Abstract
A new design of spallation target has been installed and operated at the first target station of the ISIS Neutron and Muon Source (ISIS TS-1), as part of the recently completed “TS-1 Project”. Detailed Finite Element Analysis (FEA) simulations were used to guide the design process and predict target performance. Since the TS-1 Project target began operation in November 2022, operating data has been collected and used to validate the target simulation approach. Measured temperatures of 9 out of 10 target plates showed good agreement with FEA simulations of both steady-state and transient behaviour. However, the front target plate temperature was elevated compared to predictions. Because the installed target was now too radioactive to permit hands-on inspection, FEA simulations became an indispensable tool to understand the possible causes and safety implications of this anomalous behaviour. The anomalous elevated temperature appears to be highly localised; a combination of simulations and experiments indicates the mostly likely cause is poor thermal contact between the thermocouple and the bulk of the target plate. In all other respects the target is operating as predicted, and is running reliably at up to 120 kW.
Introduction
The ISIS Neutron and Muon Source has recently completed implementation of the TS-1 Project [4], which involved the redesign and replacement of many key components of the first target station (TS-1). This included a new design of spallation target, for the first time in many years. This represented a significant risk, as spallation targets are challenging components to design, build and operate, and any failures can quickly compromise the reliable operation of a facility. Detailed FEA analysis was relied upon to guide the design process and predict target performance.
The design of the TS-1 Project target has been published previously [6], and is summarised in Fig. 1. Like previous TS-1 targets, the new target consists of a stack of tungsten plates, clad in tantalum for corrosion resistance, cooled by heavy water, and contained in a stainless steel 316L pressure vessel. The key differences from previous TS-1 targets include a reduced number of target plates (10 rather than 12), and a greatly reduced mass of stainless steel in the pressure vessel and cooling manifolds. As a result of this, all 10 plates are now cooled in parallel by a single water circuit, rather than an inlet manifold which divides the flow between plates. A secondary “case cooling” circuit which cooled the surface of earlier TS-1 targets has also been removed, as it was originally included for uranium targets which have not been used at ISIS for many years. Because of these significant changes to the cooling scheme, it is particularly important to assess whether the thermal performance of the new target is as expected.
Implementation of the TS-1 Project is now complete, and the first “new-style” TS-1 Project target began operation in November 2022. Since then, operating data has been collected which can be used to monitor target performance and validate the FEA simulation approach. TS-1 Project targets have a thermocouple at the centre of each target plate, two more thermocouples on the surface of the target vessel, and various sensors in the cooling water circuit including temperature, flowrate, and pressure sensors.
One approach to monitoring target performance from operating data has recently been published by Findlay et al. [3]; a simplified “point mass” thermal model was used to back-calculate heat transfer parameters and heat deposition values for each plate, based on transient temperature responses measured at the thermocouples. The present paper sets out a different and complimentary approach, in which detailed FEA models of the nominal target design are used to predict operating temperatures, and these are compared to observed values. The FEA approach includes more detail than the “point mass” model, but also adds more complexity and makes more prior assumptions. It is hoped that the two methods together will give the best possible assurance that the new TS-1 target is behaving as expected, and a robust error detecting capability if it is not.
Horizontal section through new-style TS-1 target, reproduced from [3].
Daily and cumulative beam current on TS1 Project target from first beam in November 2022 to the end of cycle 2023/4 in December 2023.
The first TS-1 Project target began operation in November 2022, and has now accumulated 294 mA.h of beam current as of the cycle ending 20th December 2023. The history of beam delivery to the TS-1 Project target is shown in Fig. 2. From the earliest operations it has been apparent that the temperature reported by the thermocouple on plate 1 (furthest upstream, adjacent to the beam window) has been consistently higher than expected, and cooling rate has been much slower than expected, as reported previously [3]. Because the installed target was now too radioactive to permit hands-on inspection, FEA simulations became an indispensable tool to understand the possible causes and safety implications of this anomalous behaviour.
A combination of simulations and experiments eventually provided confidence that it would be possible to increase the beam power to the target without compromising safety, as described in Section 6. The major concern had been elevated temperatures on the pressure vessel or beam window, which could have compromised the structural integrity of the target container. However, all such causes were eventually ruled out by simulation or experiment, after which approval was given to gradually increase the beam power on target. A period of successful operation at full repetition-rate followed, with beam power consistently in the range ∼108–116 kW (∼135–145 μA). During June and July 2023, the temperature on plate 1 gradually increased by around 50°C while the temperature on all other plates remained constant. This effect appears to have stopped at the end of June 2023, and plate 1 temperatures have appeared stable since, although there is some indication as of December 2023 that the plate 1 temperature may be beginning to increase again. The target has continued to run steadily since then, with reliable beam delivery to users and continuing gradual increases in beam power up to ∼120 kW (150 μA). The TS-1 target nominally operates at up to 128 kW (160 μA) when running in parallel with TS-2, and is designed to accept a maximum power of 160 kW (200 μA) at times when TS-2 is offline.
Target simulations
Thermal performance of the TS-1 target was simulated using conjugate fluid/thermal analysis in ANSYS CFX, and structural performance was simulated using ANSYS Mechanical [1]. Heat deposition due to the impinging proton beam was simulated in both MCNPX [5] and FLUKA [2], then input to CFX as a cloud of points, allowing the full detail of the heat deposition profiles to be captured in the analysis. In all simulations of heat deposition, the beam profile was assumed to be Gaussian with a nominal beam sigma of 17.9 mm. A view of the CFX analysis geometry is shown in Fig. 3. The boundary conditions are a defined mass flowrate of nominally 8.42 kg/s heavy water at the inlet, and a relative static pressure of zero at the outlet. The turbulence model was Shear Stress Transport (SST). Temperature-varying material properties were applied for all solid materials. Transient simulations of heating and cooling rates were carried out using the same CFX geometry. Transient heating simulations started from a steady state fluid solution with no beam heating. Transient cooling simulations started from the steady state beam-on simulation at the relevant power level.
The exact same CFX setup used for the original design of the target was also employed to reproduce real operating conditions in the target. The only modifications were to allow the inlet flowrate and the proton beam current to be parametrically controlled, so real operating conditions could be reproduced in the FEA simulation. Further modifications to the CFX model were made to test the effect of proposed causes of the plate 1 temperature anomaly. For example, a worst-case blockage in the front channel was simulated by replacing this fluid region with a solid volume having the properties of stationary heavy water.
Left: half-symmetry geometry and mesh of the TS-1 Project target, as used in CFX analyses. The thermocouple geometry is not modelled in detail, but is approximated by temperature measurement probes at the centre of each target plate. Right: example temperature result with thermocouples measurement locations shown.
Several possible explanations for the plate 1 temperature anomaly suggested that the measured high temperatures could be localised to the plate 1 thermocouple and its immediate surroundings. A more detailed mode of plate 1 was created to include the thermocouple geometry, as shown in Fig. 4. There are various thermal interfaces separating the thermocouple from the cooled surfaces of the plate; an unexpectedly poor thermal connection at any of these interfaces could potentially explain the anomalous readings on plate 1.
The target thermocouples are K-types, consisting of MgO insulation inside Inconel 600 sheaths. Each thermocouple is inserted into a tantalum “thimble”, which is bonded to the tungsten plate during the Hot Isostatic Press (HIP) process as part of the target manufacture. The HIP also bonds the tantalum cladding to the tungsten core. The thermocouple is nominally a slightly slack fit in the thimble; the thermocouple outer diameters measure approximately 1.9–1.95 mm and the inside diameter of the tantalum thimble has a tolerance of 2.1–2.2 mm. This thin gap is open to the target cooling channels, and is expected to fill with a small amount of heavy water during normal operation. The thermocouples are inserted horizontally, so it is also possible that a small air bubble could form around the thermocouple tip. The detailed plate model allows an arbitrary thermal resistance to be defined at each interface, so these effects and others can be investigated in the simulations.
Detailed half-symmetry model of plate 1 only, showing the geometry of the thermocouple and the various thermal interfaces which are found between the thermocouple measurement location and the cooled surfaces of the plate.
Thermocouples readings on plates 1–10 (labelled TC101-TC110) as a function of beam current, normalised to temperature difference from the bulk fluid temperature at the inlet. Thermocouple data is compared to CFX simulations using heat input from two different Monte Carlo codes; FLUKA and MCNPX. Note that plate 10 is on a different vertical scale.
A summary of recorded plate temperatures as a function of beam power is shown in Fig. 5. Excluding the anomalous behaviour of plate 1, the agreement between simulated and measured temperatures is good. The scatter in the temperature data is thought to be due in part to small variations in the position and focus of the proton beam. The measured temperatures are generally close to or slightly higher than simulated predictions, except on plate 10 where the measurements are closer to the lower bound of the simulated values. Plate 10 is different to the others as it predominantly functions as a beam stop; the Bragg peak occurs a short way into the plate, as does the peak temperature. This makes plate 10 the only plate where the thermocouple is not at the peak temperature location, and it is also the only plate with a length greater than the diameter. Therefore, it may be expected that the simulations are less accurate for this plate.
Simulations predict that plate 1 temperatures should be very similar to plates 2 and 3. The excess temperature on plate 1 is highest at low beam powers. Plate 1 is the only plate where the temperature is not directly proportional to the beam power, as can be seen in Fig. 5. This indicates that some aspect of the heat transfer from plate 1 is temperature dependent, in a way that does not occur on the other plates. Notably, there is no sudden change in the plate 1 temperature at the boiling point of the fluid (149°C at the target pressure of ∼3.5 barG), and unexpectedly high temperatures occur even when the whole target is below the boiling temperature, so the anomalous measurements do not appear to be a result of boiling.
Figure 6 shows the simulated and measured temperature rise in cooling fluid as a function of beam power. The temperature of the cooling water is measured at the target inlet and outlet. The temperature rise across the target is the most direct available measurement of heat deposition in the target, although the overall temperature rise is small, so the uncertainties e.g. from thermocouple precision are large. The predicted total heat deposition in the target at 140 μA (112 kW beam power) is 71 kW from FLUKA, 63 kW from MCNPX. This 8 kW difference in predicted power corresponds to only a 0.25°C difference in predicted temperature rise, so the uncertainty on the derived heat deposition is large. Both Monte Carlo codes look broadly consistent with the measured data, with FLUKA appearing to reproduce the trend more closely, although the uncertainty from thermocouple precision and other sources is too large to be confident in this conclusion. The inlet and outlet thermocouples are located a short distance of pipework away from the target, so heat transfer to/from the surroundings along this pipework may also affect the measured fluid temperature rise.
Temperature rise in cooling fluid between target inlet and outlet as measured on cooling circuit thermocouples, compared to simulated values from CFX, using heat loads calculated by both FLUKA and MCNPX.
Examples of transient heating and cooling results for plates 1–9 are shown in Fig. 7 and Fig. 8 respectively. Measurements on plates 2 through to 9 show good agreement with simulation results. Note that the thermocouples have an expected response time of ∼0.75 s, which may explain why the measured data consistently lags slightly behind the simulation in both heating and cooling response.
Plate 10 is shown separately in Fig. 9. The rate of temperature rise is lower and the time to reach steady state is much longer due to the lower heat load and higher thermal mass compared to other plates. This means that in some steady state measurements the plate 10 temperature looks unexpectedly low as it has not fully reached steady state.
Simulation and data of heating rates on plates 1–9 of an initially cold target with 140 μA beam current applied. Simulations use FLUKA heat deposition values.
Simulation and data of cooling rates on plates 1–9, from steady state operation with 140 μA beam current applied. Simulations use FLUKA heat deposition values.
Simulation and data of heating and cooling rates on plate 10, with 140 μA beam current. Simulations use FLUKA heat deposition values.
From simulations, the temperatures of the first 3 plates are expected to be fairly similar. This is intuitive, as these 3 plates are of similar sizes and experience similar heat depositions from the proton beam (i.e. all 3 are close enough to the front of the target that the high energy protons have not yet appreciably slowed down). Plates 2 and 3 do indeed show a similar behaviour, indicating that whatever the plate 1 anomaly is, it is fully confined to the front plate. Detailed simulations were carried out at this stage, and confirmed that the temperatures of the other 9 plates appear to be as expected, as do the cooling and heating rates. This ruled out various possible causes which would have affected multiple plates, including problems with the bulk cooling flow or the proton beam position or focus. The two thermocouples on the target surface were also reading as expected. These thermocouples are located 2 mm inside the back edge of plate 2, and halfway between the front and centre of plate 3. Both thermocouples are too far back to directly measure the circumferential temperature of plate 1, but it is reassuring that no elevated temperature was seen further along the target vessel.
During June and July 2023, the temperature on plate 1 gradually increased by around 50°C while the temperature on all other plates remained constant, as shown in Fig. 10. At other times, small temperature changes were seen on all plates simultaneously, e.g. in Fig. 10 there is a sudden temperature increase on all plate thermocouples on July 18th. These changes corresponded with adjustments in beam focus and alignment made by the accelerator group. Unfortunately, the furthest upstream beam profile monitor is not currently working, so the exact beam profile on target is not well known.
Temperatures on plates 1, 2, 5 and 8, and ambient water temperature, during June and July 2023. The order of the plot lines on Jul 31st is the same as the list of thermocouple labels at the top left.
Various experimental methods were also employed to investigate this anomaly. With the proton beam off, the secondary cooling loop was switched off, causing the pumps to gradually heat the water at a rate of a few degrees per hour. The response on the plate 1 thermocouple matched the others in this case, indicating that the thermocouples are working correctly, and the anomaly is somehow dependent on the proton beam heating. This could be because the pump heating is not fast enough to see an effect from the thermal resistance at a bad HIP bond or other poor thermal interface (especially since the water can flow inside the thermocouple wells and therefore eventually heat the thermocouple tip directly).
An alternative explanation was a blockage of the front cooling channel. The measured temperature of plate 2 is as expected, indicating that the channel between plates 1 and 2 must be flowing normally. However, there is no thermocouple on the beam entry window at the front of the target, so the condition of the front cooling channel could not be diagnosed. CFD simulations of this effect showed that a ∼100% blockage of the front channel could produce an effect of the magnitude observed. During a scheduled maintenance period, the direction of coolant flow was reversed in an attempt to flush out any blockage, but this had no effect. An inspection was then carried out by using a remotely handled endoscope to view inside the target, which confirmed that there were no visible blockages in any of the cooling channels. The blocked channel case was of particular concern because it was the only possible cause of elevated plate 1 temperature which could have a negative effect on safe operation of the target, as it would lead to elevated temperatures on the pressure vessel and beam entry window to an extent which could compromise the structural integrity of the target container. Once this case was ruled out by the endoscope inspection, approval was given to begin increasing the beam power on target. The further increase in plate 1 temperature between June and July 2023 increased the temperature anomaly to a level that is now too high to be explained by even a 100% channel blockage of the front channel.
The remaining plausible explanation is that there is poor thermal contact somewhere between the thermocouple and the cooling water. As shown previously in Fig. 4, there are several places where this could occur. The first is at the gap between the thermocouple and the inside of the thimble, which may fill with heavy water or air. Simulations were carried out with a worst-case clearance gap filled with either air or heavy water, but neither added enough thermal resistance to reproduce the measured effects.
The other possible cause of poor thermal contact is a poor bond between tantalum and tungsten, either between the tantalum thimble and the bulk of the tungsten plate, or between the plate and the clad sides. The HIP process has been known to produce incomplete bonds if setup conditions are not perfect, so all target plates undergo ultrasonic inspection of the bonded interfaces before installation. No defects were found on the current plate 1. However, it is not possible to inspect the thimble bond inside the plate. Before HIPing, the plates are electron beam welded under vacuum, hence a failed HIP bond would leave a vacuum gap between the surfaces, introducing a large thermal resistance. Such a gap may be expected to close at higher temperatures, as the tantalum thimble has a higher coefficient of thermal expansion than the surrounding tungsten. This is consistent with the observation that the excess temperature on plate 1 is higher at lower beam currents. The increase in plate 1 temperature over time can be understood in this context as continuing degradation of the bonded interface.
A prototype of plate 1 was manufactured following the same processes and passed the same ultrasonic inspection process as the installed plate 1. This was cut in half by wire electron discharge machining after manufacturing, as shown in Fig. 11. The thimble is entirely missing, indicating it did not bond even though the other interfaces bonded correctly. The surface finish inside the hole appears rough, and possibly discoloured as a result of the spark drilling process. This hole was spark drilled by an external company, whereas later plates have been manufactured in house and have a better surface finish. While this does not conclusively prove anything about the condition of the installed plate, it does show that the manufacturing processes in use at the time could produce a plate which passes ultrasonic inspection but still contains a defect consistent with the observed thermal effects.
Prototype plate 1 after cutting. The tantalum thimble is missing, and the surface finish inside the hole appears rough. The other HIPed interfaces appear to have bonded correctly.
Thermal simulations were carried out with thermal resistance added to the HIPed interface between plate and thimble. It was found that a thermal contact resistance of 4e-4 K.m2/W was required to reproduce the observed temperature at the thermocouple tip. The steady state temperature result in this case is shown in Fig. 12. The peak temperature region is highly localised around the thermocouple tip and thimble. The outer edges of the plate which contact the steel pressure vessel are well cooled, so this case will not pose a risk to the safe operation of the target.
Steady state thermal simulation results for plate 1 with a thermal contact resistance of 4e-4 K.m2/W added to the HIPed interface between plate and thimble.
With the exception of plate 1, the measured temperature results agree well with simulations of both steady state and transient performance. This was found to be the case throughout the length of the target, and across a wide range of beam current levels. This represents a thorough validation of the FEA simulation approach used to design the target, as well as the heat deposition results from MCNPX and FLUKA which the thermal simulations depend on. The heat deposition can be a significant source of uncertainty in predicting the performance of accelerator target systems. In the present case, using FLUKA gave a slightly better prediction of the measured temperatures, although the differences from MCNPX were relatively small.
Heat deposition in the target as a result of the impinging proton beam appears to be in line with expectations. A measure of the total heat deposited in the target was obtained by looking at the temperature rise in the cooling water, as shown in Section 4. The uncertainties are large, but the total heat deposition appears to be in the expected range. A separate measure of the local heat deposition at the centre of each plate can be obtained from the initial rate of transient heating when the beam is switched on. The initial gradient depends only on the heat deposition rate and the properties of the target material. The change of gradient as the system approaches steady state is also influenced by the cooling system. As shown in Fig. 7, the measured rates are in good agreement with simulations throughout the transient heating process.
The target cooling system appears to be performing correctly. The steady state temperature in each plate depends on a combination of the heat deposition rate and the heat transfer to the coolant. Heating rates in Fig. 7 and steady state temperatures in Fig. 5 are in line with expectations. A further measure of the cooling efficiency can be taken from the cooling rate when the beam is switched off. Figure 8 shows that these rates are also consistent with predictions. It is encouraging that both steady state and transient results are in good agreement with simulations, as these are influenced by different parameters of the system. An alternative “point mass” approach to estimating heat deposition and cooling performance from transient temperatures has been published previously [3] and came to the same conclusion; the target appears to be behaving as expected.
Conclusions
Thermal performance of the recently installed TS-1 Project target was measured and compared to predictions from FEA simulations. Steady state and transient thermal simulation results agree well with measured temperatures on 9 out of 10 target plates. The thermal performance of the target appears to be in line with expectations, with the exception of the front target plate, the temperature of which was elevated compared to predictions. A combination of simulations and experiments indicates the mostly likely cause is poor thermal contact at the HIP bond between the thimble containing the thermocouple and the bulk of the target plate. The anomalous elevated temperature appears to be highly localised, meaning it does not threaten the structural integrity of the target and pressure vessel. The target is now running reliably at up to 120 kW.