Method for detecting influence of temperature and mechanical deformation accumulation on anti-short-circuit performance of converter transformer

Abstract

The invention provides a method for detecting the influence of temperature and mechanical deformation accumulation on the anti-short circuit performance of a converter transformer, belongs to the technical field of converter transformer anti-short circuit, and deeply studies the mechanical response characteristics of the converter transformer under different temperature and dynamic force load conditions through precise experiments and data analysis. Firstly, material characteristic parameters are obtained, temperature and mechanical load tests are carried out on a dynamic mechanical tension press, and temperature field distribution, mechanical response, residual deformation and insulation distance change of a converter transformer component are recorded in detail by using advanced equipment such as a three-dimensional laser scanner, a temperature sensor, a displacement sensor and a high-speed camera system. Through finite element analysis and data processing, a dynamic characteristic database under the accumulated action of temperature and mechanical deformation is established, and a correlation model is trained by applying a machine learning technology, so that the problem that a detection method capable of comprehensively evaluating temperature-mechanical deformation-electrical performance interaction is lacked in the prior art is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of converter transformer anti-short-circuit performance, and in particular, relates to a method for detecting the influence of temperature and mechanical deformation accumulation on the anti-short-circuit performance of a converter transformer. Background Art

[0002] As a key equipment in the power system, the operating reliability of the converter transformer is directly related to the safety and stability of the entire power grid. In actual service, the converter transformer will be subject to complex environmental factors and mechanical stress, such as high temperature environment, short-circuit impact load, etc. These factors will cause a series of performance deterioration problems such as uneven temperature stress distribution, accumulated mechanical deformation, and insulation performance degradation of the components, which will seriously affect the service life and operational safety of the equipment.

[0003] At present, the industry usually adopts measures such as regular inspection and preventive maintenance to ensure the reliability of converter transformers. Specifically, they include: using infrared thermal imaging technology to monitor the temperature field distribution of equipment; using vibration analysis methods to diagnose the mechanical state of components; measuring insulation resistance and leakage current to evaluate insulation performance, etc. However, these methods can often only obtain the current static performance data of the equipment, and cannot fully reflect the cumulative impact of temperature and mechanical deformation on the comprehensive performance of the equipment. At the same time, these detection methods usually require shutdown inspections, and cannot achieve real-time monitoring of the operating status of the converter transformer.

[0004] In addition, some scholars have conducted special research on the temperature effect, short-circuit mechanical characteristics, insulation performance, etc. of converter transformers. For example, some scholars have proposed a thermal-mechanical coupling model based on finite element analysis to calculate the stress distribution under the temperature field; some scholars have used experimental testing methods to study the dynamic response of transformer components under short-circuit impact loads. However, most of these studies focus on a single physical field and lack comprehensive consideration and quantitative analysis of the mutual coupling effects of temperature, mechanical deformation, and electrical performance.

[0005] Therefore, there is an urgent need to establish a detection method that can comprehensively evaluate the interaction between temperature, mechanical deformation and electrical performance, so as to provide a reliable basis for condition monitoring and life management of converter transformers. Summary of the invention

[0006] In view of this, the present invention provides a method for detecting the influence of the accumulation of temperature and mechanical deformation on the short-circuit resistance of a converter transformer, which can solve the problem that the prior art lacks a detection method that can comprehensively evaluate the interaction between temperature, mechanical deformation and electrical performance.

[0007] The present invention is achieved in that: The present invention provides a method for detecting the influence of temperature and mechanical deformation accumulation on the short-circuit resistance performance of a converter transformer, comprising the following steps: S01. Obtaining material characteristic parameters of a converter transformer, wherein the material characteristic parameters include thermal expansion coefficient of a conductor material, thermal expansion coefficient of an insulating material, elastic modulus parameter of a conductor material, elastic modulus parameter of an insulating material, fracture toughness parameter of a conductor material, fracture toughness parameter of an insulating material, yield strength parameter of a conductor material, and yield strength parameter of an insulating material; S02, installing the converter transformer component to be tested on a dynamic mechanical tensile press, and heating the converter transformer component to be tested to a preset temperature through a temperature control system; S03, using a three-dimensional laser scanner to obtain the initial state parameters of the converter transformer component to be tested at the preset temperature, wherein the initial state parameters include winding size parameters, pad force parameters, support bar force parameters, and insulation layer parameters; S04, using a temperature sensor array to measure the temperature field distribution data of the converter transformer component to be tested; S05, using finite element analysis software to calculate the initial electric field distribution parameters of the converter transformer component to be tested according to the temperature field distribution data and the material characteristic parameters; S06. According to preset short-circuit impact load parameters, using the dynamic mechanical tensile press to apply a dynamic force load to the converter transformer component to be tested; S07, using a displacement sensor array to collect mechanical response data of the converter transformer component under the dynamic force load, wherein the mechanical response data includes deformation of the winding, deformation of the pad, deformation of the stay, and deformation of the insulation layer; S08, using a high-speed camera system to record the residual deformation data of the converter transformer component to be tested after the dynamic force load ends, wherein the residual deformation data includes the residual deformation of the winding, the residual deformation of the cushion block, the residual deformation of the stay, and the residual deformation of the insulation layer; S09, using an insulation distance measuring instrument to measure the insulation distance change of the component to be tested of the converter transformer; S10, adjusting the preset temperature, repeating steps S03 to S09, and obtaining mechanical response data, residual deformation data and insulation distance variation of the converter transformer component to be tested at different temperatures; S11, establishing a converter transformer dynamic characteristic database under the cumulative effect of temperature and mechanical deformation according to the mechanical response data, the residual deformation data and the insulation distance change; S12, analyzing the variation law of mechanical properties of conductors and winding components based on the converter transformer dynamic characteristics database; S13, establishing a correlation model of the mechanical and electrical properties of the converter transformer under the cumulative effect of temperature and mechanical deformation according to the mechanical property change law; S14. Based on the converter transformer dynamic characteristic database, the data is divided into a training set, a validation set and a test set in a ratio of 8:1:1, the physical model parameters and the neural network compensation terms of each equation in the association model are trained, and the trained association model is output and saved.

[0008] On the basis of the above technical solution, the detection method of the influence of temperature and mechanical deformation accumulation on the short-circuit resistance performance of the converter transformer of the present invention can also be improved as follows: The correlation model includes a temperature stress equation, a deformation accumulation equation, a mechanical strength equation, and an electrical performance equation.

[0009] Furthermore, the temperature stress equation is used to calculate the thermal stress distribution of converter transformer components at different temperatures, and the input includes the temperature field distribution data, the thermal expansion coefficient of the conductor material, the thermal expansion coefficient of the insulating material, the elastic modulus parameter of the conductor material, and the elastic modulus parameter of the insulating material, and the output is the component stress distribution data; the temperature stress equation includes a neural network compensation term based on a three-layer perceptron, which is used to compensate for the nonlinear temperature stress effect.

[0010] Furthermore, the deformation accumulation equation is used to calculate the cumulative deformation of the component under short-circuit impact, and the input includes the stress distribution data of the component, the preset short-circuit impact load parameter, the yield strength parameter of the wire material, and the yield strength parameter of the insulating material, and the output is the cumulative deformation of the component; the deformation accumulation equation includes a neural network compensation term based on a three-layer perceptron, which is used to compensate for the nonlinear deformation accumulation effect.

[0011] Furthermore, the mechanical strength equation is used to evaluate the strength margin of the component, and the input includes the cumulative deformation of the component, the fracture toughness parameter of the conductor material, and the fracture toughness parameter of the insulating material, and the output is the residual strength value of the component; the mechanical strength equation includes a neural network compensation term based on a three-layer perceptron, which is used to compensate for the nonlinear strength evolution effect.

[0012] Furthermore, the electrical performance equation is used to calculate the influence of mechanical deformation on electrical characteristics, and the input includes the residual strength value of the component, the change in the insulation distance, and the electric field distribution parameter, and the output is the insulation margin of the converter transformer; the electrical performance equation includes a neural network compensation term based on a three-layer perceptron, which is used to compensate for the nonlinear insulation characteristic effects.

[0013] Furthermore, the step of training the physical model parameters and neural network compensation terms of each equation in the association model includes: Train the temperature stress equation, use the least squares method to fit the physical model parameters, and use the back propagation algorithm to train the neural network compensation term parameters; Train the deformation accumulation equation, use the least squares method to fit the physical model parameters, and use the back propagation algorithm to train the neural network compensation term parameters; Train the mechanical strength equation, use the least squares method to fit the physical model parameters, and use the back propagation algorithm to train the neural network compensation term parameters; The electrical performance equations are trained, the physical model parameters are fitted using the least squares method, and the neural network compensation term parameters are trained using the back propagation algorithm.

[0014] Furthermore, the physical model parameters refer to the general term for material mechanical property parameters, geometric dimension parameters and boundary condition parameters.

[0015] Furthermore, for each neural network compensation term, a lightweight three-layer perceptron structure is adopted, which is built based on the GhostNet model, and an adaptive feature fusion module is added between the first Ghost module and the second Ghost module of the GhostNet model. The adaptive feature fusion module includes a channel attention submodule, a spatial attention submodule and a discriminant equation.

[0016] The core of the GhostNet network structure is the Ghost Module. The basic idea is to generate more feature maps from a small number of original feature maps through linear transformation: first, ordinary convolution is used to generate a small number of original feature maps (Primary Feature Maps), and then a series of linear transformation operations (such as 1×1 convolution, 3×3 depth convolution, simple linear transformation, etc.) are applied to these feature maps to generate Ghost feature maps. These Ghost feature maps are concatenated with the original feature maps to form the final output; the entire network starts with a standard 2D convolution layer, followed by multiple Ghost bottlenecks (Ghost The network has a main structure composed of a Ghost Bottleneck layer, each Ghost bottleneck layer contains two Ghost modules, and shortcut connections are added at appropriate locations, and finally ends with a global average pooling and a fully connected layer; the linear transformation introduced by the network in the Ghost module can greatly reduce the number of parameters and computational complexity, because it does not need to learn convolution kernels separately for each feature map, but instead derives new feature maps from the original feature maps through cheap linear operations; this design enables GhostNet to significantly reduce the model size and computational overhead while maintaining good performance, making it particularly suitable for use in resource-constrained scenarios.

[0017] Constructing neural network compensation items based on the GhostNet model has the following advantages and effects: GhostNet's Ghost module generates more feature maps from a small number of original feature maps through linear transformation, which can effectively reduce the number of parameters and computational complexity, which is very important for scenarios such as converter transformers that require real-time monitoring and rapid response. In this scheme, the four equations of temperature stress, deformation accumulation, mechanical strength and electrical performance all contain nonlinear effects. Traditional neural networks may require a deeper network structure to fit these nonlinear relationships well, while GhostNet can achieve similar fitting effects with fewer parameters through its unique feature reuse mechanism. In addition, the added adaptive feature fusion module can dynamically adjust the importance of features according to material characteristic parameters, which enables the model to better adapt to transformer performance prediction under different materials and working conditions. Combined with the evaluation mechanism of the discriminant equation, the entire compensation system not only maintains a low computational overhead, but also accurately captures various nonlinear physical effects, thereby improving the accuracy and practicality of converter transformer short-circuit performance detection.

[0018] Furthermore, the discriminant equation is used to evaluate the matching degree between the material characteristic parameters and the attention weights, the input includes the material characteristic parameters, the channel attention weights and the spatial attention weights, and the output is the effectiveness score of the feature fusion.

[0019] The models and equations involved in the present invention are described in detail below: 1. Temperature field distribution calculation model: The calculation of the temperature field distribution data is specifically expressed as follows: ; In the formula, For space point In time Temperature value at the moment; is the initial temperature; For the The temperature weight coefficient of each temperature measuring point; is the temperature attenuation coefficient; For the The spatial coordinates of the temperature measurement points; is the time influence coefficient; is the number of temperature sensors.

[0020] Parameter acquisition method: It is directly measured by a temperature sensor and the unit is Kelvin (K); Obtained through least squares fitting, ranging from 0 to 1; Obtained through experimental calibration, the range is 0.001~0.1; Obtained through time series data fitting, the range is -0.1~0.1K / s.

[0021] 2. Temperature stress equation: The temperature stress equation is specifically expressed as follows: ; In the formula, is the thermal stress tensor; is the elastic modulus tensor; is the thermal expansion coefficient tensor; is the temperature change; is the mechanical strain tensor; is the neural network compensation term.

[0022] Neural network compensation term The calculation formula is: ; In the formula, is the weight matrix; is the bias vector; is the activation function; is the temperature field vector; is the thermal expansion coefficient vector; is the elastic modulus vector.

[0023] 3. Deformation accumulation equation: The deformation accumulation equation is specifically expressed as follows: ; In the formula, is the total deformation tensor; is the elastic deformation tensor; is the plastic deformation tensor; is the creep deformation tensor; is the neural network compensation term.

[0024] The calculation formulas for each component are: ; ,when hour; ; In the formula, is the stress tensor; is the elastic modulus; is the plasticity coefficient; is the yield strength; is the work hardening index; is the material constant; is the stress index; For time; is the time index; is the activation energy; is the gas constant; is the absolute temperature.

[0025] 4. Mechanical strength equation: The mechanical strength equation is specifically expressed as follows: ; ; In the formula, is the residual strength; is the initial strength; is the degree of damage; is the neural network compensation term; For the The strain amplitude of the sub-cycle; is the fracture strain; is a material related constant.

[0026] 5. Electrical performance equation: The electrical performance equation is specifically expressed as follows: ; In the formula, is the insulation margin; is the initial insulation margin; is the attenuation coefficient; is the insulation distance change; is the electric field strength vector; is the neural network compensation term.

[0027] 6. Discriminant equation: The discriminant equation is specifically expressed as follows: ; In the formula, Score the effectiveness of feature fusion; is the weight coefficient; Score the channel attention; Score spatial attention; is the material property mapping function; is the material characteristic parameter vector.

[0028] Principle description: 1. The temperature field distribution model adopts the superposition form of Gaussian kernel function, taking into account the continuous distribution characteristics of temperature in space and the time evolution characteristics; 2. The temperature stress equation is based on the thermoelasticity theory, and a neural network term is introduced to compensate for the nonlinear effect; 3. The deformation accumulation equation comprehensively considers the three deformation mechanisms of elasticity, plasticity and creep, and is suitable for high temperature and long-term service conditions; 4. The mechanical strength equation is based on the theory of continuous damage mechanics and introduces the concept of cumulative damage; 5. The electrical performance equation uses an exponential decay model to describe the insulation performance degradation process; 6. The discriminant equation uses a weighted fusion method to evaluate the feature extraction effect.

[0029] The derivation process of each equation is described in detail below: 1. Derivation process of temperature field distribution model: Step 1: Based on Fourier's law of heat conduction, establish the basic heat conduction equation: ; In the formula, is the temperature diffusion coefficient, in m² / s, which can be calculated from the thermal conductivity, density and specific heat capacity of the material.

[0030] Step 2: Consider the influence of heat source terms and expand the heat conduction equation: ; In the formula, is the volume heat source density, in W / m³; is the material density in kg / m³; is the specific heat capacity, and its unit is J / (kg·K).

[0031] Step 3: Solve the partial differential equation using the Green function method to obtain the temperature field expression: .

[0032] 2. Derivation process of temperature stress equation: Step 1: Establish stress-strain relationship based on thermoelastic theory: ; In the formula, is the total strain; is the elastic strain; For thermal strain.

[0033] Step 2: Substitute Hooke's law and thermal strain expressions: ; In the formula, is the elastic modulus; is the coefficient of thermal expansion; is the temperature change.

[0034] Step 3: Consider the influence of mechanical deformation and introduce mechanical strain term: .

[0035] Step 4: Introduce neural network compensation term to deal with nonlinear effects: .

[0036] 3. Derivation process of deformation accumulation equation: Step 1: Based on the principle of strain decomposition, list the total deformation expression: .

[0037] Step 2: Establish the elastic deformation expression based on Hooke's law: .

[0038] Step 3: Establish the plastic deformation expression based on the Ramberg-Osgood model: ,when hour.

[0039] Step 4: Establish creep deformation expression based on Norton creep model: .

[0040] Step 5: Introduce neural network compensation term: .

[0041] 4. Derivation process of mechanical strength equation: Step 1: Based on the theory of continuum damage mechanics, establish the damage evolution equation: ; In the formula, is the number of cycles; is the damage variable; is the strain amplitude; is the fracture strain; is the material constant.

[0042] Step 2: Integrate to obtain the cumulative damage expression: .

[0043] Step 3: Establish the relationship between strength and damage: ; In the formula, is the residual strength; is the initial strength.

[0044] Step 4: Introduce neural network compensation term: .

[0045] 5. Derivation process of electrical performance equation: Step 1: Based on electric field theory, establish the relationship between electric field strength and insulation distance: ; In the formula, is the voltage; is the insulation distance.

[0046] Step 2: Consider the impact of insulation distance changes: ; In the formula, is the insulation distance change.

[0047] Step 3: Establish insulation margin degradation model: ; In the formula, is the insulation margin; is the initial insulation margin; is the attenuation coefficient.

[0048] Step 4: Introduce neural network compensation term: .

[0049] 6. Derivation process of discriminant equation: Step 1: Calculate channel attention score: ; In the formula, is the channel weight vector; is the channel feature vector; is the number of channels.

[0050] Step 2: Calculate the spatial attention score: ; In the formula, is the spatial weight vector; is the spatial eigenvector; is the number of spatial positions.

[0051] Step 3: Construct material property mapping function: ; In the formula, is the weight matrix; is the bias vector; is the material characteristic parameter vector.

[0052] Step 4: Weighted fusion to obtain the final discriminant equation: .

[0053] Supplementary instructions for parameter acquisition: Obtained through material manual query or experimental measurement; Obtained by measuring with a heat flux density sensor; Obtained through material manual query; Obtained through tensile testing; It is obtained by measuring with a thermal expansion instrument; the neural network parameters are obtained by training with a BP algorithm; Obtained through uniaxial tensile test; Obtained through high temperature creep test; is the gas constant, which is taken as 8.314 J / (mol·K); Obtained through tensile testing; Obtained through fatigue testing; Obtained by non-destructive testing methods; Obtained through insulation breakdown test; Obtained through dielectric strength test; It is obtained through cross-validation optimization; the neural network parameters are obtained through BP algorithm training.

[0054] Specifically, the steps to obtain the parameters required for the experiment are as follows: 1. Elastic modulus The acquisition steps are as follows: sample preparation, prepare the conductor material and insulation material of the converter transformer into tensile specimens that meet the GB / T228.1 standard; install the specimen on a universal material testing machine and set the tensile rate to 2 mm / min; apply a preload, eliminate the specimen clamping gap, and set the preload to 20 N; start the test and record the stress-strain curve; select 5 points in the linear section of the stress-strain curve and calculate the slope by the least squares method, which is the elastic modulus .

[0055] 2. Thermal expansion coefficient The acquisition steps are as follows: process the material sample into a cylinder with a length of 25±0.5mm; place the sample in the test chamber of the thermal expansion instrument; heat the temperature from room temperature to 200℃ at a rate of 5℃ / min; record the length change of the sample every 10℃; and calculate the length of the sample according to the formula Calculate the coefficient of thermal expansion; where, is the initial length of the sample, in mm; is the length change, in mm; is the temperature change in K.

[0056] 3. Plastic parameters The acquisition steps are as follows: prepare tensile specimens that meet the GB / T228.1 standard; perform uniaxial tensile tests on a universal material testing machine; record the complete stress-strain curve until the specimen breaks; and calculate the stress-strain curve according to the formula , perform nonlinear fitting on the data after the yield point; use the least squares method to determine the parameters and The value of .

[0057] 4. Creep parameters The acquisition steps are as follows: prepare creep specimens that meet the GB / T2039 standard; install the specimens on a high-temperature creep testing machine; conduct creep tests at different temperatures (such as 150°C, 175°C, and 200°C) and different stress levels; record the curve of creep strain changing with time; and calculate the creep strain according to the Norton creep formula. Perform multivariate nonlinear fitting on the data at different temperatures and stress levels to obtain the parameters The value of .

[0058] 5. Fracture strain and injury index The acquisition steps are as follows: prepare standard fatigue specimens; perform strain-controlled fatigue tests on a fatigue testing machine; set different strain amplitude levels, such as 0.2%, 0.4%, and 0.6%; record the fatigue life at each strain amplitude level; fit according to the Manson-Coffin formula to obtain the fracture strain ; According to the damage evolution equation Fitting value.

[0059] 6. Insulation performance parameters The steps of obtaining: prepare standard insulation breakdown test samples; install the samples on the insulation breakdown test device; gradually increase the voltage until breakdown, and record the breakdown voltage; repeat step 3 at different temperatures to obtain the relationship between temperature and breakdown voltage; according to the formula Fitting to obtain parameters .

[0060] 7. Temperature distribution related parameters Acquisition steps: Install temperature sensor array; Arrange heat flux density sensors at different locations; Collect temperature field distribution data and heat flux density data; Calculate temperature diffusion coefficient according to Fourier heat conduction equation ; Directly measure the volume heat source density through the heat flux density sensor .

[0061] 8. Discriminant equation weight coefficient The acquisition steps are as follows: prepare a training data set containing feature data under different working conditions; initialize the weight coefficients to equal values; use the cross-validation method to evaluate the model performance; use the gradient descent method to optimize the weight coefficients; repeat steps 3 and 4 until convergence.

[0062] Note: All experiments must be carried out under standard environmental conditions (temperature 23±2℃, relative humidity 50±10%); each set of experiments should be repeated at least 3 times to ensure data reliability.

[0063] Compared with the prior art, the beneficial effect of the detection method for the influence of temperature and mechanical deformation accumulation on the short-circuit resistance performance of converter transformers provided by the present invention is that by systematically measuring the mechanical response, residual deformation, insulation distance change and other indicators of converter transformer components under different temperatures and short-circuit impact loads, a comprehensive correlation model covering temperature stress, deformation accumulation, mechanical strength and electrical performance is established. Compared with the prior art, the scheme of the present invention has the following innovations and advantages: 1. The comprehensive detection and analysis of the mutual coupling of temperature, mechanical deformation and electrical performance is realized, which makes up for the shortcomings of existing research in a single physical field. The established correlation model can fully reflect the performance degradation process of key components of converter transformers under actual service conditions, providing a reliable basis for equipment status monitoring and life management.

[0064] 2. By using advanced methods such as dynamic mechanical tests, three-dimensional scanning, and high-speed photography, we obtained comprehensive performance data of the converter transformer under temperature and short-circuit impact, providing reliable basic data support for the establishment of the correlation model.

[0065] 3. A neural network-based compensation term is introduced into the correlation model, which can effectively compensate for the nonlinear temperature stress effect, deformation accumulation effect, strength evolution effect and insulation characteristic effect, thereby improving the prediction accuracy of the model.

[0066] 4. The adaptive feature fusion mechanism enables the model to dynamically adjust the importance of features according to material properties, thereby improving its ability to adapt to different working conditions and enhancing its practicality.

[0067] In summary, the present invention solves the problem that the prior art lacks a detection method that can comprehensively evaluate the interaction between temperature-mechanical deformation-electrical performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0070] like Figure 1As shown, it is a flow chart of a method for detecting the influence of temperature and mechanical deformation accumulation on the short-circuit resistance of converter transformers provided by the present invention. The specific implementation methods of the steps of the present invention are described in detail below: The specific implementation method of step S01 is to obtain various physical performance parameters of converter transformer materials. First, it is necessary to measure parameters such as the thermal expansion coefficient, elastic modulus, fracture toughness and yield strength of the conductor material. At the same time, it is also necessary to measure parameters such as the thermal expansion coefficient, elastic modulus, fracture toughness and yield strength of the insulating material. These parameters can be obtained through experimental testing or by referring to the material performance manual. Mastering these material characteristic parameters is crucial for subsequent temperature stress analysis, deformation accumulation calculation and mechanical strength evaluation.

[0071] The specific implementation method of step S02 is to install the converter transformer component to be tested on the dynamic mechanical testing machine and heat it to a preset temperature through a temperature control system. First, fix the converter transformer component to be tested on the fixture of the dynamic tensile testing machine to ensure that its position and posture are stable. Then use the temperature control system to heat the component, and the heating temperature can be set between 60°C and 120°C, which is appropriately selected according to the actual service conditions. The temperature control system can adopt electric heating, hot air flow heating or infrared radiation heating, and be equipped with a temperature sensor for real-time monitoring and feedback control to ensure that the component temperature is stable near the preset value. In this way, the working state of the converter transformer in a high temperature environment can be simulated.

[0072] The specific implementation method of step S03 is to use a three-dimensional laser scanner to obtain the initial geometric parameters of the converter transformer components at a preset temperature. First, place the component to be tested, which is heated to a preset temperature, in the scanning area of ​​the three-dimensional laser scanner to ensure that the surface of the component is not blocked. Then start the scanner for a comprehensive scan to obtain three-dimensional point cloud data on the surface of the component. Use professional three-dimensional measurement software to process and analyze the point cloud data to extract key geometric parameters such as winding size, pad stress condition, support bar stress state, insulation layer thickness, etc. These initial geometric parameters provide basic data for subsequent temperature stress analysis, deformation accumulation calculation, and insulation performance evaluation. The measurement accuracy of a three-dimensional laser scanner can usually reach about 0.1 mm, which can meet the requirements of this method.

[0073] The specific implementation method of step S04 is to use a temperature sensor array to measure the temperature field distribution of the converter transformer components. First, multiple temperature sensors, such as thermocouples or thermal resistors, are arranged on the surface of the component or at key positions inside to form a temperature sensor array. These sensors should be able to cover the main areas of the component, such as windings, pads, struts and insulation layers. Then, the temperature readings of each temperature sensor are recorded in real time through a data acquisition device, and organized into temperature field distribution data. Normally, the measurement accuracy of the temperature sensor can reach 0.1°C, and the number of sensors in the array can be determined according to the complexity of the component, generally around 10-20. These temperature field distribution data provide the required input for subsequent temperature stress calculations.

[0074] The specific implementation method of step S05 is to use finite element analysis software to calculate the initial electric field distribution of the converter transformer components according to the temperature field distribution data and material parameters. First, it is necessary to establish a three-dimensional finite element model of the converter transformer components, and set the correct geometric dimensions, material properties and boundary conditions in the model. Then the temperature field distribution data obtained in step S04 is imported into the finite element model as a thermal load condition. Using the thermal-electric coupling solver in the finite element analysis software, the initial electric field distribution of the component at a preset temperature is calculated based on the relative dielectric constant, conductivity and other parameters of the material. This electric field distribution information is crucial for the subsequent evaluation of the impact of mechanical deformation on insulation performance. Finite element analysis software can usually give numerical results of electric field intensity distribution with an accuracy of about 10V / mm.

[0075] The specific implementation method of step S06 is to use a dynamic mechanical testing machine to apply a preset short-circuit impact load to the converter transformer components. First, it is necessary to determine the typical impact load spectrum under a short-circuit accident, including parameters such as peak force and load duration. These parameters can be determined by relevant standards or engineering experience. For example, for a 500kV converter transformer, the peak force of the short-circuit impact load can be set at around 50kN, and the duration is 50-100ms. Then the component to be tested, which is heated to a preset temperature in step S02, is installed on a dynamic testing machine, and dynamically loaded according to the preset short-circuit impact load spectrum. The testing machine should be able to provide sufficient loading capacity and response speed to simulate a real short-circuit impact process. In this way, the mechanical response data of the converter transformer components under dynamic force load can be obtained.

[0076] The specific implementation method of step S07 is to use a displacement sensor array to measure the mechanical response of the converter transformer components under dynamic loads. First, multiple displacement sensors are arranged at key positions of the components such as windings, pads, stays and insulation layers to form a complete measurement array. These sensors should be able to accurately capture the deformation response of the components under dynamic loads. Typical sensors include resistive displacement sensors, optical displacement sensors or laser displacement sensors. Then, during the process of applying dynamic loads in step S06, the displacement response data of each position is recorded in real time through the data acquisition system. These data include the deformation of the windings, the deformation of the pads, the deformation of the stays and the deformation of the insulation layer. The measurement accuracy of the sensor can usually reach about 0.01 mm, which can meet the requirements of this method. These mechanical response data provide a basis for subsequent deformation accumulation calculations and mechanical strength evaluations.

[0077] The specific implementation method of step S08 is to use a high-speed camera system to record the residual deformation of the converter transformer components after the dynamic load is completed. First, place the component to be tested that has been loaded in step S06 within the shooting range of the high-speed camera system to ensure that the surface of the component is clearly visible. Then start the high-speed camera system to continuously record the deformation process of the component after the dynamic load is completed at a rate of at least 1000 frames per second. By analyzing these high-speed video images, the residual deformation data of key parts such as windings, pads, struts and insulating layers can be extracted. The measurement accuracy of the high-speed camera system can usually reach 0.05mm, which can meet the measurement requirements of the residual deformation of this method. These residual deformation data are of great significance for evaluating the mechanical strength margin and insulation performance of the components.

[0078] The specific implementation method of step S09 is to use an insulation distance measuring instrument to measure the change of the insulation distance of the converter transformer components. First, an insulation distance measuring instrument, such as a contact or non-contact capacitive or optical distance measuring sensor, is installed at the key insulation gap position of the component. These sensors should be able to accurately measure the change of the insulation distance. Then, during the process of applying the dynamic force load in step S06 and after the loading is completed, the distance change of each insulation gap is recorded in real time through the data acquisition system. The measurement accuracy can reach 0.01mm. These insulation distance change data are the key basis for evaluating the impact of mechanical deformation on insulation performance.

[0079] The specific implementation of step S10 is to adjust the preset temperature and repeat steps S03 to S09 to obtain the mechanical response data, residual deformation data and insulation distance change of the converter transformer components at different temperatures. First, the range and step size of the temperature adjustment need to be determined. Four typical temperature points, such as 60°C, 80°C, 100°C and 120°C, can be selected for testing. For each temperature point, the operation process of steps S03 to S09 is repeated, including three-dimensional scanning to obtain initial geometric parameters, temperature field distribution measurement, electric field distribution calculation, dynamic force application, mechanical response measurement, residual deformation recording and insulation distance measurement. In this way, comprehensive performance data of converter transformer components under different temperature conditions can be obtained, providing a basis for the subsequent establishment of a correlation model of temperature-mechanical deformation-electrical performance.

[0080] The specific implementation method of step S11 is to establish a converter transformer dynamic characteristic database based on the mechanical response data, residual deformation data and insulation distance variation obtained above. First, the various data obtained from the tests of steps S03 to S09 are sorted and archived to form a systematic database. The database should include winding size parameters, pad force parameters, support bar force parameters, insulation layer parameters, mechanical response data, residual deformation data and insulation distance variation under different temperature conditions. At the same time, the corresponding material characteristic parameters such as thermal expansion coefficient, elastic modulus, fracture toughness and yield strength are also required to be entered into the database. In this way, a comprehensive performance database covering multiple aspects such as temperature, mechanical deformation and electrical performance is established. This database provides basic data support for the subsequent establishment of correlation models and parameter training.

[0081] The specific implementation method of step S12 is to analyze the mechanical property change law of the conductor and winding components based on the converter transformer dynamic characteristic database. First, the mechanical response data and residual deformation data under different temperature and load conditions in the database are analyzed and compared as a whole. The typical performance change law of the conductor and winding components under the action of temperature and mechanical deformation can be extracted by statistical analysis, curve fitting and other methods. For example, the relationship between the plastic deformation degree of the conductor material and the temperature and stress changes, the evolution characteristics of the winding deformation with temperature and short-circuit impact force, etc. can be analyzed. At the same time, the influence of different material parameters such as thermal expansion coefficient, yield strength, etc. on the performance of the components can also be studied. Through this analysis, the influence mechanism of temperature and mechanical deformation on the mechanical properties of key components of the converter transformer can be deeply understood.

[0082] The specific implementation of step S13 is to establish a temperature-mechanical deformation-electrical performance correlation model based on the above-mentioned mechanical performance change law. This correlation model should include the following 4 sub-models: 1. Temperature stress equation: Based on the temperature field distribution data, material thermal expansion coefficient, elastic modulus and other parameters, the thermoelastic theory is used to calculate the internal stress distribution of the component at different temperatures. At the same time, the neural network term is introduced to compensate for the nonlinear temperature stress effect.

[0083] 2. Deformation accumulation equation: Taking into account the mechanisms of elastic deformation, plastic deformation and creep deformation, the cumulative deformation of the component under the short-circuit impact load is calculated based on parameters such as stress distribution and material yield strength. At the same time, a neural network term is introduced to compensate for the nonlinear deformation accumulation effect.

[0084] 3. Mechanical strength equation: Based on the theory of continuous damage mechanics, combined with parameters such as the fracture toughness of conductors and insulating materials, the residual mechanical strength of components under cumulative deformation is evaluated. At the same time, a neural network term is introduced to compensate for the nonlinear strength evolution effect.

[0085] 4. Electrical performance equation: Based on the residual mechanical strength of the component, the change in insulation distance, the electric field distribution and other parameters, the exponential decay model is used to calculate the degradation degree of the insulation performance. At the same time, a neural network term is introduced to compensate for the nonlinear insulation characteristic effect.

[0086] By establishing such a comprehensive correlation model, the influence of temperature and mechanical deformation on the mechanical strength and insulation performance of key components of converter transformers can be predicted more accurately, providing a theoretical basis for equipment status monitoring and life assessment.

[0087] The specific implementation method of step S14 is to train and optimize the established association model. First, the dynamic characteristic database established in step S11 is divided into a training set, a validation set and a test set in a ratio of 8:1:1. Then, for the above four sub-models, the following methods are used to train their physical model parameters and neural network compensation items: 1. Temperature stress equation: Use the least squares method to fit the physical model parameters such as thermal expansion coefficient, elastic modulus, etc.; use the back propagation algorithm to train the neural network compensation term parameters.

[0088] 2. Deformation accumulation equation: Use the least squares method to fit the physical model parameters such as yield strength, creep coefficient, etc.; use the back propagation algorithm to train the neural network compensation term parameters.

[0089] 3. Mechanical strength equation: Use the least squares method to fit physical model parameters such as fracture toughness; use the back propagation algorithm to train the neural network compensation term parameters.

[0090] 4. Electrical performance equation: Use the least squares method to fit the physical model parameters such as insulation distance attenuation coefficient, etc.; use the back propagation algorithm to train the neural network compensation item parameters.

[0091] When training the neural network compensation term, a lightweight three-layer perceptron structure is used, and an adaptive feature fusion module is added to it. This fusion module contains a channel attention submodule, a spatial attention submodule, and a discriminant equation, which is used to evaluate the matching degree between the material characteristic parameters and the attention weights, thereby dynamically adjusting the importance of the features. This design can improve the model's ability to capture nonlinear physical effects while maintaining a low computational overhead.

[0092] Specifically, the principle of the present invention is to establish a correlation model that can comprehensively consider the temperature effect, mechanical deformation effect and electrical performance coupling relationship. The model includes four main sub-models: temperature stress model, deformation accumulation model, mechanical strength model and electrical performance model.

[0093] Firstly, by measuring the temperature field distribution data and material characteristic parameters, the internal stress distribution of the component at different temperatures is calculated using the thermoelastic theory. In order to capture the nonlinear temperature stress effect, a compensation term based on a neural network is introduced on the basis of the physical model.

[0094] Secondly, combining the short-circuit impact load parameters and material strength characteristics, a mathematical model describing the cumulative deformation of components under dynamic forces was established. This model comprehensively considers the mechanisms of elastic deformation, plastic deformation and creep deformation, and also introduces neural network terms to compensate for nonlinear deformation effects.

[0095] Thirdly, based on the theory of continuum damage mechanics, the residual mechanical strength of the component under the cumulative deformation is evaluated. The neural network compensation term is used to fit the nonlinear strength evolution process.

[0096] Finally, the exponential decay model is used to calculate the degradation of insulation performance according to the residual mechanical strength, insulation distance change and electric field distribution of the component. At the same time, a neural network term is also introduced to capture the nonlinear insulation characteristic effect.

[0097] In the process of establishing the entire correlation model, full use is made of comprehensive performance data obtained by advanced means such as dynamic mechanical tests, 3D scanning, and high-speed photography. At the same time, in the design of the neural network compensation term, an adaptive feature fusion mechanism is introduced to enable the model to dynamically adjust the importance of features according to material properties, thereby improving its adaptability and generalization ability.

[0098] The following is an example of a specific application scenario of the present invention: A large power grid company is operating a 500kV converter transformer substation, which was put into operation in 2020 and has been in service for 5 years. In order to ensure the reliable operation of the equipment, the power grid company plans to use the temperature-mechanical deformation-electrical performance comprehensive detection method proposed in the present invention to conduct a comprehensive diagnosis of the operating status of the converter transformer.

[0099] First, the engineering team measured and determined the key material characteristic parameters of the converter transformer. The thermal expansion coefficient of the conductor material , elastic modulus , fracture toughness , yield strength . Thermal expansion coefficient of insulating material , elastic modulus , fracture toughness , yield strength These parameters will serve as basic input data for subsequent analysis and calculation.

[0100] After installing the converter transformer component to be tested on the dynamic mechanical testing machine, the engineering team heated it to 60°C through the temperature control system. The initial geometric parameters of the component at 60°C were obtained through a 3D laser scanner: winding length , winding diameter , winding thickness ;Pad bearing area , pad force ;Strut bearing area ,Strut force ;Insulation layer thickness .

[0101] Subsequently, the engineering team used a temperature sensor array to measure the temperature field distribution of the converter transformer components at 60°C. After data fitting, the temperature field distribution can be expressed as: ; in, Between 0.1 and 0.9, Between 0.01 and 0.05, The coordinate positions of the 15 temperature sensors are shown in Figure 2. The temperature field distribution data was imported into the finite element analysis software, and the initial electric field distribution of the component at 60°C was calculated. The maximum value is 8kV / mm.

[0102] According to the preset short-circuit impact load parameters, the engineering team used a dynamic mechanical testing machine to apply peak force to the component under test. ,Duration Through the displacement sensor array, they collected the mechanical response data of the component under the dynamic load in real time: Winding deformation: , , ; Pad deformation: , ; Deformation of stays: , ; Insulation layer deformation: ; Subsequently, the engineering team used a high-speed camera system to record the residual deformation of the component after the dynamic load ended: Winding residual deformation: , , ; Residual deformation of pad: , ; Residual deformation of the strut: , ; Residual deformation of insulation layer: ; The insulation distance measuring instrument also measures the change in insulation distance after short circuit impact. .

[0103] Based on these comprehensive performance test data, the engineering team established a database of the converter transformer dynamic characteristics, covering indicators such as initial geometric parameters, mechanical response data, residual deformation data, and insulation distance changes. At the same time, material characteristic parameters were also entered into the database.

[0104] Next, the engineering team began to establish a correlation model between temperature, mechanical deformation and electrical performance.

[0105] First, based on the theory of thermoelasticity, they established the temperature stress equation: ; In the formula, is the thermal stress tensor, is the elastic modulus tensor, is the thermal expansion coefficient tensor, is the temperature change, is the mechanical strain tensor. In order to capture the nonlinear temperature stress effect, This neural network compensation term is introduced into the model. After least squares fitting and back propagation algorithm training, the physical model parameters and neural network compensation term parameters of the temperature stress equation have been determined.

[0106] Secondly, taking into account elastic deformation, plastic deformation and creep deformation, the engineering team established the deformation accumulation equation: ; ; ; ; Also introduced Neural network terms are used to compensate for the nonlinear deformation accumulation effect. Through fitting and training, various physical parameters and neural network parameters have also been determined.

[0107] Again, based on the theory of continuum damage mechanics, the engineering team established the mechanical strength equation: ; ; in, is the residual strength, is the initial strength, is the degree of damage. Similarly, The neural network compensation term is used to fit the nonlinear intensity evolution process.

[0108] Finally, the electrical performance equation was established using the exponential decay model: ; here, is the insulation margin, is the insulation distance change, is the electric field strength. The neural network compensation term is used to capture the nonlinear insulation degradation effects.

[0109] When building the above four sub-models, the engineering team adopted an adaptive feature fusion mechanism, which includes channel attention, spatial attention, and a discriminant equation for evaluating feature importance. This enables the model to dynamically adjust the weight of features according to changes in material characteristic parameters, thereby improving its adaptability and generalization ability.

[0110] Finally, the engineering team divided the entire converter transformer dynamic characteristics database into training set, validation set and test set in a ratio of 8:1:1, and trained and optimized the above four sub-models. During the training process, the physical model parameters were fitted using the least squares method, and the neural network compensation terms were optimized using the back propagation algorithm. After repeated iterations, a relatively accurate temperature-mechanical deformation-electrical performance correlation model was finally obtained.

[0111] Using this correlation model, the engineering team conducted a comprehensive diagnosis of the operating status of the 500kV converter transformer: 1. Temperature stress analysis: At 60°C, the maximum thermal stress inside the component , mainly concentrated in the windings and supporting structures. Although this stress level is lower than the yield strength of the material, it exceeds the tensile strength of the insulation material, which will cause early damage such as microcracks in the insulation layer.

[0112] 2. Deformation accumulation analysis: Under the short-circuit impact load, the components undergo large cumulative deformation, such as radial deformation of the winding. , axial deformation These deformations have exceeded the design allowable range and will threaten the insulation performance of the winding and the geometric stability of the whole machine.

[0113] 3. Mechanical strength assessment: According to the damage accumulation model, the residual mechanical strength of the component is estimated. It has dropped to 550MPa, which is only 68.8% of the initial strength. This severe strength degradation will greatly increase the risk of equipment failure.

[0114] 4. Insulation performance prediction: The electrical performance model is calculated to obtain the insulation margin of the converter transformer under the above temperature and mechanical deformation. It is only 1.2, which is lower than the safety threshold of 1.5. The reduction of the insulation distance of the insulation layer and the increase of the local electric field strength will accelerate the aging and breakdown of the insulation material.

[0115] Based on the above analysis results, the engineering team believes that the 500kV converter transformer is already in a critical operating state and urgently needs to be repaired or replaced.

[0116] It should be noted that the variables involved in the description of the present invention are shown in Table 1 below.

[0117] Table 1 Variable explanation table

[0118] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for detecting the influence of temperature and mechanical deformation accumulation on the short-circuit resistance of a commutator, characterized in that: The following steps are involved: S01. Obtaining material characteristic parameters of a converter transformer, wherein the material characteristic parameters include thermal expansion coefficient of a conductor material, thermal expansion coefficient of an insulating material, elastic modulus parameter of a conductor material, elastic modulus parameter of an insulating material, fracture toughness parameter of a conductor material, fracture toughness parameter of an insulating material, yield strength parameter of a conductor material, and yield strength parameter of an insulating material; S02, installing the converter transformer component to be tested on a dynamic mechanical tensile press, and heating the converter transformer component to be tested to a preset temperature through a temperature control system; S03, using a three-dimensional laser scanner to obtain the initial state parameters of the converter transformer component to be tested at the preset temperature, wherein the initial state parameters include winding size parameters, pad force parameters, support bar force parameters, and insulation layer parameters; S04, using a temperature sensor array to measure the temperature field distribution data of the converter transformer component to be tested; S05, using finite element analysis software to calculate the initial electric field distribution parameters of the converter transformer component to be tested according to the temperature field distribution data and the material characteristic parameters; S06. According to preset short-circuit impact load parameters, using the dynamic mechanical tensile press to apply a dynamic force load to the converter transformer component to be tested; S07, using a displacement sensor array to collect mechanical response data of the converter transformer component under the dynamic force load, wherein the mechanical response data includes deformation of the winding, deformation of the pad, deformation of the stay, and deformation of the insulation layer; S08, using a high-speed camera system to record the residual deformation data of the converter transformer component to be tested after the dynamic force load ends, wherein the residual deformation data includes the residual deformation of the winding, the residual deformation of the cushion block, the residual deformation of the stay, and the residual deformation of the insulation layer; S09, using an insulation distance measuring instrument to measure the insulation distance change of the component to be tested of the converter transformer; S10, adjusting the preset temperature, repeating steps S03 to S09, and obtaining mechanical response data, residual deformation data and insulation distance variation of the converter transformer component to be tested at different temperatures; S11, establishing a converter transformer dynamic characteristic database under the cumulative effect of temperature and mechanical deformation according to the mechanical response data, the residual deformation data and the insulation distance change; S12, analyzing the variation law of mechanical properties of conductors and winding components based on the converter transformer dynamic characteristics database; S13, establishing a correlation model of the mechanical and electrical properties of the converter transformer under the cumulative effect of temperature and mechanical deformation according to the mechanical property change law; S14. Based on the converter transformer dynamic characteristic database, the data is divided into a training set, a validation set and a test set in a ratio of 8:1:1, the physical model parameters and the neural network compensation terms of each equation in the association model are trained, and the trained association model is output and saved.

2. The method for detecting the influence of temperature and mechanical deformation accumulation on the short-circuit resistance performance of commutation rheology according to claim 1 is characterized in that: The correlation model includes a temperature stress equation, a deformation accumulation equation, a mechanical strength equation, and an electrical performance equation.

3. The method for detecting the influence of temperature and mechanical deformation accumulation on the short-circuit resistance performance of the commutation transformer according to claim 2 is characterized in that: The temperature stress equation is used to calculate the thermal stress distribution of converter transformer components at different temperatures. The input includes the temperature field distribution data, the thermal expansion coefficient of the conductor material, the thermal expansion coefficient of the insulating material, the elastic modulus parameter of the conductor material, and the elastic modulus parameter of the insulating material. The output is the component stress distribution data. The temperature stress equation contains a neural network compensation term based on a three-layer perceptron, which is used to compensate for the nonlinear temperature stress effect.

4. The method for detecting the influence of temperature and mechanical deformation accumulation on the short-circuit resistance performance of a commutator according to claim 3 is characterized in that: The deformation accumulation equation is used to calculate the cumulative deformation of the component under short-circuit impact, and the input includes the stress distribution data of the component, the preset short-circuit impact load parameter, the yield strength parameter of the conductor material, and the yield strength parameter of the insulating material. The output is the cumulative deformation of the component; the deformation accumulation equation includes a neural network compensation term based on a three-layer perceptron, which is used to compensate for the nonlinear deformation accumulation effect.

5. The method for detecting the influence of temperature and mechanical deformation accumulation on the short-circuit resistance performance of a commutator according to claim 4, characterized in that: The mechanical strength equation is used to evaluate the strength margin of the component. The input includes the cumulative deformation of the component, the fracture toughness parameter of the conductor material, and the fracture toughness parameter of the insulating material. The output is the residual strength value of the component. The mechanical strength equation contains a neural network compensation term based on a three-layer perceptron to compensate for the nonlinear strength evolution effect.

6. The method for detecting the influence of temperature and mechanical deformation accumulation on the short-circuit resistance performance of a commutation transformer according to claim 5, characterized in that: The electrical performance equation is used to calculate the influence of mechanical deformation on electrical characteristics. The input includes the residual strength value of the component, the change in the insulation distance, and the electric field distribution parameter. The output is the insulation margin of the converter transformer. The electrical performance equation contains a neural network compensation term based on a three-layer perceptron to compensate for the nonlinear insulation characteristic effect.

7. The method for detecting the influence of temperature and mechanical deformation accumulation on the short-circuit resistance performance of a commutation transformer according to claim 6, characterized in that: The step of training the physical model parameters and neural network compensation terms of each equation in the association model includes: Train the temperature stress equation, use the least squares method to fit the physical model parameters, and use the back propagation algorithm to train the neural network compensation term parameters; Train the deformation accumulation equation, use the least squares method to fit the physical model parameters, and use the back propagation algorithm to train the neural network compensation term parameters; Train the mechanical strength equation, use the least squares method to fit the physical model parameters, and use the back propagation algorithm to train the neural network compensation term parameters; The electrical performance equations are trained, the physical model parameters are fitted using the least squares method, and the neural network compensation term parameters are trained using the back propagation algorithm.

8. The method for detecting the influence of temperature and mechanical deformation accumulation on the short-circuit resistance performance of a commutation transformer according to claim 6, characterized in that: The physical model parameters refer to the general term for material mechanical property parameters, geometric dimension parameters and boundary condition parameters.

9. The method for detecting the influence of temperature and mechanical deformation accumulation on the short-circuit resistance performance of a commutation transformer according to claim 8, characterized in that: For each neural network compensation term, a lightweight three-layer perceptron structure is adopted, which is built based on the GhostNet model, and an adaptive feature fusion module is added between the first Ghost module and the second Ghost module of the GhostNet model. The adaptive feature fusion module includes a channel attention submodule, a spatial attention submodule and a discriminant equation.

10. The method for detecting the influence of temperature and mechanical deformation accumulation on the short-circuit resistance performance of commutation rheology according to claim 9, characterized in that: The discriminant equation is used to evaluate the matching degree between the material characteristic parameters and the attention weights. The input includes the material characteristic parameters, the channel attention weights and the spatial attention weights. The output is the effectiveness score of the feature fusion.

Images