Condition monitoring in Africa: the case is the lead time, not the ROI claim
Almost every condition monitoring proposal you will read quotes a percentage — unplanned downtime cut by thirty per cent, or fifty, or seventy-five. Treat those numbers with suspicion, including when they come from us. They are averages of other people's plants, published by people selling equipment. The argument that actually holds on an African site is duller and much harder to dispute: the spare is a long way away, and knowing early is the only thing that buys you the time to get it.
By Frank Guo · Technology & Product Leadership, addanode
Distance turns a repair into a shutdown.
In a European plant, a failed bearing on a critical fan is an expensive day. The distributor holds stock, the specialist is a drive away, and the machine is back before the week ends. The condition monitoring case there really does rest on percentages, because the difference between planned and unplanned is measured in hours.
On a Zambian copper mine, a Tanzanian cement plant, a Kenyan food processor or a Zimbabwean sugar mill, the same failure behaves differently. Specialised components are ordered rather than collected. Lead times run in weeks and, for the more specialised items, considerably longer. Clearing customs is its own timeline. The technician who can commission the replacement may be on a different continent. The gap between "we knew three weeks ago" and "it stopped this morning" is not a percentage — it is the difference between a repair and a shutdown.
That is why the sequence matters more than the technology. Early warning is only valuable if it arrives before the procurement clock has to start. A method that detects a fault two days out is nearly worthless where the part takes six weeks; a cruder method that flags degradation two months out is worth a great deal. Precision is not the goal. Warning time is.
Four layers, and most sites should stop at two.
1 · Motor current and run hours
The cheapest layer and, on most plants, the highest return. Current tells you load, and load tells you when a pump is cavitating, a conveyor is dragging, a screen is blinded or a fan is fouled. Run hours turn calendar maintenance into use-based maintenance, which is what the manufacturer specified in the first place. No new sensor is needed on plant with a communicating drive or MCC.
2 · Temperature
Bearing and winding temperature, trended rather than alarmed. Temperature is a late indicator relative to vibration — but "late" here means days to weeks, which on a site with a six-week lead time is still early enough to matter. Cheap, robust, and it survives dust and heat better than most alternatives.
3 · Vibration, trended
Overall vibration level, measured continuously and watched as a trend. This is where genuine early warning starts — a degrading bearing changes its signature long before it changes its temperature. Trending answers "is this machine getting worse, and how fast", which is the question that drives a purchase order.
4 · Vibration spectra and diagnosis
Frequency analysis to identify which fault — inner race, outer race, misalignment, imbalance, looseness. Valuable, specialist, and a different discipline from trending. It answers "what is wrong" once trending has answered "something is".
Why most sites should start at layers 1 and 2 rather than 3 and 4. Not because the higher layers are worse — they are better — but because plants that begin with a full vibration programme usually end up with an expensive system nobody reads. Layers 1 and 2 come mostly from equipment already installed, produce a trend within weeks, and build the habit of looking. Once a site is acting on that data, layer 3 is an obvious extension. Before that, it is an unread dashboard.
The system works. The routine never starts.
The most common outcome of a condition monitoring project is not a wrong reading. It is a correct reading that nobody looked at. The pattern is consistent enough to predict:
- Alarms are set at manufacturer defaults rather than at levels derived from the machine's own baseline. They fire constantly for the first fortnight, everyone learns to dismiss them, and the system is now furniture.
- Nobody owns the review. Condition data needs someone whose job includes looking at the trend weekly and deciding. Where that is nobody's job it is nobody's job, regardless of how good the dashboard is.
- There is no route from a finding to a purchase order. A trend that says "this bearing has eight weeks" produces nothing if raising a requisition on a machine that is still running requires a battle. The commercial process has to accept evidence, or the evidence is decorative.
- The one machine that mattered was not instrumented, because the pilot covered what was easy to reach rather than what stops the plant.
Three of those four are organisational rather than technical, which is why the honest first question on a condition monitoring project is not "which sensors" but "who will look at this on a Monday, and what are they empowered to do about it?" A site that cannot answer that should fix it before buying anything.
Our South African team has written at length on the specifics — why most vibration-based motor monitoring projects fail, what a bearing tells you months before it stops, and choosing predictive or preventive per asset rather than per ideology.
Five things that are not in the vendor's design assumptions.
- Power events are a load on the machine, not just on production. Repeated restarts, voltage sags and swells accumulate stress on motors and drives. A plant in Uganda, where the distributor has told Parliament that outages almost doubled, or one in Nigeria running heavily on standby generation, is putting its rotating plant through duty its datasheet did not anticipate. Logging power quality alongside condition data is what lets you connect a pattern of failures to a pattern of events instead of treating each one as bad luck.
- Dust and heat shorten everything. Copperbelt dust, cement plant abrasion, coastal humidity — instrumentation has to survive the same environment as the machine, and sensors that are themselves a maintenance burden get switched off.
- Connectivity is intermittent. A monitoring system that only works when the link is up produces gaps, and gaps in a trend are worse than in a log: you cannot see a rate of change across a hole. Local storage with back-fill is not a nice-to-have here.
- Skills are concentrated, not absent. There are excellent vibration analysts across the continent, but they are scarce and their time is expensive. That argues for continuous instrumentation feeding a specialist who reviews remotely, rather than periodic site visits that catch whatever happens to be wrong that week.
- Spares strategy and monitoring are the same decision. Continuous condition data is also inventory data: it tells you which critical spares you actually need on the shelf, and which you have been holding for a machine that has never degraded. On sites where working capital is tight, that is often the faster payback.
Warning window against spares lead time
| Asset | Typical failure | Warning a sensor gives | Why it matters here |
|---|---|---|---|
| Mill / crusher bearings | Bearing damage, girth-gear wear | Weeks to months (vibration, current signature) | Bearings for large mills ship from Johannesburg, Durban or overseas — the warning is the shutdown avoided |
| Conveyor drives and idlers | Seizure, fire | Days to weeks (vibration, idler temperature) | A belt fire stops the plant; a seized idler is caught on a walk with a thermal reading |
| Slurry and dewatering pumps | Cavitation, seal and wear failure | Days to weeks (vibration, current, pressure) | Dewatering that stops floods the working; spares are shared across sites |
| Fans, compressors, refrigeration | Imbalance, bearing and valve faults | Weeks (vibration, temperature, current) | Ventilation and cold chains have no tolerance for a surprise |
| Gensets | Fuel, battery, cooling, overload | Hours to days (run-hours, temperatures, load) | On this continent the genset is a production asset, not a backup |
The list is shorter than you think.
Nearly every site we assess has between four and ten machines that genuinely stop production when they stop. Everything else has redundancy, a workaround, or an acceptable outage. The starting scope is that list, and it is worth writing down before talking to anyone about sensors:
- Which machines stop the plant? Not which are biggest or most expensive — which ones, when they fail, stop output.
- For each, how long is the spare? On the shelf, days, weeks, or a specialist order. This number sets how much warning time you need, and therefore which layer of monitoring is worth buying.
- What can be read today? Drives, MCCs and PLCs frequently already publish current, run state and sometimes temperature. On most plants a useful first trend costs no new hardware at all.
- Who reviews it, and when? Answer this before ordering anything.
Country context: mining condition monitoring in Zambia · mining and cement monitoring in Tanzania · instrumentation and control in Zimbabwe · condition monitoring across our markets.
Frequently asked questions
How much downtime does condition monitoring actually prevent?
We are not going to give you a percentage, and we would be cautious about anyone who does. Published figures are averages across other companies' plants, usually assembled by parties selling something, and they depend entirely on what was monitored, how critical it was and whether anyone acted on the data. The number that matters is specific to your site: how many production-stopping failures you had last year, how many were preceded by weeks of measurable degradation, and how long your spares take to arrive.
Is vibration monitoring necessary, or is temperature enough?
It depends on your warning-time requirement. Vibration changes earlier than temperature, so it buys more notice — often weeks more. If your critical spares are on the shelf, temperature and current may be sufficient. If a replacement takes six weeks to arrive and clear customs, the extra warning vibration provides is precisely what makes the difference between planned and unplanned. Let the lead time choose the technology, not the other way round.
What is the difference between vibration trending and vibration analysis?
Trending watches the overall level over time and answers "is this machine getting worse, and how quickly". Analysis examines the frequency spectrum and answers "what specifically is wrong" — inner race, outer race, misalignment, imbalance, looseness. Trending is continuous and can be automated; analysis is a specialist skill applied when trending raises a question. Most sites need trending on many machines and analysis on a few, not the reverse.
Can we use data we already have?
Usually a substantial part of it. Variable-speed drives, motor control centres and PLCs frequently publish motor current, run state and sometimes temperature already, and that data is often discarded rather than retained. Starting from what is already measured produces a trend within weeks at close to no hardware cost, and it tends to reveal which machines deserve dedicated sensors.
How many machines should a first project cover?
The ones that stop production — typically four to ten on a given site. Covering everything produces a system nobody reads; covering the critical list produces decisions. The selection question is not which machines are most valuable but which ones, on failing, stop output, and how long their replacement parts take to arrive.
Does condition monitoring help with spares and working capital?
Frequently more than it helps with downtime, and it is the benefit least often mentioned. Continuous condition data tells you which critical spares you genuinely need on the shelf and which you have been holding for machines showing no degradation at all. On sites where lead times force large safety stocks, converting some of that inventory into evidence-based ordering can release more cash than the avoided downtime is worth.
What does condition monitoring cost?
Driven by how many machines, which layer you need for each, how much can be read from drives and MCCs you already own, and whether power and network reach the equipment. We quote a written cost band after a remote assessment. The single largest variable is the third one — a plant whose drives can be read costs a fraction of one where every machine needs new instrumentation.
Start with the machines that stop the plant.
Send an engineer your critical machine list and the lead time on each spare. We'll tell you which layer of monitoring each one justifies — and which need none.