Dielectric Breakdown — Short Event in Immersion Tank

Kill power at the upstream PDU and breaker immediately. Trip the breaker, drop the PDU output, or pull the C19 plug — whichever is fastest. Do NOT attempt a web-UI shutdown and do NOT use the chassis power switch on a submerged miner. The tank fluid may be conducting right now and any continued energization is propagating the fault into adjacent components. Apply a LOTO tag to the breaker before walking away. This single action prevents 80% of the secondary damage that turns a one-miner fault into a four-miner repair bill.

Visual triage with the lid open and a flashlight, power confirmed off. Wait 10 minutes after power-off for any residual arcing or thermal events to settle. Wear nitrile gloves and ANSI Z87 eye protection. Open the tank lid and inspect under work light: fluid clarity (water-clear good, hazy bad, tea-coloured very bad), surface contamination (white silicone strands, floating particulate, sheen), visible damage on miner exteriors. Photograph everything. Do not touch the fluid yet.

Pull a fluid sample for lab analysis. Use a clean 250 mL glass sample bottle (Boston round amber or equivalent — never plastic, never re-used). Sample from mid-depth at the centre of the tank, not the surface (water floats up) or the bottom (sludge settles). Cap immediately, label with date, tank ID, ambient conditions, brief failure description. Ship to Insight Services, Polaris Labs, or ALS Canada for the panel: ASTM D877 dielectric breakdown, ASTM D6304 Karl Fischer water content, and ISO 4406 particle count. Roughly $80-$160 CAD, 3-7 business days.

Document the failure timeline. Pull every miner log file via SSH or fleet-management software (Foreman, Hive OS, Awesome Miner). Note the timestamp of the first fault, the order in which subsequent miners faulted, and any preceding warnings in the 60 minutes before the event. Pull the PDU event log if you have a smart PDU. This documentation tells you whether the fault was a single-miner short that propagated through the fluid (dielectric breakdown) or an upstream PDU/utility event that took the whole tank down at once.

Do not refill, do not re-energize, do not 'wait and see.' The tank stays drained or stays untouched until the fluid sample comes back from the lab. Resist the operator instinct to 'try a quick power-on test.' A tank that just shorted is in a known-bad chemistry state; powering it back on without fixing the cause is how a one-miner repair becomes a four-miner repair. Schedule the rebuild for after the lab result returns.

Drain the tank to a clean storage drum. Use the original shipping drum (always keep it) or a new HDPE 200 L drum rated for hydrocarbon dielectrics. Pump the fluid through a 5-micron filter on the way out — this catches the gross particulate and tells you immediately whether you have a particulate problem. Cap the drum, label clearly, store dust-free until the lab result comes back. Never decant fluid in an open shop area or near a humidifier.

Pull every miner from the tank. Drain each miner over the tank for 10 full minutes before carrying to the bench. Set up the bench with industrial absorbent pads and a drip tray. Open each chassis and inspect every hashboard under a loupe or stereo microscope for carbon tracking (black branched pattern across solder mask), blown vias, vaporized BGA solder balls, scorched FR4. Photograph every board with visible damage. Boards with carbon tracking are scrap — carbon paths re-fault under voltage even after solvent cleaning.

Inspect the lid seal, cable glands, and every penetration with a magnifier. Hardened, deformed, or visibly cracked seals are guaranteed water-ingress paths. The lid seal in particular is a common failure — operators slam the lid down on cable runs, crush the gasket asymmetrically, and it never seals correctly again. Source replacements before refill. Cable-gland gaskets are $5-$10 CAD each; lid seals $30-$120 CAD depending on chassis.

Bench-test each PSU under programmable load before re-deployment. A PSU that tripped its OCP during a tank short has stressed components even if visually fine. Use a programmable DC electronic load (Rigol DL3021, Itech IT8512, or equivalent) to ramp from 0 to nameplate output current in 10% steps, holding 5 minutes per step, while monitoring output voltage stability and ripple. Any PSU with output ripple above its spec or output sag below nameplate at full load is scrap. Source replacements before tank refill.

Reseal every penetration with fresh, immersion-rated gaskets. Lid seal, every cable gland, top-up port, sensor pass-through. Do not re-use old gaskets, do not 'improve' a gasket with extra silicone (silicone migration is a primary contamination mode). Use the OEM gasket spec or a known-compatible Viton/FKM substitute. Tighten cable glands to spec — under-tight leaks air, over-tight crushes the gasket asymmetrically and creates a worse leak path.

Bench-test the drained fluid with a portable D877 tester before deciding to dehydrate or replace. BAUR DTA-100C, Megger OTS80AF, or budget desktop tester gives you breakdown voltage in 20 minutes. Procedure: pour 400 mL into the tester cup, 2.5 mm electrode gap, 2 kV/s ramp, five trials, average the middle three. Compare to fresh-fill nameplate. <25 kV = scrap; 25-30 kV = marginal, dehydration may recover; ≥30 kV = bulk fluid is fine, fault was localized contamination.

Vacuum-dehydrate marginal fluid before scrapping. Rent a Pall HVP-100, Hydac FAM-25, or Des-Case unit (~$300-$800 CAD per week). Connect to the storage drum in bypass mode, run 24-72 hours per the unit's procedure, then pull a post-cycle sample for retest. A successful cycle pulls water content from 200 ppm back to under 20 ppm and recovers most of the breakdown loss. It does NOT remove silicone, particulate, or oxidation byproducts — those failure modes require fluid replacement. Post-cycle D877 still <30 kV = drain and replace.

Scrap and replace fluid that fails the post-dehydration retest. Properly dispose through an industrial fluid-recycling service (Safety-Kleen, Heritage-Crystal Clean, or your provincial equivalent). Never pour dielectric coolant down a drain; never burn it; never re-use it on another tank as 'good enough' backup. Fresh fluid is $8-$22 CAD per litre in drum quantity for engineered single-phase coolants. The math almost always favours replacement over recovery for any fluid more than 5 years in service.

Inspect the heat exchanger and pump for residue accumulation before refill. Drain and disassemble the plate-frame or coil-tube heat exchanger; flush with the OEM-specified solvent (usually fresh dielectric coolant — never water, never IPA on aluminum-bodied exchangers). Inspect the pump impeller for white silicone strands wrapped around the shaft. Replace any pump with stuck or noisy bearings; pump bearings exposed to contaminated fluid degrade fast and a stuck pump triggers thermal runaway, the fastest path to a second short event.

Refill, requalify each miner on the bench before re-submersion, log breakdown values daily for the first week. Bench-power each miner for 30 minutes at hashing load before re-submerging. Refill the tank with fresh or recovered fluid through a 1-micron filter. Resubmerge in stages — one or two miners first, run 24 hours, pull a sample, verify D877 stability before adding the rest. Daily breakdown samples for the first week catch any reseal failures before they become the next short event. Log everything.

Stop DIY and ship to D-Central when: multiple hashboards across multiple miners show carbon tracking (cascade event); PSU teardown reveals secondary-side rectifier or MOSFET damage; the fluid sample shows both water AND silicone contamination; the tank has experienced two short events within 12 months; any visible fire damage; or you don't own the bench gear (D877 tester, programmable DC load, vacuum dehydration unit) to qualify the recovery yourself.

D-Central's immersion-repair bench process: full miner teardown, per-board carbon-tracking inspection under stereo microscope, PMIC and voltage-domain electrical test under programmable DC load with active rail monitoring, per-chip SHA-256d validation on the test fixture for any board with passed visual, full PSU bench test under nameplate load with ripple capture, gasket and seal replacement, pre-deployment 24-hour burn-in at hashrate. Every miner that leaves the bench has a service record showing exactly what was tested and replaced.

Shipping immersion-exposed gear correctly. Drain every miner thoroughly — 30 minutes minimum hanging over a drip tray, then 30 minutes more on absorbent pads. Cap any open coolant fittings. Bag each miner in a heavy-duty contractor bag (the fluid film keeps weeping for hours) inside the shipping box. Double-box with at least 5 cm of foam on every side. Label outside 'IMMERSION-EXPOSED — DRAINED — FRAGILE' and 'POSSIBLE RESIDUAL DIELECTRIC FLUID — HANDLE WITH GLOVES.' Include a service note: tank fluid age, brand, last D877 reading, observed failure timeline, log excerpts.