3.1. Magnetic design

The calculations were performed using the code OPERA (Dassault Systemes SIMULIA). The magnetic design went through a long process to determine the technology that would be used for the dipole. The first idea was to design an electromagnetic dipole with standard low-carbon steel yoke Fe–Co poles for the high field and coils made of hollow conductors (Fig. 2). It gave a dipole with a 500 A range current with a power consumption of about 37 kW. As a comparison, the standard electromagnetic dipole in the SOLEIL storage ring consumes about 12 kW. Since the modification of the hydraulic infrastructure in the storage ring tunnel requested for the installation of such a device turned out to be so important, this solution was rejected. Following this decision and knowing that Sirius, the new Brazilian source (Citadini et al., 2018), made a successful PM dipole with a 3.2 T peak field, the PM solution was chosen. As mentioned previously, the dipole has two parts: the HFP, which allows the electron beam to produce the expected radiation for the ROCK beamline, and the LFP, which provides the integrated magnetic field required for the 11.25° bending (1.793 T m). The HFP comprises an iron–cobalt pyramidal pole where PMs are installed on each side. The LFP comprises a low-carbon steel pole where PMs are installed on each side and underneath (Fig. 3). Note that the size of the base of the strong field pole has been optimized with the priority of reaching the required field value for the ROCK beamline, taking into account the constraints on the gap imposed by the vacuum chamber mechanical robustness. The resulting very small size of the strong field pole tip generates a strong field roll-off and a strong sextupolar component (see Fig. 18 later). The pole has a pyramidal shape with a 120 mm × 120 mm square base to a 19.5 mm × 45 mm pole tip. This pyramidal shape is 48 mm high on top of the 145.5 mm pole block. It was not possible to significantly reduce the sextupolar component without reducing the high field value. A possible solution was to use the low field part to create an opposite integrated sextupolar component, but the curvature of the beam trajectory would have led to a complicated design to be realized.

Figure 2 View of the bottom part of the electromagnetic version of the Superbend. (Green) The yoke and the poles. (Red) The coils. X, Z and Y are the horizontal, vertical and longitudinal dimensions, respectively, expressed in millimeters. This design was rejected and replaced by a permanent-magnet version.

Figure 3 (Top) Partial view of the high field part (half-yoke) of the Superbend. (Bottom) Transverse view of the exit of the low field part (half-yoke) of the Superbend. (Light green) Pole. (Blue) PM blocks. (Dark green) Yoke. (Orange) Magnetic short circuit. (Red arrows) Direction of the magnetization of the PM blocks. X, Z and Y are the horizontal, vertical and longitudinal dimensions, respectively, expressed in millimeters.

To prevent any discrepancy between the specified field integral and the measured one, due to the sensitivity of the PMs to temperature variation, a set of dipolar coils are installed around the LFP to be able to adjust the field integral during the operation of the storage ring if necessary. These coils can be fed by a bipolar 100 A/600 W power supply with 15 turns and can vary the field integral by ±3%.

To adjust the field integral during magnetic measurements, some stainless-steel shims were inserted between the two yokes. These shims increase the gap between the poles and add an air gap in the yoke that modifies the integrated field. The designed air gap is 2.0 mm. The adjustment of the field integral by means of stainless shims is a long process in terms of manipulation of the dipole where the yoke has to be split into two to insert the shims. So, the dipole was also equipped with some magnetic short circuits (MSCs) on the exit side (Figs. 3 and 4). The MSCs consist of pieces of low carbon steel which can slide between the yoke and the pole in order to capture the flux from the PM and then induce a variation of the field integral. The PMs are made of NdFeB with a remanent field B_r of 1.38 T and a coercitive force H_cj of 1580 kA m⁻¹. The requested size for the PM block is 120 mm × 60 mm × 60 mm. Due to some manufacturing limitations, the PM blocks are made of two 120 mm × 60 mm × 30 mm blocks and the magnetization is along the 30 mm dimension. Only one type of PM block is used for both the LFP and the HFP. A high value of the remanent field was required to reach the 2.8 T peak field value and to have a reasonable volume of PM materials. This was also requested for the low field part where the vertical gap is 36 mm.

Figure 4 View of the high (1) and low (2) magnetic field parts of the Superbend, with (top) and without (bottom) the yoke.

3.2. Mechanical design

The basic mechanical concept of the Superbend dipole consists of keeping as much as possible the standard electromagnetic dipole configuration (Filhol, 2004), in terms of mass and mass center, and to keep unchanged the frequency range of the vibration modes and all the interfaces existing around the dipole. The axis of the ROCK photon beam must remain as close as possible to its initial position leading to a local increase of the magnetic field, just at the photon source point location. Then the Superbend consists of two half-yokes, each including a pole called `low magnetic field' and a pole called `high magnetic field' (Fig. 4). To remain compatible with an in situ baking of the vacuum vessel, without heating the PMs, the Superbend dipole has a C-shape, open on the outside of the ring, that allows it to be shifted on the side and away from any heating source (Fig. 5). The deformation of the C-shaped yoke at the poles is estimated to be 0.020 mm, following a first calculation made during the design of the Superbend. Note that, for the assembly of the PMs, benefits from experience of the Sirius new Brazilian synchrotron light source were taken (Citadini et al., 2018), especially with regard to the conception of dedicated tools, which was a key point for the success of the project.

Figure 5 (Top) View from inside the storage ring of the standard electromagnetic dipole. (Bottom) Same view with the Superbend dipole.

3.3. Assembly of PMs

The PMs were assembled in pairs before being positioned around the poles. All 234 PMs were first measured individually in order to be able to sort them and magnetically optimize the assembly of the magnet pairs. In particular, the angle error was minimized, or even canceled, for each pair of magnets. No automatic sorting program was used – only manual sorting was carried out using an EXCEL file. The magnetic calculations showed that attraction forces reach 8000 daN between the poles and 26000 daN between the two half-yokes. These forces are too strong to allow assembly of the PMs in the two separate half-yokes before assembly of the two half-yokes. First, after assembling the pole holding mechanism in each of the two half-yokes, PMs were assembled around the two `high-field' poles using appropriate tools (Fig. 6), designed and built at SOLEIL. Then, the two half-yokes were assembled, using dedicated tools, to take into account that the attraction force between the two high-field poles, equipped with their PMs, is 800 daN. The lower yoke was fixed on the assembly table, and it was necessary to design and build special fixtures to rigidify the system and prevent the upper yoke from swinging when approaching the lower yoke. The vertical gap between the upper and lower poles was then fine-tuned to the nominal value. Once the two half-yokes were assembled, it was possible to insert the PMs around the two `low-field' poles using dedicated tools (Fig. 6). The challenge was to correctly position the first PMs at the end of the guide (at about 0.6 m from the entrance) to avoid some blocking because removing an inserted PM was impossible due to the magnetic attraction forces involved.

Figure 6 The tools (in gray) used for the installation of PMs around the high field poles (top) and low field poles (bottom).

3.4. Magnetic measurements

3.4.1. Measurements of PM blocks

As mentioned previously, each PM block is made of two smaller PM blocks. To decide which small blocks were associated, all the small PM blocks were measured by means of a Helmholtz coil. This measurement gives the three components of the magnetization of each small block. Fig. 7 shows the data for all the small blocks. The overall standard deviations of the α angle, β angle and magnetization are 0.22°, 0.16° and 2 mT, respectively.

Figure 7 Distribution of the measured angular defect and magnetization for all the PM blocks.

3.4.2. Measurement of the magnetic field by Hall probe

The dipole has been installed on a Hall probe bench (Fig. 8). This bench provides a three-axis translation (3500 mm × 250 mm × 250 mm). All the translation axes are driven by an XPS Newport controller. The Hall voltages are read by three Keithley voltmeters. All these devices are controlled by IgorPro software together with a Tango interface. The Hall probe is an ultra-low-noise system developed and calibrated up to a 4 T field by Senis (https://www.senis.swiss). The magnetic measurement consists of a 2D field map in the median plane of the dipole. An iterative process has been used to adjust the total field integral to the one corresponding to the standard 11.25° deviation (1.793 T m), taking into account the curvature of the electron trajectory. The process consisted of two steps: (i) adjustment of the stainless-steel thickness inserted between the lower and the upper yokes, and (ii) a fine tuning with the magnetic short circuit. The final 2D field map and the variation of the field along the beam trajectory are shown in Fig. 9.

Figure 8 The PM Superbend installed on the Hall probe bench in the SOLEIL laboratory.

Figure 9 (Top) Variation of the magnetic field along the beam trajectory in the Superbend. (Bottom) Beam trajectory (in black) deduced from the measured 2D magnetic field mapping (in colors).

3.4.4. Effect of the temperature

To anticipate temperature variations in the storage ring tunnel during operation, it was important to measure the sensitivity of the integrated field to the temperature of the PM. Integrated field measurements versus temperature were performed on the single stretch wire bench, and in addition a Hall probe was fixed in the gap of the LFP to measure the local vertical field. Over a period of ten days, the integrated field, the field of the LFP and the temperature of the yoke were measured and recorded. Furthermore, the setting point of the room temperature of the lab was increased by 5°C and it took about two days for the dipole yoke's temperature to increase as well. After two days, the setting point was set back to its normal value. Fig. 10 shows the variation of the field integral during these tests. The temperature effect on the field integral is 0.07% per °C which is a bit lower than the sensitivity of the PM remanent field versus the temperature. The difference is due to the saturation of the poles.

Figure 10 Variation of the field integral during testing of the sensitivity to the yoke temperature. (Top) Versus time. (Bottom) Versus yoke temperature.

3.5. Installation

To allow installation of the Superbend dipole in the storage ring and to allow it to be shifted on its side in case of vacuum vessel baking, a dedicated mechanical support equipped with a system allowing adjustment of its horizontal and vertical positions was designed and built. This equipment has the ability to move a mass of more than 3500 kg over a lateral distance of 600 mm above cable trays and other utilities. The poles of the Superbend dipole are so close to the vacuum vessel that it was mandatory to lift the dipole vertically by about 0.4 mm above the beam axis before translation of the dipole. During all stages of placement and extraction, it was obligatory to have a look at the alignment devices to permanently check the dipole position with respect to the vacuum vessel. Finally, the Superbend was set up leaving only 0.2 mm of vertical space between the vacuum vessel and the high field magnetic pole. Note that the support was also used to extract the lower yoke of the standard electromagnetic dipole allowing the new vacuum chamber to remain unmoved and thus avoiding bake-out and the vacuum conditioning phases, thus saving almost two weeks in the planning.

4.1. Mechanical design

The `as built at the beginning' storage ring includes, among many vacuum vessels, 32 non-magnetic stainless-steel dipole vacuum vessels, with slight differences according to location on the ring itself. For this purpose, dipole vacuum vessels were designed in three parts: the first one is a common part designed to be received inside the dipole magnet strictly speaking, the second part comprises a box destined to house the crotch absorber and to bind with the third part (called quadrupole part) which can be different in terms of length, depending on the type of quadrupole and sextupole magnets that follow the dipole and on the distance between them. The vacuum chamber installed in the original ROCK electromagnetic dipole is shown in Fig. 11. As the vertical gap required between the poles of the new Superbend dipole was lower than those of the standard electromagnetic dipoles, it was not possible to keep the common first dipole part of the vacuum vessels as designed. As the quadrupole part should be identical, the original design was kept (Filhol, 2004; Level et al., 2003, 2005; Filhol et al., 2006) and only the dipole part was adapted to the new gap requirement (Fig. 12). Moreover, one of the design constraints for the Superbend implantation was to keep the identical crotch absorber, confirming the design of the new vacuum vessels. Finally, the new vacuum vessels had to accommodate the new dipole magnet gap, keeping the same crotch absorber in the same housing box that binds with the standard quadrupole part. As well as the difficulty in accommodating a very low vertical gap for the 2.8 T magnetic bore, the complementary `low-field' part of the Superbend dipole also needed a lower gap than the actual one because a magnetic field slightly higher than 1.7 T was needed to reach the nominal total field integral. Nevertheless, the dipole part was designed on a similar principle as the actual one (Filhol, 2004; Level et al., 2003, 2005; Filhol et al., 2006), keeping the same manufacturing scheme in two half shelves, assembled with an electron beam welding in the median plane. This allows the shape to be adapted to both magnetic gaps and electron beam path (Fig. 13), keeping a good mechanical rigidity and strength, and keeping the non-magnetic properties of the 316LN stainless-steel raw material. In addition, the transition between the different gaps along the chamber was designed to minimize heating due to the interaction of the electron beam with the walls of the vacuum chamber (impedance effect).

Figure 11 Original vacuum vessel installed in the previous ROCK electromagnetic dipole.

Figure 12 Top view layout of the new Superbend vacuum vessel. (1) Dipole part; (2) quadrupole part; (3) crotch box; (4) pumping sub-assembly; (5) quadrupole pumping port; (6) beamline outlet; (7) crotch absorber flange.

Figure 13 Layout of the transverse profile of the new vacuum vessel. The wall thickness of the high field part was reduced to 0.6 mm thanks to the small surface of the pole (only ∼50 mm × 50 mm). All the transition taper angles were fixed to 0.02 rad.

4.2. Construction

For the construction of the non-magnetic stainless-steel vacuum vessels, it was quite natural to entrust the same manufacturer that realized the series of 37 original SOLEIL dipole vacuum vessels. Back in the 2004–2006 period, many problems were addressed and solved for this procurement. Thus, tools to achieve mechanical tolerances were kept by the manufacturer, and already reused in 2010–2011 for the manufacturing of another specific vacuum vessel. Knowing the minimum delay of nine months, easily extendable to one year, the manufacturing of a the new Superbend vacuum vessel was launched early in May 2019. The final vacuum vessel (Fig. 14) was delivered to SOLEIL in December 2020, with a more extended planning than foreseen, partially because of the Covid-19 pandemic and because the dipole part appeared to be more difficult to machine than predicted.

Figure 14 (Top) high (1) and low (2) field gap footprint on the dipole part of the new vacuum vessel. (Bottom) Zoom of the transition from the high (1) to low (2) field parts.

4.3. Installation

The vacuum vessel was installed in the storage ring during the winter-long technical shutdown of January 2021, by only replacing the standard vacuum vessel. The new vessel was installed first inside the standard electromagnetic dipole as the Superbend dipole was not ready to be installed at that moment (Fig. 15). The vessel was baked-out in situ as is usually done at SOLEIL. Because of the notches made on the dipole part of the new vessel to accommodate the high and low field poles, the Superbend vessel could not be equipped with Kapton heating films for in situ bake-out, so the notches were temporary filled in with aluminium wedges to glue heating thin films on the vessel. In August 2021, before the final installation of the Superbend, the temporary in situ bake-out system was removed from the dipole part of the vacuum vessel (Fig. 16). As the magnetic properties of the PMs deteriorate for temperatures higher than 120°C, if some bake-out is required in the future then the Superbend should be extracted and placed on the support, beside the vacuum vessel, and conventional heating tapes should be used.

Figure 15 The Superbend vacuum vessel installed temporarily inside the standard electromagnetic dipole and equipped with heating films. The upper yoke of the electromagnetic dipole is not yet back in place in the photograph.

Figure 16 The Superbend vacuum vessel installed in the final configuration, ready to receive the Superbend.

5.1. Modeling

The lattice model of the SOLEIL storage ring includes all the magnet multipolar components, the fringing fields of the dipoles and quadrupoles and the horizontal and vertical physical apertures at each element. As the model is considered to be robust owing to comparisons with experimental results, it was mandatory to have a precise modeling of the Superbend field to anticipate linear and nonlinear effects and to compensate for them. Then the 2D measured field mapping was used to evaluate all the multipolar components generated by the Superbend field, from quadrupolar to dodecapolar ones, without neglecting the longitudinal variation of the magnetic field, due to the passage from the `high-field' region to the `low-field' region, that generates some focusing. First the trajectory in the Superbend was deduced from the magnetic field mapping (Fig. 9). Then the multipolar components were calculated around the beam trajectory as a function of the longitudinal position. The longitudinal variation of the quadrupolar component (Fig. 17) led to splitting of the Superbend in seven parts of different length in the model, each part including all the corresponding integrated multipolar components. Another important component is the sextupolar component generated mainly by the high field (Fig. 18). Due to a high value of the vertical beta function at the Superbend location, the vertical beta-beating generated by the focusing of the Superbend reached 15% and was compensated using four quadrupoles located upstream and three quadrupoles located downstream of the Superbend. Although no significant effect on the Touschek lifetime was predicted, the on-momentum dynamic aperture was significantly reduced for negative horizontal amplitudes (Fig. 19) thus inducing a reduction of the injection efficiency.

Figure 17 Variation of the quadrupolar component (solid line) along the beam trajectory in the Superbend. The magnetic field (dotted line) is plotted as a reference.

Figure 18 Longitudinal variation of the sextupolar component (solid line) generated mainly by the high field of the Superbend. The magnetic field (dotted line) is plotted as a reference.

Figure 19 On-momentum dynamic aperture and corresponding frequency map calculated at the injection point. (Top) Without Superbend. (Bottom) With Superbend after compensating for the focusing effect. Simulations were performed with the SOLEIL version of the 6D true symplectic TRACY tracking code. The color bar indicates the tune diffusion rate after 2 × 1026 turns.

5.2. Commissioning of the new vacuum vessel and the new dipole

In January 2021, the storage ring was restarted after the winter shutdown with the new vacuum vessel installed in the standard dipole as the Superbend was not yet ready. This allowed the limitations due to the reduced aperture of the vacuum vessel to be distinguished from those imposed by the new magnetic field. Despite a high pressure measured with the first stored electron beam of small intensity, the vacuum conditioning was very efficient, and the high current of 500 mA could be delivered to the users four days later. The large magnetic gap of the standard electromagnetic dipole allowed many thermocouples to be installed along the new vacuum vessel. As expected from the optimized profile, the analysis of the measured temperatures showed similar behavior to that measured with the old vacuum vessel, without any hot point. The beam loss monitors, installed upstream and downstream of the new vacuum vessel, showed that the reduction of the vertical physical aperture did not localize the losses at this place. The vertical scrapers, set at their nominal position, remain the smallest physical aperture in the whole ring. Further experiments performed by shifting the electron beam vertically in the new vacuum vessel showed that the minimum total vertical aperture was close to the expected value of 12 mm, thanks to very accurate alignment of the new vacuum vessel. The ROCK beamline found the photon beam right where it was before the installation of the new vacuum vessel, as expected. In May 2021, the blades of the X-Beam Position Monitor (XBPM) installed on the ROCK beamline were changed to anticipate the increased power of the photon beam generated by the Superbend. In August 2021, the Superbend was installed in the storage ring tunnel (Fig. 20) and on 25 August the first electron beam was stored in the presence of the Superbend. The temperature measured in the tunnel at the Superbend location was very close to that measured during the magnetic measurements, and, as expected, no field integral correction was required. The electron trajectory was adjusted using only a few tenths of an ampere on the closest dipolar corrector power supplies. This was followed by a series of experiments to characterize and optimize the performance of the storage ring. As expected, the vertical chromaticity was increased by one unit due to the sextupolar component generated by the high field roll-off. Symmetrization of the optics has shown that the quadrupolar component of the Superbend was larger than expected. This can be explained by the fact that the reference axis of the new dipole, registered on the magnetic measurement bench, may be slightly shifted in the horizontal (0.2 mm) at the location of the high field where there is a strong sextupolar component. Indeed, it was difficult to center the magnetic axis in both the high and low field regions. The high field sextupolar component also affects the injection efficiency and the beam lifetime. For the bare machine without insertion devices, the injection efficiency is reduced from 80 to 65%, and, when insertion device fields are varying during the operation, the sensitivity to the betatron tunes has been increased, making the operation less easy than before the installation of the Superbend. Finally, in daily operation, the injection efficiency and the beam lifetime were maintained at around 60% and 8 h, respectively (https://www.synchrotron-soleil.fr/fr/faisceau). Despite the expected decreases in injection efficiency and beam lifetime, the electron beam characteristics at the source points were not changed, neither in terms of transverse dimensions nor transverse position. For the standard mode of operation, the use of a multi-objective genetic algorithm such as MOGA-BMAD (Ehrlichman, 2016) has provided a new set of sextupoles that compensates for the non-linear effects of the Superbend, and since May 2022 the operating performance has returned to pre-Superbend values in terms of injection efficiency and electron beam lifetime. To measure the temperature of the Superbend dipole, temperature probes were glued onto some of the PMs. Fig. 21 shows the temperature variation during a period of two months. The temperature increases from 21°C to 21.5°C in the presence of the electron beam. This increase is due to the proximity of the vacuum chamber which is heated by synchrotron radiation. The small effect of the temperature on the integrated field is seen on the electron closed orbit and is compensated by the Slow Orbit Feedback and, as expected, only the two dipolar correctors close to the Superbend dipole (located in the upstream and downstream sextupoles) are slightly varying. There are no requirements to use the Superbend correction coils.

Figure 20 The Superbend installed in its place in the storage ring tunnel. In red, the frame equipment dedicated to extracting the dipole if a vacuum vessel bake-out must be performed.

Figure 21 Example of the temperature variation in the storage ring tunnel during operation. (Red) Air tunnel temperature. (Black) PM temperature. Temperatures on the vertical axis are given in °C.

A disappointing consequence concerns the so-called `low-alpha' mode which is used in the operation a few weeks per year. The effect on the beam dynamics of the high field sextupolar component is amplified when this optics is used because the value of the vertical beta function at the Superbend location is three times larger than the standard one. Even the use of MOGA-BMAD did not allow a return to acceptable beam dynamics and the experimental tests led to a very poor injection efficiency. This mode of operation has now been abandoned.