Design and Operation of a 3-Crystal Multi-Volume Reflection Array for Loss Reduction in CERN SPS Slow Extraction

summary

The gist

The design and operation of a 3-Crystal Multi-Volume Reflection Array for Loss Reduction in CERN SPS Slow Extraction presents a solution to mitigate induced radioactivity from beam loss on…

In short

Researchers designed and tested a 3-Crystal Multi-Volume Reflection Array (MVRA) to reduce radioactivity caused by beam loss during slow proton extraction at CERN SPS. The array uses stacked bent strips instead of a single crystal, achieving a significant loss reduction factor of four relative to the baseline crystal configuration. Experimental control maintained this reduction through an on-line feedback system.

Key concepts

Multi-Volume Reflection Array (MVRA)
This design replaces one large crystal with multiple bent strips stacked together. The deflections of these strips are arranged coherently, allowing the array to reflect and redirect beam loss more effectively than a single crystal, aiming for a fourfold reduction in intensity loss.
Multi-Fidelity Bayesian Optimization (MFBO)
This AI method combines a fast, low-fidelity surrogate model (a multilayer perceptron trained on past data) with a high-fidelity model (full particle tracking simulation). It uses this combination to intelligently search for the best array geometry by selecting test points that maximize information gain while minimizing computational cost.
Extremum-Seeking Feedback Controller
This is an intelligent control system used during operation. It constantly monitors beam loss and adjusts the array's operating point in real-time. It uses a loss-dependent amplitude schedule to keep the array at its best performance level, even when machine conditions drift.
Loss Reduction Factor
This metric quantifies how much the MVRA reduces radioactivity compared to a standard crystal setup. The goal was a factor of four reduction for high-intensity targets. The system achieved a median reduction of 3.60 times over many extraction cycles, demonstrating its effectiveness in mitigating induced radiation.

Terminology used across episodes

This episode discusses

The paper

Design and Operation of a 3-Crystal Multi-Volume Reflection Array for Loss Reduction in CERN SPS Slow Extraction · Read on arXiv

CERN

The flux of protons slow-extracted from the CERN Super Proton Synchrotron (SPS) to the North Area experiments is limited by the induced radioactivity of the beam lost on the electrostatic septum during third-integer resonant extraction. Crystal shadowing, in which a thin bent silicon crystal deflects the protons that would otherwise hit the septum wires, has been in operation at the SPS since 2021; a single crystal at a non-local position halved the extraction losses relative to the local crystal alone, short of the fourfold reduction that future high-intensity operation requires. In this paper, the design, deployment and operation of a 3-crystal Multi-Volume Reflection Array (MVRA) at the non-local position in the fourth long straight section (LSS4) are presented. The array geometry was designed by Multi-Fidelity Bayesian Optimization (MFBO), combining a multilayer-perceptron surrogate with full particle tracking; it predicts a fourfold loss reduction with three crystals and up to tenfold with an ideally aligned array of four to five. At its optimum the as-built array reaches, and at its best point exceeds, the predicted factor of four, and beam dynamics simulations matched to the measured position and angle scans reproduce their shape to within 0.04 of the crystals-retracted loss. In physics production an extremum-seeking controller has held the array on its operating plateau, with a median loss reduction by a factor of 3.6 over 1.14 x 10 5 extraction cycles relative to the pre-shadowing baseline. The remaining margin to the target, the limitations of the present installation and the upgrade to a 4- or 5-crystal array are discussed.

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: We're looking at the paper: "Design and Operation of a 3-Crystal Multi-Volume Reflection Array for Loss Reduction in CERN SPS Slow Extraction".

Mira: Here's the gist of the paper. The design and operation of a 3-Crystal Multi-Volume Reflection Array for Loss Reduction in CERN SPS Slow Extraction presents a solution to mitigate induced radioactivity from beam loss on electrostatic septum wires during resonant extraction, which is crucial for meeting future high-intensity proton delivery targets The slow extraction of high-energy proton beams from circular accelerators poses a challenge because the third-integer-resonance extraction process deposits intensity onto the wires of the electrostatic septum (ZS), dominating the local radiological budget and limiting sustainable intensity Future high-intensity fixed-target programmes call for up to 4 × 10 19 protons on target per year, requiring a factor of four reduction in extraction loss per proton compared to the present North Area delivery Crystal shadowing, particularly its non-local variant in LSS4, has been demonstrated to halve the extraction loss relative to the machine with only the local crystal in MVRA Architecture and Design The Multi-Volume Reflection Array (MVRA) replaces a single crystal with a sequence of bent strips whose deflections stack coherently, aiming for a reduction scaling toward ∼ 10× for four to five strips The array geometry was designed using Multi-Fidelity Bayesian Optimization (MFBO), combining a multilayer-perceptron surrogate with full particle tracking to predict a fourfold loss reduction with three crystals The design parameters searched jointly include crystal width, inter-crystal stagger, number of crystals, and the sign of the volumereflection deflection Optimization Methodology The optimization employs a multi-fidelity Gaussian-process (GP) model with a multi-fidelity knowledge-gradient acquisition function weighted by an affine fidelity-cost model to select evaluation points jointly with the fidelity level that maximizes information gain per unit cost The low-fidelity model is a multilayer-perceptron surrogate pretrained on prior crystal-shadowing data, while the high-fidelity model is the full tracking simulation The optimization results showed that a single crystal in volume reflection attains about 0.8 relative loss, and the 3-crystal array reaches about 0.25, a factor of four Experimental Validation and Control The experimental setup involves diagnostics using six beam loss monitors (BLMs) in the ZS region to define an intensity-normalised total loss observable An on-line extremum-seeking feedback controller was deployed to hold the array at its operating point, sustaining a median loss reduction by a factor of 3.6 over 0.1 million extraction cycles relative to the pre-shadowing baseline The controller operates by dithering the setpoint based on the local loss gradient, ensuring it remains within the working window against machine drifts Performance and Limitations The 3-crystal MVRA reached a median 3.60× reduction on the pre-shadowing baseline through routine production, which is a little short of the target set by the activation budget The performance against the factor-of-four target is limited by the cost of holding a drifting optimum with a quantised goniometer and requires an upgrade to 4 or 5 crystals The final matched configuration achieved a loss reduction of 0.184 against the crystals-retracted reference, which is better than the design prediction of 0.25 The paper demonstrates that the MVRA reaches its predicted factor-of-four loss reduction at its measured optimum and sustains a median 3.60× reduction through routine production The comparison with other shadowing configurations shows the MVRA is superior, achieving a collateral-aware figure of merit M = 0.74 against the ideal fourfold reduction of 0.

Title and authors: Mira: Now that we understand what the paper is about—the three-Crystal Multi-Volume Reflection Array for Loss Reduction in CERN SPS Slow Extraction—let’s unpack what exactly they proposed to improve it <ref:2609.23823#pg1,3-Crystal Multi-Volume Reflection Array for Loss Reduction in CERN SPS>.

Kai: They suggest several improvements, starting with using AI to perform cost-aware multi-fidelity optimization. This means combining a fast multilayer perceptron surrogate model with full particle tracking through a multi-fidelity Gaussian process model to select evaluation points based on information gain per unit cost.

Lev: That sounds like they are making the search for the best design much smarter and faster than just brute force searching all those parameters at once.

Mira: Exactly, and this optimization can jointly optimize four design parameters: crystal width, inter-crystal stagger, number of crystals, and the sign of the volume reflection deflection.

Kai: Plus two operating setpoints: the array shadowing angle and its position relative to the septum wires in that ZS region. So it’s a lot of variables they are tuning at once.

Lev: It sounds like they are building a system where we can design the crystal array geometries more intelligently based on what we know about how much computation it costs versus how much better the result is.

Mira: And then there's this extremum-seeking feedback controller that is loss dependent, using a loss-dependent amplitude schedule to hold the array at its local loss optimum by tracking the optimum without needing a model of the surface.

Kai: That’s clever because it means it doesn't need a complex model of the surface geometry just to keep it running; it reacts directly to what’s happening in real-time based on local loss.

Lev: If that controller can track the optimum without needing a surface model, that significantly reduces the overhead for running this on real hardware, which is where I get nervous sometimes.

Mira: They also talk about executing this all through on-line control by dithering and reacquiring the setpoint when drift moves it away, using a detection threshold of about zero point one above the target loss.

Kai: And that correction is driven by an inner proportional integral derivative loop driving the goniometer to make sure it stays right where it needs to be.

Lev: So they’re essentially building a self-correcting system that handles machine drift and keeps the physical setup precisely where it needs to be for those precise measurements.

Mira: Finally, they propose using AI for simulation and machine matching between their measurements and simulations by scoring candidates based on a dip-region metric, which is the mean absolute deviation between simulated and measured scans from the measured full width at half maximum down to the measured minimum.

Kai: It sounds like they are using sophisticated metrics to make sure their simulated physics actually matches what’s happening in the actual experiment.

The paper's summary: Lev: So we’ve seen the design and then the improvements; now let’s go back to what the paper is actually summarizing about this three-Crystal Multi-Volume Reflection Array for Loss Reduction in CERN SPS Slow Extraction <ref:2609.23823#pg1,3-Crystal Multi-Volume Reflection Array for Loss Reduction in CERN SPS>.

Mira: The core finding is that they successfully moved from a single crystal shadowing configuration to an array of bent strips where the deflections stack coherently, and that stacking provides a substantial reduction in loss.

Kai: They found this stacking leads to a factor of two reduction in the extracted loss per proton observable, and both configurations are reviewed together as baselines.

Mira: The non-local single crystal was their baseline, but they showed that the array of Section II B widens the working window by an order of magnitude compared to that single non-local crystal.

Lev: So they’re not just getting a better number; they are significantly improving the operational window for keeping the system stable and running reliably.

Kai: And this is directly related to their controller in Section VI, which holds the array inside its working window through routine production.

Mira: The summary points out that while stacking crystals gives them a better reduction, it also opens up a new set of design challenges because they have to optimize all those parameters simultaneously.

Lev: So the paper is summarizing that the multi-crystal approach is superior in terms of performance improvement, but it introduces complexity in the geometry itself.

Kai: They also summarize that their experimental validation showed that a single crystal in volume reflection attains about a zero point eight relative loss, while the three-crystal array reaches about a zero point two five.

Mira: So to put it plainly, they are summarizing that the three-crystal array is better for loss reduction than just using one crystal in this specific setup.

Lev: And they summarize that their experimental validation showed a median three point six factor reduction on the pre-shadowing baseline through routine production.

Kai: So, to recap the summary, they demonstrated that their design works as predicted and is sustainable in operation, even if it’s not quite hitting the ultimate fourfold target yet.

Mira: The implication here is that this geometry works well enough to be a viable path forward for achieving future high-intensity goals at CERN.

The paper's improvements: Lev: Let’s focus on the suggested improvements again, specifically how those proposed changes translate into real-world advantages for someone trying to run this kind of hardware.

Kai: The first improvement is the AI system performing cost-aware multi-fidelity optimization we just talked about, which lets them design the array geometry based on information gain per unit cost.

Mira: That means they aren't just picking a configuration that looks good on paper; they are optimizing for how much information you get back relative to how much time or simulation power it takes.

Lev: If the AI can optimize based on cost, that makes the design process much more practical for an experimentalist because it guides them toward a solution that balances performance and resource usage.

Kai: They also suggested using an extremum-seeking feedback controller that is loss dependent, which uses a loss-dependent amplitude schedule to hold the array at its local loss optimum by tracking the optimum without needing a model of the surface.

Mira: This means they can maintain stability even if machine conditions drift, as long as they know how to track that local optimum through real-time feedback.

Lev: That removes a major dependency on having an accurate model of the surface geometry, which is something I’ve seen in other control systems fail when things get messy.

Kai: And then there’s the on-line control loop, where they dither and reacquire the setpoint when drift moves it away using that detection threshold of about zero point one above the target loss.

Mira: That inner proportional integral derivative loop driving the goniometer is what really makes it robust against those small drifts that you can't always predict beforehand.

Lev: It sounds like they’ve built a system designed to handle real-world operational noise and drift with a relatively low amount of hardware complexity in the control loop itself.

Kai: And finally, the simulation and machine matching part where they score candidates based on that dip-region metric between simulations and measurements.

Mira: This is how they ensure that their theoretical predictions actually line up with what’s happening on the table, using a concrete metric to score candidates for being good matches.

Lev: So, the improvements focus heavily on making it robust against experimental noise and ensuring the design process is intelligent about its own resource usage.

Conclusion: Mira: So we’ve covered the design details of this paper, the summary of what they achieved with that three-Crystal Multi-Volume Reflection Array for Loss Reduction in CERN SPS Slow Extraction, and the specific improvements they proposed <ref:2609.23823#pg1,3-Crystal Multi-Volume Reflection Array for Loss Reduction in CERN SPS>.

Kai: The big picture is that this MVRA reaches its predicted factor of four loss reduction at its measured optimum during routine production, and it sustains a median three point six factor reduction through routine production.

Lev: I think what this means for the real world is that we have a system that can reliably manage the induced radioactivity issue in high-intensity proton delivery experiments.

Mira: It’s about making sustainable intensity targets possible by solving this specific problem at CERN SPS. This work shows that the MVRA is superior to other shadowing configurations because it achieves a collateral-aware figure of merit of zero point seven four against the ideal fourfold reduction of zero point one.

Kai: So, in conclusion, the design and operation of this three-Crystal Multi-Volume Reflection Array for Loss Reduction in CERN SPS Slow Extraction shows a path forward for tackling this challenge <ref:2609.23823#pg1,3-Crystal Multi-Volume Reflection Array for Loss Reduction in CERN SPS>.

Lev: For me, it’s encouraging to see a system that can be held at its local loss optimum by an AI controller without needing a full surface model during operation.

Mira: It’s about making sure that when we push for higher intensity, we have the tools to manage the resulting radiation budget effectively.

Kai: That's all for this paper on the three-Crystal Multi-Volume Reflection Array for Loss Reduction in CERN SPS Slow Extraction <ref:2609.23823#pg1,3-Crystal Multi-Volume Reflection Array for Loss Reduction in CERN SPS>.

More episodes

← Home