Home / Insights / Predictive maintenance for SA mining
Condition Monitoring 8 min read Published 16 June 2026

Predictive maintenance for South African mining

By Frank Guo · Technology & Product Leadership, addanode

TL;DR — Start with critical assets, confirmed fault history and the maintenance decision each measurement should support. Record operating conditions, establish a reviewed baseline and route alerts to a named owner who can obtain further evidence or raise a work request. The useful warning period depends on the developing condition and the measurement arrangement. Connect the reviewed finding with parts, access and the maintenance window, then retain inspection outcomes with the event history.

Reactive vs preventive vs predictive

(For the plant-floor version of this comparison — including how load shedding breaks calendar schedules — see predictive vs preventive maintenance.)

Three maintenance strategies, and many SA mines run a mix weighted toward the first two:

  • Reactive — run to failure, then fix. Cheapest until the failure halts production, damages adjacent components, or hurts someone. On critical equipment it's the most expensive option of all.
  • Preventive — service on a fixed schedule. Better, but you either service too early (wasting good component life and labour) or too late (the asset fails before its slot). The calendar doesn't know the machine's actual condition.
  • Predictive — use condition measurements and specialist review to identify changes that warrant further observation or maintenance. The asset owner determines the action using the measured evidence, operating context and the site's maintenance procedure.

Despite rising interest, industry practice in South African mining remains largely reactive and preventive. That gap is precisely the opportunity: the assets are extreme-duty, the downtime is expensive, and the early-warning physics is well proven.

The economics in one line: on a critical asset, the question isn't "can we afford condition monitoring?" — it's "what does one unplanned stop of this machine cost us, and how many do we need to prevent to pay for watching it?" Usually the answer is: very few.

Three strategies, and what each one actually knows Run-to-failure is the cheapest option until the failure halts production or damages what is next to it. A fixed schedule is better, but a calendar has no information about the machine. Condition data is the only one of the three that knows anything about the asset, which is why the question it invites is about the cost of one unplanned stop rather than the cost of watching. Reactive Preventive Predictive What decides when you act The failure does The calendar does The machine's own data does What it knows about the asset Nothing, until it stops Nothing — a calendar does not know a machine's condition Changes in bearing or gearbox behaviour under observed operating conditions How it goes wrong The failure takes production, adjacent components, and sometimes somebody's safety with it Service too early and you throw away good component life; too late and the asset fails before its slot A sensor that nobody trends is just another dial Where South African mining practice sits Largely here And here The gap — which is also the opportunity The economic question it invites Can we afford condition monitoring? Can we afford condition monitoring? What does one unplanned stop of this machine cost, and how many must we prevent to pay for watching it?
Three strategies, and what each one knows Only one of the three knows anything about the condition of the machine. What decides when you act Reactive The failure does Preventive The calendar does Predictive The machine's own data does What it knows about the asset Reactive Nothing, until it stops Preventive Nothing — a calendar does not know a machine's condition Predictive Changes in machine behaviour under the observed conditions How it goes wrong Reactive The failure takes production, adjacent components, and sometimes somebody's safety with it Preventive Service too early and you throw away good component life; too late and the asset fails before its slot Predictive A sensor that nobody trends is just another dial Where South African mining practice sits Reactive Largely here Preventive And here Predictive The gap — which is also the opportunity The economic question it invites Reactive Can we afford condition monitoring? Preventive Can we afford condition monitoring? Predictive What does one unplanned stop of this machine cost, and how many must we prevent to pay for watching it?
The three strategies above, against what each one actually knows about the machine. A calendar has no information about the asset — which is why the question worth asking is what one unplanned stop costs, not what watching costs.

Predictive maintenance vs oil analysis — not either/or

Oil and fluid analysis can provide evidence of lubricant condition and wear at the sampling times. Installed vibration, temperature or electrical measurements add observations at the acquisition intervals configured for the equipment. Align those records with operating conditions and the maintenance history, then have the responsible reviewer decide what additional measurement or inspection is needed.

What to monitor, and how

  • Vibration — the workhorse. Bearings, imbalance, misalignment and looseness all announce themselves in the vibration signature long before failure on mills, crushers, conveyors, pumps, fans and gearboxes.
  • Temperature — bearings and windings running hot signal trouble; cheap to measure, easy to trend.
  • Motor current — a clamp-on sensor reveals load, run-state and certain electrical and mechanical faults without wiring into the drive.

The point isn't to drown the asset in sensors. It's to pick the few measurements that reveal the failure modes that actually take that machine down, baseline each one to its own normal, and trend it.

How to start without boiling the ocean

  1. Rank your assets by failure cost. What does an unplanned stop of each critical machine cost in lost production, repair, safety and spares lead time? Instrument from the top of that list down, only while the maths holds.
  2. Baseline each machine. Capture normal behaviour first, then set thresholds per asset — never from a generic table, or you'll drown in false alarms.
  3. Route every alert to an owner and a work order. A reading that doesn't become a planned intervention with an owner and a deadline changes nothing. Close the loop into your maintenance process.
  4. Design for the site. Edge buffering for power interruptions, hardware rated for dust and heat, and trending simple enough that your existing team acts on it without a full-time analyst.
  5. Prove it, then scale. A handful of critical machines that demonstrably avoid a breakdown will fund the rollout and earn the team's trust.

This is the same disciplined, decision-first approach we cover in why most vibration-monitoring projects fail — the failure is almost never the sensor, it's the scoping.

Why it fits South African mining specifically

Imported predictive-maintenance platforms often assume reliable power, fat connectivity and a resident vibration analyst — none of which can be taken for granted on a deep-level or remote SA mine. A system engineered for local reality buffers data through outages, runs hardware that survives dust and heat, and presents trends your existing team can act on, backed by local support. That's how predictive maintenance moves from a pilot to a habit.

At addanode this runs through our asset & condition monitoring and mining optimisation solutions on one addaNet platform — so reliability sits alongside your safety and production data instead of in a silo. Because we build both the hardware and the software in-house and support it locally, we scope for payback on your most critical assets first, not for sensor count — and it's built to survive the conditions your equipment actually works in.

Asset, failure mode, sensor

AssetTypical failureCandidate measurementReview context
Mill (SAG/ball)Trunnion / pinion bearing damage, girth-gear wearVibration on bearings, motor current signatureOperating load and inspection history
CrusherMain-shaft bearing, eccentric wear, liner failureVibration, temperature, currentDuty, load changes and confirmed wear
Conveyor drives & idlersBearing seizure (fire risk), belt driftVibration on drives, temperature on idlers, drift sensorsDrive state, loading and inspection findings
Slurry and dewatering pumpsCavitation, seal failure, wearVibration, current, discharge pressureSuction/discharge conditions and speed
Main fansBlade imbalance, bearing failureVibration, motor currentDuty, speed and inspection history
CompressorsValve and bearing faultsVibration, temperature, current, pressureLoad cycle and service findings
WinderBrake wear, bearing faultsBrake temperature, drum vibration, currentApproved inspection and engineering review

Frequently asked questions

What is the difference between preventive and predictive maintenance?

Preventive maintenance uses scheduled tasks based on the equipment and maintenance plan. Predictive methods add condition evidence and specialist review to inform further observation or intervention. The asset owner combines those methods according to the consequence of failure, available evidence and the practical maintenance response.

Does predictive maintenance replace oil analysis?

Oil analysis and installed condition measurements provide different evidence. Align sample results, configured measurement intervals and operating history so the reviewer can decide whether a change needs further analysis or inspection. Retain the findings with the maintenance record.

Which mining assets should we monitor first?

Rank equipment by the cost of an unplanned stop — mills, crushers, conveyors, critical pumps, fans and gearboxes whose failure halts production or threatens safety. Instrument from the top of that list down, only while the payback holds. Starting with the most critical assets gives the fastest, easiest-to-justify return.

How quickly does it pay back on heavy mining equipment?

Usually fast, because the downtime it prevents is so expensive. When one unplanned stop of a critical mill or conveyor costs a shift of production, preventing even a handful of failures a year covers the monitoring system many times over. The business case is driven by the cost of the failures you avoid.

Do we need a full-time vibration analyst to run it?

Not for a well-scoped programme. Automated baselining and trend alerting let your existing maintenance team act on clear warnings, with specialist analysis reserved for the few most critical machines and the occasional deep diagnosis. Tools usable by your current team, plus local support, are what make it stick.

Want to move from reactive to predictive?

Send us your critical-asset list. We'll tell you honestly which machines are worth monitoring — and which aren't — and the smallest programme that proves payback on your most expensive failures.

Talk to our Mining Team See the solution →

Related reading

8 min read

Why vibration monitoring projects fail

Scope for payback, not sensor count.
9 min read

MHSA 8.10.1 collision prevention

The TMM compliance guide.
Solution

Mining Operations Optimisation

Safety, reliability and production in one.

Primary sources

Turn the maintenance method into an accepted installation

For a scoped project, see mining condition monitoring. It defines the mine and supplier responsibilities, approved measurement points, representative operating baselines and the evidence linking an alert to a reviewed work request and maintenance window.