Can Shop Floor Video Go to the Cloud? Start With What You Are Uploading

Uploading shop floor video to the cloud faces no technical obstacle; the real question is what data you are uploading. As a rule, equipment status, process parameters and cycle times carry almost no extra constraint. Footage that shows a face, a badge or an otherwise identifiable person is, in most cases, treated as personal information. Records that tie footage directly to an individual’s behaviour or performance need more care again. That is why most plants settle on "local first, cloud second" — inference and raw video stay on site, and only structured results and necessary samples are uploaded. What follows describes common engineering practice rather than legal advice; confirm the specifics with your legal team.

Step one: classify what you would upload

• Equipment and process data. Machine state, cycle time, operation results, alarm records. These point at equipment and product, usually carry no personal information, and are the least constrained.
• Footage containing people. Raw video, stills, and any frame that catches a face or a badge. Where it can be linked to a specific person, it is in most cases personal information.
• Person-linked records. Data that writes "who did which action at what time" into performance, attendance or disciplinary processes. Its purpose and retention scope need an explicit agreement.

Industrial data rules and personal information rules are two separate lines

These two are often bundled together, but they answer to different rules. Industrial data sits closer to the data-classification duties that apply to how a company identifies and protects important data and anything an industry regulator has flagged. Personal information turns on identifiability: if a clip lets a viewer connect a figure on screen to a specific person, it usually falls into that category. The practical consequence on the floor is that even when you only care whether a step was performed correctly, footage containing an identifiable face or badge name cannot simply be treated as ordinary production data. Blurring or masking a region is a common mitigation, but how much masking counts as effective is a judgement to make with your legal team.

When the cloud fits, and when it does not

Some customers in automotive and electronics write into their supply contracts that production data may not leave the plant or the country. That constraint usually bites harder than regulation. Scenes such as automotive final assembly tend to come with explicit expectations about who owns the process data. Confirm the position with the customer’s quality and legal contacts before designing the architecture.

If you do upload, how to lower the risk

There is no single right retention period. It usually falls out of three factors: the traceability years a customer or industry requires, the actual value of old data for improvement, and storage cost. A common compromise is keep all structured results for a long time (they are small), keep video and stills for a short time (they are large), and extend retention only for clips flagged as exceptions. That supports retrospective analysis without letting storage grow without bound.