In brief

  • The problem: safety alarms that sound too often, or for the wrong reasons, get ignored, muted or bypassed. The NSW Resources Regulator has named nuisance alarms as a challenge for proximity detection and collision avoidance in NSW mines.
  • The requirement: NSW law expects control measures to be fit for purpose, set up and used correctly, and protected from bypassing and interference.
  • The gap: most sites measure whether a system is installed, not whether its alerts are trusted and acted on.
  • The outcome: Flagman AI safety cameras can be tuned zone by zone, with tiered alerts and a time-stamped event log, so a site can prove its alerts mean something and fix the ones that do not.

Week three of a new safety system. The warning beacon at the workshop entry has gone off forty times this shift. Most of the time it was a parked light vehicle at the edge of the zone, or a fitter walking a path everyone agrees is safe. By Thursday, operators have stopped looking up when it flashes.

Nobody decided to ignore the alarm. The alarm taught them to. That is alert fatigue.

The problem: alarms that people learn to ignore

Vehicles feature in most Australian workplace deaths. Safe Work Australia reports 188 worker fatalities in 2024, with at least one vehicle involved in 124 of them (66 per cent). Transport, postal and warehousing recorded 54 fatalities, mining 10 and manufacturing 7. Many sites are responding with proximity detection systems and AI safety cameras.

Regulators have been candid about how that technology performs in practice. In its 2023 discussion paper on vehicle interaction controls in NSW mines, the NSW Resources Regulator recorded that an industry collision avoidance forum listed “nuisance alarms” and inadequate human factors design in operator interfaces among the challenges. The paper also noted that some mines had installed technology “with limited success”, and that proximity detection and collision avoidance technology in NSW mining was not yet reliably able to tell hazardous conditions from spurious ones.

When alarms are not trusted, people find ways around them. In December 2025 the regulator issued Safety Bulletin SB25-07 after repeated cases of mobile crushing and screening equipment running with lanyard emergency stop circuits bypassed. In one case a worker started a machine in travel mode to get around a faulty, hard-to-reset lanyard. Pre-start records said the lanyards worked. Only a live test showed they did not.

Attention is finite. In Safety Alert SA25-01 (May 2025), a front-end loader came within 10 to 15 metres of a waiting light vehicle after its operator was distracted by the two-way radio. Every extra alert competes for that same attention.

The requirement: controls that work as intended

In SB25-07, the NSW Resources Regulator pointed duty holders to three sections of the Work Health and Safety Regulation 2025 (NSW):

  • Section 37: control measures must remain fit for purpose, suitable for the work, and installed, set up and used correctly.
  • Section 205: the person with management or control of plant must, so far as is reasonably practicable, prevent unauthorised alteration of or interference with plant, including bypassed emergency stops and interlocks.
  • Section 206: reasonable steps must be taken so that safety features are used correctly.

For mines, the regulator’s 2022 compliance priorities report on proximity awareness and collision avoidance went further. It told operators to consult the workforce so that systems are effective and do not distract mobile plant operators from their core tasks. Its 2023 discussion paper also questioned whether operators could show the technology had delivered a measurable benefit after installation.

An alarm that is installed but ignored does not meet that standard.

Why existing controls leave a gap

Most safety technology projects end at commissioning. After that, three things drift:

  • Zones stop matching the site. A new laydown area, a moved walkway or a parked light vehicle creates alerts that serve no purpose.
  • Equipment degrades. Dirty lenses, a loose bracket or a faulty sensor produce false alerts or, worse, no alerts.
  • Nobody reviews the data. A system that logs thousands of events but never has them analysed cannot show which alerts are real.

The solution: how Flagman helps operators trust every alert

Flagman is a computer vision system that runs on an edge device on site, with no cloud or internet connection needed. It works with ONVIF IP, USB and GMSL cameras. Four features matter most for alert fatigue.

  • Zones drawn around real hazards. Detection zones are set in a browser-based dashboard, so they can be redrawn when the site layout changes, rather than left to trigger on areas that are no longer hazardous.
  • Tiered alerts. In forklift and mobile plant setups, Flagman can give a warning at about 2.5 metres and a critical alert at about 1.5 metres. Not every detection needs a siren. A tiered response keeps the loudest alarm for the moments that matter.
  • The right output for the right person. Alerts can go to sirens, hooters, displays, relay outputs or a PLC over Modbus or PROFINET. A slow or stop signal can be sent where the machine’s controls allow. A supervisor can receive a dashboard notification instead of the whole floor hearing a siren.
  • An event log you can audit. Every alert is time-stamped with video. That lets a site sort true alerts from false ones, find the zones that generate noise, and fix them.

Predictive maintenance: plant condition drives alarm quality

Alert fatigue is often a maintenance problem in disguise. The lanyard in SB25-07 was bypassed because it was faulty and hard to reset. A camera with a dirty or moved lens, a worn mounting or a failing power supply can generate false alerts until operators stop trusting it. On the plant side, worn brakes or failing reversing cameras change the hazard the alarm is meant to manage.

Condition-based and predictive maintenance catch these issues before they show up as nuisance alarms or bypasses. The real value comes from viewing the data together. Sharktech Global builds Business Intelligence dashboards that bring maintenance, downtime and safety data together, including Flagman events. A site can then see, for example, whether false alerts rise after a camera service is missed, or whether near misses cluster around one machine with an open work order.

Limitations: what Flagman cannot do

  • Flagman does not replace segregation, physical guarding, isolation, traffic management or certified collision avoidance systems on haul trucks. It adds detection and early warning on top of them.
  • It is not a certified safety function and has no published SIL or Performance Level rating. It must not be used in underground coal or explosive atmospheres.
  • It sees only what its cameras see. Dust, rain, glare and low light affect detection, and some false alerts will occur. Tuning reduces them but does not remove them.
  • Cameras in NSW workplaces trigger workplace surveillance obligations, including 14 days’ written notice, visible cameras and signs.

A practical plan to beat alert fatigue

  1. Measure the baseline. For two weeks, count alerts per zone per shift, and classify each as true, false or nuisance.
  2. Consult operators. Ask which alerts they ignore and why. Operators know where the noise is.
  3. Redraw zones. Remove areas that are no longer hazardous and tighten zones around real interaction points.
  4. Tier the response. Reserve sirens and stop signals for critical alerts. Route lower-level alerts to supervisors or displays.
  5. Live test, do not just inspect. Walk a test person into each zone on a set schedule and record the result, as SB25-07 recommends for lanyards.
  6. Add the cameras to maintenance. Put lens cleaning, alignment checks and alert tests on the maintenance schedule.
  7. Review monthly. Track false alert rates and response times, and report them to the safety committee.

If you are still selecting a system, our scorecard for choosing a pedestrian detection system and AI safety camera trial checklist show how to build alert quality into the trial from day one.

The outcome you can measure

  • A documented false and nuisance alert rate per zone, falling over time as zones are tuned.
  • Recorded response to critical alerts, with time-stamped video for each event.
  • Live test records that show the control works, not just that it is installed.
  • Evidence for boards and regulators that the technology delivers a measurable benefit.

Flagman in Australia: local delivery by Sharktech Global

Sharktech Global, based in Liverpool, New South Wales, is the exclusive Australian and New Zealand delivery partner for Perleybrook Labs, which designs and manufactures Flagman. Sharktech Global assesses sites, installs and integrates Flagman with existing cameras, alarms and plant controls, and provides local support during business hours. Learn more about Sharktech Global’s services for mining and resources operations.

Key takeaways

  • The NSW Resources Regulator has named nuisance alarms as a challenge for collision avoidance and proximity detection technology.
  • NSW law requires controls to be fit for purpose, used correctly and protected from bypassing.
  • Flagman’s zones, tiered alerts and event logs let a site tune alerts and prove they work, alongside, never instead of, physical controls.

Next step

Request a Flagman safety assessment or call Sharktech Global on 0468 017 373. The assessment reviews your alarms and camera coverage and shows where alerts can be made more trusted.

Sources: NSW Resources Regulator, Discussion paper: Vehicle interaction controls in NSW mines (July 2023); NSW Resources Regulator, Safety Bulletin SB25-07 (December 2025); NSW Resources Regulator, Safety Alert SA25-01 (May 2025); NSW Resources Regulator, Compliance priorities report: proximity awareness and collision avoidance (2022); Safe Work Australia, Key Work Health and Safety Statistics Australia 2025. This article is general information, not legal advice. Seek advice from a qualified professional for your site.