Microstructure gradient electric field-based intelligent detection method and device for carbon particles in insulating oil

Abstract

The invention relates to an insulating oil carbon particle intelligent detection method and device based on a microstructure gradient electric field, and the method is characterized in that the method comprises the following steps: installing an electrode array in an oil tank of a transformer through a special support, guaranteeing the full contact of electrodes and insulating oil, and connecting a signal processing unit, a display alarm module and a remote monitoring module; a monitoring period, an alarm threshold value and an alarm level are set through the touch screen, network parameters of the remote monitoring module are configured, and data uploading is ensured; applying a high-frequency alternating-current electric field to the electrode system, collecting an electric field signal in real time, analyzing the electric field signal by using a deep learning algorithm, identifying the type of the carbon particles and calculating the concentration of the carbon particles; when the detected carbon particle concentration exceeds a set threshold value, the system triggers sound-light alarm and sends a short message notification.

Description

Technical Field

[0001] The present invention relates to a field, and more specifically, to a method and device for intelligently detecting carbon particles in insulating oil based on a microstructure gradient electric field. Background Art

[0002] With the rapid development of power systems, oil-immersed transformers, as crucial power equipment, are widely used for the conversion and transmission of electrical energy. They utilize a special insulating oil as the insulating medium to ensure safe operation. However, the presence of carbon particles in the insulating oil can negatively impact the transformer's insulation performance, thus affecting the reliability and safety of the equipment.

[0003] Traditional detection methods primarily rely on physical or chemical analysis, which often suffer from deficiencies such as insufficient sensitivity and complex operations, making them incapable of meeting the requirements for real-time monitoring of tiny carbon particles. For example, common detection methods in existing technologies involve analyzing insulating oil using optical or chemical methods. While these methods can detect carbon particles, they are relatively insensitive to small particles, and the detection process is cumbersome, making online, real-time monitoring impossible. Furthermore, traditional methods are affected by various factors during the detection process, such as environmental conditions like temperature and pressure, leading to unstable monitoring results and, in turn, compromising the assessment of the transformer's safety status.

[0004] Currently, there are some new monitoring devices on the market, such as detection systems based on image recognition or acoustic signals. However, these methods rely on complex algorithms and equipment, have not yet achieved widespread application, and still have shortcomings in environmental adaptability and stability. In addition, existing technologies lack comprehensive and systematic monitoring of carbon particles in insulating oil and fail to fully consider the response characteristics of carbon particles to electric fields, resulting in inadequate assessment of potential transformer risks.

[0005] Therefore, there is an urgent need for a new detection method that can efficiently and accurately monitor the presence and concentration of carbon particles in insulating oil based on the electric field perturbation method, thereby improving the safety and reliability of transformers, timely warning of potential risks, and ensuring the stable operation of power equipment. Summary of the Invention

[0006] In order to solve the deficiencies in the prior art, the present invention provides a method and device for intelligent detection of carbon particles in insulating oil based on a microstructure gradient electric field.

[0007] The present invention adopts the following technical solutions.

[0008] The first aspect of the present invention relates to an intelligent detection method for carbon particles in insulating oil based on a microstructure gradient electric field. The method includes the following steps: installing an electrode array in the oil tank of a transformer through a special bracket to ensure that the electrodes are in full contact with the insulating oil, connecting a signal processing unit, a display alarm module and a remote monitoring module; setting the monitoring period, alarm threshold and alarm level through the touch screen, configuring the network parameters of the remote monitoring module, and ensuring data upload; applying a high-frequency AC electric field to the electrode system and collecting electric field signals in real time, using a deep learning algorithm to analyze the electric field signals, identifying the type of carbon particles and calculating their concentration; when the detected carbon particle concentration exceeds the set threshold, the system will trigger an audible and visual alarm and send a text message notification.

[0009] The electrode array is installed in the transformer oil tank through a special bracket to ensure that the electrodes are in full contact with the insulating oil, and is connected to the signal processing unit, display alarm module and remote monitoring module. The following are the steps: some electrodes in the electrode array are needle-tip shaped, and some are flat; the tip diameter of the needle-tip electrode is 50μm, the size of the flat electrode is 5mm×5mm, and the electrode spacing is adjustable; the electrode surface is covered with a layer of multi-layer graphene nanomaterial, and the electrode array is fixed in the transformer oil tank through a high-strength insulating material bracket. The installation angle of the electrode can be adjusted from 0° to 30°.

[0010] A high-frequency alternating current electric field is applied to the electrode system and the electric field signal is collected in real time, including: the electric field is applied in a high-frequency alternating current electric field with a frequency range of 1kHz to 50kHz and a voltage range of 10V to 100V.

[0011] A high-frequency AC electric field is applied to the electrode system and the electric field signal is collected in real time. A deep learning algorithm is used to analyze the electric field signal to identify the type of carbon particles and calculate their concentration. This includes: performing time-frequency transformation on the signal using the Chirp-Z algorithm; and automatically identifying and classifying the carbon particle concentration, particle size, and morphology in the electric field signal using the CLDNet fusion neural network model.

[0012] A high-frequency AC electric field is applied to the electrode system and the electric field signal is collected in real time. A deep learning algorithm is used to analyze the electric field signal, identify the type of carbon particles and calculate their concentration. This includes automatically adjusting the parameters of the processing algorithm based on actual environmental changes such as temperature, humidity, and oil flow rate.

[0013] Through CNN feature extraction and classification, the concentration, particle size and morphology of carbon particles in the electric field signal are automatically identified and classified, including: the output carbon particle information includes concentration, particle size distribution and morphology; the particle size distribution is from micron to nanometer level; the morphology includes spherical, flaky or fibrous.

[0014] When the detected carbon particle concentration exceeds the set threshold, the system will trigger an audible and visual alarm and send a text message notification, including: When the detected carbon particle concentration exceeds the set threshold, the system will remind the user through an audible and visual alarm.

[0015] The second aspect of the present invention relates to an intelligent detection device for insulating oil carbon particles based on a microstructure gradient electric field, which is implemented using the intelligent detection method for insulating oil carbon particles based on a microstructure gradient electric field described in the first aspect of the present invention; wherein, the device includes an equipment installation module, a parameter setting module, a data analysis and alarm module, and a data storage and remote monitoring module; the equipment installation module is used to install the electrode array in the oil tank of the transformer through a special bracket to ensure that the electrode is in full contact with the insulating oil, and connect the signal processing unit, the display alarm module and the remote monitoring module; the parameter setting module is used to set the monitoring period, alarm threshold and alarm level through the touch screen, configure the network parameters of the remote monitoring module, and ensure data upload; the data analysis and alarm module is used to apply a high-frequency AC electric field to the electrode system and collect electric field signals in real time, use a deep learning algorithm to analyze the electric field signals, identify the type of carbon particles and calculate their concentration; the data storage and remote monitoring module is used to trigger an audible and visual alarm and send a text message notification when the detected carbon particle concentration exceeds the set threshold.

[0016] A third aspect of the present invention relates to a terminal, comprising a processor and a storage medium; the storage medium is used to store instructions; and the processor is used to operate according to the instructions to execute the steps of the method described in the first aspect of the present invention.

[0017] A fourth aspect of the present invention relates to a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect of the present invention.

[0018] The beneficial effect of the present invention is that, compared with the existing technology, the intelligent detection method and device of insulating oil carbon particles based on microstructure gradient electric field in the present invention achieves 0.01ppm ultra-high sensitivity detection and real-time identification of particle morphology through triple technological breakthroughs of field strength enhancement-frequency domain coding-intelligent analysis, which completely solves the defects of traditional detection technology in accuracy and early warning capability.

[0019] The beneficial effects of the present invention also include:

[0020] 1. The system sensitivity is increased to 0.01ppm (8 times higher than the traditional electrochemical method), the particle size detection limit is broken through to 0.1μm, and intelligent classification of carbon particle morphology is achieved (error rate ≤3%); the time-frequency coding excitation strategy makes the multi-particle size response spectrum separation ≥80%, effectively solving the problem of missed detection of small particles in traditional single-frequency detection methods.

[0021] 2. Gradient field intensity combined with online self-calibration technology ensures ±0.5% FS accuracy even under 50m / s oil flow. The system offers excellent environmental adaptability, operating within a full temperature range of -40°C to 120°C and maintaining an IP68 protection rating. Furthermore, blind source separation technology effectively suppresses common-mode interference, improving the signal-to-noise ratio by ≥20dB, thus ensuring detection stability even in complex operating conditions.

[0022] 3. The visual interface improves fault location efficiency by 80%, and the trend prediction model reduces unplanned downtime by more than 30%. The system supports edge-cloud collaborative computing, and the incremental learning mechanism enables continuous model evolution (monthly error correction), further improving the intelligent operation and maintenance level of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a system diagram of an intelligent detection device for carbon particles in insulating oil based on microstructure gradient electric field;

[0024] Figure 2 This is a working principle diagram of an intelligent detection device for carbon particles in insulating oil based on microstructure gradient electric field;

[0025] Figure 3 is the electrode structure diagram;

[0026] Figure 4 This is the flow chart of the holographic signal analysis algorithm;

[0027] Figure 5 This is the principle diagram of the adaptive signal processing algorithm;

[0028] Reference numerals:

[0029] 1-insulating bracket; 2-platinum metal electrode plate; 3-impurity particles; 4-wire; 5-power supply; 6-signal processing unit; 7-signal transmission medium; 8-user end; 21-interdigitated electrode; 22-nano coating; 221-nano zinc oxide coating; 222-graphene coating. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solutions and advantages of the present invention clearer and more accurate, the technical solutions of the present invention are described in detail below through multiple specific embodiments. The embodiments used in the present invention are only used to explain the present invention and are not intended to limit the content of the present invention.

[0031] In view of the shortcomings of existing technologies, such as the insufficient field strength of traditional electrodes (<5×10 4 V / m) leads to missed detection of tiny carbon particles, single-frequency excitation makes it difficult to capture the multi-dimensional characteristics of particles, and offline calibration causes measurement drift (>±3% FS). The present invention provides an intelligent detection method and device for insulating oil carbon particles based on a microstructure gradient electric field.

[0032] The first aspect of the present invention relates to an intelligent detection method for carbon particles in insulating oil based on a microstructure gradient electric field, the method comprising the following steps:

[0033] Step 1: Install the electrode array in the oil tank of the transformer through a special bracket, ensure that the electrodes are in full contact with the insulating oil, and connect the signal processing unit, display alarm module and remote monitoring module.

[0034] The array uses asymmetric interdigitated electrodes manufactured using MEMS technology. The electrode line width is 30-100 μm (fine-tunable by ±1 μm) and the spacing gradient range is 10-200 μm. Nano-zinc oxide / graphene heterogeneous coatings (coating thickness is 100-500 nm) are deposited on the electrode surface by magnetron sputtering, forming a local field strength of ≥1.8×10 5 The electrode array is integrated into the transformer oil circuit via an insulating ceramic bracket. It is resistant to oil flow shock and can withstand oil flow rates of up to 50 m / s. It also features a built-in temperature-pressure dual compensation sensor with an accuracy of ±0.1% FS, effectively eliminating environmental interference and ensuring stable system operation under complex operating conditions.

[0035] The carbon particle detection device of the present invention mainly includes an electrode system, a signal processing unit, a display alarm module, a data storage unit and a remote monitoring module. Its working principle is as follows: Figure 2 shown.

[0036] The electrode system of the present invention is manufactured using MEMS (micro-electromechanical system) technology, and uses an asymmetric interdigital structure to enhance the sensitivity and responsiveness of the electrode. Figure 3 The electrode line width is 50 μm, and its spacing gradient can be adjusted in the range of 20 μm to 200 μm to meet the detection requirements of carbon particles of different sizes.

[0037] The asymmetric interdigitated electrode design manufactured by MEMS technology can form a complex electric field distribution between the electrodes. The advantage of the asymmetric design is that it can provide a stronger electric field gradient than the symmetric electrode under different working conditions, which is very important for the capture and detection of carbon particles. The line width of the electrode is 50μm, and the electrode spacing is adjustable in the range of 20μm to 200μm. This design allows the electrode to adapt to carbon particles of different particle sizes, ensuring that the concentration of carbon particles can be accurately detected within different particle size ranges. By adjusting the spacing, the gradient of the electric field distribution also changes, further improving the detection accuracy.

[0038] The electrode surface is covered with a heterogeneous coating of nano-zinc oxide and graphene (coating thickness is 100-500nm). Nano-zinc oxide has a strong charge capture ability, which can effectively enhance the electrode surface's ability to capture carbon particles, while graphene has high conductivity and excellent mechanical strength, which can enhance the durability and stability of the electrode. This coating combines the high dielectric constant of nano-zinc oxide with the conductive properties of graphene, allowing the electrode surface to more accurately sense electric field changes during the monitoring process.

[0039] Electrode Integration: These micro-nano electrodes are integrated into the transformer's insulating oil container via a dedicated insulating bracket. The electrode surface is in full contact with the insulating oil, effectively detecting the presence and concentration of carbon particles in the oil when an electric field is applied.

[0040] During implementation, users set the monitoring cycle through an augmented reality (AR) interface, ranging from 1 minute to 24 hours, and can be dynamically adjusted based on the transformer's load factor and ambient temperature and humidity. The temperature and humidity range supports environmental variations from -40°C to 120°C. The system's built-in operating condition matching algorithm automatically recommends the monitoring frequency that best suits the current operating conditions by analyzing the oil flow rate (1 to 50 m / s) and historical degradation trends. For example, under high load or high oil flow rates, the system automatically increases the monitoring frequency to respond more promptly to changes in carbon particle concentration.

[0041] To improve detection sensitivity, the present invention uses time-frequency hybrid coding to apply a broadband electric field between 0.1 and 10 MHz. A 12th-order pseudo-random phase modulation signal (with phase jumps of ±180°) is generated based on the Chirp-Z algorithm and output to the micro-nanoelectrode array via a 250MS / s high-speed DAC. The dwell time of each frequency band ranges from 0.1 to 1 second and can be adaptively adjusted to ensure system stability and responsiveness under various operating conditions.

[0042] Step 2: Set the monitoring period, alarm threshold and alarm level through the touch screen, configure the network parameters of the remote monitoring module, and ensure data upload.

[0043] The dynamic anti-interference protection system features online self-calibration, automatically performing zero and span calibration daily. Kalman filtering technology is used to compensate for temperature drift, ensuring an error of no more than ±0.5% FS across the full temperature range of -40°C to 120°C. Furthermore, blind source separation technology is used to suppress electromagnetic noise and common-mode interference, improving the signal-to-noise ratio by ≥20dB, ensuring stability and high accuracy in complex electromagnetic environments.

[0044] The system transmits data via the TSN time-sensitive network, compatible with the IEC 61850 protocol and with a latency of ≤1ms. A storage unit constructs a spatiotemporal matrix of carbon particle growth and uses an LSTM prediction model to analyze 72-hour degradation trends with an accuracy of ≥92%. Daily automatic dual-frequency calibration (1kHz / 1MHz) and Kalman filter calibration ensure a drift of ≤±0.5% FS over a five-year period.

[0045] Step 3: Apply a high-frequency AC electric field to the electrode system and collect electric field signals in real time. Use a deep learning algorithm to analyze the electric field signals, identify the type of carbon particles, and calculate their concentration.

[0046] The time-frequency hybrid coding excitation module generates a 0.1-10MHz broadband swept frequency signal based on the Chirp-Z modulation algorithm, and implants a pseudo-random phase jump (phase change ±180°, random distribution), which is output to the electrode array through a high-speed DAC (sampling rate 250MS / s). Each excitation cycle contains 12 sets of non-uniformly spaced frequency hopping signals, which can synchronously stimulate the polarization migration and relaxation effects of carbon particles, thereby increasing the separation of the response signals of carbon particles with multiple particle sizes (0.1-10μm) to 80%. This design effectively improves the detection capability of particles of various particle sizes, and has significant advantages in the detection of tiny particles (nanoscale).

[0047] The specific implementation method includes: generating a 12-order pseudo-random phase modulation signal (phase jump ±180°) based on the Chirp-Z algorithm, and outputting it to the micro-nano electrode through a 250MS / s high-speed DAC; the dwell time of each frequency band is 0.1 to 1 second (adaptively adjustable), synchronously stimulating the polarization migration (1kHz to 1MHz) and relaxation effect (1-10MHz) of carbon particles, achieving full particle size coverage of 0.1-10μm; the gradient field strength on the electrode surface is ≥1.8×10 5 V / m, with real-time correction of environmental disturbances through the temperature-pressure dual compensation module (accuracy ±0.1% FS).

[0048] The multimode resonant signal analysis unit uses a CLDNet fusion neural network model for holographic signal analysis. The signal processing process includes feature extraction, using a seven-level adaptive convolution kernel to decompose the collected ΔC / ΔQ (capacitance / charge distortion) signal into the fundamental frequency, third harmonics, and relaxation remainders. Intelligent decision-making uses a bidirectional LSTM network to analyze time-frequency ridge features and output the following three-dimensional parameters: concentration range of 0.01-500ppm, dynamic error ≤±0.5%; particle size distribution of 0.1-10μm, binning accuracy of 0.05μm; and morphology classification into spherical, lamellar, and fibrous shapes, with classification accuracy (F1 value) ≥97%.

[0049] A 7-level adaptive wavelet packet decomposition (Daubechies9 wavelet basis) is used to remove power-frequency noise, reconstructing the fundamental frequency, 3rd / 5th harmonics, and relaxation residuals. Time series modeling is then performed: a bidirectional LSTM network is used to extract time-frequency ridge features and construct a 12-dimensional dynamic tensor. Finally, dynamic decision-making is performed: outputs include carbon particle concentration (0.01-500ppm, ±0.5% accuracy), particle size distribution (0.05μm resolution), and morphological classification (F1 value ≥ 97%). Furthermore, if the output confidence level is less than 95%, the system triggers an incremental learning mechanism to update the model parameters through resampling.

[0050] In the signal excitation unit, the system uses an FPGA as the control core and has a built-in Chirp-Z algorithm processing module. By online generating and updating a 12th-order pseudo-random phase sequence (phase jump ±180°), it achieves broadband (0.1MHz to 10MHz) frequency domain sweeping. The FPGA sends the digital baseband signal to a 250MS / s, 14-bit resolution DAC, which is then output to the micro-nano electrode array after cascading amplifiers. The amplification chain includes a two-stage RF power amplifier with a bandwidth flatness of ±1dB and an output range of up to ±50V. It is also equipped with an adjustable capacitor / inductor impedance matching network to ensure optimal matching with the 50Ω electrode impedance. The dwell time of each frequency band is initially set to 0.5s and can be dynamically adjusted within the range of 0.1s to 1s through an "adaptive dwell algorithm" based on the oil temperature and preset sensitivity indicators to achieve a balance between high SNR and fast response.

[0051] In the feature noise reduction stage, the signal processing unit uses the 7-level adaptive wavelet packet decomposition technology to denoise the collected electric field signal. Specifically, the signal x(t) is decomposed into a series of frequency band signals x i (t), each frequency band signal corresponds to a specific scale. The wavelet packet transform is performed using the Daubechies9 wavelet basis, using the formula:

[0052]

[0053] Among them, h i represents the filter for the i-th wavelet packet, * denotes the convolution operation, and N is the number of decomposition layers. This method separates the low-frequency signal from the high-frequency noise, thereby eliminating power frequency noise and environmental interference. The reconstructed signal retains the main components related to changes in carbon particle concentration, while effectively removing the noise.

[0054] Next, we enter the time series modeling phase. The signal processing unit uses a bidirectional LSTM (long short-term memory) network to analyze the time-frequency ridge features of the signal. To capture the long-term dependencies in the signal, the LSTM network uses the following state update formula:

[0055] f t=σ(W f ·[h t-1 ,x t ]+b f )

[0056] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0057] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0058] c t =f t *c t-1 +i t *tanh(W c ·[h t-1 ,x t ]+b c )

[0059] h t =o t *tanh(c t )

[0060] Among them, x t is the input signal of the current time step, h t-1 and c t-1 are the output and memory state of the previous moment, W f ,W i ,W o ,W c is the weight matrix in the network, b f ,b i ,b o ,b c is the bias term, σ is the sigmoid activation function, and tanh is the hyperbolic tangent activation function. Through this process, the LSTM network can extract long-term temporal dependencies from the signal, further extract the carbon particle response characteristics in the signal, and generate a temporal dynamic feature representation of the electric field signal.

[0061] In another embodiment, in the signal acquisition and preprocessing unit, the micro-nanoelectrode feedback signal is sampled at 250MS / s using a 12-bit ADC. The FPGA then performs framing and time-stamping of the 1024-sample signal, embedding a time-frequency hybrid coding label. Primary noise reduction is performed within the FPGA using Daubechies-8 wavelet decomposition to perform an 8-level process, removing noise components exceeding the 0.1MHz to 10MHz bandwidth. The denoised data is transferred to the ARM Cortex-A53 master CPU via DMA. On the CPU, the system first uses an ARIMA (p, d, q) model (p = 2, d = 1, q = 2) to perform prediction residual analysis on each frame of the signal, automatically determining the start t0 and end t1 of each frequency band. Mutual information (MI) time series curves are then calculated to remove abnormal glitches exceeding the threshold. Finally, maximum-minimum normalization is performed on each signal segment, outputting a zero-mean, unit-variance data block.

[0062] The time-frequency feature extraction unit is deployed on the GPU accelerator board and performs a fusion operation of short-time Fourier transform (STFT) and Chirp-Z transform in parallel for each signal block. The specific steps include: selecting a switchable window function (Hamming window or Blackman window) with a length of N = 512 for the signal block and calculating:

[0063]

[0064] where φ n The linear frequency modulation phase quantity generated online by the Chirp-Z controller. The system further performs cubic spline interpolation and smoothing polynomial fitting on the resulting amplitude Af(t) and phase φf(t) curves to extract 16 time-frequency features, such as amplitude peaks, slope change points, and second-order derivative extremes. The feature vectors of each frequency band are concatenated to form an overall input vector X of dimension M ≈ 256.

[0065] The deep fusion signal processing unit, based on the innovative CLDNet network structure, consists of a feature denoising layer, a time series modeling layer, and a dynamic decision layer. The feature denoising layer utilizes a two-layer multi-head self-attention mechanism (MHSA, Head = 8), each followed by a residual connection and LayerNorm. The time series modeling layer consists of three layers of bidirectional LSTM (HiddenSize = 256) to capture the spatiotemporal correlations of features across different frequency bands. The dynamic decision layer employs an XGBoost (number of trees = 100, Depth = 6) regression branch and a Bayesian optimized classifier in parallel, simultaneously outputting regression values and discrete categories for carbon particle concentration C, particle size distribution D, and morphological index M. The entire network is trained offline on a manually prepared calibration sample set (2,000 samples, concentration range 0-100 ppm, particle size 50-500 nm), with minor iterative fine-tuning after each batch measurement to adapt to online environmental changes. The overall input vector X is fed into the trained deep fusion signal processing unit.

[0066] The result output and feedback control unit visualizes the three data sets (C, D, and M) in real time via a touchscreen UI, displaying concentration-time curves, particle size histograms, and morphology distribution pie charts. When C exceeds the user-set threshold (Cthr) or the network detects an abnormal morphology probability >0.9, the system automatically triggers an audible and visual alarm, simultaneously recording an event log and activating the closed-loop injection module for a reference oil sample, completing an automatic calibration test. All measurement data is stored in CSV format on a local SSD and uploaded to the cloud platform via encrypted 4G / 5G communication modules, enabling remote monitoring and historical backtracking.

[0067] In order to improve the accuracy of time-frequency signals, the system will further perform deep fusion processing on the time series features to construct a 12-dimensional dynamic tensor T 12 To represent the complex response of the signal in various frequency bands and time scales:

[0068]

[0069] Among them, F i is the time-frequency characteristic of the i-th frequency component, g i (x t ) is the corresponding time series response function, α i is a weighting coefficient that reflects the contribution of different frequency components to the overall characteristics. Through this combination of multi-dimensional features, the system can more comprehensively understand the time-frequency ridges of the signal and extract accurate carbon particle information from it. In the final dynamic decision-making stage, the system combines the features after time series modeling and noise reduction processing, and outputs the carbon particle concentration, particle size distribution and morphology classification through an optimized neural network. Carbon particle concentration C n The calculation formula is:

[0070]

[0071] Among them, |F i ∣ 2 Represents the energy of the i-th frequency band, k1 is a constant coefficient, α i and β i are the weighting coefficient and response index of each frequency band, respectively, reflecting the concentration contribution of carbon particles at different frequencies.

[0072] The calculation formula for particle size distribution is:

[0073]

[0074] Among them, γi and δi are the weight coefficients of particle size distribution, g i (x t) is the time-frequency response function, which represents the particle size response corresponding to each frequency band. The particle size distribution accurately distinguishes carbon particles from microscopic to nanoscale by analyzing multi-frequency signals.

[0075] The output of morphological classification is optimized using the following formula:

[0076]

[0077] Among them, P i represents the probability of the i-th form, d i is the particle size under this morphology, d avg is the average particle size of all particles, λ i and η i is the weighting coefficient of morphological classification, and this formula can be used to accurately classify spherical, lamellar and fibrous morphologies.

[0078] In particular, in order to eliminate the influence of temperature and pressure fluctuations on the accuracy of electric field measurement, the present invention has a built-in temperature T(t)-pressure P(t) dual compensation module in the main circuit. First, the system uses a constant temperature and constant pressure test bench at the equipment factory stage to test the E(t) at temperature T (-40 ~ 85 ° C) and pressure P (0.8 ~ 1.2 atm). m (t) Perform multi-point calibration and use the least squares method to fit the baseline compensation coefficients α0 and β0;

[0079] During operation, 4-way Pt1000 platinum resistance temperature sensors and 2-way MEMS piezoresistive pressure sensors sample at 1kS / s, and after Sigma-Delta A / D conversion, they are cached in FPGA channels and sent to the calibration unit in real time. This unit combines the one-dimensional Kalman filter to measure the E m (t) is recursively corrected, and the least squares method is triggered every 10 seconds to update α(t) and β(t). corr (t) is calculated as follows:

[0080] E corr (t) = E m (t)-α(t)[T(t)-T0]-β(t)[P(t)-P0]+K KF (E m (t),t)

[0081] Among them, T0 and P0 are reference temperature and pressure states, K KF The Kalman gain function is used, and its state and measurement equations are set according to the standard one-dimensional model. Through this module, the system can control the electric field drift caused by temperature and pressure errors to within ±0.5%.

[0082] Step 4: When the detected carbon particle concentration exceeds the set threshold, the system will trigger an audible and visual alarm and send a text message notification.

[0083] The intelligent early warning interactive platform features an augmented reality interface that dynamically displays a carbon particle deposition heat map (refresh rate 60fps). Using a transfer learning model, it predicts the risk of insulating oil degradation within 72 hours and proactively triggers a three-level alarm system, including audible and visual alarms, SMS notifications, and cloud-based alarms (false alarm rate ≤ 0.1%). Furthermore, the system transmits data back to the main control system via the TSN protocol, supporting seamless integration with the IEC 61850 standard protocol, ensuring compatibility with existing power equipment management systems and efficient data transmission.

[0084] Users can set the monitoring cycle (adjustable from 1 minute to 24 hours) through an augmented reality (AR) interface, and dynamically optimize the detection frequency based on the transformer's load factor and ambient temperature and humidity (supporting a range of -40°C to 120°C). The system's built-in operating condition matching algorithm automatically recommends the optimal monitoring plan based on oil flow rate (1 to 50 m / s) and historical degradation trends.

[0085] The AR interface dynamically displays a heat map of carbon particle deposition (60fps refresh rate), with areas exceeding the standard marked with 3D pulsed light spots, and supports gesture-controlled zooming and analysis. The early warning system includes a three-level alarm linkage strategy: Primary Alert (concentration exceeds the threshold by 50%): triggers a 105dB audible and visual alarm; Intermediate Alert (exceeding the standard for three consecutive cycles): automatically pushes a text message and sends a work order to the equipment; Advanced Alert (CPHI index > 0.8): initiates a cloud-based expert consultation and links to oil regeneration equipment (via OPC UA protocol). The alarm system has a built-in environmental adaptation module that dynamically adjusts the alarm volume based on background noise (30-90dB).

[0086] The signal processing unit outputs carbon particle concentration, particle size distribution, and morphology classification with an accuracy of up to ±0.5%. If the confidence level of the system output falls below 95%, an incremental learning mechanism is triggered, using new signal samples to resample and update the parameters of the neural network model to achieve self-optimization of the model.

[0087] Through the above-mentioned complex signal analysis process, the present invention can efficiently and accurately extract relevant information about carbon particles, while improving the adaptability and robustness of the system in complex environments, providing strong data support for carbon particle detection and equipment early warning.

[0088] The AR interface dynamically displays a heat map of carbon particle deposition at a refresh rate of 60fps. Exceeding standards are marked with 3D pulsed light spots, and gestures are supported for zooming in and out, facilitating in-depth analysis and location tracking.

[0089] The early warning system includes a three-level alarm linkage strategy:

[0090] Primary warning (concentration exceeds the threshold by 50%): triggers a 105dB sound and light alarm to inform the user of the possible risk of carbon particle accumulation.

[0091] Intermediate warning (exceeding the standard for three consecutive cycles): Automatically push text messages and equipment work orders to remind maintenance personnel to pay attention to equipment status.

[0092] Advanced early warning (CPHI index > 0.8): Initiates cloud-based expert consultation, provides remote diagnosis and decision support, and can be linked to oil regeneration equipment for processing (linked via the OPC UA protocol).

[0093] In order to adapt to the interference of different environmental noises, the system has a built-in environment adaptive module that can dynamically adjust the alarm volume according to the background noise (such as changes within the range of 30-90dB). The algorithm principle is as follows Figure 5 This module performs noise adaptive adjustment using the following formula:

[0094]

[0095] Among them: A adjusted is the adjusted alarm volume; Abase is the basic alarm volume (the volume in a standard noise environment); N is the current ambient noise level; N0 is the noise baseline value; ΔN is the noise tolerance range; α is the adjustment coefficient for sensitivity adjustment.

[0096] In addition to detecting carbon particle concentration, the system can also trigger abnormal pattern detection based on changes in parameters such as oil flow rate, temperature, and pressure. When the system detects changes in these parameters outside the normal range, there may be a risk of equipment degradation, and the early warning mechanism will be activated. This detection mechanism is calculated based on the following formula:

[0097]

[0098] Where: X is the current monitored parameter value (such as oil flow rate, temperature, etc.); X0 is the historical standard value; and σ is the standard deviation of the parameter. When the deviation value exceeds the set threshold, the system triggers an early warning mechanism to further analyze the potential risk of carbon particle accumulation.

[0099] The alarm system has a built-in environmental adaptive module that can dynamically adjust the alarm volume according to the background noise (30-90dB), ensuring that the alarm can be clearly received in different noise environments.

[0100] The system transmits data via the TSN time-sensitive network, compatible with the IEC 61850 protocol, with a data transmission latency of ≤1ms. The storage unit constructs a spatiotemporal matrix of carbon particle growth and combines it with an LSTM prediction model to analyze degradation trends over a 72-hour period. The model achieves an accuracy rate exceeding 92%.

[0101] The system automatically performs dual-frequency calibration (1kHz / 1MHz) and Kalman filter calibration daily to ensure drift ≤±0.5% FS over a five-year period. The Kalman filter formula is as follows:

[0102]

[0103] Among them, P(t+1) is the prediction error covariance matrix of the next moment, is the system state transfer matrix, P(t) is the error covariance matrix at the current moment, and Q is the process noise covariance matrix.

[0104] To ensure the accuracy of the system in long-term operation, especially in response to environmental changes and new failure modes, this paper introduces an incremental learning mechanism. When the system's prediction accuracy decreases, a self-optimization process is triggered to retrain the model using new monitoring data. The core idea of incremental learning is to update the model using the following formula:

[0105]

[0106] θ t+1 =θ t +Δθ t

[0107] Where: θ t is the model parameter at the current moment; η is the learning rate; L(θ t ,D t ) is the loss function; is the gradient of the loss function.

[0108] By applying these multimodal early warning, data fusion, and predictive maintenance technologies, this invention effectively improves substation equipment health management capabilities. Through precise fault warnings and predictive maintenance, the system minimizes equipment failure risks, optimizes maintenance processes, and improves substation operational safety and reliability.

[0109] The second aspect of the present invention relates to an intelligent detection device for insulating oil carbon particles based on a microstructure gradient electric field; it is implemented using an intelligent detection method for insulating oil carbon particles based on a microstructure gradient electric field according to the first aspect of the present invention; the device includes an equipment installation module, a parameter setting module, a data analysis and alarm module, and a data storage and remote monitoring module; the equipment installation module is used to install the electrode array in the oil tank of the transformer through a special bracket to ensure that the electrode is in full contact with the insulating oil, and to connect the signal processing unit, the display alarm module and the remote monitoring module; the parameter setting module is used to set the monitoring period, alarm threshold and alarm level through the touch screen, configure the network parameters of the remote monitoring module, and ensure data upload; the data analysis and alarm module is used to apply a high-frequency AC electric field to the electrode system and collect electric field signals in real time, use a deep learning algorithm to analyze the electric field signals, identify the type of carbon particles and calculate their concentration; the data storage and remote monitoring module is used to trigger an audible and visual alarm and send a text message notification when the detected carbon particle concentration exceeds the set threshold.

[0110] The carbon particle detection device includes a micro-nano composite electrode array, a time-frequency hybrid coding excitation module, a multi-mode resonant signal analysis unit, a dynamic anti-interference protection system and an intelligent early warning interaction platform. The carbon particle detection device mainly includes an electrode system, a signal processing unit, a display alarm module, a data storage unit and a remote monitoring module, aiming to further improve the safety and reliability of substation transformers.

[0111] The electrode system utilizes an improved micro-nano electrode array made of a highly conductive platinum alloy to enhance the uniformity and stability of the electric field. To enhance sensitivity and detection accuracy, the electrode array adopts a needle-point-plane structure, with some electrodes in the array being needle-point and others being plane. This design enhances the electrodes' ability to capture carbon particles.

[0112] Electrode Size and Spacing: The tip diameter of the needle tip electrode is 50 μm, and the flat electrode is 5 mm x 5 mm. The electrode spacing is adjustable from 50 μm to 300 μm. By adjusting the distance between the needle tip and the flat electrode, the detection effect of carbon particles of different sizes can be optimized.

[0113] Electrode Materials and Surface Coating: The electrode surface is coated with a multilayer graphene nanomaterial. This coating effectively improves the electrode's charge capture capability and ensures its stability after long-term use. Graphene's excellent electrical conductivity makes the electrode more sensitive to changes in the electric field.

[0114] Electrode Installation: The electrode array is fixed to the transformer tank using a high-strength insulating material bracket to ensure full contact between the electrodes and the insulating oil. The electrode installation angle can be adjusted from 0° to 30° to accommodate different types of transformer containers.

[0115] The electric field is applied using a high-frequency AC electric field with a frequency range of 1kHz to 50kHz and a voltage range of 10V to 100V. By adjusting the frequency and voltage, the intensity and uniformity of the electric field are effectively optimized, ensuring sensitive detection of various carbon particles.

[0116] The signal processing unit uses a multi-level processing algorithm based on deep learning and time-frequency analysis to accurately analyze the collected electric field signals. The optimization points of this algorithm include:

[0117] Multi-band time-frequency analysis: By using the Chirp-Z algorithm to perform time-frequency transformation on the signal, it is possible to accurately capture tiny fluctuations in the electric field signal and identify the impact of different types of carbon particles.

[0118] Deep Convolutional Neural Network (CNN): CNN performs feature extraction and classification, enabling automatic identification and classification of carbon particle concentration, particle size, and morphology in electric field signals. The CNN processes signal features through multiple convolutional layers, performs dimensionality reduction through pooling layers, and ultimately outputs relevant information about carbon particles through fully connected layers.

[0119] Dynamic Algorithm Adjustment: This signal processing unit automatically adjusts the parameters of the processing algorithm based on actual environmental changes, such as temperature, humidity, and oil flow rate. For example, if the oil flow rate is high, the system will automatically increase the monitoring frequency to more promptly respond to changes in carbon particle concentration.

[0120] The algorithm calculates within 0.05 seconds, meeting the needs of real-time monitoring. The final output of carbon particle information includes concentration, particle size distribution (micrometer to nanometer scale), and morphology (spherical, flaky, or fibrous).

[0121] The display alarm module uses a 7-inch high-definition touch screen design, providing a convenient user interface. Users can use the touch screen to set the monitoring period (1 minute to 24 hours), adjust the alarm threshold, and view real-time monitoring data and historical trends.

[0122] Display interface: The user interface is simple and intuitive, including real-time monitoring data, historical data curves, alarm settings, and system configuration. Users can set up and operate directly through the touch screen, enhancing the operability of the device.

[0123] Alarm function: When the detected carbon particle concentration exceeds the set threshold, the system will alert the user through audible and visual alarms. The alarm volume is adjustable (60dB to 90dB), and the alarm light color is customizable (red, yellow, green). In addition, the system supports SMS notifications. When the carbon particle concentration is abnormal, an SMS message will be automatically sent to the user's preset mobile phone number, including the carbon particle concentration, detection time, and device location.

[0124] The remote monitoring module has expanded its functionality to support remote data access and control via a cloud platform. Monitoring data is uploaded to a cloud server via a 4G / 5G network, allowing users to view device operating status, historical data, and analysis reports in real time via a mobile app or website.

[0125] Cloud Storage and Data Analysis: The cloud platform has a storage capacity of 1TB, supporting long-term storage of monitoring data. The cloud platform integrates data analysis tools to perform trend prediction and fault diagnosis, and provides equipment maintenance recommendations based on trends in carbon particle concentration.

[0126] Remote control function: Users can remotely control the device through the mobile app or website, including starting / stopping monitoring, adjusting monitoring cycles, modifying alarm thresholds, and other operations. The remote monitoring module supports multi-user permission management, allowing multiple users to view and analyze the same set of data simultaneously.

[0127] Specific implementation steps include:

[0128] Equipment Installation: Install the electrode array in the transformer's oil tank using a dedicated bracket, ensuring full contact between the electrodes and the insulating oil. Connect the signal processing unit, display and alarm module, and remote monitoring module.

[0129] Parameter settings: Set the monitoring period, alarm threshold, and alarm level through the touch screen. Configure the network parameters of the remote monitoring module to ensure that data can be uploaded to the cloud.

[0130] Electric field application and signal acquisition: The electrode system applies a high-frequency AC electric field and collects electric field signals in real time. The signals are analyzed by the signal processing unit.

[0131] Data Analysis and Alarms: The signal processing unit uses a deep learning algorithm to analyze the electric field signal, identify the type of carbon particles, and calculate their concentration. When the detected carbon particle concentration exceeds the set threshold, the system triggers an audible and visual alarm and sends a text message notification.

[0132] Data storage and remote monitoring: Monitoring data is automatically uploaded to a cloud server, allowing users to view data and analysis reports in real time through the app or website. The system analyzes trends based on historical data and provides maintenance recommendations.

[0133] A third aspect of the present invention relates to a terminal, comprising a processor and a storage medium; the storage medium is used to store instructions; and the processor is used to operate according to the instructions to execute the steps of the method described in the first aspect of the present invention.

[0134] A fourth aspect of the present invention relates to a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect of the present invention.

[0135] Finally, it should be noted that the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art will appreciate that the technical solutions of the present invention still include modifications or equivalent substitutions that may be made to the specific embodiments of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention are intended to be covered by the claims of the present invention.

Claims

1. An intelligent detection method for carbon particles in insulating oil based on microstructure gradient electric field, characterized in that: The method comprises the following steps: Install the electrode array in the transformer's oil tank using a dedicated bracket, ensuring full contact between the electrodes and the insulating oil, and connect the signal processing unit, display alarm module, and remote monitoring module; Set the monitoring cycle, alarm threshold and alarm level through the touch screen, configure the network parameters of the remote monitoring module, and ensure data upload; Apply a high-frequency AC electric field to the electrode system and collect electric field signals in real time. Use a deep learning algorithm to analyze the electric field signals, identify the type of carbon particles, and calculate their concentration. When the detected carbon particle concentration exceeds the set threshold, the system will trigger an audible and visual alarm and send a text message notification.

2. The intelligent detection method for carbon particles in insulating oil based on microstructure gradient electric field according to claim 1, characterized in that: The electrode array is installed in the oil tank of the transformer through a special bracket to ensure that the electrodes are in full contact with the insulating oil, and connected to the signal processing unit, display alarm module and remote monitoring module, including: Some electrodes in the electrode array are needle-tip shaped, and some are planar; The tip diameter of the needle-tip electrode is 50 μm, the size of the planar electrode is 5 mm × 5 mm, and the electrode spacing is adjustable; The surface of the electrode is covered with a layer of multi-layer graphene nanomaterial. The electrode array is fixed in the transformer tank by a high-strength insulating material bracket. The installation angle of the electrode can be adjusted from 0° to 30°.

3. The intelligent detection method for carbon particles in insulating oil based on microstructure gradient electric field according to claim 2, characterized in that: The method of applying a high-frequency alternating electric field to the electrode system and collecting electric field signals in real time includes: The electric field is applied in a high-frequency alternating current electric field with a frequency range of 1kHz to 50kHz and a voltage range of 10V to 100V.

4. The intelligent detection method for carbon particles in insulating oil based on microstructure gradient electric field according to claim 3, characterized in that: The method of applying a high-frequency alternating current electric field to the electrode system and collecting electric field signals in real time, analyzing the electric field signals using a deep learning algorithm, identifying the type of carbon particles and calculating their concentration includes: Perform time-frequency transformation on the signal by using Chirp-Z algorithm; The CLDNet fusion neural network model is used to extract and classify features, and the concentration, particle size and morphology of carbon particles in the electric field signal are automatically identified and classified.

5. The intelligent detection method for carbon particles in insulating oil based on microstructure gradient electric field according to claim 4, characterized in that: The method of applying a high-frequency alternating current electric field to the electrode system and collecting electric field signals in real time, analyzing the electric field signals using a deep learning algorithm, identifying the type of carbon particles and calculating their concentration includes: Automatically adjust the parameters of the processing algorithm according to changes in the actual environment, such as temperature, humidity, oil flow rate, etc.

6. The intelligent detection method for carbon particles in insulating oil based on microstructure gradient electric field according to claim 5, characterized in that: The feature extraction and classification by CNN is used to automatically identify and classify the concentration, particle size and morphology of carbon particles in the electric field signal, including: The output carbon particle information includes concentration, particle size distribution and morphology; The particle size distribution is from micrometer to nanometer level; The morphology includes spherical, lamellar or fibrous.

7. The intelligent detection method for carbon particles in insulating oil based on microstructure gradient electric field according to claim 6, characterized in that: When the detected carbon particle concentration exceeds the set threshold, the system will trigger an audible and visual alarm and send a text message notification, including: When the detected carbon particle concentration exceeds the set threshold, the system will alert the user through sound and light alarms.

8. An intelligent detection device for carbon particles in insulating oil based on microstructure gradient electric field; characterized by: The method is implemented by using the intelligent detection method of carbon particles in insulating oil based on microstructure gradient electric field according to any one of claims 1 to 8; wherein The device includes an equipment installation module, a parameter setting module, a data analysis and alarm module, and a data storage and remote monitoring module; The equipment installation module is used to install the electrode array in the oil tank of the transformer through a dedicated bracket to ensure that the electrodes are in full contact with the insulating oil, and to connect the signal processing unit, the display alarm module and the remote monitoring module; The parameter setting module is used to set the monitoring period, alarm threshold and alarm level through the touch screen, configure the network parameters of the remote monitoring module, and ensure data upload; The data analysis and alarm module is used to apply a high-frequency AC electric field to the electrode system and collect electric field signals in real time, analyze the electric field signals using a deep learning algorithm, identify the type of carbon particles and calculate their concentration; The data storage and remote monitoring module is used to trigger an audible and visual alarm and send a text message notification when the detected carbon particle concentration exceeds a set threshold.

9. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images