7.1. Mesh Convergence Analysis
A mesh convergence study was first carried out with the proposed VUMAT and VUHARD models to find the optimum mesh density to give the most accurate results in the simulation of the dynamic properties of DMLS Ti6Al4V(ELI) using the SHPB test. Continuum elements C3D8R were used which have meshes with global sizes of 1 mm, 0.5 mm, 0.25 mm and 0.18 mm. The mesh convergence study was carried out by imposing an impact velocity of 8 m/s through the striker bar at a simulation temperature of 25 °C. The corresponding arising equivalent plastic strain contours are shown in
Figure 5, while the stress-strain curves resulting from the simulation using the four mesh sizes are shown in
Figure 6. The summary of the computation time in the SHPB test simulations using different mesh sizes is presented in
Table 4.
The equivalent plastic strain profiles for the SHPB test specimen are seen in
Figure 5 to be almost the same for all four mesh sizes. The maximum equivalent plastic strain on the test specimen is seen to be located at the surfaces in contact with the incident and transmission bars, and a distinct strain zone that forms an “X-shape” pattern through the diameter of the deformed model sample is visible in all cases.
The VUMAT subroutine is seen in
Table 4 to require more computation time compared to the VUHARD subroutine. As expected, the computation time increases as the mesh size decreases. It is seen from
Figure 6 that the stress-strain curves for various mesh sizes nearly overlap for the better part of the profiles, and the curves for the 0.25 mm and 0.18 mm mesh sizes are indistinguishable. It should be noted here that extensive mesh convergence analysis for the simulation of multiple element mesh models using the implemented VUMAT and VUHARD subroutine were carried out in Muiruri et al. [
43]. The researchers in the referred study found that the values of equivalent plastic strain and von Mises stress converged for mesh sizes less than 0.5 mm.
The effect of mesh sizes in
Figure 6 is seen to be significant at the initiation of equivalent plastic strain, which is consistent with the observations also made in Muiruri et al. [
43]. It is seen in this figure that the value of initial plastic strain increases with the increase in mesh size. These initial values of strain are very small, thus the difference in computed initial von Mises stress from these strain values is 1.7% between 1 mm and 0.18 mm mesh sizes and 0.87% between 0.5 mm and 0.18 mm mesh sizes. Considering the computation time shown in
Table 4 and the observation made with reference to
Figure 6 and in [
43], it can be concluded that 0.25 mm mesh size is sufficient to obtain a good prediction. This is the case, since a smaller mesh size would show a negligible improvement of the stress-strain curve and require more computation resources. Therefore, all ensuing simulations to generate stress-strain curves for various forms of DMLS Ti6Al4V(ELI) were conducted with a 0.25 mm mesh size in the present study.
7.2. The SHPB Simulation Test Results and Discussion
Simulations were carried out with striker velocities of 8 m/s, 15 m/s and 25 m/s at temperatures of 25 °C, 200 °C and 500 °C following the experimental setup described in
Section 4.2. To ensure sufficient plastic deformation and at the same time avoid excessive distortion of the model, different total step times of 1000 µs, 300 µs and 200 µs for the velocities of 8 m/s, 15 m/s and 25 m/s, respectively, were used. Typical curves of equivalent plastic strain, plastic strain rate and von Mises stress against simulation time at a velocity of 15 m/s are shown in
Figure 7.
From
Figure 7, the model is seen to have experienced near linearly increasing strains over most of the period of strain buildup. During this period, the simulation results eventually averaged at an equivalent plastic strain rate of about 1500 s
−1, a value similar to that recorded during the experimental results reported in [
44]. This demonstrates great confidence in the SHPB numerical model setup and the implemented subroutines developed in this work.
Figure 8,
Figure 9 and
Figure 10 show a comparison between the experimental results and results from the numerical model at average plastic strain rates of approximately 750 s
−1, 1500 s
−1 and 2450 s
−1 and temperatures of 25 °C, 200 °C and 500 °C. The numerically predicted values at the equivalent strain of 0.1 and 0.2 for the same test conditions were also plotted against one another and are presented in (d) of
Figure 8,
Figure 9 and
Figure 10.
The ability of the implemented numerical model in the VUMAT and VUHARD subroutines developed in this study to accurately predict the flow stress of DMLS Ti6Al4V(ELI) was assessed from the statistical measures of correlation coefficient (
) and absolute average error (
), based on the plot of
Figure 8d,
Figure 9d and
Figure 10d. The correlation coefficient (
) can provide details on the strength of the linear relationship between the experimental and predicted values. However, the
-value may not reliably show better performance of the model, due to a tendency of the linear fit to be biased toward lower or higher values. This suggests that the
-value may be misleading if outlier values are present. The
-value, on the other hand, is calculated through a term-by-term comparison of the relative error and is thus an unbiased statistical parameter for measuring the predictability of the model.
The standard statistical performance measures of
and
were obtained from the following expressions [
42]:
The correlation coefficients and average absolute errors obtained from these plots are summarised in
Table 5.
It is evident from
Table 5 that the implemented microstructural-based dislocation model in the VUMAT and VUHARD subroutines shows a very high degree of correlation as the
values are above 0.97. It is also observed in this table that the absolute percentage errors between the numerical and the experimental values are all below 4%. These measures of correlation suggest that the numerical model developed in the present study accurately predicts the flow properties of the various microstructures of DMLS Ti6Al4V(ELI) tested here. The numerical model developed in this study is therefore suitable for use in designing the dynamic strength of DMLS Ti6Al4V(ELI) by controlling the morphology of its microstructure and the initial dislocation density present in the alloy.
7.3. Simulation of SHPB Tests Using the Johnson–Cook Model In-Built in ABAQUS
The Johnson–Cook (J–C) model is a constitutive law that is commonly used to define isotropic flow properties of metals and metallic alloys during plastic deformation. This model is in-built in ABAQUS/CAE due to its simplicity of use. The model comprises an empirical form of the strain-hardening law, rate dependence and thermal softening and is normally of the form [
36,
63]:
where
where the parameters
,
,
,
and
are yield stress, strain-hardening factor, strain-hardening exponent, dimensionless strain rate hardening coefficient and thermal-softening exponent, respectively. The symbols
and
denote the equivalent plastic strain and strain rate, respectively, while
is the reference strain rate at which parameters
,
and
are determined. This is usually a low strain rate (typically
1 s
−1) where the effects of such strain rate on the plastic flow are negligible. Parameters
and
are the test and melting temperatures of the materials, respectively, while
is the room temperature. In this model, the transition temperature (
) is usually defined as a temperature at or below which there is no temperature-dependence of the yield stress and is usually taken as room temperature (
). This is the temperature used to determine the parameters
,
and
. As the model is in-built in ABAQUS for modelling isotropic flow properties, the user needs to provide the values of parameters
,
,
,
,
and
as part of metal plasticity material definition. The user also needs to provide the values of parameters C and
when defining Johnson–Cook rate dependence. A typical properties module for the J–C model with rate dependence is shown in
Figure 11.
The J–C plasticity model is also extended for high-strain-rate transient dynamic applications where the temperature change (
) in the model is generally computed internally by assuming adiabatic conditions using the following expression [
63].
Here the parameters , and are inelastic heat fraction (% of plastic work converted into heat and is normally taken as 0.9 for metal), density and specific heat capacity of a material, respectively. These three parameters must be provided for a simulation step that includes adiabatic heating effects.
The 3D finite element model of the SHPB test discussed in
Section 6, which was used to investigate the high-strain-rate deformation behaviour of Ti6Al4V(ELI) samples, was modelled using the set of J–C parameters derived from literature and shown in
Table 6. Beside these parameters, the values of specific heat, density and inelastic heat fraction 560 J/Kg·K, 4420 Kg/m
3 and 0.9 were used [
63,
64] since the simulation was taken to include adiabatic heat effects. Interestingly, the sets of parameters in
Table 6 differ considerably, even though they describe the behaviour of the same material, Ti6Al4V(ELI), though with different microstructures in certain cases.
As was the case for the numerical models developed in this work, three different striker velocities of 8 m/s, 15 m/s and 25 m/s, and a temperature of 25 °C were used to perform simulations and determine the equivalent von Mises stress and strain reported for the model at each striker velocity.
Figure 12,
Figure 13 and
Figure 14 present the plots of the results obtained using different sets of J–C model parameters in
Table 6 together with those from the numerical models developed in this study, using the VUMAT and VUHARD subroutines.
It is worth noting that the shape of the flow stress curve (relationship between the flow stress and plastic strain) seen in
Figure 12,
Figure 13 and
Figure 14 for the J–C model is established empirically by isolating the effects of temperature and strain rate. The strain rate
and temperature
parts are functions that scale the flow stress without necessarily influencing the shape of the flow stress curve. This implies that the J–C model is not adequate to represent flow characteristics of a material with the existence of recovery and recrystallization that allow stress to saturate as strain increases, as seen for samples C, D and E in this study. A similar observation was reported in the work of Yingnan et al. [
65].
Table 6.
A summary of J–C model parameters used to model the SHPB Ti6Al4V(ELI) samples.
Table 6.
A summary of J–C model parameters used to model the SHPB Ti6Al4V(ELI) samples.
J–C Model Parameter | Microstructure | α-Lath Size | A(MPa) | B(MPa) | n | C | m |
---|
Lee and Lin [66] | [-] | [-] | 782.7 | 498.4 | 0.28 | 0.028 | 1 |
Lee and Lin [67] | [-] | [-] | 724.7 | 683.1 | 0.47 | 0.035 | 1 |
Xin et al. [68] | [-] | [-] | 920 | 380 | 0.578 | 0.042 | 0.633 |
Yu et al. [39] | Lamellar | | 984.32 | 601.1 | 0.512 | 0.025 | 0.987 |
Yu et al. [39] | Lamellar | | 829.30 | 524.32 | 0.621 | 0.024 | 0.715 |
Yu et al. [39] | Bimodal | | 907.03 | 752.97 | 0.502 | 0.20 | 0.904 |
Yu et al. [39] | Bimodal | | 849.71 | 776.73 | 0.742 | 0.026 | 0.829 |
Figure 12.
Results of SHPB test simulation at an impact velocity of 8 m/s and a temperature of 25 °C with the test samples modelled using various sets of J–C model parameters (note: B and L stand for bimodal and lamellar microstructures, respectively) and the numerical models developed in the present work.
Figure 12.
Results of SHPB test simulation at an impact velocity of 8 m/s and a temperature of 25 °C with the test samples modelled using various sets of J–C model parameters (note: B and L stand for bimodal and lamellar microstructures, respectively) and the numerical models developed in the present work.
Figure 13.
Results of SHPB test simulation at an impact velocity of 15 m/s and a temperature of 25 °C with the test samples modelled using various sets of J–C model parameters (note: B and L stand for bimodal and lamellar microstructures, respectively) and the numerical models developed in the present work.
Figure 13.
Results of SHPB test simulation at an impact velocity of 15 m/s and a temperature of 25 °C with the test samples modelled using various sets of J–C model parameters (note: B and L stand for bimodal and lamellar microstructures, respectively) and the numerical models developed in the present work.
Figure 14.
Results of SHPB test simulation at an impact velocity of 25 m/s and a temperature of 25 °C with the test samples modelled using various sets of J–C model parameters (note: B and L stand for bimodal and lamellar microstructures, respectively) and the numerical models developed in the present work.
Figure 14.
Results of SHPB test simulation at an impact velocity of 25 m/s and a temperature of 25 °C with the test samples modelled using various sets of J–C model parameters (note: B and L stand for bimodal and lamellar microstructures, respectively) and the numerical models developed in the present work.
The Taylor strain-hardening model that allows for strain-hardening and dynamic recovery as strain increases through calibrated parameters
and
was used in the numerical models developed in the present research. This explains the flow stress saturation observed in samples C, D and E as opposed to continuous strain-hardening for the curves of the J–C model seen in
Figure 12,
Figure 13 and
Figure 14.
It is very clear from these three figures that each set of J–C model parameters gives a different flow stress curve. The microstructural details of Ti6Al4V(ELI) from the work of Lee and Lin [
66], Lee and Lin [
67] and Xin et al. [
68] are not inserted in
Table 6 as they are not specified in these references. However, specific microstructural information of this alloy is available from the work of Yun et al. [
39]. Each of these microstructures show different stress-strain curves in
Figure 12,
Figure 13 and
Figure 14. For instance, lamellar and bimodal microstructures with average α-lath grain sizes of 0.5 µm and 0.4 µm, respectively, show higher values of flow stress in comparison with similar microstructures with average α-lath grain sizes of 2.0 µm and 2.1 µm, respectively. The lamellar microstructure with an average α-lath grain size of 2.0 µm shows lower values of flow stress in comparison to the bimodal microstructure with an average α-lath grain size slightly larger at 2.1 µm. The initial values of flow stress at lower strain for the lamellar microstructure with 0.5 µm α-lath average grain size are higher than those for the bimodal microstructure with a 0.4 µm α-lath average grain size. At higher strains, the lamellar microstructure with a 0.5 µm α-lath average grain size shows lower values of flow stress than those for the bimodal microstructure with a 0.4 µm α-lath average grain size, especially at the higher impact velocities of 15 m/s and 25 m/s in
Figure 13 and
Figure 14.
As seen in
Figure 13 and
Figure 14, at low values of strain and at striker velocities of 15 m/s and 25 m/s, the flow stress curves of samples type C (an average α-lath grain size of 2.5 µm) from the two subroutines developed here are close to those of Lee and Lin [
67] and Yu et al. [
39] for lamellar microstructure (average α-lath grain size of 2.0 µm). Whereas the flow stress curves of samples D and E, with much larger average α-lath grain sizes of 6 µm and 9 µm, respectively, are much lower than the rest, as seen in
Figure 13 and
Figure 14.
The preceding discussion suggests that no single set of J–C model parameters can be adequate in accurately describing the flow properties of any given microstructure of Ti6Al4V(ELI). This is mainly because each microstructure is related to a different set of J–C model parameters. In contrast, the microstructure- and dislocation-based constitutive numerical models presented in this study can be used to predict the flow properties of alloys, such as Ti6Al4V(ELI), that show a wide range of microstructures. The numerical models developed in the present study offer a few advantages including:
- (a)
The critical microstructural parameters of initial dislocation density and grain size are part of the few input parameters that are needed to adequately describe the flow stress.
- (b)
Both strain-hardening and dynamic recovery that occur for deformation at high temperatures and high strain rates are articulated in these numerical models through calibrated parameters and . These parameters are insensitive to the different microstructures of samples C, D and E, as opposed to the case of the J-C model where its many parameters vary with microstructure.
- (c)
There are only four input parameters that are influenced by the microstructure in the numerical models developed in this study. These include initial dislocation density and grain size, which are determined directly from the microstructure, as well as two viscous drag stress-fitting parameters ( and ) determined empirically from the experimental data. This is less than the five microstructure-sensitive parameters of the J–C model that are empirically determined from the experimental data.