Guest comment: Why mining safety systems fail

Mining is getting smarter. From autonomous haul trucks to real-time fleet management systems, digital transformation is reshaping how mine sites operate. But despite this wave of automation, some of the industry’s most serious safety risks remain stubbornly present. “Struck-by” and “caught-in/between” incidents continue to account for a significant share of mining injuries and fatalities, particularly those involving heavy mobile equipment.

While new technologies and processes have raised the floor on safety, they haven’t eliminated blind spots altogether. 

Lack of data or intelligence isn’t a problem, but applying it effectively remains a stubborn roadblock. We have alert systems, but they tend to overwhelm operators rather than support them. According to one university-backed research paper, consistent exposure to audible alerts is one of the primary contributors to mental fatigue in the industry, impacting miners’ performance and their ability to do their jobs safely.

Where safety automation does exist, it’s often designed in a way that disrupts workers rather than support them, and technologies like video monitoring tend to be too primitive to cope with such a busy environment. All of these shortcomings conspire to allow risk to hide in plain sight, and a solution is long overdue.  

The limits of traditional safety systems 

Mining is such a unique environment that few traditional safety systems were designed to deal with the pace and pressure that come with it. Machine vision exists, but object detection tools are often so generic that they are unable to distinguish between a person and a piece of equipment with the reliability required in such fast-moving, high-risk zones.

Meanwhile, alert systems tend to overcompensate, issuing frequent warnings that don’t always reflect real danger. Over time, this can train operators to dismiss or mute alerts altogether, creating a dangerous gap between what the system sees and what the worker perceives. This isn’t a worker problem, but a systemic one.  

This “alert fatigue” is particularly acute in mining environments, where dust, vibration, poor lighting, and constantly shifting activity make precision more difficult to achieve. A well-intentioned alert that triggers needlessly, especially during peak work cycles, can break concentration, delay progress, or be seen as a nuisance.

And when alerts are ignored or overridden, the entire safety net starts to unravel. Mining needs safety technology that reacts, but it also needs safety technology that understands the environment it’s working in. 

Why visual intelligence needs simple, rugged design 

In theory, AI can make split-second decisions with superhuman accuracy. Advances in machine vision can allow AI to distinguish between humans and other objects with overwhelming accuracy. But in mining, theory doesn’t count for much unless the system can survive the real world. That means hardware that holds up in dust-choked air, on vibrating machines, in low-light tunnels and open pits alike.

It also means software that doesn’t rely on remote servers or constant connectivity. Decisions need to happen at the edge – on the vehicle, in the moment, and without latency or dependency on external infrastructure. 

Equally important is how the system communicates with the operator. In a cab already filled with gauges, radios, and movement, a new screen or complex dashboard might seem helpful on the surface but could unintentionally become a hazard in and of itself.

What works best in mining tends to be simple: clear visual cues like LED indicators, or voice alerts – used sparingly – that cut through background noise without overwhelming the operator. When systems stay quiet until something truly demands attention, workers are more likely to trust what they hear. 

Lessons from the pit 

AI safety systems are already being used across a growing number of mining operations, often as fully embedded tools on everything from underground loaders to light vehicles. But rather than rely on off-the-shelf object recognition models, which can easily mistake a shadow or a cone for a person, they’re trained specifically to detect pedestrians using stereoscopic vision and edge-based neural processing.

That narrows the focus to what truly matters, making the system and any alerts it produces more impactful and trustworthy.  

This is important, because false alerts are perhaps one of the biggest hazards of all. One false alert and workers will continue, but two or three false alerts in quick succession will lead most workers to power down the safety system. Again, it’s important to note that this isn’t about blaming workers themselves – it’s about designing technology and safety protocols that support them rather than hinder them.  

In a mining environment, safety should never be the sole responsibility of operators and “boots on the ground” – it should be built into the working environment as standard.