LOTO and machine guarding compliance with AI video

What LOTO and machine guarding compliance covers

Machine guarding keeps people away from the moving parts of a machine while it runs: fixed guards, guard doors with interlocks, light curtains, two-hand controls. Lockout/tagout (LOTO) covers the times when someone has to go past those guards — to clear a jam, clean, adjust, repair or change a tool. Before they do, the machine is stopped, its energy sources are isolated, each isolation point is locked and tagged, stored energy is released and the zero-energy state is verified. Compliance is about both holding in everyday work. It breaks down in six recurring situations.

Guards opened or bypassed

A guard that slows the work tends to get opened, propped up or defeated: an interlock switch taped down, a spare actuator key left in place, a fixed panel not put back after maintenance. The machine keeps running as if nothing changed, and the guard that is supposed to stop it no longer does.

Reaching into a running machine

A carton jams on the conveyor, a part sits crooked in a fixture, a label roll runs out. The fastest fix is to reach in and sort it out without stopping the line. When it works, it becomes a habit — until the one time the machine moves while the hand is inside.

Maintenance and cleaning on energized equipment

Maintenance and cleaning are the tasks that need people inside the machine, and they can end up being done in a hurry: between batches, at product changeover, in a short maintenance window. Cleaning a mixer, a slicer or a conveyor with the drive only switched off at the control panel — not isolated — leaves the machine one button away from starting.

LOTO steps skipped

A LOTO procedure is a chain of steps: stop the machine, isolate every energy source (electrical, pneumatic, hydraulic, gravity, stored), apply a personal lock and a tag at each isolation point, release stored energy, and verify that the machine cannot start. Each step is easy to shorten: one isolation point is missed, the lock stays in the locker "because it will only take a minute", the verification try-start is skipped. A skipped step leaves no visible change to the machine until someone is inside it.

Shift-close pressure

The end of a shift adds its own pressure: the line has to be cleared, the area cleaned, the handover done, and everyone wants to leave on time. Shortcuts that would not be taken mid-shift become tempting in the last minutes — and the next shift inherits whatever state the machine was left in.

Contractors and non-routine work

Contractor technicians, installers and cleaning crews may not know the site's machines, isolation points or LOTO procedure, and site staff may assume someone else has isolated the equipment. Non-routine jobs — a breakdown at night, an unusual repair — are where procedures are less practiced and easier to improvise.

Why traditional methods miss LOTO and guarding violations

Safety rules for work equipment set a clear baseline: guards must not be easy to remove or make ineffective; maintenance is to be done only with the equipment switched off, its energy connection cut and all movement stopped, or, where that is not possible, with the necessary precautions taken or outside the danger zone; and the means of cutting off energy sources must be easily visible and recognizable. A LOTO procedure puts these requirements into practice at the machine. The gap is not the rule — it is seeing whether the rule holds in the few seconds when someone reaches in.

• Interlocks can be bypassed. An interlock stops the machine when the guard door opens — unless the switch has been defeated. A taped switch or a spare key looks the same to the machine as a closed door, and nothing records that the protection was off.
• Permits and LOTO records live on paper. A signed permit or a lock log shows that isolation was planned and signed off. It cannot show what happened at the machine: whether every energy source was isolated, or whether someone reached in before the work was signed off at all.
• Periodic audits see a moment, not the shift. A safety walk observes a line for a short time, and people work by the book when they know they are being watched. The jam cleared by hand in the last minutes of the evening shift is not on the audit route.
• Manual CCTV review doesn't scale. The cameras may already be pointed at the lines. But checking them means knowing which camera and which minute to watch, so footage is reviewed after an injury — not to find the interventions that came before it.
• Rule-based detectors don't know intent. A light curtain or a camera zone trigger reacts when something crosses a plane or enters an area. It can't tell an operator feeding a part through a designed opening from one reaching past the guard to clear a jam, and it has nothing to say about a guard that was simply left open. Light curtains are safety devices and stay in place; as a way of seeing violations they produce either a stream of alarms for normal work or silence for the situation nobody drew a zone for.

What these methods lack is the reading a safety professional makes at a glance: the guard is open, the machine is still running, and that person is about to put a hand inside.

How vision-language models make it solvable

Primarch builds on and enhances vision-language models (VLMs): models that read video and language together. That changes what can be checked on a production floor. Instead of drawing boxes around objects frame by frame, the model reads the scene the way an experienced safety professional would — and does it on every camera, on every shift.

• It reads the action, not just the object. An open guard is an object state; an operator about to reach into a machine that is still energized is an action with intent. The model reads the action — and that is the moment that matters.
• It reads movement over time, not a single frame. A VLM reads a window of time, so the before and after of a movement are read together: the line still running, the guard swung open, the hand moving toward the infeed.
• It reads guards in context. An open barrier or a disabled guard means something different on a machine whose parts have stopped than on one that is still running with an operator beside it. Context is what separates normal work around a machine from a reach into moving parts.
• It names the violation and writes it in plain language, with a clip. The finding is a sentence a supervisor can read — what happened, where, and which rule it breaks — with a time-stamped evidence clip, so everyone discussing it sees the same thing.
• It chains events by shift, zone and time. Single events are flagged and alerted for review; chained together, they show where and when the same intervention repeats, and help diagnose why: an attention lapse, or a process gap.

A safety professional at the line doesn't stop at whether the guard is open. They look at whether the parts behind it are still moving and whether a hand is heading there — and that is what the seconds of video before the hand goes in can show.

Coverage depends on the cameras. An intervention can be read only where a camera sees the machine's access points and the person working at them. Video reads what is visible; it does not replace the verification step of the LOTO procedure, where the person doing the work confirms the machine cannot start. From video, 'still energized' is read from what the camera shows — parts still moving, a guard open on a running line; a machine stopped at the panel but not isolated can look the same as an isolated one.

The same chaining by shift, zone and time applies to vehicle–pedestrian risk — see Forklift and pedestrian safety with AI video.

What to look for in a solution

Whichever vendor you talk to, these are fair questions to ask about a system meant to improve LOTO and machine guarding compliance with video:

• Works on your existing cameras. Can it connect to the IP cameras already covering your lines, or does every machine need new hardware?
• Reads the intervention, not just the guard. Does it understand that a person is about to reach into a running or energized machine, or does it only report a door open or a zone entered?
• Is clear about what video can and can't verify. Ask which LOTO steps a camera can actually show at your site, and which still rest on the procedure and the person doing the work.
• Handles false alarms openly. Ask how it separates normal operation — loading a part, a planned intervention on an isolated machine — from a violation, how findings are reviewed by a person, and how single events are separated from repeating patterns.
• Explains repeats. Can events be grouped by shift, zone and time, and does it help separate an attention lapse from a process gap such as shift-close pressure?
• Produces records your safety team can use. Findings with the regulatory reference, corrective actions with owners, and zone-level summaries for the safety committee and audits.
• Respects privacy. Is footage anonymized? Is there an on-premise option where footage never leaves the site? Does it meet GDPR and other data-protection requirements, and does it report situations rather than individuals?
• Can be piloted on your own floor. One zone, a few of your cameras, your own machines — not a demo video.
• Gets alerts to the right role. An instant alert for the shift supervisor, a digest for the safety and maintenance teams, a dashboard for plant management.

Example use cases

The walk-throughs below are illustrative examples of how a finding is built. They are not customer cases and contain no performance figures.

Example 1: clearing a jam at shift close

1. 01

   Moment

   A packaging line, 23:42, in the last minutes of the evening shift. A camera covers the case erector and its infeed conveyor.

2. 02

   What the camera shows

   A carton jams at the infeed. The operator opens the access guard, leaves the conveyor running and reaches in to free the carton.

3. 03

   The finding + evidence clip

   "LOTO violation at the case erector: the access guard is open and the operator is reaching into the infeed while the conveyor is still running." The clip covers the seconds before and after; earlier repeats of the same scene at this station are recalled from memory.

4. 04

   Who is alerted

   The shift supervisor receives an instant notification; the event goes into the safety team's weekly digest.

5. 05

   What the shift and zone pattern shows

   Similar interventions at this station fall in the last part of the evening shift, while the line is being cleared for close.

6. 06

   Corrective action

   Diagnosis: not an attention lapse but a process gap at shift close. Recommendation for the production manager: a close-out checklist that puts isolation before any guard is opened, and an interlock on the access guard.

Example 2: cleaning a mixer that is still turning

1. 01

   Moment

   A food plant, 02:15, during the night sanitation shift. A camera covers the mixer platform.

2. 02

   What the camera shows

   A sanitation worker opens the mixer's guard and leans in with a hose and brush while the agitator is still turning.

3. 03

   The finding + evidence clip

   "Intervention on a still-energized mixer during cleaning: the guard is open and the worker is leaning into the bowl while the agitator is turning." Clip attached.

4. 04

   Who is alerted

   The sanitation supervisor, instantly; the safety team in the weekly digest.

5. 05

   What the shift and zone pattern shows

   Flagged events on the mixer platform fall at the start of the sanitation shift, when cleaning begins right after production stops.

6. 06

   Corrective action

   Diagnosis: a process gap — the cleaning instruction starts with the guard, not with isolation. Recommendation: rewrite the sanitation procedure so isolation and verification come first, assigned to the sanitation supervisor and the maintenance manager.

Example 3: a guard door that no longer stops the machine

1. 01

   Moment

   A machining cell, 14:10, during a setup. A camera covers the front of the machine and its guard door.

2. 02

   What the camera shows

   The guard door stands open while the spindle keeps turning. The operator reaches in to adjust the workpiece in the fixture.

3. 03

   The finding + evidence clip

   "Guard door open on a running machine — the interlock appears to be defeated; the operator is reaching to the workpiece while the spindle turns." Clip attached.

4. 04

   Who is alerted

   The shift supervisor and the maintenance team, instantly.

5. 05

   What the shift and zone pattern shows

   The same scene repeats at this machine on every shift, and only during setups.

6. 06

   Corrective action

   Diagnosis: a process gap, not individual carelessness — the setup cannot be done comfortably with the door closed. Recommendation for the maintenance manager: restore the interlock and review the setup method and fixture so that adjustment does not require reaching in on a running machine.

Example 4: a maintenance window with an outside crew

1. 01

   Moment

   A distribution center's sortation conveyor, Saturday 09:30, during a scheduled maintenance window. A camera covers the conveyor's drive section.

2. 02

   What the camera shows

   A side panel of the drive section has been removed. A technician works with an arm inside the drive while the belt is still moving.

3. 03

   The finding + evidence clip

   "Intervention on a still-energized conveyor: the drive-section panel is off and the technician is reaching into the drive while the belt is moving." Clip attached.

4. 04

   Who is alerted

   The maintenance manager responsible for the zone, instantly; the event also appears in the zone's weekly score.

5. 05

   What the shift and zone pattern shows

   The safety team asks the event memory for last month's interventions on running equipment in the sortation zone and lays the answer over the maintenance schedule: every entry falls in a window when an outside crew was working.

6. 06

   Corrective action

   A recommendation for the plant manager, with the regulatory reference: include the site's LOTO procedure in contractor induction, and start contractor work on machines only after isolation has been done together with site maintenance.

How Primarch's OHS Expert helps

This section is about our product. The OHS Expert is the Primarch expert module that reads the floor like a safety professional, on the cameras a site already has. For LOTO and machine guarding it:

• Doesn't just see an open barrier: it understands that an operator is about to intervene on a machine that is still energized — and flags that moment.
• Reads open barriers and disabled guards in context, as the action happens.
• Names the violation — LOTO — and writes findings with the regulatory reference; narrates each finding in natural language with a time-stamped evidence clip, and recalls past repeats of the same scene from memory.
• Chains events by shift, zone and time and diagnoses why they repeat — an attention lapse or a process gap.
• Writes corrective actions in the language an inspector would use, assigns owners, and prepares regulation-referenced records, weekly zone safety scores and executive summaries.
• Needs no rule set. It comes pre-configured with OHS domain knowledge and regulation; site-specific thresholds are tuned together at onboarding, and violations nobody wrote a rule for are still caught through scene understanding.
• Checks whether a lock and tag are visible on the isolation point before work on the machine starts, where a camera covers that point.

Because it reads the scene in context, it is designed to remove the biggest source of noise in classic systems — rules firing without context — and multi-stage validation and temporal noise suppression are added on top.

Putting it to work

1. 01

   Connect

   Connect your existing RTSP cameras; no new cameras or special hardware.

2. 02

   Pilot on your own floor

   Pilots typically start with a single zone and a few cameras — one line or a group of machines, say. Thresholds are tuned to your site together, and you review the findings with your safety team.

3. 03

   Roll out

   Scale facility-wide as the system proves out, with dozens of streams watched in parallel.

• Alerts, digests and dashboards. Instant notifications, periodic digests and role-based dashboards, so each finding reaches the person who acts on it.
• Ask the past in plain language. Findings are written into a queryable event memory: "Which guard-open interventions were recorded on line 2 last week?"
• Records you can defend in an audit. Every finding carries its time-stamped clip; reports carry the regulatory reference and the corrective action with its owner.
• Privacy by design. Footage is anonymized in a GDPR-compliant way, and the system is designed to report risky situations and process gaps, not individuals. With on-premise deployment, footage never leaves the facility.