Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Computed maintenance durations

New in FARIS 0.2. The download includes a recorded result for the four demo arrangements: open Maintenance in the top bar and choose Open the recorded maintenance result.

Most plant models give every replacement the same fixed outage length. In a real plant the length depends on how long the parts around the replaced component take to cool down, and that grows as the parts that stay in the machine collect long-lived activation. FARIS can compute each replacement’s outage from the decay heat of those parts and compare the result with the fixed durations, design by design.

What FARIS computes

For each design and each replacement over the plant’s life:

  1. Cooldown. FARIS asks ACTINV for the decay heat of every governing component (the parts that must cool before work can start) as they are at that shutdown, after the irradiation history up to it. The governing set’s decay heat per volume is followed for up to 365 days. The cooldown ends when it falls below the class threshold q*.
  2. Outage. The outage is the cooldown plus the class’s work time.
  3. Iteration. Longer outages shift every later event and change the irradiation history, so FARIS reruns the operating history with the new outages and repeats until no outage changes by more than a day (at most 10 passes).

It then reports, under both the fixed and the computed model, every design’s total replacement downtime, availability and lifetime net electricity, plus the downtime difference between each pair of designs.

The comparison makes no ranking claim. In the validation runs, computed durations changed the downtime differences between designs by factors of 2 to 14. Whether they change which design produces more electricity depended on how long outages are assumed to be.

What you need

  • ACTINV and its nuclear data. See actinv.avilalabs.org.
  • Python 3 for FARIS’s activation-input builder, which writes the ACTINV inputs. The package has it as tools/build_activation_inputs.py, and FARIS finds it there.
  • Per design: a scenario, its physics file, a verified transport run for the history, a 709-group spectrum run and operating-history assumptions. These are the same files the history and transport chapters use. The download does not hold the demo arrangements’ 709-group spectrum runs, so recomputing needs your own transport runs.

The designs file

A designs file names each design and its five files. Paths are relative to the file.

{
  "schema_version": "faris-maintenance-designs/v0.1",
  "designs": {
    "port__reference": {
      "scenario": "scenarios/arc-inspired/cold-reference-port.scenario.json",
      "physics": "scenarios/arc-inspired/cold-reference-port.reference.physics.json",
      "history_run": "runs/port-reference/run.json",
      "spectrum_run": "runs/port-reference-709/run.json",
      "history_assumptions": "scenarios/arc-inspired/demountable-magnet-assumptions.json"
    }
  }
}

The maintenance assumptions file

One entry per replacement class: the replaced component, its governing components, the work time and the threshold.

{
  "schema_version": "faris-maintenance-assumptions/v0.1",
  "governing_quantity": "heat",
  "classes": {
    "blanket": {
      "component_id": "blanket",
      "governing": ["first-wall", "blanket"],
      "work_s": 2592000,
      "threshold": {"kind": "calibrate", "design": "port__reference", "target_cooldown_s": 2592000}
    },
    "magnet": {
      "component_id": "magnets",
      "governing": ["first-wall", "blanket", "shield", "vessel"],
      "work_s": 5184000,
      "threshold": {"kind": "q_star", "q_star_w_per_m3": 2179}
    }
  },
  "cooling": {"min_s": 3600, "max_s": 31536000, "points": 40},
  "max_iterations": 10,
  "convergence_s": 86400
}

A threshold is set in one of two ways:

  • calibrate: q* is chosen so that this class’s first replacement in the named design, at the fixed durations, cools down in exactly target_cooldown_s. Durations are then relative to that one outage: FARIS shows how they differ between designs and over the plant’s life, not how long they are in absolute terms.
  • q_star: a threshold in W/m³ that you supply, for example from a plant’s remote-handling limits.

The fixed duration of each class is the replacement_duration_s in that design’s operating-history assumptions. cooling, max_iterations and convergence_s may be left out; the values above are the defaults.

Run it from the desktop

Open Maintenance in the top bar, then Run…:

  1. Choose the designs file and the maintenance assumptions file.
  2. Choose the actinv executable and the ACTINV data directory (the folder above the catalogue version). FARIS remembers them until you close it. You can also pass them at launch with --actinv and --actinv-data.
  3. Optionally choose the result file. By default it is maintenance-result.json next to the designs file; an existing name gets a number, so nothing is overwritten.
  4. Press Run.

Anything missing is listed under the form with what to do next. While it runs, the window shows the design, the pass, ACTINV runs and cached curves, and the elapsed time. Cancel stops ACTINV and writes no result. You can close the window and reopen it to check progress. When the run finishes, the result opens in the window.

ACTINV results are cached in maintenance-work next to the designs file, keyed by the exact ACTINV input, so a rerun with unchanged inputs takes seconds.

Run it from the command line

faris maintenance run --designs designs.json --assumptions maintenance.json \
  --actinv /path/to/actinv --data-dir /path/to/actinv-data --output result.json [--workers 2]
faris maintenance report result.json

run refuses an existing output file. Its work folder defaults to result.json.work; --cache points several runs at one cache. --impurities uses the specification-maximum impurity variant instead of the bare materials. report prints markdown tables, or the result itself with --format json. See Command line for the other commands.

Read the result

Open a result with Maintenance → Open result…, or start FARIS with --maintenance result.json. In the downloaded package, Open the recorded maintenance result opens the result recorded for the four demo arrangements after checking its SHA-256. It was computed with ACTINV at the coupling test’s settings: each class’s work time and calibration target are half its fixed duration, with q* calibrated on the reference-port arrangement, and the “Demountable magnets” operating assumptions (in the package as maintenance/operating-assumptions.json).

  • Thresholds: q* per class and where it came from.
  • Designs: downtime, availability and lifetime net electricity, fixed beside computed, and whether the computed model could be evaluated.
  • Replacement outages: per replacement, a grey bar for the fixed duration beside the computed one, split into work (blue) and cooldown (orange), against plant time.
  • Replacements: every replacement with its start, both durations, and why: each governing component’s share of the decay heat when the cooldown ends, and how long it had been in the machine. A cooldown that grows from one replacement to the next is usually explained here by a part that is never replaced. In the validation designs the first wall gave essentially all of it.
  • Between designs: the downtime difference under each model and their ratio. The ratio is left out when the fixed difference is under 30 days, where it would not mean much.

A result marked not evaluated says why and what to try next. Common reasons:

  • The decay heat never falls below q* within 365 days. Raise q* or reconsider which components govern.
  • The outages still change after the last pass. Raise max_iterations, or look at the last two passes in All iterations for durations that alternate.
  • A cooldown is cut short by the next restart (window-limited), so it is only a lower bound.

Literature durations and remote-handling limits

Computed durations are one choice. FARIS also carries published ones, labelled literature wherever they appear, with the source link and the conditions they were derived under. They are inputs taken from a paper, not FARIS results.

PresetWhole outageConditions
EU-DEMO blankets and divertor, 4 remote-handling systems7.9 months5.9 months of handling (Table 1, p. 5; the paper’s “about 6 months”) + 1 month cooling + 1 month conditioning and pump-down (Sec. 5.2)
EU-DEMO blankets and divertor, 8 systems5.2 months3.2 months of handling, the shortest in Table 1, + the same 2 months
EU-DEMO blankets and divertor, 1 system24 months22 months of handling, one system working all 16 sectors in turn (about 1,000 hours per sector, p. 4), + the same 2 months

All three come from Crofts and Harman, Fusion Eng. Des. 89 (2014), arXiv:1412.4008, read in full. Each replaces all blanket segments and divertor cassettes of the 2012 EU-DEMO (5 segments and 3 cassettes in each of 16 sectors) through all 16 upper and 16 divertor ports; replacing the blankets alone takes as long, because the divertor must come out first (p. 4). The handling months already include the paper’s equipment-failure recovery (about 20 % of the total, p. 5), transporter delays and shift and operator productivity. The paper states no percentage contingency, so the presets add none and say so.

Remote-handling constraints. The outage length follows the number of parallel remote-handling systems, ports and crews. Handling time is a fixed part plus the shared work divided by the systems that can work at once, which is the smallest of the systems, the ports and the crews; cooling, conditioning and contingency are added. The model assumes the systems share the work evenly and never fail or wait for each other. Fitted to the five points in Table 1 of the paper (22, 11, 5.9, 4.4 and 3.2 months for 1, 2, 4, 6 and 8 systems), it stays within 10 % of each. The fit is a calibration to those points, not a derivation from the paper’s handling sequence.

Sensitivity to each duration. For a chosen range of each replacement duration, FARIS reruns the history at the low and high value of one duration at a time and reports lifetime net electricity and availability for each, largest effect first (a tornado). Availability here is one minus the replacement downtime over the history. Net electricity is absent, with the reason, when the history lacks its energy inputs.

Limits

  • Decay heat governs; contact dose does not. With dose as the governing quantity and a threshold fixed at the first outage, the never-replaced first wall kept the dose above it for more than a year at later replacements in the validation designs. FARIS therefore refuses "governing_quantity": "dose" for now. Testing a remote-handling dose rule needs its own validation.
  • Screening, not a maintenance plan. Work times, governing sets and thresholds are your assumptions. FARIS does not model crews, spare parts, queues or remote-handling equipment.
  • Run time. The calculation runs ACTINV once per replacement and governing component, and again on each pass. A four-design, 30-year case took about ten minutes from an empty cache and about a minute from a full one on a laptop.
  • Checked against the validation test. On the maintenance coupling test’s four designs, FARIS reproduces the reference calculation’s outages, downtimes and lifetime electricity to 1 part in 10¹⁴. The validation itself is described in the repository’s docs/notes/MAINTENANCE_COUPLING_VALIDATION_RESULT.md.