A collision avoidance system (CAS) detects people and vehicles around a mobile machine and reacts before they meet. At its simplest it warns the operator. At its most advanced it slows and stops the machine without waiting for anyone. On paper, few safety investments are easier to justify in an underground mine. In practice, the most common way a CAS fails has nothing to do with its sensors. It fails because the operators switch it off.
The System Operators Learned to Defeat
The worst problem I have seen on a CAS deployment was not detection. The hardware did what it was specified to do: whenever a tagged pedestrian came within range, the operator got an alarm. The trouble was that range was the only thing the system understood. A worker standing in a refuge station while a scoop drove past triggered the same alarm as a worker stepping into its path. So did someone behind the machine while it was travelling forward, away from them.
Operators heard the alarm over and over through a shift, for people who were in no danger at all. It did not take long before they found ways to disable it.
It would be easy to call that a discipline problem. It is a design problem. An alarm that sounds when there is no danger teaches the person hearing it that the alarm means nothing, and the operator is the one person whose trust the system cannot work without. Worse, a disabled CAS does not look disabled from the office. The mine still believes the layer is there.
Everything below comes back to that one question: does the alarm still mean something to the operator six months after commissioning?
Three Levels, Three Different Costs
The industry reference is the Vehicle Interaction model from EMESRT, the Earth Moving Equipment Safety Round Table. It describes nine layers of control. The first six are design and operating controls: segregated travelways, traffic rules, procedures, operator competence and fitness. Only layers 7, 8 and 9 involve collision avoidance technology, and each one asks more of the machine than the last.
Level 7 is operator awareness. The system detects something nearby and tells the operator. It can be retrofitted on almost anything because it needs very little from the machine itself, but it also knows very little: distance, and not much else. That is the level where the refuge station alarm lives.
Level 8 is advisory. The system reads the machine’s own data (direction of travel, gear selection, speed) and uses it to judge whether a detection matters. A pedestrian behind a scoop travelling forward is not the same risk as a pedestrian behind a scoop in reverse, and Level 8 can tell the difference. That requires a connection into the vehicle, which is a bigger project than bolting on a display.
Level 9 is intervention. If the operator does not respond to the warnings, the system slows the machine and brings it to a controlled stop. That means an interface into the machine’s controls, and in a mixed fleet from several manufacturers, that interface is where projects stall. ISO 21815 exists to standardize the connection between a third-party CAS and the machine, but a standard on paper and a working integration on every model in your fleet are two different things.
Climbing the levels also raises the price of a false positive. At Level 7, a false alarm is noise. At Level 9, it is a loaded scoop stopped mid-cycle, and it shows up in the production numbers the same week. The adoption problem does not go away as you move up. It gets more expensive.
| Level | What it does | What it needs from the machine | Cost of a false alarm |
|---|---|---|---|
| Level 7: awareness | Detects a tag nearby and warns the operator | Almost nothing. Retrofits on to most equipment | Noise in the cab |
| Level 8: advisory | Judges whether a detection matters, using direction of travel, gear and speed | A data connection into the vehicle | Noise, plus lost trust in a system that should have known better |
| Level 9: intervention | Slows the machine and brings it to a controlled stop if the operator does not react | An interface into the machine controls, which is what ISO 21815 standardizes | A loaded scoop stopped mid-cycle, visible in the week’s production numbers |
Fixing False Positives Is the Real Project
Two changes did more for adoption than any sensor upgrade.
The first is direction. A single detector on a machine can tell you how far away a tag is, not where it is. Adding tags to the machine itself, at the front and at the rear, lets the system work out which side of the machine a pedestrian is on. Combined with direction of travel, the alarm for someone behind a scoop moving away from them can be dropped, while the alarm for someone in its path stays loud.
The second only works if the mine also runs an RTLS. The location system already knows where the refuge stations and other safe locations are, so it can tell the CAS to ignore pedestrian tags inside them. A worker standing in a refuge while a scoop passes is doing exactly what they are supposed to do, and the operator should not be getting an alarm for it.
That exclusion has to be designed with care. It should be tied to locations that are physically protected, a refuge cut-out, not a spot along the drift that simply feels out of the way. It also depends on the RTLS being accurate and available, since a position error in the wrong direction suppresses an alarm that should have sounded. This is also the point where CAS and RTLS stop being two separate purchases. The same location infrastructure that drives location-driven Ventilation on Demand can make the collision system quieter and more trustworthy.
The Scoop Bucket Problem
Detection technology matters too, and the clearest example I know is the space directly in front of a scoop bucket.
Ultra-wideband (UWB) measures distance by timing a radio pulse between the machine and a tag. In an open drift with a clear line of sight, it is precise. The bucket takes that line of sight away. A person standing right in front of it has a large mass of steel between their tag and the machine’s antennas, and the UWB signal gets blocked or bounced around it. Ranging becomes unreliable or drops out, in the one spot where the operator’s own view is also blocked by the bucket.
Low-frequency (LF) electromagnetic systems handle that spot better. The machine generates a magnetic field around itself and the tag measures its strength. Because detection does not depend on a clear radio path, a person in front of the bucket is still inside the field. The field’s shape is also predictable, which keeps the protective zones consistent from one machine to the next.
Both have limited reach. LF covers the immediate area around the machine, and UWB reaches roughly 50 to 80 metres. That is enough in front of the bucket. It is not always enough for a loaded scoop coming down a ramp toward an intersection, where the warning has to arrive well before the stopping distance runs out. Close-range detection needs help from a longer-range layer, typically chirp spread spectrum (CSS) radio, which reaches around 250 metres. The trade is precision: the farther a system reaches, the less precisely it places the tag. CSS tells the operator something is coming well before UWB could, but not exactly where it is. It also tolerates the reflections off drift walls that degrade other radio signals, and those same reflections are what let an RF system detect a tag around a corner, where anything relying on line of sight sees nothing.
Radar, LiDAR and cameras fill a different gap. They detect things that carry no tag: a visitor, a contractor who was never issued one, a stalled vehicle. They need line of sight, and dust, water spray and fog all work against them underground. They complement a tag-based system rather than replace it.
The architecture that holds up is layered: long range to warn early, short range to be certain up close, and a machine that knows which way it is going.
Where Canada Sits
No Canadian regulator that I am aware of currently prescribes a CAS level. Ontario’s Regulation 854 requires mines to run a traffic management program with measures to prevent collisions, which addresses the first six EMESRT layers without mandating any technology. Quebec’s mining regulation (RSSM) similarly deals with pedestrian circulation and visibility underground. The regulatory push toward Level 9 is coming from elsewhere, South Africa most notably, while in the United States the federal mandate for proximity detection still covers only continuous miners in underground coal.
That leaves adoption in Canada driven by operators themselves: corporate standards, the commitments of large miners that belong to bodies like ICMM, and each site’s own risk assessment.
Every site’s situation is different. Underground and surface operations face different interactions, and the fleet mix and layout change the answer again. The broad picture, though, is that Levels 8 and 9 are not the trend in Canada yet. The hesitation is the same problem this article opened with: a concern that the system will slow operations and throw too many false positives. Until those two concerns are answered on site, the higher levels remain the exception.
One trend makes this more pressing here than in most places. Canadian underground mines have been early adopters of battery-electric equipment, and battery-electric machines are quiet and deliver torque instantly. A pedestrian loses the audible cue a diesel engine used to give, and the machine closes distance faster when it moves off. The margin that noise and slow acceleration used to provide now has to come from somewhere else.
What CAS Does Not Solve
A tag-based system only sees tags. A visitor who was never issued one, a cap lamp left on the charger, a tag with a dead battery: to the system, that person does not exist.
It does not replace the first six layers either. Segregated travelways, traffic rules and procedures still carry most of the load, and a CAS installed to compensate for weak traffic management will spend its life raising alarms the layout should have prevented. ICMM’s own position is that technology is not a silver bullet and that operating discipline is a prerequisite.
Tuning is never finished. New headings, new equipment and changes to the mine layout all shift where false positives come from, and the exclusion zones and thresholds need someone who owns them after the commissioning team leaves.
Disabling has to be visible. If a machine can run with its CAS bypassed and nobody outside the cab knows, the mine’s safety case rests on a layer that may not exist on any given shift.
The sensors get most of the attention in a CAS project, and the scoop bucket shows they deserve some of it. The deployments that last, though, are the ones where the alarm still means something: the system knows direction, knows which locations are safe, and saves its loudest warning for a real threat.
Most of the systems described here work machine to machine, with no network underground at all. What you can and cannot run in a mine without LTE is a separate question, and a longer one.
Evaluating a collision avoidance system, or struggling with adoption on one already installed? Get in touch. No commitment, no sales pitch.
Leave a Reply