Talk to any maintenance superintendent at a large open-cut and they’ll tell you the same thing: conveyors are the boring infrastructure that destroys their week when they fail. Run-of-mine and overland conveyors at major Australian iron ore and coal operations have replacement values in the hundreds of millions. A belt rip on a key trunk conveyor can pull production for 12 to 72 hours depending on how bad the damage gets and how many splice crews you can mobilise.

So why is the state of conveyor monitoring at most sites still a mix of weekly visual inspections and reactive callouts when something is already wrong?

The legacy of cheap belts and expensive failures

Twenty years ago, if a belt cost $400,000 and the downtime was four hours, the maintenance philosophy was simple: run to failure, hot-vulcanise the splice, get back into production. Bigger sites have been doing this forever and the operational muscle memory is still there.

Three things have changed that calculus:

• Belt costs on long overland conveyors are now north of $5 million in some cases.
• Production rates per hour mean every hour of unplanned downtime hurts more.
• The downstream cost — ports demurrage, lost ship slots, missed contract obligations — has grown.

That’s pushed sites to take condition monitoring more seriously. But the gap between what’s marketed and what works is wide.

What “monitoring” actually means

The shopping list looks something like this:

Belt rip detection — embedded loops or RFID strips that detect longitudinal tears. Mature tech, broadly reliable, mostly catches catastrophic events rather than warning ahead.

Magnetic belt scanning — tracks cord damage in steel-cord belts. Useful for identifying belts that need replacing on schedule rather than at random.

Idler temperature monitoring — thermal cameras at fixed locations or running on inspection trolleys. Bearing temperature is a strong predictor of failure, but distinguishing a hot bearing from solar load on a hot day in the Pilbara is a real problem.

Acoustic monitoring — listens for the high-frequency signatures of bearing degradation. The good systems are remarkable. The bad ones generate so many false alerts that crews ignore them.

Computer vision-based belt edge and tracking analysis — newer, growing fast. Cameras at strategic points feed a model that looks for off-tracking, mistracking events, material spill, and surface damage on the belt.

Power signature analysis — drive motor current and torque give you indirect data on belt condition, blockages, and load distribution. Underrated and cheap.

Where the money is actually going wrong

Three failure modes show up over and over again in post-incident reviews at Australian mines:

Failure to fuse the signals. A site might have idler temperature, belt rip detection, and acoustic monitoring all installed, but they live in three separate dashboards owned by three different vendors. When a bearing fails and takes out an idler that ultimately damages the belt, the post-mortem inevitably finds that all three systems flagged something in the 48 hours before the event. Nobody connected them.

False alarm fatigue. This is the most common reason expensive monitoring systems get switched off in the field. If your idler temperature alarms five times a day and four of them are sunshine, the fifth one gets ignored. Reliable sites have invested heavily in tuning and contextualising alerts, often using local ML models that account for ambient conditions, time of day, and known idler quirks. CSIRO and a few specialist consultancies have done good work here, and the ACARP-funded research on conveyor monitoring is worth looking at.

Inspection trolleys that never get used. Sites buy an autonomous inspection trolley, run it for three months, then the wheels need maintenance, the cameras need cleaning, the data nobody is reviewing piles up, and quietly it goes back in the shed. The same pattern shows up at dozens of operations.

What the better operations are doing

The Australian mines that have actually moved the needle on conveyor reliability share a few habits.

First, they treat data integration as a first-class problem. The various monitoring systems all feed into a single reliability platform that the maintenance planner uses to prioritise work. That sounds obvious; it almost never happens in practice without someone forcing it.

Second, they pair each technology with a clear failure mode it’s expected to catch and a service level for response. If acoustic monitoring flags a bearing as Stage 2 degradation, there’s a written workflow: replace within X shifts, log the prediction, log the actual condition on removal. Without that feedback loop, the monitoring system never improves.

Third, they accept that some failures are still going to happen and they invest in fast-response splice capability, on-site belt stockholdings for critical conveyors, and clear escalation paths. The goal is reducing the impact of failures as much as preventing them.

The next two years

The promising development is the application of computer vision and acoustic models to conveyor data, where modern transformer-based models can do something useful with messy multi-sensor streams that older techniques struggled with. A few Australian sites are trialling these on overland conveyors with promising results. AFR ran a useful piece on autonomous mining technology adoption earlier this year that touches on some of the same themes.

What’s not going to change: the basic discipline of belt management, splice quality, idler replacement on schedule, and good drive maintenance still does more for conveyor reliability than any monitoring system. The monitoring catches what the discipline misses. It doesn’t replace it.

If your sites are still running reactive maintenance on critical conveyors in 2026, the conversation to have isn’t about which sensor package to buy. It’s about why the maintenance philosophy hasn’t moved. Sensors are easy. Operating habits are hard.