PSFC/JA-16-60

Linear probe drive system with real-time self-adaptive position control for the Alcator C-Mod tokamak

D. Brunner, A.Q. Kuang, B. LaBombard, and W. Burke

October 2016

Plasma Science and Fusion Center Massachusetts Institute of Technology Cambridge MA 02139 USA

This work was supported by DoE Contract DE-FC02-99ER54512 on Alcator C-Mod, a DoE Office of Science user facility. Reproduction, translation, publication, use and disposal, in whole or in part, by or for the United States government is permitted.

Linear servomotor probe drive system with real-time self-adaptive position control for the Alcator C-Mod tokamak D. Brunner, A.Q. Kuang, B. LaBombard, and W. Burke

[email protected]

Plasma Science and Fusion Center, Massachusetts Institute of Technology, Cambridge, Massachusetts, 02139, USA A new servomotor drive system has been developed for the horizontal reciprocating probe on the Alcator C-Mod tokamak. Real-time measurements of plasma temperature and density—through use of a mirror Langmuir probe bias system—combined with a commercial linear servomotor and controller enables self-adaptive position control. Probe surface temperature and its rate of change are computed in real time and used to control probe insertion depth. It is found that a universal trigger threshold can be defined in terms of these two parameters; if the probe is triggered to retract when crossing the trigger threshold, it will reach the same ultimate surface temperature, independent of velocity, acceleration, or scrape-off layer heat flux scale length. In addition to controlling the probe motion, the controller is used to monitor and control all aspects of the integrated probe drive system. 1 Introduction Plasma probes, i.e., measurement instruments that make direct contact with the plasma, have been one of the primary tools to study boundary plasma physics in experimental fusion reactors. Probes enable excellent temporal resolution (up to ~1 MHz with present technology1) and spatial location (sub- millimeter) of many quantities of interest: ion and electron temperatures, plasma density, plasma potential, flow velocities, and magnetic field. Reciprocating systems allow probes to measure and survive plasma in conditions too intense for stationary probes. Fast reciprocation of the probe into and out of the plasma limits the probe exposure to high heat flux. The reciprocations are typically fast enough to take advantage of ‘inertial cooling’, i.e., a situation in which only a thin surface layer is heated. In this case, active cooling of the probe will not enable deeper scanning. It is desirable to scan the probe as deep as possible. However, the exponentially increasing heat flux profile a tokamak scrape-off layer presents a considerable challenge, requiring accurate control of the probe trajectory – in particular the insertion depth. There are two effects that limit scan depth: (1) Melting the probe should be avoided. Melting can cause the probe tip to deform (making the collection area and thus the inferred density uncertain), shorting of the electrode (rendering it useless), and termination of the tokamak plasma (through a large impurity injection). (2) Even before melting, as the surface temperature is raised the probe can undergo thermionic electron emission. The electrons emitted from the probe reduce the sheath potential (rendering plasma potential measurements uncertain) and increase the net current collected during ion saturation measurement (making plasma density measurements uncertain). Fortunately, enhanced electron emission occurs at a surface temperature that is significantly below melting temperature. Thus, the onset of electron emission can often be used as a marker to identify the maximum useful scan depth without irreversible damage to the probe. Reciprocating probe systems have been implemented and used on many plasma physics experiments. A comprehensive list would include nearly every magnetic fusion experiment. For a partial list see Ref. 2. Most systems have relied on a pneumatic or hydraulic drive to reciprocate the probe into and out of the plasma. Pneumatic/hydraulic drives have the advantage of being a simple system with few components.

1 They are relatively easy to implement and can readily achieve velocities (~1 m/s) as well as accelerations (>100 m/s2) that enable effective operation. The major disadvantage of pneumatic/hydraulic systems is the control of their ultimate plunge depth. This is typically done by trial and error. The plasma boundary shape from a previous plasma pulse is taken into account and the position of the pneumatic cylinder that drives the probe is adjusted (typically with a slow ) such that it scans deep enough to measure interesting plasma physics but not so deep as to destroy the probe. In order to minimizing the dwell time at the probe’s end-of-stroke (during maximum heat flux), the C-Mod pneumatic systems additionally engage a mechanical return spring that provides a large turn-around acceleration (~600 m/s2). Some other systems rely solely on toggling air or hydraulic pressure to the opposite side of the cylinder to change probe direction, resulting in a much lower turn-around acceleration and shallower scan depth into the plasma. Neither of these pneumatic systems have sufficient time response for precise, real-time control of probe position in narrow scrape-off layers. Electromagnetic systems, however, can have sufficient performance for real-time control. Electronic circuits can respond fast enough (<1 ms) to turn the probe around in time. Electromagnetic control of probe movement can been implemented in three ways: (1) Using standard (either linear3 or rotary4) with a series of permanent to provide a background field; (2) Using the magnetic field of the experiment to provide the background field5–10; (3) Using pulsed electromagnets to induce currents in a ‘drive hoop’11. In first two cases a current is passed through a coil or series of coils and the Lorentz force between the current in the coil(s) and the background magnetic field provides the acceleration. In the third case the image currents in the ‘drive hoop’ act against the pulsed electromagnets. Both an electromagnetic, the so-called ‘pecker-probe’, and a hydraulic reciprocating probe with real- time control were operated on Tore Supra10,12. The output of a real-time calculation of the magnetic equilibrium was used to control the ultimate depth of each probe scan. This is not a viable technique for use on Alcator C-Mod; the magnetic equilibrium reconstruction of the plasma boundary and the probe- measured boundary plasma can drift over the course of a plasma shot (it is presumed that this is due to currents redistributing in the copper toroidal field coils as they heat up). That is, for an equilibrium with a constant boundary plasma position at the outer plasma-wall gap, the probe-measured plasma temperature and density profile can drift away from the probe. The profiles can drift a few millimeters— a few heat flux decay lengths. That size of variation in the actual plasma position relative the calculated equilibrium does not provide for a useful feedback technique. Since probe insertion depth is practically limited by surface temperature, a better approach is to control the probe depth based on real-time measurements of probe surface temperature. Experiments indicate that this should be feasible: an IR camera was used to measure the heat flux on a reciprocating probe on ASDEX-U13 and surface thermocouples14 could be integrated into a probe. A less direct, but perhaps simpler solution, would be to use real-time measurements of the plasma quantities at the probe tip combined with a thermal model of the probe to calculate surface temperature. The Mirror Langmuir Probe1 (MLP) bias system provides real-time measurements of plasma temperature and flux. An MLP is an analog computer that quickly switches (~3 MHz) through three bias states of a Langmuir probe I-V (current-) trace. The system optimizes each bias voltage through use of circuitry that mirrors the Langmuir probe response in the plasma. A by-product of this is real-time

2 outputs proportional to the three fit parameters to the I-V: ion saturation current (Isat, proportional to the plasma flux), electron temperature (Te), and floating potential (Vf, related to the plasma potential through sheath theory). To complement the MLP analog computer we have developed an analog computer that ‘solves’ a thermal model of the Langmuir probe body, outputting the surface temperature evolution of the probe head in real time. It uses real-time MLP Te and Isat measurements to compute the incident plasma heat flux, based on a simple sheath heat flux model. The scanning probe system must be capable of diagnosing a wide variety of scrape-off layer plasma profiles. As discussed below, this makes it impractical to use surface temperature alone as a parameter to trigger a turn-around response. Fortunately, the scrape-off layer profiles are close to exponential. In this case, a fairly reliable estimate of the surface temperature at a future time can be made based on the current temperature and its time derivative. Using these two parameters, we’ve found a robust, universal control algorithm that is able to trigger the probe’s turn-around such that its surface temperature stays below a specified peak temperature. This paper is organized as follows. Section 2 explores the optimization of probe acceleration and velocity for maximizing the depth of probe scan into the plasma while ensuring controllability. A universal control algorithm is developed in Section 2. Section 3 describes the mechanical design and implementation of the linear servomotor system. Section 4 covers the control (software and electrical) interface, including the surface temperature analog computer. Section 5 describes operational results of the linear servomotor drive system on Alcator C-Mod. 2 Thermal performance The thermal performance is the crucial concern when considering designing a reciprocating probe system with feedback control. The exponentially increasing heat flux profile of the boundary plasma presents a formidable challenge: The surface temperature of the probe results from the time integrated heat flux as it traverses across the profile. For time scales typical of a probe scan (few milliseconds) the probe surface behaves as semi-infinite body with the temperature rise localized to the surface15. Each millimeter deeper into the plasma causes a surface temperature increase that is larger than all previous. Since the probe must retrace its path out of the plasma, the peak surface temperature does not occur until after its maximum insertion depth. Thus, in order to avoid melting the probe, the control system must issue a return command at a time when the surface temperature is well below the peak temperature desired. In the following subsections, we examine the optimization of velocity and acceleration as well as the required response times needed to attain a target peak temperature. This is done with the help of a thermal model for the probe head exposed to an exponentially-increasing heat flux profile. Based on this, we construct a universal control algorithm that can project the peak surface temperature based on the present value of the surface temperature and its temporal derivative. 2.1 Optimizing velocity and acceleration Ideally, for feedback control of a reciprocating probe system, the return command should be issued at a time that is very close when the probe reaches its peak surface temperature, minimizing uncertainty in what happens to the probe in the ensuing time interval. This requirement pushes the drive system towards attaining the maximum acceleration available. The requirements for velocity optimization, as we shall see, are not so clear.

3 Naïvely one may assume that simply moving the probe as fast as possible through the plasma boundary would minimize the chances of melting and optimize the scan depth. In the limit of high acceleration and low velocity, this does help to maximize the ultimate depth. However, for a given acceleration, the time between when the probe begins to slow down and when it changes direction increases linearly with initial velocity. Since the turn-around signal must be generated before peak insertion is attained, increasing the velocity increases the difficulty of implementing a feedback control system. To quantitatively assess this, we have undertaken a survey using a simple 1D, semi-infinite thermal model for a tungsten electrode. The surface temperature evolution (Ts) was systematically calculated for various initial probe velocities (v = 0.5 to 5.0 m/s), accelerations (a = 100 to 1,000 m/s2), and exponential scrape-off layer heat flux scale lengths (λ = 1, 4, and 10 mm). The probe trajectory is modeled as having a constant insertion velocity, followed by constant acceleration, and finally constant retraction velocity:

− − �� ; � < − � � = �� ; − < � < . (1) − + �� ; < � The heat flux to the surface of the probe is assumed to be a simple exponential profile:

(2) � � = �� . Solving the heat conduction equation with Eqns. 1 and 2 must be done numerically. There is no closed 16 analytic solution (although there is a series solution ). For each simulation, the peak heat flux q0 was adjusted such that the peak temperature reached the melting temperature of tungsten (3,695 K; it started at room temperature, 300 K). An example of the trajectory, surface heat flux and temperature are shown in Figure 1.

4000 surface temperature [K] 3000 2000 1000 0 surface heat flux [GW/m2] 0.8 0.6 0.4 0.2 0.0 probe position [mm] v=1.5 m/s 20 a=600 m/s2 15 λ=4 mm 10 5 0 -15 -10 -5 0 5 10 15 time [ms] Figure 1 Example of a typical probe trajectory for the old C-Mod pneumatic system and a calculated surface heat flux and temperature history. The results for each velocity, acceleration, and heat flux decay length are plotted in the top panels of Figure 2. The depths are indicated as relative to the distance for the C-Mod pneumatic system (v = 1.5 m/s and a = 600 m/s2). From these graphs, it is evident that there are diminishing returns for

4 increasing either the velocity or acceleration over certain ranges. There is a tradeoff on what factors dominate the time spent at the highest heat fluxes. At slow velocities, the turnaround time is short and the time spent in high heat flux is dominated by the slow-moving probe; here there is no advantage in increasing the acceleration to increase the scan depth. At high velocities, the time spent at the highest heat fluxes is dominated by the amount of time it takes to turn the probe around; here there is no advantage in increasing the velocity to increase the scan depth. Optimizing the velocity and acceleration to increase the maximum scan depth provides only a marginal benefit. It would only change the ultimate depth by a fraction (~1/4) of the heat flux scale length over the ranges of practical interest considered. For example, if the pneumatic system velocity was doubled, it would increase the maximum scan depth by <0.25 mm in the λ = 4 mm case. This is not enough to make an appreciable difference to physics that will be accessible. It is the shape of the heat flux profile that limits the maximum scan depth. Narrower heat flux scale lengths allow for scanning to higher peak surface heat fluxes. There is a factor ~2 difference in the peak surface heat flux obtainable for 1 and 10 mm heat flux scale lengths. λ = 1 mm λ = 4 mm λ = 10 mm acceleration [m/s2]

0.5 depth relative to pneumatic [mm] 800 0.5 2 pneumatic system 600 0.0 1 0.0 -0.1 400 0 -0.2 -0.5 -1 200 -0.3 -1.0 -3 -2 -1.5 0 peak surface heat flux [GW/m2] 0.7 0.6 800 1.3 0.9 0.5 1.2 600 0.4 1.1 400 0.8 1.0 0.6 200 0.9 0.8 0.6 0 1500 500 2500 1500 500 2000 1000 3000 3000 2500 2000 1500 1000 800 2000 1000

600 500

tdelay 3.0 ms 400 2000 0.3 ms 2500 500 2000 1000 minimum temp 1000 200 1500 500 1500 500 to trigger return [K] 0 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 5 velocity [m/s] velocity [m/s] velocity [m/s] Figure 2 Survey of the optimization of acceleration (vertical axis) and velocity (horizontal axis) for three different heat flux scale lengths (columns). Top row is the ultimate depths into the plasma relative to the ultimate depth of the old pneumatic probe system. Middle row is the peak surface heat fluxes needed to get the peak surface temperature to the melting point of tungsten. Bottom row is the surface temperature at the time to trigger the turn-around signal for two delay times (0.3 and 3.0 ms). Optimization of velocity and acceleration matters primarily in assessing feasibility of feedback control. This comes in two parts: (1) the amount of time it takes to accelerate the probe, (2) the delay time between the feedback signal reaches its threshold and the application of the acceleration to turn around the probe. Two time delays (tdelay = 3.0 and 0.3 ms, based on operating modes of the probe controller discussed later) were examined for the latter point.

5 The threshold temperature for the return signal is determined for each velocity, acceleration, and heat flux decay length by taking the time of the beginning of the acceleration pulse, subtracting the delay time, and taking the surface temperature at that time. This is plotted in the bottom panel of Figure 2. As the heat flux decay length gets smaller, the threshold temperature gets smaller. For some of the range considered (higher velocities and lower accelerations) the return threshold temperature is so low (< 500 K) as to be useless. The turn-around signal would need to be triggered as soon as the probe experiences a modestly-small temperature, which would result in a large uncertainty in the projected peak surface temperature. The simulations indicate that there is an optimum for the velocity: one that allows for sufficient margin on the return temperature threshold yet does not restrict the maximum depth scanned into the plasma. Given that the return temperature threshold is much more sensitive than the ultimate scan depth to the velocity, a slower scan velocity is better, allowing for more margin on the threshold temperature. The servomotor system in this paper has an acceleration of ~250 m/s2 and operates in heat flux scale lengths of ~1 to 4 mm. In this region of operational space, the ultimate scan depth is insensitive to the velocity. As such, we have chosen a relatively slow velocity of 0.5 m/s to maximize the margin on the turn- around temperature. From these plots, it is clear that the delay time between the feedback signal and application of the acceleration makes a large difference in the threshold temperature. For a 1 mm heat flux decay length, a 3.0 ms delay time is useless. The probe drifts too far into the steep heat flux profile before the system can react. However, a 0.3 ms delay or less is serviceable, allowing for a turn-around command at a significant fraction (~1/2) of the melting temperature. Of special note is a recent ‘ultra-fast’ coil gun reciprocating probe system for SUNIST11, with a peak velocity of 21 m/s and average acceleration of 11,000 m/s2. Since the SUNIST pulse length is short (<20 ms), it was necessary to have the probe move that fast to scan through ‘stationary’ plasma profiles during the discharge. And it was necessary to have such an enormous acceleration to ensure that the probe turns around before scanning too deep. In cases where the discharge duration is much longer than the reciprocation duration a system with this large of velocity would not be a large advantage over slower systems. However, such a large acceleration with a ~10x slower velocity would be advantageous for its fast turn-around. 2.2 Universal trigger condition As shown in Figure 2, there is not a unique temperature threshold that can be used to generate a turn- around trigger. It has a strong dependence on the plasma heat flux scale length. Searching for the possibility of projecting forward the temperature time history from its present value, we have found a unique relationship between the change in surface temperature (DTs, the difference between the temperature at a given time and the starting temperatures) and its time derivative (dTs/dt) at the time at which a turn-around signal must be issued. It accommodates all heat flux scale lengths, velocities, and accelerations considered. Based on this observation, we construct a universal trigger condition – a turn- around trigger should be issued when the probe’s trajectory in this space crosses a well-defined threshold line. Following the procedure of the previous subsection, we explore the space of velocity (0.25 to 2 m/s), acceleration (50 to 600 m/s2), and heat flux scale length (0.5 to 5 mm). We again adjust the peak heat flux (q0) so that the peak surface temperature reaches the melting temperature of tungsten (3,695 K), this 6 time tracking the change in temperature relative to room temperature. In this case, we examine both the surface temperature and its time derivative at the time that a turn-around signal must be issued. The results are shown in Figure 3. Again, it is clear that high acceleration, low velocity, and long heat flux scale lengths result in a temperature at turn-around closest to the ultimate temperature. If the temporal derivative of the surface temperature is normalized by the time to turn around (v/a) we find a remarkable result: there is a universal line defining a turn-around trigger condition in [DTs, dTs/dt (v/a)] space, bottom panel of Figure 3. If the turn-around acceleration is triggered when the probe trajectory crosses this line, the probe will reach the same ultimate surface temperature, independent of velocity, acceleration, or heat flux scale length. A useful analogy for tokamak physicists is the heat flux factor used for ELM heat pulses in the divertor.

The ultimate temperature rise only depends on the total energy in the heat pulse (E0), the time over which it is spread, and the shape of the heat pulse. For the simple case of a uniform heat pulse the -1/2 temperature change as a function of time as T~E0t . This can be multiplied by a shape factor for arbitrary pulse shapes17. Some insight into the what determines the shape factor for the case considered here can be gained by recasting the surface heat flux in non-dimensional form:

� ; � < −1 � � = � ; −1 < � < 1 , (3) � ; 1 < � where � = � �, � = �� �, and � = � ��. The shape of the heat pulse is entirely determined by the value of �. It is unfortunate that there is not an analytic solution for this problem, as the numeric solution hides the parametric dependences of this trigger threshold line. Recognizing that (1) we do not have the parametric dependencies of the turn-around line and (2) the thermal model will only be qualitatively accurate due to simplifying assumptions (mainly the sheath heat flux transmission coefficient), the turn-around trigger threshold line is approximated as a simple quadratic curve in [DTs, dTs/dt (v/a)] space for the purposes of implementing it in the control system.

7 2000 2 acceleration 50 [m/s ] heat flux scale 1500 length [mm] 0.5 dTs 1 [K/s]1000 2 dt 3 4 500 5 velocity [m/s] 0 2.0 2000 1.5 acceleration 200 [m/s2] 1.0 0.75 1500 0.5 0.25 dTs[K/s]1000 dt 500 0 2000 acceleration 400 [m/s2] 1500 dT s[K/s]1000 dt 500 0 2000 locus of points defining a universal 1500 trigger condition dT v s [K]1000 dt a 500

00 1000 2000 3000 4000 ∆Ts [K] Figure 3 Results from 216 simulations scanning velocity, acceleration, and heat flux scale length. These values shown occur at the time that turn-around acceleration is applied and result in the ultimate surface temperature change (DTs,max) being the same. Colors indicate different velocities and symbols indicate different heat flux scale lengths. Top three plots show the trends due to different accelerations in the value of the surface temperature (DTs) and its temporal derivative (dTs/dt) at the time the turn-around acceleration is applied. Bottom plot with the temperature derivative normalized to the time it takes to deaccelerate (v/a). Remarkably, all values are clustered along a universal line. Figure 4 shows four simulated trajectories using parameters similar to the new servomotor system 2 (actual system: v = 0.5 m/s, a = 250 m/s , λ ~ 1 to 4 mm) scanning into heat flux profiles with q0=1 2 GW/m with all adjusted to scan to the same ultimate surface temperature change, DTs,max. Even with factors of 2 changes in the velocity (green) or acceleration (orange) from the nominal case (red) only result in small (<0.5 mm, < λ/4) changes in the ultimate scan depth and relatively large (+/- ~1000 K) differences in the turn-around temperature. The value of going to lower velocities is clear in comparing the green (slow) and red (fast) lines for fixed values of acceleration, the turn-around temperature is much closer to the final temperature for the slower velocity. Similarly, the faster acceleration (red) has a higher turn-around temperature than the slower (orange). Wider heat flux scale lengths (purple) results in shallower scan depths but higher turn-around temperatures. Interestingly, although not surprisingly, constant ratios of v/λ result in identical temperature time histories (green and purple).

8 position [mm] 0 -1

-2

-3 -4

∆Ts [K] 4000 3000 nominal 2000 v = 0.5 m/s a = 250 m/s2 λ = 2 mm 1000 λ×2 v = 0.5 m/s 0 a = 250 m/s2 λ = 4 mm dTs/dt*v/a [K] v/2 2000 v = 0.25 m/s 1500 a = 250 m/s2 λ = 2 mm 1000 a/2 v = 0.5 m/s 500 a = 125 m/s2 λ = 2 mm 0 -500 -15 -10 -5 0 5 10 15 time [ms] dTs/dt*v/a [K] 2000 trigger 1500 time 1000 500 0 -500 0 1000 2000 3000 4000 ∆T [K] s Figure 4 Comparison of 4 simulations at parameters similar to those in the servomotor system covered in this paper. Top panel is the trajectory. Second panel is the history of the change in the surface temperature. Third panel is the temporal derivative of the surface temperature normalize to the turn-around time v/a. And the bottom panel is trajectory is normalized surface temperature derivative versus change in surface temperature. The diamonds mark out the ‘turn-around trigger’ points at which time the trigger threshold is passed (dashed gray line) and the acceleration is applied. In the bottom panel time increases in clockwise trajectories as indicated by the black arrow; the dots on each trajectory mark out a cadence of 1 ms intervals.

3 Mechanical system This mechanical system for the servomotor-actuated reciprocating probe is nearly identical to the previous pneumatic system on Alcator C-Mod, having been designed from the outset to be a drop-in replacement for the old system. Since the servomotor track was longer than the pneumatic cylinder, the only major change to the system was a 0.33 m increase in length and new brackets to account for the additional weight of the servomotor system. The whole system takes up an approximately 0.43 m wide by 0.23 m tall footprint extending 2.6 m radially from a horizontal vacuum

9 port and weights ~90 kg. Two vertical aluminum U-channels provide gravity support for the system and take up minimal floor space. The probe system is electrically isolated from the vertical supports to prevent electrical ground loops. It is connected to the Alcator C-Mod vacuum chamber by a 2.75” CF flange.

SMA feedthrough magnet track "fast" bellows forcer (a)

probe signals; forcer current, sensors, & cooling "slow" bellows (b) guide cables tube lead screws

stepper motor

pneumatic ionization vacuum pump pulley gate valve gauge (c) hand gate valve

vacuum air

Alcator C-Mod electric isolation vacuum vessel (d)

detachable probe head linear bearings rail fixed stand Figure 5 Simplified diagram of the major mechanical and vacuum functions for the servomotor probe system. Blue outline highlights the moving parts for each step. (a) Fast motion: servomotor actuates probe through plasma. (b) Slow motion: stepper motor actuates probe system between scan rest position and gate valve. (c) Probe behand gate valve. (d) Hand motion: probe system slides back to allow replacement of probe head.

magnet track

forcer

carriage

bearings

1 cm

Figure 6 Internal assembly of the servomotor system. Optical encoder is behind the carriage. Nut and threaded rod on the left attach to the feedthrough to couple motion to the components internal to the vacuum system. 10 servo- motor

stepper motor

potentiometer

ionization gauge

10 cm drive tube

probe head

Figure 7 Assembled probe system. The compact design (2.6 m length, 0.26 m width, 0.23 m height) connects to the torus vacuum chamber via a 2-3/4” CF flange. The mechanical system has three primary functions: (1) fast motion of the probe into and out of the plasma while maintaining vacuum; (2) slow motion of the probe between its rest position and ready-to- scan position; (3) isolation of the probe vacuum chamber from the main vacuum chamber and removal of the probe. Each function is explained in detail below. 1. Fast, Plasma Scanning Servomotor Motion The system for fast reciprocation of the probe has the following requirements: provide fast (v ≈ 500 mm/s), repeatable (Δx < 0.1 mm) motion (d < 120 mm) with large acceleration (a ≈ 250 m/s2) while maintaining Ultra High Vacuum (UHV) and transmitting 4 signals from the probes to atmosphere. The fast motion into and out of the plasma is performed with the linear servomotor, Figure 6. The servomotor subsystem is contained within a precision-machined aluminum U-channel to maintain alignment among the components. The linear servomotor is an Aerotech18 BLMH-142 model with a 480 mm long permanent magnet track, allowing for up to 200 mm of travel. The servomotor forcer, containing the electromagnet coils that interact with the permanent magnets, is attached to a custom aluminum carriage. Carriage motion is constrained to be linear with a pair of THK19 SSR ball cage linear bearings. A matching rail for the bearings is bolted to the U-channel. Mechanical limits with sorbothane bumpers are placed at extremes in motion to prevent mechanical interferences and over- extension/compression of the bellows (future versions of this system will use more deliberately- engineered end-stops to absorb energy in worst-case faults, likely a spring in the front and shock absorber in the rear). A Renishaw20 RGH22 optical encoder (Δx ≈ 0.02 mm) is also located within the aluminum carriage. The corresponding encoder strip as well as home and limit position indicators are mounted to the side of the U-channel. The servomotor carriage is connected by a high-strength aluminum rod and a titanium yoke to a 2.75” CF vacuum feedthrough with 4 SMA plugs. This is then reduced down to a 1.33” CF vacuum flange and

11 is mated to the ‘fast’ bellows. The ‘fast’ bellows are custom BellowsTech21 edge-welded stainless steel bellows with a 0.550” ID and 6.0” stroke. The other end of the ‘fast’ bellows is secured to the stepper motor carriage. The probe tube is welded to a flange mounted between the ‘fast’ bellows and the reducer. A high- temperature POCO carbon bearing at the front and Teflon bearings in the middle ensure smooth, purely linear motion of the probe tube through its guide tube. At the end of the probe tube is a custom threaded- fitting to which various scanning probe heads are mounted. There is an asymmetric key pattern in the fitting to fix the probe orientation around the scanning axis. Four ceramic-insulated, stainless steel-sheathed 50 W coaxial cables run from the SMA feedthrough down the probe tube. At the probe-end the center conductors of the coaxial cables are terminated with beryllium copper plugs. The plugs are electrically isolated from the probe tube and maintained in position with a zirconia spacer. The air-side 50 W coaxial cables from the SMA feedthrough and the servomotor power and sensor cables as well as a cooling air tube are constrained in an Igus22 cable tray. The cooling air tube provides compressed air (~12 psi) to ensure that the forcer does not overheat if left on for an extended period. When the probe is not scanning, the carriage (attached to the forcer, bearings, and aluminum push rod) is held in place with an Alpro23 EB250KO security deadbolt lock. This lock can be controlled either locally with a lever or remotely through digital signals. The lock not only provides personnel protection against unwanted servomotor motion, it also holds the probe in its rest position against the force from the pressure differential on the vacuum flange when the coil is powered off. The system requires ~1.5 A to maintain the forcer in position against the pressure differential when the brake is disengaged. The magnetic ‘in-position’ sensor for this lock failed in the ambient magnetic field of Alcator C-Mod; the position of the forcer measured with the optical encoder was used to confirm that it was in place and the bolt could be engaged. Future versions of this system will likely use a clamping friction brake to avoid mechanical interference faults possible with a deadbolt type lock along with a mechanical limit switch to ensure the probe is in the proper position when locked. 2. Slow, Positioning Stepper Motor Motion The purpose of the slow (~10 mm/s) positioning system is to move the probe from its rest position behind the gate valve to its ready-to-scan position at the end of the guide tube while maintaining vacuum integrity. For the old pneumatic system, the placement of the slow positioning system also determines the final scan depth of the probe because the pneumatic cylinder always had the same scan distance. Since the servomotor does not have a fixed scan distance, this level of control is no longer necessary. All of the elements of the servomotor ‘fast’ system described in the previous section are contained within an aluminum carriage. The carriage is actuated with a pair of lead screws connected by pulleys and a timing belt to a stepper motor. The motor and screws set the carriage position relative to the outer probe frame and vacuum vessel. Limit switches wired to the stepper are placed at extremes in motion to prevent mechanical interferences and over-extension/compression of the bellows. Position of the stepper motor carriage is measured with a linear potentiometer. Vacuum motion of this system is maintained by a ‘slow’ bellows, which is another custom BellowsTech21 edge-welded stainless steel bellows with an 1.000” ID and 23” stroke. The moving end

12 of the ‘slow’ bellows is mounted on the opposite side of the stepper motor carriage to the fixed end of the ‘fast’ bellows. The other end of the ‘slow’ bellows is mounted to the outer probe frame. Between the ‘slow’ bellows and the probe is another guide tube, which is constrained to linear motion by a high-temperature POCO carbon bearing at the front and a pair of Teflon bearings in the middle and rear. 3. Roll-Back, Maintenance The purpose of the roll-back system is to allow removal and replacement of the probe head in atmosphere while maintaining UHV within the main vacuum chamber. When it is time to change the probe head, the stepper motor pulls the carriage all the way to the rear limit switch, this places the probe behind the pneumatic gate valve. The gate valve is then closed, isolating the main vacuum chamber from the local probe vacuum. Dry nitrogen is flowed through the probe system from a valve by the turbo pump to prevent water vapor build up while it is opened to atmosphere. Next the vacuum flange on the probe side of the pneumatic gate valve is opened. Opening this vacuum flange releases the outer probe frame from its secured position to the vacuum vessel. It can then be rolled back by hand on a pair of rails and the four linear bearings that are affixed to the probe stand. The frame can be rolled back far enough that the entire probe head is free from the vacuum tubes. The securing nut is then loosened, the old probe is removed, and a new probe is secured to the reciprocating drive. If the probe has exposed electrodes, electrical continuity is checked between the electrode and the airside cables. When the new probe is attached, a new copper gasket is placed on the vacuum flanged and it is then bolted and sealed. The probe vacuum system has a dedicated vacuum pump that is then used to evacuate the probe chamber. An ionization pressure gauge indicates when the probe chamber vacuum pressure is sufficiently low to open the pneumatic gate valve to the main vacuum chamber. A probe head can be changed in ~30 minutes and is typically pumped down to 10-5 torr, sufficient to open it to the main vacuum chamber, within 2 hours. New probe heads are stored under vacuum to minimize absorbed water and pump down time on installation. 4 Control system The use of a commercial servomotor and controller enabled quick development of the new system. In order to expedite development time, we purchased a turn-key system from Aerotech based on the ACT165DL-0200 linear , Soloist HPe50 servo motor controller, and associated connecting cables. After gaining experience with this system, the actuator was disassembled and the subcomponents reconfigured for Alcator C-Mod probe drive. As delivered, the controller is already setup to run the forcer coils and is interfaced with the encoder and sensors. The user only needs to write a simple script (in Aerobasic, a BASIC-like language specific to Aerotech controllers) to perform the desired actions. In addition, the controller used here came with 16 digital inputs, 16 digital outputs, 4 analog inputs, and 4 analog outputs. All of these I/O channels are available within the scripts running on the controller. This allows for a simple scripting of the real-time feedback system. In addition, it allows for the controller to oversee all operation of the reciprocating probe system, including the stepper motor, gate valve, and vacuum pressure gauge. This section describes the interface of the controller with the Alcator C-Mod data system, including a GUI for high-level user control and interfaces with the MDSplus data structure. A schematic of the interfaces is given in Figure 8.

13 MDSplus tree Python GUI Numeric Nodes Probe Scan Parameters Named Numbered Global Global Variables Variables Move Stepper Read Sensor Motor States Action Nodes Run Load Numbered Probe Global Variables Scans

Servomotor Controller Vacuum Gauge, ASCII Ethernet Gate Valve, User-Created Communication Stepper Motor Controller, Software Interface Stepper Motor Brake, Numbered Limit Switches, Global Potentiometer, Variables Solenoid Lock, Dictionary Cooling Air Valve, Mirror Langmuir Probe, Named Thermal Response Circuit, Global Shot Trigger, Variables External Go/Return, Stepper I/O bus Auxilliary I/O Servomotor Internal Sensors Probe Scans Control System Servomotor Forcer Figure 8 Diagram of the interface among the Python GUI, MDSplus tree, servomotor controller, and the servomotor system. 4.1 Servomotor controller The servomotor controller is an Aerotech Soloist HPe50 with the optional IO expansion board. It is powered by two separate circuits, a 3-phase 208V 30A-fused circuit for the forcer coils and a 120V for the controller. This allows the 3-phase power to be cut to the motor during an emergency stop while maintaining operation of the controller. All communication to the controller is through the local Ethernet. Within the controller is a system that interfaces with the servomotor forcer and sensors. This internal control system has three modes of operation available to the user: programmable position, velocity, or current. Within the user software a mode is selected and commands are sent to the internal control system. The internal control system does all of the ‘hard work’ of feedback control of the forcer based on the sensor signals, ensuring that it maintains the user-programed position trajectory, velocity or current. The ASCII Ethernet communication interface is the hub of the controller system. During the setup, the controller is assigned an IP address on the local network. It is then accessible by TCP through a port (UDP communication is also available). The ASCII interface to the controller accepts various strings that launch actions on the controller; e.g., executing servomotor motions, reading and writing to I/O ports, and launching user-loaded scripts. After receiving the ASCII string, the controller returns a string 14 giving the state of the action (i.e., fault, invalid, success, terminating, or timeout) followed by an ASCII string containing information (e.g., the value of an I/O channel or the system fault state) if appropriate for the user sent string. 4.2 Python GUI

Figure 9 Screenshot of GUI used to control servomotor system. A GUI was written in Python to control the system Figure 9. Python was chosen because it is a widely used open source high-level programing language with an MDSplus interface24. This allows it to be easily ported to other labs independent of their local software site licenses. The GUI is designed to be a high-level user interface, giving the operator access to all of the actions and information necessary to control the probe yet ensuring that the system can only be operated in a safe manner. At the servomotor controller level, it is too easy for an inexperienced operator to command the system to perform an unsafe task (e.g., send the forcer into the mechanical limits at the full 50 A current). Thus, it’s crucial to have this safe control interface. The GUI interfaces directly through the ASCII interface to the controller. This allows it to read and reset controller fault states, read and set I/O states, control the gate valve separating reciprocating probe vacuum chamber from the main vacuum chamber, and control the stepper motor position. Some of these controls are done directly through the ASCII interface to the I/O bus or the internal control system and some call user-created scripts on the controller that further process information. In all cases, sufficient interlocks are in place to ensure that it only operates safely, e.g., to prevent the gate valve from shutting

15 on the reciprocating probe shaft, the gate valve will only close if the forcer is locked in position and the rear limit switch is engaged (the rear limit fails open to ensure that there are no false-positive engaged readings). The GUI also controls the programing of the scans for the next plasma pulse. There are two graphic panels related to the scans displayed to the operator. The first displays the plasma boundary trajectory and the probe scan trajectory from a shot (typically the previous) along with the programmed probe trajectory for the next shot. The second displays the surface temperature and its time derivative calculated by the analog computer for the same shot as the trajectory plot, along with the user-set threshold boundary to issue a return command. There are two scan modes available to the user, each with its advantages: plunge or position control. The plunge control mode makes use of the controller’s current control mode. This mode had the advantage of nearly instantaneous (~0.3 ms) response of the applied current, allowing for the quickest available return. Its disadvantage is a lower precision in probe position control. Through this interface, the user is able to set the scan times, maximum scan depths, and the option to have the feedback control algorithm decide when to turn around. Under feedback control, the probe will scan with its peak insertion at approximately the user set time. While scanning, the feedback script on the controller monitors the probe position as calculated by the internal control system as well as its trajectory in [Ts, dTs/dt (v/a)] space based on the calculated surface temperature evolution. When one of these two values (position or universal trigger) reaches the user-set limit the software will issue a command to the internal control system to turn on the reverse current, full- throttle (50 A). There are two features to ensure fail-safe operation: If the probe electrode is shorted, it draws a very large current, which results in the analog computer outputting an enormous heat flux. This causes the probe to turn around almost immediately once it enters the plasma with a shorted electrode. Also, the user sets a minimum probe temperature signal level corresponding to a certain probe insertion distance. If the signal does not reach this level at when it crosses that distance, as would occur if the probe electrode is open-circuited or the MLP system fails, the probe is instructed to turn-around immediately. In the position control mode, the user sets knots in time and space for the probe to move between. The probe will trace a trajectory through these knots with set velocities and accelerations. This mode makes use of the controller’s ability to precisely track the position and velocity of the forcer with the optical encoder. This allows for a very accurate scan trajectory and is good for ‘dwelling’ the probe in the far boundary plasma where the heating is not as intense. It also allows for more complicated ‘dwell and poke’ modes to whatever complexity the operator desires within the limits of the given velocity and acceleration. The biggest limitation of position control mode is the delay in response time (~3.0 ms) between the software issuing a position change command (i.e., a turn-around from the feedback script) and the internal control system applying the current to the controller. This is because the system must interrupt its programmed trajectory to execute a new trajectory. As shown in the thermal analysis section, this delay is too large to be useful for setting a return command. So, although the thermal feedback technique could be used in position control mode, in practice it is not responsive enough due to this delay.

16 4.3 MDSplus tree interface MDSplus is the standard software for data acquisition and control for the majority of the magnetic fusion experiments24,25. Data is communicated between the GUI and the controller through a MDSplus tree by way of numbered global variables. This is due to a limitation of the ASCII interface, which does not allow communication by named variables. The ASCII interface only allows for assignment of values to a set of global integers and doubles. This communication works as follows: the GUI writes the scan parameters to named nodes in the tree (e.g., scan times and maximum depths). The tree then has a dictionary, created automatically by a script based on the tree structure, that links these named nodes to a series of numbered global nodes. Then, during the init phase of the next pulse, an action node is triggered that calls a script to load the numbered global variables from the tree into the controller through the ASCII interface. When any user-created software on the controller is run, it first calls a local dictionary program (created by a script every time the MDSplus tree structure for the controller is changed) that translates the numbered global variables into named global variables. The named global are then used in the user software. The logic of this is that any time a developer makes changes to the system the software needs to only be updated in two places: the node references within the tree and the dictionary on the controller. This allows both the GUI and controller software to contain named variables, which makes for a much easier to read code than generic numbered variables. It also minimizes the number of places one needs to update when changes are made and the likelihood of an error while making changes. If the GUI, tree, and controller all directly communicated with the generic numbered variables, then the developer would need to search (or keep track of) the numerous places throughout the code where a variable is used. Besides the scan parameters, the variable system also includes all of the information on the assignment of the I/O channels. For the analog I/O’s it simply has assigned channel numbers to named variables, e.g., the vacuum pressure gauge voltage is on analog in channel 0. For the digital I/O’s it has assigned ports, bits, and true values (due to system differences some are high for true and some are low for true) for each channel, e.g., the shot trigger is on port 1, bit 3, and its true value is high. Having this information within the tree allows information on the I/O assignments to not only be used by the controller, but the GUI also has easy access to it. As with the other variables, any time a change is made to I/O channels, only the tree and the controller dictionary need to be updated. In addition to facilitating the flow of control variables, the MDSplus tree also initiates the command to the controller to launch the user-created script to control the probe scans. 5 Surface temperature analog computer As discussed in Section 2, the parameter of interest for controlling the reciprocating probe movement through the plasma is the peak surface temperature on the probe. Initial attempts at programing the

Soloist servomotor controller to digitally calculate the surface temperature from real-time outputs of Isat and Te reported by the Mirror Langmuir probe system were not successful. The time between calculations was found to be too long to generate a useful control signal. So, we created an analog computer that solves a 1-D heat transport model in real-time, computing probe surface temperature from surface heat flux, using the Mirror Langmuir Isat and Te analog data signals as input. First, a real-time signal of surface heat flux is calculated from the MLP signals. The ion saturation current is transformed into the ion saturation current density (�,) assuming a constant probe projected 17 area (1 mm2, which is accurate within 10% throughout the probe scan). The surface heat flux is 26 calculated through sheath theory (� = ���,), assuming a sheath heat flux transmission coefficient � = 7. This calculation does not take into account secondary electron emission, non-unity ion-to-electron temperature ratio, or probe bias; all of which will increase the surface heat flux (ion-to-electron temperature ratio is almost always greater than unity in the SOL27). As such, although the system reports quantitative values of heat flux and surface temperature, the reported magnitudes are likely smaller than reality. The main goal of this system is to provide qualitatively accurate time responses. Operational set points (i.e., the shape of the turn-around threshold line) are determined empirically based on real scanning conditions. The surface temperature analog computer is an adaptation of the analog computer that was developed to compute real-time heat fluxes based on surface temperature measurements in C-Mod28. This computer (Figure 10) makes use of the direct analog between the thermal and electric diffusion equations where voltage corresponds to temperature and current to heat flux with appropriately chosen conversion factors. Thermal diffusivity of the RC-nodes is set to match that of tungsten at 400 K. The node spacing is optimized with small nodes at the front to resolve the large temperature gradients from the surface heat flux and increasing node spacing to the back to accurately simulate the long-term temperature evolution. The Langmuir probe electrodes are electrically isolated from the probe head with mica insulation, which also acts as thermal isolation. Thus, the plasma-deposited energy on the electrodes only slowly transfers to the rest of the probe head and the bulk temperature of the electrodes ratchets-up through the course of successive scans through a plasma shot. To ensure that this is accurately captured, the total length of the nodes is set such that the total effective thermal mass of the simulated Langmuir probe matched that of the actual probe. A numerical simulation (Figure 11) indicates excellent agreement between very finely spaced nodes and the optimized 8-node network, both in fast-time response and the long-term ratcheting up. Using the optimized node spacing greatly reduces the size of the RC-network.

R /2 R0/2 R0/2 R1/2 R1/2 R2/2 R2/2 n I V q ... T C0 C1 C2 Cn

Figure 10 Schematic of the RC-network used as an analog computer of the surface temperate evolution. A current proportional to the surface heat flux is applied to the first node. The resulting voltage is proportional to the surface temperature.

18 4000 ∆Ts [K] 3000

2000

1000 fine node spacing 0 optimized 8-nodes 800 dTs/dt×v/a [K]

400

0

-400 0.6 0.8 1.0 1.2 1.4 time [s] Figure 11 Comparison of thermal simulations of three successive probe scans with a very fine node spacing and an 8-node simulation with optimized spacing. The optimized spacing (small in the front and growing ~2.2x towards the rear) accurately matches both the short time- scale temperature rises and the long-term temperature evolution.

19 6 Use on Alcator C-Mod The servomotor system was integrated onto Alcator C-Mod during the summer 2016 experimental campaign. The feedback control system was not fully integrated until late in the campaign. For much of the time the system was used in a feed-forward manner, programing the ultimate depth of each scan individually. This in itself was a large improvement over the old pneumatic system, which allowed only a uniform depth for all scans. Each scan was individually tailored to the plasma profile to reach the same ultimate depth (i.e., plasma temperature) into the plasma. There was no need to fine-tune the plasma equilibrium through the plasma control system as was required with the pneutatic system, saving many shots at the beginning of an experiment run-day. The feedback control system proved to be very useful when doing single scans, as shown in Figures 12 and 13. Four successive scans are shown. The plasma equilibrium was adjusted between shots for the first three scans (green, orange, and blue), pulling the plasma further away from the outer wall and reciprotacting probe. With each scan the feedback system allows the probe to plunge deeper until it finds the proper level in the turn-around plane to return. A fourth, repeated shot is also shown, having the same plasma equilibrium as the previous shot (blue and pink). Here, the excellent repeatability of the system is evident, turning the probe around at esetntially the identical plasma conditions.

r [mm] 120 110 100 v [m s−1] 0.5 0.0 0.5 − Isat [A] 1.0 0.5 0.0

Te [eV] 60 30 0

∆Ts [K] 1500 1000 500 0

dTs/dt*v/a [K] 400 200 0 200 − 0.49 0.50 0.51 0.52 0.53 time [s]

20 Figure 12 Comparison of four successive scans where the feedback system was used to turn the probe around. The circles indicate the time the acceleration was applied in each case. The equilibrium was changed between the first three scans (green, orange, and blue), pulling the boundary away from the probe. With each scan the probe plunged deeper to reach the same ultimate depth with respect to the plasma. The equilibrium was not changed between the third and fourth scans (blue and pink). Here we see excellent repeatability between the two scans.

Figure 13 shows the four scans in [DTs, dTs/dt (v/a)] space along with the programmed threshold line. The circles indicate the time at which the current was applied to turn the probe around. The delay in time between passing through the threshold line and the current being applied ranged from ~0.3 to 0.6 ms. The minimum amount of time between the command to apply the turn-around current and its application is ~0.3 ms, as determined in offline testing of the control script. The remainder of the spread is likely due to variability in the timing of the control script running on the servomotor controller, since it takes a finite amount of time to execute (this is challenging to determine in practice).

600 dTs/dt*v/a [K] 500 400 trigger 300 200 100 time 0 100 − 200 − 0 500 1000 1500

∆Ts [K]

Figure 13 Comparison of the four scans from the previous plot on the normalized temperature derivative versus temperature plane along with the turn-around trigger control boundary. The circles indicate the time the turn-around acceleration was applied. There is a ~0.3 ms delay from crossing the trigger boundary to the earliest turn-around and ~0.3 ms variation on the delay in turning around. Figures 14 and 15 show one of the important limitations of this feedback control scheme. In this case, it was used on three successive scans within one plasma pulse, spaced 300 ms apart. With each scan, the feedback control system scans to shallower depths. This is because the probe is starting successive scans pre-heated with a non-uniform temperature profile through the body from the previous scans. The universal trigger threshold condition was developed for a probe that starts out at room temperature with a uniform temperature throughout its body and does not account for this pre-heating. A simple way of overcoming this may be to have different threshold lines for each scan. However, this would still depend on the details of the temperature history within the probe body. Another, more complicated method would be to have fast numerical thermal simulations of the probe body. At each time step the simulation would project what the ultimate temperature would be, given extrapolations from existing conditions. Such a system would be robust to differences in past temperature history and its strength would rely in the accuracy of a forward model to extrapolate the heat flux. In any case, since the probe starts out hotter in successive scans, it may not be able to safely penetrate to the same depth for each scan.

21 r [mm] 120 100 80 v [m s−1] 0.5 0.0 0.5 − Isat [A] 1.0 0.5 0.0 Te [eV] 60 30 0

∆Ts [K] 2000 1000 0 dTs/dt*v/a [K] 400 200 0 200 − 0.6 0.8 1.0 1.2 1.4 time [s]

Figure 14 Use of the feedback system with three successive scans in one shot. Each successive scan goes shallower due to the inability of the present algorithm to account for the temperature history of the previous scan.

22 dTs/dt*v/a [K] 400 trigger 300

200 time 100

0

100 − 200 − 0 500 1000 1500 2000 2500 ∆T [K] s Figure 15 Three successive plunges, the begging of each marked with an arrow, during one shot on the normalized temperature derivative versus temperature plane. Having the temperature ratchet up provides an additional challenge to the control system, the present algorithm is not able to reach the same ultimate temperature for each scan. 7 Conclusions We have developed a new servomotor controlled reciprocating probe system for the Alcator C-Mod tokamak. The system is largely based on the original C-Mod pneumatic scanning probe system. The servomotor allows for both independent scan depths as well as adaptive feedback control of the probe position based on real time measurements of the plasma conditions at the probe tip. The latter being a significant advance in reciprocating probe capability. Through the development of this system we have examined in detail the issues of feedback control of a probe through an exponentially increasing heat flux profile. We’ve found that there exists a universal trigger condition for initiating the turn-around acceleration. It results in the same peak surface temperature of the probe – independent of the probe velocity, acceleration, or the heat flux scale length in the plasma. This condition corresponds to a well-defined curve in the plane of the temporal derivative of the surface temperature normalized to the turn-around time (v/a) versus the surface temperature. When the probe trajectory crosses this curve, the turn-around command should be issued. Practical use of this control algorithm shows good results when using it for single probe scans. However, for multiple scans in close succession, the control algorithm exhibits limitations due to pre-heating of the probe. Acknowledgements The authors wish to express their gratitude to the following people for the helpful discussions: Bob Mumgaard, Mark Chilenski, Christian Haakosen, and Rui Viera at the PSFC along with Klaus Koster and Jeremy Donatell at Aerotech. Thanks to Kevin Cole for his work on solutions to the heat equation for a reciprocating probe system. Thanks to a reviewer for pointing out the parallel with ELM heat flux factors. Thanks to the entire C-Mod staff for providing some of the most challenging plasma conditions in which to operate probes. This work was supported by DoE Contract DE-FC02-99ER54512 on Alcator C-Mod, a DoE Office of Science user facility.

23 References 1 B. LaBombard and L. Lyons, Rev. Sci. Instrum. 1 (2007). 2 J.A. Boedo, N. Crocker, L. Chousal, R. Hernandez, J. Chalfant, H. Kugel, P. Roney, and J. Wertenbaker, Rev. Sci. Instrum. 80, 123506 (2009). 3 J.M. Haas, A.D. Gallimore, K. McFall, and G. Spanjers, Rev. Sci. Instrum. 71, 4131 (2000). 4 W. Zhang, J.F. Chang, B.N. Wan, G.S. Xu, C.J. Xiao, B. Li, C.S. Xu, N. Yan, L. Wang, S.C. Liu, M. Jiang, and P. Liu, Rev. Sci. Instrum. 81, 113501 (2010). 5 C.S. Pitcher, H.-S. Bosch, A. Carlson, A. Field, A. Herrmann, J. Neuhauser, T. Richter, and W. Schneider, in Proc. 20th EPS Conf. Control. Fusion Plasma Phys. (European Physical Society, Lisboa, 1993), p. 291. 6 D. Desideri, G. Serianni, V. Antoni, M. Bagatin, C.S. Pitcher, and L. Tramontin, Rev. Sci. Instrum. 70, 403 (1999). 7 N. Smick, B. LaBombard, and C.S. Pitcher, J. Nucl. Mater. 337–339, 281 (2005). 8 N. Smick and B. LaBombard, Rev. Sci. Instrum. 80, 23502 (2009). 9 A. Schmid, A. Herrmann, V. Rohde, M. Maraschek, H.W. Müller, and the ASDEX Upgrade Team, Rev. Sci. Instrum. 78, 53502 (2007). 10 J.P. Gunn and J.-Y. Pascal, Rev. Sci. Instrum. 82, 123505 (2011). 11 W. Liu, Y. Tan, W. Wang, and Z. Gao, Rev. Sci. Instrum. 87, 11D437 (2016). 12 J.P. Gunn, J.Y. Pascal, F. Saint-Laurent, and C. Gil, Contrib. to Plasma Phys. 51, 256 (2011). 13 M. Kočan, H.W. Müller, B. Nold, T. Lunt, J. Adámek, S.Y. Allan, M. Bernert, G.D. Conway, P. de Marné, T. Eich, S. Elmore, F.. Gennrich, a. Herrmann, J. Horacek, Z. Huang, a. Kallenbach, M. Komm, M. Maraschek, F. Mehlmann, S. Müller, T.T. Ribeiro, V. Rohde, R. Schrittwieser, B. Scott, U. Stroth, W. Suttrop, and E. Wolfrum, Nucl. Fusion 53, 73047 (2013). 14 D. Brunner and B. LaBombard, Rev. Sci. Instrum. 33501, 33501 (2012). 15 D. Brunner, B. LaBombard, R. Ochoukov, and D. Whyte, Rev. Sci. Instrum. 84, 33502 (2013). 16 K.D. Cole, Exact Anal. Conduct. Toolbox, Exact.unl.edu (2016). 17 J.H. Yu, G. De Temmerman, R.P. Doerner, R.A. Pitts, and M.A. van den Berg, Nucl. Fusion 55, 93027 (2015). 18 Aerotech, Www.aerotech.com (2016). 19 THK, Www.thk.com (2016). 20 Renishaw, Www.renishaw.com (2016). 21 BellowsTech, Www.bellowstech.com (2016). 22 Igus, Www.igus.com (2016). 23 Alpro, Www.alpro.co.uk (2016). 24 T. Fredian, J. Stillerman, and G. Manduchi, Fusion Eng. Des. 85, 568 (2010). 25 J.A. Stillerman, T.W. Fredian, K.A. Klare, and G. Manduchi, Rev. Sci. Instrum. 68, 939 (1997). 26 P.C. Stangeby, The Plasma Boundary of Magnetic Fusion Devices (Taylor & Francis, 2000). 27 D. Brunner, B. Labombard, R.M. Churchill, J.W. Hughes, B. Lipschultz, R. Ochoukov, T.D. Rognlien, C. Theiler, J.R. Walk, M. V. Umansky, and D.G. Whyte, Plasma Phys. Control. Fusion 55, 24 95010 (2013). 28 D. Brunner, W. Burke, A.Q. Kuang, B. Labombard, B. Lipschultz, and S. Wolfe, Rev. Sci. Instrum. 23504, (2016).

25