Manufacturing and process industries

Manufacturing safety is mostly about controlling energy: the moving parts of machines, the stored energy released during maintenance, and the mass of forklifts and loads moving through the same space as people. This chapter sets out what regulators' data says about harm in manufacturing, the hazards that drive serious injuries, the rules that apply in the US, UK and EU, and where AI video, wearables and other technology are being used. It ends with the questions a buyer should put to any vendor.

Manufacturing here means discrete manufacturing (automotive, metals, machinery, electronics, packaging) and process industries such as food and beverage, chemicals, paper and cable. The hazards overlap heavily. A meat plant and a stamping plant both have conveyors, forklifts, sharp tooling and repetitive handling, even though their products have nothing in common.

How many workers are killed and injured in manufacturing?

United States

The US Bureau of Labor Statistics (BLS) Census of Fatal Occupational Injuries recorded 5,070 fatal work injuries across all industries in 2024, down from 5,283 in 2023. The figures were released on February 19, 2026, later than usual ^[1]. Manufacturing accounted for 353 of those deaths, a 9.7 percent fall from the previous year ^[2].

Manufacturing Dive's analysis of the same release breaks the manufacturing total down by event. Contact incidents, which BLS defines to include being caught or compressed by running powered equipment and being struck by falling or suspended objects, caused 105 deaths. Transportation incidents, which in a plant setting include forklift and other vehicle events, caused 65 ^[3]. Together those two categories account for roughly half of manufacturing deaths, and both are events a camera can often see developing.

Non-fatal harm is far larger. Employers reported 2.5 million injury and illness cases in private industry in 2024, a rate of 2.3 cases per 100 full-time equivalent workers. BLS described this as the lowest count in the series going back to 2003 ^[4]. Manufacturing remains one of the larger contributors to that total because of its size and its mix of machine, handling and vehicle work.

OSHA's list of most frequently cited standards gives a different view: what inspectors actually find. For fiscal year 2025 (October 2024 to September 2025), four of the ten most cited standards apply directly to factory floors: control of hazardous energy (lockout/tagout) in fourth place, respiratory protection fifth, powered industrial trucks eighth and machine guarding tenth. Hazard communication, which covers chemical labeling and safety data sheets, was second ^[5].

Great Britain

The Health and Safety Executive (HSE) recorded 126 worker deaths in 2025/26 (provisional figures). Manufacturing accounted for 18 of them, third behind construction and agriculture, forestry and fishing. Across all sectors, 10 workers died from contact with moving machinery and 21 from being struck by a moving object ^[6].

For non-fatal outcomes, HSE puts manufacturing's self-reported non-fatal injury rate at 2,110 per 100,000 workers, above the all-industry estimate of 1,780, and its work-related ill health rate at 3,670 per 100,000, below the all-industry 4,170. HSE does not list manufacturing among the sectors with statistically significant differences on either measure ^[7]. Nationally, 511,000 workers were suffering from a work-related musculoskeletal disorder in 2024/25, out of 1.9 million with any work-related illness ^[8]. Lifting, repetitive assembly and awkward postures make manufacturing one of the sources of those cases.

European Union

Eurostat's 2024 data, published in September 2026, records 3,367 fatal accidents at work in the EU, 45 more than in 2023. Manufacturing accounted for 13.8 percent of fatal accidents and 18.3 percent of the 2.78 million non-fatal accidents that caused at least four days' absence. That makes manufacturing the largest single sector for non-fatal accidents in the EU ^[9].

What the numbers mean for technology buyers

Three points stand out across the three jurisdictions:

What are the main hazards in manufacturing?

Machine guarding and amputations

OSHA's guidance states that moving machine parts can cause "crushed fingers or hands, amputations, burns, or blindness" and that any machine part, function or process that may cause injury must be safeguarded ^[10]. The hazard points are familiar to any production engineer: points of operation on presses and saws, in-running nip points on rollers and conveyors, rotating shafts, and reciprocating parts.

Guarding failures tend to follow a pattern. A guard is removed for a changeover and not replaced, an interlock is bypassed because it trips too often, or an operator reaches past a guard to clear a jam while the machine is still running. Manufacturing Dive's coverage of the 2024 figures gives one example: a supervisor at a GE Appliances plant in Alabama died after being caught in machinery, and OSHA found the company had allowed workers to bypass safety doors. The agency proposed $193,585 in penalties ^[3].

OSHA ran a National Emphasis Program on amputations in manufacturing industries from December 2019 (CPL 03-00-022). The directive is now archived ^[13], but the targeting logic it used, focusing on industries with high amputation rates, still shapes how inspectors approach presses, conveyors and food slicing equipment.

Lockout/tagout and hazardous energy

Lockout/tagout (LOTO) is the procedure for isolating and securing all energy sources before someone services or maintains equipment. It covers electrical, hydraulic, pneumatic, mechanical, thermal and gravitational energy, and stored energy in springs, capacitors and raised loads. OSHA notes that failure to control hazardous energy during maintenance can cause serious or fatal injuries, and that millions of workers service equipment routinely ^[11]. The general industry standard is 29 CFR 1910.147, which ranked fourth among OSHA's most cited standards in fiscal year 2025 ^[5].

The common failure modes are procedural. Written machine-specific procedures are missing or out of date, the verification step (trying to start the machine after isolation) is skipped, contractors and employees use different lock systems, or "minor servicing" exceptions are stretched to cover jobs that need full isolation. These are documentation and supervision problems first.

Powered industrial trucks

Forklifts appear in almost every plant. OSHA lists the typical ways workers are hurt: trucks driven off loading docks, trucks falling between a dock and an unsecured trailer, pedestrians struck by a truck, and workers falling from elevated pallets and tines. The standard, 29 CFR 1910.178, requires employers to certify that each operator is competent through training and evaluation, and federal law prohibits anyone under 18 from operating a forklift ^[12].

Manufacturing adds problems that a pure warehouse may not have: forklifts crossing production aisles where operators work at fixed stations, mixed traffic with tuggers and automated guided vehicles (AGVs), and loading areas that double as material staging zones.

Ergonomics and musculoskeletal disorders

EU-OSHA's 2019 report on work-related musculoskeletal disorders calls them the most common work-related health problem in the EU ^[15]. In manufacturing, the sources are well understood: repeated lifting at line pace, reaching into bins and fixtures, sustained awkward wrist and shoulder postures in assembly, and pushing and pulling carts.

These injuries rarely make headlines, but they drive a large share of lost workdays and workers' compensation costs, and they are where many wearable and video-based products focus.

Other recurring hazards

• Chemical exposure and hazard communication, particularly in process plants and in maintenance work on chemical lines ^[5].
• Respiratory hazards from dusts, fumes and mists, reflected in respiratory protection's place in OSHA's top ten ^[5].
• Heat stress near furnaces, ovens and in unconditioned buildings. OSHA proposed a federal heat injury and illness prevention rule for indoor and outdoor work in August 2024 and held its public hearing in June and July 2025. No final rule had been published at the time of writing ^[25].
• Robot cells and collaborative robots, where the hazard is a person inside the robot's working envelope while it can still move.
• Hot work, combustible dust and confined spaces in process industries.

Which regulations apply?

United States

OSHA's general industry standards (29 CFR Part 1910) apply to most manufacturing. The ones most relevant to technology projects are:

Consensus standards fill in the engineering detail. ANSI B11 covers machine tool safety in the US, and ISO 13849 and IEC 62061 set out how safety-related control systems are designed and rated. A safety-rated function, such as a light curtain that stops a press, has a defined performance level. A typical AI video alert does not.

United Kingdom

The Health and Safety at Work etc. Act 1974 sets the general duties, and the Provision and Use of Work Equipment Regulations 1998 (PUWER) apply them to machinery. Regulation 11 requires measures that either prevent access to dangerous parts of machinery or stop their movement before any part of a person enters a danger zone. It sets a hierarchy: fixed guards where practicable, then other guards or protection devices, then protection appliances such as jigs and push sticks, all backed by information, instruction, training and supervision ^[14]. A camera system that sends an alert after someone has entered a danger zone sits outside that hierarchy. It is a monitoring and supervision aid.

European Union

The Machinery Directive 2006/42/EC is being replaced by the Machinery Regulation (EU) 2023/1230, which applies from 20 January 2027 ^[16]. Two changes matter for safety technology. First, the regulation lists safety components with fully or partially self-evolving behavior using machine learning, where they ensure safety functions, among the categories that need third-party conformity assessment. Second, it adds requirements to protect safety functions and control systems against corruption, including malicious attempts ^[16].

The EU's Digital Omnibus on AI, Regulation (EU) 2026/1744, entered into force on 27 July 2026. According to Eurogip's analysis, it makes the Machinery Regulation the primary framework for AI systems that perform a safety function in a machine and requires high-risk AI requirements to be built into the Machinery Regulation by delegated act by 2 August 2028. The January 2027 application date is unchanged ^[17].

For buyers, this has a practical consequence. A camera product that only monitors and reports is unlikely to be a machinery safety component. A product that is wired to stop machines in order to protect people may be, depending on how it is designed and marketed. Ask the vendor how it classifies its product under the Machinery Regulation and who carries the conformity duty when you integrate it into an existing line. Privacy and AI Act obligations for monitoring workers are covered in chapter 13.

What safety technology is used in manufacturing?

AI video analytics on existing CCTV

AI video analytics runs computer vision models on camera feeds to detect defined events such as a person in a restricted zone, a missing hard hat or a forklift passing too close to a pedestrian. Chapter 4 explains how these systems work. In manufacturing, the use cases cluster around the hazards above:

Intenseye raised a $64 million Series B led by Lightspeed Venture Partners in February 2024 ^[21]. Its automotive page lists use cases including robot cell envelope intrusion, stamping press reach-in, conveyor pinch points, AGV and tugger traffic, and LOTO and energy isolation. The company says the system can trigger a machine stop at a robot cell or press in about 0.8 seconds. On the same page, Intenseye describes an anonymized global manufacturer with more than 20 plants that it says cut its recordable incident rate by 40 to 45 percent over two years and reduced lost-time days from about 500 to about 120 a year ^[18]. The company lists Bridgestone, Mercedes-Benz, Nestlé, Henkel, Mars and Nexans among its customers ^[18].

Two Intenseye case studies show the range of what vendors publish. For a US-headquartered global meat producer, Intenseye says the system flagged 597 potential hazards over nine months in 2022 against three found through manual documentation, and that hazard detections then fell by 61 percent ^[19]. Its Nexans case study, from a cable plant in France, describes forklift slowdown zones, pedestrian and equipment interaction tracking and heat maps, and reports a decrease in unsafe behaviors and near-miss notifications without giving numbers ^[20]. A fall in detections can reflect real behavior change, changes in how alerts are configured, or both. A buyer should ask which.

Voxel raised a $44 million Series B led by NewRoad Capital Partners in June 2025, bringing its total funding to $61 million. Its announcement names Berry Global, a packaging manufacturer, among customers alongside logistics and retail users, and claims a 91 percent reduction in recordable injuries and up to an 80 percent reduction in high-risk behaviors after deployment ^[22]. The release does not state the baseline period, sample size or how many sites those figures cover.

Protex AI describes a platform that runs on existing CCTV and connects to manufacturing execution systems (MES), warehouse management systems and PLCs. Its website lists General Motors, Toyota, Tesla, Nucor and Cummins among its customers and cites results including an 80 percent reduction in near-miss events in 54 days and a 44 percent reduction in lost-time injury rate, without naming the sites behind each figure ^[23].

Other vendors with manufacturing deployments are covered in chapter 12. The pattern across all of them is that published outcomes come from the vendors themselves, often for anonymized customers, and none of the figures above has been independently verified.

Connecting video to machine state and plant systems

Most false alarms in manufacturing come from the system not knowing what the machine is doing. A person inside a robot cell during a planned maintenance window under full lockout is not a violation. The same person inside the cell during production is a serious one. The better deployments read machine state from the PLC or MES, or from a simple I/O signal, so the model's rule can include "and the equipment is not isolated."

The same integrations allow the reverse flow: sending a signal to slow a forklift, sound a horn, or request a machine stop. Here the buyer needs a clear answer on whether the stop is part of a safety-rated function or a standard control action. Engineering and legal teams should review any design that relies on an AI output to protect a person from a hazard, especially for machinery sold or modified in the EU after January 2027 ^[16].

Wearables and sensors

Wearable devices are used in manufacturing for two main jobs: ergonomic coaching and proximity warning. Chapter 6 covers the technology in depth.

StrongArm Technologies sells a sensor worn on the body that gives haptic feedback when a worker bends or twists in a high-risk way, combined with training content. The company claims an average 35 percent year-over-year reduction in soft tissue injuries and lists 3M, Schneider Electric, Tyson and Walmart among its customers ^[24]. These are the company's own figures.

Proximity systems use ultra-wideband (UWB) or RFID tags on workers and vehicles to warn both parties when they get too close. They work where cameras cannot see, such as around racking and blind corners, but require everyone to wear a tag and the tags to be charged.

EHS software, permits and digital LOTO

Many plants still manage LOTO procedures, permits to work and confined space entry on paper. EHS software platforms digitize machine-specific energy control procedures, track periodic inspections required under 1910.147, and link permits to the equipment register. The value is in keeping procedures current and auditable. Chapter 7 covers EHS software in detail.

Digital twins and simulation

Some manufacturers use digital twins, virtual models of a line or plant fed with live data, to plan layouts and check traffic flows before changing a floor. For safety, the useful applications are planning forklift routes and pedestrian walkways, checking robot reach and fencing positions, and testing changeovers in simulation. These tools are mostly part of engineering and operations software rather than standalone safety products. Chapter 16 discusses where this is heading.

How do manufacturers deploy safety technology in practice?

The deployments described in public case studies tend to follow a similar path:

1. Start with incident and near-miss history to pick two or three hazards that drive serious injuries at the site.
2. Use existing CCTV where camera angles and resolution are adequate, and add cameras only where coverage is missing.
3. Run in a silent or observation mode first, so the team can see baseline event rates before alerts go to supervisors.
4. Tune zones and rules with operations, maintenance and worker representatives, especially around maintenance and changeover activity.
5. Route alerts to the person who can act (a shift supervisor or team leader), with a short review loop, rather than to a central inbox.
6. Use the trend data in toolbox talks and engineering reviews, and fix layout and process problems the data reveals.

The step most often skipped is the last one. A system that produces thousands of forklift close-pass alerts but never leads to a changed aisle layout or a new pedestrian crossing has produced data without reducing risk. Chapter 15 covers how to measure whether the technology is making a difference.

Worker trust and unions

Many manufacturing sites have recognized unions or works councils. Cameras that watch behavior raise concerns about discipline and surveillance. Deployments that last usually agree in writing what the data will and will not be used for, blur or anonymize faces where possible, involve safety representatives in setting rules, and use the data mainly to fix conditions rather than to punish individuals. Chapter 13 covers the legal side, including the EU AI Act and GDPR.

What should buyers ask vendors?

About detection and accuracy

• Which of our specific hazards does your model detect out of the box, and which need custom training?
• What precision and recall do you measure for each use case, on what data, and can you measure it on our footage during the pilot?
• How does the system handle occlusion by racking, machines and stacked product?
• How do you avoid flagging authorized maintenance work as violations? Can you read machine state from our PLCs or MES?

About machine integration and safety functions

• If the system can stop a machine or slow a vehicle, is that function safety rated? To what standard and performance level?
• How do you classify the product under the EU Machinery Regulation 2023/1230, and who holds the conformity duty when it is added to an existing machine?
• What happens to alerts and any machine interlock when the camera, network or edge server fails?

About outcomes and evidence

• For each outcome figure you publish, what was the baseline period, how many sites and workers were included, and what else changed at the site at the same time?
• Can we speak to a reference customer in our sub-sector, ideally one that has used the system for more than a year?
• How do you separate a real fall in unsafe events from a change in alert thresholds or camera coverage?

About data, privacy and workforce

• Where is video processed and stored, and for how long?
• Can faces be blurred, and can we turn off identification of individuals entirely?
• What support do you provide for consultation with unions and safety representatives?

About cost and fit

• What hardware is required (edge servers, new cameras, network upgrades) and who maintains it?
• How is pricing structured: per camera, per site, per use case?
• How does the system send events to our EHS software and maintenance management system?

Chapter 14 sets out how to turn these questions into a pilot plan with acceptance criteria.

Summary

Manufacturing kills fewer US workers than construction or transportation and warehousing, but injuries such as amputations are severe and permanent. BLS recorded 353 US manufacturing deaths in 2024, with contact with machinery and objects the leading cause, and OSHA's most cited standards for fiscal year 2025 include lockout/tagout, forklifts and machine guarding ^[2][3][5]. In Great Britain and the EU, manufacturing is a major source of fatal and non-fatal injuries, and musculoskeletal disorders account for a large share of long-term harm ^[6][8][9].

The technology now in use, mainly AI video on existing cameras, wearables for ergonomics and proximity, and EHS software for procedures and permits, is best at making unsafe conditions visible and measurable. It works alongside guards, interlocks and energy control procedures, which remain the primary controls. Vendors including Intenseye, Voxel and Protex AI publish large reductions in injuries and unsafe behaviors, but those figures are their own and should be tested against a site's baseline in a structured pilot. The EU Machinery Regulation, applying from January 2027, adds a specific compliance question for any system that uses machine learning to perform a safety function.