Device and method for detecting dissolved gas in oil

Abstract

The invention relates to the technical field of transformer fault diagnosis, in particular to a device and a method for detecting dissolved gas in oil. The device comprises a detection sensor, a signal conversion module and a diagnosis module. Compared with the prior art, the performance bottleneck of a traditional gas sensor is fundamentally broken through through an original cantilever beam double-sided heterogeneous MOFs modification structure and an in-situ growth process: the in-situ growth process ensures firm combination of a sensitive coating and a substrate, and the defects that the coating is easy to fall off and poor in uniformity in a traditional coating method are remarkably overcome; a reliable hardware basis is provided for long-term online monitoring; and a double-sided heterogeneous modification strategy enables a single sensing unit to generate collaborative sensing and differential response to various fault gases, so that the core problems of poor selectivity and serious cross interference of a traditional sensor are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of transformer fault diagnosis technology, and in particular to a device and method for detecting dissolved gas in oil. Background Technology

[0002] Electricity is a core driving force for my country's economic and social development. Statistics show that in 2022, total electricity consumption reached 8.64 trillion kilowatt-hours, a year-on-year increase of 3.6%. With the continuous growth in electricity demand and increasingly stringent requirements for power grid stability, the safe operation of transformers, as key equipment for power transmission, is of paramount importance. Transformers themselves have complex structures and high costs; once a fault occurs, not only are repair costs high, but it can also trigger large-scale power outages, severely impacting the entire power system. Compositional analysis of dissolved gases in transformer oil can effectively diagnose potential equipment faults—different gas proportions often correspond to specific fault types. Therefore, accurately predicting the gas concentration in the oil has become an important technical means of predicting the operating status of transformers, and the maturity and widespread adoption of online monitoring systems provide a solid data foundation for achieving this goal.

[0003] Transformer oil plays multiple roles in oil-immersed transformers, including insulation, heat dissipation, and arc extinguishing. Currently, mineral-based transformer oil is widely used. During transformer operation, the oil produces small amounts of gas due to normal aging. However, in abnormal conditions such as overheating or discharge, the insulating oil and solid insulating materials undergo decomposition, generating fault gases with distinct characteristics. Although transformer oil has a complex chemical composition, its molecules are mainly composed of carbon-carbon and carbon-hydrogen bonds. When localized high temperatures or discharges occur inside the equipment, these chemical bonds break and recombine to form gases including methane, ethane, ethylene, hydrogen, carbon monoxide, and carbon dioxide. It is noteworthy that the generation of different gases is closely related to temperature: methane dominates at medium and low temperatures; as the temperature increases, the proportions of ethylene and hydrogen gradually rise; and acetylene is only produced in large quantities under high-temperature arcs exceeding 800°C, making it a key indicator for identifying serious faults. Summary of the Invention

[0004] This invention discloses a dissolved gas detection device in oil, which is implemented through the following technical solution: it includes a detection sensor, which includes a cantilever beam. A layer of macroporous MOFs material is constructed on the front side of the cantilever beam using in-situ growth technology, and a layer of microporous MOFs material is constructed on the back side of the cantilever beam using in-situ growth technology.

[0005] Furthermore, the macroporous MOF material is MIL-100(Fe), which has a window pore size of 2.5-2.9 nm and a three-dimensional cage-like channel, preferentially adsorbing and enriching macromolecular gases in transformer insulating oil.

[0006] Furthermore, the macromolecular gas includes ethylene (C2H4) and ethane (C2H6).

[0007] Furthermore, the microporous MOF material is ZIF-8, with a pore size of approximately 0.34 nm, which can repel large molecular gases and allow small molecular gases to enter its pores and be adsorbed.

[0008] Furthermore, the small molecule gas includes hydrogen (H2) and carbon monoxide (CO).

[0009] Furthermore, the cantilever beam is fabricated using MEMS technology and is made of silicon or silicon nitride material;

[0010] The dimensions of the cantilever beam are: length 100-500μm, width 20-100μm, and thickness 1-5μm.

[0011] One end of the cantilever beam is rigidly fixed to the base, while the other end is a free end, forming a resonant / deformation structure that is sensitive to mass and surface stress.

[0012] Furthermore, a macroporous MOF material and a microporous MOF material were respectively constructed on the front and back of the cantilever beam using in-situ growth technology. The specific method is as follows:

[0013] The MEMS chip with the bare cantilever beam was placed in acetone, ethanol and deionized water in sequence for ultrasonic cleaning to remove organic pollutants and particulate matter.

[0014] Dry with high-purity nitrogen or in a critical point dryer to avoid damage to the beam structure due to surface tension;

[0015] The cleaned chip is placed in an oxygen plasma treatment instrument and treated for 1-5 minutes at a specific power to remove residual organic matter and introduce hydroxyl groups (-OH) onto the cantilever beam surface to enhance hydrophilicity and chemical activity.

[0016] The activated chip was immersed in a toluene solution containing 1% (3-aminopropyl)triethoxysilane (APTES) and reacted at 60-80°C for 2-4 hours. This process grafted amino functional groups onto the surface, which served as anchoring sites for heterogeneous nucleation of MOF crystals, thereby improving the adhesion of the MOF film.

[0017] Photolithography or a precision mechanical mask fixture is used to physically cover and protect the reverse side of the cantilever beam.

[0018] The protected chip was placed in a DMF / H2O mixed solution of ferric chloride hexahydrate and trimesic acid;

[0019] By maintaining a temperature of 100 degrees Celsius for 12 hours, MIL-100(Fe) undergoes heterogeneous nucleation and epitaxial growth only on the unprotected front side, forming a uniform thin film.

[0020] After completing the front-side growth, remove the protective layer on the back side;

[0021] Protect the already grown front side and expose the back side;

[0022] The chip was placed in a methanol solution of zinc nitrate hexahydrate and 2-methylimidazole.

[0023] The reverse ZIF-8 coating was grown after being kept at room temperature for 4 hours.

[0024] After growth is complete, the chip is gently soaked and rinsed in the appropriate solvent to remove loose crystals physically adsorbed on the surface.

[0025] Heat treatment at 80-150°C under an inert atmosphere or vacuum to completely remove solvent molecules from the pores and activate adsorption sites.

[0026] The prepared sensor undergoes preliminary electrical aging treatment to stabilize its sensitivity characteristics.

[0027] Furthermore, the device also includes a signal conversion module and a diagnostic module;

[0028] The signal conversion module states that after the front and back sides of the cantilever beam adsorb different gas molecules, the equivalent mass of the cantilever beam increases, the resonant frequency decreases, and the cantilever beam undergoes bending deformation. A composite electrical signal containing the resonant frequency Δf and the bending deformation ΔZ is obtained through an integrated piezoresistive sensor or optical displacement detection device at the root of the cantilever beam.

[0029] The diagnostic module analyzes the composite electrical signal and uses a deep learning model to obtain the types and contents of dissolved gases in the oil.

[0030] This invention also discloses a method for detecting dissolved gas in oil, characterized by the following specific method:

[0031] Place the cantilever beam in the insulating oil to be tested and leave it for a preset time;

[0032] The front side of the cantilever beam adsorbs and enriches large molecular gases in the transformer insulating oil, while the back side of the cantilever beam adsorbs and enriches small molecular gases in the transformer insulating oil.

[0033] Due to the mass loading effect of adsorbed molecules, the weight of the cantilever beam changes. The composite electrical signal is obtained by an integrated piezoresistive sensor or optical displacement detection device at the root of the cantilever beam. The composite electrical signal has the characteristics of changes in the cantilever beam's resonant frequency Δf and bending deformation ΔZ.

[0034] The composite electrical signal is analyzed, and the dissolved gas content in the oil is output through a deep learning model.

[0035] Furthermore, the deep learning model is based on the CPO-1DCNN convolutional neural network model.

[0036] Due to the adoption of the above technical solutions, this application has the following beneficial effects:

[0037] 1. Compared with existing technologies, this application fundamentally breaks through the performance bottleneck of traditional gas sensors through an original cantilever beam double-sided heterogeneous MOFs modification structure and in-situ growth process: the in-situ growth process ensures a firm bond between the sensitive coating and the substrate, significantly overcoming the defects of easy coating peeling and poor uniformity in traditional coating methods, and providing a reliable hardware foundation for long-term online monitoring; while the double-sided heterogeneous modification strategy enables a single sensing unit to generate collaborative sensing and differentiated response to multiple fault gases, effectively solving the core problems of poor selectivity and serious cross-interference of traditional sensors.

[0038] 2. This application combines the CPO-1DCNN intelligent algorithm to dynamically correct and decouple the multidimensional response signal, and finally realizes highly sensitive and selective identification and accurate quantification of various dissolved fault gases in transformer oil under near-real operating conditions, providing an unprecedented reliable technical means for the diagnosis of early latent faults in transformers.

[0039] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0040] The accompanying drawings of this invention are described below.

[0041] Figure 1 This is a schematic diagram of the overall process for the detection method of dissolved gas in oil.

[0042] Figure 2 A schematic diagram of the material flow for constructing MOFs.

[0043] Figure 3 A schematic diagram of the signal processing process for the dissolved gas in oil detection device.

[0044] Figure 4 This is a schematic diagram of a cantilever beam gas sensor structure. Detailed Implementation

[0045] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0046] Example 1:

[0047] A dissolved gas detection device for oil, the device comprising:

[0048] Part 1, Detection Sensors

[0049] The detection sensor is a cantilever beam gas sensor core, such as Figure 4 As shown, this core structure is designed to achieve highly selective detection of multi-component gases through a single cantilever beam structure.

[0050] The cantilever beam is fabricated using MEMS technology, preferably with silicon or silicon nitride materials. Its typical dimensions are: length 100-500 μm, width 20-100 μm, and thickness 1-5 μm. One end of the cantilever beam is rigidly fixed to the base, while the other end is free, forming a resonant / deformation unit that is extremely sensitive to mass and surface stress. Its core innovation lies in transforming the traditional single-sensing surface into a bifacial sensing structure with heterogeneous functional regions.

[0051] This module forms the physicochemical basis for selective detection. Specifically: Front functional layer: On the front side of the cantilever beam, a macroporous MOF material, such as MIL-100(Fe), is constructed using in-situ growth technology. This material has a window pore size of approximately 2.5-2.9 nm and three-dimensional cage-like channels, enabling it to preferentially adsorb and enrich large molecular gases, such as ethylene (C2H4) and ethane (C2H6), from transformer fault gases through size sieving and affinity. The adsorption process mainly occurs in this functional layer. Back functional layer: On the back side of the cantilever beam, a microporous MOF material, such as ZIF-8, is modified using in-situ growth technology. Its pore size is approximately 0.34 nm, effectively repelling the aforementioned macromolecular gases while allowing small molecular gases, such as hydrogen (H2) and carbon monoxide (CO), to enter its channels and be adsorbed. This gives the back functional layer high selectivity for small molecules such as H2.

[0052] This application also provides a method for manufacturing a cantilever beam gas sensor core, such as... Figure 2 As shown, the details are as follows:

[0053] Cleaning and Drying: The MEMS chip with the bare cantilever beam is ultrasonically cleaned sequentially in acetone, ethanol, and deionized water to remove organic contaminants and particulate matter. It is then dried with high-purity nitrogen or in a critical point dryer to avoid damage to the beam structure due to surface tension. Surface Activation Unit: The cleaned chip is placed in an oxygen plasma treatment instrument and treated at a specific power (e.g., 100W) for 1-5 minutes. This step effectively removes residual organic matter and introduces abundant hydroxyl groups (-OH) onto the cantilever beam surface, greatly enhancing its hydrophilicity and chemical activity. Surface Functionalization Unit (Optional but Preferred): The activated chip is immersed in a toluene solution containing 1% (3-aminopropyl)triethoxysilane (APTES) and reacted at 60-80°C for 2-4 hours. This allows for the grafting of amino functional groups onto the surface, serving as "anchor points" for heterogeneous nucleation of MOF crystals, significantly improving the adhesion of subsequent MOF films.

[0054] MOS seed layer deposition: Provides gas-sensitive activity and nucleation sites for MOF growth. Deposition process unit: Atomic layer deposition (ALD) technology is used to deposit a 5-20 nm thick metal oxide (such as SnO2 or ZnO) film on the entire chip surface (including both sides of the cantilever beam). ALD technology ensures excellent conformability of the film in the three-dimensional structure and precise control of its thickness. This MOS layer is not only an ideal substrate for subsequent MOF growth, but also constitutes the underlying gas-sensitive functional body of the sensor.

[0055] Step-by-step protection and in-situ growth: This is the core step, enabling region-selective growth at the micrometer scale. Specifically, the process involves: front-side protection: using photolithography or a precision mechanical mask fixture, the reverse side of the cantilever beam is physically protected. If photolithography is used, the steps include: spin-coating photoresist, soft baking, UV exposure through a mask, and development, thus forming a protective film only on the reverse side of the cantilever beam. Front-side MOF growth unit: The protected chip is placed in a reactor containing a specific precursor solution (such as a DMF / H2O mixture of ferric chloride hexahydrate and trimesic acid). Under precisely controlled temperature (e.g., 100°C) and time (e.g., 12 hours), MOF crystals (e.g., MIL-100(Fe)) undergo heterogeneous nucleation and epitaxial growth only on the unprotected front side, forming a uniform thin film. Switching between protection and reverse-side MOF growth unit: After completing the front-side growth, the protective layer on the reverse side is first removed (e.g., by stripping the photoresist with acetone). Then, the reverse steps are performed to protect the already grown front side and expose the back side. Subsequently, the chip is placed in another precursor solution (such as a methanol solution of zinc nitrate hexahydrate and 2-methylimidazole) and a reverse MOF coating (such as ZIF-8) is grown under milder conditions (such as room temperature, 4 hours).

[0056] Post-processing and packaging: After growth, the chip is gently immersed and rinsed in a suitable solvent to remove loose crystals physically adsorbed on the surface. Subsequently, it undergoes heat treatment at 80-150°C under an inert atmosphere or vacuum to completely remove solvent molecules from the pores and activate adsorption sites. Performance aging unit: The fabricated sensor undergoes preliminary electrical aging treatment to stabilize its sensitivity characteristics and ensure consistent and reliable performance upon delivery.

[0057] Part 2, Signal Conversion Module

[0058] Gas adsorption events directly alter the physical properties of the cantilever beam, thus converting chemical signals into physical signals. Mass loading effect: Gas molecules are adsorbed into the double-sided MOF coating, increasing the equivalent mass of the cantilever beam. According to resonance theory, its resonance frequency (Δf) decreases accordingly. This effect is particularly significant for the adsorption of small-molecule gases (such as H2) on the reverse side; therefore, monitoring changes in resonance frequency is a primary method for detecting small-molecule gases. Surface stress effect: The interaction between gas molecules and the MOF framework causes minute expansion or contraction of the material lattice, generating differential stress on the cantilever beam surface. This stress forces the cantilever beam to undergo bending deformation (ΔZ). For the adsorption of larger molecules with strong interactions with MOFs (such as C2H4) on the front side, the surface stress effect is the dominant response mechanism.

[0059] The physical signal is converted into a measurable electrical signal and preliminarily analyzed. An integrated piezoresistive sensing unit (or an external optical displacement detection system) at the root of the cantilever beam monitors its frequency and deformation in real time. When the resonant frequency of the cantilever beam changes or it bends, the resistance value of the piezoresistive unit changes accordingly, which is then converted into a voltage signal output through an external Wheatstone bridge circuit. This composite electrical signal (containing Δf and ΔZ information) is then transmitted to the subsequent signal processing system. Because the positive and negative sensitive layers have different adsorption preferences for different gases and generate different signal weights, the coexistence information of multiple gases can be analyzed from the composite signal obtained from a single cantilever beam, laying the physical foundation for realizing multi-gas detection on a single beam.

[0060] Part 3, Diagnostic Module

[0061] This module acts as the brain of the diagnostic process, utilizing advanced algorithms to extract fault information from the raw signal. The data preprocessing unit standardizes, filters, and performs sliding window segmentation on the input raw digital signal, preparing it for feature extraction. The feature extraction unit extracts various time-domain and frequency-domain features from the preprocessed signal, such as, but not limited to, resonant frequency offset, static bending amplitude, signal energy, and the slope of the response curve, forming a feature vector characterizing the current gas response state.

[0062] The signal processing flow of Part 2 and Part 3 is as follows: Figure 3 As shown.

[0063] Example 2:

[0064] A method for detecting dissolved gases in oil, such as Figure 1 As shown, the specific steps are as follows:

[0065] S1. Place the cantilever beam in the insulating oil to be tested and leave it for a preset time.

[0066] S2. The front side of the cantilever beam adsorbs and enriches large molecular gases in the transformer insulating oil, while the back side of the cantilever beam adsorbs and enriches small molecular gases in the transformer insulating oil.

[0067] S3. Due to the mass loading effect of adsorbed molecules, the weight of the cantilever beam changes. The composite electrical signal is obtained by an integrated piezoresistive sensor or optical displacement detection device at the root of the cantilever beam. The composite electrical signal has the characteristics of changes in the cantilever beam resonant frequency Δf and bending deformation ΔZ.

[0068] Steps S1 to S3 are responsible for extracting the target gas from the complex oil sample and physically signaling it. Insulating oil is continuously or intermittently drawn from the transformer tank via a micro-pump valve system and transported to the oil-gas separation unit. The oil-gas separation unit employs a gas permeation chamber constructed from a membrane with selective permeability (such as a polytetrafluoroethylene membrane). Oil flows through one side of the membrane, where dissolved fault gases (H2, CO, CH4, C2H2, C2H4, etc.) diffuse through the membrane based on their concentration gradient and enter the clean gas chamber on the other side, while oil molecules are effectively blocked, achieving online, non-destructive gas extraction. The Janus sensing core unit guides the separated mixed gas to a sealed micro-gas chamber, directly contacting the cantilever beam sensor modified with double-sided heterogeneous MOFs as described in this invention. The sensor generates corresponding frequency change signals (Δf) and deformation signals (ΔZ) based on the different gas compositions.

[0069] S4. Analyze the composite electrical signal and output the dissolved gas content in the oil through a deep learning model.

[0070] Step S4 is responsible for converting the weak physical signal into a high-quality digital signal. Signal conditioning circuit unit: The piezoresistive element at the root of the cantilever beam is connected to a precision Wheatstone bridge. This circuit converts minute changes in piezoresistive resistance into a voltage signal. Subsequent low-noise amplifiers and filters amplify the signal and filter out power frequency and high-frequency noise, improving the signal-to-noise ratio. Data acquisition unit: The conditioned analog voltage signal is received by a high-precision data acquisition card and converted from analog to digital, generating a raw digital signal sequence that can be processed by a computer.

[0071] This embodiment employs the CPO-1DCNN intelligent diagnostic algorithm. A one-dimensional convolutional neural network (1DCNN) is designed to automatically learn the deep mapping relationship between the aforementioned feature vectors and the fault type and gas concentration. Its input layer receives the feature vectors, automatically extracts deeper features through multiple convolutional and pooling layers, and finally outputs the diagnostic results through a fully connected layer. To further optimize network performance, the Crowned Porcupine Optimizer (CPO) algorithm is introduced to automatically and globally optimize key hyperparameters of the 1DCNN (such as learning rate, number of convolutional kernels, and number of nodes in the fully connected layer), ensuring that the diagnostic model is always in an optimal performance state, thereby achieving higher recognition accuracy and generalization ability.

[0072] Finally, the output of the CPO-1DCNN algorithm is received to generate a structured diagnostic report, which includes at least: a) the quantitative concentration values ​​of each fault gas; b) the identification results of potential fault types inside the transformer (such as partial discharge, low-temperature overheating, high-temperature overheating, arc discharge, etc.). The early warning and output unit compares the diagnostic report with preset safety thresholds. Once an anomaly is detected or the warning level is reached, the system immediately sends an early fault warning signal to the remote monitoring center via a communication interface (such as 4G / 5G, LoRa, Ethernet), and can trigger light and sound alarms locally. Simultaneously, the system's health status assessment results can be displayed in real time on a local or remote human-machine interface. Through the above modular design, this system achieves a closed loop from physical perception to intelligent decision-making, perfectly achieving the invention's objective of providing early warning of latent transformer faults.

[0073] The CPO-1DCNN algorithm combines the CPO algorithm and the 1DCNN neural network. It utilizes the CPO algorithm to optimize the weights and thresholds of the 1DCNN neural network, thereby improving the model's prediction accuracy. The specific process is as follows:

[0074] S41. Input and segment the sample data, and normalize these data. Specifically, Bi2O3@Ni3(HHTP)2, CeO2@Ni3(HHTP)2, and NiO@Ni3(HHTP)2 sensors collected response test results to dissolved gases in different oils, and used these results as a data sample set. This paper uses the data collected by the gas sensors as input variables, and extracts five features from the response curves of the gas sensors to construct a feature space for characterization: the transient frequency value at the 6th second (S... 6s ), response time (T) res ), response extreme value (S) max Recovery time (T) rec ) and the sum of frequencies (S sumThe concentration of the introduced gas is then used as the output variable. Extracting meaningful features from the sensor response curve helps to construct a model with strong generalization ability and a high coefficient of determination (R²). 2 The regression model was constructed. To ensure the randomness of the data, the original samples were shuffled using the randperm function, and 80% of the data was selected as training samples, with the remaining 20% ​​used as test samples. Finally, the mapminmax function was used to normalize these sample data, scaling them to the range [0,1] to improve the training efficiency and prediction performance of the model.

[0075] S42. To construct a model for predicting dissolved gases in oil, a three-layer 1DCNN neural network with a single hidden layer was designed and its parameters were initialized. The output layer of the model has three nodes, corresponding to different gas concentrations to be predicted. The weights and thresholds of each layer are randomly initialized in the interval (-1,1) to increase the initial diversity of the model.

[0076] S43. The weights and threshold parameters of the 1DCNN neural network are optimized using the Crowned Porcupine optimization algorithm. The main steps of this optimization process are as follows:

[0077] S431. When invoking the CPO optimization algorithm, parameter initialization is performed first. The initial population contains 80 individuals, and the maximum number of generations for algorithm iteration is set to 100. For each individual in the population, the algorithm randomly initializes its position vector and records the score of its historical best position. Simultaneously, a convergence curve is initialized to track the performance changes of the algorithm as iterations progress. Furthermore, this paper sets upper and lower bounds for the search range of each independent variable involved in the algorithm to ensure that the optimization process is carried out within a predefined parameter space. These initializations provide the necessary starting point for the execution of the CPO algorithm, enabling it to effectively explore the solution space in subsequent iterations and gradually approach the optimal solution.

[0078] In S432 and 1DCNN neural networks, the evaluation function is a key metric used to measure network performance, and it is usually directly related to the network's prediction error. For 1DCNN neural networks optimized using CPO, the fitness value of an individual can be calculated using a specific formula.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A dissolved gas detection device for oil, characterized in that, The device includes a detection sensor, which includes a cantilever beam. A layer of macroporous MOFs material is constructed on the front side of the cantilever beam using in-situ growth technology, and a layer of microporous MOFs material is constructed on the back side of the cantilever beam using in-situ growth technology.

2. The dissolved gas detection device in oil as described in claim 1, characterized in that, The macroporous MOFs material is MIL-100(Fe), which has a window pore size of 2.5-2.9 nm and three-dimensional cage-like channels, preferentially adsorbing and enriching macromolecular gases in transformer insulating oil.

3. The dissolved gas detection device in oil as described in claim 2, characterized in that, The macromolecular gases include ethylene (C2H4) and ethane (C2H6).

4. The dissolved gas detection device in oil as described in claim 2, characterized in that, The microporous MOF material is ZIF-8, with a pore size of approximately 0.34 nm. It can repel large molecular gases while allowing small molecular gases to enter its pores and be adsorbed.

5. The dissolved gas detection device in oil as described in claim 4, characterized in that, The small molecule gases include hydrogen (H2) and carbon monoxide (CO).

6. The dissolved gas detection device in oil as described in claim 1, characterized in that, The cantilever beam is fabricated using MEMS technology and is made of silicon or silicon nitride material. The dimensions of the cantilever beam are: length 100-500μm, width 20-100μm, and thickness 1-5μm. One end of the cantilever beam is rigidly fixed to the base, while the other end is a free end, forming a resonant / deformation structure that is sensitive to mass and surface stress.

7. The dissolved gas detection device in oil as described in claim 1, characterized in that, A macroporous MOF material and a microporous MOF material were respectively constructed on the front and back of the cantilever beam using in-situ growth technology. The specific method is as follows: The MEMS chip with the bare cantilever beam was placed in acetone, ethanol and deionized water in sequence for ultrasonic cleaning to remove organic pollutants and particulate matter. Dry with high-purity nitrogen or in a critical point dryer to avoid damage to the beam structure due to surface tension; The cleaned chip is placed in an oxygen plasma treatment instrument and treated for 1-5 minutes at a specific power to remove residual organic matter and introduce hydroxyl groups (-OH) onto the cantilever beam surface to enhance hydrophilicity and chemical activity. The activated chip was immersed in a toluene solution containing 1% (3-aminopropyl)triethoxysilane (APTES) and reacted at 60-80°C for 2-4 hours. This process grafted amino functional groups onto the surface, which served as anchoring sites for heterogeneous nucleation of MOF crystals, thereby improving the adhesion of the MOF film. Photolithography or a precision mechanical mask fixture is used to physically cover and protect the reverse side of the cantilever beam. The protected chip was placed in a DMF / H2O mixed solution of ferric chloride hexahydrate and trimesic acid; By maintaining a temperature of 100 degrees Celsius for 12 hours, MIL-100(Fe) undergoes heterogeneous nucleation and epitaxial growth only on the unprotected front side, forming a uniform thin film. After completing the front-side growth, remove the protective layer on the back side; Protect the already grown front side and expose the back side; The chip was placed in a methanol solution of zinc nitrate hexahydrate and 2-methylimidazole. The reverse ZIF-8 coating was grown after being kept at room temperature for 4 hours. After growth is complete, the chip is gently soaked and rinsed in the appropriate solvent to remove loose crystals physically adsorbed on the surface. Heat treatment at 80-150°C under an inert atmosphere or vacuum to completely remove solvent molecules from the pores and activate adsorption sites. The prepared sensor undergoes preliminary electrical aging treatment to stabilize its sensitivity characteristics.

8. The dissolved gas detection device in oil as described in claim 1, characterized in that, The device also includes a signal conversion module and a diagnostic module; The signal conversion module states that after the front and back sides of the cantilever beam adsorb different gas molecules, the equivalent mass of the cantilever beam increases, the resonant frequency decreases, and the cantilever beam undergoes bending deformation. A composite electrical signal containing the resonant frequency Δf and the bending deformation ΔZ is obtained through an integrated piezoresistive sensor or optical displacement detection device at the root of the cantilever beam. The diagnostic module analyzes the composite electrical signal and uses a deep learning model to obtain the types and contents of dissolved gases in the oil.

9. A method for detecting dissolved gas in oil, characterized in that, The specific method is as follows: Place the cantilever beam in the insulating oil to be tested and leave it for a preset time; The front side of the cantilever beam adsorbs and enriches large molecular gases in the transformer insulating oil, while the back side of the cantilever beam adsorbs and enriches small molecular gases in the transformer insulating oil. Due to the mass loading effect of adsorbed molecules, the weight of the cantilever beam changes. The composite electrical signal is obtained by an integrated piezoresistive sensor or optical displacement detection device at the root of the cantilever beam. The composite electrical signal has the characteristics of changes in the cantilever beam's resonant frequency Δf and bending deformation ΔZ. The composite electrical signal is analyzed, and the dissolved gas content in the oil is output through a deep learning model.

10. The method for detecting dissolved gas in oil as described in claim 9, characterized in that, The deep learning model is based on the CPO-1DCNN convolutional neural network model.

Images