The use of Nb₃Sn in fusion: lessons learned from the ITER production including options for management of performance degradation

Nb₃Sn superconductor properties are highly dependent on the lattice distortion of the Nb₃Sn. Nb₃Sn is also well known as a brittle compound that easily fractures under tension [35]. After forming the cable from the strands it (and the conductor jacket) are then heat treated for a few hundred hours at about 650 °C to form the Nb₃Sn compound [2]. This on its own produces many challenges for the jacket material and the coil insulation. At 650 °C, some components of the conductor are formed 'new'(the Nb₃Sn), some are unaffected (Ta), or unaffected if separate from Sn (Nb), some undergo some stress relaxation (the steel jacket) and some are fully annealed (Cu and bronze).

In ITER, strands have to be separated to allow Helium at ∼5 K to provide cooling, and to limit AC losses. But the strands must be strongly supported for magnet/thermal loads. Achieving the correct balance is extremely difficult, and, due to cost pressures, there is a tendency to err on the 'overload' side. Figure 10 shows the situation, with the loads.

Zoom In Zoom Out Reset image size

This configuration is difficult to analyse mechanically, especially when cables become large with more than 1000 strands. Typical early simple attempts are shown in [3] and more recent in [36]. Broadly the strands are compressed by the thermal loads and due to the curvature, bend sideways. The (sideways) magnetic loads then add to this bending.

The causes of conductor degradation were discussed extensively between 2002 and 2012, along with ways to mitigate/control/eliminate. Early work [6] focused on bending and non-uniform strand properties which created inter filament current transfer (Type 1 Degradation). By 2013 clear evidence was obtained [3] that in some conductors filament fracture is playing a major role (i.e. Type 2 degradation).

However, even at this point, the mechanism by which both thermal and magnetic cycles could contribute to fracture was not clear. Subsequent tests (including the ITER conductor samples in 2018, described in the next section) show clearly that at least in the ITER TF conductors, it is a result of mechanical coupling between electromagnetic (or I × B) loads and thermal compression. Deformation and cracking under one causes repeated deformation and cracking from the other when conductor undergoes a load or thermal cycle. It is a form of coupled buckling [3] which is likely to be related both to the copper stabilizer on the outside of the strands and mechanical weakening as filaments fracture on the inside.

The nature of the cracks, and their distribution, are clearly identified in [8]. Figure 11 shows the result of micrographs made from strands extracted from a cable. These examinations also suggest that it is the local magnetic forces that cause the fracture, not (as was previously expected) the cumulative pressure build up inside the cable from one side to the other. However, there are also indications in more recent investigations that the cumulative pressure can create damage too [37].

Zoom In Zoom Out Reset image size

Unfortunately, it is only possible to assess degradation with an I × B cycle, which makes separating the impact of thermal and magnetic cycles quite difficult. As shown in the next section, it appears that a thermal cycle is more critical than an electromagnetic one. Electromagnetic triggered degradation stabilizes after ∼10 cycles but is reset by each thermal cycle, which also stabilizes after ∼10–15 thermal cycles. This multiple dependency makes data reduction difficult, especially when trying to show visually a correlation between one of the measurements of degradation described below and the 'number of cycles'.

Since 2002, various methods have been developed to characterize the performance of degraded Nb₃Sn strands [28]. The intent of all of them is to find a parameter that is independent of the strand operating conditions and can be deduced from the experimental data. The directly measured performance parameter is usually current sharing temperature but this is obviously dependent on current density and field, as well as being linked to the definition of the critical electric field at which it is measured. The two most useful measurements have been

• 1.  
  Effective strain (sometimes as a function of electromagnetic load).
• 2.  
  Fractured (or intact) filament area.

Both being supplemented by a deduction of the exponent of the resistive transition n which is defined below for a strand [14].

$E={E}_{c}{\left(I/{I}_{c}\right)}^{n}$ where E_c (or E₀) is conventionally 10 μV m⁻¹

The total effective strain was originally developed [28, 30] because at least part of the strand degradation being observed (Type 1) was expected to be associated with current transfer. It is not reversible but nor is it fully what is expected as a 'degradation'. It is convenient because strand-in-conductor performance in Nb₃Sn is always associated with a relatively unknown parameter, the filament strain. Lumping Type 1 degradation together with intrinsic strain and operational strain has the advantage of putting all the unknowns related to strain into one parameter. However, for Type 2 degradation caused by filament fracture, the effective strain is not a representative physical parameter, as it tends to vary with the operating conditions, and the fractured filament area appears more useful.

The second set of CS insert tests give a good example of a coil where using a strain based interpretation of performance, we find a dependence of extra strain on I × B, when operating strains are taken into consideration. The cable also shows, relative to the strand, a low n (although not as low as displayed in the second TF insert). The CSI-2 conductor showed low (negligible) filament fracture (as defined below) and no change with thermal or electromagnetic cycling. Therefore, there is no Type 2 degradation but significant pointers to Type 1.

The fractured filament area was first used in 2002 [26] and, because of its direct relation to the main factor behind the irreversible degradation found in the TF conductor, is the parameter used here. It is not entirely satisfactory since the filaments do not actually fracture to provide a full current transport barrier, but instead provide many (low) resistive interruptions to the current flow, where the nature of the current transfer depends again on the strand operating conditions. To derive the fractured filament area requires a knowledge of the Nb₃Sn thermal and operational strain (ε_eff = ε_op + ε_th). Rather than (as above) deducing a total effective strain for the conductor based on the strand database and the coil measurements, the effective strain becomes an input parameter determined by a mechanical/thermal analysis. The deduced parameter is the fractured filament area. These strains are a form of average since the strains on the actual filaments are complicated by bending of the strands within the cable [3, 4, 6, 36].

Recent results from the ITER insert coils [38, 39] have finally reduced the uncertainty by confirming that the differential thermal strain between Nb₃Sn and jacket (from 650 °C to 4 K) as well as the coil operational strains are 80% transmitted to the filaments [40]. The total strain in the TF conductor in operation at the peak field point is now assessed as −0.55% which includes any Type 1 degradation.

The fractured filament area is deduced from the coil performance measurements as follows

$$\begin{eqnarray*}&&E\left(T\right)=\displaystyle \frac{{E}_{0}}{L{A}_{tot}}\displaystyle {\int }_{L}dz\displaystyle {\int }_{{A}_{tot}}{\left(\displaystyle \frac{{J}_{op}}{{J}_{c}(B,T,{{\epsilon }}_{eff})}\right)}^{n}d{A}_{tot}.\end{eqnarray*}$$

With j_c(B, T, ε) coming from the strand parameterization (well defined for ITER, [2])

$$\begin{eqnarray*}&&{J}_{op}=\displaystyle \frac{{I}_{op}}{k{A}_{tot}}=\displaystyle \frac{{I}_{op}}{A}\end{eqnarray*}$$

with L being the total length (direction z) over which the average electric field E(T) is being calculated, at a temperature T. A_tot is the total conductor non-copper area, A is the intact non copper area and k is the fraction of the area that is intact. A and n are derived typically by measuring E(T) as a function of T and carrying out a best fit of the resulting curve, typically at E values from 5 to 50 μV m⁻¹. Implicitly, n is not a function of B or T and is constant for all strands over the conductor. For single un-degraded strands (so n ≫ 10) n is well known to be a function of the critical current density but this dependence is expected to disappear in degraded strands.

Two items are critical to this process:

• 1.  
  The availability of an accurate strand parameterization. During production of the ITER strands a very large effort was made to characterize the critical current of all types of strands, as a function of field, temperature and strain [41]. This work included cross-checking between laboratories carrying out the measurements and accurate parameterization of the results.
• 2.  
  The knowledge of the cable strain state ε_eff. This is essentially including the Type 1 degradation before deducing the extra Type 2. This is obtained from pre-cycling measurements on the coils together with assessments from conductor samples of the transmission factor of operational and differential thermal strain to the strands [33].

The fracture filament area method was first applied to the first ITER CS insert coil [42] with the result shown below in figure 12. The order of fracture (25%–50%) is quite similar to that which will be estimated in later sections for the ITER TF conductors.

Zoom In Zoom Out Reset image size

The ITER strand production was accompanied by a strong quality control programme that required the test of multiple full size Sultan conductor samples [2]. Since 2010 around one hundred have been tested. However, these samples were intended to be a form of production sampling, and were tested only at peak field and current, after a lifetime simulation of 1000 electromagnetic cycles (and one final thermal cycle that was not part of the acceptance test). This is in one sense an extreme form of EM loading since no part loads were carried out at the start, and in another sense optimistic, in that no intermediate Warm-Up, Cool-Down (WUCD) cycles were carried out. The risks in the procedure were indicated already in Sultan tests in 2012 but did not become fully known until the test of a second TF insert coil, TFI2, in 2017 [34].

This insert showed a pronounced degradation (certainly Type 2), with apparently a coupling between EM and WUCD cycles so that each WUCD cycle triggered a new set of EM degradations over the following 10 (or so) EM cycles before a stabilization. The loading was however again not representative of that to be imposed on the real coils, with a long series of full current full field EM cycles carried out before the start of WUCD, and only full current and full field EM cycles being performed. This insert eventually showed stabilization [39] but left open questions about the behaviour of different strands and the possibility of managing the Type 2 degradation as well as of course the mechanism causing the coupling between WUCD and EM cycles.

Since 2018 therefore, a series of new SULTAN samples have been tested with more realistic mixture of low EM cycles and WUCD than with model coils, insert coils and production conductor samples. These samples are made from the ITER conductor archive (2 samples are stored for every unit length). The aim was to look for thresholds for the onset of the WUCD-EM degradation cycles as well as to investigate the behaviour of strands from different suppliers. A summary table of these samples is given in a later section, table 6. In addition, some of the production samples [2] were similar to these recent samples and could be used for comparison. These appear in section 6.2 and are designated as TFEU8 and 10 and TFJA7 and 8.

A typical example of the tests is shown in figure 13. Because of the coupling between EM and WUCD cycles, it is difficult to find a plot that enables trends to be picked out. In figure 13, both EM and WUCD are counted as 'equal' cycles. This at least shows the procedure that was followed but does not allow the different weight of EM and WUCD cycles to be observed. The plot shows that degradation sets in between 0.25 and 0.5 of the full load (current) × (field) kAT, BxI. It is worth noting that WUCD cycles in Sultan are a lot more time consuming (and expensive) than EM cycles. Even an 80 K WUCD takes 1 day; a 300 K cycle takes 3.

Zoom In Zoom Out Reset image size

One of the overall features shown by these latest samples is that the degradation in current sharing temperature is either less than, or comparable to, the degradation found in the production samples. However, the production samples are subject to only one WUCD cycle and are clearly not stabilized. This opens the intriguing possibility that it is possible to condition (i.e. train) the Nb₃Sn CICC conductors to better resist the combined magnetic and thermal load cycles.

Evidence of training in Nb₃Sn conductors is still subject to interpretation and further tests are needed to confirm it. Here we can set out, using the ITER TF conductor database, some of the evidence.

• (1)  
  Thermal cycling of internal tin conductor during ITER production [43]

This was an early investigation of WUCD effects which was carried out in 2 steps. The sample has 2 identical legs, made of WST internal tin strand. In the first step, one leg (left) was subject to 10 thermal cycles to 80 K at the start and then to a further 30 cycles to 80 K before step 2 [43]. In the first step, both legs went through 1000 full current/field magnetic load cycles, in the second step both legs went through 3 × 300 K WUCD interspersed with 500 full current/field magnetic cycles.

Figure 14 shows the results. There is a difference of about 0.2 K between the two legs which is maintained over the life, with the 'thermally conditioned' leg remaining above the other leg. However, other samples from the same type of strand showed also a difference between the 2 legs that was maintained over the EM cycling and 1 WUCD [44]. So it is not clear if it is the thermal conditioning that is improving the Left Leg in figure 14, or a difference in the sample preparation. The difference in figure 14 is maintained through 2 WUCD whereas production samples generally converge with load cycling [45].

• (2)  
  ITER conductor samples: trend comparisons

Zoom In Zoom Out Reset image size

Although several ITER TF conductor samples have been used in the degradation management programme, the training investigations have focused on two (both bronze route) where the samples have undergone testing up to full I × B and where there is a collection of production sample data available for comparison.

The TFIO1 sample was composed of two conductors based on bronze route strand. One of those was made by Jastec (JADA), the same strand used for the TFI manufacturing. The second leg was selected from Bruker-made strand (EUDA) that demonstrated the most similar performance to Jastec during qualification and production tests.

The sample was subjected to different combinations of thermal cycles (WUCD to 80 and 300 K) and electro-magnetic (EM) cycles as illustrated in figure 13. These increased the magnetic load progressively in steps from of 25%, 50%, 63%, 80% and 100% of the maximum EM load required in ITER.

The main goal of the TFIO2 test was to confirm the reproducibility of the TFIO1 results. Only 300 K WUCD cycles were implemented on TFIO2 during testing at lower EM combinations and then the effect of 80 versus 300 K was compared at the end of the campaign at 100% of EM load.

The final results are summarized in table 6. This table also includes the results of a 'TFI' sample which matched the conductor used in the second TF insert coil.

Table 6 shows clearly that TFIO1R and TFIO2L, although subjected to many more WUCD cycles, degrade to the same extent (or less) than the TFI sample.

TFIO1 and TFIO2 underwent, correspondingly, 28 and 12 WUCD cycles and that these WUCD cycles were mixed with different EM load levels. The TFIO1 and 2 are at or near stabilization after this. Production and qualification Jastec samples (81JNCxxx) were exposed to one WUCD cycle only and did not reach a stable level. There is a steep drop at the WUCD after 1000 EM cycles already close to the final level of the TFIO1 and 2, and this first drop may not have stabilized. The TFI samples were warmed up and cooled down a second time and already after this their T_cs is below that of the TFIO1 and 2 samples. This shows that the degradation of the conductor seems to decrease when the conductor is exposed to lower EM loads at the beginning of the testing series before being exposed to the highest ELM loads (i.e. a training effect).

The problem with quantitative interpretation of the test data is that there is an element of variability with single EM load tests and single WUCD, linked to delayed onset of fracture. A series of tests is much more reliable than an individual test. So in this section we have tried to find a way to look at trends rather than try to correlate changes with a specific cycle. This has the further advantage that the fairly large errors associated with looking at changes in intact filament area (or T_cs) between specific cycles can be reduced by looking at an overall trend.

To assess the data we have to find a way of simplifying the variability in cycles. There are too many parameters to enable trends to be established. The solution proposed is to define compound cycles. This concept follows from the observation that WUCD cycles, when followed by an EM cycles (unavoidable, to allow the extent of fracture to be measured) is much more severe than one (or 100) full load EM cycles. One conversion (of course, there are many possibilities) that gives consistent results is given below:

• One WUCD from 4 to 300 K and back to 4 K, followed by a full load electromagnetic cycle is defined as 1 compound cycle.
• 1000 EM cycles (whatever the electromagnetic load value) are defined as one compound cycle.

The contribution of a series of EM cycles is then determined by dividing their number by 1000. A WUCD plus the first EM cycle after the WUCD is weighted by (I × B)/(I × B_max) with I being the conductor current and B the field, and additionally by 0.2 for an 80 K WUCD. This definition of compound cycles gives a reasonable similarity between TFIO1 and 2. Although there is no reason that the two should be identical, we could expect that they are at least similar in form.

A number of production samples are available for comparison with TFIO1 and 2, denoted as TFEU 8 and 10, and TFJA7 and 8. The definition of compound cycles normally reduces the production samples to 2 or 3 points (which was of course the reason for its definition, to take account of the higher degradation weighting found with WUCD than with EM). However, this limits the extent of the comparisons that can be made. To provide a wider set of data, an old production sample from 2012 (TFEU8) was reassembled in 2019 and put through further electromagnetic and thermal cycling.

Figures 15 and 16 show the results. The extended test of the TFEU8 production sample confirms the expected trend.

We can make some observations:

• On the first EM cycle after first cooldown, full load produces a drop of 0.3 of intact area (1 to 0.7–0.6), 0.25 full load produces a drop of 0.1 (1 to 0.9–0.8)-approximately.
• This difference is maintained throughout rest of life regardless of level of EM load (i.e. loading in steps reduces fracture by about 0.15).

• (3)  
  ITER conductor samples: performance loss/cycle comparison

In order quantify better this (possible) training effect with a different assessment of the data, a performance loss indicator was defined and calculated for each series (at constant electromagnetic load) of TFIO1 and TFOI2 measurements. In line with the definition of 'compound cycles' above, the presentation focuses on thermal cycles followed by a few EM cycles, not on a long sequence of EM cycles. The initial T_cs value of a series was considered as T₀, the final T_cs of the same series T_f. The difference between T₀ and T_f is then divided by T₀ providing the performance loss (%) = (T₀ − T_f)/T₀. For the TFI sample, the performance loss was determined for warm-up cool-downs only. To focus on compound cycles, T_cs at 1000 EM was taken as T₀. The correlation of the performance loss with EM load cycles is presented in figure 17.

The results above confirm that the performance loss increases with EM load level, and that the conductors that were exposed to lower ELM loads first show a lower performance loss at the full (100%) EM load. The performance loss of the TFIO1 at 100% EM load is significantly lower than the TFI-Sultan sample, although the TFIO1 sample was subjected to 29 WUCD cycles, while TFI Sultan sample went through 2 WUCD cycles only.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Although the test data indicates a possible training effect, that could be used to reduce the extent of filament fracture in coil applications, we also need to establish at least conceptually a possible mechanism for the effect.

The mechanism proposed is based around the mechanical behaviour of the copper outside the diffusion barrier that makes up about 50% of the area of the strand. Being furthest from the bending axis, the copper has also the largest potential impact on the bending stiffness of the strands.

The mechanical properties of Nb₃Sn strands are dominated by the reaction heat treatment (at 650 °C) and the subsequent cooldown to 4 K [46, 47]. The copper in the strands is initially fully stress relieved and has a very low yield stress whereas the Nb₃Sn (and any residual Nb material) are fully elastic even at 650 °C. As discussed in section 5.1, the copper is fully bonded to the unyielding Nb₃Sn/Nb/Ta components which contract much less than the steel, and do not yield. In the cable, the strands can undergo bending (see figure 10) due both to the thermal loads (longitudinal) and the magnetic loads (transverse) [46]. This bending creates tensile strains, even though the Nb₃Sn is compressed by the steel jacket. The Nb₃Sn deforms elastically under this bending up to the point when it fractures, a tensile strain of around 0 to 0.002. The copper within the strands has a thermal contraction close to that of the steel jacket but, being attached to the Nb₃Sn, is put into tension by it and due to its very low yield stress, it plastically deforms. The stress strain behaviour of copper at different temperatures is shown in figure 18 based on the mechanical property parameterization as a function of temperature given in [46]. Overlaid on these lines is the approximate stress–strain trajectory of the copper in the strands during the first cooldown. To produce this, we have assumed that the accumulated copper strain is one half of the integrated differential contraction from 650 °C, using the parameterization of the copper, Nb₃Sn and steel thermal contraction reported in [46]. The copper work hardens substantially at 4 K and has a significant role in supporting the Nb₃Sn filaments. So tensile strain on the copper may also be associated with filament fracture (of course, depending on the cable configuration).

Zoom In Zoom Out Reset image size

Following this initial cooldown, it is possible to condition the conductor by carrying out thermal cycles from 4 K to either 77 or 293 K and then back to 4 K, or for the conductor to be subjected to EM loads immediately.

Figure 19 shows the approximate trajectory followed by the copper with this thermal cycling to 77 K. The situation is quite similar to a high temperature annealing process. Due to the bending, softening of the copper allows a strand deflection (in the form of a higher tensile strain). With 2–3 thermal cycles, the copper will stabilize on the 77 K stress–strain line.

Zoom In Zoom Out Reset image size

Figures 20 and 21 show then the effect of an EM cycles with or without the thermal conditioning. We have assumed that the EM loads create an additional bending stress of 15 MPa. The copper that has been conditioned (i.e. work hardened) initially deforms elastically (due to the scale, this appears as a near vertical line) and then plastically. The un-hardened copper deforms immediately plastically, with the result that the copper at the end has a strain change under the EM loading of 0.0027, compared to 0.0005 with the work hardened copper. This translates into reduced Nb₃Sn fracture for the conditioned strand.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

These diagrams are of course largely qualitative and a gross approximation to the real mechanical situation in the strands, which is a mix of bending and direct stresses. However, they have a sufficiently quantified base to give confidence that a potential mechanism to explain the experimental results exists.

The original design criteria for the ITER conductors were common to NbTi and Nb₃Sn and were developed in the period 1988–1991 [44, 48]. They were based very much on the problems found in the 1980s largely with composite NbTi conductors and problems with stability [19]. The five factors considered were:

• (1)  
  Cryogenic stability (Stekly criterion and limiting current) which gave Cu:nonCu requirements for the strands >1.5.
• (2)  
  Quench and hot spot.
• (3)  
  Derivation of Nb₃Sn intrinsic strain based on differential thermal expansion with the jacket material from the heat treatment temperature, about 650 °C.
• (4)  
  React and wind versus wind and react: maximum allowable strain on reacted strands.
• (5)  
  Conventional temperature margin between operating temperature and current sharing temperature (defined by an electric field of 10 μV m⁻¹) of >1 K for Nb₃Sn, with a strand in cable n-value >20.

Nb₃Sn conductors were being designed as if a kind of high field version of NbTi, leading to a failure to exploit the capabilities of Nb₃Sn.

These criteria were unsatisfactory in a number of respects, even with the extent of our knowledge in 1991. What we did not know then (or could not quantify, and therefore ignored) can be summarized as follows:

• The concept of Nb₃Sn filament fracture except as an abrupt limit (below, no impact, above, fully broken and no current flow).
• Low 'n' behaviour (i.e. degraded) of Nb₃Sn conductors and its impact on definition of critical current and operational regime.
• Current non-uniformity (inherent to any superconductor) and its effects on stability during pulsed (or even near steady state). There were several noted failures in the 1980s and the conditions to allow (or to block) current transfer between strands within a composite cable using strand coatings was the subject of much discussion in the 1990s with insulated strands still being proposed [49]. Eventually a Cr coating was selected for Nb₃Sn and this has become standardized, allowing controlled and reproducible current transfer between strands in composite cables.
• The strain behaviour of the filaments inside the jacket. We expected significant gains in strand jc by choosing jacket materials that would allow the Nb₃Sn to operate near the peak of the well known jc-strain curves [41]. Theoretically, based on differential contraction coefficients, the filaments in a steel jacket would experience a strain of −0.83% which reduces the jc from the peak of the curve by a factor of about 3 [35] whereas for a Titanium jacket the value is around −0.25% (and the jc is within 20% of the maximum value). Even in the 1990s, we expected some reduction of the 0.83% for steel, due to the impact of cabling, the positive effect of operation tensile strain and the extension of the jacket by the cable, hence the choice (rather arbitrary) of −0.5%.

The points at which these shortcomings were realized can be picked out in the timeline in figure 3 (the trash cans).

It was not until 2003 that we were able to obtain an agreement with (at that time) the only 3 ITER participants EU, US and JA to revise the Nb₃Sn criteria to reflect the results of R&D in the 1990s and in particular the results from the ITER model coils 2000–2002 [21, 22]. The new guidelines were documented internally for ITER but have not been published as they have not been adequately generalized (i.e. they are applicable to variations around a specific ITER CICC). They can be divided into 3 groups with the last, the 'conductor-in-coil' reflecting the need for a more holistic approach to the conductor design.

7.3.1. Strand

7.3.2. Cable

7.3.3. Conductor-in-coil