Skip to content

September 14 run and Kikusui telemetry

The preanode short is solved by the owner's physical repair to the plasma source (confirmed in this session). The remaining findings concern control software and measurement quality, not an unresolved preanode short.

Experiment record

Read the September 14 PIHTI journal, its ControlUnit screenshots, the completed cu_20260914_173253.csv, the event/launcher logs and the source installed on the Pi. The rig reported 4.14.1; its checkout was 32959f27745c2b0548af49781ad7706580d412b8. All event times below are JST.

  • 49,295 complete 26-field rows, no missing parsed values, strictly increasing timestamps, 17:32:53.711849–19:25:58.689335. Stop was deliberate; a new slow acquisition began at 19:26:02.
  • Sampling changed to 1 s at 17:33:43 and back to 0.1 s at 18:07:32. Do not interpret the whole-file average interval as a missed 10 Hz target.
  • During the plasma-test window 18:44:46–18:53:16: 5,098 rows, mean interval 0.100043 s, median 0.099863 s, maximum 0.182482 s; no interval above 0.2 s.
  • Upstream Baratron pressure at 18:45–18:53: median 4.854e-3 Torr, 5th–95th percentiles 4.812e-3–4.988e-3 Torr.
  • Two ADC errors before the test, at 18:31:35 and 18:36:05: OSError(121, Remote I/O error), each automatically recovered after one failed read. Adjacent CSV gaps were 0.606 and 0.623 s (the log rounds to one second). Switching-transient causation is possible, not established. No later reader failure was found in this completed run.
  • Other fast-segment timing outliers did occur outside the discharge window, including 1.216 s at 19:12:01. The good plasma cadence is not a claim that the entire evening had no timing delays.

The CSV SHA-256 is 05b976b8f74046c32d28d170ff879c49b9b030a1287302d6bd30ae1b4b08c7a1. Source files were copied for analysis and not modified.

PID handover

The 18:52 journal entry describes the 18:50:03 command to engage a 0.50 A PID target. Manual drive before it was 1900 mV. The first CSV row showing the PID command, at 18:50:04.113832, records 5500 mV, sustained until manual control resumes at 18:50:11.897668. Ip returns near its off-plasma baseline during the episode.

Installed source shows that manual commands recreate the PID, but enabling it does not reset its clock or initialize the output from the held drive. Its first update counts roughly 169 s of manual operation as elapsed time. With P=30 and I=40, the integral saturates at 4500 mV; the controller then adds a 1000 mV offset outside that limit. An offline replay with 169 s and input -0.25 A reproduces 5500 mV. The installed DAC driver clamps this to code 4095, nominally 4.9988 V; it does not output 5.5 V or wrap to zero.

This establishes a software handover fault, not the complete physical cause of extinction. There was no simultaneous PSU voltage/current/protection record, so a full-scale command must not be presented as a measured filament current pulse. The owner clarified that the student intended ignition plus maintenance; the replacement should engage only on an established discharge.

Two further record caveats:

  • Off-plasma Ip median at 18:41–18:43 was -0.325891 A, with no zero action logged in this run. Absolute plasma-current values and PID targets need a valid zero reference.
  • PresetV_cathode stays zero during manual drive: that path bypasses the ADC worker's command record. The event log supplies the manual commands. Existing analog Ci/Cv channels are unwired placeholders.

Read-only LAN proof and next work

The connected PWR401L, firmware VER01.24 BLD0056, answered *IDN?, MEAS:VOLT?, MEAS:CURR? and OUTP? on TCP 5025. Output was off on both checks, with -0.0045 V and approximately 0.0075–0.0083 A readback. Ten voltage/current pairs from the office PC took 4.58 ms median and 19.53 ms maximum. This establishes connectivity, not reliability during plasma.

Owner decision: log several normal manual discharges and bakes first, then use those records to understand filament current, voltage, warm-up behaviour and LAN reliability before fixing PID. The initial 4.15.0 recorder writes a separate timestamped CSV, records missing polls and connection recovery, and never sends output, setpoint, reset or remote-mode commands. Its setup and data contract are in Kikusui over LAN.

Later filament-condition warnings should compare resistance V/I and power VI at similar operating and thermal conditions against a healthy baseline. Temperature, contacts, leads and geometry also affect resistance; a warning cannot uniquely diagnose thinning, and today's file cannot establish a resistance trend without the absent PSU measurements.

The owner also requested PIHTI Log's existing attachment flow be extended to all available ControlUnit parameters at the instant of capture, as a table and optionally a picture, with a possible labelled averaging window. Fleet post 20260914-86630ead-525f3f carries that request to code/pihti-log. Kikusui live readouts and a snapshot API are follow-up integration work.