Gob Watch: Gob Heating Early Warning for Longwall Mines

VERTICAL-DRIVEN ARCHITECTURES · MINING

Gob Watch

A system design for mine managers, ventilation and safety engineers, and the engineers who would build it. It shows how a longwall mine can detect a gob heating as a trend over days, separate it from the pressure, fan and nitrogen events that mimic one, and keep every advisory outside the certified alarm path.

THE CASE IN BRIEF

• A gob heating does not trigger an alarm. It appears as a slow drift in carbon monoxide behind the seals over days to weeks.
• The records that would reveal it are split across the monitoring historian, tube bundle and bag sample sheets, nitrogen injection logs, barometric readings and fan data, much of it on paper.
• No existing system separates a developing heating from a barometric swing, a nitrogen injection response or a fan pressure change.
• The cost of missing it is a sealed panel, a lost longwall and the safety exposure that comes with both.
• Gob Watch fuses every channel into one model the mine owns and issues advisories that sit beside the certified monitoring system, never in place of it.
• The first phase replays the mine's last event and stops the work cheaply if no trend signal is found.

WHY THIS MATTERS

The certified system is built for a different job. The atmospheric monitoring system watches for fast exceedances against fixed alert and alarm levels, and it is deliberately blind to slow trends in places its sensors do not reach.

Context can mimic a heating. A carbon monoxide rise behind a seal can be oxidizing coal or a pressure swing squeezing gas out of the gob, and nitrogen injection suppresses the very trend it responds to. Only the channels read together can tell them apart.

Monitoring changes belong to the regulator. In a gassy coal mine the ventilation plan, not IT, controls how monitoring changes, so any new system must be provably an addition and never a substitute.

And the record is safety-critical. Ineffective controls, rather than hazards themselves, are what kill people in high-hazard industries, which makes the paper trail behind every control a record the mine must keep.

This creates a structural exposure.

• Gob-side gas readings arrive as scanned sheets that no system computes on.
• Barometric, fan and nitrogen injection data sit apart from the gas record they explain.
• The certified alarm path has no view of trends over days.
• Re-entry and reseal records are assembled after the event rather than as the work happens.
• No one can show auditors and insurers that safety data never leaves the property.
• A new tool risks becoming a de facto substitute for required monitoring.

See a gob heating in days, not discover it in a fire.

Every channel is read together, every advisory is decided by a person, and the certified alarm path is never touched.

WHAT YOU'LL LEARN

Inside the architecture, you'll find:

• Why a gob heating appears as a slow drift across records held in different systems and different people's heads
• How seven monitoring systems each see one slice of the gob, and why none separates signal from context
• The four goals that fix the design, from never touching the alarm path to keeping only records the mine will keep
• Three scoping decisions and what each costs, from a replay verdict to one central inference node
• What the design chose against, and what it ruled out of scope entirely
• The layered stack from eight sources through two adapter families to one fused trend panel
• The fifteen-object model of the mine, with the longwall panel shown as Praxis records it
• How paper tube bundle and bag sample sheets become comparable objects through digitization and two-keyer reconciliation
• How a one-way data diode turns a no-egress posture into something auditors and insurers can verify
• Why all inference runs at the surface facility and nothing new enters the classified underground
• The four open-weight models, the memory arithmetic for the node, and why license terms decide the sovereign footprint
• How the replay verdict, written in advance, can stop the work cheaply
• How shadow mode, an alert budget and a disagreement path keep the advisory channel trusted
• The six named failure modes, five lessons the design encodes and the questions still open
• How Praxis contextualized the requirement, read it through eight lenses and reasoned every choice in the design

DOWNLOAD THE ARCHITECTURE

Read the complete architecture, from the problem every monitoring system sees only part of, through the object model, ingestion, inference placement and model register, to the replay-first rollout and the reasoning Praxis recorded for every choice.

A heating caught as a trend leaves time to act; a heating caught as a fire does not. Mines that read the record they already hold, on hardware they own and beside the systems the regulator certifies, can see it in time.