Solid medium cracking condition prediction method and insulation parameter test method

Abstract

The invention provides a solid medium cracking condition prediction method which comprises the following steps: connecting an insulation sample, a main electrode, a high-voltage electrode, a piezoelectric sensor, a high-voltage direct-current power supply, a high-voltage pulse generator, a picoammeter and a computer together to form a specific test circuit, and stretching the insulation sample through a stretching device; therefore, the body conductivity, trap density and space charge of the solid medium body at different temperatures and in different stretching states can be measured; then taking the measured data as the input of an adversarial network model, taking the cracking condition of the solid medium as the output of the adversarial network model, training the model, enabling the input and the output to present a corresponding mapping relationship, and taking the model as a prediction model; and finally, inputting one piece of insulation parameter data into the prediction model to realize prediction of the cracking condition of the solid medium. Correspondingly, the invention further provides a solid medium insulation parameter testing method.

Description

Technical Field

[0001] The present invention belongs to the technical field of solid dielectric material parameter testing, and in particular relates to a solid dielectric cracking prediction method and an insulation parameter testing method. Background Art

[0002] Cable intermediate joints are a vulnerable link in HVDC power cable lines. Statistics show that joint failures account for nearly 40% of all failures. During power cable installation, cable joints require manual on-site installation. Therefore, compared to the cable itself, these joints are subject to more uncontrollable factors. Furthermore, due to their complex structure, the probability of failure in these joints is far greater than that of the cable itself. During daily operation, the insulation of cable joints is subjected to a variety of stresses, including electrical stress caused by applied voltage, thermal stress caused by transmitted current, and mechanical stress from the assembly and relaxation of various components. This can easily lead to insulation degradation, electrical tree formation, and ultimately, breakdown. Furthermore, epoxy components are a key component of prefabricated joints, supporting the cable accessories and maintaining interfacial pressure. Cable accessories in northern China often experience extreme temperatures ranging from -10°C to -30°C, or even lower. Therefore, prefabricated joints typically utilize an epoxy-metal embedded composite, with the metal as the conductor and the epoxy as the insulation. However, because epoxy materials have a significantly higher coefficient of thermal expansion than metals, alternating hot and cold operating conditions can lead to stress concentration zones at the epoxy-metal interface, inducing cracking there. For example, low-temperature cracking of epoxy components was discovered during commissioning of the ±535 kV HVDC cable for the Zhangbei Flexible DC Project.

[0003] The common technical challenge of low-temperature cracking in epoxy-coated structures for large-scale metal inserts requires a systematic breakthrough through multidisciplinary collaborative research. Currently, improvements have been made in areas such as the mechanism of epoxy low-temperature cracking, the coupling mechanism of the epoxy resin-metal interface, the development of domestic epoxy casting systems that resist low-temperature cracking, casting process control, product design optimization, on-site operation and maintenance measures, and the improvement of testing methods. However, the results have not been particularly significant, and the extent of low-temperature cracking cannot be accurately analyzed. Summary of the Invention

[0004] In order to overcome the above problems existing in the prior art, the present application provides a solid medium cracking condition prediction method; the method measures the volume conductivity, trap density and space charge of the solid medium body under different temperatures and different stretching states by using a specific test circuit; then the measured data are taken as the input of the adversarial network model, the cracking condition of the solid medium is taken as the output of the adversarial network model, the model is trained to make the input and the output present a corresponding mapping relationship, and the model is taken as a prediction model; at this time, only one insulation parameter (volume conductivity, trap density, space charge) data needs to be input into the prediction model, and the corresponding cracking condition can be represented, so that the prediction of the cracking condition of the solid medium can be realized. Correspondingly, the present application also provides a solid medium insulation parameter test method.

[0005] For the prediction method, the technical solution of the present application is:

[0006] The solid medium cracking condition prediction method specifically comprises the following steps:

[0007] S1, an insulation sample with a specified thickness is prepared; a test main electrode and a high-voltage electrode are arranged on the upper and lower sides of the insulation sample, and a piezoelectric sensor is arranged below the insulation sample; the high-voltage electrode is connected with a high-voltage direct current power supply and a high-voltage pulse generator respectively, the main electrode is connected with a picoammeter, and the picoammeter and the piezoelectric sensor are electrically connected with a computer respectively, so as to form a test circuit;

[0008] S2, the insulation sample is stretched by different lengths at different positions by using a stretching device, and the corresponding stretching rate is tested; then the high-voltage pulse generator is turned on, the main electrode is grounded, the high-voltage pulse excitation is applied to the insulation sample, the sound pressure wave is excited, the sound pressure wave signal is collected by the piezoelectric sensor and sent to the computer, and the space charge data are obtained;

[0009] S3, the high-voltage pulse generator is turned off, the high-voltage direct current power supply is turned on, and the insulation sample is polarized; after a period of time, the high-voltage direct current power supply is turned off, and the insulation sample is grounded and short-circuited for a period of time;

[0010] S4, the temperature of the insulation sample is controlled to linearly increase at a specified rate by using a temperature impact box, and in the process, the short-circuit current value of the insulation sample at different times and temperatures is obtained by using the picoammeter, and then the data are plotted into an SPD curve;

[0011] S5, the current peak value and the potential are obtained according to the SPD curve, and the trap density and the volume conductivity of the insulation sample under different stretching and temperatures are calculated;

[0012] S6, using the space charge, trap density and bulk conductivity data as the input layer of the adversarial network model, and the cracking condition as the output layer of the adversarial network model, the model is trained so that the input layer and the output layer present a corresponding mapping relationship, thereby obtaining a cracking condition prediction model; then, the space charge, trap density or bulk conductivity data of the solid medium is input into the prediction model to predict its cracking condition.

[0013] Compared with the existing technology, the solid medium cracking prediction method of the present application first uses a specific test circuit to measure the bulk conductivity, trap density and space charge of the solid medium body at different temperatures and different tensile states; then the measured data is used as the input of the adversarial network model, and the cracking condition of the solid medium is used as the output of the adversarial network model. The model is trained so that its input and output present a corresponding mapping relationship, and this model is used as a prediction model; at this time, only one of the insulation parameters (bulk conductivity, trap density, space charge) data needs to be input into the prediction model to obtain the corresponding cracking condition characterization, thereby realizing the prediction of the cracking condition of the solid medium, with high prediction accuracy and low implementation difficulty.

[0014] As an optimization, in the aforementioned solid dielectric crack prediction method, a grounded ring guard electrode is provided outside the main electrode. During testing, more of the bulk current in the insulating sample flows toward the guard electrode, thereby shielding the insulating sample from leakage current.

[0015] Furthermore, the calculation formula of the trap density is as follows: ; ; ; where μ is the carrier mobility; ρ is the charge density; E is the electric field strength; ε is the permittivity of the insulating sample; x is the distance from the upper surface of the insulating sample to the integration point; V0 is the DC voltage; L is the thickness of the insulating sample after stretching; V s is the surface potential; N t is the trap density, E t is the energy level; e is the electron volt, k is the Boltzmann constant, and ν is the electron escape frequency. The calculation formula for the bulk conductivity is as follows: ; Where I is the current; V is the corresponding DC voltage; L is the thickness of the insulating sample after stretching; d is the diameter of the main electrode; g is the gap between the main electrode and the circular guard electrode.

[0016] Furthermore, the adversarial network model includes a generator G and a discriminator D; both the generator G and the discriminator D are implemented by a multi-layer perceptron; the loss function formula of the generator G is: ; where x is the real sample data, including space charge, trap density and bulk conductivity data, z is the noise input, G(z) is the generated sample data, and λ is the equilibrium hyperparameter; ∥G(z)−x∥ 2 is the mean square error, log(1−D(G(z))) is the error of the generated sample data being misclassified as true by the discriminator D; the loss function formula of the discriminator D is: Where logD(x) is the probability of maximizing the true sample data to be judged as true, and log(1−D(G(z))) is the probability of maximizing the generated sample data to be judged as false. During the training process of the adversarial network model, if sample loss occurs, the successive layer gradient descent method is used. The specific formula is: ; First, calculate the gradient ∇ of the loss function L with respect to the parameter θ θ L is used to locate the parameter update direction and constrain its norm ∥∇θL∥≤c; then, the update step size is controlled based on the learning rate η, and the parameter θ is adjusted along the negative gradient direction; finally, the generated distribution is corrected through gradient backpropagation, so that the samples lost during training fall back into the high probability area to optimize the model convergence efficiency.

[0017] As an optimization, in the aforementioned solid medium crack prediction method, the stretching device includes a fixed ring and a set of spaced apart stretching rods; the insulating sample is positioned within the fixed ring; one end of the stretching rod is fixed to the inner wall of the fixed ring, and the other end is connected to the insulating sample. This configuration provides a simple structure and ease of assembly.

[0018] As an optimization, in the aforementioned method for predicting cracking of solid media, a voltage regulator and a current-limiting resistor are further provided in the test circuit, and the current-limiting resistor is greater than 1MΩ. The voltage regulator plays the role of stabilizing the voltage, and the current-limiting resistor is used to prevent the circuit from short-circuiting, thereby avoiding damage to the picoammeter when the insulating sample breaks down or flashes along the surface during the test. Furthermore, the piezoelectric sensor is coupled to the bottom surface of the high-voltage electrode; the top surface of the high-voltage electrode is in contact with the insulating sample. Therefore, when setting up the test circuit, there is no need to use additional accessories to fix the piezoelectric sensor. The high-voltage electrode directly transmits the sound pressure wave signal to the piezoelectric sensor, which has higher accuracy.

[0019] As for the testing method, the technical solution of this application is:

[0020] A method for testing the insulation parameters of a solid dielectric, wherein the insulation parameters include space charge, trap density, and bulk conductivity, specifically comprises the following steps:

[0021] S1, preparing an insulating sample with a specified thickness; placing a test main electrode and a high-voltage electrode on the upper and lower sides of the insulating sample, and placing a piezoelectric sensor below the insulating sample; the high-voltage electrode is connected to the positive electrode of a high-voltage DC power supply and a high-voltage pulse generator, respectively, the main electrode is connected to a picoammeter, the picoammeter is connected to the negative electrode of the high-voltage DC power supply, and the picoammeter and the piezoelectric sensor are electrically connected to a computer, thereby forming a test circuit;

[0022] S2: Use a stretching device to stretch different parts of the insulating sample to different lengths and measure the corresponding stretching rate. Then, turn on the high-voltage pulse generator and ground the main electrode to apply high-voltage pulse excitation to the insulating sample to stimulate acoustic pressure waves. The acoustic pressure wave signals are collected by the piezoelectric sensor and sent to the computer to obtain space charge data.

[0023] S3, disconnect the high-voltage pulse generator and connect the high-voltage DC power supply to polarize the insulating sample; after a period of time, disconnect the high-voltage DC power supply and short-circuit the insulating sample to the ground for a period of time to eliminate the charge on the surface of the insulating sample, thus completing the polarization and depolarization of the insulating sample;

[0024] S4, using a temperature shock chamber to control the temperature of the insulation sample to rise linearly at a specified rate. During this process, the short-circuit current values ​​of the insulation sample at different times and temperatures are obtained using a picoammeter, and these data are then plotted into an SPD curve;

[0025] S5. Obtain the current peak and potential based on the SPD curve, and calculate the trap density and bulk conductivity of the insulating sample under different stretching and temperatures.

[0026] Compared with the prior art, the solid dielectric insulation parameter testing method of the present application connects an insulating sample, a main electrode, a high-voltage electrode, a piezoelectric sensor, a high-voltage DC power supply, a high-voltage pulse generator, a picoammeter and a computer together to form a specific test circuit, and stretches the insulating sample through a stretching device to obtain different stretching rates; then, by controlling the switching of the high-voltage pulse generator and the high-voltage DC power supply, the space charge detection, as well as polarization and depolarization treatment of the insulating sample are performed; then, a temperature shock box is used to linearly increase the temperature to control the temperature of the insulating sample, and a computer is used to record the short-circuit current value of the insulating sample and the temperature value and time corresponding to the current value in real time to draw an SPD curve; finally, by analyzing the SPD curve, the trap density and bulk conductivity of the insulating sample at different temperatures and different stretching rates are calculated; thereby, the test of insulation parameters such as space charge, trap density, and bulk conductivity of the solid medium is realized, and the test circuit is simple and easy to implement; and each insulation parameter is strictly calculated through a formula, the data is more intuitive, and the accuracy is also higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1is a flow chart of the solid medium cracking prediction method of the present application;

[0028] Figure 2 This is a schematic diagram of the solid medium cracking prediction method of the present application;

[0029] Figure 3 This is a diagram showing the assembly relationship between the insulating sample, the high-voltage electrode, and the main electrode in this application;

[0030] Figure 4 is a schematic diagram of the assembly of the stretching device and the insulation sample in this application;

[0031] FIG5 is a schematic diagram of the adversarial network model in this application;

[0032] Figure 6 This is the surface morphology of epoxy resin under different aging days and different stresses;

[0033] Figure 7 This is the distribution diagram of traps on the epoxy resin surface under different stress aging effects;

[0034] Figure 8 is the electrical conductivity of epoxy resin under different stresses;

[0035] Figure 9 This is the spatial charge distribution diagram of epoxy resin under different thermal aging effects;

[0036] Figure 10 is the characteristic influence weight of different electrical properties;

[0037] Figure 11 Comparison chart of predicted cracking conditions and actual values.

[0038] The markings in the accompanying drawings are: 1-main electrode; 2-circular protective electrode; 3-insulating sample; 4-high-voltage electrode; 5-high-voltage DC power supply; 6-high-voltage pulse generator; 7-switch; 8-picoammeter; 9-stretching device, 91-fixing ring, 92-stretching rod; 10-computer; 11-voltage regulator; 12-current limiting resistor; 13-support table; 14-temperature shock box. DETAILED DESCRIPTION

[0039] The present invention is further described below with reference to the accompanying drawings and examples, but is not intended to be limiting of the present invention. Any details not described in the following examples are common technical knowledge in the art or can be implemented using conventional technical means in the art.

[0040] Aiming at the common technical problem of low-temperature cracking of epoxy-coated structures in large-sized metal inserts, and laying a technical foundation for the safe and stable operation of high-voltage DC cables, this application provides a method for predicting the cracking of solid dielectrics, including the following steps (see Figures 1 to 5).

[0041] S1, prepare an insulating sample 3 with a specified thickness; place a main electrode 1 on the upper surface of the insulating sample 3 and a high-voltage electrode 4 on the lower surface; a grounded ring protective electrode 2 is provided on the outside of the main electrode 1, so that the body current in the insulating sample 3 can flow more to the protective electrode 2, thereby shielding the leakage current in the insulating sample 3; a piezoelectric sensor is coupled on the bottom surface of the high-voltage electrode 4, which can respond to the pressure wave generated by the high-voltage pulse; the high-voltage electrode 4 is respectively connected to the positive pole of the high-voltage DC power supply 5 and the high-voltage pulse generator 6; the high-voltage DC power supply 5 and the high-voltage pulse generator 6 are connected in parallel, and the on and off of the high-voltage DC power supply 5 and the high-voltage pulse generator 6 are controlled by a switch 7; the main electrode 1 is connected to the negative pole of the picoammeter 8 and the high-voltage pulse generator 6; the picoammeter 8 is connected to the negative pole of the high-voltage DC power supply 5, and the picoammeter 8 and the piezoelectric sensor are respectively electrically connected to the computer 10, thereby forming a test circuit. The test circuit also includes a voltage regulator diode 11 and a current-limiting resistor 12. The voltage regulator diode 11 stabilizes the voltage, while the current-limiting resistor 12, which is greater than 1MΩ and connected in series with the high-voltage pulse generator 6 and the high-voltage DC power supply 5, prevents short circuits and damage to the picoammeter 8 during insulation sample breakdown or surface flashover during testing. The main electrode 1 and the annular guard electrode 2 are designed to be as thin as possible. The high-voltage electrode 4 is mounted on a support surface 13.

[0042] The diameter d of the main electrode 1 , the gap g between the main electrode 1 and the annular guard electrode 2 , and the initial thickness L0 of the insulating sample 3 were measured respectively.

[0043] In step S2, various parts of the insulation sample 3 are stretched to different lengths using a stretching device 9, and the corresponding stretch ratios are measured. The stretching device 9 comprises a fixed ring 91 and a set of spaced stretching rods 92. The insulation sample 3 is positioned within the fixed ring 91. One end of each stretching rod 92 is fixed to the inner wall of the fixed ring 91, and the other end is connected to the insulation sample 3. The stretch ratio is calculated as ΔL / L0, where ΔL is the stretched length.

[0044] Turn on the high-voltage pulse generator 6 and ground the main electrode 1. Apply high-voltage pulse excitation (output amplitude 1-10 kV, pulse width 1-100 μs) to the insulating sample 3, so that the space charge inside the insulating sample 3 generates mechanical vibration due to the Lorentz force, which excites acoustic pressure waves. The acoustic pressure wave signals are collected by the piezoelectric sensor and sent to the computer 10 to obtain space charge data.

[0045] The acoustic pressure wave P(x,t) excited by the force on the charge propagates in the medium. Its intensity decays with distance and carries information about the charge position. When the acoustic pressure wave reaches the piezoelectric sensor, it is converted into a voltage signal, and the charge distribution is inverted through time domain integration. The specific formula is: , ; ; where q is the space charge density; ΔE p is the pulse electric field intensity; δ(t) is the Dirac function, which characterizes the nanosecond pulse; α is the acoustic attenuation coefficient; h(t) is the medium transmission function; γ is the piezoelectric constant, which represents the sensitivity of the piezoelectric sensor; Vq(t) is the output voltage signal; the upper limit of integration d represents the maximum effective distance that the sound pressure wave propagates in the medium, which is generally expressed as the medium thickness or the effective detection depth (when the sample to be measured is thick).

[0046] S3, disconnect the high-voltage pulse generator 6, connect the high-voltage DC power supply 5, and polarize the insulating sample 3; after a period of time (about 20 minutes), disconnect the high-voltage DC power supply 5, and short-circuit the insulating sample 3 to the ground for a period of time (about 2-3 hours) to eliminate the charge on the surface of the insulating sample 3, thereby completing the polarization and depolarization of the insulating sample 3.

[0047] S4, linearly increase the temperature at a specified rate through the temperature shock box 14 to control the temperature of the insulation sample 3, and in this process, obtain the short-circuit current value of the insulation sample 3 at different times and temperatures through the picoammeter 8, and then plot these data into an SPD curve.

[0048] S5. Obtain the current peak and potential based on the SPD curve, and calculate the trap density and bulk conductivity of the insulating sample under different stretching and temperatures.

[0049] The calculation formula of the trap density is as follows: ; ; ; where μ is the carrier mobility; ρ is the charge density; E is the electric field strength; ε is the permittivity of the insulating sample; x is the distance from the upper surface of the insulating sample to the integration point; V0 is the DC voltage; L is the thickness of the insulating sample after stretching; V s is the surface potential; N t is the trap density, E t is the energy level; e is the electron volt, k is the Boltzmann constant, and ν is the electron escape frequency; T represents the thermodynamic temperature (absolute temperature), measured in Kelvin (K), which refers to the ambient temperature of the material or device and together with k (Boltzmann constant) constitutes the thermal energy scale. The calculation formula for bulk conductivity is as follows: ; Where I is the current; V is the corresponding DC voltage; L is the thickness of the insulating sample after stretching; d is the diameter of the main electrode; g is the gap between the main electrode and the circular guard electrode.

[0050] S6, uses the space charge, trap density and bulk conductivity data as the input layer of the adversarial network model, and the cracking situation as the output layer of the adversarial network model; after the data is imported into the input layer, it will enter the hidden layer of the adversarial network model to start training. If sample loss occurs during the training process, the sample will be re-trained through the successive layer gradient descent method to meet the training requirements more quickly, and finally the input layer and the output layer will present a corresponding mapping relationship, thereby obtaining a cracking situation prediction model.

[0051] In actual operation, the obtained space charge, trap density and bulk conductivity data can be divided into a training sample set and a test sample set; the training sample set is input into the adversarial network model for training, and the trained model is tested using the test sample set.

[0052] The adversarial network model includes a generator G and a discriminator D; both the generator G and the discriminator D are implemented by a multi-layer perceptron.

[0053] The loss function formula of the generator G is: ; where x is the real sample data, including space charge, trap density and bulk conductivity data, z is the noise input, G(z) is the generated sample data, and λ is the equilibrium hyperparameter; ∥G(z)−x∥ 2 is the mean square error, log(1−D(G(z))) is the error of the generated sample data being misclassified as true by the discriminator D. The loss function formula of the discriminator D is: Where logD(x) is the probability of maximizing the true sample data to be judged as true, and log(1−D(G(z))) is the probability of maximizing the generated sample data to be judged as false. The specific formula of the successor layer gradient descent method is: ; First, calculate the gradient ∇ of the loss function L with respect to the parameter θ θ L, to update the positioning parameters in the direction; to prevent gradient explosion and unstable training, the gradient needs to be clipped and its norm ∥∇ θ L∥≤c (c is a constant chosen empirically; typically between 1 and 10, depending on the number of training runs). Next, the learning rate η controls the update step size (determined by both the learning rate and the gradient, representing the actual parameter update step size and controlling the magnitude of each update). The parameter θ is adjusted along the negative gradient. Finally, the generated distribution is corrected through gradient backpropagation, allowing samples lost during training to fall back into high-probability regions, optimizing model convergence efficiency. After adversarial training, the generator G establishes a mapping from input noise z to output data x = G(z), confirming the generalization capability of the input-output mapping.

[0054] Finally, by inputting one of the data of space charge, trap density or bulk conductivity of the solid medium into the prediction model, its cracking situation can be predicted.

[0055] In order to verify the accuracy of the prediction method of this application, the applicant conducted the following related experiments.

[0056] ① Select three epoxy resins with different aging days (0 days, 10 days, and 30 days) as insulating samples and build a test circuit. Among them, the diameter d of the main electrode 1 is 50mm, the gap g between the main electrode 1 and the circular protective electrode 2 is 2mm, and the initial thickness L0 of the insulating sample 3 is 0.5mm. Use the stretching device 9 to stretch the insulating sample 3, and obtain the surface morphology of the epoxy resin under different aging days and different stresses through SEM testing, as shown in the following figure. Figure 6 As shown (where a is 0 MPa / 0 day; b is 0 MPa / 10 days; c is 0 MPa / 30 days; d is 0.1 MPa / 30 days; e is 0.2 MPa / 30 days; f is 0.3 MPa / 30 days); it can be found that microcracks will exist on the surface of the sample after aging and under stress.

[0057] The increased free volume of epoxy resin allows charges to gain greater kinetic energy through electric field acceleration, making them more susceptible to trapping and detrapping. The shallow trap density in epoxy resin is primarily related to the number of trapping and detrapping events, so the overall shallow trap density tends to increase. Research indicates that as microcracks expand, more severe traps and defects are generated within them, significantly increasing the density of shallow and deep traps, exacerbating molecular chain breakage and significantly increasing the resulting disorder. This introduces more traps within the epoxy resin, making it easier for charges to be captured.

[0058] The electrical conductivity of epoxy resins primarily depends on two parameters: carrier mobility and concentration. Carrier mobility is related to the size of the internal free volume; the smaller the free volume, the lower the carrier mobility. Carrier concentration is primarily determined by a small amount of unreacted monomers and impurities. Crack expansion partially releases stress within the epoxy resin. As the crack widens and deepens, the free volume increases. When the free volume increases to a level similar to that of certain structural units, these units gain more space to move, which increases mobility and leads to an increase in bulk conductivity.

[0059] According to steps S3-S5, the trap density and bulk conductivity of the epoxy resin are calculated. The results are as follows: Figure 7 (a aging 0 days, b aging 10 days, c aging 30 days) and Figure 8 As shown; depending on the differences in the crack conditions of the samples, their trap density and conductivity also show corresponding rules.

[0060] ② According to step S2, the space charge density of epoxy resin under three different aging effects is tested, such as Figure 9 As shown in the figure (a: aging for 0 days, b: aging for 10 days, c: aging for 30 days); it can be found that the presence of cracks will significantly change the spatial charge distribution characteristics inside the epoxy resin, and the spatial charge distribution diagram can indirectly reflect the structural damage state of the epoxy resin: when microcracks initiate inside the material, the charge will be highly enriched along the edge of the crack, forming a localized peak in the spatial charge distribution diagram (this charge aggregation phenomenon can be used to predict the cracking state of the sample).

[0061] See also Figure 10 and Figure 11 Based on the space charge, trap density and bulk conductivity data obtained from experiments ① and ②, a quantification is performed according to the aging time and cracking condition of the epoxy resin, and each sample is input into the adversarial network model in the form of a data matrix for training to obtain a cracking prediction model. Finally, one of the data items, bulk conductivity, trap density or space charge, is selected and input into the prediction model to predict the cracking condition of the epoxy resin and compare it with the actual cracking condition to obtain a regression equation. Figure 11 It can be seen that the difference between the predicted value and the actual value is not large, which shows that the above method is feasible to predict the cracking of solid media and has high prediction accuracy.

[0062] The above descriptions are merely descriptions of preferred embodiments of the present invention, and should not be construed as limiting the specific embodiments of the present invention. Those skilled in the art will appreciate that improvements and modifications can be made without departing from the principles of the present invention, and such improvements and modifications should be considered within the scope of protection of the present invention. For example, the prediction method of this application can also be applied to the prediction of breakdown voltage and dissipation factor.

Claims

1. A method for predicting cracking of solid media, characterized in that: The method specifically comprises the following steps: S1, preparing an insulating sample with a specified thickness; placing a test main electrode and a high-voltage electrode on the upper and lower sides of the insulating sample, and placing a piezoelectric sensor below the insulating sample; the high-voltage electrode is connected to a high-voltage DC power supply and a high-voltage pulse generator, respectively, the main electrode is connected to a picoammeter, and the picoammeter and the piezoelectric sensor are electrically connected to a computer, respectively, to form a test circuit; S2: Use a stretching device to stretch different parts of the insulating sample to different lengths and measure the corresponding stretching rate. Then, turn on the high-voltage pulse generator and ground the main electrode to apply high-voltage pulse excitation to the insulating sample to stimulate acoustic pressure waves. The acoustic pressure wave signals are collected by the piezoelectric sensor and sent to the computer to obtain space charge data. S3, disconnect the high-voltage pulse generator, connect the high-voltage DC power supply, and polarize the insulating sample; after a period of time, disconnect the high-voltage DC power supply and short-circuit the insulating sample to the ground for a period of time; S4, using a temperature shock chamber to linearly increase the temperature at a specified rate to control the temperature of the insulation sample. During this process, the short-circuit current values ​​of the insulation sample at different times and temperatures are obtained using a picoammeter, and these data are then plotted into an SPD curve; S5, obtain the current peak and potential according to the SPD curve, and calculate the trap density and bulk conductivity of the insulating sample under different stretching and temperature; S6, using the space charge, trap density and bulk conductivity data as the input layer of the adversarial network model, and the cracking condition as the output layer of the adversarial network model, the model is trained so that the input layer and the output layer present a corresponding mapping relationship, thereby obtaining a cracking condition prediction model; then, the space charge, trap density or bulk conductivity data of the solid medium is input into the prediction model to predict its cracking condition.

2. The method for predicting cracking of solid media according to claim 1, characterized in that: A grounding ring protection electrode is provided on the outer side of the main electrode.

3. The method for predicting cracking of solid media according to claim 2, characterized in that: The calculation formula of the trap density is: ; ; ; where μ is the carrier mobility; ρ is the charge density; E is the electric field strength; ε is the permittivity of the insulating sample; x is the distance from the upper surface of the insulating sample to the integration point; V0 is the DC voltage; L is the thickness of the insulating sample after stretching; V s is the surface potential; N t is the trap density, E t is the energy level; e is the electron volt, k is the Boltzmann constant, and ν is the electron escape frequency.

4. The method for predicting cracking of solid media according to claim 3, characterized in that: The calculation formula of the bulk conductivity is as follows: ; Where I is the current; V is the corresponding DC voltage; L is the thickness of the insulating sample after stretching; d is the diameter of the main electrode; g is the gap between the main electrode and the circular guard electrode.

5. The method for predicting cracking of solid media according to claim 4, characterized in that: The adversarial network model includes a generator G and a discriminator D; both the generator G and the discriminator D are implemented by a multi-layer perceptron; the loss function formula of the generator G is: ; where x is the real sample data, including space charge, trap density and bulk conductivity data, z is the noise input, G(z) is the generated sample data, and λ is the equilibrium hyperparameter; ∥G(z)−x∥ 2 is the mean square error, log(1−D(G(z))) is the error of the generated sample data being misclassified as true by the discriminator D; the loss function formula of the discriminator D is: Where logD(x) is the probability that the real sample data is judged to be true, and log(1−D(G(z))) is the probability that the generated sample data is judged to be false.

6. The method for predicting cracking of solid media according to claim 5, characterized in that: During the training process of the adversarial network model, if sample loss occurs, the successive layer gradient descent method is used. The specific formula is: Where ∇ θ L is the gradient of the loss function L with respect to the parameter θ, and its norm is ∥∇ θ L∥≤c; η is the learning rate.

7. The method for predicting cracking of solid media according to claim 2, characterized in that: The stretching device includes a fixed ring and a group of stretching rods arranged at intervals; the insulating sample is located in the fixed ring; one end of the stretching rod is fixed on the inner wall of the fixed ring, and the other end is connected to the insulating sample.

8. The solid dielectric insulation parameter testing method according to claim 2, characterized in that: The test circuit is further provided with a voltage-stabilizing tube and a current-limiting resistor, and the current-limiting resistor is greater than 1MΩ.

9. The solid dielectric insulation parameter testing method according to claim 2, characterized in that: The piezoelectric sensor is coupled to the bottom surface of the high-voltage electrode; the top surface of the high-voltage electrode is in contact with the insulating sample.

10. A method for testing the insulation parameters of a solid dielectric, characterized in that: The insulation parameters include space charge, trap density and bulk conductivity; the test steps are as follows: S1, preparing an insulating sample with a specified thickness; placing a test main electrode and a high-voltage electrode on the upper and lower sides of the insulating sample, and placing a piezoelectric sensor below the insulating sample; the high-voltage electrode is connected to the positive electrode of a high-voltage DC power supply and a high-voltage pulse generator, respectively, the main electrode is connected to a picoammeter, the picoammeter is connected to the negative electrode of the high-voltage DC power supply, and the picoammeter and the piezoelectric sensor are electrically connected to a computer, thereby forming a test circuit; S2: Use a stretching device to stretch different parts of the insulating sample to different lengths and measure the corresponding stretching rate. Then, turn on the high-voltage pulse generator and ground the main electrode to apply high-voltage pulse excitation to the insulating sample to stimulate acoustic pressure waves. The acoustic pressure wave signals are collected by the piezoelectric sensor and sent to the computer to obtain space charge data. S3, disconnect the high-voltage pulse generator and connect the high-voltage DC power supply to polarize the insulating sample; after a period of time, disconnect the high-voltage DC power supply and short-circuit the insulating sample to the ground for a period of time to eliminate the charge on the surface of the insulating sample, thus completing the polarization and depolarization of the insulating sample; S4, using a temperature shock chamber to linearly increase the temperature at a specified rate to control the temperature of the insulation sample. During this process, the short-circuit current values ​​of the insulation sample at different times and temperatures are obtained using a picoammeter, and these data are then plotted into an SPD curve; S5. Obtain the current peak and potential based on the SPD curve, and calculate the trap density and bulk conductivity of the insulating sample under different stretching and temperatures.

Images