跳到正文
viAct · 視覺 AI 及工地安全· Barnali Sharma·· 2026-10-05AI 評分29

地下礦場安全:AI 如何在風險升級前識別隱患

Underground Mine Safety: How AI Helps Identify Risks Before They Escalate

AI 導讀

AI 可在地下礦場事故發生前識別風險,將安全監察由事故後調查提前至警示、前兆和未遂事故階段。文章指出,地下環境視線受阻、網絡連接不穩,需結合攝影機、穿戴式裝置、邊緣設備及環境感測器,並由 Edge AI 在現場處理影片和感測器數據。

正文

Underground Mine Safety: How AI Helps Identify Risks Before They Escalate

Underground Mine Safety: How AI Helps Identify Risks Before They Escalate

“Quick AI-Powered Insights on the Topic— Freshly Updated!”

ChatGPT      Perplexity     Google AI Mode    Claude

TLDR

On April 2026, a shuttle car operator with over 23 years of experience in the coal mining industry lost his life when rock from the roof of the mine fell on him in an underground coal mine in West Virginia. The final investigation report by Mine Safety & Health Administration (MSHA) revealed three violations: the person was found to be working or travelling under the unsupported roof, the approved roof control plan was not followed, and the on-shift examination was not adequate.

Each of these findings indicate conditions that existed prior to the roof falling down. This is the strongest case for the use of AI for underground mine safety. Most underground incidents undergo visible warning stages first. The absence of a rule is not what causes the problem. It is the absence of constant supervision in areas where checking every shift, everytime is not possible by a supervisor.

Underground mines have another gap: connectivity. Network coverage below ground is often weak or missing, so safety systems that rely fully on cloud processing can go silent exactly when they are needed. That is why Edge AI, which processes video and sensor data on site, sits at the centre of any underground mine safety setup.

This blog focuses on underground mining operations, covering how underground risk differs from surface mining, how incidents escalate, risks in underground mines that AI can detect early, what role does Edge AI play in underground mining safety, and where the technology still has limits.

Underground Mining Hazards vs. Surface Mining Risks: What Actually Changes

Underground mining hazards differ from surface risks in one fundamental way: the environment itself limits what people can see, hear, and reach. On an open pit, a supervisor can often spot a problem from a distance. Underground, the same problem may sit behind a pillar, around a crosscut, or in a heading with no signal.

Here is how the key risk factors compare between surface and underground mining operations, and what each one means for AI monitoring:

Risk Dimension

Surface or Open Pit Mines

Underground Mines

What It Means for AI Monitoring

Line of sight

Long, open sightlines

Broken by bends, crosscuts and pillars

AI detects risks around bends and corners that workers themselves cannot see, and alerts them in time

Ground hazard

Slope and wall failure

Roof and rib falls, and rock burst in deep hard rock

Detection focuses on people entering unsupported or restricted ground

Vehicle interaction

Haul trucks in open space

Scoops, shuttle cars and loaders in narrow drifts

Proximity alerts must work at short range with little room to escape

Atmosphere

Open air

Gas, heat and dependence on ventilation

Processing must happen on-site, not only in the cloud

Connectivity

Generally available

Dead zones and patchy coverage

Edge AI must process data on site so alerts do not depend on a network link

Egress and rescue

Multiple routes out

Limited routes and refuge chambers

Knowing where every worker is becomes critical in an emergency

Worker location

Usually visible

Often unknown without tracking

Wearables and access logs fill the gaps camera cannot cover

The practical takeaway is simple: on the surface mines, cameras alone can cover a large share of the risk. On the other hand, in underground mines, AI can only work when cameras, wearables, edge devices and environmental sensors operate as one connected system. This is the approach a mining AI safety solution needs to take, bringing every monitoring layer into a single view across the mine.

Because connectivity cannot be guaranteed underground, the system also needs to run at the edge, processing data where it is captured instead of sending every frame to the cloud.

How Underground Mining Risks Escalate into Incidents?

Most underground incidents develop in four stages:

  • A warning sign

  • A precursor

  • A near miss

  • The incident itself.

Each stage offers a chance to intervene, and each chance is easier to act on than the one after it.

Stage

What it Looks Like in Underground Mines

Example

What AI Can Do?

1.     Warning Sign

A small deviation from the plan

A miner walks a travel way without checking in at the section entry

Log the entry and flag missing check-ins

2.     Precursor

A hazard and a person start to overlap

A scoop begins toward a crosscut where people walk

Detect the person and the moving machine in the same zone

3.     Near Miss

The overlap becomes dangerous but no one is hurt

The scoop passes within arm’s reach of the miner

Trigger alerts to the operator and supervisor, record the event

4.     Incident

Injury, damage or fatality

Pinning or striking

Support the investigation with the footage and the event timeline

Traditional safety systems generally function at level 4, that is, through investigating accidents and implementing corrective measures. In contrast, AI moves the focus to levels 1 to 3, where underground mining hazards are still cheap to fix and nobody has been hurt.

Here, it should be noted that near misses are of particular importance. Underground crews are accustomed to seeing near misses, and they tend to go unreported. However, with automated detection, every near miss turns into a data point that safety teams can respond to before the next shift starts.

The value of AI for underground mining safety is clearest when it is tied to specific risks. Each of the seven risks below has a recognisable warning stage that AI can pick up before it turns into an incident.

1. Roof and Rib Falls

Roof and Rib Falls

Similarly, as reported by Industry Queensland, in Australia, the regulator’s early information on the January 2026 Mammoth Underground collapse suggested the roof failed while a multi-bolter was being moved back from the bolted face. 

In this regard, AI fulfils two fundamental roles. The first is the ground itself. LiDAR builds a 3D image of the roof and ribs of the mines, as well as provides data about the change in the ground over time that cannot be observed by the naked eye. Research in this area is active: a 2026 study from Missouri University of Science and Technology, supported by NIOSH, combines LiDAR and thermal imaging with deep learning to detect unstable ground in the underground mines. 

viLID – viAct’s LiDAR solution, brings this capability to site teams. It captures high-resolution geometry inside shafts, tunnels and enclosed structures, and detects movements and instability in roof and rib conditions before they lead to a fall of ground.

The second front consists of the people. The warning sign in many roof fall cases is that a person enters the ground that is not yet supported. AI-powered Danger zone detection can identify anyone who crosses a marked boundary at the face of the underground site. At the same time, red zone monitoring can control such boundaries at all working shifts.

2. Machine Pinning and Crushing in Narrow Drifts

Machine Pinning and Crushing in Narrow Drifts

Underground machines work in spaces where a worker has little room to step aside. According to MSHA, two underground incidents of this type took place in the US in 2026. On March 5, a continuous mining equipment operator was pinned against a coal rib in Hillsboro, Illinois. In a similar incident in Dallas, West Virginia, on April 3, a scoop reversed into another scoop carrying a longwall shield, and the shifting shield pinned a miner.

Such incidents do not happen out of the blue. A person gets inside a pinch point formed by the equipment and the rib, while a scoop reverse operates without any spotter. Either two pieces of machinery collide in a crosscut. AI is intended to detect these moments when it is still possible to act on them.

By using cameras located at intersections, loading areas, and pathways, Vehicle control software powered by AI can monitor the movement of both individuals and vehicle/ equipment within the same visuals. It is capable of identifying when the worker moves too close to the machinery or when a vehicle goes too fast, or enters a zone where it is not allowed to.

Similarly, the proximity detection solution can flag unsafe distances between worker and machinery or worker and equipment, and trigger immediate alerts to the concerned supervisors, via on-site alerts and buzzers as well as mobile notifications. Since detection is camera-based, it can recognize a person even if they are not wearing a tag. 

With AI safety systems in place, even close calls get recorded. AI logs the event along with the footage, so the safety teams can see which intersections and tasks produce repeated near misses and fix them before they turn into serious injuries and fatalities (SIFs).

3. Gas, Heat and Ventilation Failure

Gas, Heat and Ventilation Failure

Atmospheric hazard is not visible on camera; therefore, this is where AI in underground mining depends on sensors. A gas leak detection system tracks gas concentration levels at fixed points. IoT-powered smart helmets and smart watches monitor exposure at the worker level, while also proving continuous insights on the well-being of the underground miners.

However, this does not mean that cameras are completely irrelevant when gas exposure affects a worker. AI-powered unconscious worker detection can identify an underground miner who has collapsed. This matters especially in areas where no one else is nearby or in case of lone miners.

4. Lone Worker and Missing Miners

Lone Worker and Missing Miners

In underground mines, a missing worker is often noticed only at the end of the shift. However, lone worker monitoring solutions, track movement and location of lone workers across the tunnels and shafts, so that the supervisor knows who is where at any given point in time.

Further, AI can record the entry, exit, and time spent inside the confined space mines, closing the gap between a permit on paper and what happens on the ground. 

5. Fire and Smoke in Underground Mines

Fire and Smoke in Underground Mines

A fire underground is more dangerous than one on the surface because smoke travels with the ventilation and escapes routes are limited in underground mines. Conveyor transfer points, electrical installations, and diesel equipment are common areas to watch. AI-powered fire and smoke detection for underground mines picks up visual signs of fire or smoke at an early stage and triggers safety alerts.

Further, solutions like worker smoking detection in mining sites helps enforce no-smoking rules wherever they apply on the site.

6. Explosives Handling and Blast Zone Entry

Explosives Handling and Blast Zone Entry

Explosives underground carry two risks: unauthorized access at storage points, and early re-entry after a blast.

At explosive magazines and handling areas, area control systems can detect tailgating, unauthorized entry, and handle violations. Further, IoT enabled smart locks can help ensure that access is both restricted and recorded.

Around a blast site, AI can confirm the zone is clear before firing and flag anyone who enters before clearance. Further, gas sensors can confirm that the air is safe before the crew return to the blast site. Every entry and violation gets logged, so safety teams can see where clearance procedures are being bypassed.

7. Fatigue and Unsafe Behaviour on Long Shifts

Fatigue and Unsafe Behaviour on Long Shifts

Long shifts, heat and low light wear down attention, and tired workers are more likely to take shortcuts such as skipping checks or standing in pinch points. Computer vision can flag visible signs of tiredness so supervisors can step in early. Additionally, smart watch can track signs of worker fatigue even outside camera coverage. Further, AI can monitor and detect unsafe acts such as climbing on equipment or risky positioning immediately as it happens.

Over time, the logged data shows whether violations rise late in a shift or cluster in a particular area/ zone. This data can then be used by EHS teams to adjust shift lengths, breaks, and supervision.

Edge AI for Underground Mine Monitoring: How Detection Works Without Reliable Connectivity 

Connectivity is the biggest practical barrier to underground mine monitoring. In fact, an alert that waits for a cloud connection may arrive too late due to weak connectivity, making it completely useless Thus, processing data at the site of detection is the solution to this problem.

Edge AI does exactly this. Instead of transferring all visual information to a remote/ cloud server, edge devices process video and sensor data on-site. Edge processing brings three practical benefits:

  • Alerts keep working offline. Detection continues even when the connection is interrupted.

  • Reduced latency. Processing on site removes the round trip to the cloud, so alerts reach the concerned person in time to prevent the hazard from happening.

  • Limited data traveling. Raw video is analysed and it stays on-site. Only alerts and reports are sent off-site, which keeps bandwidth needs low and sensitive footage under the mine’s control.

viAct has three different deployment modes: Cloud, On-Prem and Hybrid. In the case of underground mining sites, on-prem deployment is mostly preferred where edge AI analyses data on-site and triggers alerts. The Hybrid deployment maintains the continuous detection on-site and syncs reports to the cloud when the connection is established.

In the case of remote or temporary work sites, viMOV, viAct’s mobile edge AI device, facilitates the real-time detection of hazards in hard-to-reach areas. It is useful for development headings and places where no permanent installation has been made.

How AI Alerts Lead to Faster Response in Underground Mines

Detection makes sense only if there is someone to act on it. AI can send alerts to concerned personnel wherever they are. viAct does this through on-site alerts, SMS, WhatsApp notification, and email to different stakeholders simultaneously. 

The next step is investigating without travelling. Getting to a heading underground mine can take a long time, which slows down decision. EHS AI Agents can pull live footage and check where workers are on request. That lets a control room supervisor to confirm what happened in the underground mine, instead of sending someone down the decline to check the situation.

In an underground mining operation, alerts are most effective when they get integrated into the procedures teams already follow during every shift:

  • Permits: A permit-to-work software makes sure that all entries into confined spaces, such as ore passes, sumps and ventilation raise, and hot work near conveyors match what cameras and wearables actually record.

  • Incident reporting: Incident management software record alerts with correlated footage, timestamp, and other details, so a near miss at crosscut gets an owner and a closing date.

  • Planning: Job hazard analysis software feeds recurring alerts back into planning. If one intersection keeps producing near misses, the next shift’s task plan can change traffic routes and add a spotter there.

Limitations of AI in Underground Mine Safety

AI is a strong early warning layer, but it is not a complete safety system. Mines should plan around four limits:

  • Coverage gaps: Cameras only see what is in frame. Bends, pillars and long drifts create blind spots that wearables, drones, and sensors must cover.

  • Harsh conditions: Dust, water and vibration can reduce image quality. Lenses need cleaning schedules and camera placement needs regular review.

  • Geotechnical monitoring: Video analytics detects people entering unsafe ground. It does not measure rock stress. Predicting events such as rockburst relies on methods like micro seismic monitoring. AI in underground mining complements ground control plans; it does not replace them.

  • People still decide: An alert without a clear response procedure becomes noise. Every alert type needs an owner and an agreed action.

Conclusion & Key Takeaways

The underground incidents covered in this blog share one thing: none of the hazards were new. Unsupported roof, a machine reversing near a worker, a person inside a zone they should not have entered – each was a known risk, already covered by rules and plans. What failed was not knowledge but visibility. Underground conditions break the line of sight, the signal and the supervision that safety depends on, and incidents grow in those gaps.

That is where AI for underground mine safety earns its place. It does not replace the ground control plan, or an experienced supervisor. It covers the hours and places they cannot reach, and it catches early warning signs while there is still time to act. Mines that treat AI as a connected layer, with cameras, wearables and sensors processed at the edge and tied to clear response procedures, will move their safety effort from investigating incidents to preventing them.

Key Takeaways

  • Underground incidents rarely come from unknown hazards, but from known risks nobody could see in time.

  • Visibility, not knowledge, is the biggest safety gap below ground.

  • No single technology is enough: cameras, wearables, sensors and LiDAR need to work together.

  • Edge processing is essential underground, because alerts cannot depend on a network link in such risky environment.

  • Near misses are a mine's most valuable safety data when they are recorded automatically.

  • AI strengthens ground control plans, permits and proximity systems rather than replacing them.

  • Every alert needs an owner and an agreed action, or it becomes noise.

viAct WhatsApp Channel

FAQs

1. What camera setup does viAct AI monitoring need in an underground mine?

viAct AI is hardware and brand agnostic and connects to existing IP cameras and CCTV through RTSP, so most mines do not need a new camera network. For underground mines, favourable considerations are infrared or low-light capability, protection against dust and water, and placement at chokepoints such as section entries, intersections and conveyor transfer points.

2. How much does viAct AI for underground mine safety cost?

There is no fixed price. viAct offers software subscription per camera × per AI module × per month, plus a one-off on-site implementation fee. This means the cost depends completely on   the number of cameras, the number of AI modules selected and the contract duration. Further, it also depends on the deployment model (Cloud, On-Prem or Hybrid), as hardware is separate capex.

3. How is miner privacy protected when viAct AI tracks workers underground?

At viAct, data privacy is as important as the safety of the workers on site. For this, viAct adopts methods like automated face masking, end-to-end encryption, On-Prem deployment options and GDPR and other regional compliance rules. Further access to data is only allowed to the authorized personnel and not all.

4. Does AI monitoring work the same way in underground coal and hard rock mines?

The detection principles are the same, but the priorities differ. Underground coal mines deal with methane and coal dust, so equipment certification for gassy areas and close monitoring of the atmosphere come first. Hard rock mines are often deeper, rely heavily on diesel equipment and face risks such as rock burst, so vehicle interaction, ventilation and ground movement need more attention. A good deployment starts by matching the monitoring layers to the mine type rather than applying one setup everywhere.

5. Can one AI system monitor both the underground and surface operations of a mine?

Yes. Most mines run both, with the portal, processing plant, workshops and haul roads above ground and the workings below. A connected platform, can bring cameras, wearables and sensors from both environments into one dashboard. Safety teams then see a single view of risk across the whole site instead of managing separate systems. viAct's mining safety solution is built this way, covering underground, open pit and surface operations. Each area can still use the monitoring layers that suit its conditions.

Read more:

來源:viAct · 視覺 AI 及工地安全 · viact.ai