Microstructure and low cycle fatigue of a Ti₂AlNb-based lightweight alloy

1. Introduction

With the rapid development of aerospace vehicles towards fuel-efficiency, high-speed and long-voyage, the lightweight and dynamic loading-resistant materials have received more and more attention in the field of aircraft engine while retaining or enhancing safety and durability [[1], [2], [3], [4], [5]]. For this purpose, a new class of titanium-based lightweight intermetallic compound, Ti₂AlNb-based alloy, has recently received increasing attention as a promising and attractive material due to its unique ordered orthorhombic (O-Ti₂AlNb) phase [[6], [7], [8], [9], [10], [11], [12], [13], [14]]. Since its discovery in a Ti₃Al-xNb alloy in the late 1980s [15], O-Ti₂AlNb phase has been reported to improve room-temperature ductility, oxidation- and creep-resistance [10,[15], [16], [17], [18]].

As one of the most promising aerospace materials with superior balanced mechanical properties [10,17,[19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30]], the application of Ti₂AlNb-based alloy will inevitably experience fluctuating or cyclic loading. Indeed, fatigue failure of gas turbine engine blades has been recognized to be a serious problem [[31], [32], [33], [34], [35], [36]]. For example, one of fan blades in an engine on Southwest Airlines Flight 1380 bound from New York to Dallas with 149 people aboard failed by fatigue, causing catastrophic explosion of Boeing 737 CFM-56-B engine on April 17, 2018, according to the US National Transportation Safety Board. Therefore, it is essential and urgent to understand the behavior of Ti₂AlNb-based alloy when it is subjected to alternating or cyclic loading in service. The aim of this study is to evaluate the strain-controlled low cycle fatigue (LCF) behavior and lifetime of Ti₂AlNb-based alloy, since no such information is available for the Ti₂AlNb-based alloy in the open literature, to the best of the authors’ knowledge, despite some data existing for the traditional titanium alloys like Ti-6Al-4 V alloys [[37], [38], [39], [40], [41]]. This study would pave the way for the safe and durable applications of Ti₂AlNb-based lightweight intermetallic compounds in the aerospace industry.

2. Experimental procedures

2.1. Materials and sample preparation

Ti-22Al-20Nb-2V-1Mo-0.25Si (at.%) alloy (hereafter referred to as Ti₂AlNb-based alloy). The as-cast Ti₂AlNb-based alloy in this study was prepared using the vacuum consumable arc melting process for three times. X-ray diffractometer (XRD) was performed to identify the alloy phases present with CuKα radiation with a step of 0.02° and 3 s in each step. The diffraction angle (2θ) varied from 30° to 90°. SEM (JSM-6380LV) and Oxford EBSD were used to observe microstructures with a scanning step size of 0.4 μm. The samples for microstructural observations were first ground using a series of silicon carbide abrasive sandpapers, followed by electro-polishing in an electrolyte containing 6% perchloric acid, 34% n-butanol and 60% methanol at 0.6 A and 253 K for about 40 s.

2.2. Tensile and fatigue testing

The specimens with a gauge section of Φ6 × 32 mm were used for both tensile and fatigue tests, which were machined from the rods cut from the cast ingot. Tensile tests were performed at room temperature at a strain rate of 1 × 10^-2 s^-1. Two samples were repeated at the same strain rate to determine the yield strength (YS), ultimate tensile strength (UTS), and tensile elongation (%EL). Low-cycle fatigue (LCF) tests were carried out at ambient temperature under strain control using a computerized Instron 8801 fatigue testing system at a fixed strain rate of 1 × 10^-2 s^-1, using a triangular waveform under completely reversed cycling (i.e., strain ratio R_ε = -1). The total strain amplitude was applied from 0.2% to 1.0% at an interval of 0.2% until final failure or up to 10⁶ cycles. However, the LCF tests at lower strain amplitudes (i.e., 0.2% and 0.4% in this study) were initially conducted under strain control mode until 10⁴ cycles. Then it was transferred to load control mode at a frequency of 50 Hz. Sinusoidal waveform was used to accommodate the testing at the higher frequency, in accordance with ASTM E606 standards [42]. A dynamic extensometer with a gauge length of 25 mm was used to measure the strains. At least two samples were tested at each level of strain amplitude for the accuracy of the results. The fracture surface of some typical samples (failed at a strain amplitude of 0.4% and 1.0%) were chosen to examine and identify the fracture mechanisms.

3. Results and discussion

3.1. Phase constitution and microstructural characterization

XRD spectrum of the as-cast Ti₂AlNb-based alloy is presented in Fig. 1. It can be seen that the cast alloy consisted of O-Ti₂AlNb, α₂-Ti₃Al and B2 phases. Some peaks of O-Ti₂AlNb and α₂-Ti₃Al are basically the same in the analysis due to the similar crystal structure. Since α₂-Ti₃Al was reported to have a low tolerance to cracking with a lower fatigue resistance in comparison with O-Ti₂AlNb phase [7,[26], [27], [28], [29], [30],[43], [44], [45]], it is necessary to explore the morphology and fraction of each phases.

View original graphic| Download| PPT slide

Fig. 1.   XRD spectrum showing the phases present in the as-cast Ti₂AlNb-based alloy.

Fig. 2(a) displays the microstructure of the experimental Ti₂AlNb-based alloy in the SEM-BSE mode. The grains of the as-cast alloy were very large, inside which uniformly distributed fine lamellar colonies were present. According to the XRD results in Fig. 1 and the principle of backscattering, the matrix phase with white contrast to the grey colony is identified as B2 phase due to the heaviest average atomic number [7], while the lath or needle structure with specific angles is considered as O-Ti₂AlNb phase, but more analysis is needed to confirm this, since α₂-Ti₃Al has also a similar contrast.

View original graphic| Download| PPT slide

Fig. 2.   (a) SEM micrograph of as-cast Ti₂AlNb-based alloy; EBSD maps of (b) phase distribution and (c) orientation of O-Ti₂AlNb phase; (d) grain boundary distribution of O-Ti₂AlNb phase (LAGBs and HAGBs).

A better view of the grain morphology and boundary distribution of the as-cast Ti₂AlNb-based alloy is provided by EBSD mapping at a higher magnification. First, Fig. 2(b) presents the phase distribution maps, which verified the previous results that abundant refined lamellar structure of O-Ti₂AlNb phase was present. The amount of primary O-Ti₂AlNb phase is approximately 90% in the scanned region. The B2 phase is the second most abundant phase which precipitated along lath-shaped O-Ti₂AlNb phase, and only a small amount of α₂-Ti₃Al phase is distributed randomly. Fig. 2(c) shows an inverse pole figure (IPF) map of O-Ti₂AlNb phase. It is clear that the O-Ti₂AlNb phase mainly exhibited four different orientations. This is because the lath or needle-like O-Ti₂AlNb phase was formed along certain defined directions of B2 matrix with such an orientation relationship: {001}O-Ti₂AlNb//{110}B2 and <110 > O-Ti₂AlNb//<$\bar{1}$11 > B2 [16,30,[46], [47], [48], [49]], and the morphology was dependent on the cooling rate. Fig. 2(d) illustrates the distribution of low-angle grain boundaries (LAGBs) and high-angle grain boundaries (HAGBs), where the red and green lines represent HAGBs (≥15°) and LAGBs (2-15°), respectively. It can be seen that the fraction of HAGBs is much higher than that of LAGBs, which means that there were few recrystallized grains in the cast alloy.

3.2. Tensile properties

The tensile test results of the as-cast Ti₂AlNb-based alloy obtained at different strain rates from 1 × 10^-2 s^-1 to 1 × 10^-5 s^-1 at room temperature are summarized in Table 1. The experimental Ti₂AlNb-based alloy is observed to be nearly strain-rate independent considering the experimental error, indicating that the strain-rate sensitivity of this intermetallic compound is weak. The hardening capacity (H_c) of the alloy could be assessed via the following equation [50],

H_c=(σ_UTS-σ_y)/σ_y, (1)

where σ_UTS and σ_y is the UTS and YS of the material. Table 1. shows the value of H_c obtained at different strain rates from 1 × 10^-2 s^-1 to 1 × 10^-5 s^-1, with an average value of ~0.15. Plastic deformation of a material can also be depicted by strain hardening exponent (n), which could be calculated using the following Hollomon equation,

σ=Kεⁿ, (2)

Where σ is the true stress, ε is the true strain, and K represents the strength coefficient. The experimental data of σ and ε in the uniform deformation phase from yield point to ultimate tensile point are selected to evaluate n. The value of n obtained in the present alloy is ~0.16 at a strain rate of 1 × 10^-2 s^-1, corresponding to that imposed in the strain-controlled LCF tests.

3.3. Cyclic deformation response

Fig. 3(a) and (b) illustrate typical stress-strain hysteresis loops of the first cycle and mid-life cycles of samples fatigued with varying applied strain amplitudes ranging from 0.2% to 1.0%, respectively. Compared with the asymmetrical hysteresis loops of some hexagonal close-packed (hcp) alloys with a strong texture (such as extruded Mg alloys) [[51], [52], [53], [54], [55], [56]], the present Ti₂AlNb-based alloy exhibits symmetric hysteresis loops at all strain amplitudes applied. The asymmetrical deformation behavior was considered to be associated with the twinning during cyclic deformation in the compressive phase, and detwinning in the tensile phase [[51], [52], [53], [54], [55], [56]], which is due to the fact that the limited basal slip systems cannot meet the requirement of five independent slip systems based on the von Mises criterion for an arbitrary homogeneous straining [57]. In contrast, the symmetric hysteresis loop in the present study would be due to the weak texture in an as-cast state. The amount of hexagonal α₂-Ti₃Al is also very low based on the above microstructural analysis. Furthermore, the crystal structure of α₂-Ti₃Al in Ti₂AlNb-based alloys has a small c/a ratio (~0.8) [15,17], which is very different from that of a typical hexagonal close-packed α-Ti (c/a = 1.587) [[58], [59], [60], [61]]. Therefore, the symmetry in the stress-strain hysteresis loops in the present as-cast Ti₂AlNb-based alloy suggests a mode of dislocation slip-dominated deformation in the orthogonal structured O-Ti₂AlNb phase, which will be discussed in the later section. Besides, the symmetrical shape of stress-strain hysteresis loops can also facilitate the modeling of fatigue behavior since fewer fatigue parameters would be required. The initial quarter of the hysteresis loop in the first cycle at the higher strain amplitudes could be employed to evaluate the yield stress of the alloy.

View original graphic| Download| PPT slide

Fig. 3.   Typical stress-strain hysteresis loops at various total strain amplitudes of as-cast Ti₂AlNb-based alloy in the (a) first cycle, and (b) mid-life cycle.

Fig. 4 presents the change of cyclic stress amplitude with respect to the number of cycles on a semi-log scale at varying strain amplitudes from 0.2% to 1.0%. It can be seen that a slight cyclic hardening occurs at higher strain amplitudes (0.8% and 1.0%) in the as-cast Ti₂AlNb-based alloy. However, the stress amplitude is seen to be basically stable at lower strain amplitudes (from 0.2% to 0.6%). To better illustrate the cyclic hardening characteristics, the change of plastic strain amplitude with the number of cycles is shown in Fig. 5. The plastic strain amplitude exhibits a slight decrease with increasing number of cycles at higher strain amplitudes (0.8%-1.0%), reflecting the occurrence of cyclic hardening behavior. Similarly, the plastic strain amplitude also remains roughly stable at lower strain amplitudes from 0.2% to 0.6%. The average Young’s modulus in the initial check prior to LCF is obtained to be 127.8 ± 2.2 GPa. The value of Young’s modulus remains almost unchanged with increasing number of cycles and increasing strain amplitude during cyclic deformation, unlike the change of Young’s modulus in the extruded magnesium alloys [[51], [52], [53], [54]].

View original graphic| Download| PPT slide

Fig. 4.   Stress amplitude versus the number of cycles of the as-cast Ti₂AlNb-based alloy tested at different total strain amplitudes at a strain ratio of R_ε = -1.

View original graphic| Download| PPT slide

Fig. 5.   Plastic strain amplitude versus the number of cycles of the as-cast Ti₂AlNb-based alloy tested at different total strain amplitudes at a strain ratio of R_ε = -1.

3.4. Fatigue life analysis

The fatigue life of the as-cast Ti₂AlNb-based alloy is shown in Fig. 6, in comparison with that of other available intermetallic compounds reported in the literature, since there is no such information for Ti₂AlNb-based alloys, to the best of the authors’ knowledge. Because the specimens subjected to a strain amplitude of 0.2% did not fail until 10⁶ cycles, the run-out data points are labelled with horizontal arrows. As can be seen in Fig. 6, the fatigue life of the intermetallic compounds increases with decreasing strain amplitude. The present as-cast Ti₂AlNb-based alloy has a significantly higher fatigue strength or longer fatigue life, in comparison with several other intermetallic compounds reported in [[62], [63], [64]]. Obviously, the significantly higher fatigue and monotonic strengths of the present Ti₂AlNb-based alloy are attributed to the superior deformability or ductility of O-Ti₂AlNb phase [65], along with its strain hardening capacity, as mentioned earlier. The finer lath or needle O-Ti₂AlNb microstructure in the present alloy as shown in Fig. 2 would be another key factor for the superior fatigue performance. Furthermore, the overall coarse grains/colonies in the present alloy also plays an important role for preventing the initiation of fatigue crack. Therefore, the present as-cast Ti₂AlNb-based alloy with such a unique heterogeneous structure exhibits a superb fatigue performance at room temperature.

View original graphic| Download| PPT slide

Fig. 6.   Total strain amplitude versus the number of cycles to failure for the as-cast Ti₂AlNb-based alloy, in comparison with the data available for intermetallics reported in the literature [[62], [63], [64]].

Fig. 7 shows the strain-life response of the present alloy, where the total, elastic and plastic strain amplitudes as a function of the number of reversals to failure (2N_f) are plotted with black, red and blue color, respectively. The values of stress and strain in the evaluation are taken from the mid-life cycles for the sake of ensuring that cyclic saturation has already occurred. The obtained fatigue life parameters calculated by using Coffin-Mason-Basquin equation are listed in Table 2. Additionally, the value of fatigue strength exponent (b) and fatigue ductility exponent (c) can also be estimated from cyclic hardening exponent values (nʹ) based on the following equations proposed by Morrow and Tomkins [66]

View original graphic| Download| PPT slide

Fig. 7.   Total, plastic, and elastic strain amplitude-fatigue life response of the as-cast Ti₂AlNb-based alloy.

Morrow:

b≈-n'/(1+5n'), (3a)

c≈-1/(1+5n'). (3b)

Tomkins:

b≈-n'/(1+2n'), (3c)

c≈-1/(1+2n)'. (3d)

The b and c values calculated using Equ.3(a)-3(d) are listed in Table 2. as well. The obtained values are in good agreement with the experimental values.

Like the monotonic tensile stress-strain curve, the cyclic stress-strain curve (CSSC) of a material is another vital characteristic in the fatigue properties [51], which could be expressed by the following equation,

Δσ/2=K'(Δε_p/2)^n', (4)

where Δσ/2 is the mid-life stress amplitude, K' is the cyclic strength coefficient, Δε_p/2 is the mid-life plastic strain amplitude and n' is the cyclic strain-hardening exponent. Fig. 8 shows both CSSC and monotonic stress-strain curve at the same strain rate of 1 × 10^-2 s^-1. The cyclic and monotonic stress-strain curves are overlapped at nearly all strain amplitudes applied, suggesting the presence of fairly stable cyclic deformation. The obtained cyclic strain hardening exponent n' is ~0.09, which is somewhat lower than the monotonic strain hardening exponent n of ~0.16.

View original graphic| Download| PPT slide

Fig. 8.   Cyclic stress-strain curve (CSSC) for the as-cast Ti₂AlNb-based alloy, where the corresponding monotonic stress-strain curve is also potted for comparison.

3.5. Fatigue life modeling

In view of the fact that the majority of the failure of structural components is fatigue failure, more accurate model for predicting fatigue life is of vital importance. There are plentiful of models of fatigue life prediction based on stress, strain and energy parameters have been proposed [67]. One of the common models is to use Coffin-Mason-Basquin equation, as mentioned above. In general, the Coffin-Mason relationship is usually applied in LCF regime where the specimens are subjected to high strain and stress. Hence, the value of strain is more suitable than the stress in this situation, where the value of plastic strain amplitude (Δε_p/2) is required. It is more appropriate to use the Basquin’s equation in the high cycle fatigue (HCF) regime, since the stress level is below the elastic limit, being equivalent to elastic strain amplitude (Δε_e/2). Therefore, the incorporation of Coffin-Mason relationship and Basquin’s equation leads to the following equation for the total strain amplitude,

Δε_t/2=Δε_e/2+Δε_p/2=σ'_f (2N_f)^b/E+ε'_f (2N_f)^c, (5)

where σ'_f is the fatigue strength coefficient, N_f is the number of cycles to failure, b is the fatigue strength exponent (also referred to as Basquin exponent), ε'_f is the fatigue ductility coefficient and c is the fatigue ductility exponent and E represents the Young’s modulus with an average value of 127.8 ± 2.2 GPa for the present as-cast alloy as mentioned earlier. As seen from Fig. 9, Eq.(5) could be used to estimate the fatigue life fairly well.

View original graphic| Download| PPT slide

Fig. 9.   Predicted fatigue life from Equ.(5) based on Basquin’s equation and Coffin-Mason relationship, in comparison with the experimental data.

Energy-based fatigue damage models have also been used to predict the fatigue life, which involves the evaluation of some constants from energy-life curve. Park et al. [68] used the following plastic strain energy density (ΔW_p) as a function of the number of cycles to failure (N_f),

ΔW_p∙N_f^m=C, (6)

where m and C are two fitting constants. Nevertheless, the effect of mean stress during fatigue loading could not be taken into consideration in Equ.(6), which would lead to a weak life prediction. To solve this problem, Park et al. [68] provided a modified equation below,

ΔW_t∙N_f^m=C, (7)

where ΔW_t (total strain energy density) equal to ΔW_e (elastic strain energy density) add ΔW_p (plastic strain energy density). The values of ΔW_t of the present as-cast Ti₂AlNb-based alloy are calculated from the 1 st, 10th and mid-life cycle at different strain amplitudes varying from 0.2% to 1.0%, and shown in Fig. 10(a). It can be observed that ΔW_t increases with increasing strain amplitude. It has also been noted that the role of elastic strain energy density could be somewhat weakened during cyclic deformation; this could be taken into account by multiplying a weighting factor of 0.25 (a quarter of a cycle). Then the slightly adjusted total strain energy, referred to as the combined strain energy density (ΔW_comb(t)), could be expressed as [69,70],

ΔW_comb(t)=0.25ΔW_e+ΔW_p. (8)

View original graphic| Download| PPT slide

Fig. 10.   (a) Total strain energy density of the as-cast Ti2AlNb-based alloy versus total strain amplitude, and (b) the predicted fatigue life based on a combined (or weighted) total strain energy density approach.

Dallmeier et al. [69] and Mohammed et al. [70] reported that this weighting function is better for more accurately predicting fatigue life. The weighting function is indeed used to account for the mean stress sensitivity of the material during cyclic loading [65]. Hence, the weighted energy-based model (Equ.(8)) is used for predicting fatigue life of the present alloy. As can be seen from Fig. 10(b), the experimental data is in close agreement with the predicted line. Although the fatigue results are usually scattered, such a weighted energy-based model can be used for fatigue life prediction with fairly good accuracy, since the experimental data points are all situated in-between the 1.5x scatter band. It should be mentioned that the accuracy of the life prediction would be improved if the fatigue crack initiation life could be taken into consideration during modeling, because fatigue failure consists of two main stages: crack initiation and crack propagation.

3.6. Fractography

An overall view of fracture surfaces of samples fatigued at both lower and higher levels of strain amplitudes (i.e., 0.4% and 1.0%) is shown in Fig. 11(a) and (c), respectively. Fig. 11(b) and (d) show the corresponding fracture surfaces characteristics at a higher magnification. The crack initiation was observed to occur basically from the specimen surface, especially at the lower strain amplitude (Fig.11(a)). However, at the higher strain amplitude multiple crack initiation sites tend to be observed (Fig.11(c)), where the fracture surface demonstrates a chaotic wavy feature. Also, fatigue crack growth area is larger at the lower strain amplitude of 0.4% than at the higher strain amplitude of 1.0%, since crack growth lasted longer with a relatively slow growth rate. While the crack growth was mainly in the mode of facet formation, indicating strong crystallographic cracking along certain slip planes, like 2195 Al-Li alloy [71], nickel-based superalloys [72,73], and titanium single crystals [74], the rapid propagation region was rougher with river-like patterns because of the increased plastic deformation and the accelerated crack growth rate. As seen from the secondary-electron image at a higher magnification (Fig. 11(b)), fatigue striation-like features also appeared in the present as-cast Ti₂AlNb-based intermetallic compound, which were also observed in a recent high-cycle fatigue experiment of Ti₂AlNb-based alloy via stress-controlled rotating bending tests [75]. These fracture characteristics suggest that some degree of plastic deformation is present ahead of the crack tip during fatigue crack growth [76]. Fracture morphologies at the higher strain amplitude seemed to exhibit more brittle features with quasi-cleavage facets and tearing ridges as seen in Fig. 11(d).

View original graphic| Download| PPT slide

Fig. 11.   SEM micrographs of the as-cast Ti₂AlNb-based alloy tested at a strain amplitude of (a, b) 0.4% and (c, d) 1.0%, respectively, where (a) and (c) show an overall view of fracture surfaces of the specimens fatigued at 0.4% and 1.0%, respectively, while (b) and (d) show a magnified view of the surface surfaces near the crack initiation site.

The structure of the O-Ti₂AlNb phase is closely related to the Ti₃Al ordered hexagonal structure (D0_l9). The [100], 1/2[110] and ½ [1$\bar{1}$0] translations are almost equivalent in the O-Ti₂AlNb phase, which are similar to 1/3 < 1$\bar{1}$20> in the ordered hexagonal phase. To avoid the unwanted confusion, the [100] and the 1/2[110] (and ½ [1$\bar{1}$0]) translations are referred to as ‘a’ and ‘a*’ dislocations [25], respectively. The deformation mechanism of Ti₂AlNb-basd alloy at room temperature was reported by Nandy and Banerjee [77] as follows: the general microstructure of the as-cast Ti-22A1-27Nb (at.%) O-Ti₂AlNb phase after deformation could be represented by the activation of ‘a’ and ‘a*’ dislocations gliding on the (010) and the (1-10) planes at room temperature, respectively. In some cases, ‘a’ and ‘a*’ dislocations would be split on their glide planes into two super-partials to decrease energy by starting the slip systems [21,24,77]. This kind of slip of dislocations in O-Ti₂AlNb phase would provide the necessary condition for the formation of fatigue striations.

4. Conclusions

(1) The microstructure of the present Ti₂AlNb-based intermetallic compound is composed of abundant fine lamellar O-Ti₂AlNb phase which had a specific orientation relationship with B2 phase. This kind of crystallographic relationship showed a significant influence in the subsequent cyclic deformation characteristics.

(2) Stress-strain hysteresis loops were symmetrical at all levels of strain amplitudes. Cyclic stabilization basically occurred in the present alloy except slight cyclic hardening at higher strain amplitudes, which could also be reflected by the good overlap between the monotonic stress-strain curve and cyclic stress-strain curve.

(3) In comparison with other intermetallic-based alloys, the as-cast Ti₂AlNb-based alloy exhibited a superb fatigue resistance at all strain amplitudes due to the presence of profuse fine lamellar O-Ti₂AlNb phase, which possessed a superior combination of strength and plasticity along with different grain orientations.

(4) Fatigue life increased with decreasing strain amplitude, which could be described by the commonly-used Coffin-Mason-Basquin equation. The weighted total strain energy-based model could be used to better predict the fatigue life.

(5) Fatigue crack initiation occurred from the specimen surface. Crack growth was characterized mainly by crystallographic cracking mode along with fatigue striation-like features. The slip of dislocations in the plastic zone ahead of crack tip provided the necessary condition for the formation of fatigue striations due to the relatively good ductility of the Ti₂AlNb-based alloy.

Acknowledgment

This work was supported by the National Natural Science Foundation of China (NSFC) (Grant No. 51871168) and the Natural Sciences and Engineering Research Council of Canada (NSERC) in the form of international research collaboration. Y.L. Zhang thanks China Scholarships Council (CSC) for providing a PhD student scholarship. D.L. Chen is also grateful for the financial support by the Premier's Research Excellence Award (PREA), NSERC-Discovery Accelerator Supplement (DAS) Award, Canada Foundation for Innovation (CFI), and Ryerson Research Chair (RRC) program. The authors would also like to thank Q. Li, A. Machin, J. Amankrah and R. Churaman for their assistance.

Reference

View Option

• By original order
• By published year
• By cited within times
• By Impact factor