Whatsminer M30S – Water Cooling Leak Detection

Kill the PDU breaker first — not the miner UI, not the pump controller, the mains breaker. Hydro leaks on a live miner are a race: the longer 14 V DC energizes through a glycol path, the more copper traces are being electrolytically eaten. Wait 60 seconds for LEDs to fully die before touching the chassis. This single step is the difference between a $60 rinse and a $900 hashboard replacement.

Isolate the loop at the miner. Close the inlet and outlet ball valves or disconnect the quick-disconnects (QDs) as close to the miner as your rack allows. If you don't have QDs or valves, put a bucket under the miner and accept you're going to spill the loop volume between the miner and the nearest upstream fitting. Every litre of `MicroBT Hydro Coolant` / `propylene glycol` you keep in the reservoir is one you don't have to source fresh.

Stop the pump. If the pump controller is on the same rack PDU as the miner, one breaker kill covers both. If it's on a separate circuit, kill it manually at the pump controller. Residual pump momentum keeps coolant moving for 30-60 seconds even after power-off — wait for complete stop before opening any fitting. Never open a QD or unscrew a fitting with the pump running.

Contain and measure the spill. Towels, shop vac (wet/dry), drip tray. Measure the drained volume if possible — M30S++ Hydro internal volume is roughly `150-200 mL`, so anything significantly more points at loop volume, not just miner volume. Propylene glycol is low-toxicity but wear nitrile gloves and splash goggles — the dye/inhibitor package isn't as kind as the base glycol.

Walk the loop with the lights low and a UV torch (365 nm). Most data-center hydro coolants include a UV tracer dye. Scan reservoir → pump → pre-filter → supply manifold → miner inlet → miner outlet → return manifold. Fluorescing residue anywhere other than the expected leak point means you've got a second, slower leak hiding. Note every spot, photograph for the repair log.

Pressurize and decay-test the isolated loop section with the miner removed. Cap the miner ports, hand-pump the remaining loop to `1.0 bar`, close the valve, and watch for 10 minutes. Pressure decay `>0.1 bar` in 10 minutes = active leak somewhere else in the loop. Pressure decay `<0.05 bar` = loop is healthy, miner was the only source.

Open the miner chassis on a dry bench, AC confirmed off. Inspect for: glycol pooled on the PCB (sticky, fluoresces under UV), salt crystals near QDs, discoloration or green oxidation on copper busbars, milky residue on the hashboard cold-plate mating surface, corrosion on the control-board connectors. Smell for the sweet glycol odour — fastest way to locate a slow leak that dried overnight.

Clean and dry the leak-detect sensor pad. Wipe the conductivity pad (underside of chassis on M30S++ Hydro — two-wire pad bonded to chassis floor) with `99% IPA` on a lint-free wipe. Let it fully dry. Clear the latched alarm via `WhatsminerTool → Remote Ctrl → Clear Fault`. If the alarm re-triggers within 60 seconds on a verified-dry sensor, the sensor or its wiring is the next suspect.

Measure sensor continuity with a DMM on the `MΩ` range. Dry sensor pad should read open (`>10 MΩ`) between the two pads. Anything under `1 MΩ` dry = contaminated or damaged sensor. Wet reading should be near-short. If the dry reading is stuck low even after IPA clean, the sensor assembly needs replacement — it's a cheap part (typically `$15-$40 CAD`) and a 10-minute swap.

Re-seat every quick-disconnect on the miner. QD seal wear is the #1 hydro leak source in D-Central's repair queue (outranks cold-plate failure roughly 3:1). Disconnect, wipe both halves clean with IPA, inspect the internal seat for scoring or deformation, apply a thin film of silicone grease rated for glycol service (never petroleum grease — it swells rubber), and re-mate with a clean click. No wiggle, no weep.

Replace the cold-plate O-ring if the leak is traced to the plate-to-board seam. Correct durometer matters — most M30S++ Hydro cold plates use a 70 Shore A EPDM o-ring; a nitrile (NBR) substitute will fail early under glycol attack. Source from the hydro loop vendor, not a hardware-store assortment tray. Hand-tight plus 1/8 turn on the retaining hardware — over-torque crushes the o-ring, under-torque weeps.

Rinse a glycol-contaminated PCB. If you saw wet traces in Step 7, this is not optional — residual glycol + ambient humidity = continuing electrolysis. Rinse with deionized or distilled water first (flood, don't spray), follow with `99% IPA` rinse, drain thoroughly, then oven-dry at `55 °C` for `24 hours` minimum. Do not skip the oven — trapped moisture in BGA pockets causes delayed failures weeks later. Tap water is banned; it leaves mineral residue that conducts.

Inspect copper busbars and hashboard traces for electrolytic damage post-rinse. Green discoloration = copper acetate = trace has been eaten. Salt crystals = evaporite from the glycol/inhibitor mix, usually indicates a days-to-weeks contamination. Either finding means Tier 4 — bench-level trace repair, not field-repairable. Stop, ship.

Torque the copper busbar screws to `2.0-2.5 Nm` on reassembly. MicroBT publishes no busbar torque spec; D-Central's consensus number from repair queue data is `2.5 Nm`. Loose busbars arc under load, run hot, and in a post-leak context they'll pull current through any remaining moisture film. A torque screwdriver is $80 and pays for itself on the first miner it saves.

Verify coolant chemistry before re-charging the loop. Confirm you have `MicroBT Hydro Coolant` or a vendor-qualified `20-30% propylene glycol + DI water` blend with a data-center-rated inhibitor package. Refractometer-check the concentration — the target is within `2-3 %` of spec. Canadian installs that might see below `-15 °C` need a minimum of `30 % propylene glycol`. Never use automotive coolant. Never use undiluted glycol. Never use tap water.

Pressure-test the repaired loop before re-commissioning. Charge to `1.5× working pressure` (typically `1.5-2.0 bar`) with pump off, close the isolation valve, and watch for `30 minutes`. Decay `<0.05 bar` = seal holds, proceed to bleed + start. Anything more = find the remaining leak, don't re-energize. Bleed air from the high point of the loop after charge; trapped air causes pump cavitation and false low-pressure alarms.

Cold-start protocol for any install that saw ambient below `0 °C` during downtime. Run the pump in anti-freeze / keep-warm mode (coolant temp `5-10 °C` target per WhatsminerTool manual) for at least `30 minutes` before starting the miner. Cold glycol is viscous, cold electrolytics have elevated ESR, and the combination of cold pump-start pressure spikes + cold-cap inrush is a known cold-plate killer. No MicroBT docs cover this; D-Central's does.

Firmware verification. Confirm miner firmware is `20230411` or newer via WhatsminerTool → status. Older builds have a documented leak-sensor false-positive issue under `>85 % RH` rack conditions. If on older firmware, upgrade via a known-good wired LAN — never over WiFi, never with a flaky switch in the path. Interrupted firmware flashes on Whatsminers trigger codes 800/801/802 and compound your recovery.

24-hour loaded burn-in before returning to production. Start the miner, watch hashrate stabilize to nameplate, monitor hashboard temperature (`60-75 °C` band under full hydro flow), confirm no leak-sensor re-alarm, confirm flow rate stable within `±0.5 L/min` of nominal. Log the numbers. Any anomaly in the first 24 hours — pull it down, don't chase it live. Post-leak miners fail in clusters if rushed back into production.

Stop DIY and ship to D-Central when: (a) any copper trace is green or any busbar shows salt crystallization, (b) the miner ran wet for more than 30 seconds before AC was killed and PCB contamination is likely, (c) a cold-plate crack is visible, suspected, or the loop has seen freeze-cycle history, (d) the leak alarm persists after a confirmed-clean sensor and confirmed-good wiring (control-board channel fault), or (e) you don't have the bench + oven + DI water + PPE to do a proper rinse. Pack the miner upright, drain the loop side first, include a packing slip with: observed code, hex flags from `power.log`, date of leak, coolant chemistry in use, and how long the miner ran wet before kill. That metadata saves diagnostic time and cuts your repair cost.