The Arrhenius model highlights that the key to constructing the constitutive equation under specific strain is determining the mathematical relationship among stress, strain rate, and temperature. However, the Arrhenius model itself is an implicit equation, making it relatively complex to solve in engineering applications. Additionally, the fitting accuracy of this model for the stress–strain values of the GH4169 alloy requires improvement. Theoretically, the relationship between stress, strain rate, and temperature can be represented as
. However, due to the significant nonlinearity among these three variables, this study did not directly use this functional form. Before constructing a new constitutive model, the influence of strain rate and temperature on stress was investigated. To this end, experimental data at low strain (0.136), medium strain (0.518), and high strain (0.9) levels from
Table 3 were analyzed. To reveal the intrinsic relationship between logarithmic stress, logarithmic strain rate, and temperature, the partial derivatives of logarithmic stress with respect to temperature and logarithmic strain rate were calculated using numerical discretization methods, such as finite differences. Common numerical discretization methods include forward difference, backward difference, and central difference. To ensure computational accuracy and stability, forward and backward differences were used at boundary points, while central differences were used at internal nodes.
4.1. Mathematical Principles
Partial derivatives are an important concept in multivariable calculus, indicating the rate of change in a function in the direction of a specific variable. For a two-variable function
, the partial derivative in the
-direction is defined as follows [
39]:
Similarly, the partial derivative of the function in the
-direction is defined as follows:
In practical calculations of partial derivatives, numerical discretization methods are commonly employed, approximating derivatives using finite differences. Similarly, second-order and higher-order derivatives can be obtained through the recursive application of first-order finite differences. To ensure accuracy when calculating derivatives at boundary points, one-sided difference schemes are typically used, whereas central difference schemes are often employed at internal nodes to balance accuracy and stability.
The Taylor series is an important local expansion of functions used to approximate complex nonlinear functions. If a function
has continuous derivatives up to the nth order at the point
, its Taylor expansion is given as follows [
40]:
Nickel-based superalloys exhibit significant nonlinear mechanical behavior under high strain rates and high-temperature conditions, with complex nonlinear coupling effects between stress levels, strain rates, and temperatures. Experimental data at strain levels of 0.136, 0.518, and 0.9 from
Table 3 were selected for analysis. To quantitatively describe the functional relationship between logarithmic stress, temperature, and logarithmic strain rate, the partial derivatives of logarithmic stress with respect to temperature and logarithmic strain rate were calculated using discrete formulas. The discrete formulas include forward difference, backward difference, and central difference. Forward difference and backward difference were used at the initial and final boundaries, respectively, while central difference was used in the intermediate region. Based on the discrete difference method, the nth-order partial derivatives of logarithmic stress with respect to temperature and logarithmic strain rate at any point on the experimental stress–strain curve can be calculated using Formula (9) as the central difference.
where
represents the order of the partial derivative, and
and
represent the indices of strain rate and temperature, respectively. For example, when
= 1 and
= 1, ln
denotes the logarithmic stress at a strain rate of 0.01 s
−1 and a temperature of 850 °C. The maximum values of
and
correspond to the number of strain rates and temperatures, respectively.
When
= 1 and
= 1, Formula (9) transforms into the backward difference Formula (10).
When
= 4 and
= 8, Formula (9) transforms into the forward difference Formula (11).
From Formulas (9)–(11), it is evident that the accuracy of calculating the partial derivatives of stress is closely related to the step size of the finite differences. The denser the experimental points for temperature and strain rate, the closer the numerical calculation results are to the actual partial derivatives. Maximizing the number of experimental data points is crucial for accurately describing the constitutive behavior of the wrought GH4169 superalloy. The first, second, and third-order partial derivatives of logarithmic stress with respect to temperature and logarithmic strain rate were calculated at strain levels of 0.136, 0.518, and 0.9. The relevant results are shown in
Figure 7,
Figure 8 and
Figure 9, respectively. For the three strain levels, logarithmic stress decreases with increasing deformation temperature (as shown in
Figure 7a,
Figure 8a and
Figure 9a). This indicates that, at a given strain level, higher strain rates lead to higher stress values. At a specific strain level, the first-order partial derivatives of logarithmic stress with respect to temperature (
) show little difference at different temperatures, with a magnitude of about 0.001, as shown in
Figure 7b,
Figure 8b and
Figure 9b. This indicates that the rate of change in logarithmic stress with temperature (i.e., the slope of the logarithmic stress–temperature curve) remains relatively stable within the studied temperature range. The second-order partial derivatives (
) and third-order partial derivatives (
) of logarithmic stress with respect to temperature are close to zero at all strain levels, with smaller magnitudes, as shown in
Figure 7c,d,
Figure 8c,d and
Figure 9c,d.
From the Taylor series Formula (8), it can be seen that if the Taylor series expansion of
near the point
is written out and the fourth and higher-order terms are ignored, the following can be obtained:
The constant term , which represents the value of the function at , determines the vertical position of the Taylor series expansion but does not affect the shape of the function. The first-order term is , which is a linear function with a slope given by the first derivative . When is close to , this term causes the function to exhibit approximately linear behavior near . The second-order term is a quadratic function with a coefficient determined by the second derivative . If is not zero, this term introduces curvature, causing the function to deviate from linearity. However, if is close to zero, the contribution of this term becomes very small, and the function f(x) still exhibits approximately linear behavior near . The third-order term is a cubic function with a coefficient determined by the third derivative . Similar to the second-order term, if is close to zero, the contribution of this term also becomes very small. This implies that even if the function exhibits slight nonlinearity near , its rate of curvature change is very small, and it can still be well approximated by a linear function overall.
The first-order partial derivative of logarithmic stress with respect to temperature () is relatively stable within the studied temperature range. This indicates that the logarithmic stress–temperature curve can be well approximated by a straight line with a constant slope within this temperature range. The second-order partial derivative () and third-order partial derivative () of logarithmic stress with respect to temperature are very close to zero within the studied temperature range. This suggests that the logarithmic stress–temperature curve has almost no curvature or rate of curvature change within this temperature range, exhibiting characteristics very close to linearity. However, to fit the experimental curve more precisely, higher-order partial derivatives are considered. Therefore, the relationship between logarithmic stress ln and temperature within the studied temperature range can be described with high accuracy by a second-order function.
The importance of this conclusion is that it provides a simple and effective method for describing the law of variation of material flow stress with temperature. By analyzing the higher-order partial derivatives of logarithmic stress, the linearity of the ln- relationship can be directly judged, thereby selecting an appropriate mathematical model to fit the experimental data. This not only greatly simplifies the process of establishing the constitutive model but also improves the computational efficiency and predictive capability of the model. It provides a theoretical foundation for establishing efficient and practical material constitutive models. This method is not only applicable to the high-temperature alloy studied in this paper but can also be extended to other types of materials, offering broad application prospects.
For the studied strain levels (low strain
= 0.136, medium strain
= 0.518, and high strain
= 0.9), logarithmic stress increases with the increase in logarithmic strain rate (as shown in
Figure 10a,
Figure 11a and
Figure 12a). The first-order partial derivative of logarithmic stress with respect to logarithmic strain rate (
) varies significantly at all strain levels (as shown in
Figure 10b,
Figure 11b and
Figure 12b), and the absolute value of the first-order partial derivative increases with the increase in temperature and logarithmic strain rate. This indicates that the impact of strain rate on flow stress becomes more significant with an increase in temperature and strain rate. In other words, the material is more sensitive to strain rate under high temperature and high strain rate conditions. Furthermore, it can be found that the second-order partial derivative of logarithmic stress with respect to logarithmic strain rate (
) is relatively small at all strain levels (as shown in
Figure 10c,
Figure 11c and
Figure 12c), while the third-order partial derivative (
) is also relatively small (as shown in
Figure 10d,
Figure 11d and
Figure 12d). To make the predicted values more accurately describe the experimental values, the relatively small third-order partial derivatives are also taken into account. Therefore, the relationship between logarithmic stress and logarithmic strain rate can be described by a cubic function.
4.2. Establishment of the New Constitutive Model
Based on the previous analysis, the deformation of the constitutive equation with two variables can be expressed by the following formula:
Then, using the Taylor series expansion and more accurately considering the coupling effect of temperature and strain rate, a new constitutive model is proposed, as shown in the following Formula (14):
Further simplifying Formula (15):
where
,
,
,
,
,
,
,
,
, and
are material parameters. The constant term
represents the reference value of stress with respect to the strain rate and temperature. The first-order term
represents the first-order response of stress to strain rate, describing the linear relationship between stress and logarithmic strain rate. The first-order term
represents the first-order response of logarithmic stress to temperature, reflecting the thermal softening effect of the material. The second-order term
represents the nonlinear change in logarithmic stress with respect to the logarithmic strain rate, reflecting the nonlinear behavior of the material under strain rate. The cross term
represents the interaction between strain rate and temperature on stress, describing how stress is influenced by both strain rate and temperature, and reflecting the coupling effect of the temperature and strain rate. The second-order term
represents the nonlinear change in logarithmic stress with respect to temperature, reflecting the nonlinear characteristics of the material’s thermal softening behavior. The third-order term
represents the third-order response of stress to strain rate, also describing the complex nonlinear relationship between stress and logarithmic strain rate. The term
represents the interaction between the second-order term of strain rate and temperature, describing how stress is influenced by the square of logarithmic strain rate and temperature. The term
represents the second-order interaction between strain rate and temperature, describing how stress is influenced by logarithmic strain rate and the square of temperature. The term
represents the interaction between the third-order term of the logarithmic strain rate and the first-order term of temperature, capturing the high-order nonlinear relationship between logarithmic strain rate and temperature. The term
represents the interaction between the first-order term of the logarithmic strain rate and the third-order term of temperature. The term
represents the interaction between the second-order term of the logarithmic strain rate and the second-order term of temperature, reflecting the more complex interaction between these two variables.
Based on the data provided in
Table 3, the material parameters in Formula (15) can be obtained using multiple linear regression. Additionally, since different strains have different material parameters, the relationship between material parameters and strain can be fitted using a polynomial. The degree of the polynomial can be determined based on the regression accuracy. In this study, a seventh-degree polynomial was used to fit these material parameters, and the relationship between each material parameter and different strains was obtained, thus incorporating stress, strain, strain rate, and temperature into the new constitutive equation. The coefficients of each term are shown in
Table 5, and the polynomial fitting curves of the quadratic model coefficients at different strain levels are shown in
Figure 13. For different materials, the number of coefficients and the order of partial derivatives must be determined based on the complexity of their mechanical behavior. The accuracy of the new constitutive model is also ensured by these factors. Additionally, the coefficients must be solved mathematically.
By substituting the relationship between the parameters of the new constitutive model for the GH4169 alloy and strain from
Table 5 into Formula (15), a complete new constitutive model is obtained. This equation considers the nonlinear relationship between material parameters and strain, allowing the model to more accurately describe the mechanical behavior of the material under different deformation conditions. To evaluate the predictive ability of the newly established constitutive model, the stress predicted by the new model was compared with the experimentally measured data, as shown in
Figure 14. In the figure, the solid lines of different colors represent the data obtained from experiments, reflecting the true mechanical behavior of the material under different temperatures and strain rates. The scatter points represent the stress values predicted by the new model under the corresponding conditions. By visually comparing the degree of agreement, the predictive accuracy and reliability of the new model can be qualitatively assessed.
4.3. Evaluation of Predictive Performance
To comprehensively evaluate the overall predictive accuracy of the classical model and the new model, a series of standard statistical parameters were introduced, including the correlation coefficient (
), root mean square error (RMSE), sum of squared errors (SSE), and sum of absolute errors (SAE). These parameters quantify the deviation between the model predictions and experimental measurements from different perspectives, providing an objective basis for comparing the performance of different models. The correlation coefficient (
) measures the linear correlation between the model predictions and the experimental values, and its calculation formula is as follows:
where
and
represent the predicted and experimental values of the
th data point, respectively,
and
represent the mean predicted and experimental values, respectively. The value of
ranges from [−1, 1]. If
is closer to 1, it indicates a stronger positive correlation between the predicted and experimental values. If
is closer to −1, it indicates a stronger negative correlation, while values close to 0 indicate no significant linear correlation between them.
Both the root mean square error (RMSE) and the sum of squared errors (SSE) measure the overall magnitude of the deviation between the model predictions and the experimental values. However, RMSE focuses more on reflecting local deviations, while SSE emphasizes reflecting global deviations. Their calculation formulas are shown in Formulas (17) and (18), respectively:
where
represents the total number of data points. The smaller the values of RMSE and SSE, the smaller the deviation between the model predictions and the experimental values, indicating higher overall predictive accuracy of the model.
The sum of absolute errors (SAE) calculates the cumulative amount of absolute errors between the model predictions and the experimental values. Its calculation formula is shown in Formula (19):
This indicator is similar to RMSE and SSE. The smaller the value of SAE, the smaller the absolute deviation between the model predictions and the experimental values, indicating higher overall predictive accuracy of the model.
By calculating the values of these four statistical parameters for both the classical model and the new model, and comparing them, the overall predictive performance of the two models can be quantitatively evaluated. Generally, if the new model achieves better results in these parameters (e.g.,
closer to 1, RMSE, SSE, and SAE smaller), it indicates that incorporating the nonlinear relationship between material parameters and strain effectively improves the overall predictive accuracy of the constitutive model. The quantitative comparison of the predictive accuracy of different models is shown in
Table 6.
According to the quantitative comparison of the predictive abilities of different models in
Table 6, the new model performs the best in all indicators. Its correlation coefficient (
) is 0.9948; the root mean square error (RMSE) is 22.5; the sum of squared errors (SSE) is 16,356; and the sum of absolute errors (SAE) is 5561 MPa. These values indicate that the model has superior fitting and predictive capabilities. The Arrhenius model follows closely, with a correlation coefficient (
) of 0.9888, a root mean square error (RMSE) of 44.1020, a sum of squared errors (SSE) of 19,450, and a sum of absolute errors (SAE) of 8832 MPa. However, its overall performance is slightly inferior to the new model. The HS model shows the poorest performance in all indicators, with a correlation coefficient (
) of 0.9827, a root mean square error (RMSE) of 44.0915, a sum of squared errors (SSE) of 38,881, and a sum of absolute errors (SAE) of 10,877 MPa, indicating its relatively weaker predictive ability.
Figure 15 shows the relationship between the predicted values of different models and the experimental data. The solid line represents the experimental data, hollow circles represent the predictions of the HS model, plus signs represent the predictions of the Arrhenius model, and solid dots represent the predictions of the new model. By analyzing
Figure 15a–d, it can be seen that under different strain rates and temperatures, the new model has the best fitting effect, accurately predicting the stress–strain relationship of the material under various conditions. The Arrhenius model follows, fitting the experimental data well under most conditions but performing slightly worse than the new model under high temperatures and high strain rates. The HS model has relatively poor predictive performance, especially under high temperature and high strain rate conditions, where its predictions deviate significantly from the experimental data. Therefore, the new model shows a clear advantage in predicting material behavior.
Figure 16 shows the predictive ability of different models for flow stress. The coefficient of determination
for the Arrhenius model is 0.977; for the HS model, it is 0.967; and for the new model, it is 0.99. Combined with the previous analysis results, the new model has the best fitting effect under various strain rate and temperature conditions, accurately predicting the stress–strain relationship of the material and having the highest predictive accuracy. The Arrhenius model follows, with good predictive ability, but it is slightly inferior to the new model. The HS model has the poorest predictive ability, with significant deviations, especially under high-stress conditions. Therefore, the new model demonstrates the highest reliability and accuracy in practical applications. Consequently, this high-precision model, with its mathematical expression, can be integrated into forming simulation software. This integration provides accurate simulations for optimizing existing products or developing new ones, thereby shortening the production cycle.