Artificial intelligence in construction health & safety – from compliance monitoring to hazard prediction

Traditional challenges in construction safety management

Limitations of conventional H&S control methods

The traditional approach to health and safety on construction sites rests on a few basic pillars, which today are proving insufficient:

• Manual inspections – the site manager or H&S specialist physically walks the site, observing workers
• Reactive reporting – hazards and accidents are documented only after the fact, once they have already happened
• Subjective assessments – whether a situation is dangerous depends solely on the inspector’s experience
• Delayed analysis – the accident report goes into an archive instead of being analysed alongside similar incidents
• Human fatigue – an H&S officer cannot be everywhere at once, and their alertness drops over time

The impact of fatigue on worker safety

It should also be stressed that research shows 40% of construction workers suffer from fatigue and stress, which significantly increases the risk of accidents. Consequently, traditional H&S control methods – spot-check inspections and self-reporting – are ineffective in an environment where hazards are dynamic and ever-present.

Low level of digitalisation in the construction industry

McKinsey shows that construction is one of the least digitised industries – on average only 14% of workers have access to advanced technological tools that support safety. Meanwhile, artificial intelligence in health & safety can transform the approach to safety management – from reactive to proactive, from subjective to data-driven, from delayed to real-time.

Computer vision for monitoring compliance with safety regulations

Automatic detection of missing personal protective equipment (PPE)

One of the most serious problems on construction sites is failure to use personal protective equipment (PPE). A worker “forgets” their helmet, gloves or high-visibility vest – which can be the result of fatigue, haste or simple reluctance.

Health & safety software equipped with computer vision systems can monitor the site 24/7, looking for workers without the required protective equipment:

• Helmet detection – AI analyses camera images and identifies workers without a helmet
• Glove and boot detection – the system recognises whether a worker is wearing the right protective gloves and boots
• High-visibility vest monitoring – AI checks whether a worker is wearing a visible vest in hazard zones
• Comprehensive PPE analysis – the system can verify whether a person has all the required items for a given work zone

Real-time alerts and immediate intervention

When a health & safety department app detects a worker without PPE, the system takes immediate action:

• Sends an alert to the supervisor or the worker (via a wearable device – smartwatch, phone)
• Blocks access to the hazard zone – where possible, the system controls gates or airlocks
• Documents the incident – records a photo, time, worker and location
• Analyses trends – if the same worker regularly forgets PPE, the system flags it as a training issue

Computer vision for detecting behavioural hazards

Beyond PPE control, health & safety software can monitor unsafe worker behaviour:

• Workers close to machinery – AI raises an alarm when a person gets too close to moving machine parts
• Falls from height – detecting a worker who moves too close to the edge of a platform or roof
• Incorrect equipment operation – an excavator operator manoeuvres the machine in a dangerous way
• Forklift driving – AI monitors whether a forklift driver is going too fast or getting too close to workers

Cority (a company specialising in safety) shows that computer vision can recognise dangerous body positions – such as a low stance when lifting heavy loads – and send immediate ergonomic warnings.

Monitoring worker health – wearable devices

Wearable devices with biometric sensors

In addition to observing behaviour, artificial intelligence in health & safety can directly monitor workers’ health through wearable equipment. Modern smartwatches and specialised bands can be fitted with sensors that monitor:

• Heart rate – detecting excessive exertion or stress
• Body temperature – early detection of overheating or hypothermia
• Blood oxygen saturation – detecting fatigue or lack of oxygen
• Fatigue level – through analysis of movements and biometric signals
• Falls – an accelerometer detects a sudden fall

A practical example

A construction worker is working on a roof in full summer sun. Their fitness band monitors body temperature. When the temperature reaches 38.5°C (the safety threshold during physical exertion), the H&S app sends an alert: “Elevated body temperature. Take a break in the shade.” Alternatively, the system can broadcast a message and notify the site manager.

Worker location and geofencing

Smart devices can track the location of every worker (via GPS or Bluetooth) and set virtual boundaries (geofencing):

• Hazard zones – an alert when an unauthorised worker enters a roof, a shaft or a zone with hazardous gas
• Emergency evacuation – the system knows exactly where each worker is and can evacuate them quickly
• Working time control – automatic reporting of time spent in different work zones

Predictive analytics for accident prevention

Predictive models based on historical data

Instead of waiting for an accident to happen, artificial intelligence in health & safety can predict the risk and suggest interventions. Machine learning analyses historical accident data and identifies patterns:

• High-risk hours – data shows that accidents are 3x more frequent towards the end of a shift (fatigue)
• High-risk days – Mondays and Thursdays record more accidents
• Weather conditions – rain increases the risk of falls by 40%
• High-risk types of work – work at height, excavation, machine operation
• High-risk workers – new workers, workers returning from sick leave

Automatic allocation of H&S resources

On this basis, a health & safety department app can recommend:

• Increased supervision – assign an additional H&S observer on Friday afternoon
• Worker rotation – worker fatigue is observed towards the end of a shift, so they should be moved to other tasks
• Postponing work – if the forecast indicates an 80% risk for work at height in the rain, it is better to postpone such work to the next day
• Additional training – a new worker or one returning from leave requires special supervision

Natural language processing for analysing incident reports

Automating report processing

Every construction site produces near-miss and incident reports. Traditionally, the H&S manager would have to read each report manually, try to categorise it and draw conclusions. Meanwhile, health & safety software using AI can automate this process:

• Report categorisation – is it an ergonomic hazard, a fall from height or contact with machinery?
• Extraction of key data – which worker, what time, which zone, what conditions
• Root cause identification – the system analyses the worker’s natural-language description and tries to find the cause
• Comparison with history – the system checks whether the same type of problem has occurred before

A practical example of report processing

Autonomous warning systems – workers approaching machinery

Proximity detection systems

One of the most dangerous scenarios on a construction site is a worker getting too close to moving machinery (excavators, pneumatic hammers, concrete mixers). For this reason, H&S automation combined with sensors can:

• Monitor distance – radar/LiDAR tracks all objects within a range of several metres
• Identify people – AI recognises whether it is a worker or material
• Calculate trajectory – is the person heading towards the machine?
• Send alerts – graduated warnings: yellow (approaching), orange (close), red (switch off the machine immediately!)

In the latest systems, if a worker does not respond to the alerts, the machine switches itself off. This can be controversial (the worker may not have heard the alarm or may have unknowingly stood in the wrong place), but it is the last line of defence.

Sources and bibliography