Machine Guards Get Bypassed: How Flagman AI Vision Protects the Hands That Run the Line

In brief

• Australian work health and safety law sets a clear order for guarding machinery: permanently fixed guards first, then interlocked guards, then guards removable only with a tool, and presence sensing systems only where physical barriers are not practicable.
• Hands and fingers are the body parts most exposed on production lines, particularly around saws, rollers, presses and conveyors.
• AI vision can create virtual zones around machines, warn operators and, through relays or programmable logic controllers, send stop or hold signals. It is a valuable extra layer of protection and evidence.
• AI vision is not a certified safeguarding device. It should strengthen compliant guarding, not substitute for it, unless a competent engineer has assessed it as a safety function.

Permanently fixed guards come first. Presence sensing systems are the last option, used only where physical barriers are not reasonably practicable.

Code of Practice: Managing the risks of plant in the workplace

Walk through almost any Australian factory and you will find a machine where the guard has been modified, propped open or removed “just for a minute” to clear a jam. That minute is when many of the worst manufacturing injuries happen. Machine guarding is one of the oldest safety disciplines in industry, yet it remains one of the most common findings in regulator inspections.

This article explains what the law requires, why guarding still fails in practice, and how AI vision can help manufacturers protect the line and the hands that run it.

The guarding hierarchy set by law

Regulation 208 of the model Work Health and Safety Regulations, which applies in most states and territories, sets out the order in which guarding must be chosen. The approved Code of Practice for managing the risks of plant in the workplace explains it in practical terms. Each option is used only if the one before it is not reasonably practicable:

1. A permanently fixed barrier, where access to the danger area is not needed during operation, maintenance or cleaning.
2. An interlocked physical barrier, where access is needed. Opening the guard stops the hazardous motion.
3. A physical barrier that can only be altered or removed using a tool.
4. A presence sensing safeguarding system, such as a light curtain, laser scanner or pressure mat, that detects a person entering the danger area and stops the machine.

The code also says plant should usually be isolated before maintenance or cleaning starts, and describes lock-out as the most effective isolation procedure. Regulation 226 requires records of safety integrity tests, inspections and maintenance for plant fitted with presence sensing systems.

Victoria applies equivalent duties under its Occupational Health and Safety Act 2004 and Regulations 2017. Across the country, the technical detail is set out in the AS/NZS 4024 series of machinery safety standards.

Why guarding still fails

Guards rarely fail because nobody installed them. They fail because production pressure creates reasons to work around them. Common patterns include:

• interlocks bypassed to keep a line running while clearing jams or making adjustments
• guards removed for maintenance and not refitted correctly
• new or modified machines commissioned before guarding is complete
• operators reaching around or under guards on older equipment where gaps have opened over time
• cleaning carried out while machinery is still energised

Each of these is a behaviour, not a design failure, which is why they are so hard to catch. Supervisors cannot watch every machine on every shift.

The solution: how Flagman AI vision watches the danger zone

AI vision adds continuous observation to a guarding system. Cameras watch the area around a machine, and software running on a local edge device recognises when a person, a hand or an arm enters a defined virtual zone.

Platforms such as Flagman can then respond in several ways. They can trigger local alarms and visual warnings. Through relays or programmable logic controllers, they can send a stop or hold signal to the machine and allow restart only when the zone is clear. Flagman’s hand and finger detection creates a “no hand zone” around rollers and saws, and its machine guarding module can restrict access to authorised operators. Every event is logged with time-stamped video, which gives safety teams evidence of where and when guarding is being challenged.

That evidence is often the most valuable part. If one press records repeated hand intrusions during changeovers, the solution may be a redesigned guard, a better jig or a revised changeover procedure. AI vision shows managers where to act.

Know the limits before you rely on it

There is an important distinction between AI vision and a certified presence sensing safeguarding device. Light curtains and safety laser scanners used as safeguarding are designed and rated to functional safety standards so that their response time and reliability are known. A stop signal from any system into a machine’s controls is a safety function, and under AS/NZS 4024 it needs a defined performance level justified by a competent engineer.

In practical terms:

• Use AI vision to strengthen compliant fixed or interlocked guarding, not to replace it.
• Do not use AI vision as the only safeguard on high-risk machinery unless it has been assessed as a safety function for that application.
• Remember that cameras need a clear view. Steam, dust, glare and blocked lenses can reduce detection.
• Never let any monitoring system become a reason to work on energised plant. Isolation and lock-out still apply.

A practical guarding review for your site

1. Inventory your machines. List every machine with moving parts and record the guarding type currently in place.
2. Check against the hierarchy. For each machine, ask whether a higher-order guard is reasonably practicable.
3. Look for bypass evidence. Interview operators and maintenance staff about where and why guards are opened during production.
4. Prioritise by consequence. Focus first on saws, presses, rollers, conveyors and any machine with a history of hand injuries.
5. Add monitoring where behaviour is the risk. Use AI vision on machines where guards are regularly challenged, to warn, record and reveal patterns.
6. Review monthly. Use the data to improve guard design, procedures and training.

The outcome you can measure

For manufacturers, the outcome is visibility of a risk that is otherwise invisible. Flagman shows which machines have hand or body intrusions, when they happen and during which tasks, while warning operators in real time and, where engineered and assessed, holding the machine. Production and safety teams can then fix the root cause, whether a guard redesign, a better changeover method or retraining, and use the event data to confirm the fix has worked.

Key takeaways

Australian law is clear about how machinery must be guarded, and physical guarding remains the foundation. AI vision earns its place by watching what supervisors cannot, warning operators in real time and producing the evidence needed to fix guarding problems at their source. Used within the guarding hierarchy, it helps manufacturers protect both production and the people who run it.

Next step

Want to understand where guarding is being challenged on your line? Request a Flagman safety assessment. In Australia, Flagman is delivered and supported locally by Sharktech Global.

Sources: SafeWork NSW, Code of Practice: Managing the risks of plant in the workplace (2022); model Work Health and Safety Regulations 2011, regulations 208, 210 and 226; PACE Today, changes to AS 4024.1. This article is general information, not legal or engineering advice for your site.