On the influence of test speed and environment in the fatigue life of small diameter nitinol and stainless steel wire

Abstract

To better understand the complex interplay of speed and environment on metals commonly used in implants, rotary bend fatigue tests were conducted on stainless steel and nitinol wires. A range of alternating strains was tested to create ε-N curves at two speeds (physiologic and accelerated) and in three environments (deionized water at body temperature, phosphate buffered saline at body temperature, and laboratory air at ambient room temperature). Results indicate that speed and environment can affect the observed fatigue life in nuanced ways. An electropotential monitoring technique was demonstrated to characterize fatigue crack growth which may be useful in future investigations.

1. Introduction

Patient safety and device performance are the main objectives of preclinical medical device testing. As part of a suite of preclinical testing, manufacturers are required to demonstrate that implanted medical devices can withstand cyclic mechanical loads expected to be seen by the implant during clinical use over its lifetime. Each device’s risk profile determines the required reliability as the potential patient risk from fracture can vary depending on the design and the anatomical location [1–4]. To demonstrate acceptable reliability, manufacturers perform preclinical fatigue testing in accordance with consensus standards and FDA Guidance Documents. To accomplish the testing in the laboratory setting, the experimental test frequency is accelerated and the environmental conditions are idealized.

Cold-drawn metallic wire is widely used in medical devices (endovascular grafts, braided stents, orthodontic archwires, septal occluders, etc.). As a part of medical device development, manufacturers often characterize the material’s durability using rotary bend fatigue tests to enable fatigue safety factor calculations and device design evaluations using whole device testing in conjunction with finite element analysis [5]. In general, preclinical fatigue testing seeks to characterize the fatigue life of the device’s finished material by imparting anatomical loading modes as part of a full durability evaluation often in conjunction with statistical analysis. A common fatigue test for wires subjected to cyclic bending is ASTM E2948 which describes a method to conduct rotary bend fatigue tests. The rotary bend fatigue test of wire is relatively simple to setup and has the benefit of allowing for highly accelerated test speeds which makes it well suited for research investigations. Regarding the test environment, in general, consensus standards and FDA Guidance Documents that describe fatigue tests for implanted medical devices suggest using a phosphate buffered saline (PBS) solution at 37 °C as the test environment in order to mimic the in vivo environment (e.g., ASTM F2477, ASTM F2942). The use of a liquid environment is especially important for nitinol because of the heat generated during the martensite-austenite phase transformation [6,7]. For ease of setup or other considerations, air or deionized water (DIW) are sometimes used as a test environment for fatigue testing of medical devices or materials. While it may be desirable to conduct fatigue tests as quickly as feasible and in a non-physiologic environment, it is important to understand how differences in speed and test environment might influence fatigue performance measured experimentally. Furthermore, since the publication of ASTM F3211 on the “Fatigue-to-Fracture” methodology in 2017, there has been an effort in the medical device community to assess the appropriateness of extrapolation from low cycle to high cycle fatigue. This is because the “Fatigue-to-Fracture” methodology employs ‘hyperphysiological’ loading to induce fatigue fractures in fewer cycles than the expected implant lifetime. This methodology is in contrast to fatigue testing at expected physiological loading where fractures are not anticipated, commonly referred to as “Test-to-Success,” to demonstrate safety and performance. If “Fatigue-to-Fracture” data are to be used to make or inform statistical estimates of fatigue life at higher cycle counts, then an improved understanding of the effect of test conditions on both low and high cycle fatigue would aid in that effort.

Given its unique thermomechanical properties, the influence of environment and test speed on nitinol fatigue has been investigated previously. In 1997, Tobushi et al. conducted rotary bend fatigue tests of nitinol wire in air and water at speeds varying from 100 to 1,000 RPM out to 10⁶ or 10⁷ cycles [8]. Tests in water were insensitive to speed, but faster tests in air had a reduced fatigue life that was attributed to an increase in specimen temperature during testing. Test environment was found to influence the fatigue life with shorter fatigue life observed in air as compared to water when the cycles to fracture were lower than approximately 10,000 to 100,000. In contrast, higher cycle fatigue tests conducted in air had a longer fatigue life when compared to the water environment. These differences in fatigue life were attributed to corrosion although no evidence of corrosion was provided. In 2004, Wagner et al. conducted tests of nitinol wire at speeds between 36 and 800 RPM in either air or a silicone oil bath out to 10⁶ cycles [7]. Similar to Tobushi et al., the authors found a decrease in fatigue life with faster test speeds in air. Tests conducted in silicone oil, however, did not show a dependence on rotational speed between 100 and 800 RPM. Norwich conducted rotary bend fatigue tests of nitinol wire at speeds between 500 and 5,000 RPM in 2014 in water out to 10⁶ cycles and similarly found no dependence of fatigue life on test speed [9]. In contrast, our laboratory resolved an increase in fatigue life with faster test speed for nitinol wire rotary bend fatigue tests conducted in DIW at speeds between 36 and 3,600 RPM out to 10⁶ or 10⁷ cycles and at alternating strains below 1.5% [10]. In 2015, we expanded testing to include stainless steel and cobalt chromium [11]. Nitinol (with alternating strain of 1.0%) and stainless steel and cobalt chromium (both with alternating strain of 0.5%) all showed an increase in fatigue life in DIW as test speed increased from 100 to 35,000 RPM at relatively low cycle fatigue of less than 200,000 cycles. Lastly, although the study was not undertaken to evaluate the effect of test speed on fatigue life, Jerina and Mitchell published relevant results of an ASTM interlaboratory study on the ASTM E2948 method [12] which was conducted with nitinol and cobalt-chromium alloy (35N LT) wires across 10 different laboratories. All tests were run until fracture occurred with no sample surviving longer than 300,000 cycles. Of the 9 laboratories that used DIW for their test environment, speeds ranged from 1,000 to 6,000 RPM. Similar to the results obtained in our laboratory, when grouping by speed, an increase in the fatigue life can be observed, although this was not specifically analyzed by the authors. For example, when grouping the 35N LT results by speed, the average fatigue life at an alternating strain of 0.5% at 1,000, 3,600, and 6,000 RPM was 64,984, 110,542, and 122,655 cycles, respectively. For nitinol tested at 1.0% alternating strain at 1,000, 3,600, and 6,000 RPM, the average fatigue life was 9,640, 11,023, and 12,300 cycles, respectively.

Altogether, however, none of these studies have characterized the entire ε-N (strain life) curve for commonly used medical device materials in a PBS environment. Furthermore, none of the studies highlighted above have fully examined high cycle fatigue beyond 10⁷ cycles which is critically important for medical devices that are commonly tested to cycle counts in the hundreds of millions. Since it may be the case that different mechanisms influence fatigue crack nucleation and propagation at low and high cycle fatigue life, it is important to understand how test environment and speed influence fatigue life under different conditions. In order to compare a simulated in vivo environment (PBS) with environments commonly used for in vitro testing (air or DIW) as well as to better understand the effect of test speed for all three environments, rotary bend fatigue tests of nitinol and stainless steel wires in all three environments at both a slow and an accelerated rate were conducted. Additionally, after initial results were obtained, supplemental experiments of nitinol in PBS were performed utilizing an electropotential monitoring technique to better understand differences between crack initiation and propagation between the slow and fast test speeds.

2. Methods

2.1. Materials

316LVM stainless steel (ASTM F138) and superelastic nitinol (ASTM F2063) wires were selected for this study as representative materials commonly used in medical implants. The stainless steel wires were 0.178 mm in diameter while the nitinol wires were 0.500 mm in diameter. The stainless steel had been subjected to 75.8% cold work while the nitinol was provided in an annealed state. The nitinol material was provided with a mechanically polished surface finish. These material selections allowed for a wide range of fatigue life to be tested ranging from low to high cycle fatigue with our equipment and fixtures. Tensile tests to failure were conducted on both materials at room temperature and at 37 °C in air using an actuator displacement rate of 2 mm/min. Stainless steel wires were pulled directly to failure whereas nitinol wires were pulled to approximately 6% strain, unloaded, and then pulled to failure similar to the method described in ASTM F2516. Nitinol wires were also characterized with digital scanning calorimetry to determine the austenite finish temperature, A_f, utilizing a method similar to ASTM F2004. These material characterization results are presented in Table 1 for stainless steel and Table 2 for nitinol. A sample size of n = 3 was used for each characterization test and all strains and stresses presented are engineering values. Fig. 1 shows a representative stress–strain curve for each material and test temperature.

Table 1.

Material properties for stainless steel wires organized by test temperature and shown as average value ± standard deviation.

	Room temperature	37 °C
Ultimate strength (MPa)	1,344.6 ± 24.5	1,316.6 ± 47.9
Yield strength (MPa)	1,189.9 ± 31.7	1,174.2 ± 47.4
Elastic modulus (GPa)	154.6 ± 0.9	151.4 ± 4.0
Elongation at failure (%)	2.3 ± 0.2	2.1 ± 0.1

Table 2.

Material properties for nitinol wires organized by test temperature and shown as average value ± standard deviation.

	Room temperature	37 °C
Ultimate strength (MPa)	1,375.9 ± 1.9	1,361.3 ± 2.1
Upper plateau stress (MPa)	473.8 ± 1.5	559.1 ± 13.7
Lower plateau stress (MPa)	142.8 ± 17.0	255.8 ± 5.6
Elongation at failure (%)	12.9 ± 0.1	12.6 ± 0.1
Austenite finish temperature (°C)	9.7 ± 0.8

Fig. 1.

Representative tensile test results for the stainless steel (top) and nitinol (bottom) materials. Room temperature results are shown with a solid line and 37 °C results are shown with a dashed line.

2.2. Rotary bend fatigue

Rotary bend fatigue tests of wire were conducted per ASTM E2948 in one of three environments: DIW at 37 °C, PBS (Fisher Scientific – Hampton, NH) at 37 °C, or room temperature air. The PBS environment had a nominal pH of 7.4 which mimics typical blood pH levels. Tests were conducted with either the Blockwise Engineering FTX or FTXH wire fatigue tester (Tempe, AZ). All nitinol wire tests were completed in an unguided configuration. Stainless steel tests in DIW and PBS were conducted in a guided configuration while the air tests were done in an unguided configuration since friction between wire and the guides in air may generate heat.

Tests were conducted at a speed of either 72 RPM (1.2 Hz) or 9,000 RPM (150 Hz) with the former meant to mimic an approximate in vivo cardiac loading rate (72 beats per minute) and the latter meant to mimic an accelerated rate that might be selected in order to complete high cycle fatigue testing in a more reasonable timeframe. The 72 RPM tests were run until fracture occurred or 10⁶ cycles were reached. The 9,000 RPM tests, on the other hand, were run until fracture occurred or until 10⁸ cycles were reached. For the stainless steel wires, imposed alternating strains varied from 0.3 to 0.96%. For the nitinol wires, imposed alternating strains varied from 0.6 to 2.0%. Similar to previous investigations [11,13], alternating strain, ε_a, was determined per Eq. (1) below where d is the wire diameter and R is the radius of curvature. For guided configuration tests the radius of curvature was taken from the center of the guide mandrel to the centerline of the bent wire. For unguided configuration tests, the radius of curvature was calculated analytically as described in Appendix X1 of ASTM E2948. A total of six wires were tested at each condition, i.e. alternating strain, speed, and environment, for a total of 180 stainless steel and 252 nitinol specimens.

$${\epsilon }_a=\frac{d}{2R}$$ (1)

2.3. Rotary bend fatigue with electropotential monitoring

Supplemental rotary bend fatigue tests of nitinol wire at 1.2% and 0.8% alternating strain were conducted per the method described by Sivan et al. 2017 [14]. A single, characteristic test was conducted at 72 RPM and 9,000 RPM in PBS only, because of the need for a conductive test environment. All electrochemical measurements were made with a three-electrode system using a saturated calomel electrode (SCE) as a reference electrode and a carbon rod as a counter electrode. The electrical connection to the rotating nitinol wire was maintained with a graphite clasp which was located above the PBS environment. All electrochemical measurements were made with a Gamry Interface 1000 potentiostat (Warminster, PA). The open circuit potential (OCP) of the nitinol wires was measured for approximately 60 s before the fatigue test begun and then throughout the test until just after wire fracture.

2.4. Microscopy

Images of the fracture surfaces of stainless steel and nitinol wires were taken with a JEOL JSM-3690LV (Tokyo, Japan) scanning electron microscope (SEM). Due to the large number of specimens, only a subset of the fractured specimens was imaged. The 0.68% alternating strain for stainless steel and 0.9% alternating strain for nitinol were imaged for all speeds and environments. For those two alternating strain conditions, a representative specimen with a fatigue life near the average of the group was selected from each speed and environment condition. Additionally, three nitinol wires tested at 9,000 RPM at 0.7% alternating strain were imaged. Specimens were selected to include one from each environment and to encompass the range of fatigue life at 0.7% alternating strain. The DIW, PBS, and air specimens had fractured at 36,362, 29,683,979, and 99,504,792 cycles, respectively.

2.5. Statistical procedures

A statistical fatigue strength model was fit to the data based on a simplified Coffin-Manson equation. The standard Coffin Manson model relates median strength $\overline{\epsilon }$ as a function of life, N,

$$\overline{\epsilon }\left(N\right)=A_{el}{\left(2N\right)}^b+A_{pl}{\left(2N\right)}^c$$ (2)

where parameters b and c are, respectively, the limiting slopes of the high cycle and low cycle segments on a log–log plot. The significance of parameters A_el and A_pl are the intercepts of the high cycle and low cycle segments on a log–log plot. Based on the modelling assumption that the materials in this study may demonstrate endurance limits, the Coffin-Manson equation was simplified with b = 0. In this case A_el becomes the endurance limit ε_∞

$$\overline{\epsilon }\left(N\right)={\epsilon }_{\infty }+A_{pl}{\left(2N\right)}^c$$ (3)

Fatigue data models used the strength distribution rather than the life distribution as given in Falk 2019 [15]. This method has benefits in terms of goodness of fit to fatigue data showing substantial changes in ε-N function slope, as is observed with nitinol and stainless steel. In addition, the method allows the statistical implementation of the Coffin-Manson function as well as other well known functions. Variation in strength was modeled as normal. The standard deviation of the strength distribution, σ, is the single parameter which attempts to bracket the data scatter in the observed lifetimes of samples tested at all strain levels.

To summarize, the four statistical model parameters used are: {ε_∞, A_pl, c, σ}. The first three define the median ε-N curve and the last defines the standard deviation of the strength normal distribution. The maximum likelihood estimate (MLE) is the choice of model parameters that maximizes the likelihood of the observed lifetime vs. strain data for each combination of material, environment, and test speed. These parameters were fit to the fatigue data using standard maximum likelihood estimation techniques implemented in MATLAB. This technique allows for analysis of fracture and censored data. The confidence intervals were determined for each parameter using the Profile Likelihood methodology, defining the confidence interval for each scalar parameter as the highest and lowest value which maintains negative log-likelihood within limits based on the chi-squared distribution. For the 95% confidence interval, the limit corresponds to 1.92 log-likelihood units.

3. Results

Fatigue life results for the stainless steel wires are presented in Table 3 and Figs. 2 and 4; nitinol wire results are in Table 4 and Figs. 3 and 5. Coefficients for the MLE curves in Figs. 2 through 5 based on Eq. (3) above can be found in Tables 5 and 6; it should be noted that the coefficients for the faster test speed were calculated twice. The first calculation was conducted with the fatigue life data censored at 10⁸ cycles per the experimental protocol. The second calculation was conducted with the fatigue life data censored at 10⁶ cycles to facilitate direct comparisons (i.e., Figs. 2 and 3) between the two test speeds since those had different run-out cycle counts. SEM images for both materials including all speed and environmental conditions are shown in Figs. 6 and 7. In some of the stainless steel fracture surfaces, especially PBS tested at 9,000 RPM, evidence of torsional fracture can be noted. These observations are similar to what we observed in an earlier study [11] and appear to be limited to tests conducted in a guided configuration with one wire end driven and the other wire end free.

Table 3.

Median fatigue life ± standard deviation of stainless steel wires for all test conditions.

Alternating Strain, εa	72 RPM			9,000 RPM
	DIW	PBS	Air	DIW	PBS	Air
0.96	2,656 ± 227	2,811 ± 517	3,844 ± 2,415	4,306 ± 238	2,102 ± 89	3,986 ± 1,943
0.68	6,464 ± 602	7,213 ± 544	6,238 ± 46,516	12,333 ± 1,273	7,014 ± 1,821	7,495 ± 311
0.49	14,080 ± 3,060	32,393 ± 4,432	27,319 ± 2,895	43,629 ± 2,417	26,907 ± 1,100	39,825 ± 3,420
0.4	84,793 ± 30,490	169,787 ± 30,099	120,057 ± 334,170	171,802 ± 48,236	60,820 ± 12,400	238,521 ± 60,876
0.3	1,000,000 ± 0	1,000,000 ± 0	1,000,000 ± 0	100,000,000 ± 51,346,853	100,000,000 ± 13,799,767	100,000,000 ± 0

Fig. 2.

Stainless steel ε-N curves grouped by environment. Blue triangles represent fractures at 72 RPM and orange Xs fractures at 9,000 RPM. The solid blue line is the MLE curve for 72 RPM and the dashed orange line is the MLE curve for 9,000 RPM. Run-outs are indicated with an arrow on the right and the number of run-outs at each condition.

Fig. 4.

Stainless steel ε-N curves grouped by test speed. Blue circles represent fractures in DIW, orange diamonds fractures in PBS, and gray squares fractures in air. The dotted blue line is the MLE curve for DIW, the solid orange line is the MLE curve for PBS, and the dashed gray line is the MLE curve for air. Run-outs are indicated with an arrow on the right and the number of run-outs at each condition.

Table 4.

Median fatigue life ± standard deviation of nitinol wires for all test conditions.

Alternating Strain, εa	72 RPM			9,000 RPM
	DIW	PBS	Air	DIW	PBS	Air
2.0	762 ± 75	774 ± 31	1,020 ± 91	1,352 ± 304	1,587 ± 220	1,126 ± 190
1.2	3,107 ± 222	3,694 ± 471	6,828 ± 578	5,417 ± 554	4,795 ± 175	6,959 ± 703
1.0	5,927 ± 408	7,914 ± 567	15,341 ± 1,141	10,953 ± 890	8,884 ± 624	16,041 ± 2,673
0.9	10,068 ± 825	15,435 ± 1,023	23,697 ± 4,817	15,504 ± 2,234	15,992 ± 1,378	21,256 ± 1,751
0.8	15,747 ± 2,594	31,226 ± 18,827	48,678 ± 494,567	25,936 ± 625	23,172 ± 6,350	29,665 ± 2,694
0.7	1,000,000 ± 341,989	1,000,000 ± 392,398	1,000,000 ± 0	55,141 ± 8,927,357	18,920,441 ± 11,236,586	52,451 ± 40,601,500
0.6	1,000,000 ± 0	1,000,000 ± 0	1,000,000 ± 0	100,000,000 ± 0	100,000,000 ± 7,875,174	100,000,000 ± 0

Fig. 3.

Nitinol ε-N curves grouped by environment. Blue triangles represent fractures at 72 RPM and orange Xs fractures at 9,000 RPM. The solid blue line is the MLE curve for 72 RPM and the dashed orange line is the MLE curve for 9,000 RPM. Run-outs are indicated with an arrow on the right and the number of run-outs at each condition.

Fig. 5.

Nitinol ε-N curves grouped by test speed. Blue circles represent fractures in DIW, orange diamonds fractures in PBS, and gray squares fractures in air. The dotted blue line is the MLE curve for DIW, the solid orange line is the MLE curve for PBS, and the dashed gray line is the MLE curve for air. Run-outs are indicated with an arrow on the right and the number of run-outs at each condition.

Table 5.

Strain-life coefficients for stainless steel wires.

	MLE parameters				106 cycle fatigue strength (% strain) 95/90 C/R
	ε∞	Apl	c	σ	Lower bound	MLE	Upper bound
72 RPM, DIW	0.3400	445.3256	−0.7674	0.0330	0.2887	0.3043	0.3169
72 RPM, PBS	0.2932	30.9966	−0.4514	0.0361	0.2763	0.2913	0.3060
72 RPM, Air	0.3262	134.2905	−0.5989	0.0986	0.1496	0.2224	0.2657
9,000 RPM, DIW	0.2979	89.2736	−0.5398	0.0200	0.3000	0.3076	0.3149
9,000 RPM, PBS	0.3021	42.4522	−0.4979	0.0248	0.2911	0.3012	0.3129
9,000 RPM, Air	0.3426	162.8000	−0.6194	0.0686	0.2580	0.2750	0.2984
9,000 RPM, DIW with 106 Cycle Censor	0.2656	56.9705	−0.4857	0.0183	0.2846	0.2916	0.2990
9,000 RPM, PBS with 106 Cycle Censor	0.2594	27.5488	−0.4392	0.0222	0.2682	0.2780	0.2893
9,000 RPM, Air with 106 Cycle Censor	0.3221	105.0250	−0.5678	0.0690	0.2191	0.2614	0.2871

Table 6.

Strain-life coefficients for nitinol wires.

	MLE parameters				106 cycle fatigue strength (% strain) 95/90 C/R
	ε∞	Apl	c	σ	Lower bound	MLE	Upper bound
72 RPM, DIW	0.6655	163.0978	−0.6591	0.0481	0.5935	0.6152	0.6400
72 RPM, PBS	0.6864	122.6414	−0.6173	0.0313	0.6457	0.6620	0.6801
72 RPM, Air	0.6811	76.4316	−0.5325	0.0517	0.6253	0.6486	0.6748
9,000 RPM, DIW	0.5975	165.8111	−0.6034	0.0811	0.4826	0.5197	0.5573
9,000 RPM, PBS	0.6189	267.5954	−0.6618	0.0676	0.5205	0.5503	0.5815
9,000 RPM, Air	0.5613	60.2614	−0.4841	0.0692	0.4928	0.5263	0.5610
9,000 RPM, DIW with 106 Cycle Censor	0.5948	164.5788	−0.6021	0.0824	0.4825	0.5156	0.5520
9,000 RPM, PBS with 106 Cycle Censor	0.6763	518.3436	−0.7489	0.0701	0.5730	0.5963	0.6238
9,000 RPM, Air with 106 Cycle Censor	0.5327	53.0413	−0.4650	0.0666	0.4796	0.5097	0.5450

Fig. 6.

SEM images of fractured stainless steel wires with alternating strain of 0.68%.

Fig. 7.

SEM images of fractured nitinol wires with alternating strain of 0.9%.

4. Discussion

A myriad of factors is known to affect the fatigue of metals at low and high cycle life and the relative influence of these factors can vary depending on the metal being tested as well as the environmental conditions. Competing phenomena such as crack closure, strain rate sensitivity, corrosion assisted fatigue, and others may occur simultaneously and lead to surprising results in some cases [16–20]. In this study, we attempted to evaluate the effects of test speed and environment on two metals commonly used in implants with a rotary bend fatigue loading mode. As such, we were unable to evaluate mean strain effects that are important for in vivo loading of implanted medical devices [5]. While we expect our primary conclusions as described below to be independent of mean strain effects, we recognize that some effects from test environment or speed could be different under non-zero mean strain conditions. Thus, we would recommend future work include a focus on non-zero mean strain states given the relevance to complex in vivo loading. A large number of experiments were conducted in an attempt to elucidate whether test speed or environment acted in isolation or whether there were combinatorial effects where fatigue life was only affected for particular combinations of factors. Given the large dataset including two metals, three environments, and two test speeds, we have organized the sub-sections of the discussion to focus on areas where differences in fatigue life were noted.

In our study it was important to use small diameter wire specimens to have relevance to medical implants which tend to be rather small as compared to other engineering domains like automotive or aerospace. For relatively small test specimens such as ours, it is important to consider that the bulk of the fatigue life may be spent in crack initiation and not in crack propagation [13]. Rahim et al., for example, showed that for nitinol wires of size similar to the ones we used, the number of loading cycles spent in fatigue crack propagation never exceeded 3,000 even as the total cycles to fracture went beyond 10⁶ cycles [21]. Relatedly, when studying nitinol crack propagation, Robertson and Ritchie found that the crack growth rate da/dN was relatively insensitive to frequency in a saline solution (1 Hz vs. 50 Hz) and that da/dN in air was also similar to what they measured in saline [22]. Thus, in our study, we expect that most differences in fatigue life noted are likely due to differences in crack initiation, although as alternating strain increases and fatigue life decreases, the proportion of the fatigue life spent in crack propagation may increase, so differences in crack propagation could manifest in test conditions with lower cycles to fracture.

4.1. Role of test speed

When examining the ε-N curves for both materials in Figs. 2 and 3, one of the most prominent trends in the data is an increase of fatigue life at the faster test speed in DIW at higher alternating strains. This can be seen clearly as a separation of the MLE curves for the slower and faster speeds above alternating strains of ~0.5% in stainless steel and ~0.8% in nitinol. At lower alternating strains, in contrast, the MLE curves for both materials seem to converge as the run-out cycle count is reached. Although this trend was in line with some of our previous work on rotary bend fatigue in DIW [10,11] and an analysis of the ASTM E2948 interlaboratory study [12] as described above in Section 1, it was unexpected to not see a similar trend for test speed in PBS. While determining the underlying cause of this behavior lies outside the scope of this work, we hypothesize that the observed trends in DIW and PBS might be caused by preferential dissolution of metal ions at slip bands [23]. Since the dissolution of ions is a time dependent phenomenon, it may be that the slower test speed allowed for more dissolution at slip bands which accelerated fatigue crack nucleation relative to the faster tests in DIW. Thus, a smaller number of cycles would be needed to nucleate a fatigue crack in DIW at the slower speed which may have manifested in our results with both materials showing ‘reduced’ fatigue life at slower test speeds. Similar to our observations here, in a study which examined the fatigue performance of AISI 347 stainless steel in air and several aqueous environments, preferential dissolution at slip bands was proposed as a controlling mechanism for fatigue crack initiation and to explain the reduction of fatigue life in aqueous environments as compared to air [24]. Additionally, it could be that the reason this trend with test speed was observed in DIW, but not PBS, is due to the relative rate of ion release in each of those environments. Unrelated work in our laboratory on the release of nickel ions from nitinol in different environments [25] found that nitinol in DIW tended to release more nickel ions that nitinol in PBS suggesting greater surface activity. In contrast, another study of nickel ion release in orthodontic wires completed under static and dynamic conditions in either water or saline did not find a difference in nickel ions released for the two environments [26]. This seeming contradiction could be attributable to differences in surface finish and methods including the timescale of experiments. Both studies examined nickel ion release over the course of days whereas the fatigue experiments in our study which showed a difference in fatigue life between speeds for DIW were completed in a matter of minutes or hours. Thus, it could be the case that preferential dissolution of ions occurs more readily in DIW than in PBS over the timescale of our fatigue experiments. In that case it would be logical to conclude that the slowest test speed in DIW would have the shortest fatigue life if preferential dissolution at slip bands is the controlling mechanism for fatigue crack nucleation. However, as stated above, the confirmation of this hypothesis lies outside the scope of the present study and furthermore other mechanisms including differences in repassivation kinetics or oxide-induced crack closure could be responsible for the observed differences in fatigue life. Oxide-induced crack closure, however, might be expected to have minimal influence since it only affects crack growth and not crack initiation [27] and since we expect that the majority of the measured fatigue life was spent in initiation and not growth in our experiments. Overall, we would recommend future efforts to better understand the observed differences in fatigue life in DIW and the underlying cause(s). In particular, it would be interesting to examine why this effect seems to diminish as the alternating strain is reduced. We suspect it could be that different mechanisms exert more control over the fatigue crack initiation process as the alternating strain is reduced and the material approaches run-out. In general, test speed was not found to have a large effect on fatigue performance in either PBS or air, although there were some minor differences observed with the nitinol material that are discussed below in Section 4.3.

4.2. Role of environment

In general, the DIW and PBS environments seemed to result in similar fatigue performance as evidenced by their near overlap in the ε-N curves in Figs. 4 and 5. For both materials tested at the slower test speed, the fatigue life in air was larger than when tests were conducted in DIW or PBS. Given the potential for corrosion fatigue in DIW or PBS at the slower test speed, this seems reasonable and in line with previous investigations of nitinol and stainless steel [20,28–30]. At the faster test speed in air, the nitinol likely experienced some self-heating which is discussed in more detail in Section 4.3. Stainless steel when tested at the faster speed, on the other hand, seemed less sensitive to environment with a fair amount of overlap in fatigue life. The only exception to that might be the highest alternating strain condition where the MLE curve for PBS seems shifted towards lower fatigue life as compared to DIW or air. This could either be due to a corrosive effect of the PBS environment or possibly an anomalous observation due to the relatively small standard deviation of the data at that condition. The standard deviation was 89 cycles in PBS vs. 238 cycles in DIW or 1,943 cycles in air, which could have affected the statistical modelling of the dataset. Lastly, for stainless steel tested in air at the slower speed, several outliers were observed that had a much higher fatigue life than other specimens with the same alternating strain (see top portion of Fig. 4). It is unclear if this is an anomalous observation or if something unique to that test condition was leading to the presence of the outliers and an increased variability in fatigue life.

Fig. 8 highlights the fractured surfaces from nitinol specimens tested in each environment at the faster test speed. Even though the specimens had drastically different fatigue lives, the morphology of the fractured surfaces appear similar. In all cases, the maximum crack depth was between 241 and 255 μm. Since crack growth in nitinol has been shown to not depend strongly on environment [22], this suggests that the differences in cycles to fracture were due to differences in the number of cycles to nucleate the fatigue crack.

Fig. 8.

SEM images of fractured nitinol wires with alternating strain of 0.7% tested at 9,000 RPM in different environments: DIW (top), PBS (middle), and air (bottom). The fatigue life of each specimen, from top to bottom, was 36,362, 29,683,979, and 99,504,792 cycles.

4.3. Nitinol-specific phenomena

Previous investigations of nitinol fatigue life in air have found a clear reduction in fatigue life for tests conducted at faster test speed because of the heat generated during the pseudoelastic martensite-austenite phase transformation [7,8]. This trend, however, was not observed in our results, as can be noted in Fig. 3 where the number of cycles to fracture in air for the slow and fast test speeds between approximately 0.8% and 2.0% alternating strain seem to overlap greatly. We suspected that this surprising result was due, at least in part, to a fan inside the Blockwise Engineering wire fatigue tester which is meant to keep the electronic components cool during operation. Since the exhaust from the electronics cooling fan exits onto where the rotating wire is located, we decided to repeat experiments in air at the faster test speed for several alternating strain conditions, but with that opening blocked to prevent any airflow from cooling off the specimen. Results from these repeat tests (n = 6 for each strain condition) along with the original nitinol air tests are shown in Table 7. At the 2% alternating strain, a clear and statistically significant reduction in cycles to fracture is observed when compared to the original results (p < 0.05 from T-test). This is in line with expectations since the 2% alternating strain condition would be expected to generate the most heat. The 1.2% and 0.8% conditions, in contrast, did not demonstrate any difference in cycles to fracture with the airflow blocked. From this, we can surmise that there was no effect from the fan on the nitinol fatigue results in air at strains below approximately 1.2%. Since the nitinol tensile tests in Fig. 1 demonstrated that the stress plateau began around 1% strain in tension, it seems reasonable to conclude that only minimal phase transformation was taking place at the 0.8% and 1.2% alternating strain conditions in our tests and, for that reason, heat generation was not a factor that greatly influenced the results in air. It is also worth noting that for our relatively thin test specimens (0.5 mm diameter nitinol wire), air cooling with room temperature air was sufficient to dissipate the heat caused during phase transformation such that the original slow and fast test speed experiments resulted in similar fatigue life at 2% alternating strain. The same may not be true, however, for thicker specimens where airflow may be insufficient to dissipate any heat generated.

Table 7.

Median fatigue life ± standard deviation of original and supplemental nitinol tests in air.

	Air, 72 RPM	Air, 9,000 RPM	Air, 9,000 RPM without airflow
2.0	1,020 ± 91	1,126 ± 190	752 ± 44
1.2	6,828 ± 578	6,959 ± 703	6,720 ± 393
0.8	48,678 ± 494,567	29,665 ± 2,694	33,901 ± 3,108

It should also be noted that the nitinol MLE curves in air for 72 and 9,000 RPM seem to diverge below ~ 0.8%. Whereas 6 of 6 specimens tested at 0.7% strain for the slow speed reached 10⁶ run-out without fracture, 5 of 6 specimens tested at the faster speed fractured before 10⁶ cycles. Given the small size of our specimens, we would expect that this difference is due to differences in fatigue crack initiation and not in fatigue crack growth. Relatedly, Fitzka et al. recently examined the mechanical response of nitinol with a novel X-ray diffraction synchrotron experiment at low frequency (0.1 Hz) and ultrasonic frequency (20 kHz) and found that the same deformation mechanisms took place at both frequencies [31]. Specifically, the authors observed stress induced phase transformation from austenite to martensite through an intermediate R-phase. Specimen temperature was controlled in their experiments by means of pulsing and pausing the ultrasonic loading in combination with forced-air cooling. Since our experimental speeds were well within those examined by Fitzka et al., it would stand to reason that the same mechanical deformation processes would occur at the slow and fast speeds in the present study and thus it is difficult to determine a specific mechanism for this apparent reduction in fatigue life. Since the alternating strain condition of 0.7% would not be expected to generate bulk cyclic phase transformation during loading, it could be that localized heating is taking place due to small regions undergoing phase transformation, possibly near inclusions, which led to accelerated crack nucleation at the faster test speed. However, we did not confirm this in our study and would recommend further investigations into the specific mechanism(s) in the future. Lastly, it should be noted that while the DIW and PBS experiments were conducted at 37 °C, all experiments in air were done at room temperature which was slightly cooler. All else being equal, we would expect nitinol fatigue tests conducted at cooler temperatures to have longer fatigue lives [8,13]. However, given the complicated interplay between environment and speed discussed above, it is difficult to attribute any particular shift in fatigue life between air and the PBS or DIW environments solely to the difference in temperature between the experiments. Nevertheless, if our nitinol fatigue experiments in air were to be repeated at 37 °C, we might expect to see a downward and/or leftward shift in the ε-N curve as compared to our present results.

In Fig. 3, a divergence between test speeds in PBS can be observed in the nitinol wires tested at higher alternating strains. Interestingly, this divergence was not noted at lower alternating strains or with the stainless steel experiments. We hypothesize this is due to the unique properties of nitinol and specifically that the phenomenon may be explained by the fatigue damage mechanism proposed by Racek et al. [29]. Summarizing their proposed mechanism, the fracture of nitinol’s surface oxide leads to a network of cracks which open and close periodically exposing the underlying nitinol to the test environment. Hydrogen is then generated during the passivation reaction which moves into the bulk nitinol leading to embrittlement, loss of strength, and eventually propagation of fatigue cracks leading to fracture. Since the passivation reaction is a time-dependent process, we suspect that at our slower test speed more time per loading cycle would allow for more hydrogen to penetrate into the bulk nitinol and accelerate the fatigue process as compared to the faster speed. Furthermore, that this divergence between test speed was only observed at higher alternating strains where more surface oxide cracks would be likely to form, suggests that Racek’s proposed fatigue mechanism for nitinol in aqueous environments may help explain our results.

In order to better understand the fatigue mechanisms taking place in PBS at both the slow and fast test speeds, supplemental nitinol fatigue tests were conducted at 0.8% and 1.2% alternating strain. For these supplemental tests, we used an electropotential monitoring method developed in our laboratory for rotary bend fatigue [14] which is similar in nature to what other investigators have utilized [29,32,33]. Results from these experiments are provided in Table 8 and Fig. 9. Immediately apparent from the OCP vs. cycles in Fig. 9 is the initial drop in potential as the experiment is started which is presumed to be due to the fracture of the surface oxide layer and reaction of the bulk nitinol with the PBS environment. Also apparent is that the drop in potential depends both on the alternating strain and the test speed with the faster test speed and higher alternating strain resulting in a larger OCP drop. In all cases investigated herein, as the metal begins to repassivate, the OCP begins to recover to a local maximum and then begins to drop off again until final fracture of the wire. In a related study, Racek et al. found a reduction in wire strength when fatigue tests were prematurely stopped after the OCP maximum and before final fracture [29] suggesting that once the OCP begins to drop again, the fatigue crack has begun to propagate further into the bulk material. If the local maximum does represent approximately when the fatigue crack has begun to propagate more deeply into the bulk material, our results are somewhat in line with those from Rahim et al. [21] wherein the authors counted fatigue striations in SEM to measure how many loading cycles were needed to propagate the fatigue crack to final fracture. In that study, the authors measured the number of cycles to propagate to be between approximately 1,500 and 3,000 for alternating strains between approximately 0.8% and 1.8% with a relatively small range (plus or minus a few hundred cycles) for each alternating strain. Similarly, as seen in Table 8, we find that the number of cycles between local maximum and fracture is in the same range as was measured for crack propagation by Rahim et al. In contrast, the cycle count at local maximum OCP, which might be thought of as a proxy for the crack initiation phase, shows greater variability than that of the cycle count between local maximum and fracture. Our results also suggest that for lower alternating strain conditions where the fatigue life is greater than approximately 10,000 cycles, the majority of the fatigue life is spent in the fatigue crack nucleation phase and not in the fatigue crack growth phase [13]. In contrast, when the alternating strain is higher and the fatigue life is less than approximately 5,000 cycles, the number of cycles spent nucleating and propagating fatigue cracks may be more evenly split.

Table 8.

OCP data from nitinol tests at 1.2% and 0.8% alternating strain.

	1.2%, 72 RPM	1.2%, 9,000 RPM	0.8%, 72 RPM	0.8%, 9,000 RPM
Starting OCP (mV)	−263.0	−200.6	−46.6	−47.3
Local Minimum OCP (mV)	−347.5	−524.2	−60.5	−172.8
Cycle Count at Local Minimum OCP	54	534	84	686
Local Maximum OCP (mV)	−316.2	−431.2	6.2	−148.1
Cycle count at Local Maximum OCP	968	2,211	25,894	16,089
OCP at Fracture (mV)	−481.4	−574.5	−87.1	−360.4
Cycle Count Between Local Maximum and Fracture	2,898	2,401	3,321	5,615
Cycles to Fracture	3,866	4,612	29,215	21,704
Test Duration (sec)	3,222	31	24,346	145

Fig. 9.

OCP signal from nitinol fatigue tests at 1.2% (left) and 0.8% (right) alternating strain in PBS. The solid blue line represents 72 RPM test speed and the dashed orange line 9,000 RPM test speed. Black dots indicate the locations of local minimum and maximum as described in Table 8.

4.4. Fatigue strength calculations and the statistical significance of test speed and environment on reliability

For many implants, testing in PBS at the slow speed would provide the most relevant estimation of expected fatigue performance. However, time considerations necessitate the use of accelerated frequencies and experimental practicalities may require the use of DIW, air, or some other non-physiologic test environment. Fatigue datasets, like the ones collected here, are often used to make statistical comparisons to test the presence of a significant difference in reliability, for example, when evaluating potential effects of a manufacturing change or new material supplier on implant durability. To help ensure credibility of those results, it is therefore important to understand whether test speed or environment could affect the predicted fatigue strength for each material. Using the statistical methods described above, we calculated confidence intervals around the estimated fatigue strength for each material, speed, and environment with fatigue strength defined as the alternating strain with 90% reliability at 10⁶ cycles. Fig. 10 illustrates fatigue strength MLE values as well as the 95% upper and 5% lower confidence bounds. For the faster test speed, two fatigue strengths were calculated for each dataset – the first with run-outs censored at 10⁸ cycles and the second with run-outs censored at 10⁶ cycles. This allows for direct comparisons between test speeds for a particular environment, but censors the fractures that were observed to occur between 10⁶ and 10⁸ cycles in our testing. This technique also allowed us to investigate the statistical implications of data censoring on the estimated fatigue strength depending on the particular material and environment combination.

Fig. 10.

Estimated 10⁶ cycle fatigue strength with 90% reliability for stainless steel (top) and nitinol (bottom). The MLE 10⁶ cycle fatigue strength is shown in black while the colored bars represent the 90% confidence interval. DIW results are shown in blue, PBS in orange, and air in gray. The red diamonds indicate the lowest alternating strain where at least one fracture was observed prior to run-out at each condition.

Although, as discussed above, in some cases the low cycle data exhibited differences among test speeds and environments, the high cycle regime appeared less sensitive to these variables with many of the MLE curves in Figs. 2–5 converging at higher numbers of cycles. As seen in Fig. 10, stainless steel 10⁶ cycle fatigue strength remained largely independent of test speed and environment. An increase in data scatter in the air environment had the effect of reducing the 90% reliability fatigue strength even though the median fatigue behavior is almost identical to the other two environments. Nitinol, on the other hand, demonstrated some small differences in the 10⁶ fatigue strength of approximately 0.1% or less as the test conditions were varied. For nitinol tested at the faster test speed in any environment, the predicted 10⁶ fatigue strength was slightly reduced when compared to the slower test speed. Overall, since the stainless steel fatigue strength was relatively insensitive to test speed and since the nitinol tests at faster speed resulted in slightly more conservative estimated fatigue strength values, the data and analysis herein are supportive of the use of accelerated test frequencies during preclinical testing phases of device development for implanted medical devices. Interestingly, comparing across the three environments at the faster test speed, the estimated fatigue strength values for each material overlap highly suggesting that fatigue tests at faster speed are relatively insensitive to environment. This, however, may not be true for other materials and test conditions or for fatigue strength at higher cycle counts that may be more relevant for implanted medical devices.

Fig. 10 also highlights the importance of understanding the performance of statistical models for the given material prior to making fatigue life predictions. For stainless steel, the censoring of 9,000 RPM data to 10⁶ cycles resulted in a more conservative estimate of fatigue strength for all test environments. However, for nitinol, censoring had the effect of either increasing, decreasing, or no effect depending on which environment was being considered. Comparing the effect of data censoring on the DIW and PBS nitinol results provides an informative example of the statistical model performance. Censoring the DIW data resulted in the removal of two fracture datapoints between 10⁶ and 10⁸ cycles whereas censoring the PBS data resulted in the removal of seven such fractures. This disparity in how many fracture datapoints were censored may explain why the estimated fatigue strength for nitinol tested at 9,000 RPM in DIW did not change with data censoring whereas the estimated fatigue strength for nitinol tested at 9,000 RPM in PBS increased with data censoring. Other statistical models might be more or less sensitive to data censoring, but in the context of ASTM F3211, it is critical to understand the implications of using fatigue data censored at 1 million cycles to predict durability for device lifetimes which are commonly at 400 million cycles or beyond. In our analysis, we made an effort to avoid extrapolation since it appeared that subtle differences in the ε-N curve shape could affect high cycle fatigue predictions.

5. Conclusions

Overall, this work presents a thorough examination of the interplay between test environment and speed on two commonly used implant materials. Some of the most important conclusions from this work include:

• When tested in DIW, low cycle fatigue in both stainless steel and nitinol is dependent on test speed with faster tests resulting in longer fatigue life.

• The effect of corrosion fatigue seems to be stronger at slower test speeds as evidenced by the increase in fatigue life seen in air as compared to DIW or PBS.

• Tests conducted at the accelerated speed resulted in similar or slightly more conservative estimated 10⁶ cycle fatigue strength as compared to tests conducted at the slower speed.

• Electropotential monitoring is a powerful tool for characterizing the fatigue process and may assist in distinguishing between the crack initiation and propagation phases.