This visibility gap is an uncomfortable weakness in modern manufacturing. Connected machines, sensors and data platforms alone do not make a factory truly smart. A smart factory must be able to detect what is happening, understand why it is happening and support the right response while there is still time to affect the outcome. That matters when 57% of UK manufacturing decision-makers recently surveyed identified rising operational costs as a major challenge, as highlighted in research by Axis Communications.
‘Visual intelligence’ offers an opportunity to close that gap, by connecting production events with what is physically happening across the plant.
Real-time systems, delayed explanations
Factories generate an enormous amount of data, but how much of it provides context? Only 16% of respondents said data from security technology was fully integrated into operational decision-making. For most respondents, that potentially useful source of information is only partly connected to the production workflow, if it is used there at all.
An alarm may pinpoint the second that a line slowed, but if the cause is not reflected in routinely measured values, engineers must begin a manual investigation. Comparing logs and speaking to the shift team may uncover the reason, but the process begins with the problem already having an impact.
Managing by hindsight may explain yesterday’s disruption, but it does not help operators intervene while the event is still developing. Visual intelligence can solve this.
Associate a machine alarm or production timestamp with vision-based sensors, and an operator can go straight to the relevant visual overview. If the event is still developing, that same view can show whether immediate intervention is required. Recorded video then preserves the lead-up for engineers investigating the underlying cause.
Analytics can make the response even faster. A camera might detect an object appearing where the process does not expect it and send a time-stamped alert, with the relevant visual context, to an operator.
Putting sight into production
Manufacturers are already putting this approach into practice. BMW Group, for example, uses network cameras within its AIQX platform for automated quality inspection. Cameras along the assembly line capture detailed vehicle images, synchronised with their location in the plant. AIQX analyses them for defects and assembly errors, allowing staff to address identified problems before the line moves on.
Nestlé uses compact cameras to supervise robots within coffee-jar filling operations at its French production sites. When an incident occurs, teams can examine the visual context to diagnose the cause and refine machine settings. Cameras also give operators a live view of equipment in sensitive production areas, reducing unnecessary entry.
In both cases, visual context has become part of how the plant operates. Vision-based sensor output directly feeds into a quality or production workflow instead of simply remaining as evidence for a later investigation. With the right integration, a visual detection could prompt an operator to intervene, trigger an inspection or even create a maintenance work order. The camera’s output becomes part of the plant’s normal response to a production problem.
Make the view fit the question
Existing security cameras can provide a starting point, but each view must be assessed against the operational question. The first step is to ensure visual sensors are detecting the right thing. Angle, field of view, resolution and frame rate must match the task. Lighting, vibration, dust or steam might also affect what can be seen.
The processing architecture should also follow the application. Edge analytics may suit a defined condition requiring a fast local response. Server- or cloud-based processing may be preferable where the application needs greater computing capacity or information from several sources. Whichever approach is chosen, there also needs to be a clear process for responding. Who receives the alert? What do they see? And what action can they take? Open platforms allow camera events and metadata to pass into compatible industrial and maintenance systems, enabling deep, customised integrations.
Health and safety requirements demonstrate the need for that specificity. If the aim is to protect people in restricted areas, assure safe processes or identify missing personal protective equipment, the visual sensor needs a clear view of the relevant zone, and its analytics must be configured for that condition. The resulting alert should reach a supervisor who can verify it and respond, giving the system a defined role in areas that one person cannot watch continuously.
This is about protecting them and assuring processes in a way that supports proactive safety. That distinction must be designed into the project from the start. Connected cameras need appropriate access controls, software maintenance and lifecycle management. The purpose and retention rules for archived visual context should be clear, and worker privacy should be treated as a core design requirement rather than an afterthought.
Privacy masking can preserve a view of the process while protecting workers, or be used in reverse to protect the privacy of the environment. Early consultation can prevent an operational project being mistaken for personal surveillance and help build confidence that visual intelligence is being used to improve safety, resilience and operational understanding.
Start with the costliest unanswered question
Trying to create total visibility across a factory is a difficult place to begin. Plant leaders should start with one persistent problem: unexplained short stops or a defect whose cause takes hours to establish. Before piloting, record its frequency, investigation time and potential impact. Design the sensor view, analytics and integration around the decision an operator or engineer must make.
Comparing the results with that established baseline will show whether investigation times and disruption have reduced. That small-scale pilot also tests image quality, alert handling and system links. It provides evidence for investment and establishes data flows before wider rollout.
The best place to start is with a problem that already costs the plant time and money. Connecting visual information with existing production data can give teams a clearer understanding of what happened and, more importantly, help them respond while there is still time to make a difference.
Want the best engineering stories delivered straight to your inbox? The Professional Engineering newsletter gives you vital updates on the most cutting-edge engineering and exciting new job opportunities. To sign up, click here.
Content published by Professional Engineering does not necessarily represent the views of the Institution of Mechanical Engineers.