A deep learning-based piezoelectric hydrogel performance optimization method

Abstract

The application discloses a piezoelectric hydrogel performance optimization method based on deep learning, and relates to the technical field of material preparation, which comprises the following steps: preparing a hydrogel precursor solution of a bifunctional monomer, adding a crosslinking agent N,N'-methylene bisacrylamide and stirring and dissolving to obtain a uniform transparent solution; injecting the uniform transparent solution into a polytetrafluoroethylene mold, controlling the wet film thickness by adopting a doctor blade coating method, and performing ultraviolet initiation polymerization to form a gel thin film; cutting the gel thin film into samples, measuring initial piezoelectric coefficients, dielectric constants and dielectric loss factors, and obtaining initial performance data. Through the innovative combination of the adversarial network and the reinforcement learning technology, the application constructs a complete piezoelectric hydrogel intelligent optimization system, the adversarial network dynamic polarization control step utilizes the physical constraint adversarial training to generate an asymmetric triangular wave polarization scheme, and the accurate matching between the polarization parameters and the material dynamic response is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of material preparation, and particularly relates to a piezoelectric hydrogel performance optimization method based on deep learning. BACKGROUND

[0002] As a new type of intelligent material, piezoelectric hydrogel has a broad application prospect in the fields of flexible electronics and biosensing due to its unique force-electric coupling characteristics. Traditional performance optimization methods are mainly based on component regulation and process optimization. A double network structure is constructed by copolymerization of acrylic acid and acrylamide, and a d33 coefficient of 15 pC / N is achieved by combining with direct current polarization processing. The monomer ratio and crosslinking density are optimized by using the orthogonal test method. Although the basic performance is improved, the polarization process still uses a fixed waveform loading mode, and the dynamic response characteristics of the material in the non-equilibrium polarization process are not fully considered. The polarization efficiency of the piezoelectric polymer is closely related to the time sequence characteristics of the electric field, and in particular, the rising and falling edge rates have a decisive influence on the orientation of dipoles. However, the traditional fixed waveform polarization cannot accurately control such parameters.

[0003] The current technical system has significant limitations in interface regulation. When it comes to ionic piezoelectric hydrogel, existing methods mostly use empirical methods to adjust pH and ionic strength, and lack quantitative correlation analysis of the double-layer formation mechanism. The synergistic effect of Zeta potential and Stern layer thickness can significantly affect the interface charge transfer efficiency. However, the traditional technology neither establishes a data mapping relationship between polarization and interface characteristics, nor develops a regulation strategy that can dynamically adapt to different chemical compositions. This limitation leads to a long performance optimization cycle. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a piezoelectric hydrogel performance optimization method based on deep learning, which solves the problem of long performance optimization cycle and large interface impedance caused by the lack of data mapping relationship between polarization and interface characteristics in the traditional empirical regulation method.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a piezoelectric hydrogel performance optimization method based on deep learning, which comprises: preparing a hydrogel precursor solution of a double-functional monomer, adding a crosslinking agent N,N'-methylenebisacrylamide and stirring to dissolve, to obtain a uniform transparent solution;

[0008] The uniform transparent solution is injected into a polytetrafluoroethylene mold, a doctor blade coating method is used to control the wet film thickness, and ultraviolet initiation polymerization is performed to form a gel thin film;

[0009] Cut the gel film into samples to measure the initial piezoelectric coefficient, dielectric constant and dielectric loss factor, and obtain the initial performance data;

[0010] Input the initial performance data into the adversarial network to generate an asymmetric triangular wave polarization scheme, and load it to the sample through a high-voltage amplifier for polarization treatment;

[0011] Immerse the sample after polarization treatment in a buffer solution with adjustable pH to obtain a sample with a specific Zeta potential range, measure the sample with a specific Zeta potential range to obtain the interfacial double layer characteristic data;

[0012] Input the interfacial double layer characteristic data into the reinforcement learning model to obtain the pH value and ion concentration conditions, and process the sample with a specific Zeta potential range; test the piezoelectric coefficient, power density and cycle stability of the sample after processing to obtain the performance data.

[0013] As a preferred scheme of the piezoelectric hydrogel performance optimization method based on deep learning, wherein: a hydrogel precursor solution of bifunctional monomer is prepared, a crosslinking agent N,N'-methylene bisacrylamide is added and stirred to dissolve, and a uniform transparent solution is obtained, including the following steps,

[0014] The acrylic acid monomer is weighed, placed in a dry beaker, and AMPS monomer is added to form a bifunctional monomer mixture;

[0015] Mix deionized water with dimethyl sulfoxide, then add disodium ethylenediaminetetraacetate after mixing to form a mixed solvent;

[0016] Add the bifunctional monomer mixture to the mixed solvent to form a preliminary solution, add N,N'-methylene bisacrylamide to the total mass of the monomer to the preliminary solution, and transfer it to a vacuum degassing machine to eliminate bubbles to obtain a uniform transparent solution.

[0017] As a preferred scheme of the piezoelectric hydrogel performance optimization method based on deep learning, wherein: a hydrogel precursor solution of bifunctional monomer is prepared, a crosslinking agent N,N'-methylene bisacrylamide is added and stirred to dissolve, and a uniform transparent solution is obtained, including the following steps,

[0018] Spray APTES ethanol solution on the polytetrafluoroethylene mold and bake to form a hydrophilic modified surface, inject the uniform transparent solution into the treated mold, adjust the speed and gap of the automatic doctor blade coater, and set the parameters for the doctor blade coating solution to form the doctor blade coated wet film;

[0019] Move the doctor blade coated wet film into the UV curing box for low intensity UV irradiation, and then switch to full intensity UV irradiation to form the gel film.

[0020] As a preferred embodiment of the deep learning-based piezoelectric hydrogel performance optimization method described in this invention, the method includes the following steps: cutting the gel film into samples, measuring the initial piezoelectric coefficient, dielectric constant, and dielectric loss factor to obtain initial performance data.

[0021] The gel film was cut into samples, and the cut surfaces of the samples were treated with a plasma cleaner to remove micro-burrs.

[0022] The processed sample is set in d 3 3 In the tester fixture, the voltage output is recorded to obtain the initial piezoelectric coefficient. The processed sample is then transferred to the impedance analyzer electrode fixture to measure the dielectric constant and loss factor.

[0023] The initial piezoelectric coefficient, dielectric constant, and loss factor are integrated to obtain the initial performance data.

[0024] As a preferred embodiment of the deep learning-based piezoelectric hydrogel performance optimization method described in this invention, the method includes the following steps: inputting initial performance data into an adversarial network to generate an asymmetric triangular wave polarization scheme, and then applying the scheme to the sample via a high-voltage amplifier for polarization processing.

[0025] The initial performance data is range-planned to generate a standardized performance vector. The adversarial network is then trained using a loss function to obtain the trained adversarial network.

[0026] The standardized performance vector is input into the adversarial network to obtain the asymmetric triangular wave parameters, forming an asymmetric triangular wave polarization scheme. The electric field uniformity of the generated waveform is checked by COMSOL simulation, and unqualified asymmetric triangular wave polarization schemes are removed.

[0027] Initialize the Trek 610E high-voltage amplifier and FLIR A655sc thermal imager, immerse the sample in a silicone oil bath, apply an electric field according to the generated waveform, and record the actual voltage and current curves to obtain the polarized sample.

[0028] As a preferred embodiment of the deep learning-based piezoelectric hydrogel performance optimization method described in this invention, the method includes the following steps: immersing the polarized sample in a pH-adjustable buffer solution to obtain a sample within a specific Zeta potential range; measuring the sample within the specific Zeta potential range to obtain interfacial double-layer characteristic data.

[0029] Disodium hydrogen phosphate and sodium dihydrogen phosphate were mixed in a molar ratio to prepare a phosphate buffer solution. The buffer solution was monitored using a pH meter, and the pH was adjusted by adding hydrochloric acid and sodium hydroxide to obtain the target pH buffer solution.

[0030] The polarized sample is washed with ion water, then immersed in a target pH buffer, and the Zeta potential instrument is used for real-time monitoring to obtain a sample with a specific Zeta potential range;

[0031] The double-layer structure is calculated by molecular dynamics simulation to obtain the Stern layer thickness;

[0032] The diffusion layer charge density is calculated by electrochemical impedance spectroscopy test;

[0033] The interface resistance is obtained by equivalent circuit fitting based on electrochemical impedance spectroscopy;

[0034] The interface double-layer characteristic data is integrated from the interface resistance, the sample with a specific Zeta potential range, the Stern layer thickness and the diffusion layer charge density.

[0035] As a preferred scheme of the piezoelectric hydrogel performance optimization method based on deep learning, the interface double-layer characteristic data is input into a reinforcement learning model to obtain pH value and ion concentration conditions, and the sample with a specific Zeta potential range is processed, including the following steps,

[0036] The interface double-layer characteristic data is normalized, and the normalized interface double-layer characteristic data is input into the reinforcement learning model to output an adjustment strategy;

[0037] According to the adjustment strategy, hydrochloric acid and sodium hydroxide are added to the target pH buffer to adjust the pH, and lithium perchlorate is dissolved to adjust the ion concentration to obtain a new buffer;

[0038] The sample with a specific Zeta potential range is immersed in the new buffer, and the Zeta potential instrument is used for online monitoring to obtain a treated sample with a specific Zeta potential range.

[0039] As a preferred scheme of the piezoelectric hydrogel performance optimization method based on deep learning, the piezoelectric coefficient, power density and cycle stability of the treated sample with a specific Zeta potential range are tested to obtain performance data, including the following steps,

[0040] The treated sample with a specific Zeta potential range is installed in a d 3 3 tester fixture, the output voltage is measured by a charge amplifier to obtain the piezoelectric coefficient;

[0041] The d 3 3 tested sample with a specific Zeta potential range is transferred to a vibration table to simulate a human motion environment, a power point is scanned by an adjustable load resistance box, an open circuit voltage and a short circuit current are recorded, and a power density is calculated;

[0042] The sample after power test is installed to a micro force fatigue machine to obtain a cycle number-performance attenuation curve;

[0043] The piezoelectric coefficient, power density and cycle number-performance attenuation curve are integrated to obtain performance data.

[0044] In a second aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the computer program, when executed by the processor, implements any step of the deep learning-based piezoelectric hydrogel performance optimization method according to the first aspect of the present application.

[0045] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any step of the deep learning-based piezoelectric hydrogel performance optimization method according to the first aspect of the present application.

[0046] The present application has the following beneficial effects: the present application innovatively combines an adversarial network and reinforcement learning technology to construct a complete piezoelectric hydrogel intelligent optimization system, the adversarial network dynamic polarization control step uses physical constraint adversarial training to generate an asymmetric triangular wave polarization scheme, realizes accurate matching of polarization parameters and material dynamic response, and the reinforcement learning optimization interfacial double layer step cooperates with a data-driven interfacial precise control method, an accurate material preparation process and a standardized performance test process to form a closed-loop optimization of material preparation-performance test-machine learning optimization. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0048] Fig. 1 It is a flowchart of the deep learning-based piezoelectric hydrogel performance optimization method.

[0049] Fig. 2 It is a schematic diagram of interfacial double layer characteristic data.

[0050] Fig. 3 It is a schematic diagram of a homogeneous transparent solution.

[0051] Fig. 4 It is a schematic diagram of integrated initial performance data. DETAILED DESCRIPTION

[0052] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0053] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details given herein, that the present application can be practiced with other different ways, and that the present application is not limited to the specific embodiments described herein.

[0054] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an embodiment that is separate or alternative to other embodiments.

[0055] Reference Figs. 1-4 For one embodiment of the present application, the embodiment provides a deep learning-based piezoelectric hydrogel performance optimization method, comprising the following steps:

[0056] S1, a hydrogel precursor solution of a bifunctional monomer is prepared, a crosslinking agent N,N'-methylene bisacrylamide is added and stirred to dissolve, and a uniform transparent solution is obtained.

[0057] S1.1, the acrylic acid monomer is weighed and placed in a dry beaker, and the AMPS monomer is added to form a bifunctional monomer mixture.

[0058] Further, 12.5g of acrylic acid monomer is accurately weighed using an analytical balance, and the weighed acrylic acid monomer is transferred to a 100mL glass beaker washed with acetone and dried. According to the molar ratio of acrylic acid monomer to 2-acrylamide-2-methylpropanesulfonic acid monomer 4:1, 2-acrylamide-2-methylpropanesulfonic acid monomer is quantitatively added using a pipette, and the beaker is placed on a magnetic stirrer. Stir at 300 revolutions per minute in a 25℃ environment for 30 minutes until the acrylic acid monomer and 2-acrylamide-2-methylpropanesulfonic acid monomer are completely mixed to form a uniform transparent liquid, and a bifunctional monomer mixture is obtained.

[0059] S1.2, deionized water and dimethyl sulfoxide are mixed, and then disodium ethylenediaminetetraacetate is added after mixing to form a mixed solvent.

[0060] Further, 70 milliliters of deionized water and 30 milliliters of dimethyl sulfoxide were added to a 200 milliliter glass beaker, and a glass rod was used to stir at a uniform speed of 60 revolutions per minute for 10 minutes to form a uniform transparent mixed solution. After stirring was completed, 0.1 grams of disodium ethylenediaminetetraacetate powder was weighed using an electronic balance with a precision of 0.0001 grams, and the weighed disodium ethylenediaminetetraacetate powder was slowly added to the mixed solution, and stirring was continued for 20 minutes until the disodium ethylenediaminetetraacetate was completely dissolved, obtaining a clear transparent mixed solvent.

[0061] S1.3, the bifunctional monomer mixture was added to the mixed solvent to form a preliminary solution, and N,N'-methylenebisacrylamide was added to the preliminary solution, and transferred to a vacuum degassing machine to eliminate bubbles, obtaining a uniform transparent solution.

[0062] Further, the prepared bifunctional monomer mixture was slowly poured into a 500 milliliter three-necked flask containing the mixed solvent, and a mechanical stirrer was used to continuously stir at a speed of 200 revolutions per minute for 30 minutes to fully dissolve the bifunctional monomer mixture in the mixed solvent to form a preliminary solution. N,N'-methylenebisacrylamide crosslinking agent equivalent to 0.6% of the total mass of the bifunctional monomer mixture was accurately weighed by an electronic balance (for example, when the total mass of the bifunctional monomer mixture is 20 grams, 0.12 grams of N,N'-methylenebisacrylamide is weighed), and the weighed N,N'-methylenebisacrylamide was added to the preliminary solution in three times with an interval of 5 minutes, and the stirring speed was kept unchanged during the interval. After the addition was completed, the three-necked flask was transferred to a vacuum degassing machine, and degassing treatment was carried out under a vacuum degree of -0.095 MPa for 45 minutes, and the solution temperature was maintained within the range of 25±1°C during the degassing treatment. After the degassing was completed, a uniform transparent solution without bubbles and with a transmittance greater than 95% was obtained, which was immediately transferred to a light-proof container for sealing and storage. The whole operation process was carried out in a yellow light environment to avoid ultraviolet light induced pre-polymerization.

[0063] S2, the uniform transparent solution was injected into a polytetrafluoroethylene mold, and a doctor blade coating method was used to control the wet film thickness and perform ultraviolet induced polymerization to form a gel thin film.

[0064] S2.1, the polytetrafluoroethylene mold was sprayed with an APTES ethanol solution and baked to form a hydrophilic modified surface, the uniform transparent solution was injected into the treated mold, the speed and gap of the automatic doctor blade coater were adjusted, and the parameters were set for blade coating the solution to form a wet film after blade coating.

[0065] Further, 1 wt% of (3-aminopropyl)triethoxysilane ethanol solution is uniformly sprayed on the surface of the polytetrafluoroethylene mold using a spray gun, the spraying distance is kept at 20 cm, the spraying pressure is controlled at 0.2 MPa, the sprayed polytetrafluoroethylene mold is placed in a 80°C air drying oven for baking for 60 minutes, a covalently bonded hydrophilic modified surface is formed, the uniform transparent solution is slowly injected into the treated polytetrafluoroethylene mold along the center line of the mold, the injection speed is controlled at 5 ml per minute, the scraping speed of the automatic doctor blade coater is adjusted to 5 mm per minute, the knife gap is set to 500 microns, the doctor blade is started to scrape the solution at a constant speed in one direction, the ambient humidity is kept within the range of 40±5% during the scraping process, and the temperature is maintained at 25±1°C, after the scraping is completed, the wet film thickness is detected using a laser thickness gauge to ensure that the thickness uniformity deviation is not more than ±10 microns, and a scraped wet film with a smooth surface and uniform thickness is obtained.

[0066] S2.2, the scraped wet film is moved into a UV curing box for low-intensity UV irradiation, and then switched to full-intensity UV irradiation to form a gel film.

[0067] Further, the scraped wet film is transferred to a nitrogen-protected UV curing box together with the polytetrafluoroethylene mold, first turn on the UV LED array with a wavelength of 365 nm, adjust the light intensity to 5 mW / cm2 for pre-curing irradiation, last for 2 minutes, keep the oxygen concentration in the box below 50 ppm during the pre-curing stage, and control the temperature at 25±1°C, immediately increase the UV intensity to 12 mW / cm2 for main curing irradiation after the pre-curing is completed, last for 4 minutes, use an infrared thermal imager to monitor the surface temperature of the wet film in real time during the main curing process to ensure that it does not exceed 50°C, turn off the UV light source after the curing is completed, keep the nitrogen atmosphere for 10 minutes to slowly cool the sample to room temperature, obtain a gel film with uniform crosslinking degree and no surface cracks, immediately immerse the gel film in a 0.1 mol / L sodium chloride solution after it is peeled off from the mold for storage to avoid dehydration shrinkage, and the entire curing process is operated in a light-proof environment to prevent the photoinitiator from being depleted prematurely.

[0068] S3, cut the gel film into samples to measure the initial piezoelectric coefficient, dielectric constant and dielectric loss factor to obtain initial performance data.

[0069] S3.1, cut the gel film into samples, and put the cut samples into a plasma cleaning machine to treat the cut surface and remove micro burrs.

[0070] Further, the gel film was precisely cut into 20 mm x 20 mm square samples using a CO2 laser cutting machine, with the laser power set to 10 watts and the cutting speed controlled at 50 mm / s. The cut samples were immediately placed in the reaction chamber of a plasma cleaner, with argon gas used as the working gas at a flow rate of 20 standard milliliters per minute. The power of the plasma cleaner was set to 50 watts, with a processing time of 1 minute and a chamber pressure of 30 pascals. During processing, the samples were placed on a rotating sample stage at a speed of 10 revolutions per minute to ensure uniform treatment of the cut edges. After plasma treatment, the cut edges were examined under a light microscope at 100x magnification to confirm the absence of fiber fraying or microcracks. The treated samples were stored in a desiccator with a humidity of 50% for 24 hours before subsequent testing. The entire cutting and processing process was performed in a cleanroom environment with a temperature controlled to 25 ± 1 °C.

[0071] S3.2, the treated samples were placed in a d 3 3 The initial piezoelectric coefficient was recorded by the voltage output in the tester fixture. The treated samples were transferred to the electrode fixture of the impedance analyzer for measurement of the dielectric constant and loss factor.

[0072] Further, the 20 mm x 20 mm square samples after plasma treatment were placed flat in a d 33 The samples were placed in the special fixture of the tester (model: ZJ-6A) and a pre- pressure of 10 Newtons was applied using a precision pressure device to ensure tight contact between the sample and the electrode. The dynamic load parameters were set to a sinusoidal wave load (frequency 0.5 Hz, amplitude 5 Newtons). After starting the test program, the voltage output signal of 5 complete cycles was recorded by the charge amplifier (model: Kistler 5015), and the average value was taken to obtain the initial piezoelectric coefficient. The d 3 3 After testing, the samples were carefully transferred to the electrode fixture of the impedance analyzer (model: Agilent 4294A), and conductive silver paste was used to ensure good electrode contact. The test frequency range was set to 1 Hz to 1 MHz, and an AC bias voltage of 0.5 volts was applied. The dielectric constant and loss factor curves as a function of frequency were measured. The test environment temperature was controlled to 25 ± 0.5 °C, and the relative humidity was maintained within the range of 50 ± 5%.

[0073] S3.3, the initial piezoelectric coefficient, dielectric constant, and loss factor were integrated to obtain the initial performance data.

[0074] Further, the initial piezoelectric coefficient, dielectric constant and loss factor are introduced into the data analysis software (example: Origin 2022) to establish a data matrix containing the three-dimensional relationship of frequency-dielectric constant-loss factor. The statistical analysis of the 5 measurement results of the initial piezoelectric coefficient is carried out to obtain the average value and standard deviation, and the abnormal data points deviating from the average value by more than 3% are removed. The dielectric constant and loss factor data are taken as 20 logarithmically uniformly distributed data points every ten times of frequency range from 1 hertz to 1 megahertz. The physical rationality of the data is verified by using Kramers-Kronig transformation. The initial piezoelectric coefficient, dielectric constant and loss factor values of the characteristic frequency points (example: 1 kilohertz, 10 kilohertz, 100 kilohertz) verified are integrated into a structured data set, stored as a JSON format file, and the data set contains sample number, test time, environmental temperature and humidity, etc. Metadata information. After integration, SHA-256 algorithm is used to generate data verification code.

[0075] S4, input the initial performance data into the adversarial network, generate an asymmetric triangular wave polarization scheme, and load it to the sample through a high-voltage amplifier for polarization treatment.

[0076] S4.1, range planning is performed on the initial performance data to generate a standardized performance vector, and the adversarial network is trained through a loss function to obtain a trained adversarial network.

[0077] Specifically, the expression is,

[0078]

[0079] wherein 3 total is the loss function, is the adversarial loss, λ is a hyperparameter balancing the adversarial loss and the physical constraint term, is the dielectric loss gradient constraint.

[0080] S4.2, the standardized performance vector is input into the adversarial network to obtain asymmetric triangular wave parameters, form an asymmetric triangular wave polarization scheme, and check the electric field uniformity of the waveform through COMSOL simulation to remove unqualified asymmetric triangular wave polarization schemes.

[0081] Further, the normalized standardization performance vector is input into the pre-trained generative adversarial network generator to generate an asymmetric triangular wave parameter group containing the rising slope, peak field strength, retention time, and falling slope. The generated parameter group is imported into COMSOL Multiphysics software to establish a two-dimensional axisymmetric model containing sample geometric dimensions, electrode configuration, and material parameters. Boundary conditions are set to simulate the actual polarization environment, and transient electric field simulation calculation is performed to extract the sample thickness direction electric field strength distribution curve. The electric field uniformity index (example: electric field strength standard deviation to average value ratio) is calculated. When the electric field uniformity index exceeds the 5% threshold, it is determined as an unqualified scheme and is rejected. The simulated asymmetric triangular wave parameter group is converted into time-electric field strength sequence data, and stored as a polarization scheme file in CSV format. Each polarization scheme file is accompanied by a corresponding COMSOL simulation report, recording key parameters such as maximum electric field strength, minimum electric field strength, and electric field uniformity index.

[0082] S4.3, initialize Trek 610E high-voltage amplifier and FLIR A655sc thermal imager, immerse the sample in a silicone oil bath, apply an electric field according to the generated waveform, record the actual voltage and current curves, and obtain the polarized sample.

[0083] Further, turn on the Trek 610E high-voltage amplifier power switch, set the output voltage range to 0-20 kilovolts after preheating for 30 minutes, connect the control computer and load the asymmetric triangular wave polarization scheme CSV file verified by COMSOL simulation, install the FLIR A655sc thermal imager on a fixed support 10 centimeters away from the sample, set the sampling frequency to 1 frame per second, the temperature measurement range to 20-100℃, and the accuracy to ±0.5℃. Immerse the sample to be polarized in a 50 centistokes viscosity dimethyl silicone oil bath, and maintain a 5 millimeter distance between the sample and the high-voltage electrode. After starting the polarization program, the high-voltage amplifier applies an electric field according to the loaded waveform scheme, while the data acquisition card synchronously records the actual output voltage and current signals at a sampling rate of 1000 times per second. The FLIR A655sc thermal imager monitors the sample surface temperature distribution in real time, and automatically triggers a safety protocol to pause the polarization program when the local temperature exceeds 50℃. After completing the entire waveform loading, gradually reduce the electric field to zero at a rate of 2 kilovolts per minute, and keep the sample in the silicone oil bath for 10 minutes to eliminate residual charges. After removing the sample, wash the surface with n-hexane to remove the silicone oil, and obtain the polarized sample after nitrogen drying. Attach a two-dimensional code label containing the polarization parameters to the sample surface for subsequent tracing.

[0084] S5, immerse the polarized sample in a pH-adjustable buffer solution to obtain a sample with a specific Zeta potential range, measure the sample with a specific Zeta potential range, and obtain interfacial double layer characteristic data.

[0085] S5.1, Mix disodium hydrogen phosphate and sodium dihydrogen phosphate in a molar ratio, prepare a phosphate buffer solution, monitor the buffer solution using a pH meter, add hydrochloric acid and sodium hydroxide dropwise for pH adjustment, and obtain a target pH buffer solution.

[0086] Further, accurately weigh disodium hydrogen phosphate and sodium dihydrogen phosphate powders using an analytical balance, mix them in a molar ratio of 4:1 (example: weigh 5.68 grams of disodium hydrogen phosphate and 1.56 grams of sodium dihydrogen phosphate), dissolve the mixed powders in 500 milliliters of deionized water, and stir until completely dissolved using a magnetic stirrer at a speed of 300 revolutions per minute. Transfer the prepared phosphate buffer solution to a 1-liter glass beaker, insert a three-point calibrated pH meter electrode, monitor the solution pH value in real time, and adjust the pH using a micropipette by adding 0.1 mol / L hydrochloric acid solution or 0.1 mol / L sodium hydroxide solution dropwise. After each addition, stir for 2 minutes to mix the solution evenly. Stop adjusting when the pH value reaches the target range (example: 6.8±0.1), add deionized water to a final volume of 1 liter, and filter the adjusted phosphate buffer solution through a 0.22-micron pore size mixed cellulose ester filter membrane to remove bacteria. Store the prepared phosphate buffer solution in sterile glass bottles at 4°C in the dark for future use.

[0087] S5.2, rinse the polarized sample with ionized water, then immerse it in the target pH buffer solution, and use a Zeta potential instrument to monitor in real time to obtain a sample with a specific Zeta potential range.

[0088] Further, ultrasonically clean the polarized sample in 100 milliliters of ultrapure water for 3 times, 5 minutes each time, to remove surface residues. Immerse the cleaned sample in a previously prepared and adjusted phosphate buffer solution with a target pH value (example: 6.8±0.1), and make sure the sample is completely immersed and the buffer volume ratio is 1:10. Place the sample container in the Zeta potential instrument (example: Malvern Zetasizer Nano ZS) sample cell, set the measurement temperature to 25°C, and the electric field strength to 40 volts per centimeter. Measure each sample 3 times. The Zeta potential instrument monitors the interfacial potential in real time. When the measurement results are stable within the target range (example: -40±5 millivolts) for 5 consecutive times and the standard deviation is less than 2 millivolts, it is determined that the specific Zeta potential range has been reached. Immediately remove the sample after reaching the target, and gently absorb the excess buffer on the surface with filter paper to obtain a sample with a specific Zeta potential range.

[0089] S5.3, calculate the double-layer structure by molecular dynamics simulation to obtain the Stern layer thickness.

[0090] Specifically, the expression is,

[0091]

[0092] where d is the Stern layer thickness, k is the Boltzmann constant, T is the absolute temperature, e is the elementary charge, Stern where d is the Stern layer thickness, k is the Boltzmann constant, T is the absolute temperature, e is the elementary charge,

[0093] S5.4, electrochemical impedance spectroscopy is used to calculate the charge density of the diffusion layer.

[0094] Specifically, the expression is,

[0095]

[0096] where A is the charge density of the diffusion layer, ∈1 is the relative dielectric constant, ∈0 is the vacuum dielectric constant, c i is the molar concentration of the i-th ion in the solution, c i is the charge number of the i-th ion, and i is the ion index.

[0097] S5.5, based on electrochemical impedance spectroscopy, the interface resistance is obtained by fitting the equivalent circuit.

[0098] Further, the sample with a specific Zeta potential range is installed in a four-electrode test cell of an electrochemical workstation, the electrode spacing is set to 5 mm, and the electrochemical impedance spectroscopy test is performed in the frequency range of 0.1 Hz to 100 kHz, a sinusoidal wave disturbance signal with an amplitude of 10 mV is applied, the collected impedance spectroscopy data is imported into the equivalent circuit fitting software, an equivalent circuit model is selected for fitting, the circuit element parameters are iteratively optimized by the Levenberg-Marquardt algorithm, and when the chi-square value is less than 0.01, the fitting is determined to be converged, and the interface charge transfer resistance Rct value obtained by fitting is recorded as the interface resistance.

[0099] S5.6, the interface resistance, the sample with a specific Zeta potential range, the Stern layer thickness, and the diffusion layer charge density are integrated into the interface double-layer characteristic data.

[0100] Further, the interface resistance obtained by electrochemical impedance spectroscopy, the specific Zeta potential range sample data measured by the Zeta potential instrument, the Stern layer thickness calculated by molecular dynamics simulation, and the diffusion layer charge density derived by electrochemical impedance spectroscopy are imported into a data analysis platform, a metadata framework including sample number, test timestamp, and environmental parameters (temperature 25±0.5℃, humidity 50±5%) is established, the interface resistance is measured three times and the average value is taken, the Zeta potential data is kept for five consecutive stable measurement results, the Stern layer thickness is the mean value of three independent calculations by molecular dynamics simulation, and the diffusion layer charge density is verified by electrochemical impedance spectroscopy fitting error analysis.

[0101] ​S6. Inputting the interface double electric layer characteristic data into the reinforcement learning model to obtain the pH value and ion concentration conditions, and processing the sample with a specific Zeta potential range.

[0102] S6.1. Normalizing the interface double electric layer characteristic data, inputting the normalized interface double electric layer characteristic data into the reinforcement learning model, and outputting the adjustment strategy.

[0103] Further, the interface resistance, Zeta potential, Stern layer thickness and diffuse layer charge density in the interface double electric layer characteristic data are extracted respectively, the maximum and minimum value normalization method is used to linearly transform each parameter to the interval [0, 1], the parameter value is subtracted by the difference of the historical minimum value and divided by the difference of the historical maximum and minimum value, the historical extreme value comes from the 500 group experimental database established in the early stage, the normalized interface double electric layer characteristic data arranged in a fixed order as an 8-dimensional vector (example: [pH value, ionic strength, Zeta potential, Stern layer thickness, diffuse layer charge density, interface resistance, dielectric constant, loss factor]) is input into the pre-trained PPO (proximal policy optimization) reinforcement learning model, the PPO reinforcement learning model contains three layers of fully connected neural network (256-128-64 nodes), the Adam optimizer (learning rate 0.0003) is used to update the policy network parameters, and the adjustment strategy is output after 100 iterations of calculation.

[0104] S6.2. According to the adjustment strategy, hydrochloric acid and sodium hydroxide are added dropwise to the target pH buffer to adjust the pH, and lithium perchlorate is dissolved to adjust the ion concentration, to obtain a new buffer.

[0105] Further, the adjustment strategy parameters output by the PPO reinforcement learning model are imported into an automatic titration module (example: METTLER TOLEDO DL50), and under the condition of continuous magnetic stirring (200 revolutions / minute), 0.1 mol / L hydrochloric acid solution or 0.1 mol / L sodium hydroxide solution is added dropwise into the target pH buffer solution through a precision peristaltic pump for pH adjustment, the volume of each drop is controlled at 10 microliters, the drop interval is 30 seconds, a three-point calibrated pH meter (example: METTLER TOLEDO FE28) is used to monitor the pH change in real time, and the target pH value (example: 6.8±0.05) specified by the strategy is reached until the pH adjustment is completed. After the pH adjustment is completed, according to the ion concentration change amount required by the strategy, the corresponding mass of lithium perchlorate powder is weighed using an analytical balance (accuracy 0.0001 grams) (example: 0.053 grams of lithium perchlorate is required for adjustment +5 mmol / L), and the weighed lithium perchlorate powder is added to the buffer solution in three portions with an interval of 5 minutes, and stirred until completely dissolved. The newly prepared buffer solution is filtered through a mixed cellulose ester filter membrane with a pore size of 0.22 microns to remove bacteria, and sampled for ion chromatography (example: Thermo Scientific Dionex ICS-5000+) detection to verify the actual ion concentration, and after confirming that it meets the requirements of the strategy, it is stored in a brown glass bottle and stored at 4°C for standby use.

[0106] S6.3, immerse the sample with a specific Zeta potential range in a new buffer solution, and monitor online using a Zeta potential instrument to obtain the sample with a specific Zeta potential range after treatment.

[0107] Further, the sample with a specific Zeta potential range is completely immersed in a newly prepared buffer solution, the sample to buffer volume ratio is maintained at 1:10, and it is placed in a constant temperature magnetic stirrer (25±0.5°C) for continuous stirring at 150 revolutions / minute, a flow cell connected to a Zeta potential instrument (example: Malvern Zetasizer Nano ZS) is set, the measurement parameters are set as electric field strength 40 volts / cm, temperature 25°C, and equilibrium time 60 seconds, dynamic light scattering mode is started for online monitoring, Zeta potential data is automatically collected every 5 minutes, and when the measurement results are stable in the target range (example: -40±2 millivolts) for three consecutive times and the standard deviation is less than 1 millivolt, the treatment is terminated. After the sample is taken out, it is quickly rinsed with ultrapure water for three times, the surface liquid is blown dry with nitrogen, and it is immediately transferred to a humidity control box (50±5%RH) for 30 minutes of equilibrium. The sample with a specific Zeta potential range after treatment is obtained. The whole treatment process is operated in an anti-static environment to avoid external charge interference with the measurement results. Before each measurement, the Zeta potential instrument is calibrated using standard latex particles (example: -50±5 millivolts). The sample label contains the original number and treatment parameters, and is stored in an anti-static aluminum foil bag for standby use.

[0108] S7. Perform piezoelectric coefficient, power density and cyclic stability test on the treated sample with specific Zeta potential range, and obtain performance data.

[0109] S7.1. Install the treated sample with specific Zeta potential range on a d 3 3 Tester fixture, measure output voltage through charge amplifier, and obtain piezoelectric coefficient.

[0110] Further, place the treated sample with specific Zeta potential range on a d 3 3 Tester (Model: ZJ-6A) fixture, apply a constant pre-pressure of 10 Newtons using a precision pressure device to ensure good contact between the sample and the electrode, set the dynamic test parameters to sinusoidal wave loading (frequency 0.5 Hz, amplitude 5 Newtons), collect voltage output signals for 5 complete loading cycles through a charge amplifier (Model: Kistler 5015), set the sampling rate to 1000 points per second, calibrate the test module using a standard piezoelectric ceramic before each test to ensure measurement error is less than ±1%, control the test environment temperature at 25±0.5℃, and maintain the relative humidity within the range of 50±5%, repeat the measurement for each sample 3 times and take the average value as the final piezoelectric coefficient, immediately after the test, place the sample back in the humidity control box for storage to avoid environmental changes affecting subsequent test results, and automatically record the test data including sample number, test time, environmental parameters, etc.

[0111] S7.2. Install the treated sample with specific Zeta potential range on a d 3 3 Transfer the treated sample with specific Zeta potential range to a vibration table to simulate human motion environment, scan power points through an adjustable load resistance box, record open circuit voltage and short circuit current, and calculate power density.

[0112] Specifically, the expression is,

[0113]

[0114] Where P max is the maximum output power, V oc is the open circuit voltage, and R internal is the internal resistance of the material.

[0115]

[0116] Where p is the power density, and L is the sample volume.

[0117] S7.2. Install the sample after power test to a micro-force fatigue machine to obtain cycle number-performance attenuation curve.

[0118] Further, the specific Zeta potential range sample that has completed the power density test is installed in the fixture of the micro force fatigue testing machine, the cyclic loading parameters are set as 1 Hz sine wave, 20% strain amplitude, pre-load 0.5 Newton, the equipment is paused after completing 1000 cycles, the sample is transferred to the d 3 3 The piezoelectric coefficient is retested by the tester under the same test conditions as the initial test (0.5 Hz, 5 Newton dynamic load), and at the same time, the sample surface is scanned using a laser confocal microscope (example: Olympus LEXT OLS5000), three fixed observation areas are selected (example: the center area and 5 mm at both ends), and the crack length and density changes are recorded. After each detection, the sample is returned to the original position of the fatigue machine for continued testing until 1 million cycles are completed or the piezoelectric coefficient attenuation exceeds 20%, and the corresponding relationship curve of cycle number and piezoelectric coefficient retention rate is arranged to obtain the cycle number-performance attenuation curve.

[0119] S7.3, integrate the piezoelectric coefficient, power density and cycle number-performance attenuation curve to obtain performance data.

[0120] Further, the d 3 3 The piezoelectric coefficient data measured by the tester, the vibration table power density test results and the cycle number-performance attenuation curve obtained by the micro force fatigue testing machine are imported into a unified analysis platform, a metadata database containing the complete preparation history of the sample (from the ratio of bifunctional monomer to polarization treatment parameters) is established, the piezoelectric coefficient data is stored in association with the load resistance and vibration frequency parameters during the power density test, and the cycle number-performance attenuation curve data is extracted with a characteristic value as a node every 1000 cycles.

[0121] The embodiment also provides a computer device suitable for the deep learning-based piezoelectric hydrogel performance optimization method, which comprises a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the deep learning-based piezoelectric hydrogel performance optimization method proposed in the above embodiment.

[0122] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected by a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved by WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0123] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to implement the method for optimizing performance of piezoelectric hydrogel based on deep learning proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk or an optical disk.

[0124] In summary, the present application combines the adversarial network and the reinforcement learning technology innovatively to construct a complete piezoelectric hydrogel intelligent optimization system. The adversarial network dynamic polarization control step generates an asymmetric triangular wave polarization scheme through adversarial training of physical constraints, so as to realize accurate matching of polarization parameters and material dynamic response. The reinforcement learning interface double electric layer optimization step establishes a data-driven interface precise control method through multi-parameter collaborative search of the optimal control strategy, and cooperates with the accurate material preparation process and the standardized performance test process to form a closed-loop optimization of material preparation-performance test-machine learning optimization.

[0125] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A deep learning-based piezoelectric hydrogel performance optimization method, characterized by: Comprising, A hydrogel precursor solution of bifunctional monomers is prepared, and a crosslinker N,N'-methylene bisacrylamide is added and stirred to dissolve, to obtain a uniform transparent solution; The uniform transparent solution is injected into a polytetrafluoroethylene mold, and a doctor blade coating method is used to control the wet film thickness and perform ultraviolet initiation polymerization to form a gel film; The gel film is cut into samples, and the initial piezoelectric coefficient, dielectric constant and dielectric loss factor are measured to obtain initial performance data; The initial performance data is input into the adversarial network to generate an asymmetric triangular wave polarization scheme, and the high-voltage amplifier is loaded to the sample for polarization treatment; The sample after polarization treatment is immersed in a buffer solution with adjustable pH value to obtain a sample with a specific Zeta potential range, and the sample with a specific Zeta potential range is measured to obtain the characteristics of the electric double layer; The characteristics of the electric double layer are input into the reinforcement learning model to obtain the pH value and ion concentration conditions, and the sample with a specific Zeta potential range is treated; The piezoelectric coefficient, power density and cycle stability of the treated sample with a specific Zeta potential range are tested to obtain performance data; The initial performance data is input into the adversarial network to generate an asymmetric triangular wave polarization scheme, and the high-voltage amplifier is loaded to the sample for polarization treatment, Comprising the following steps, The initial performance data is range planned to generate a standardized performance vector, and the adversarial network is trained through a loss function to obtain a trained adversarial network; The standardized performance vector is input into the adversarial network to obtain asymmetric triangular wave parameters, form an asymmetric triangular wave polarization scheme, and check the electric field uniformity of the generated waveform through COMSOL simulation to remove unqualified asymmetric triangular wave polarization schemes; Initialize the Trek 610E high-voltage amplifier and FLIR A655sc thermal imager, immerse the sample in a silicone oil bath, apply an electric field according to the generated waveform, record the actual voltage and current curves, and obtain the polarization-treated sample.

2. The deep learning-based piezoelectric hydrogel performance optimization method of claim 1, wherein: A hydrogel precursor solution of bifunctional monomers is prepared, and a crosslinker N,N'-methylene bisacrylamide is added and stirred to dissolve, to obtain a uniform transparent solution, comprising the following steps, The acrylic monomer is weighed and placed in a dry beaker, and AMPS monomer is added to form a bifunctional monomer mixture; Deionized water and dimethyl sulfoxide are mixed, and then disodium ethylenediaminetetraacetate is added after mixing to form a mixed solvent; The bifunctional monomer mixture is added to the mixed solvent to form a preliminary solution, and the total mass of N,N'-methylene bisacrylamide is added to the preliminary solution, and transferred to a vacuum degassing machine to eliminate bubbles to obtain a uniform transparent solution.

3. The deep learning-based piezoelectric hydrogel performance optimization method of claim 2, wherein: The uniform transparent solution is injected into a polytetrafluoroethylene mold, and a doctor blade coating method is used to control the wet film thickness and perform ultraviolet initiation polymerization to form a gel film, comprising the following steps, The polytetrafluoroethylene mold is sprayed with an APTES ethanol solution and baked to form a hydrophilic modified surface, the uniform transparent solution is injected into the treated mold, the speed and gap of the automatic doctor blade coater are adjusted, and the parameters are set for doctor blade coating. The wet film after doctor blading is moved into a UV curing box for low-intensity UV irradiation, and then switched to full-intensity UV irradiation to form a gel thin film.

4. The deep learning-based piezoelectric hydrogel performance optimization method of claim 3, wherein: The gel thin film is cut into samples, and the initial piezoelectric coefficient, dielectric constant and dielectric loss factor are measured to obtain initial performance data, including the following steps, The gel thin film is cut into samples, and the cut samples are placed in a plasma cleaning machine to treat the cut edges and remove microscopic burrs; The treated sample is placed in the d 33 In the test instrument fixture, the voltage output is recorded to obtain the initial piezoelectric coefficient. The treated sample is transferred to the impedance analyzer electrode fixture to measure the dielectric constant and loss factor. The initial piezoelectric coefficient, dielectric constant and loss factor are integrated to obtain the initial performance data.

5. The deep learning-based piezoelectric hydrogel performance optimization method of claim 1, wherein: The polarized sample is immersed in a buffer solution with adjustable pH to obtain a sample with a specific Zeta potential range, and the sample with a specific Zeta potential range is measured to obtain interface double layer characteristic data, including the following steps, Mix sodium phosphate dibasic and sodium phosphate monobasic in a molar ratio to prepare a phosphate buffer solution, monitor the buffer solution using a pH meter, and add hydrochloric acid and sodium hydroxide to adjust the pH to obtain a target pH buffer solution; The polarized sample is rinsed with ionized water and then immersed in the target pH buffer solution, and a Zeta potential instrument is used for real-time monitoring to obtain a sample with a specific Zeta potential range; The Stern layer thickness is obtained by calculating the double layer structure through molecular dynamics simulation; The diffusion layer charge density is calculated by electrochemical impedance spectroscopy test; The interface resistance is obtained by equivalent circuit fitting based on electrochemical impedance spectroscopy; The interface double layer characteristic data, the sample with a specific Zeta potential range, the Stern layer thickness and the diffusion layer charge density are integrated into interface double layer characteristic data.

6. The deep learning-based piezoelectric hydrogel performance optimization method of claim 5, wherein: The interface double layer characteristic data is input into a reinforcement learning model to obtain pH and ion concentration conditions, and the sample with a specific Zeta potential range is treated, including the following steps, The interface double layer characteristic data is normalized, and the normalized interface double layer characteristic data is input into the reinforcement learning model to output an adjustment strategy; According to the adjustment strategy, hydrochloric acid and sodium hydroxide are added to the target pH buffer solution to adjust the pH, and lithium perchlorate is dissolved to adjust the ion concentration to obtain a new buffer solution; The sample with a specific Zeta potential range is immersed in the new buffer solution, and a Zeta potential instrument is used for online monitoring to obtain a treated sample with a specific Zeta potential range.

7. The deep learning-based piezoelectric hydrogel performance optimization method of claim 6, wherein: The piezoelectric coefficient, power density and cycle stability of the treated sample with a specific Zeta potential range are tested to obtain performance data, including the following steps, The sample of a specific Zeta potential range after treatment was installed in d 33 The tester clamp measured the output voltage through a charge amplifier to obtain the piezoelectric coefficient; d 33 The sample with the specific zeta potential range after the test is transferred to a vibration table to simulate the human motion environment. The power point is scanned by an adjustable load resistance box, the open circuit voltage and short circuit current are recorded, and the power density is calculated. The sample after power test is installed into a micro-force fatigue machine to obtain a cycle number-performance attenuation curve; The piezoelectric coefficient, power density and cycle number-performance attenuation curve are integrated to obtain performance data.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that: The processor executes the computer program to implement the steps of the deep learning-based piezoelectric hydrogel performance optimization method of any one of claims 1-7.

9. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to implement the steps of the deep learning-based piezoelectric hydrogel performance optimization method of any one of claims 1-7.

Images