Overview and first measurements of the MAST Upgrade bolometer diagnostic

Power loss via short wavelength photon emission is an important factor in the overall power balance in fusion devices. Next generation devices will require large fractions of the input power to be radiated, primarily in the divertor, to avoid damage to plasma facing components (PFCs). ITER, for example, is expected to require 60%–70% of the total loss power to be radiatively dissipated, and DEMO will require $>95%$ radiative dissipation.¹ Inferring the power crossing the separatrix by measuring the core radiated power loss sets an important boundary condition for on going divertor physics research and is expected to be a necessary component for the control of future power producing reactors.² 

Bolometers are widely used in fusion experiments to provide absolute measurement of the power radiated by the plasma, integrated over a wide energy range from visible light to soft x-ray emission. The bolometer system implemented on MAST Upgrade (MAST-U) uses arrays of resistive bolometers to provide measurements of the radiated power in both the main chamber and lower divertor of the device. The bolometers will be used in a range of physics studies on MAST Upgrade, such as exploring the effectiveness of the new Super-X divertor (SXD) to dissipate the loss power into the divertor through radiation and charge exchange losses and the presence of impurity radiation from the core plasma and its effect on confinement. A prototype infrared video bolometer (IRVB) that complements the resistive bolometer has also been deployed on MAST-U and is the subject of a companion paper.³ 

This paper describes the resistive bolometer diagnostic as implemented in MAST Upgrade for its first experimental campaign. It is organized as follows. First, the diagnostic hardware is described in Secs. II A and II B, including both the in-vessel sensors and the ex-vessel acquisition electronics. Analysis of the theoretical performance and coverage of the system using a synthetic diagnostic framework is presented in Sec. II C. The first measurements made using the system are presented in Sec. III, illustrating the capabilities and challenges experienced in the first campaign. Future plans to enhance the diagnostic’s capabilities are then discussed in Sec. IV based on the learning outcomes from the initial operation of the diagnostic.

The sensors used in the diagnostic are compact, four-channel integrated units of foil resistive bolometers, first developed in 1991.⁴ Each channel features two foils, one exposed to the plasma radiation and one shielded from it. Each foil is thermally connected to a pair of meander resistors such that the four resistors form a Wheatstone bridge. Plasma radiation heats the exposed foil, which in turn heats the two exposed meander resistors on opposite sides of the bridge: the increase in resistance due to the temperature rise of these resistors unbalances the bridge, and this bridge imbalance is measured by applying an AC excitation voltage across one diagonal of the bridge and measuring the voltage across the other diagonal. This sensor design is in widespread use in tokamaks.⁵ These sensors are grouped into cameras: one camera is sited at the midplane (Fig. 1) and views the main chamber and two cameras view the lower divertor chamber from the baffle (Fig. 2) and the outer wall (Fig. 3), respectively. The midplane camera is referred to as the “core” system throughout this paper, and the lower Super-X divertor cameras are referred to as the “SXDL” system.

The main chamber bolometers use 8 μm thick gold foil absorbers on a Kapton substrate since this was the standard construction of these sensors when the system was installed on MAST in 2001. These foils are sensitive to photons with energies in the range 2.5 eV–10 keV.⁶ The divertor bolometers use 4.5 μm platinum absorbers on a SiN substrate based on the design intended for use on ITER at the time the system was designed.⁷ These foils are sensitive to photons with energies in the range 20 eV–10 keV; they do retain some sensitivity at energies as low as 1.5 eV but at reduced absorption efficiency.⁷ Neither the main chamber nor the divertor bolometers are blackened, so the reflectivity of the foils at lower photon energies needs to be considered when interpreting the data from each system.⁸ In particular, relating measurements between the two systems is complicated by the different reflectivities of gold⁶ and platinum.⁹ 

All bolometer sensors are compatible with being baked to at least 150 °C, which is above the average vessel temperature of 110 °C when MAST-U is baked following a vacuum break. The bolometers are not actively operated during these bake cycles. During the routine operation of MAST-U, the vessel is not heated, so the cameras operate in an environment that is approximately at room temperature.

Figure 4 shows the viewing geometry of the system. A significant fraction of the poloidal cross section of the plasma is covered through the combination of the main chamber poloidal array and the lower divertor arrays. The combination of vertical and horizontal viewing channels in the divertor arrays enables tomographic reconstructions of the measured brightness profiles to produce 2D emissivity profiles in the lower divertor. No vertically viewing poloidal arrays are installed in the main chamber due to the difficulties imposed by the double-null design of MAST-U: the locations where a vertically viewing array would be usefully placed are extremely difficult to access and have insufficient space to site a camera without conflicting with the poloidal field coils. Instead, the main chamber has tangentially viewing channels, which provide radial profiles of the emission at the midplane to complement the vertical profiles of the horizontally viewing core poloidal array. While this does preclude performing reconstructions of the 2D emissivity profile in the core using the same technique as the divertor arrays, it is possible to make some estimations of the 2D profile by combining the profiles from the poloidal and tangential arrays and making assumptions about the structure of the core radiation on flux surfaces although the details of how such analysis may be performed and interpreted are out of the scope of this work.

For many of the plasmas run in the first MAST Upgrade campaign—which were typically low elongation (κ ∼ 1.8–2.2)—the X points are viewed by the main chamber array, though in subsequent campaigns where the vertical control has been optimized it is anticipated that higher elongation plasmas (κ ∼ 2.5) will be typical and, thus, the X points will be closer to the divertor plates and, thus, out of the view of the core array. There is a region around the divertor throat where neither the core nor divertor arrays have coverage. There is also no coverage of the upper divertor for the first MAST-U campaign although an upper divertor bolometer is to be installed for the second campaign.

The spatial resolution of the core poloidal array is ∼5–10 cm with the uppermost four and lowermost four channels having finer resolution than the central eight channels due to the internal layout of the camera. The tangential array has a spatial resolution of 10–14 cm for the co-NBI channels and 15–25 cm for the counter-NBI channels. In the lower divertor, spatial resolutions are ∼5 cm for the vertical channels and 3 cm for the horizontal channels. Since the lines of sight for each camera fan out, the spatial resolution varies with distance from the camera: the values quoted here are with respect to the magnetic axis in the main chamber at R = 0.9 m and the center of the lower divertor chamber at R = 1.2 m, Z = −1.8 m.

The in-vessel bolometer sensors are connected to the data acquisition electronics by screened twisted pair cables manufactured by Habia, which have braided copper shields around each twisted pair and around the 16 twisted pairs carried per cable. The sensors are earthed to the vessel through bolometer cameras, and the screen in approximately one quarter of the length of the cable run is earthed to the same single earth point through the same camera as the sensor. The same cable assembly and materials are used in this portion of the ex-vessel cable, with the exception of an additional polymer sheath on the outside of the cable. The electronics are bonded to the same earth reference as the sensors in a different location, and the remaining three quarters of the length of the cable run screen are earthed through the electronics. This cable run uses Cat 7 cables, which are also doubly screened twisted pair but have aluminum braided outer shielding with aluminum foil screens in the internal four twisted pairs. At the joint between the two parts of the cable run, the earth is broken by using an insulating connector to avoid a ground loop. The level of noise resulting from this electrical arrangement is described in Sec. III D.

The data acquisition system uses recently developed compact electronics based on the field-programmable gate array (FPGA) technology; the electronics is described in detail in other publications.^10,11 This system provides an AC excitation voltage to the sensors, digitizes the output voltage, and performs digital signal processing of the received signal. The drive frequency of the AC excitation is configurable and can be set to any frequency of up to 61 kHz in ∼1 Hz intervals to reduce the impact of narrow band noise pickup. Digital signal processing recovers the output amplitude of the bridge voltage through AC synchronous detection, and the low-pass filters used in this process can be configured with any user-supplied cut-off frequency of up to 5 kHz. This enables the system to optimize, on a shot by shot basis, the trade-off between the signal-to-noise ratio and time resolution. Furthermore, a parallel data pathway in the signal processing pipeline enables real-time calculation of the power incident on bolometer sensors using per-channel deconvolution filters, which makes the system suitable for integration into real-time control systems to provide feedback control of the plasma radiation.

The processed data are recorded at 10 kSPS, enabling a theoretical maximum time resolution of 0.2 ms (5 kHz Nyquist frequency). Such high time resolutions are likely only useful for the specialist study of high intensity, quickly evolving events, such as disruptions due to the increase in noise and the reduction in higher frequency components of the signal due to the thermal inertia of the bolometer chips. Prior work¹⁰ has demonstrated operation of this system using a 1 kHz filter bandwidth, corresponding to a time resolution of 1 ms, and this is the goal for routine operation on MAST Upgrade too.

The use of the FPGA technology has several advantages. The integration of all the device logic onto a single chip and the digitization of the processing pipeline mean that the system is very compact: up to 48 channels can be integrated into a single 1U, 19 in. unit. In fact, MAST-U uses two of these units, each with 32 channels, for the core and divertor systems. A full 64-channel system, therefore, only takes up 2U of cubicle space. The flexibility of the programmable logic and Linux-based firmware in the chip provides a highly configurable system as described above and also enables system upgrades with new features and enhancements without having to replace the hardware.

The electronics also provides a mechanism to calibrate the bolometer sensors in situ. The procedure is described in detail in Ref. 11 and is summarized here for convenience. A DC voltage is applied across both diagonals of the Wheatstone bridge in the sensor such that there is a voltage difference across each of the two resistors in contact with the exposed foil and no voltage difference across the resistors in contact with the reference foil. This arrangement allows a current to flow through the measurement resistors to provide Ohmic heating with a known power. At the same time, the AC excitation voltage is applied and the amplitude of the bridge output is measured. This electrical configuration is shown schematically in Fig. 5.

The change in output voltage amplitude with Ohmic power determines the sensitivity S (in V/W) of the foil, and the time constant of the reduction in voltage when the Ohmic heating is switched off determines the cooling time τ (in s) of the foil. These quantities can then be used in the bolometer equation to calculate the incident power on the foil P, given a measured voltage amplitude V,

The advantage of this method is that no correction for AC effects or foil operating temperature needs to be applied: the foil voltage measurement is made in the same way as in regular operation, making the calibration highly representative of plasma operation.

One other important feature to note is that the AC synchronous detection is done in quadrature. The output voltage from the sensor is of the form

Here, A is the amplitude, ω is the drive frequency, and ϕ is the phase delay due to the capacitance in the signal cables. We calculate the in-phase (I) and quadrature-phase (Q) components of the signal by multiplying it by a reference signal of the same frequency and filtering out the 2ω component,

For integer n, the final two simplifications are exact, but even for non-integer n ≫ 1, the approximation error is small. In practice, this is achieved by ensuring that the filter cut-off frequency is significantly less than the excitation frequency.

From Eqs. (3) and (4), the amplitude and phase can be extracted,

Both the amplitude and phase are recorded. This means that the voltage can be treated as a complex quantity,

Post-processing operations, such as smoothing the voltage, removing offsets—non-zero voltage at zero input power typically due to intrinsic imbalances in the bridge resistances—and differentiating for use in the bolometer equation, can be applied to this complex signal in the same way as traditionally one might treat a real voltage. The amplitude of the resulting voltage and power signals is the physical quantity of interest.

Recording both the in-phase and quadrature components means that no manual hardware correction of the phase is required: the correction can be applied in post-processing. The calibration process enables calculation of the components of the “offset voltage,” and these can be applied during the signal processing pipeline stage if post-shot offset correction is not practical (for example, when the system is used for real-time feedback control).

A synthetic diagnostic model of bolometer cameras has been developed using the Cherab framework.¹² 3D computer aided design (CAD) models of MAST-U—including the PFCs and in-vessel coils—and bolometer assemblies are used to produce the model. The model allows us to calculate the field of view of each channel in 3D, incorporating the volumetric line of sight effects. With this information, we can produce geometry matrices for use in tomographic reconstructions of the plasma emissivity, accurate calculations of the detector étendues, and forward-modeled measurements of synthetic plasma emissivities.

Figure 6 shows the result of calculating the geometry matrix on a grid of toroidally symmetric volume elements (“voxels”) with rectangular poloidal cross sections for a single channel. Each element of the sensitivity matrix is calculated by performing a weighted sum of the path lengths of a distribution of randomly generated rays, which pass through the corresponding voxel: the method is described in detail in Ref. 13. The grid cells closest to the detector have high weights concentrated in a small area, and the weights then reduce as the field of view spreads out further from the detector. Previous work has shown that proper consideration of the finite volume of the detector field of view reduces the need for additional regularization when performing tomographic inversions of measured line-integral brightness profiles and so more accurately represents the spatial resolution of the system.¹³ 

Figure 7 shows the result of summing the geometry matrix weights for all channels in both the core and divertor regions. This illustrates the full volumetric extent of the viewing geometry in a way that the approximation using single lines of sight does not. It complements Fig. 4 and shows that the system covers the majority of the main and lower Super-X chambers, with a gap in coverage around the divertor throat.

An important quantity for calculating line-integral brightness measurements and then tomographically inverting these to obtain the emissivity is the étendue. For each bolometer channel consisting of a slit and a foil in vacuum, the étendue is defined as

Here, dA_s is an element of the slit surface and θ_s is the angle between the slit normal and the vector from a foil surface element to a slit surface element: dA_s cos θ_s represents the projected area of the surface element of the slit on the foil. Likewise, dA_d is an element of the foil surface and θ_d is the angle between the foil normal and the vector from a slit surface element to a foil surface element: dA_d cos θ_d, in turn, represents the projected area of the surface element of the foil on the slit. d_sd is the distance between the foil element and the slit element. The integral is performed over both the slit and foil surfaces. When the foil and slit areas are small compared to the separation between them, it is reasonable to take the approximation that the projected area of the slit is the same at all points on the foil (and vice versa) and also that the distance between the foil element and the slit element is constant, and we obtain the following simplified analytic expression:

Here, A_det and A_ap are the areas of the foil detector and slit aperture, respectively; θ_det and θ_ap are the angles between a vector from the center of the detector to the center of the aperture and the normal vectors of the detector and aperture, respectively; and d is the distance between the foil and the slit.

The simplified analytic expression is a widely used approximation, but we are able to also calculate the étendue without making any assumptions about the relative size and separation of the slit and foil with Cherab using a ray tracing method.¹³ The results of both the analytic approximation and the ray-traced calculation are shown in Fig. 8. Generally, the two methods are in good agreement for the core arrays, but the analytic model over-estimates the étendue for the divertor arrays by ∼5%. The divertor sensors have slightly more rounded openings in front of the foils, which reduces the effective foil area. The étendue of the counter-NBI tangential sensors in the core is also over-estimated by the analytic model compared with the rest of the core sensors. These have a smaller separation between the aperture and the detector, and so the variation in the projected areas and slit–foil separation that is neglected in Eq. (9) becomes more significant. This behavior matches what has been observed in previous work.¹³ Accounting for these effects is important as we seek to minimize the systematic error on the calculation of the brightness measurements, which linearly depend on the étendue.

Using the calculated geometry matrices and étendues, we can forward-model the bolometer measurements for a range of radiation emissivity profiles. The use of synthetic profiles—phantoms—with known metrics, such as the total and mean position and size of the radiation, means that we can assess the quality of reconstructions possible with the system. This technique additionally provides a means of quantifying the uncertainties on calculated metrics, such as the total radiated power, either by directly integrating the reconstructed emissivity profile or using simplified calculations, such as a weighted sum of brightness measurements.¹⁴ 

Uncertainty quantification has recently been performed using this technique on data from the JET tokamak,¹⁵ where the total radiated power metric was computed for a large number of phantoms of varying positions and shapes. Here, we focus on only a couple of phantoms to illustrate the expected performance and limitations of the system with a full systematic scan left for future work. Figure 9 shows an example of a phantom and the inversion of the forward-modeled brightness profile. Gaussian phantoms, such as these, were generated by using an emissivity ϵ parameterized by mean radius r₀, mean height z₀, width dr, and height dz as follows:

This particular example used r₀ = 1.2 m, z₀ = −1.8 m, dr = 0.1 m, and dz = 0.1 m. Synthetic brightness profiles were produced using the synthetic model of the bolometers implemented with Cherab, and then, these were inverted using a non-negative least squares routine with Tikhonov regularization¹⁶ using a Laplacian second derivative operator as the regularization operator as has been done in previous work.¹³ The regularization operator was chosen by hand through an iterative process to minimize the amount of regularization while avoiding the presence of strong artifacts in the reconstruction.

From both the phantom and the inversion, we compute the zeroth order (total power), first order (mean position), and second order (width and height) moments of the emissivity distribution. For this particular case of an “isotropic” phantom (i.e., one that has similar spatial variation in both the R and Z dimensions), the emissivity profile itself and all its moments are well reproduced by the inversion. While not necessarily representative of expected emissivity profiles in the divertor chamber as it is not based on a physics-informed model, this simple Gaussian example does at least provide a useful test of the workflow and suggests that there are no fundamental issues with its capability to reconstruct large scale 2D radiation profiles and report the total radiated power.

To study more realistic radiation profiles, we turn to predictive SOLPS simulations of MAST-U plasmas. In the rest of this subsection, we focus, in particular, on a simulation of an L-mode, Super-X plasma with 5 MW of NBI heating power.¹⁷ We take as the radiation profile the sum of the line emission and recombination for all species in the simulation, which includes all charge states of deuterium and carbon; the impact of charge-exchange neutrals on the bolometers is neglected. Figure 10 shows the results of inverting this phantom using the same regularization operator as in the Gaussian case above. While the total power and mean position of the phantom are rather well reproduced in the inversion, the spatial extent and the structure of the distribution are not. The inversion shows the radiation much closer to the divertor target tiles than in the phantom and with a larger poloidal spread. This is unsurprising as the phantom is very anisotropic, and so attempting to find a solution that maximizes isotropic smoothness is unlikely to lead to a high fidelity reconstruction. However, the good agreement of the lowest order moments indicates that a standard Laplacian regularization operator may still be useful as a general technique when only these global quantities are of interest.

We can improve the reconstruction of higher order moments by exploiting a priori knowledge of the anisotropy of the phantom. Figure 11 shows a reconstruction of the same phantom’s brightness profiles where the isotropic 2D regularization operator has been replaced with one that encourages more smoothness in the direction parallel to the poloidal magnetic field lines than the perpendicular direction. This anisotropic regularization technique has previously been successfully applied to bolometer tomographic reconstructions on other devices.¹⁸ We can see now that not only are the moments of the distribution well-reproduced, but the 2D profile itself is also faithfully reconstructed with the exception of a few artifacts.

The inversions presented so far use regularization parameters hand-tuned for optimal results. This is a labor-intensive effort, which requires a priori knowledge of the result, and, therefore, is not practical for routine analysis. Figure 12 shows the same SOLPS phantom inverted using anisotropic regularization, but using the FTIKREG algorithm.¹⁹ This algorithm performs a Tikhonov-regularized inversion with a regularization parameter, which is estimated using the “self-consistence” method described, which minimizes the discrepancy between measured and back-calculated brightnesses with random uncertainties added: details are given in Refs. 19 and 20. This enables an unsupervised analysis, which is more feasible to perform on a routine basis, say, for an entire experimental campaign dataset. With a nominal estimate of the errors of 0.5%—a best case scenario since real data will likely have higher errors than this, even heavily smoothed—the inversion algorithm produces an emissivity profile whose moments are again in good agreement with the input phantom. The profile itself is rather smoothed out, but this is not unexpected since these highly anisotropic phantoms are pushing the spatial resolution of the diagnostic to the limits. This final example is perhaps closest to what one might expect when routine tomographic inversions of MAST-U data become available.

It is important to note that these examples are representative only and not an exhaustive study of the system’s limits. While the results are encouraging, they rely on a number of idealized assumptions, such as minimal measurement noise, no loss of individual channels, and radiation distributions positioned optimally within the field of view. In reality, it is anticipated that the bolometry may more reliably function as a consistency check for radiation structures, confirming that a modeled emissivity distribution is possible within measurement uncertainty without being able to fully disprove the hypothesized radiation structure in detail. A detailed study of the achievable spatial resolution of reconstructions, as well as the quantification of uncertainties due to factors such as spatial misalignment, real-world-informed noise levels, and missing channels, is planned for future work.

The calibration procedure described in Sec. II B and external references¹¹ is routinely performed in about 1 min before every MAST-U shot. This enables an analysis of the variation of the foil calibration over time as the condition of the sensors, vessel, and external environment (temperature and humidity, for example) evolves. Repeated calibrations also provide good statistics for evaluating the uncertainty in the calibration parameters due to the noise in the voltage and current measurements and imperfect fits of the cooling curves.

Figure 13 shows the bolometer sensitivity and cooling time calibration values, obtained from fits of the recorded calibration data, for many shots over a period of 3 months, for a representative sample of main chamber channels. There are clearly visible differences in the calibration parameters between channels due the variations in the manufacturing process of the individual bolometer chips, with channels 6 and 7, in particular, coming from a much older sensor, which was used to replace previously failed hardware when the array was re-used for MAST Upgrade. However, the individual channels all show only small variations in both the sensitivity and cooling time, with the former having slightly more scatter than the latter across all channels. The scatter provides a useful quantification of the real world uncertainty on these calibration parameters (and therefore on the uncertainty of the measured radiated power). For the period and channels shown in Fig. 13, we obtain fractional uncertainties of $δSS$ = 1%–2% and $δττ=0.1%to0.4%$ = 0.1%–0.4%.

Another advantage of performing the calibration before every shot is that it provides almost continuous monitoring of the state of the system. This means that when faults occur, they can be identified quickly and easily. By way of example, Fig. 14 shows the sensitivity values for a selection of channels in the baffle-mounted camera in the lower divertor. In contrast to the main chamber bolometers, these bolometers show sudden large changes in the sensitivity, which indicate a problem with the channels. Looking at the average measured DC Ohmic heating voltage applied during the calibration, we can see that some of these channels experienced a sudden drop in the applied voltage, which hints at a partial short circuit in the bolometer chip. This was confirmed by multimeter measurements at the vacuum feedthrough after the failures were observed in routine data analysis.

We can pinpoint the date and (in some cases) the first shot number after the failure of these channels, which may aid in the determination of the cause of these faults. For example, sudden calibration changes after a particularly large disruption, or some other off-normal event, such as the plasma divertor strike point position striking unintended surfaces or rapid pressure changes, could indicate the cause of a failure and the likely damage. In this instance, nothing unexpected was observed in plasma shots around the time of the failures of these particular channels, but the capability is still useful to have for future operations.

We present here some examples of measurements routinely made by the diagnostic in the first campaign. A key quantity is the total radiated power; this is calculated for both the core and divertor arrays using a weighted sum of brightness measurements.¹⁴ We report separately the total radiated power measured by the core and lower divertor arrays. For the core, the 16 poloidal channels are used. For the lower divertor, the 16 baffle-mounted, vertically viewing channels were used in the first MAST-U experimental campaign. We would ideally use the wall-mounted horizontally viewing channels,²¹ but hardware failures (see Sec. III E for details) precluded this and necessitated post-campaign repairs.

Figure 15 shows some examples of the total measured radiated power for three different plasma scenarios with both conventional and Super-X divertors and Ohmic and NBI heating. Due to the viewing geometry in the divertor, the measured divertor radiation in conventional divertor shots is very low and almost certainly significantly lower than the total emitted radiation along the divertor leg and near the strike point. The Super-X scenario shows more measured radiation, but again misses a significant amount further from the target toward the X point. Both these shots show low levels of radiated power compared with the NBI-heated scenario as one might expect, given the significant reduction in input power.

Figure 16 shows the core and divertor brightness profiles for an Ohmically heated plasma, which started in a conventional divertor configuration and transitioned to a Super-X divertor configuration, after which a small density ramp was applied from 600 ms until the end of the discharge. Despite being in a connected double null configuration, there is a significant asymmetry in the core radiation, with radiation near the top of the core plasma reducing as the divertor outer leg target radius increases from conventional to Super-X, while the bottom part of the core plasma sustains about the same radiation levels. The movement of the radiation front with a strike point position can also be clearly seen in the divertor profiles although the movement stops at around 400 ms and remains somewhere around 2.7 m (hardware failures here limit the spatial resolution with which we can determine the position of the front). The density ramp in this shot was too small to move the radiation front further up the leg to where it could be detected by the neighboring working bolometer channels.

As described in Sec. II, there are two tangentially viewing arrays of bolometers at the midplane, viewing clockwise and counter-clockwise. During NBI heating periods, there is a significant asymmetry in the brightness measurements recorded with bolometers, even for channels with similar viewing geometries as defined by their tangency radii. Figure 17 shows an example of this behavior for a shot that had significant Ohmic and NBI-heated phases. The large increase in the measured brightness of the clockwise-facing channels, by an order of magnitude, is strongly correlated with the injected beam power. The brightness levels are in good agreement before and after the beam is injected into the plasma. Note that the slow ramp up is due to the smoothing of the bolometer signals and does not indicate an increase in radiation shortly before and after beam injection.

It is postulated that this significant increase in the apparent brightness is due to high energy neutral particles produced by beam–plasma interactions, which are preferentially emitted in the co-beam direction. The effect is seen not only when the on-axis beam is used but also when the off-axis beam is used, which is 60° toroidally and 65 cm vertically displaced from the on-axis beam: this indicates some degree of transport of fast particles before they leave the plasma. These fast ions have large Larmor orbits that drift across the plasma cross section into loss orbits, where they charge exchange with cold neutral atoms. The resulting fast neutrals are no longer confined and are able to reach the bolometer foils, imparting their energy onto the foils as they collide.

An important point to note about these observations is the strong toroidal asymmetry. Power balance calculations that assume toroidally symmetric radiation losses and heat loads would likely miss this power channel loss. It is not possible to determine the total power lost through this mechanism as we are unable to confirm what fraction of the fast particles are actually being detected, but it could potentially represent a significant amount of power, which would have an appreciable effect on the power balance. It could, for example, at least partially explain previous observations of an energy deficit on devices heated by NBI.²² 

The bolometer system has suffered from excessive noise pickup due to the electro-magnetic interference in the first MAST Upgrade campaign. This has limited the time resolution achievable by the diagnostic as heavy smoothing must be applied to achieve an acceptable signal-to-noise ratio. A filter with 15 Hz bandwidth has been used for routine inter-shot analysis in the first experimental campaign, resulting in effective time resolutions of ∼65 ms. While the bolometer system is not the only system on MAST-U to suffer interference, the typically small signal levels (of order 1 mV) mean that it has been one of the more seriously affected diagnostics.

The AC synchronous detection and high performance digital filtering performed by the ex-vessel electronics are effective at removing 1/f noise and sources of interference localized in the frequency domain. However, it is unable to remove broad band noise, which overlaps the frequency domain in which the synchronous detection is performed. Hence, the characteristics of the interference on MAST-U limit the extent to which this interference can be removed from the signal.

Figure 18 shows the noise picked up by a typical bolometer channel for a typical plasma shot. These data were taken with the AC synchronous detection described in Sec. II B disabled, so it represents the raw bridge voltage pickup. The noise appears for the duration of the window when MAST-U’s switched mode power supplies are enabled, from −0.8 to 2.3 s, and is strongly correlated with this enable window. As can be seen from the zoomed time trace (b), the noise is non-sinusoidal, regular but not exactly repeatable and has a rather large duty cycle of almost 50%. This makes it very difficult to use time-domain-based noise removal techniques, such as spike removal.

The bottom half of Fig. 18 shows the spectrum of the noise during the window the power supplies are active. It is a very broad band, with many individual peaks and a noise floor significantly above that in the absence of the power supplies (which sits at around −130 dBm). Of particular note is a strong 6 kHz component with many harmonics (higher harmonics are aliased due to the down-sampling required to acquire these data), corresponding to the switching frequency of MAST-U’s switched mode power supplies. Efforts to identify the route by which this noise enters the bolometer electronics and devise effective hardware mitigations are under way.

We can mitigate the effect of the noise by optimizing the AC synchronous detection. The choice is frequency for the excitation voltage, along with the bandwidth of the low pass filter used for the time integral, determines from which region of the spectrum the diagnostic will show noise pickup. Figure 19 illustrates this, in practice: we take the previously shown raw voltage data and apply a post-shot sinusoidal excitation voltage waveform on top and then perform the synchronous detection in software. At 18 kHz, the frequency of one of the large noise harmonics, the residual noise is as large as 20 mV, but by moving to 22.35 kHz—away from strong line noise sources in the spectrum—the residual noise is reduced by an order of magnitude. This requires no hardware changes, merely a change to the settings on the BOLO8 acquisition system.

Early in the operation of the system, it was discovered that there was a very low frequency (sub-1 Hz) beat voltage on the bolometer channels. This occurred even in shots with no power supplies enabled (such as gas calibration shots), which indicated a different source of noise to that already discussed. Investigation showed that this noise pickup was, in fact, coming from the other bolometer camera. An internal clock is used as a reference when generating the excitation voltage of the requested frequency. This design decision was made in order to enable excitation voltage to be applied continually to reduce thermal cycling of the bolometers in installations where the acquisition clock is only active during the pulse sequence, but it does mean that two different ACQ2106 units will have small variations in their internal clock frequencies due to manufacturing tolerances and environmental variations.

On MAST-U, the main chamber, and divertor arrays are connected to different ACQ2106 units. Both were originally set to use the same excitation frequency, but in reality, there was a sub-Hz difference in the frequencies of the analog output voltages. The excitation voltage from the core system was picked up by the divertor system and vice versa, and after the AC synchronous detection was performed, the frequency mismatch appeared as a low frequency beating on the measured voltages, as shown in Fig. 20.

The solution adopted was to set the core and divertor systems to use different excitation frequencies, far enough apart in frequency that the digital filtering performed as part of the synchronous detection suppressed the resulting beat voltage. We set the two excitation frequencies to be a little over 1 kHz apart and set the digital filter bandwidth to 500 Hz. As can be seen in Fig. 20, this was sufficient to eliminate the beat. The downside of this solution is that it requires two suitable frequency operating points to be available, which limits options in the high noise environment of MAST-U. An alternative solution would be to synchronize the excitation voltage to the same external clock that is used for data acquisition, which would require some firmware modifications to the BOLO8 system and careful treatment of the situation where the acquisition clock is intermittent.

After ∼2 months of routine operations, failures of the wall-mounted divertor bolometer channels were observed. This was first noticed by the failure of the routine calibration performed before every shot, as described in Sec. III A. Subsequent measurements of inter-pin resistances on the vacuum feedthroughs on the vessel showed pins, which should have been connected across the bridge having open circuit resistances. This failure severely limited the capability of the divertor bolometer system and precluded performing tomographic inversions of the radiation emissivity profile, as well as limiting measurements of the radiation in conventional divertor configuration plasmas.

During the engineering break after the end of the first experimental campaign, the wall-mounted divertor bolometer camera was removed from the vessel, disassembled, and inspected. It was noted that the measurement foils appeared to be missing from the bolometer chips, and upon disassembling one of the sensors and extracting the chips, it was ascertained that the foils had broken away from the chips, as shown in Fig. 21. In Fig. 21, all but one of the foils in the four-channel collection of chips failed, leaving a ragged edge where the foil assembly was stretched across and suspended in place.

Failure due to the mechanical shock was ruled out: the foils were designed to withstand greater shocks than the camera had been subjected to during the campaign. A pressure differential causing an unacceptably high force normal to the foil face and pushing the foils out was also considered, but the scale and speed of pressure changes in the divertor during plasma operations are insufficient to cause this failure, and the foils survived the (two) initial pump downs from air pressure to a vessel pressure of 1 × 10⁻⁸ mbar unscathed. Fatigue from thermal cycling due to the plasma radiation was also considered, but the level and duration of radiative power loads on these foils in MAST-U plasmas would have only resulted in about a 1 K temperature rise.

It was ultimately determined that the foils were suffering thermally induced failures due to the elevated operating temperatures, which is generally more of a concern for bolometers used in long-pulse or reactor environments. High operating temperatures in this instance are a result of applying the 18 V AC excitation voltage across the Wheatstone bridge, which continually heats all four resistors and both foils in each channel. Using the measured DC current and voltage in the calibration, it is possible to estimate the resistance of the meander resisters through which the current is flowing, which will to a good approximation be the same as the resistance of all four bridge resisters, using simply R_{foil, active} = V_DC/I_DC. The resistance of the foils with no current R_{foil, idle} was measured regularly throughout the campaign using a multimeter attached to the vacuum feedthrough socket. Along with the temperature coefficient of resistance α, we can determine the temperature rise of the meander resistors (and hence the foils) via the following simple relation:

We take α = 2.4 × 10⁻³ Ω/K, which is a little bit lower than values found in the literature for platinum but similar to those measured during the manufacture of this type of bolometer chip.²³ The calculated temperature rise is shown in Fig. 22 for a representative sample of baffle-mounted divertor, wall-mounted divertor, and main chamber channels. The main chamber channels run coolest as their intrinsic resistance is higher, and so the Ohmic heating current is lower. The baffle mounted channels have a 175 nm Al heat conducting layer applied to improve thermal conductivity between the foils and the heat sink, which also helps them run cooler. The wall mounted sensors do not have this layer, and this means that they run significantly hotter at up to 160 K above ambient temperature. Allowing for a range of α of 2.1 × 10⁻³ to 2.6 × 10⁻⁶ Ω/K resulting from iterative improvements to the manufacturing process over the time range this bolometer chip type was manufactured²³ gives a corresponding temperature rise of between 150 and 190 K for the wall-mounted foils. At the ambient temperatures of 20 °C, this means that the foils are operating at around 200 °C.

Elevated foil temperatures have been shown to lead to premature failure in testing.²⁴ Previous work has demonstrated these types of bolometer chips operating at temperatures in excess of 300 K,^4,25 but these were in baked environments, which have two key differences to the situation encountered by the MAST-U detectors. The first is that the temperature of the baked bolometers was raised slowly, at a rate of around 1 K/h, whereas the Ohmic heating applied to the bolometers by the MAST-U electronics will heat the foils significantly more quickly than this: temperature rises of ∼1 K/s on the measurement resistors were observed with 1 mW of (DC) Ohmic heating during the calibration, and ∼140 mW of Ohmic heating from the AC excitation will cause an even faster temperature rise. The rapid and wide temperature cycling would have occurred repeatedly as channels were plugged in and unplugged or the electronics are powered off and on during commissioning.

The second important difference is that the baked bolometers tested previously were in thermal equilibrium with their surroundings as both the foil chips, the sensor, and the surroundings are heated together. In contrast, the Ohmic heating due to the AC excitation specifically heats the bolometer chips but not the surroundings. This results in large thermal gradients, which are associated with higher thermal stresses. These thermal gradients will be exacerbated by the fast temperature rise, compounding the stresses placed upon the chips. We therefore theorize that it is the thermal stresses caused by these elevated operating temperatures of the chips in otherwise ambient temperatures, likely exacerbated by the rate of change of temperature, which caused the premature failure of these sensors.

The failed bolometer chips were replaced and the sensors were reinstalled in the vessel during the engineering break. In order to reduce the risk of a repeat of these failures, we have replaced the chips with the ones that do feature the heat conduction layer, which should reduce the operating temperature. As an additional precaution, we have halved the amplitude of the excitation voltage, which will reduce the Ohmic heating by a factor of 4 and should greatly reduce the operating temperature. This does come at the expense of signal levels, so we hope to raise the voltage amplitude back up to the maximum after demonstrating that the operating temperature is safe.

Despite the challenges encountered during the first MAST-U experimental campaign, the bolometer system has produced useful measurements of the radiated power, which are enabling analysis of a range of experiments. Using the knowledge gained from these early operations, a number of improvements have been or are planned to be implemented.

Repairs to failed channels and adopting a more cautious operating regime should enable us to achieve the full spatial resolution for which the system was designed and infer information about the radiation emissivity profile in the lower divertor using the tools described in Sec. II C. Modifications to the ex-vessel cabling and earthing have been made during the engineering break with the aim of reducing the electrical interference, in parallel with the investigative work on reducing the noise at its source. This work should help us to improve the achievable time resolution of the system with an acceptable signal-to-noise ratio.

To address the lack of coverage of the lower X point, development of integrated analysis techniques using the resistive bolometers and MAST Upgrade’s infra-red video bolometer (IRVB), which has a viewing geometry optimized for X point coverage, will be developed. This should provide better estimates of the total radiated power than either diagnostic can achieve in isolation.

Finally, a new bolometer camera has been installed in the upper divertor. This new camera is designed so that the lines of sight mirror those of the wall-mounted lower divertor camera, as shown in Fig. 23. This will be known as the “SXDU” camera. By designing the camera to have the same viewing geometry relative to the divertor leg in both the upper and lower divertors, we can make measurements of the up/down radiation asymmetry by comparing both the line-integral brightness profiles and an estimate of the total divertor radiated power. The camera is located in sector 1, the same as the main chamber arrays and immediately adjacent to sector 12 where the lower divertor bolometer are located, to minimize the impact of any toroidal asymmetries (particularly during transient events).

The camera design is based on that of the lower divertor, as shown in Fig. 24 for comparison with Fig. 3. Modifications to the design were required to permit a different mounting position resulting from the different locations of port on the upper and lower halves of the machine: the lower bolometer is mounted to the port plate, whereas the upper bolometer is mounted directly on the vacuum vessel inner wall. The new camera does, however, feature the same sensor arrangement, aperture shape, and anti-reflection internal components as the lower camera. The sensors used feature gold foils; this does pose a challenge for interpretation as the different reflectivity of platinum and gold in the lower and upper divertors, respectively, at low photon energies must be accounted for. However, it was decided that the less brittle gold foils represented a lower risk of failure than the platinum foils, and the interpretation complexity was considered an acceptable price to pay for what is expected to be improved robustness of the system.

In addition to direct up/down radiation asymmetry measurements, the addition of the upper divertor bolometers improves the ability to perform whole device power balance measurements. Radiation in the upper divertor is likely to be a significant power loss mechanism in many MAST-U plasmas, particularly up/down symmetric double null and upper single null Super-X plasmas, which are standard MAST-U operating scenarios, so this new installation contributes significantly toward the goal of full device radiation coverage. With the main chamber, lower divertor, upper divertor, and IRVB diagnostics, the whole plasma poloidal cross section has radiated power coverage with the exception of the upper X point. Future diagnostic development will prioritize adding this last piece of required coverage.

MAST Upgrade features a resistive bolometer system to measure the spectrally integrated power radiated by the plasma. A poloidal array in the main chamber measures the core plasma radiation and up/down asymmetries, while tangential arrays provide radial profiles of the core radiation. Two arrays of bolometers in the lower divertor chamber provide measurements of the divertor radiation and can be inverted to produce 2D poloidal emissivity profiles. Bolometer measurements are digitized using FPGA-based smart, compact electronic units.

The main chamber and lower divertor systems were commissioned for the first experimental campaign and provided routine radiated power measurements. Difficulties were encountered with high noise levels and in-vessel hardware failures, with investigation, repair, and mitigation undertaken throughout the campaign and during the engineering break after the end of the campaign.

The repaired lower divertor system is complemented by a new upper divertor camera, providing up/down asymmetry measurements and improved whole-device radiated power measurements.

This work was supported by the U.S. Department of Energy under Award Nos. DE-AC05-00OR22725 and DESC0014264. This work was partly funded by the EPSRC Energy Programme (Grant Nos. EP/T012250/1 and EP/W006839/1). To obtain further information on the data and models underlying this paper, please contact PublicationsManager@ukaea.uk. Support for M. L. Reinke’s contributions was, in part, provided by Commonwealth Fusion Systems. The authors would like to thank Stefan Schmitt, Hans Meister, and Matthias Bernert for discussions of possible failure modes of the sensor shown in Sec. III E.

The authors have no conflicts to disclose.

J. Lovell: Conceptualization (equal); Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Project administration (supporting); Software (lead); Validation (lead); Visualization (lead); Writing – original draft (lead); Writing – review & editing (lead). M. L. Reinke: Conceptualization (supporting); Formal analysis (supporting); Funding acquisition (lead); Investigation (supporting); Methodology (supporting); Project administration (equal); Resources (equal); Supervision (equal); Writing – review & editing (equal). A. R. Field: Conceptualization (equal); Funding acquisition (supporting); Investigation (supporting); Methodology (supporting); Project administration (supporting); Resources (equal); Writing – review & editing (equal). B. A. Lomanowski: Methodology (equal); Project administration (equal); Resources (supporting); Writing – review & editing (supporting).

Raw data were generated at MAST Upgrade. Derived data supporting the findings of this study are available from the corresponding author upon reasonable request.