or monitoring interface.

[0193] (3) The data processing and artificial intelligence module can be centrally deployed in the substation monitoring host or deployed in the field control cabinet using an edge computing device;

[0194] (4) In addition to hydrogen, the present invention can also be extended to monitor other key gases (such as CO, C2H2, etc.), but the focus of the present invention is to use hydrogen characteristics for fault location.

[0195] Any equivalent substitutions and improvements made within the spirit and principles of this invention shall be included within the scope of protection of this invention.

Claims

1. A method for locating faults in oil-filled bushings based on a palladium thin-film hydrogen sensor array, characterized in that, include: A thin-film hydrogen sensor array is set at multiple monitoring points connected to the oil cavity of the oil-filled casing to obtain the hydrogen content of oil samples from multiple points. Time synchronization, temperature compensation, and feature extraction were performed on the sensor-acquired signals to obtain multi-point hydrogen response arrival time series data and curve morphology characteristics; A digital twin simulation model of an oil-filled casing was established to simulate the hydrogen diffusion process under different fault types and locations of the oil-filled casing, so as to generate a virtual dataset for algorithm training. The digital twin simulation model includes a thermal-electrical-diffusion coupling model, which includes the potential control equation, the heat conduction equation, and the diffusion equation. Artificial intelligence algorithms are trained to form a fault location prediction model using a virtual dataset generated by a digital twin simulation model; In actual operation, the oil-filled casing sensor is input into the fault location prediction model to monitor the time series data in real time, and the location result of the oil-filled casing fault is output to achieve early warning.

2. The method for fault location of oil-filled bushing based on palladium thin-film hydrogen sensor array according to claim 1, characterized in that, The method for simulating hydrogen diffusion processes under different fault types and locations in oil-filled casing to generate a virtual dataset for algorithm training is as follows: The electric potential control equation is used to calculate the electric field distribution and help identify high field strength regions. The heat conduction equation is used to calculate the temperature field and its effects on the diffusion coefficient and oil flow. Diffusion equations were used to simulate the spatiotemporal evolution of hydrogen concentration. The fault release model includes a type function, a position function, and a time-gated function; The initial values ​​of the diffusion model can be obtained by combining the fault release model. Subsequently, the diffusion model is used to perform time-domain calculations to obtain the time-series characteristics of hydrogen at each point after diffusion, so as to generate a virtual dataset for training artificial intelligence algorithms.

3. The method for fault location of oil-filled bushings based on a palladium thin-film hydrogen sensor array according to claim 1, characterized in that, The digital twin simulation model includes a thermal-electrical-diffusion coupling model, which includes the potential control equation, the heat conduction equation, and the diffusion equation. Electric potential control equation: Parameter definitions: electric potential φ(x,t), electric field strength E(x,t), ε(x,T) is the dielectric constant; electric potential V(t); Heat conduction equation Q e (x,t)=σ(T,x)||E(x,t)|| 2 Parameter definitions: Temperature field T(x,t), ρ(x) is density, c(x) is specific heat capacity, k(x) is thermal conductivity; Heat source term Q(x,t), electrical loss heat source Q e (x,t), the additional heat source Q caused by the fault. f (x,t), conductivity σ(T,x), h is the convective heat transfer coefficient, T ∞ The ambient temperature; Diffusion equation: Parameter definition: Hydrogen concentration field Diffusion coefficient D(T,x), source term Sum / Decrease Term R sink (x,t), the escape boundary passes through the mass transfer coefficient k m Environmental hydrogen concentration 4. The method for fault location of oil-filled bushings based on a palladium thin-film hydrogen sensor array according to claim 1, characterized in that, When training an artificial intelligence algorithm to form a fault location prediction model, a microchannel transmission delay Δt is added. i With low oil content and rapid enrichment characteristics τ e Correction: The artificial intelligence algorithm introduces a low-oil feature parameter vector Θ={τ} for low-oil equipment such as oil-filled casing. e ,κ top ,r ref ,Δt i }, where τ e For rapid enrichment time constant, κ top r is the top air gap concentration factor. ref The boundary reflection causes a second-order increase coefficient, Δt i The microchannel transmission delay compensation amount for the i-th monitoring point; and the hydrogen response signal C of each monitoring point i (t) according to the formula Perform parameterization correction / feature mapping to obtain the feature vector f = Ψ({C i (t)};Θ) is used as input to the fault location model to improve the robustness of the location.

5. The method for fault location of oil-filled bushing based on palladium thin-film hydrogen sensor array according to claim 4, characterized in that, The microchannel transmission delay Δt i : The original sampling sequence C i (t) or the equivalent input sequence C after enrichment with less oil i,en (t) according to t→t-Δt i Time shift compensation is performed to obtain the delayed-corrected sequence C. i,en (t-Δt i ), where Δt i The data is obtained through oil circuit calibration tests, online correlation alignment, or joint estimation using digital twins and field data. The low-oil rapid enrichment characteristic τ e The observed signal is represented as the superposition of the actual hydrogen concentration input and the equivalent response of enrichment / mass transfer kinetics, using τ e Constructing a first-order enrichment model for C i,en (t;τ e Perform inversion or reparameterization mapping; τ e The parameter obtained by fitting mass transfer calibration experiments under different oil volume / temperature conditions can also be used as a low-oil characteristic parameter in deep networks as a conditional variable for learning.

6. The method for fault location of oil-filled bushing based on palladium thin-film hydrogen sensor array according to claim 3, characterized in that, The top air gap aggregation coefficient κ top By constructing a position-related weight function w i (κ top ) Scale correction is applied to the response of the i-th monitoring point, κ top The data is obtained by simulation using digital twins under different oil level / air gap volume fraction conditions, or by fitting and estimating long-term statistical data from the field, and can be used as a conditional parameter to participate in the adaptive modulation of cross-sensor fusion weights. The boundary reflection second-order rise coefficient r ref Introducing the boundary reflection coefficient r ref The secondary component is modeled and corrected, and the observed response of the i-th channel is represented as the superposition of the delayed-corrected principal component and its delayed replica: This allows for explicit characterization of the second rise intensity and suppression of its interference with arrival time and localization features during feature extraction or deep network encoding. r ref With τ ref It can be obtained by statistical analysis of the second peak amplitude ratio / peak interval under pulse excitation, or by inverse calculation from the boundary mass transfer and reflection condition parameters of digital twin.

7. The method for fault location of oil-filled bushing based on palladium thin-film hydrogen sensor array according to claim 1, characterized in that, The preferred artificial intelligence algorithm is a multi-sensor temporal deep network, which performs temporal encoding (1D-CNN / TCN / Transformer) on the synchronous response curves of each palladium thin-film hydrogen sensor, then performs cross-sensor fusion according to the casing axial topology, and integrates the low-oil characteristic parameter Θ={τ e ,κ top ,r ref ,Δt i The probability distribution of the output fault location under the given conditions.

8. The method for fault location of oil-filled bushing based on palladium thin-film hydrogen sensor array according to claim 1, characterized in that, The digital twin simulation model also includes a fault gas release model, which includes a type function, a position function, and a time-gated function. Type functions: Fault type s∈{thermal fault, partial discharge, electric arc, ...}; location x f The fault center is represented by δ, which is the radius of influence / initial diffusion scale. Position function ω(x;x f ,δ) is used to characterize the spatial distribution weight of the fault release source term, where x is the spatial coordinate, x f The fault center location is δ, and the influence scale parameter is δ. Time Gating Function u(t;t0,τ)=H(t-t0)-H(t-t0-τ) Where H(·) can be the Heaviside step function, t is the simulation time, t0 is the fault initiation time, and τ is the fault duration.

9. The method for fault location of oil-filled bushings based on a palladium thin-film hydrogen sensor array according to claim 1, characterized in that, Including type-related release intensity functions (1) Thermal failure type Arrhenius: (2) Power-law type of field strength for discharge / partial discharge: Parameter definition: Thermal failure type release intensity function g th (T), discharge / partial discharge type gas release intensity function g pd (||E||,T), electric field strength ||E||.

10. A fault location system for oil-filled casing based on a palladium thin-film hydrogen sensor array, characterized in that, include: The multi-point sensing data acquisition module is used to acquire the raw hydrogen content signal of each monitoring point from the palladium thin film hydrogen sensor array set at multiple monitoring points connected to the oil cavity of the oil-filled casing. The signal preprocessing and feature extraction module is used to perform time synchronization correction and temperature compensation on the original signal, and extract the arrival time series data and curve morphology features of hydrogen response at multiple measurement points to form a structured monitoring feature vector. The digital twin simulation modeling module is used to construct a digital twin simulation model of an oil-filled casing. The digital twin simulation model includes a thermal-electrical-diffusion coupling model and a fault gas release model. The thermo-electric-diffusion coupling model includes an electric potential control equation, a heat conduction equation, and a hydrogen diffusion equation, which are used to simulate the coupled evolution of the electric field, temperature field, and hydrogen concentration field inside the casing. The fault release model includes a type function, a location function, and a time-gated function, which are used to characterize the hydrogen release behavior under different fault types, spatial locations, and temporal characteristics, and to provide initial source terms for the diffusion equation. The virtual training data generation module is used to traverse various fault types and fault location combinations in the digital twin simulation model, simulate and generate corresponding multi-measurement point hydrogen concentration time series response data, and combine the measurement link distortion parameterization model to perform system error correction on the simulation data to form a labeled virtual training dataset. The fault location model training module is used to train an artificial intelligence algorithm based on the virtual training dataset to generate a fault location prediction model that can predict the spatial location of the fault based on the hydrogen response characteristics of multiple measurement points. The online fault location and early warning module is used to receive the real-time monitoring feature vector processed by the signal preprocessing and feature extraction module during the actual operation of the oil-filled casing, input it into the fault location prediction model, output the fault location result of the oil-filled casing, and trigger an early warning based on the location result.

Images