HyperQ AI Safety connects to existing CCTV in under 1 hour, applying continuous zone and intrusion detection to the camera feeds already pointed at production equipment. On a guarded machine, that means the area inside or around the guard becomes a monitored zone — any breach is timestamped, categorised, and routed to an alert within seconds, not discovered at the next audit.
Most manufacturing plants have good guards on their machines. The ISO 45001 audit showed them compliant. The DOSH inspection last year found no violations. What neither the audit nor the inspection captured is what happens at 02:30 on a Thursday night shift when a maintenance technician tapes a magnet over the interlock to stop the door sensor from killing the machine mid-run, finishes the job, and forgets to remove it.
The interlock is intact. The guard is physically present. The compliance record shows a compliant plant. And the guard has been effectively disabled for 11 weeks.
This is the structural gap in guarding compliance: audits and inspections are point-in-time observations of a continuous-state property. A guard is compliant or non-compliant at every moment of every shift, not only during the observation window. The audit verifies the guard exists. It does not verify the guard functions for the 8,760 hours between one audit and the next.
How bypass happens
Guard bypass is not primarily a rogue-behaviour problem. Most bypass events in manufacturing are improvised workarounds by workers responding to a real operational constraint, not deliberate safety circumvention. The patterns are consistent:
A machine requires adjustment during a run. The normal adjustment procedure requires the guard to be open, which kills the machine, which stops the line, which triggers a delay notification to the production supervisor. The experienced operator knows what needs adjusting and knows they can do it safely with the machine running if they're careful. They prop the guard open with a piece of stock, make the adjustment, and move on. The guard returns to its position. The interlock sensor reads closed. Nothing is logged.
A maintenance technician is troubleshooting an intermittent fault. The fault only manifests under running conditions. The technician needs to observe the mechanism in motion to isolate the cause. The safe procedure requires a second person, a formal permit, and a supervisor sign-off — a 45-minute process for a 3-minute observation. The technician uses a bypass jumper from the toolbox, which the team has used for years for this exact purpose. The fault is fixed. The jumper is removed. No record exists.
A guard panel is removed for a planned maintenance task and reinstalled with one of the four fasteners missing. The panel seats flush, the interlock closes, the machine runs. The panel looks compliant. It will shift under vibration within two weeks of normal operation.
In each case, the guard bypass was transient, motivated by a real operational need, and invisible to any audit or inspection not conducted at the specific moment it occurred. The bypass that causes an injury is statistically the one that overlapped with an unexpected machine event (a jam, a misfeed, a maintenance-triggered fault mode) that the bypassing worker didn't anticipate.
What continuous monitoring provides
Continuous zone monitoring around guarded machines converts guarding compliance from a point-in-time observation to a time-series record. When HyperQ AI Safety is configured with a monitored zone inside or adjacent to a guarded machine area, any human presence detected in that zone while the machine is in an active or maintenance state generates a timestamped event with a clip.
This does not replace the interlock or the guard. The engineering control hierarchy under ISO 45001 and local WSH regulations places physical guards and interlocks above administrative and monitoring controls. Continuous AI monitoring is a verification layer, not an alternative to hardware controls.
What it provides that hardware controls cannot: a documented record of when the guarded zone was accessed, under what conditions, and by whom — continuously, not just at audit dates.
The detection logic for a guarded machine zone is straightforward. During normal production (machine running), any person detected inside the guard perimeter is a breach event — the guard is supposed to prevent this. During scheduled maintenance (machine in LOTO), a person inside the guard perimeter is expected and should not trigger an alert; the zone monitoring is suspended for the LOTO duration in the system. When LOTO ends and the machine returns to production state, the zone monitoring reactivates.
Events that fall outside these two states are the highest-value entries for a continuous monitoring record: a person detected in the guard zone while the machine is transitioning from maintenance to production, or a person detected after LOTO sign-off but before formal production restart. They capture the gap between paperwork and physical state.
Guard bypass detection: what the event record shows
The following is an illustrative scenario, not a documented client deployment. It reflects the event pattern the system would capture.
A press line has two guarded zones: the die area (hard guard with interlock) and the feed ramp (light curtain). HyperQ AI Safety monitors both zones on the two cameras already covering the press. The event log for a 30-day period shows:
| Event type | Count | Time pattern | Finding |
|---|---|---|---|
| Person in die area — production state | 3 | 23:00–01:00, night shift | Consistent: same camera angle, consistent silhouette. Investigated: technician adjusting feed stock with guard propped. Guard prop removed, procedure retrained. |
| Person in die area — LOTO end gap | 1 | 06:45 | Machine transition from LOTO to production. Operator re-entered before formal restart confirmed. Procedure gap identified. |
| Light curtain zone — multiple occupancy | 7 | 14:00–16:00, afternoon shift | Two workers feeding material simultaneously during peak output. Light curtain is single-person by design. Retrained. |
| No events | — | 06:00–22:00 day shift | Normal production. No guard zone breaches. |
The 3 night-shift die-area events and the LOTO-end event are findings that a quarterly audit conducted during day-shift hours would not see. The light-curtain double-occupancy events are a systematic finding across multiple workers on one shift — a procedure gap, not an individual behaviour issue.
A 30-day event record of this specificity is not producible by walk-around audit. It is the difference between a compliance snapshot and a compliance record.
Audit trail for ISO 45001, WSH, and DOSH
The export from a 12-month continuous monitoring record is an audit evidence file with timestamps, event categories, alert routing logs, and clip references for every guarding-zone event across the year. For ISO 45001 surveillance audits, this addresses two clauses directly: Clause 8.1.1 (operational planning and control — demonstrate that controls for hazardous processes are active, not just documented) and Clause 9.1.1 (monitoring and measurement — demonstrate that the organisation monitors the performance of its operational controls at defined intervals). A 12-month timestamped event record satisfies both better than an annual inspection sign-off because it shows the control was operating continuously, not just verified on the audit date.
For MOM WSH Inspectorate site visits in Singapore, the audit trail demonstrates that the guarding compliance record extends beyond the physical guard inspection. The record shows that zone access is monitored in real time, that breaches generate alerts to the responsible person, and that corrective actions are documented.
For DOSH inspections in Malaysia under the OSH Act 1994 and the Factories and Machinery Act, a documented continuous monitoring record for guarded equipment is a substantive addition to the guarding compliance file. DOSH inspection citations for guarding non-compliance often stem from evidence of bypass behaviour discovered after an incident — a continuous monitoring record provides the counter-evidence that the organisation was actively detecting and responding to bypass events before any incident occurred.
The role of smartband monitoring for maintenance technicians
For the specific gap period most relevant to guard bypass — planned and unplanned maintenance access — wearable monitoring for the maintenance technician supplements camera zone monitoring. HyperQ AI Safety supports smartbands from $35–250 per unit, monitoring heart rate, SpO2, and skin temperature.
During LOTO work on a machine with known heat or confined-space exposure (electrical panel interiors, enclosed drive systems), physiological monitoring records whether the technician's vitals were normal throughout the access period. For the scenario where a technician bypasses a guard to troubleshoot under live conditions — which involves elevated stress — heart-rate elevation during the event window is logged alongside the zone breach record. The combined record (zone event + vitals log) provides more complete evidence for an incident investigation than camera footage alone.
For the paper-audit compliance argument, the combined record also addresses the question of whether the person present during a guard zone breach was working calmly through a procedure or responding to a fault under stress — a distinction that affects how the incident review characterises the event and whether it triggers a procedure update or a culture intervention.
Zone configuration and calibration
Setting up a guarded machine zone in HyperQ AI Safety requires three inputs: the camera covering the guard area, the zone boundary drawn around the guard perimeter, and the production/LOTO state integration. The production state is communicated either through a direct signal (integration with the machine's PLC output or safety relay) or through a manual schedule in the system (machine in production 06:00–22:00, LOTO window 22:00–06:00). The manual schedule is less precise but functional for lines with fixed shift patterns.
Zone sensitivity should be calibrated for the machine's typical proximity pattern. Workers on adjacent lines or maintenance technicians walking past the guard perimeter (not entering it) should not generate events. The first week of operation typically requires 2–3 sensitivity adjustments. After calibration, the false-positive rate for a guarded machine zone is low relative to a high-traffic zone like a loading dock, because the expected occupancy pattern is tighter.
Where this approach has limits
AI zone monitoring on CCTV has a line-of-sight requirement. Guards that are fully enclosed (tunnel guards, machine enclosures with no camera coverage of the interior) are not verifiable by camera — the zone monitoring covers the entry point, not the interior. For enclosed guards, the interlock remains the only continuous verification control; camera monitoring covers access patterns but not conditions inside.
Physical guard condition (a missing fastener, a cracked guard panel, a misaligned interlock) is not detectable by camera zone monitoring. Camera monitoring detects human access events, not mechanical guard integrity. Physical inspection at a defined interval (weekly check of guard fasteners, quarterly interlock function test) remains necessary alongside continuous access monitoring.
The LOTO state integration requires accurate input. If the maintenance team signs off LOTO on the permit but does not update the system state (manual schedule or PLC integration), the monitoring will generate false events during legitimate LOTO work. EHS officer calibration and consistent LOTO state entry discipline is required for the event record to be reliable.
What the continuous record adds to the ISO 45001 corrective action cycle
ISO 45001 Clause 10.2 requires that the organisation investigate incidents and near-misses, determine root causes, and implement corrective actions. The standard also requires that the corrective action's effectiveness be verified — not just documented.
A 30-day event record showing bypass behaviour in a specific zone, followed by a corrective action (retraining, physical barrier addition, procedure update), followed by a second 30-day event record showing reduced or eliminated bypass events in that zone, is a documented Clause 10.2 effectiveness verification. The event record before and after the corrective action is the evidence that the intervention worked — or that it didn't, which is equally important for the next cycle.
This before-and-after structure is difficult to produce from walk-around audits because the audit frequency is too low to show trend change. It is straightforward to produce from a continuous monitoring record because the baseline and the post-intervention data are both generated automatically from the same system.
For organisations pursuing ISO 45001 certification (rather than maintaining it), auditors reviewing Clause 10.2 implementation will look for documented examples of the corrective action cycle operating. A before-and-after guard zone event record is one of the clearest demonstrations available.
Guarding compliance as a continuous property
An interlock that passes the annual inspection and then gets bypassed in week 3 has provided 11 weeks of false compliance. The audit record is clean. The compliance record is not.
Continuous zone monitoring does not change the guard hardware, the interlock requirement, or the LOTO procedure. It adds a persistent record of whether the guard's intended function — preventing access to the danger zone during production — was actually maintained between the audit dates.
That record is what changes the compliance question from "did the guard pass inspection?" to "was the guard effectively in service during the year?" The second question is the one that matters when the injury investigation starts.
ISO 45001 and AI safety monitoring: the compliance relationship · Singapore WSH Act AI safety monitoring context · HyperQ AI Safety
Send us your guard layout — machine count, camera coverage map, and shift pattern — and we'll return a zone-monitoring configuration plan with suggested LOTO integration method, sensitivity thresholds, and a 30-day event record template formatted for your ISO 45001 or WSH audit file — no contract until you've reviewed the configuration against your actual machines. Send us your guard layout