Intelligent terminal for fault monitoring of dry-type transformer

Abstract

The invention discloses an intelligent terminal for fault monitoring of a dry-type transformer, and the terminal comprises a LoRa communication module, an FRAM storage module, a power management module, an MCU main control module, a transformer fault monitoring terminal module, and a temperature-vibration-magnetism integrated sensing module. The temperature-vibration-magnetism integrated sensing module is used for monitoring an acceleration value of vibration of the dry-type transformer, a magnetic induction intensity value around an iron core and a temperature value of an internal low-voltage side winding; and the FRAM storage module is used for storing the acceleration value, the temperature value and the magnetic induction intensity value measured by the temperature-vibration-magnetism integrated sensing module. The pressure working range of the temperature-vibration-magnetism integrated sensing module is 1-5 N, and the voltage peak value under the pressure of 2N reaches more than 4000 mV. According to the terminal, the temperature-vibration-magnetic sensor is integrated for the first time, synchronous acquisition of multi-physical field data is achieved, layered preprocessing can be conducted on different signals, and the accuracy and real-time performance of terminal fault diagnosis are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of transformer fault monitoring, and in particular is an intelligent terminal for dry-type transformer fault monitoring. Background Art

[0002] Dry-type transformers are widely used in manufacturing, new energy, and other sectors, playing a vital role in ensuring power stability and safety. With the development of power systems and the growing demand for electricity, the use of dry-type transformers is also increasing. During long-term operation, dry-type transformers are prone to internal hazards such as winding overheating, partial discharge, and insulation degradation, which can affect the normal operation of the transformer.

[0003] At present, traditional transformer fault monitoring methods face many problems. There are significant defects in multi-parameter coordination, real-time performance, accuracy, miniaturization and lightweighting, and intelligent diagnostic algorithms. For example, most existing transformer monitoring sensors are based on a single monitoring data, which makes it difficult to reflect the actual operation of the transformer. For example, Reference 1 (Li Hao, Wei Fanrong, Wang Hao, et al. Power transformer fault diagnosis method based on vibration signal and deep learning [J]. New Technology of Electrical Engineering and Energy, 2024, 43(10): 1-12) proposes a power transformer fault diagnosis method based on vibration signal and deep learning. This method uses a single vibration sensor for fault judgment, but the application of a single sensor cannot make an accurate judgment on the transformer fault. For example, when the vibration signal of the transformer is abnormal but the temperature is stable, it may indicate that the transformer is mechanically loose rather than overloaded. Or it mainly relies on traditional electrical parameters (such as voltage, current) and oil temperature and gas monitoring, lacks physical quantity sensors such as vibration and magnetic field, and cannot detect faults such as abnormal winding mechanical vibration or core magnetic saturation, resulting in a single fault diagnosis dimension and a high missed detection rate. Secondly, traditional transformer monitoring often relies on offline monitoring or periodic inspections, which fail to reflect the transformer's real-time operating status and cannot capture transient faults. Regarding equipment installation convenience, most sensors are large and cannot be installed in the optimal location for parameter monitoring. In terms of intelligent diagnostic algorithms, traditional transformer monitoring relies on fixed thresholds or simple logical judgments, ignoring and unable to handle multi-physics field coupling effects (such as the correlation between temperature, vibration, and magnetic fields). Fault classification capabilities are weak, making it difficult to identify the root causes of faults such as insulation aging and partial discharge, resulting in low fault diagnosis accuracy. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes an intelligent terminal for dry-type transformer fault monitoring. This terminal integrates temperature, vibration, and magnetic sensors for the first time, enabling simultaneous acquisition of multi-physics field data and enabling layered preprocessing of different signals, improving the accuracy and real-time performance of terminal fault diagnosis.

[0005] The present invention solves the technical problem by adopting the following technical solutions:

[0006] A smart terminal for dry-type transformer fault monitoring includes a LoRa communication module, a FRAM storage module, a power management module, an MCU main control module, and a transformer fault monitoring terminal module. The terminal also includes a temperature-vibration-magnetic integrated sensing module for monitoring the acceleration value of the dry-type transformer vibration, the magnetic induction intensity value around the iron core, and the temperature value of the internal low-voltage side winding; the FRAM storage module is used to store the acceleration value, temperature value, and magnetic induction intensity value measured by the temperature-vibration-magnetic integrated sensing module.

[0007] Furthermore, the temperature-vibration-magnetic integrated sensing module includes, from top to bottom, an upper substrate 1, a negative friction layer 2, a positive friction layer 3, a lower substrate 4, and a temperature-magnetic field composite sensing layer 5; the substrate is made of a PDMS film, the negative friction layer is made of a BiFeO3 film modified with PDMS-MXene-PDA, and the positive friction layer uses a copper-plated silver electrode. The temperature-magnetic field composite sensing layer is divided into a temperature zone and a magnetic field zone. The temperature zone is made of a PT100 thermal resistor embedded in a PDA-modified AlN-PDMS film, and the magnetic field zone is composed of a TMR2651 tunnel magnetoresistive sensor encapsulated with an insulating film.

[0008] When in use, the temperature zone of the temperature-vibration-magnetic integrated sensing module is attached between the low-voltage side winding and the iron core of the transformer.

[0009] Furthermore, the preparation process of the temperature-vibration-magnetic integrated sensing module is:

[0010] 1. Preparation of negative electrode friction layer:

[0011] S1: Preparation of PDA-modified nano-BiFeO3 powder: Nano-BiFeO3 powder and dopamine hydrochloride were sequentially added to Tris-HCl buffer solution and mixed evenly, and then centrifuged and dried to obtain PDA-modified nano-BiFeO3 powder, wherein the mass ratio of nano-BiFeO3 powder to dopamine hydrochloride and buffer solution was 200:240:50;

[0012] S2: Preparation of PDA-modified BiFeO3 and PDMS-MXene composite films:

[0013] First, a 5 mg / mL MXene solution was dissolved in the PDMS-a solution (stock solution) and mixed evenly. The amount of MXene solution added was 30% of the stock solution mass. Then, the PDA-modified nano-BiFeO3 powder was added at a mass ratio of 2-3.5%, and magnetic stirring was performed to uniformly mix. Then, the PDMS-b solution (curing solution) was added and magnetic stirring was performed to uniformly mix. Finally, ultrasonic oscillation was performed to obtain a mixed solution of PDA-modified BiFeO3 and PDMS-MXene.

[0014] The mixed solution was injected into a slotted mold, and then placed in a dryer for drying to obtain a composite film of PDA-modified BiFeO3 and PDMS-MXene;

[0015] 2. Preparation of flexible substrate:

[0016] The PDMS-a solution (stock solution) and the curing solution were mixed evenly, ultrasonicated, and printed to obtain a PDMS film;

[0017] 3. Preparation of temperature-magnetic field composite sensing layer:

[0018] S3: Preparation of sensing substrate: PDMS solution is injected into the slotted mold and dried to obtain a sensing substrate PDMS film with a cavity for accommodating the temperature zone;

[0019] S4: Preparation of PDA-modified AlN and PDMS composite films:

[0020] Nano-AlN powder and dopamine hydrochloride were sequentially added to Tris-HCl buffer solution and mixed evenly, and then centrifuged and dried to obtain PDA-modified nano-AlN powder, wherein the mass ratio of nano-AlN powder to dopamine hydrochloride and buffer solution was 3000:240:50;

[0021] The PDA-modified nano-AlN powder is added to the PDMS-a solution (stock solution) at a mass ratio of 50-65%, and after magnetic stirring, the PDMS-b solution (curing solution) is added, and after stirring, ultrasonic vibration is performed to obtain a mixed solution of PDA-modified AlN and PDMS; the mixed solution is then injected into the cavity of the sensing substrate PDMS film in step S3, and dried to obtain a composite film of PDA-modified AlN and PDMS, wherein the composite film of PDA-modified AlN and PDMS is flush with the height of the sensing substrate PDMS film in step S3;

[0022] S5: Encapsulating the PT100 thermal resistor with a polyimide film, and using silver nanowires to lead out the positive and negative electrodes, and fixing them on the composite film of the PDA-modified AlN and PDMS to form a temperature zone;

[0023] S6: A TMR2651 tunnel magnetoresistive sensor encapsulated by a PI film is placed 4-5 mm away from the temperature zone to form a magnetic field zone, and all interfaces are led out with silver nanowires to complete the temperature-magnetic field composite sensing layer;

[0024] 4. Integrated sensor assembly:

[0025] The composite film of BiFeO3 modified by PDA and PDMS-MXene is used as the negative electrode friction layer 2, and the copper-plated silver electrode is used as the positive electrode friction layer 3; conductive glue is pasted on the upper surface of the composite film of BiFeO3 modified by PDA and PDMS-MXene and the lower surface of the copper foil to lead out the positive and negative electrodes; then the flexible substrate is used as the outer package and encapsulated outside the positive and negative electrode friction layers; then the temperature-magnetic field composite sensing layer is encapsulated on the bottom layer to complete the assembly of the integrated sensor.

[0026] Furthermore, the thickness of the upper and lower substrates 1 is 0.2-1 mm, the thickness of the positive and negative friction layers 2 is 0.2-1 mm, and the thickness of the temperature-magnetic field composite sensing layer 5 is 0.5-1 mm; the distance between the adjacent edges of the temperature zone and the magnetic field zone is 4-5 mm.

[0027] Furthermore, the pressure operating range of the temperature-vibration-magnetic integrated sensing module is 1-5N, and the voltage peak reaches more than 4000mV under a pressure of 2N.

[0028] Furthermore, the processing process of the transformer fault monitoring terminal module is:

[0029] (1) placing the temperature-vibration-magnetic integrated sensing module between the low-voltage winding and the iron core of the transformer to collect and obtain the temperature value of the low-voltage side winding of the transformer, the magnetic induction intensity value around the iron core, and the acceleration value of the transformer vibration;

[0030] (2) Layered preprocessing: The temperature signal is preprocessed using a sliding window filter. The filtering formula is as follows:

[0031]

[0032] Among them, N=5 is the sliding window length, X k-i The original temperature sampling value at time ki; y k is the filtered temperature data;

[0033] The acceleration signal is filtered using Butterworth bandpass filtering to extract the energy of the 100Hz-2kHz characteristic frequency band;

[0034] Perform fast Fourier transform and harmonic energy extraction on the magnetic field signal, specifically using the Blackman-Harris window function to suppress spectrum leakage and extract the amplitude characteristics of the 50Hz fundamental wave and the 2nd to 9th harmonics;

[0035] Normalize the signal values ​​after filtering and noise reduction, and use the normalized samples for model training;

[0036] (3) Construct a BKA-CNN-LSTM neural network model integrating attention mechanism to achieve real-time monitoring and early warning of dry-type transformer faults;

[0037] The BKA-CNN-LSTM neural network model integrating the attention mechanism includes a one-dimensional CNN layer for extracting local frequency domain features of the vibration signal, an LSTM layer for capturing temporal variation patterns of temperature / magnetic field, an attention mechanism, and a fully connected layer. The one-dimensional CNN layer includes three one-dimensional convolutional layers and three maximum pooling layers. A ReLU activation function is added after the convolutional layer, and the convolution kernel moves in a single direction from left to right.

[0038] The output of the LSTM layer is processed by the attention mechanism, the fully connected layer, and the softmax function to obtain the fault classification result;

[0039] The hyperparameters of the BKA-CNN-LSTM neural network model integrating the attention mechanism are used as optimization variables of the BKA algorithm. The optimal hyperparameter combination is searched through the BKA algorithm, and the model with the best performance of the hyperparameter combination is selected for fault classification.

[0040] Furthermore, the fitness function of the BKA algorithm is the root mean square error (RMSE) value on the model validation set.

[0041] Furthermore, the fault classification includes five categories: normal, partial discharge, insulation aging, loose core, and winding overheating.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] 1. Regarding integrated sensors for dry-type transformer fault monitoring, we have innovatively developed a temperature-vibration-magnetic integrated sensing module and applied it to dry-type transformer fault detection. Compared to traditional rigid sensors, this module is more corrosion-resistant and resistant to electromagnetic interference, and offers a wider range of installation locations and greater convenience. It utilizes collaborative sensing of temperature, vibration, and magnetic fields to achieve multi-dimensional data fusion of winding temperature, mechanical vibration, and core magnetic field, improving the accuracy and robustness of transformer fault diagnosis compared to traditional single-sensor monitoring.

[0044] 2. The terminal integrates LoRa wireless communication and a FRAM non-volatile memory module, supporting long-distance real-time transmission and historical data storage. Compared to traditional offline monitoring systems, it reduces response latency and can capture transient fault characteristics. Furthermore, it utilizes dynamic power management technology, enabling dual-channel power supply from a battery, enabling low-power operation and extending battery life compared to similar devices. A current transformer collects current from the transformer's low-voltage busbar. When the current is sufficient for power supply, it is processed by the subsequent circuitry and directly supplied to the monitoring device, charging the lithium battery. If the current is insufficient for power supply, the lithium battery is used for power supply, achieving dynamic dual-channel self-powering.

[0045] 3. In terms of transformer fault monitoring, a layered processing method is applied to pre-process temperature and acceleration signals, which significantly reduces the false alarm rate of each signal. The BKA optimization algorithm is used to optimize the model's hyperparameters, which has a strong global search capability and effectively avoids falling into the local optimal solution. A one-dimensional CNN network is used to extract local features of the vibration spectrum. Compared with the multi-dimensional CNN network, the one-dimensional CNN network performs particularly well in extracting vibration signal features. The LSTM layer is used to capture the temporal variation of the temperature signal / magnetic field (such as the temperature slow change trend and the harmonic periodicity of the magnetic field), and the attention mechanism is integrated to dynamically weight and enhance key features, reducing the attention to other information and improving the efficiency and accuracy of task processing. It supports the identification and classification of five types of dry-type transformer faults, such as insulation aging and partial discharge, and realizes real-time monitoring and fault warning of dry-type transformers, improving the efficiency and stability of transformer operation and reducing maintenance costs and safety risks.

[0046] 4. In the hierarchical preprocessing of the present invention, different signals are optimized separately (such as temperature smoothing with a sliding window, vibration extraction of the 100Hz-2kHz frequency band with bandpass filtering, and magnetic field extraction of harmonics with FFT), and the original signal is converted into features suitable for model input. This enables the effective fusion of multimodal data with large differences in the physical properties of temperature, vibration, and magnetic field signals (such as slow temperature changes, high frequency vibration signals, and magnetic fields containing harmonics), avoiding direct fusion that may cause the model to confuse key features, thereby improving signal reliability.

[0047] 5. Use a one-dimensional CNN instead of a two-dimensional one to process vibration signals, reducing the number of parameters. At the same time, set the LSTM hidden state dimension to 32 to balance computational efficiency and feature expression capabilities. The entire network model works together to enable the terminal device (MCU main control module) to meet low power consumption and real-time requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a structural block diagram of an intelligent terminal for dry-type transformer fault monitoring according to the present invention;

[0049] Figure 2This is a schematic diagram of the overall structure of a temperature-vibration-magnetic integrated sensing module according to an embodiment of the present invention;

[0050] Figure 3 Schematic diagram of the preparation process of a temperature-vibration-magnetic integrated sensing module according to an embodiment of the present invention;

[0051] Figure 4 Schematic diagram of the structure of the temperature-magnetic field composite sensing layer of the temperature-vibration-magnetic integrated sensing module of the present invention;

[0052] Figure 5 This is a 200-fold magnified SEM image of the composite film of BiFeO3 and PDMS-MXene modified by PDA of the present invention;

[0053] Figure 6 Graph showing the output voltage of the temperature-vibration-magnetic integrated sensing module under different pressures of the present invention;

[0054] Figure 7 This is a comparison chart of the response signals of the nano-BiFeO3 powder modified with PDA and the temperature-vibration-magnetic integrated sensing module without PDA;

[0055] Figure 8 This is a schematic diagram of the structure of a BKA-CNN-LSTM neural network model integrating an attention mechanism according to an embodiment of the present invention;

[0056] Figure 9 Schematic diagram of the BKA algorithm in the present invention.

[0057] Explanation of the accompanying symbols: 1. upper substrate; 2. negative electrode friction layer; 3. positive electrode friction layer; 4. lower substrate; 5. temperature-magnetic field composite sensing layer. DETAILED DESCRIPTION

[0058] Specific embodiments are given below in conjunction with the accompanying drawings. The specific embodiments are only used to introduce the technical solutions of the present invention in detail and are not intended to limit the scope of protection of the present application.

[0059] Example 1

[0060] This embodiment is used for the integrated terminal for dry-type transformer fault monitoring (referred to as terminal, see Figure 1), including a temperature-vibration-magnetic integrated sensing module, a LoRa communication module, a FRAM storage module, a power management module, an MCU main control module, and a transformer fault monitoring terminal module. The MCU main control module is electrically connected to the temperature-vibration-magnetic integrated sensing module, the LoRa communication module, the storage module, and the power management module, and the MCU main control module communicates the collected data with the transformer fault monitoring terminal module. The temperature-vibration-magnetic integrated sensing module is used to monitor the acceleration value of the dry-type transformer vibration, the magnetic induction intensity value around the iron core, and the temperature value of the internal low-voltage side winding. The FRAM storage module is used to store the acceleration value, temperature value, and magnetic induction intensity value measured by the temperature-vibration-magnetic integrated sensing module.

[0061] When in use, the temperature zone of the temperature-vibration-magnetic integrated sensing module is attached between the low-voltage side winding and the iron core of the transformer.

[0062] Example 2

[0063] The LoRa communication module in this embodiment includes an antenna switch and a main control chip, the model of which is SX1276IMLTRT. The model of the antenna switch is PE4259.

[0064] The FRAM storage module includes a storage chip, the model of which is CY15B104Q-LHXI.

[0065] The power management module includes a charging management chip and a linear voltage regulator. The charging management chip model is TP4056, and the linear voltage regulator model is LM1117.

[0066] Example 3

[0067] like Figure 2 As shown, in this embodiment, the temperature-vibration-magnetic integrated sensing module includes an upper substrate 1, a negative friction layer 2, a positive friction layer 3, a lower substrate 4, and a temperature-magnetic field composite sensing layer 5. The upper and lower substrates are made of PDMS thin films, the negative friction layer is made of a BiFeO3 thin film modified with PDMS-MXene-PDA, and the positive friction layer is made of a copper-plated silver electrode. The temperature-magnetic field composite sensing layer is divided into a temperature zone and a magnetic field zone. The temperature zone is made of a PT100 thermal resistor embedded in a PDA-modified AlN-PDMS thin film, and the magnetic field zone is composed of a TMR2651 tunnel magnetoresistive sensor encapsulated with an insulating film. The acquisition of temperature and magnetic field is not affected by the upper layer, and the distance between the temperature zone and the magnetic field zone is 4-5mm, which can achieve interference-free detection of temperature and magnetic field. The temperature-vibration-magnetic integrated sensing module is a flexible porous integrated sensor that simultaneously collects temperature, vibration, and magnetic field signals, and has the monitoring function of three parameters: vibration, temperature, and magnetic field.

[0068] Example 4

[0069] The preparation process of the temperature-vibration-magnetic integrated sensing module in this embodiment is as follows: Figure 3 As shown, the following steps are included:

[0070] 1. Preparation of negative electrode friction layer:

[0071] S1: Preparation of PDA-modified nano-BiFeO3 powder:

[0072] 1) Place 50 mg of Tris-HCl buffer in a beaker, add 0.2 g of BiFeO3 powder, then stir for 20 minutes and ultrasonicate for 30 minutes.

[0073] 2) Take 240 mg of dopamine hydrochloride and add it to the beaker in step 1), and stir with a magnetic stirrer for 12 hours.

[0074] 3) The liquid obtained in step 2) was centrifuged twice with deionized water at 3000 rpm for 5 min.

[0075] 4) The liquid obtained in step 3) was vacuum filtered, and then rinsed with deionized water and ethanol until the solution was clear and colorless.

[0076] 5) The BiFeO3 obtained in step 4) was placed in a vacuum oven and dried at 80° C. for 24 h to obtain PDA-modified nano-BiFeO3 powder.

[0077] S2: Preparation of PDA-modified BiFeO3 and PDMS-MXene composite films

[0078] 1) Take 5 g of PDMS-a solution (stock solution) in a beaker, add 1.5 g of MXene solution, and stir the mixed solution in a magnetic stirrer for 10 min.

[0079] 2) 0.15 g of the PDA-modified nano-BiFeO3 powder prepared in S1 was added to the liquid obtained in step 1) and stirred for 30 min.

[0080] 3) 0.5 g of PDMS-b solution (curing solution) was added to the mixed liquid obtained in step 2), and then stirred with a magnetic stirrer for 30 minutes.

[0081] 4) The mixed solution obtained in step 3) is subjected to ultrasonic oscillation to obtain a mixed solution of PDA-modified BiFeO3 and PDMS-MXene.

[0082] 5) The mixed solution obtained in step 4) was injected into a rectangular (18 mm × 9 mm × 1 mm) slotted mold printed by a 3D printer, and then placed in a dryer and dried at 70°C for 4 h to obtain a composite film of PDA-modified BiFeO3 and PDMS-MXene with a thickness of 1 mm.

[0083] 2. Flexible substrate preparation:

[0084] Take 5g of PDMS-a solution (original solution) in a beaker, take 0.5g of PDMS-b solution (curing solution) and add it to the beaker. After magnetic stirring, ultrasonic and printing can obtain a PDMS film with a thickness of 1mm on the upper and lower substrates.

[0085] 3. Preparation of temperature-magnetic field composite sensing layer:

[0086] The temperature sensor model is PT100 thermal resistor, and the magnetic field sensor model is TMR2651 tunnel magnetoresistive sensor. The temperature-magnetic field composite sensing layer is divided into a temperature zone and a magnetic field zone, such as Figure 4 shown.

[0087] The temperature zone is made of PT100 thermal resistor and PDA modified AlN-PDMS film:

[0088] S3: Preparation of sensing substrate: injecting PDMS solution into the slotted mold and drying to obtain a sensing substrate PDMS film having a cavity for accommodating the temperature zone; the thickness of the sensing substrate is the same as that of the flexible substrate.

[0089] A rectangular (18mm × 9mm × 1mm) slotted mold with a hollowed-out temperature zone was printed using a 3D printer. The slotted plane dimensions were 4mm × 4mm, meaning the temperature zone was 4mm × 4mm. PDMS solution was injected into the slotted mold and dried in a dryer at 70°C for 4 hours. This yielded the hollowed-out PDMS film on the right, which served as the sensing substrate.

[0090] S4: Preparation of PDA-modified AlN and PDMS composite films:

[0091] The preparation of PDA-modified nano-AlN powder is similar to that of PDA-modified nano-BiFeO3 powder, except that the weight of AlN powder is changed to 3g, and the process will not be described again. Finally, PDA-modified nano-AlN powder is obtained. The PDA-modified nano-AlN powder is added to the PDMS-a solution (stock solution) at a mass dosage of 60% (based on the mass percentage of nano-powder in the stock solution), and after magnetic stirring, the PDMS-b solution (curing solution) is added, and after stirring evenly, ultrasonic vibration is performed to obtain a mixed solution of PDA-modified AlN and PDMS; the mixed solution is then injected into the cavity of the sensing substrate PDMS film in step S3, and dried to obtain a composite film of PDA-modified AlN and PDMS. The composite film of PDA-modified AlN and PDMS is flush with the height of the sensing substrate PDMS film in step S3.

[0092] The mixed solution of PDA-modified AlN and PDMS was injected into the 4×4 mm empty groove, and the drying conditions were: drying at 60° C. for 12 h.

[0093] S5: Encapsulating the PT100 thermal resistor with a polyimide film, and using silver nanowires to lead out the positive and negative electrodes, and fixing them on the composite film of the PDA-modified AlN and PDMS to form a temperature zone;

[0094] S6: A TMR2651 tunnel magnetoresistive sensor encapsulated by a PI film is placed 4-5 mm away from the temperature zone to form a magnetic field zone, and all interfaces are led out with silver nanowires to complete the temperature-magnetic field composite sensing layer;

[0095] 4. Integrated sensor assembly:

[0096] The composite film of BiFeO3 modified by PDA and PDMS-MXene is used as the negative electrode friction layer 2, and the copper-plated silver electrode is used as the positive electrode friction layer 3; conductive glue is pasted on the upper surface of the composite film of BiFeO3 modified by PDA and PDMS-MXene and the lower surface of the copper foil to lead out the positive and negative electrodes; then the flexible substrate is used as the outer package and encapsulated outside the positive and negative electrode friction layers; then the temperature-magnetic field composite sensing layer is encapsulated on the bottom layer to complete the assembly of the integrated sensor.

[0097] In the present invention, the composite film of BiFeO3 modified by PDA and PDMS-MXene is used as the negative electrode friction layer. Among them, the porous structure made of PDMS-MXene enhances the specific surface area and conductivity compared with the pure PDMS film. After the MXene is composited with PDMS, a conductive network is formed, which improves the charge transfer efficiency of the friction electrode and enhances the output intensity of the vibration signal. It also has excellent mechanical flexibility. The groups on its surface have good hydrophilicity and are easy to be chemically modified on the surface and composited with other materials. Figure 5As shown in the 200x SEM image, the magnified image clearly shows a relatively smooth surface with some bubbling. The added nano-BiFeO3 exhibits excellent ferroelectric properties, enhancing the film's conductivity and further improving its response speed without affecting magnetic field detection. It also increases the film's specific surface area, improving its dielectric and ferroelectric properties, and increasing the film's effective specific surface area and frequency response range. Polydopamine (PDA) is used to modify nano-BiFeO3, enhancing its surface adhesion and improving the interfacial compatibility between BiFeO3 and PDMS-MXene. Filling the porous structure creates a tighter bond between the two and inhibits the oxidative activity of MXene, improving its triboelectric properties. The copper-plated silver electrode, used as the positive electrode friction layer, reduces contact resistance compared to pure copper electrodes, improves the electrode's antioxidant capacity, and forms a triboelectric effect with the negative electrode friction layer, enabling the conversion of mechanical vibrations into electrical signals. The synergistic effect of these substances enables highly sensitive detection of mechanical vibrations.

[0098] Figure 6 This graph shows the output of the integrated temperature-vibration-magnetic sensor module under different pressures. The sensor exhibits excellent sensitivity at relatively low pressures (1-5N), with the signal increasing linearly. However, the increase decreases when the force exceeds 5N, no longer being linear. Therefore, the optimal operating range for this sensor is considered to be 1-5N.

[0099] Figure 7 This graph compares the response signals of BiFeO3 materials modified with PDA and those without PDA at a pressure of 2 N. It can be seen that the sensor with the modified ferromagnetic material exhibits a higher signal during testing than the sensor without it. A numerical comparison shows that the temperature-vibration-magnetic integrated sensing module without the ferromagnetic material achieves approximately 85% of the performance of the sensor with the modified material, with a peak voltage exceeding 4000 mV at a pressure of 2 N. This demonstrates that the addition of the modified ferromagnetic material can significantly improve sensor performance at lower pressures and frequencies (1 to 10 Hz).

[0100] The PT100 thermal resistor in the temperature-magnetic field composite sensing layer has a wide temperature range of -50 to +200°C, meeting the full temperature range of the transformer. By using the characteristic that the resistance value changes with temperature, it can monitor the temperature changes of the transformer winding in real time.

[0101] A composite film of PDA-modified AlN and PDMS was selected as the temperature zone substrate. The PDA-modified AlN exhibits strong surface adhesion, high thermal conductivity, and insulation properties, enabling the temperature sensor to quickly respond to changes in internal transformer temperature rise, improving response speed. This also avoids delamination issues caused by differences in thermal expansion coefficients, while also enhancing the film's mechanical strength and environmental tolerance.

[0102] Based on the tunnel magnetoresistance effect, the sensor is highly sensitive to weak magnetic field changes caused by core magnetic saturation or partial discharge, with a wide dynamic range of ±50mT, meeting the fluctuation range of abnormal magnetic fields in transformer cores. Polyimide packaging further enhances the sensor's insulation protection and electromagnetic interference resistance, ensuring stable operation in high-voltage environments and extending its lifespan.

[0103] Example 5

[0104] In this embodiment, the processing process of the transformer fault monitoring terminal module is:

[0105] (1) The above-mentioned integrated sensor is placed between the low-voltage winding and the iron core of the transformer to collect and obtain the temperature value of the low-voltage side winding of the transformer, the magnetic induction intensity value around the iron core, and the acceleration value of the transformer vibration, which are used as the original signal.

[0106] (2) Performing hierarchical preprocessing on the above raw signals, including filtering, noise reduction and normalization. The hierarchical preprocessing method includes preprocessing of temperature signals, acceleration signals and magnetic field signals.

[0107] The temperature signal is preprocessed using sliding window filtering, and the filtering formula is as follows:

[0108]

[0109] Among them, N=5 is the sliding window length, X k-i The original temperature sampling value at time ki; y k is the filtered temperature data.

[0110] Sliding window filtering is a time-domain filtering method that statistically processes a certain number of consecutive data points to smooth the original signal and reduce the effects of random noise. This method is simple, computationally inefficient, and well-suited for resource-constrained microcontroller applications. Its core concept is to maintain a fixed-size data window. Whenever new data arrives, the oldest data point is removed and the new data point is added to the window. The data within the window is then processed to produce the filtered output.

[0111] The acceleration signal is filtered using a Butterworth bandpass filter to extract the energy of the 100Hz-2kHz characteristic frequency band. The transfer function H is:

[0112]

[0113] Among them, w c is the cutoff frequency.

[0114] The Butterworth filter is a commonly used filter characterized by a flat passband response with no fluctuations and a gradually increasing attenuation characteristic. On a Bode plot of the logarithm of amplitude versus diagonal frequency, starting from a certain boundary angular frequency, the amplitude gradually decreases as the angular frequency increases, tending towards negative infinity.

[0115] The magnetic field signal is subjected to fast Fourier transform and harmonic energy extraction, specifically using the Blackman-Harris window function to suppress spectrum leakage and extract the amplitude characteristics of the 50 Hz fundamental wave and the 2nd to 9th harmonics.

[0116] The mathematical expression of the Blackman-Harris window function is as follows:

[0117]

[0118] Its coefficient configuration is: a0=0.35875, a1=0.48829, a2=0.14128, a3=0.01168.

[0119] The Blackman window is a weighted window function that limits the amplitude attenuation outside the window boundaries by weighting and truncating the signal. It has low sidelobe amplitude and a narrow mainlobe width, which can reduce the impact of spectral leakage to a certain extent. It is suitable for applications requiring high frequency resolution.

[0120] The signal value after filtering and noise reduction is normalized. The normalization formula is as follows:

[0121]

[0122] Among them, x n It represents the characteristic value obtained by normalizing the collected temperature value, magnetic induction intensity value and acceleration value. x is the original characteristic data, x max is the maximum value of x, x min is the minimum value of x. The normalized samples are divided into two parts, used as training set and test set respectively.

[0123] The role of normalization is to map the value of a certain feature to a fixed interval between [0,1], eliminating the impact of dimension on the final result, making different features comparable, and making features that may have originally had large distribution differences have the same weight impact on the model, thereby improving the convergence speed of the model. In deep learning, data normalization can prevent model gradient explosion.

[0124] (3) Construct a BKA-CNN-LSTM neural network model that integrates the attention mechanism to perform feature extraction, data classification, and data prediction on the training set data to achieve real-time monitoring and early warning of dry-type transformer faults.

[0125] The BKA-CNN-LSTM neural network model with integrated attention mechanism (see Figure 8 ) includes a one-dimensional CNN layer for extracting local frequency domain features of vibration signals, an LSTM layer for capturing temporal variations in temperature / magnetic field, an attention mechanism, and a fully connected layer; the one-dimensional CNN layer includes three one-dimensional convolutional layers and three maximum pooling layers, with a ReLU activation function added after the convolutional layer, and the convolution kernel moves in a single direction, from left to right;

[0126] The output of the LSTM layer is processed by the attention mechanism, the fully connected layer, and the softmax function to obtain the fault classification result;

[0127] The hyperparameters of the BKA-CNN-LSTM neural network model integrating the attention mechanism are used as optimization variables of the BKA algorithm. The optimal hyperparameter combination is searched through the BKA algorithm, and the model with the best hyperparameter combination is selected for fault classification;

[0128] Furthermore, the mathematical expression of the ReLU activation function is as follows:

[0129] ReLU(x)=max(0,x)

[0130] When the input function is less than or equal to 0, the output is 0. When the input function is greater than 0, the output is equal to the input. Applying the ReLU activation function can solve the problem of vanishing gradients, improve the nonlinear capabilities of the model, and reduce computational complexity, saving computational costs and accelerating model learning.

[0131] The unidirectional LSTM part is connected after the CNN neural network.

[0132] The attention mechanism is connected after the LSTM part and is used to focus on the information that is more critical to the current task among a large amount of input information, reduce attention to other information, and improve the efficiency and accuracy of task processing.

[0133] A fully connected layer with a Softmax activation function is connected after the attention mechanism part, with an output channel of 5, in order to achieve the purpose of classifying transformer faults.

[0134] The Softmax activation function is used for the output of multi-classification problems. It is essentially an activation function that normalizes a numerical vector into a probability distribution vector, where the sum of all probabilities is 1. Its mathematical expression is as follows:

[0135]

[0136] The input is a vector containing K elements. z represents an element of a vector.

[0137] The root mean square error (RMSE) value on the model validation set is defined as the fitness function of BKA.

[0138] The calculation formula of the root mean square error is:

[0139]

[0140] In step S3, the model hyperparameters are used as optimization variables of the BKA algorithm, and the optimal hyperparameter combination is searched through the BKA algorithm.

[0141] The Black Kite Optimization Algorithm (BKA) is a new bio-inspired algorithm that not only captures the hunting and migration behavior of black kites, but also simulates their high adaptability to environmental changes and target locations. The algorithm works as follows:

[0142] 1. Random distribution of population locations

[0143] Generate random individual positions in the initial solution space, the formula is as follows:

[0144] P i =L b +rand(U b -L b )

[0145] Among them U b and L b are the upper and lower limits of the search range, and rand is a random number between [0,1].

[0146] 2. Hunting behavior

[0147] The capture behavior uses sine and exponential functions to achieve global and local searches and quickly locate potential optimal areas. The formula is as follows:

[0148]

[0149] Where q is a random number between [0, 1], n is the current iteration number, N is the total iteration number, and the random number c = 0.9.

[0150] 3. Migration behavior

[0151] The algorithm simulates the migration behavior of black kites. If the current population fitness value is less than the random population fitness value, a new leader is selected to escape the local optimal solution. The formula is as follows:

[0152]

[0153] z=2sin(q+π / 2)

[0154] Among them, F v is the current population fitness value, F r is the fitness value of the random population.

[0155] 4. Optimal value selection

[0156] The algorithm selects the individual with the best fitness value in the current population as the leader. The formula is as follows:

[0157] f min =min(f(P i ))

[0158] P l =P(find(f min = =f(P i )))

[0159] like Figure 9 As shown in the figure, the BKA algorithm is used to optimize the hyperparameters of the CNN-LSTM neural network model with integrated attention mechanism, including the number and size of convolution kernels, the number of LSTM hidden units, the number of attention heads, and the learning rate.

[0160] The training set is used to train the CNN-LSTM neural network model with fusion attention mechanism, and the validation set is used to evaluate the performance of the model under different hyperparameters. The best hyperparameter combination is selected to obtain the final BKA-CNN-LSTM model with fusion attention mechanism.

[0161] Furthermore, based on the constructed BKA-CNN-LSTM model integrating the attention mechanism, the test set data is used to test it, and various numerical evaluation indicators are output to determine whether they reach the expected values.

[0162] The numerical evaluation indicators include mean square error (MSE) and mean absolute error (MAE), and their formulas are as follows:

[0163]

[0164] The present invention uses a temperature-vibration-magnetic integrated sensing module to collect three characteristic signals of the dry-type transformer and predict its fault condition, providing certain technical support for improving the reliability of the transformer, reducing maintenance costs and safety risks in actual work, and at the same time providing a corresponding theoretical basis for the intelligent fault monitoring of dry-type transformers.

[0165] Any matters not described in the present invention are applicable to the prior art.

Claims

1. An intelligent terminal for dry-type transformer fault monitoring, comprising a LoRa communication module, a FRAM storage module, a power management module, an MCU main control module and a transformer fault monitoring terminal module, characterized in that: The terminal also includes a temperature-vibration-magnetic integrated sensing module, which is used to monitor the acceleration value of the dry-type transformer vibration, the magnetic induction intensity value around the iron core, and the temperature value of the internal low-voltage side winding; The FRAM storage module is used to store the acceleration value, temperature value and magnetic induction intensity value measured by the temperature-vibration-magnetic integrated sensing module.

2. The intelligent terminal according to claim 1, characterized in that: The temperature-vibration-magnetic integrated sensing module comprises, from top to bottom, an upper substrate, a negative friction layer, a positive friction layer, a lower substrate, and a temperature-magnetic field composite sensing layer. The substrate is made of a PDMS film, the negative friction layer is made of a BiFeO3 film modified with PDMS-MXene-PDA, and the positive friction layer is a copper-plated silver electrode. The temperature-magnetic field composite sensing layer is divided into a temperature zone and a magnetic field zone. The temperature zone is made of a PT100 thermal resistor embedded in a PDA-modified AlN-PDMS film, and the magnetic field zone is composed of a TMR2651 tunnel magnetoresistive sensor encapsulated with an insulating film. When in use, the temperature zone of the temperature-vibration-magnetic integrated sensing module is attached between the low-voltage side winding and the iron core of the transformer.

3. The intelligent terminal according to claim 1, wherein: The preparation process of the temperature-vibration-magnetic integrated sensing module is as follows:

1. Preparation of negative electrode friction layer: S1: Preparation of PDA-modified nano-BiFeO3 powder: Nano-BiFeO3 powder and dopamine hydrochloride were sequentially added to Tris-HCl buffer solution and mixed evenly, and then centrifuged and dried to obtain PDA-modified nano-BiFeO3 powder, wherein the mass ratio of nano-BiFeO3 powder to dopamine hydrochloride and buffer solution was 200:240:50; S2: Preparation of PDA-modified BiFeO3 and PDMS-MXene composite films: First, a 5 mg / mL MXene solution was dissolved in the PDMS-a solution and mixed evenly. The amount of MXene solution added was 30% of the original solution mass. Then, the PDA-modified nano-BiFeO3 powder was added at a mass ratio of 2-3.5%, and magnetic stirring was performed to uniformly mix. Then, the PDMS-b solution was added and magnetic stirring was performed to uniformly mix. Finally, ultrasonic oscillation was performed to obtain a mixed solution of PDA-modified BiFeO3 and PDMS-MXene. The mixed solution was injected into a slotted mold, and then placed in a dryer for drying to obtain a composite film of PDA-modified BiFeO3 and PDMS-MXene; 2. Preparation of flexible substrate: The PDMS-a solution and the PDMS-b solution were mixed evenly, ultrasonicated, and printed to obtain a PDMS film; 3. Preparation of temperature-magnetic field composite sensing layer: S3: Preparation of sensing substrate: PDMS solution is injected into the slotted mold and dried to obtain a sensing substrate PDMS film with a cavity for accommodating the temperature zone; S4: Preparation of PDA-modified AlN and PDMS composite films: Nano-AlN powder and dopamine hydrochloride were sequentially added to Tris-HCl buffer solution and mixed evenly, and then centrifuged and dried to obtain PDA-modified nano-AlN powder, wherein the mass ratio of nano-AlN powder to dopamine hydrochloride and buffer solution was 3000:240:50; The PDA-modified nano-AlN powder is added to the PDMS-a solution at a mass ratio of 50-65%, and the PDMS-b solution is added after magnetic stirring. After stirring, ultrasonic vibration is performed to obtain a mixed solution of PDA-modified AlN and PDMS; the mixed solution is then injected into the cavity of the sensing substrate PDMS film in step S3, and dried to obtain a composite film of PDA-modified AlN and PDMS, wherein the composite film of PDA-modified AlN and PDMS is flush with the height of the sensing substrate PDMS film in step S3; S5: Encapsulating the PT100 thermal resistor with a polyimide film, and using silver nanowires to lead out the positive and negative electrodes, and fixing them on the composite film of the PDA-modified AlN and PDMS to form a temperature zone; S6: A TMR2651 tunnel magnetoresistive sensor encapsulated by a PI film is placed 4-5 mm away from the temperature zone to form a magnetic field zone, and all interfaces are led out with silver nanowires to complete the temperature-magnetic field composite sensing layer; 4. Integrated sensor assembly: The composite film of BiFeO3 modified by PDA and PDMS-MXene is used as the negative electrode friction layer, and the copper-plated silver electrode is used as the positive electrode friction layer; conductive glue is pasted on the upper surface of the composite film of BiFeO3 modified by PDA and PDMS-MXene and the lower surface of the copper foil to lead out the positive and negative electrodes; then the flexible substrate is used as the outer package and encapsulated outside the positive and negative electrode friction layers; then the temperature-magnetic field composite sensing layer is encapsulated on the bottom layer to complete the assembly of the integrated sensor.

4. The intelligent terminal according to claim 2, characterized in that The thickness of the upper and lower substrates is 0.2-1mm, the thickness of the positive and negative friction layers is 0.2-1mm, and the thickness of the temperature-magnetic field composite sensing layer is 0.5-1mm; the distance between the adjacent edges of the temperature zone and the magnetic field zone is 4-5mm.

5. The intelligent terminal according to claim 2, characterized in that: The pressure operating range of the temperature-vibration-magnetic integrated sensing module is 1-5N, and the voltage peak reaches more than 4000mV under a pressure of 2N.

6. The intelligent terminal according to claim 1, characterized in that: The processing process of the transformer fault monitoring terminal module is: (1) placing the temperature-vibration-magnetic integrated sensing module between the low-voltage winding and the iron core of the transformer to collect and obtain the temperature value of the low-voltage side winding of the transformer, the magnetic induction intensity value around the iron core, and the acceleration value of the transformer vibration; (2) Layered preprocessing: The temperature signal is preprocessed using a sliding window filter. The filtering formula is as follows: Among them, N=5 is the sliding window length, X k-i The original temperature sampling value at time ki; y k is the filtered temperature data; The acceleration signal is filtered using Butterworth bandpass filtering to extract the energy of the 100Hz-2kHz characteristic frequency band; Perform fast Fourier transform and harmonic energy extraction on the magnetic field signal, specifically using the Blackman-Harris window function to suppress spectrum leakage and extract the amplitude characteristics of the 50Hz fundamental wave and the 2nd to 9th harmonics; Normalize the signal values ​​after filtering and noise reduction, and use the normalized samples for model training; (3) Construct a BKA-CNN-LSTM neural network model integrating attention mechanism to achieve real-time monitoring and early warning of dry-type transformer faults; The BKA-CNN-LSTM neural network model integrating the attention mechanism includes a one-dimensional CNN layer for extracting local frequency domain features of the vibration signal, an LSTM layer for capturing temporal variation patterns of temperature / magnetic field, an attention mechanism, and a fully connected layer. The one-dimensional CNN layer includes three one-dimensional convolutional layers and three maximum pooling layers. A ReLU activation function is added after the convolutional layer, and the convolution kernel moves in a single direction from left to right. The output of the LSTM layer is processed by the attention mechanism, the fully connected layer, and the softmax function to obtain the fault classification result; The hyperparameters of the BKA-CNN-LSTM neural network model integrating the attention mechanism are used as optimization variables of the BKA algorithm. The optimal hyperparameter combination is searched through the BKA algorithm, and the model with the best performance of the hyperparameter combination is selected for fault classification.

7. The intelligent terminal according to claim 6, characterized in that: The fitness function of the BKA algorithm is the root mean square error (RMSE) value on the model validation set.

8. The intelligent terminal according to claim 6, characterized in that: The fault classification includes five categories: normal, partial discharge, insulation aging, loose core, and winding overheating.

Images