Safety Design of Accelerator High-Voltage Power Supplies in Particle Physics and Medical Treatment Equipment
An accelerator power supply at a few hundred kilovolts carries enough stored energy to kill, and its safety design cannot be borrowed from industrial supplies that happen to reach the same voltage. The differences are structural: the load is a long vacuum column with distributed capacitance, the return path runs through a concrete vault, and the people who service the equipment include physicists, technicians and clinical staff whose relationship to the hazard varies. Designing for the worst of those cases changes the topology, not just the labels.
Start with the energy budget. A two-hundred-kilovolt supply with two microfarads of output capacitance stores forty joules; a spark at the terminal delivers that energy in microseconds, and the acoustic and pressure effects inside a vacuum vessel can damage the ceramic stack. Reducing stored energy by lowering capacitance conflicts with ripple and transient performance, so the design compromise is usually a moderate capacitance plus a crowbar that diverts energy away from the load within a few microseconds of a fault.
The crowbar is the single most important protective element and also the one most often omitted for cost. A triggered spark gap or a thyristor stack that shorts the output to a dump resistor converts an uncontrolled arc into a controlled discharge. The gap must be rated for the full voltage and for repeated operation without erosion, and its trigger circuit must be immune to the electromagnetic noise that accompanies the fault it is meant to handle.
Interlocks have to be layered and independent. The first layer is physical: keys, hasps and door switches wired in series into a relay that removes primary power. The second layer is electronic: a watchdog in the controller that removes the enable signal if the state machine stops progressing. The third layer is procedural: a written sequence for lockout, discharge, verification and re-entry. Any single layer can fail; the design assumes each will.
Discharge verification is not a formality. A bleeder resistor sized for a ten-minute discharge is adequate for the vault but useless for the technician who wants to work in ten minutes. A dedicated discharge stick with a resistive tip, a meter that reads the actual voltage, and a mandatory second reading after a delay, together form a sequence that catches the case where a bleeder has failed open.
Grounding must be designed as a system. The high-voltage return, the vacuum vessel, the vault rebar and the building earth are separate nodes connected by impedance, and the fault current path between them determines the transient voltage that appears on instrumentation. A single-point reference for the measurement electronics, bonded to the vault floor, with fibre links to the control room, prevents instrument damage and prevents operators from being exposed to a step potential.
Medical installations add requirements that physics installations do not have. The equipment sits in a treatment room with a patient, and the interlocks must be tied to the door, the beam-off logic and the dosimetry chain. Any fault that could leave the beam on while the door is open is a design error, and the interlock chain has to be verified by actual test, not by simulation, before the first patient is treated.
Leakage current limits are tighter in medical use. Patient-connected circuits must not exceed the limits of the applicable standard under normal and single-fault conditions, and the high-voltage supply is one of the paths that can violate them. Isolation transformers with adequate creepage, double insulation on the measurement path, and a leakage monitor on the output return are the practical measures.
Availability and safety are not opposites in this context. A supply that trips frequently because of over-sensitive protection is a supply that operators learn to reset without investigating, which is the behaviour safety design tries to prevent. Protection thresholds set with margin, clear alarm codes, and a fault log that distinguishes transient from persistent events encourage investigation instead of repetition.
Radio frequency interference from the accelerator itself is a design input. Cavities, klystrons and pulsed magnets produce fields that couple into the control electronics, and a supply whose protection logic responds to those fields will shut down at the wrong moment. Shielded enclosures, fibre control links, and filtering on every external cable are baseline requirements, not refinements.
Component selection follows the fault mode rather than the datasheet margin. A resistor that fails open in a bleeder chain removes one third of the discharge capacity, so bleeders are specified with two parallel strings and monitoring of each. A capacitor that fails shorted in a filter bank dumps energy into the transformer, so filter capacitors are fused individually or selected with self-healing dielectrics.
Documentation of the safety design matters as much as the design. A single drawing that shows the energy sources, the stored energy at each node, the discharge paths and the interlocks, kept current through every modification, is what a service engineer needs at two in the morning. Drawings that lag the equipment are a hazard in themselves.
Verification should be periodic and recorded. Annual tests of interlock function, crowbar firing, discharge time and leakage current, with results kept for the life of the installation. A crowbar that fired correctly five years ago proves nothing about the capacitor that has aged since.
Training completes the design. The people who operate the supply need to know where the energy is stored, how long it persists, and what the interlocks do not protect. A supply with excellent hardware and no training produces exactly the incidents the hardware was meant to prevent.
Cost of the safety features is concentrated in the crowbar, the interlock chain and the documentation. In a research installation those items are a small fraction of the total project; in a small medical facility they can dominate the budget of the supply. Reducing them is a decision that should be made explicitly and recorded, not a silent consequence of procurement pressure.
Limits should be stated plainly. No interlock prevents a failure inside the high-voltage tank from exposing a person who has already defeated the door switch, and no crowbar protects a load whose own insulation fails at a voltage far below the supply rating. The design protects against the faults that can be anticipated; the procedure protects against the rest.
A useful commissioning sequence exists and is worth writing down. Verify interlocks with the output disabled; verify the discharge path with a calibrated meter; fire the crowbar into a dummy load and measure the dump time; measure leakage with the load connected; then run at half voltage for an hour and repeat the crowbar test. Five steps, one day, and a supply whose safety behaviour is known rather than assumed.
Insulation coordination inside the tank deserves independent review. Creepage distances that satisfy a clean, dry environment fail when oil or condensate accumulates, and the tank is the one place where inspection is hardest. Specifying distances for the polluted condition, and adding a guard electrode that intercepts leakage before it reaches a measurement node, removes a failure mode that otherwise appears only after years of operation.
The control interface is a safety element, not a convenience. A remote enable that can be issued from a keyboard outside the vault must be paired with a local, hardware-only disconnect that no software can override. When the two disagree, the hardware wins, and the wiring must make that true rather than the firmware.
Fault recording closes the loop on safety design. A supply that logs the sequence of events leading to a trip, with timestamps to the microsecond, allows the cause to be established without reproducing the fault. Without that record, every trip becomes a discussion, and the discussion usually ends with a raised threshold.
Replacement parts must preserve the safety properties. A crowbar gap substituted with a different model, or a bleeder replaced with a lower power rating, silently degrades the protection. Part numbers for safety-critical items belong on a controlled list, and substitutions require the same review as a design change.
Finally, the design should be tested against a fault that the designer did not anticipate. A deliberate short at the output, applied while the supply is at full voltage and the interlocks are enabled, is the only test that exercises the whole chain at once. It is uncomfortable, it is loud, and it is the test that finds the gap between the drawing and the hardware.
Redundancy should be considered at the level of the protection chain rather than the power stage. Duplicating the crowbar is expensive; duplicating the trigger circuit, the interlock relay and the discharge monitor is affordable and covers the failures that actually occur, which are usually in the small components that decide whether the large component acts.
Ageing of safety components is predictable and should be scheduled. Trigger capacitors lose capacitance, spark gap electrodes erode, and bleeder resistors drift upward. A five-year replacement schedule for these items, independent of measured failure, keeps the protection chain at design strength instead of letting it decay until a test reveals it.
Spares and skills belong in the safety case. A vault with one trained person and no spare crowbar gap has a safety margin that depends on a single individual's availability. Cross-training two people and holding one spare of each safety-critical item converts a fragile arrangement into a resilient one at negligible cost.

