World Sleep Industry ENCYCLOPEDIA

Underground Mining Personnel Fatigue Monitoring Systems

4.2.4

System purpose, covered workers, and management boundary

An underground mining personnel fatigue monitoring system is a supporting system that combines roster and task information, worker reports, appropriate physiological or behavioral signals, and equipment-operation data to identify possible fatigue risk and trigger human verification and field controls. It belongs in WSIC 4.2.4. It may serve underground haulage, extraction, development, ground support, drilling and blasting, ventilation and dewatering, electrical and mechanical maintenance, surveying, and supervision, but the mine must declare the commodity, mining method, equipment, areas, rosters, population, and jurisdiction. A model developed for surface mining or public-road driving should not be transferred underground without validation.

Monitoring is only one tool inside mine fatigue risk management. It does not replace staffing, roster and recovery design, ventilation and illumination, traffic organization, equipment isolation, ground and gas controls, work authorization, training, supervision, emergency response, or required communications and tracking. It cannot diagnose a sleep disorder or use a single score to decide fitness, misconduct, or accident causation. Production goals must not override a miner's report, peer observation, or supervisory judgment. No alert means only that the system produced no alert; it does not prove wakefulness, a safe environment, or permission to continue the task.

Underground tasks, environmental interference, and complete exposure

Risk assessment should connect a fatigue indication to task consequences. Trackless haulage, shuttle cars and load-haul-dump machines, continuous mining, development machinery, roof support, drilling and charging, hoisting, conveyors and crushing, maintenance, walking inspections, and emergency response present different collision, entanglement, pinch, ground-control, blasting, electrical, fire, and escape hazards. Narrow entries, restricted sight distance, grades and curves, dust, diesel emissions, noise, vibration, heat and humidity, low illumination, glare, monotony, and heavy protective equipment can affect both human performance and sensor readings.

Complete exposure starts when the worker prepares to leave home or camp and covers bus or personal travel, the pre-entry meeting, changing and equipment issue, cage or decline transport, travel to the work area, production, waiting, breakdowns, temporary reassignment, overtime, return to surface, and safe travel afterward. It also combines second jobs, training, standby, and actual rest. Waiting in a vehicle, an equipment stoppage, or a cage queue is not necessarily recovery. Long commuting, rotation across day and night, consecutive nights, physical load, and environmental heat should be interpreted together. When the planned roster differs from actual records, the system should use verifiable actual exposure and preserve the source and reason for missing data.

Data sources, signal meaning, and limits of use

Predictive approaches may use actual work and recovery opportunity, circadian timing, and task history to estimate future risk. Detection approaches may use eyelid or head features, reaction testing, heart-related signals, posture, vehicle control, or work behavior to indicate a current state. Location tags, inactivity detection, fall or man-down functions, proximity detection, gas instruments, and production telemetry have different purposes. Sharing a cap lamp, belt, vehicle terminal, or network does not turn them into fatigue evidence. A person may remain still during maintenance, observation, permit waiting, or traffic avoidance, while loss of communications does not mean sleep.

Each input needs a documented sensor position, sampling and analysis window, time synchronization, algorithm version, threshold source, and unavailable state. A hard-hat brim, eye protection, respirator, facial variation, coal dust, or mist may obscure a camera. Rough travel, looking down for footing, reversing, and turning to confirm clearance change head and eye features. Sweat, skin contact, temperature, and exertion influence wearable signals. Steering and braking patterns change with vehicle, road, load, and automation. A tag's associated zone is not necessarily a precise coordinate; node coverage, reflection, obstruction, power loss, and damage change location accuracy. Missing information must not be silently imputed as low risk.

Deployment design, miner participation, and field calibration

The objective should be set before a technology is chosen: identifying high-risk rosters, warning an equipment operator immediately, supporting a supervisor's intervention, evaluating a roster control, or analyzing group trends. These purposes need different latency, sensitivity, evidence, and response. A device intended for research or post-shift analysis must not be presented as real-time protection. Miners and their representatives, ventilation and electrical teams, occupational health, information security, rescue personnel, and contractors should participate before deployment. The evaluation should cover comfort, field of view, auditory load, understandable alerts, language, operation with gloves, charging, cleaning, and maintenance.

Equipment must be suitable for every intended underground area and meet applicable requirements for explosion protection, intrinsic safety, electromagnetic compatibility, batteries, and wireless devices. A small consumer wearable is not thereby acceptable in a restricted area. Field calibration should cover shifts, skin tones and facial features, glasses and protective equipment, machine types, roads, lighting, dust, mist, temperature and humidity, and communications coverage. Recognition must not be improved by asking miners to remove necessary protection. Pilots should begin in shadow mode, comparing alerts with observations, self-reports, tasks, and outcomes before operational expansion.

Alert chain, human verification, and safe response

For each alert class, procedures should state who receives it, how it is acknowledged, the allowed response time, how loss of contact escalates, and who may close it. An alert should first prompt brief human verification at a safe location, not demand screen interaction while a vehicle is moving, machinery is running, explosives are being handled, a worker is at height, or the person is in a hazardous zone. A mobile-equipment operator follows the work procedure to stop at a suitable place, stabilize the machine, and prevent unintended motion. Other workers first bring the task to its defined safe state and then contact supervision through reliable communications. The system must not directly brake a vehicle, stop ventilation or hoisting, or control another safety-critical function without a separate safety case.

Responses may include qualified relief, an added partner or supervision, task rotation, deferral of high-consequence work, suitable recovery, altered surface transport, or safe travel home. A brief pause, caffeine, cold air, or a loud alarm does not prove that risk has resolved. A miner who believes they cannot work safely needs a protected route that is not overruled by a normal algorithm result. The supervisor must also distinguish possible fatigue from acute illness, gas or heat exposure, medication effects, and other emergencies, using the relevant rescue or medical process rather than allowing the monitoring system to diagnose the cause.

Validation, performance evaluation, and event investigation

Release validation should evaluate availability, latency, false alerts, missed alerts, calibration, repeatability, and actual post-alert action in representative people and tasks. It should include nights, shift boundaries, long commutes, heat and humidity, dust, rough travel, low illumination, communications dead zones, failed nodes, and power loss. Performance must be reported by task consequence; aggregate accuracy must not hide misses during high-consequence work. Wear rate, alert count, or production alone cannot show effectiveness. Evaluation should also determine whether false alerts create alarm disregard, bypass behavior, distraction, or disproportionate impact on a group.

An event or near-miss investigation should reconstruct the actual roster, sleep and recovery opportunity, commute, task, machine and roadway, environmental monitoring, ventilation and lighting, uncertainty in personnel location, self-reports, alerts, communications, human decisions, delivered controls, and software version. A fatigue indication is one item of evidence and cannot replace analysis of equipment, management, and environmental causes. A new mining district, entry layout, vehicle, automation function, illumination system, ventilation plan, communications network, cap lamp, protective equipment, roster, contractor, sensor, or algorithm requires an impact assessment and revalidation where necessary.

Data governance, cybersecurity, and emergency continuity

Location history, video, biometrics, health, roster, and performance information can create a persistent portrait of a person when combined. Collection should be minimized for a stated safety purpose, with defined retention, access, correction, appeal, and deletion. Individual health information should be separated from production supervision, discipline, and pay decisions. Data should not default to work-intensity ranking, identification of union activity, insurance, or commercial profiling. Small crews and rare events can reidentify individuals even in aggregate reports. Miners should understand the collection scope, automated inferences, recipients, and route to human review.

The platform should have assessed interfaces and separation from ventilation, hoisting, vehicle control, personnel tracking, emergency communications, and other operational technology so a fatigue function cannot propagate failure into a life-safety system. Testing should include identity and access, clock drift, malicious or replayed data, offline caches, node power loss, backup power, network partition, software updates, audit, rollback, and supplier exit. When connectivity or monitoring is unavailable, stale status should be explicit and approved manual check-in, self-report, pre-shift review, communications, and field supervision must continue. During evacuation, emergency communications, personnel accounting, and the mine rescue plan take priority; the fatigue platform must not delay them.