Methods, apparatus, systems, and storage media for predicting the thermo-oxidative aging properties of polyamide

Abstract

This invention discloses a method, device, system, and storage medium for predicting the thermo-oxidative aging performance of polyamide, including: acquiring experimental data of polyamide thermo-oxidative aging under various coupled conditions such as oxygen pressure, temperature, and sample thickness, and the corresponding tensile strength retention rate; constructing a modified response prediction closed-form formula based on the diffusion-reaction synergistic mechanism and introducing it as a soft physics constraint into the model; employing a multi-objective joint training strategy to simultaneously minimize data fitting error and physical equation residuals; evaluating and visualizing the performance of the trained model, solving for and outputting the model parameters. Using the technical solution of this invention, PINN driven by the reaction-diffusion synergistic closed-form formula is used to predict the thermo-oxidative aging performance of polyamide under multiple coupled conditions.

Description

Technical Field

[0001] This invention belongs to the field of polymer material performance prediction technology, and particularly relates to a method, device, system, and storage medium for predicting the thermo-oxidative aging performance of polyamide, which is used to predict the thermo-oxidative aging performance of polyamide under multiple coupling conditions based on PINN driven by a reaction-diffusion synergistic closed-loop formula. Background Technology

[0002] Currently, the thermo-oxidative aging assessment of polymer materials mainly relies on accelerated aging experiments under conventional hot air conditions. This method achieves acceleration by increasing the ambient temperature, but it often requires several weeks or even months. In today's rapidly iterating market environment, this time-consuming assessment method cannot meet the urgent needs of rapid material development and screening. In particular, how to achieve a quantitative evaluation of the thermo-oxidative aging behavior of polymer materials under complex service environments while balancing evaluation efficiency and extrapolation reliability has become a key scientific problem that urgently needs to be solved by academia and industry.

[0003] In recent years, machine learning (ML) technology has demonstrated strong application potential in the field of materials science. However, its "black box" nature makes it difficult for models to reveal the mechanism of thermo-oxidative aging. Physics-guided neural networks (PINNs), in addition to traditional data fitting loss, automatically differentiate and calculate the residuals of mass conservation, energy conservation, momentum conservation, and constitutive relations, and embed these residuals as physical constraints during training. This achieves a seamless integration of data-driven and mechanism-driven approaches, providing a new paradigm for predicting thermo-oxidative aging behavior under complex service conditions that combines accuracy and interpretability.

[0004] However, existing technologies mainly focus on macroscopic characterization and qualitative analysis, failing to incorporate multi-dimensional coupled data such as temperature, oxygen pressure, and sample thickness obtained from high-fidelity accelerated aging experiments into mechanism-driven quantitative modeling processes, making it difficult to reveal the synergistic mechanisms of various factors. On the other hand, machine learning models that rely solely on data-driven approaches lack physical constraints, resulting in insufficient extrapolation reliability under complex service conditions. Therefore, there is an urgent need for a PINN method based on a reaction-diffusion synergistic closed-loop formula, which can seamlessly integrate multi-coupled experimental design with physical equation residuals to achieve rapid, accurate, and mechanistically interpretable quantitative prediction of the thermo-oxidative aging behavior of polyamides, thus meeting the dual needs of scientific research and industrial applications for efficient assessment and mechanistic analysis. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method, apparatus, system, and storage medium for predicting the thermo-oxidative aging performance of polyamide.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for predicting the thermo-oxidative aging properties of polyamide includes:

[0008] Aging data of polyamide under different environments were collected and the aging data were preprocessed.

[0009] The preprocessed aging data is used to calculate the data-driven loss and physical residuals respectively; the physical residuals are calculated based on the modified reaction-diffusion closed-form response formula and are integrated as physical constraint terms into the training process of the physical information neural network.

[0010] Within the cross-validation framework, the Adam optimizer is invoked to simultaneously optimize both the data fitting loss and the physical residual loss, and the coefficient of determination R is evaluated periodically on the validation set. 2 And automatically save the optimal model parameters;

[0011] Forward inference is performed on the trained PINN model to generate prediction results of the thermo-oxidative aging performance of polyamide, and reaction-diffusion mechanism parameters are automatically extracted and output.

[0012] Preferably, a modified closed-form formula for response prediction is constructed based on the diffusion-reaction synergistic mechanism, which includes:

[0013]

[0014] ε′=aε0

[0015] A = A0P n

[0016]

[0017] m = cz

[0018]

[0019]

[0020] Where k is the total rate constant, c is the diffusion kinetic parameter, A0 is the reaction intensity factor under the reference state, n is the pressure correction coefficient, a is the empirical correction coefficient used to reflect the deviation between the initial state of the spline and the standard reference, τ is the characteristic diffusion time, D is the diffusion coefficient, z is the anomalous diffusion exponent, and D0 is the pre-factor, also known as the diffusion frequency factor; E D It is the diffusion activation energy, E a The activation energy is given by R, the gas constant is given by R, which is 8.314 J·mol⁻¹·K⁻¹, and T is the thermal aging temperature.

[0021] As a preferred approach, dimensionless processing is introduced, replacing the time term t in the response prediction formula with the diffusion control time scale t / τ, where τ reflects the diffusion delay of oxygen in the polymer system, to demonstrate the dominant role of sample thickness d and diffusion coefficient D in the aging rate. Simultaneously, in the atmosphere variable processing, the total pressure parameter is replaced with oxygen pressure PO2. Oxygen pressure is dimensionless based on standard atmospheric pressure, expressing the actual oxygen pressure as a multiple relative to atmospheric pressure. The modulation effect of oxygen pressure on the aging rate is quantitatively characterized through an oxygen pressure correction index, thus constructing a corrected closed-form formula for response prediction.

[0022]

[0023] Wherein, PO20 is the standard reference oxygen pressure, which is 0.101325 MPa, and E... eff It is the apparent activation energy.

[0024] The present invention also provides a device for predicting the thermo-oxidative aging performance of polyamide, comprising:

[0025] The first processing module is used to collect aging data of polyamide under different environments and preprocess the aging data;

[0026] The second processing module is used to calculate the data-driven loss and physical residual based on the preprocessed aging data; wherein, the physical residual is calculated based on the modified reaction-diffusion closed-form response formula and is integrated as a physical constraint term into the training process of the physical information neural network.

[0027] The third processing module, within the cross-validation framework, calls the Adam optimizer to simultaneously optimize the data fitting loss and the physical residual loss, by periodically evaluating the coefficient of determination R on the validation set. 2 And automatically save the optimal model parameters;

[0028] The fourth processing module is used to perform forward inference on the trained PINN model to generate prediction results of the thermo-oxidative aging performance of polyamide, and automatically extract and output reaction-diffusion mechanism parameters.

[0029] Preferably, a modified closed-form formula for response prediction is constructed based on the diffusion-reaction synergistic mechanism, which includes:

[0030]

[0031] ε′=aε0

[0032] A = A0P n

[0033]

[0034] m = cz

[0035]

[0036] Where k is the total rate constant, c is the diffusion kinetic parameter, A0 is the reaction intensity factor under the reference state, n is the pressure correction coefficient, a is the empirical correction coefficient used to reflect the deviation between the initial state of the spline and the standard reference, τ is the characteristic diffusion time, D is the diffusion coefficient, z is the anomalous diffusion exponent, and D0 is the pre-factor, also known as the diffusion frequency factor; E D It is the diffusion activation energy, E a The activation energy is given by R, the gas constant is given by R, which is 8.314 J·mol⁻¹·K⁻¹, and T is the thermal aging temperature.

[0037] As a preferred approach, dimensionless processing is introduced, replacing the time term t in the response prediction formula with the diffusion control time scale t / τ, where τ reflects the diffusion delay of oxygen in the polymer system, to demonstrate the dominant role of sample thickness d and diffusion coefficient D in the aging rate. Simultaneously, in the atmosphere variable processing, the total pressure parameter is replaced with oxygen pressure PO2. Oxygen pressure is dimensionless based on standard atmospheric pressure, expressing the actual oxygen pressure as a multiple relative to atmospheric pressure. The modulation effect of oxygen pressure on the aging rate is quantitatively characterized through an oxygen pressure correction index, thus constructing a corrected closed-form formula for response prediction.

[0038]

[0039] Wherein, PO20 is the standard reference oxygen pressure, which is 0.101325 MPa, and E... eff It is the apparent activation energy.

[0040] The present invention also provides a polyamide thermo-oxidative aging performance prediction system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a polyamide thermo-oxidative aging performance prediction method when executed by the processor.

[0041] The present invention also provides a storage medium storing a computer program, which executes a method for predicting the thermo-oxidative aging performance of polyamide when running.

[0042] This invention fully integrates multi-factor accelerated aging experimental data with machine learning modeling technology. By constructing a nonlinear mapping relationship between key environmental parameters such as temperature, oxygen pressure, and thickness and the tensile strength retention rate of polyamide, it achieves efficient prediction and dynamic evaluation of aging performance.

[0043] This invention incorporates the modified reaction-diffusion closed-loop response formula as a physical residual term into PINN training, achieving simultaneous minimization of data fitting error and physical equation residuals, thereby ensuring that the model output strictly follows the diffusion-reaction synergistic mechanism; at the same time, it can automatically extract key mechanism parameters and has the ability to rapidly migrate between other polyamide or similar polymer systems, meeting the aging performance prediction needs of different material systems.

[0044] This invention not only significantly shortens the cycle and cost of traditional aging performance evaluation, but also, compared with purely data-driven or purely empirical fitting methods, combines high precision, strong physical consistency and good extrapolation performance, providing systematic, data-driven and physically interpretable theoretical support for the durability design, service performance prediction and engineering application of polyamide materials. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0046] Figure 1 This is a flowchart of the method for predicting the thermo-oxidative aging performance of polyamide according to an embodiment of the present invention;

[0047] Figure 2 This is the model evaluation result of the method for predicting the thermo-oxidative aging performance of polyamides under multiple coupling conditions based on the reaction-diffusion synergistic closed-loop formula driven by PINN in this invention.

[0048] Figure 3 This is the model evaluation result of the method for predicting the thermo-oxidative aging performance of polyamides under multiple coupling conditions based on the reaction-diffusion synergistic closed-loop formula driven by PINN in this invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0051] Example 1:

[0052] like Figure 1 As shown, this embodiment of the invention provides a method for predicting the thermo-oxidative aging performance of polyamide, comprising:

[0053] Step S100: Obtain aging data of polyamide PA6 under different environments: thickness, temperature, pressure, oxygen concentration, oxygen pressure, aging time, and corresponding tensile strength retention rate.

[0054] Aging data includes: thickness d, temperature T, pressure P, oxygen concentration VOL, oxygen pressure PO2, aging time t, and the corresponding tensile strength retention rate. Specific data are shown in Table 1.

[0055] Table 1

[0056]

[0057]

[0058] Oxygen pressure PO2 is obtained by calculating pressure P * oxygen concentration VOL.

[0059]

[0060] Where ε is the tensile strength after stretching, and ε0 is the initial tensile strength.

[0061] Remove outlier / missing values, discard rows containing any null value (NaN), and delete all records with positive and negative infinity.

[0062] Unit conversion: Temperature is converted from degrees Celsius (°C) to Kelvin (K), T K =273.15+T ℃ The aging time is converted from hours (h) to seconds (s), t s =t h ×3600;

[0063] Boundary trimming: To avoid numerical instability caused by zero or negative pressure, a lower limit trimming is applied to the oxygen pressure PO2 to ensure that PO2 ≥ 1 × 10⁻⁶. -6 MPa.

[0064] Step S200: Based on the diffusion-reaction synergistic mechanism, construct a modified closed-form formula for response prediction and introduce it into the model as a soft physics constraint.

[0065] The extraction of the diffusion-reaction synergistic mechanism includes:

[0066]

[0067] ε′=aε0

[0068] A = A0Pn

[0069]

[0070] m = cz

[0071]

[0072] Where k is the total rate constant, c is the diffusion kinetic parameter, A0 is the reaction intensity factor under the reference state, n is the pressure correction coefficient, a is the empirical correction coefficient used to reflect the deviation between the initial state of the spline and the standard reference, τ is the characteristic diffusion time, D is the diffusion coefficient, z is the anomalous diffusion exponent, and D0 is the pre-factor, also known as the diffusion frequency factor; E D It is the diffusion activation energy, E a The activation energy is given by R, the gas constant is given by R, which is 8.314 J·mol⁻¹·K⁻¹, and T is the thermal aging temperature (in K).

[0073] Dimensionless processing is introduced, replacing the time term 't' in the response prediction formula with the diffusion control time scale 't / τ', where 'τ' reflects the diffusion delay of oxygen in the polymer system, to demonstrate the dominant role of sample thickness 'd' and diffusion coefficient 'D' on the aging rate. Simultaneously, in the atmosphere variable processing, the total pressure parameter is replaced with oxygen pressure 'PO2'. Oxygen pressure is dimensionlessized based on standard atmospheric pressure, expressing the actual oxygen pressure as a multiple relative to atmospheric pressure. The modulation effect of oxygen pressure on the aging rate is quantitatively characterized by an oxygen pressure correction index, resulting in the corrected closed-form formula for response prediction:

[0074]

[0075] Wherein, PO20 is the standard reference oxygen pressure, which is 0.101325 MPa, and E... eff It is the apparent activation energy.

[0076] The obtained closed-form formula for response prediction is embedded as a soft constraint in the training of the PINN network. That is, the loss function includes both the traditional data fitting error term and the physical residual term derived from the closed-form formula. The network parameters are optimized by jointly minimizing these two errors to ensure that the model can accurately fit the experimental data and always follow the reaction-diffusion synergistic mechanism, thereby further improving the physical consistency of the prediction results and the cross-condition generalization ability.

[0077] Step S300: Adopt a multi-objective joint training strategy to minimize both data fitting error and physical equation residuals.

[0078] A set of standardized sample features is input into the PINN network, and the corresponding logarithmic tensile strength retention rate is predicted by forward calculation.

[0079] A multi-objective joint training strategy is adopted, and the loss function consists of:

[0080]

[0081]

[0082] Where λ phys It is the physical weighting coefficient, ln nn R i ln true R i ln phys R i These are the logarithmic tensile strength retention rates calculated using the neural network prediction, the experimental actual value, and the corrected response prediction closed-form formula.

[0083] The gradient of the total loss L is calculated by backpropagation over all trainable parameters of the network, and the Adam optimizer is called to perform a parameter update, thereby reducing the total loss.

[0084] Step S400: Evaluate and visualize the performance of the trained model, solve for and output the model parameters.

[0085] The optimal model, saved after training, is applied to both the entire training set and the independent test set to perform forward inference to obtain the predicted logarithmic tensile strength retention rate. Subsequently, the coefficient of determination R is calculated between the predicted values ​​and the corresponding experimental values. 2 The root mean square error (RMSE) and mean absolute error (MAE) are used to quantify the model's performance in terms of fitting accuracy and generalization ability.

[0086] Extract all physically trainable parameters from the network parameters—a, A0, n, m, c, D0, E. eff .

[0087] R 2 The results are as follows Figure 2 As shown, the closed-form formula for the PA6 corrected response prediction obtained after training is as follows:

[0088]

[0089] Example 2

[0090] This invention also provides a method for predicting the thermo-oxidative aging properties of polyamides under multiple coupling conditions based on a reaction-diffusion synergistic closed-loop formula driven by PINN, comprising:

[0091] Step S100: Obtain aging data of polyamide PA66 under different environments: thickness, temperature, pressure, oxygen concentration, oxygen pressure, aging time, and corresponding tensile strength retention rate.

[0092] Aging data includes: thickness d, temperature T, pressure P, oxygen concentration VOL, oxygen pressure PO2, aging time t, and the corresponding tensile strength retention rate. Specific data are shown in Table 1.

[0093] Oxygen pressure PO2 is obtained by calculating pressure P * oxygen concentration VOL.

[0094]

[0095] Where ε is the tensile strength after stretching, and ε0 is the initial tensile strength.

[0096] Remove outlier / missing values, discard rows containing any null value (NaN), and delete all records with positive and negative infinity.

[0097] Unit conversion: Temperature is converted from degrees Celsius (°C) to Kelvin (K), T K =273.15+T ℃ The aging time is converted from hours (h) to seconds (s), t s =t h ×3600;

[0098] Boundary trimming: To avoid numerical instability caused by zero or negative pressure, a lower limit trimming is applied to the oxygen pressure PO2 to ensure that PO2 ≥ 1 × 10⁻⁶. -6 MPa.

[0099] Step S200: Based on the diffusion-reaction synergistic mechanism, construct a modified closed-form formula for response prediction and introduce it into the model as a soft physics constraint.

[0100] The extraction of the diffusion-reaction synergistic mechanism includes:

[0101]

[0102] ε′=aε0

[0103] A = A0P n

[0104]

[0105] m = cz

[0106]

[0107] Where k is the total rate constant, c is the diffusion kinetic parameter, A0 is the reaction intensity factor under the reference state, n is the pressure correction coefficient, a is the empirical correction coefficient used to reflect the deviation between the initial state of the spline and the standard reference, τ is the characteristic diffusion time, D is the diffusion coefficient, z is the anomalous diffusion exponent, and D0 is the pre-factor, also known as the diffusion frequency factor; E DIt is the diffusion activation energy, E a The activation energy is given by R, the gas constant is given by R, which is 8.314 J·mol⁻¹·K⁻¹, and T is the thermal aging temperature (in K).

[0108] Dimensionless processing is introduced, replacing the time term 't' in the response prediction formula with the diffusion control time scale 't / τ', where 'τ' reflects the diffusion delay of oxygen in the polymer system, to demonstrate the dominant role of sample thickness 'd' and diffusion coefficient 'D' on the aging rate. Simultaneously, in the atmosphere variable processing, the total pressure parameter is replaced with oxygen pressure 'PO2'. Oxygen pressure is dimensionlessized based on standard atmospheric pressure, expressing the actual oxygen pressure as a multiple relative to atmospheric pressure. The modulation effect of oxygen pressure on the aging rate is quantitatively characterized by an oxygen pressure correction index, resulting in the corrected closed-form formula for response prediction:

[0109]

[0110] Wherein, PO20 is the standard reference oxygen pressure, which is 0.101325 MPa, and E... eff It is the apparent activation energy.

[0111] The obtained closed-form formula for response prediction is embedded as a soft constraint in the training of the PINN network. That is, the loss function includes both the traditional data fitting error term and the physical residual term derived from the closed-form formula. The network parameters are optimized by jointly minimizing these two errors to ensure that the model can accurately fit the experimental data and always follow the reaction-diffusion synergistic mechanism, thereby further improving the physical consistency of the prediction results and the cross-condition generalization ability.

[0112] Step S300: Adopt a multi-objective joint training strategy to minimize both data fitting error and physical equation residuals.

[0113] A set of standardized sample features is input into the PINN network, and the corresponding logarithmic tensile strength retention rate is predicted by forward calculation.

[0114] A multi-objective joint training strategy is adopted, and the loss function consists of:

[0115]

[0116]

[0117] Where λ phys It is the physical weighting coefficient, ln nn R i ln true R i ln phys R i These are the logarithmic tensile strength retention rates calculated using the neural network prediction, the experimental actual value, and the corrected response prediction closed-form formula.

[0118] The gradient of the total loss L is calculated by backpropagation over all trainable parameters of the network, and the Adam optimizer is called to perform a parameter update, thereby reducing the total loss.

[0119] Step S400: Evaluate and visualize the performance of the trained model, solve for and output the model parameters.

[0120] The optimal model, saved after training, is applied to both the entire training set and the independent test set to perform forward inference to obtain the predicted logarithmic tensile strength retention rate. Subsequently, the coefficient of determination R is calculated between the predicted values ​​and the corresponding experimental values. 2 The root mean square error (RMSE) and mean absolute error (MAE) are used to quantify the model's performance in terms of fitting accuracy and generalization ability.

[0121] Extract all physically trainable parameters from the network parameters—a, A0, n, m, c, D0, E. eff .

[0122] R 2 The results are as follows Figure 3 As shown, the closed-form formula for the PA66 corrected response prediction obtained after training is as follows:

[0123]

[0124] Example 3:

[0125] This invention also provides a device for predicting the thermo-oxidative aging performance of polyamide, comprising:

[0126] The first processing module is used to collect aging data of polyamide under different environments and preprocess the aging data;

[0127] The second processing module is used to calculate the data-driven loss and physical residual based on the preprocessed aging data; wherein, the physical residual is calculated based on the modified reaction-diffusion closed-form response formula and is integrated as a physical constraint term into the training process of the physical information neural network.

[0128] The third processing module, within the cross-validation framework, calls the Adam optimizer to simultaneously optimize the data fitting loss and the physical residual loss, by periodically evaluating the coefficient of determination R on the validation set. 2 And automatically save the optimal model parameters;

[0129] The fourth processing module is used to perform forward inference on the trained PINN model to generate prediction results of the thermo-oxidative aging performance of polyamide, and automatically extract and output reaction-diffusion mechanism parameters.

[0130] As one embodiment of the present invention, a modified closed-form formula for response prediction is constructed based on the diffusion-reaction synergistic mechanism, which includes:

[0131]

[0132] ε′=aε0

[0133] A = A0P n

[0134]

[0135] m = cz

[0136]

[0137] Where k is the total rate constant, c is the diffusion kinetic parameter, A0 is the reaction intensity factor under the reference state, n is the pressure correction coefficient, a is the empirical correction coefficient used to reflect the deviation between the initial state of the spline and the standard reference, τ is the characteristic diffusion time, D is the diffusion coefficient, z is the anomalous diffusion exponent, and D0 is the pre-factor, also known as the diffusion frequency factor; E D It is the diffusion activation energy, E a The activation energy is given by R, the gas constant is given by R, which is 8.314 J·mol⁻¹·K⁻¹, and T is the thermal aging temperature.

[0138] As one embodiment of the present invention, dimensionless processing is introduced, replacing the time term t in the response prediction formula with the diffusion control time scale t / τ, where τ reflects the diffusion delay of oxygen in the polymer system, to show the dominant control effect of the sample thickness d and diffusion coefficient D on the aging rate; simultaneously, in the atmosphere variable processing, the total pressure parameter is replaced with oxygen pressure PO2; the oxygen pressure is dimensionless based on standard atmospheric pressure, expressing the actual oxygen pressure as a multiple relative to atmospheric pressure, and the modulation effect of oxygen pressure on the aging rate is quantitatively characterized by the oxygen pressure correction index, thus constructing a corrected response prediction closed-form formula:

[0139]

[0140] Wherein, PO20 is the standard reference oxygen pressure, which is 0.101325 MPa, and E... eff It is the apparent activation energy.

[0141] Example 4:

[0142] This invention also provides a polyamide thermo-oxidative aging performance prediction system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a polyamide thermo-oxidative aging performance prediction method when executed by the processor.

[0143] Example 5:

[0144] This invention also provides a storage medium storing a computer program that executes a method for predicting the thermo-oxidative aging performance of polyamide during runtime.

[0145] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for predicting the thermo-oxidative aging performance of polyamide, characterized in that, include: Aging data of polyamide under different environments were collected and the aging data were preprocessed. The preprocessed aging data is used to calculate the data-driven loss and physical residuals, respectively. The physical residuals are calculated based on the modified response prediction closed-form formula and are integrated as physical constraint terms into the training process of the physical information neural network. Under the cross-validation framework, the Adam optimizer is called to optimize both the data fitting loss and the physical residual loss. The coefficient of determination R² is evaluated periodically on the validation set and the optimal model parameters are automatically saved. Forward inference is performed on the trained PINN model to generate prediction results of the thermo-oxidative aging performance of polyamide, and reaction-diffusion mechanism parameters are automatically extracted and output. A modified closed-form formula for response prediction is constructed based on the diffusion-reaction cooperative mechanism, namely: Where k is the overall rate constant, c is the diffusion kinetic parameter, A0 is the reaction intensity factor under the reference state, n is the pressure correction coefficient, and a is an empirical correction coefficient used to reflect the deviation between the initial state of the spline and the standard reference. Where D is the characteristic diffusion time, z is the diffusion coefficient, and D0 is the pre-diffusion factor, also known as the diffusion frequency factor; E D It is the diffusion activation energy, E a The activation energy is R, the gas constant is 8.314 J·mol⁻¹·K⁻¹, and the thermal aging temperature is T. Dimensionless processing is introduced, replacing the time term t in the response prediction formula with the diffusion control time scale t / τ, where τ reflects the diffusion delay of oxygen in the polymer system, to show the dominant role of sample thickness d and diffusion coefficient D on the aging rate. Simultaneously, in the atmosphere variable processing, the total pressure parameter is replaced with oxygen pressure PO2. Oxygen pressure is dimensionlessly represented using standard atmospheric pressure as a reference, expressing the actual oxygen pressure as a multiple relative to atmospheric pressure. The modulation effect of oxygen pressure on the aging rate is quantitatively characterized by an oxygen pressure correction index, thus constructing a corrected closed-form formula for response prediction. in, The standard reference oxygen pressure is 0.101325 MPa. It is the apparent activation energy.

2. A device for predicting the thermo-oxidative aging performance of polyamide, characterized in that, include: The first processing module is used to collect aging data of polyamide under different environments and preprocess the aging data; The second processing module is used to calculate the data-driven loss and physical residual based on the preprocessed aging data; wherein, the physical residual is calculated based on the modified response prediction closed formula and is integrated as a physical constraint term into the training process of the physical information neural network. The third processing module is used in the cross-validation framework to call the Adam optimizer to optimize both the data fitting loss and the physical residual loss. It periodically evaluates the coefficient of determination R² on the validation set and automatically saves the optimal model parameters. The fourth processing module is used to perform forward inference on the trained PINN model to generate prediction results of the thermo-oxidative aging performance of polyamide, and automatically extract and output reaction-diffusion mechanism parameters. A modified closed-form formula for response prediction is constructed based on the diffusion-reaction cooperative mechanism, namely: Where k is the overall rate constant, c is the diffusion kinetic parameter, A0 is the reaction intensity factor under the reference state, n is the pressure correction coefficient, and a is an empirical correction coefficient used to reflect the deviation between the initial state of the spline and the standard reference. Where D is the characteristic diffusion time, z is the diffusion coefficient, and D0 is the pre-diffusion factor, also known as the diffusion frequency factor; E D It is the diffusion activation energy, E a The activation energy is R, the gas constant is 8.314 J·mol⁻¹·K⁻¹, and the thermal aging temperature is T. Dimensionless processing is introduced, replacing the time term t in the response prediction formula with the diffusion control time scale t / τ, where τ reflects the diffusion delay of oxygen in the polymer system, to show the dominant role of sample thickness d and diffusion coefficient D on the aging rate. Simultaneously, in the atmosphere variable processing, the total pressure parameter is replaced with oxygen pressure PO2. Oxygen pressure is dimensionlessly represented using standard atmospheric pressure as a reference, expressing the actual oxygen pressure as a multiple relative to atmospheric pressure. The modulation effect of oxygen pressure on the aging rate is quantitatively characterized by an oxygen pressure correction index, thus constructing a corrected closed-form formula for response prediction. in, The standard reference oxygen pressure is 0.101325 MPa. It is the apparent activation energy.

3. A polyamide thermo-oxidative aging performance prediction system, characterized in that, include: A memory and a processor, wherein the memory stores a computer program executed by the processor, the computer program performing the polyamide thermo-oxidative aging performance prediction method as described in claim 1 when executed by the processor.

4. A storage medium, characterized in that, The storage medium stores a computer program, which executes the polyamide thermo-oxidative aging performance prediction method as described in claim 1 when running.

Images