Coal Mine Health: Data, Decisions and How It All Connects

Planning explainer for the pilot. Mine practice described here is typical; confirm every point (systems, timings, roles) at the actual mine.

1. What does “healthy” mean? Six vital signs

A mine is healthy when it digs what it planned, safely, at a sensible cost, and can ship it. Each vital sign needs its own data. The tags show what we have covered so far.

Output

Did we dig the planned coal?

Covered

Movement

Did the coal leave: road, rail, stock yard?

Covered

Machines

Are shovels, dumpers and the surface miner working?

Partly: breakdowns only, not hours or idle time

People and safety

Enough operators? Is the pit safe?

Not yet

Quality and cost

Right grade? What does a tonne cost?

Not yet

Surroundings

Weather, power, water, railways, community.

Partly: weather and rakes
Honest answer to “have we covered everything?” No. Version 1 covers the daily plan-versus-actual loop well. Machines, people, safety, quality and cost are the next layers. That is deliberate: start with one loop that works, then widen.

2. Every data source, in plain language

Trust level: how much you can rely on it alone, before cross-checking.

SourceWhat it tells youBest reliable way to get itTrust
Covered in version 1
Production plan and actuals (SAP)How much coal we meant to dig versus how much we dug.Scheduled nightly extract arranged with SAP and IT. Cross-check against survey or stock measurement.High for totals. It never says why.
Dispatch (Coalnet: RFID, weighbridge, rakes)How much left the gate, by road or rail, and to whom. Shows queues and rake delays.Scheduled extract from SAP/Coalnet owners. Machine-recorded, so the most dependable source.High
Shift reports (paper)The shift’s own story: output, machines working, problems faced.Photo at shift end, then OCR. An officer checks a sample. Keep paper as the legal record.Medium. Handwritten, often incomplete.
WhatsApp groupsLive events with timestamps: breakdown started, rain began, rake placed.Export now, company business number later, own app eventually. AI extracts, humans confirm doubtful items.Medium-low alone, strong once matched with other sources.
Meetings (evening and morning)Priorities, decisions, who owns which action.Record with consent, transcribe, manager approves the minutes.Medium. Decisions need approval before they count.
Phone callsVerbal escalations: spares, railway, urgent fixes.Mostly invisible today. Ask officers to log a short voice note after key calls.Low until logged.
WeatherRain stops work and wrecks haul roads.On-site rain gauge plus a weather service. Easy and automatic.High
Equipment list and breakdown logWhich machine failed, when, why, what fixed it, whether it came back.Maintenance register or SAP maintenance records, linked to the chat events.Medium-high
Next to add
Machine hours and fleet tracking (GPS or fleet system, if installed)Where time is truly lost: waiting, queuing, idle, travel.Hour meters and fleet-management exports; otherwise a simple shift sheet.High if automatic
Manpower and attendanceDo we have enough trained operators today?Attendance or HR system, plus contractor muster.Medium-high
Spares and storesWhy a repair is waiting: part not in stock.Stores and inventory system extract.High
Safety and statutory recordsIncidents, near misses, inspections, slope or ground movement, gas, dust. Whether it is safe to keep working.Safety registers and monitoring systems, strictly permissioned. The AI stays advisory only.Medium. Under-reporting is common.
Later
Drilling, blasting and overburden removalIs coal exposed for tomorrow? A leading indicator of future output.Blast and drill records, survey measurements.Medium-high
Survey and mine planWhere the coal is, how deep, how much waste rock to move per tonne.Survey department files and plan drawings.High
Coal quality (grade, ash, moisture)Whether coal is sellable at the promised grade. Affects revenue and rejections.Lab and sampling reports, including third-party samples.High if sampling is sound
Fuel, power and pumpingDiesel efficiency, power cuts, water in the pit.Fuel logs, power records, pump and sump readings.Medium
Rail and customer demandWhether anyone can take the coal: rake availability, orders, auctions.Dispatch cell, railway and sales records.Medium-high
Cost per tonne and contractorsIs it economical? Are hired machines delivering?Finance and contract records.Medium
Outside disruptionsLand, community, law and order, clearances, environmental limits.Manager’s log; hard to automate.Low, but matters

3. How to get data you can trust

1. Machines beat people

Rank sources: sensors, then system transactions (SAP, Coalnet), then forms, then chat and voice, then memory. Lean on the top of the list.

2. Capture once, at the source

No retyping. Every retyping step adds errors and delay. This is why the future app matters.

3. Two sources per claim

Output from SAP should roughly match dispatch plus stock change. A breakdown in chat should appear in the maintenance log. Mismatches get flagged, not hidden.

4. Keep the evidence link

Every structured row points back to the original message, photo or record so a manager can check it.

5. Humans verify the doubtful bits

The AI marks its confidence. Only low-confidence items go to a 10-minute daily review.

6. Start small and permissioned

Ask for a limited, approved extract per source for one mine. Written permission first, then data.

4. How the sources connect

Everything joins on a few shared keys. If these are clean, the sources fit together like a puzzle. If they are messy, nothing matches.

Sources
SAP · Coalnet · shift reports · WhatsApp · meetings · calls · weather · breakdown log · (later) machine hours, manpower, stores, safety, quality
↓
Clean and match
Read chats, OCR photos, turn voice into text, fix names (EX3 becomes EX-03), align times to shifts.
Date + shiftEquipment IDLocation (face, bench, siding)Rake / vehicle
↓
Operational event store
One row per thing that happened (breakdown, rain, delay, decision), each linked to its evidence and a confidence score.
↓
Four questions it can answer
What was planned? What actually happened? Why was there a gap? What did we do last time this happened, and did it work?
↓
Manager decides
The system prepares and suggests; the human approves. The approved plan becomes tomorrow’s “planned” number, which closes the loop.

Worked example: 10,000 t planned, 8,000 t actual (synthetic data, 18 Sep)

SourceThe clue it gives
Plan and SAPPlanned 10,000 t, produced 8,000 t, so 2,000 t short. Nothing about why.
Shift reportsB shift notes “hose burst” and “rain”. A shift mentions the late rake. Some details missing.
Breakdown logEX-03 hydraulic hose burst, 15:10 to 20:40.
Weather38 mm of rain, 17:30 to 21:00, overlapping the breakdown.
CoalnetRake placed about 4 hours late, so dispatch fell.
WhatsAppExact timestamps for when each thing started and ended.
Past incidents and manager notesSame hose failure recurred through Jul–Sep (most recently 1 Sep); experience says hoses fail more in heavy rain.
Result: three evidenced contributors (machine, rain, rake) and a small unexplained remainder. In a real mine the causes overlap, so the system should show the evidence and its uncertainty, never an exact tonne split per cause.

5. How decisions are made in a mine, day to day

The logic never changes: what we owe (target) compared with what we can actually do (machines, people, weather, roads, rakes, buyers), minus what is blocking us. Every shortfall should get a reason and an owner.

Before each shift

Is the shift ready?

Officials check the workings and safety, confirm which machines are fit and how many operators are present, brief the crew, and receive the day’s programme. Decides: shift in-charge and mine officials.

Data used: plan, machine status, attendance, weather, safety checks. Later: auto briefing sheet.
During the shift

Keep coal moving

Assign machines to faces, manage the haul cycle, react to breakdowns (move a shovel, call maintenance), stop or restart for rain, manage weighbridge queues and rake loading. Decides: shift in-charge, minute by minute.

Data used: WhatsApp, phone calls, weighbridge, fleet data. Later: instant logging and alerts.
Shift change

Handover

Output, open problems and pending jobs are handed to the next shift and written in the report.

Today paper; later auto-drafted from the day’s events.
Morning meeting

What went wrong yesterday?

Review target versus actual, name the causes, assign fixes with owners and deadlines. Decides: manager with mining, maintenance, dispatch and safety.

Product: plan-versus-actual with cited evidence, similar past incidents.
Daytime

Fix and prepare

Schedule maintenance, drilling and blasting, spares, and rake and road dispatch with the railway and customers. Stock yard level shapes how hard to cut versus how hard to dispatch.

Evening meeting

Set tomorrow’s programme

Fix the target, face and machine allocation, dispatch priority, maintenance windows and safety jobs, using weather and rake forecasts and open breakdowns. Decides: manager, approved by senior officers.

Product: transcript to draft programme and action list, manager approves.
Weekly and monthly

Bigger resources and targets

Compare monthly target with achievement, request equipment, manpower or funds from area or headquarters, and report upwards.

Product later: cross-mine comparison and recurring-problem views.
Two rules to keep: when safety and production conflict, safety wins. And the AI only advises; it never issues commands for equipment, blasting or any safety-critical action.