Metallographic autopsies of full-scale ITER prototype cable-in-conduit conductors after full testing in SULTAN: 1. The mechanical role of copper strands in a CICC

The ITER magnet system will be the largest superconducting magnet system in the world and with a stored magnetic energy of up to 51 GJ (compared to 11 GJ for the LHC) inside a building as large as the Jefferson memorial in Washington DC, it presents many challenges that have been studied extensively [1]. The possibility of nuclear heating and other electromagnetic (EM) disturbances expected during plasma operation puts a high demand on the cooling system of the superconducting magnets and therefore a cable in conduit conductor (CICC) design was chosen to effectively remove heat [2]. The magnet system will be comprised of four coil systems [3]: the toroidal field (TF) coils, the central solenoid (CS) coils, the poloidal field (PF) coils and the correction coils (CC). The PF and CC systems will have magnetic fields low enough to operate with Nb–Ti, which is the most robust and inexpensive superconductor material available [4]. The TF and CS coils on the other hand, must generate magnetic fields of 11.8 and 13 T respectively, and therefore require higher critical field superconductors, of which Nb₃Sn is by far the most economical [5]. Figure 1 shows photographs of the TF and the CS conductors showing the composite Nb₃Sn strands and copper strands coming out of the steel, force-support and He containment jacket. The cables are fully transposed in order to ensure uniform current distribution.

Zoom In Zoom Out Reset image size

The use of Nb₃Sn presents an important challenge for the CICC design, mainly because this A15 structured intermetallic is much more sensitive to strain than ductile Nb–Ti [6]. Therefore reversible and irreversible strain effects may degrade the performance if the mechanical effects of the Lorentz force are not properly controlled. The first evidence of uncontrolled degradation in an ITER conductor was found during the testing of the model coils in 2000 [7] when the performance degraded due to strain accumulation as the coils were energized on and off. Although this degradation was a significant concern, it appeared that the conductor performance would stabilize after about 2000 EM cycles [8].

The fact that the TF coils are to be made using strands made by eight different Nb₃Sn suppliers under ITER contract with different strand architectures makes the CICC production even more challenging. Therefore, it was required to validate all conductor designs before coil fabrication [9] by full qualification tests at the SULTAN (SUpraLeiter TestANlage) testing facility in CRPP, Switzerland [10, 11] where short full-size ITER conductors can be tested in a magnetic field generated by three concentric pairs of superconducting split coils that can provide a homogeneous magnetic field of 10.8 T within 2% over a length of ∼410 mm [12]. With this magnet at full field, the CICC current can then be ramped up, to match the transverse Lorentz force produced in ITER operation, and subsequently ramped down. Such event is called an EM cycle. In addition to EM cycling SULTAN also allows the temperature to be raised to find the current-sharing temperature, T_cs, which is used to qualify the conductor performance at a given EM cycle. A TF CICC qualification run consists in a thousand EM cycles with an occasional T_cs check, and this is done until T_cs stabilization is achieved or the critical number of coil energization cycles is exceeded [13] while still providing an acceptable T_cs [1]. By 2010 all the TF conductor designs were validated and despite the fact that some of them did not reach stabilization, all the parties met the requirement of a T_cs, higher than 5.8 K after 1000 cycles [14], which is the life expectancy of the TF coils [3]. Accordingly, production of the TF conductor could proceed [9]. Interestingly enough, the TF qualification process showed a wide variety of degradation behavior despite the fact that some conductors were technically identical [15]. There were also a few conductors showing very little sensitivity to cycling [16] which will be an issue addressed in the third article of this study.

The CS coils on the other hand must withstand 60 000 cycles during the 30 000 inductively driven plasma pulses expected from the ITER machine [1, 2, 9], and they therefore present a much greater challenge than the TF coils. In 2010 the first CS prototype, denominated CSJA1, was tested in SULTAN and the T_cs fell below the required limit quite rapidly [9, 17]. The autopsy performed on this conductor [17] observed a significant permanent displacement of the strands in the direction of the Lorentz force, leading to the conclusion that the CS design was not optimized against the transverse mechanical effects of the Lorentz force under extended cycle testing. The second CS prototype tested used a '3rd generation' bronze process strand with improved J_c at the cost of a slight increase in hysteresis losses [18]. This design made a slight improvement in the conductor behavior, but it did not show signs of T_cs stabilization even after 10 000 cycles [9, 17, 19].

These results raised broad general concerns about the lifetime performance of the CS coils and, in response, a comprehensive R&D program was launched by the ITER International Organization with the support of the US ITER Project Office and the Japan Atomic Energy Agency as the ITER Japan Domestic Agency, as well as Oxford Superconducting Technologies in Carteret, New Jersey, who donated a significant amount of superconducting strand for the production of new CS prototype cables [1, 9]. The task called for a collaboration across many laboratories around the world [9] and a variety of techniques including very sophisticated finite element analysis [20], mechanical testing [21, 22], metallographic analysis [23] and magnetic susceptibility measurements [24]. The program aimed to find the influence of the following conductor parameters: the strand architecture (internal tin versus bronze process), the first triplet configuration (the presence of copper strands) and the cabling twist pitch [19].

This paper compares two cables, CSIO1-2 and CSIO1-1. The former had both Cu and superconducting strands and served as a baseline for internal tin conductor performance, while the latter used only superconducting strands with an increased Cu to non-Cu ratio of 1.5:1 that resulted in an overall superconducting area increase of about 20% [19] in an attempt to offset the degradation as much as possible. In addition, the design with no Cu strands (CSIO1-1) has the advantage of reducing the force per strand from 919 to 613 N m⁻¹ and it was hoped that this significant reduction in strand-level Lorentz force would reduce the strand peak strain under SULTAN testing. The superconductor-only design of CSIO1-1 was also thought to have the benefit that the composite strands would be stiffer than the pure Cu strands [22] and thus should have more rigidity than CSIO1-2.

The conductor design parameters are listed in table 1. All testing conditions, sample preparation and conductor parameters were essentially identical for both CSIO1-1 and baseline CSIO1-2 with the major exception being the design of the first triplet configuration (Cu strands or not) and the Cu to non-Cu ratio of the strands.

The two different strands of these prototype CICCs are shown in figure 2, while the CICC's T_cs degradation profiles are shown in figure 3. The higher initial T_cs of the CSIO1-1 CICC is consistent with its higher overall superconducting area (∼40% superconductor in CSIO1-1, compared to ∼33.3% in CSIO1-2). It is worth noting that approximations of T_cs(1)–T_cs(∞) through data fitting [19, 25] show that the T_cs degradation is lower for the CSIO1-1 sample indicating a more robust conductor versus cycling. In this study we have made a detailed quantitative metallographic study to try to understand the machanism for degradation in the two conductors.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The structural complexity of the samples (CICCs) and the narrow window of observation (last EM cycle evidence) presents many challenges and limitations to this study. Fortunately, as it will be shown below, the sections of conductor which underwent magnetic fields lower than 1 T remain mechanically identical to the sample before SULTAN testing, providing us with an additional data point that can be regarded as zero EM cycles, something that we did not have in our previous CICC autopsy [23]. Unfortunately, solely having a before-and-after picture may not explain the degradation behavior in its entirety, and this is something that needs to be kept in mind at the time of the discussion and conclusion of the study.

The metallographic techniques presented here are not conventional, and although briefly introduced in [23], they will be explained here thoroughly in order to fully demonstrate their capabilities. They consist on two main observations. (1) Transverse cross-section analysis which can provide evidence of permanent transverse movement due to Lorentz force and (2) single strand extraction which can provide evidence of the mechanical effects of the Lorentz force on the composite strands (mainly filament fractures).

Let us remind the reader that filament fractures in Nb₃Sn strands are indications of very high stresses and therefore a large degradation on the transport current capabilities of the strands, however, a great deal of transport current degradation (sometimes as large as 30%) can still be induced by bending even when filament fractures are not produced [22], making filament fractures merely the tip of the T_cs degradation iceberg. Another consideration is that transport current in extracted strands may not be reliable since the removal of the jacket can alter the strand's stress state significantly [26–28] therefore we focused on metallographic studies only. Despite all these limitations many relevant conclusions can be drawn from these seemingly simple observations.

An additional technique that was not included in [23] will be introduced. It is used to measure the curvature of the extracted strand segments—a curvature which is mostly induced during the cabling and compaction of the CICC.

Two ∼150 mm long conductor segments (one from the HFZ and one from the LFZ) were cut from each conductor length using electric discharge machining (EDM). These pieces were then vacuum impregnated with epoxy to fill the ∼33% void space inside the jacket. Once the epoxy was set, each was further cut using a diamond saw so as to make three cross-sections about 50 mm apart (transposed 40°) of each of the four distinct samples. All were carefully polished and imaged at 50x magnification using a laser scanning confocal microscope (LSCM) with a computer controlled stage that allows taking very large matrices of fully focused images with enough resolution to accurately measure the areas and the coordinate positions of every strand within each CICC. To perform these measurements, the widely used open source software package, Fiji [31] (based on ImageJ [32]) was used to isolate the different components (i.e. strands, void, petal wrap and jacket) into binary (single tone) images like those shown in figure 5(b). Once isolated, a different color was assigned to each component of the image as shown in figure 6(a).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

After all components were measured, we calculated the void fraction in each of the six petals or sub-cables, the strand movement and the geometric distortions of the strands. The autopsies performed in [17] and [23] showed that the strands migrate in the direction of the Lorentz force and therefor a rearrangement of the void fraction is produced. So, in order to quantify the local change in void spacing produced by the Lorentz force we assign an angular value to each petal where 0° (or 360°) marks the circumferential position at which the forces accumulate the most (as the strands migrate in the direction of the Lorentz force), which makes 180° the side of the conductor from which the strands have migrated, as is shown in figure 6(b). We refer to the 0° and 180° positions as the high pressure (HP) and low pressure (LP) conductor sides. Which are (in self-field) slightly lower field and slightly higher field respectively [33].

The void fraction of each petal was measured individually and their deviation from the overall void fraction was used to determine whether the petal had undergone a void fraction expansion (strands moving away from this region) or a void fraction compression (strands moving towards this region). Regions outside the petal wrap (the black features in figure 6(a) were not assigned to any particular petal and therefore do not play a role in the void deviation measurements but they were taken into account for calculation of the overall void fraction.

Another way to track strand movement within the CICC is by measuring the coordinates of each individual strand with respect to the center of the whole CICC. These geometric strand centers were averaged and compared to the center of the conductor itself.

2.2.1. Introduction

Another set of conductor segments (∼150 mm long) from the HFZ and from the LFZ were cut using EDM but this time the jacket was removed using an end mill. The released sub-cables were then submerged in a 15% HCl solution for two hours to remove the chromium plating so as to release any sintering that might have occurred during CICC heat treatment. The petals were then separated, the stainless steel wrap removed and the strands color coded with enamel paint according to position. The unraveling of the strands was done with extreme care in order to minimize any additional strain (a video was made of this crucial step). Strand segments about 12 mm long were then cut using wire cutters and these segments were polished in longitudinal orientation. The goal was to extract pieces from the most representative regions of the conductor, so it was decided to pick strands from the HP side, the LP side and four different petal positions as shown in figure 7. In the modeling studies of Bajas et al [20] it was predicted that the strands that curved the most during cabling are those that would be most prone to filament fracture, this was also confirmed in [23]. Therefore all extracted strands were classified as 'bent' or 'straight' depending on the degree of curvature (see strand curvature section). A total of eighty strands were extracted from each conductor, forty from the HFZ and forty from the LFZ. All strands were imaged using the LSCM at 200x magnification which was enough to be able to see filament fractures.

Zoom In Zoom Out Reset image size

There are two main pieces of information that were obtained from the extracted strands, their crack count and their curvature.

2.2.2. Crack count and analysis

Every LSCM image was carefully inspected for filament fractures and a small digital marker was placed on each crack for subsequent automatic counting. The strands showing filament fractures were then imaged again with backscattered electrons (BSE) in a field emission scanning electron microscope (FESEM) in order to image the cracks at higher magnification and check for additional signs of any post heat treatment strain, in particular cold worked copper grains. Figure 8 shows an example of a strand that developed multiple filament fractures, which also possesses local Cu regions that are significantly cold worked and elsewhere quite strain free. This will be discussed further in the next section.

Zoom In Zoom Out Reset image size

2.2.3. Strand curvature

In all ITER cables the principal strand curvature is produced by the multiple cabling and twisting stages that are designed to transpose all of the strands to reduce ac losses and to guarantee a uniform current distribution. The curvature mostly depends on the cable twist prameters as well as final compaction of the cable. In addition, some permanent strand curvature may be introduced by the Lorentz force loading as shown qualitatively by the autopsy of the CSJA1 cable [17]. In our earlier autopsy of a SULTAN-tested TF cable [23], we found that almost all filament cracking occurred in curved sections of the strands. Therefore in order to find the probability of having a 'bent' strand segment inside the CICC, a separate and much larger set of strand segments around 12 mm long were extracted. To do this, we epoxy impregnated a small length (∼10 cm) from one of the six petals for each CICC. Once cured, cuts were done to obtain three 12 mm long pieces of this epoxy impregnated petal. The epoxy was dissolved to release ∼450 small strand segments from each sample and these were then submerged in 50% nitric acid in order to completely dissolve the copper strands as we only wanted to measure the curvature of the superconducting strands. Dissolving the copper strands of the CSIO1-2 sample was a crucial step since the pure copper strands are much easier to bend than the Cu/superconductor composite strands [22] and therefore it was expected that the superconducting strands would have less average curvature than the soft copper strands. High resolution flatbed scanner images were then taken of these Cu-stripped strand segments and these images used to measure the end to end separation, d, and the length, l, of each strand as seen in figure 9 for a small number of strands. The ratio d/l defines the strand curvature, c, which in a straighter strand approaches 1

$$\begin{eqnarray}&&c=d/l.\end{eqnarray} \tag{ 1 }$$

Zoom In Zoom Out Reset image size

We measured overall void fractions of 33.5 ± 0.35% and 34.0 ± 0.30% for the CSIO1-1 and CSIO1-2 CICC, values very close to the calculated values of 33.4% in [1]. We found no difference in these overall void fractions between the LFZ and HFZ. The individual petal void fractions for all cross-sections are plotted in figure 10. They show a significant void fraction expansion on the LP side of the HFZ, indicating a migration of the strands from the LP side towards the HP side. This strand migration reduces the strand-to-strand support on the LP side, which is known to be detrimental to cable performance [20, 30].

Zoom In Zoom Out Reset image size

The deviation of the center of mass of the strands from the center of the cable is shown in figure 11(a) for all cross-sections and their average deviations are shown in figure 11(b). In the LFZ the average positions of the strands are much smaller, slightly off-center (<100 μm overall) and occur in random directions. In the HFZ, however, the strand offsets lie within ±30° from the direction of the Lorentz force (0°). Clearly most of the (∼320–400 μm) offset in the HFZ is due to strand migration in the direction of the Lorentz force. And although we cannot exclude a small ∼100 μm random component (as found by the LFZ measurements), the lower offset values in the CSIO1-1 point to a higher stiffness than the CSIO1-2.

Zoom In Zoom Out Reset image size

Another way of appreciating the influence of pure copper strands is by looking at the strand contact points. Figure 12 shows transverse cross-sections of the two conductors with the strands shaded according to their nearest neighbor number [23]. The number ranges from 0 for strands with no supporting neighbors (i.e. a strand which is not in contact with any other strand) to 5 for strands with five contact points. The segregation of the light and dark regions is much more discernable for the less stiff, Cu-strand-containing CSIO1-2 than for all-superconducting CSIO1-1—indicating more plasticity coming from the copper strands.

Zoom In Zoom Out Reset image size

The bell curves that correspond to the statistical values (i.e. average and standard deviation) obtained from the curvature measurements c = d/l (1) of each sample are shown in figure 13 where it can be seen that the probability of finding a curved strand (c < 0.95) is much higher for the all superconducting CSIO1-1 CICC (0.21) than for the mixed strand CSIO1-2 CICC (0.08).

Zoom In Zoom Out Reset image size

3.3.1. Introduction

3.3.2. Filament fracture classification

We found no evidence of filament fracture or cold work in any strands extracted from the LFZ of the CICC samples. We take this to indicate that the axial compression produced during cool-down (the main source of strain in the LFZ) is not high enough to produce filament fractures. It also shows that our CICC deconstruction techniques do not produce false positive cracking.

There was, however, a small population of strands in the HFZ of both CICCs that developed filament fractures especially in strands with curvature ratios smaller than 0.95. To our surprise, some fractures were not perfectly orthogonal to the filament axis, which is not expected from a brittle material fracturing under tension. We have observed this type of unusual crack in strands tested in TARSIS under conditions designed to simulate pinching or high contact stresses [22]. These high-angle cracks are readily distinguishable from low angle cracks produced in a pure bending configuration where the tensile stresses produced by bending initiate cracks. Figure 14 shows examples of cracks at high angles produced by pinching and low angle cracks produced by bending during TARSIS testing. For this paper, based on the TARSIS results, we classify the cracks with angles of less than 15° as Type A strand bending cracks, while those cracks with angles higher than 15° are defined as Type B strand pinching or contact cracks. The angle is measured between the local wire axis and the direction of the crack.

Zoom In Zoom Out Reset image size

3.3.3. Crack count and analysis

The crack counts for each piece of strand extracted from the HFZ are shown in tables 2 and 3. It became apparent very quickly that the population of strands with filament fractures is not directly related to the performance of the cables—since more of these were found in the better-performing CSIO1-1. In tables 2 and 3 strands labeled as 'bent' have a stronger curvature with ratios lower than 0.95, while straight strands have a curvature ratio higher than 0.98 (strand pieces with intermediate values of curvature were not used). The strands are arranged according to their respective petal and cable positions denoted in figure 7. Empty cells in tables 2 and 3 are strand pieces where we failed to obtain an acceptable cross-section.

Table 2.  Filament fracture count on each strand extracted from the CSIO1-2 (Cu and superconducting strands) sample.

	HP bent	HP straight	LP bent	LP straight
a	0	0	0	3
a	0	0	0	37
b	0	0	79	0
b	0	0	152	20
a	0	0	460	9
a	0	0	—	0
c	0	0	0	0
c	0	0	—	0
d	0	0	0	0
d	0	0	0	0
Ave.	0	0	86.4	6.9

Table 3.  Filament fracture count on each strand extracted from the CSIO1-1 (superconducting strands only) CICC.

	HP bent	HP straight	LP bent	LP straight
a	—	16	8	4
a	22	3	3	2
b	68	4	28	0
b	4	15	0	0
a	0	0	0	0
a	334	0	165	7
c	—	5	—	2
c	—	0	5	26
d	—	25	7	0
d	6	0	0	3
Ave.	72.3	6.8	24	4.4

An important result on strands extracted from CSIO1-2, that contains both Cu and superconducting strands, is that we found no filament fractures on the HP side but we did find some strands with a significant crack density on the LP side. The most damaged strand from CSIO1-2 is shown in figure 15 and the crack locations are highlighted with blue dots. This strand had more than 95% Type A cracks and since the fractures were located on a single side of the strand we can assume that this was on the tensile side of the bend, as previously observed in TF cable [23]. The overall population of filament fractures in this CSIO1-2 sample is 93% Type A and 7% Type B which are ratios comparable to our observations on samples tested in bending with TARSIS [22].

Zoom In Zoom Out Reset image size

In contrast to the CSIO1-2, the all superconducting CSIO1-1 has several strands with a wide variety of crack densities (table 3). The most damaged strand extracted from the HP side is shown in figure 8 where only 20% of the cracks were Type A. In fact all the strands showing fractures in the HP side were Type B crack dominant. Due to their evident frequency in the CSIO1-1, it appears that Type B cracks do not have a strong impact on the performance of the cable. Type A cracks on the other hand have a similar pattern in the LP side of both conductors, with the CSIO1-1 having an overall lower average number (see LP side in tables 2 and 3). A strand from the LP side of the CSIO1-1 sample is shown in figure 16. It presents a very similar fracture pattern to the CSIO1-2 strand from figure 15, consistent with bending-induced cracks.

Zoom In Zoom Out Reset image size

In summary, the overall crack population in the HP side of this CSIO1-1 sample is 36% Type A bending cracks and 64% Type B pinch cracks, while on the LP side, we found 80% Type A and 20% Type B—yielding a much lower overall number of Type A cracks in the CSIO1-1 sample.