Omada in Harsh Environments: Networks and Cameras for Factories

Factories, warehouses and cold rooms are hard on network equipment: temperature swings, dust, moisture and long cable runs. They are also where businesses are most tempted by "AI vision", cameras that spot problems automatically. TP-Link's Omada platform can do a lot of the underlying work, and some of the camera analytics. Real defect inspection is a custom project rather than a built-in feature, but VIGI's open interface gives it something to build on. This guide separates what works out of the box from what needs development.

What "AI vision" means on a factory floor

The term covers two quite different things:

• Security analytics. Cameras that recognise people and vehicles, alert on someone entering a restricted zone or crossing a line, read licence plates at a gate, or count people. This is what TP-Link's VIGI cameras offer. TP-Link lists smart detection such as human and vehicle classification, intrusion, line-crossing, loitering, abandoned or removed objects and tampering, and names retail, education, hospitality and residential as its target scenarios.
• Machine vision. Inspecting products for defects, checking that parts are present or that people are wearing protective equipment. That normally needs cameras suited to the job, controlled lighting and a computer running inference models. VIGI's built-in analytics don't include defect inspection, but VIGI isn't a closed box either: see the next section.

For a manufacturer, the first is useful for site security and safety monitoring out of the box. The second is a custom project that VIGI can be part of, and Omada's role in it is the network underneath.

Customising VIGI: the Open API

VIGI cameras and recorders publish an Open API. TP-Link documents it as an interface that lets third-party applications communicate with the camera for device configuration, event subscription and video streaming: a secured control interface to query and change camera settings, a way to subscribe to events as they trigger, and RTSP streams for live preview and playback. TP-Link publishes separate API documents for its cameras and its recorders, and a list of supported models.

For a manufacturer, that opens the door to custom work. An integrator can build an application that pulls a VIGI camera's stream into its own inspection software, runs its own defect-detection model, and acts on the result, such as raising an alert or stopping a line. TP-Link's pages describe the Open API as an integration interface, not as a ready-made defect-inspection product or an open-source toolkit, so the inspection logic itself has to be developed and trained for your product. It also has to be validated: whether a given camera's resolution, lens and lighting suit a particular inspection is something to test on your line, not assume.

One network, one dashboard

Omada Central is TP-Link's cloud platform. TP-Link says it manages gateways, switches, access points, cameras and network video recorders from one place. The free Essentials tier covers core management, including industrial PoE switches, VIGI cameras and VIGI recorders; the paid Standard tier, which needs a per-device licence, adds features such as IDS/IPS, deep packet inspection, an open API and the AI video features like people counting and searching by person or vehicle attributes. Camera features are identical for Omada and VIGI cameras in Omada Central, but third-party cameras connected through a recorder get only basic live view and playback.

Controller options also include hardware and self-hosted software controllers, which suit sites that don't want cloud management.

Built for cold and heat: what the datasheets say

Ratings matter more than marketing in a harsh site, so here are TP-Link's published figures:

• EAP650-Outdoor access point: operates from -30°C to 70°C, at 10 to 90% relative humidity non-condensing, with an IP68 rating on TP-Link's current product page. A directional version, the EAP650 D120-Outdoor, is aimed at warehouses; TP-Link states its roaming support suits handheld scanners and automated vehicles.
• Industrial PoE switches (IES210GPP and related models): aluminium, DIN-rail or wall mounted, rated from -40°C to 75°C, with redundant DC power input.
• VIGI C350 camera (an example): IP67, -30°C to 60°C, non-condensing.

Every one of those datasheets specifies non-condensing humidity. That is the important caveat: a temperature rating doesn't mean a device tolerates condensation forming on it. In our cold-room deployment we mounted the access points below the compressor to avoid condensation, and that placement decision mattered as much as the rating.

Power and cabling come first

Industrial switches are only useful if the power budget and the cable runs are designed properly:

• PoE budget depends on the input. The IES210GPP's datasheet gives a PoE budget of 60, 120 or 240 W depending on input voltage, and 240 W applies only at the higher voltage range. Add up the devices before choosing.
• Reach has limits. Standard PoE runs to 100 m over copper. The datasheet describes an extended 250 m mode, but at 10 Mbps, and reach depends on cable quality. Long factory runs may need fibre or an intermediate switch.
• It's a basic managed switch. The IES210GPP supports VLANs, QoS and spanning tree, but TP-Link classes it as Easy Managed, so it isn't a substitute for a full-featured core switch.
• Ground and shield properly. TP-Link's outdoor datasheet notes that lightning and ESD protection depends on proper grounding and cable shielding.

Cabling, PoE and heatmapped access-point placement are design work, and they are where most harsh-site problems originate. We use AP heatmap analysis on projects for that reason.

Keeping production equipment apart from the office

Machines, controllers and cameras shouldn't share a flat network with office laptops. Omada supports VLANs, so production equipment, cameras and office traffic can sit on separate segments with rules between them. The right segmentation design depends on the site and the equipment, and should be planned rather than left to defaults.

Bringing inference on-site

Cameras generate a lot of data. VIGI's own datasheet lists main-stream bitrates of roughly 256 Kbps to 6 Mbps per camera, which adds up across dozens of cameras on a site with limited upload capacity. For custom machine-vision work built on the Open API, processing on-site rather than sending video to a cloud service is often the practical answer. That means an inference machine on the plant network: for prototyping and small-scale work, a desktop AI system such as the ones covered in our GB10 guide is one option, and higher-volume production lines need proper industrial or server-class hardware. In both cases Omada's contribution is the network they run on.

A real example, and its limits

We installed outdoor-rated Omada access points in a -30°C cold room and its office for a Sabah cold-chain logistics company, positioned to avoid condensation, and in-room Wi-Fi went from under 10 Mbps to around 400 Mbps. The case study has the detail. It's the same class of environment as a demanding factory floor, but it was a logistics site and a Wi-Fi project, with no cameras or AI video involved, and it isn't a manufacturing case study.

How Hakobi helps

We design and install networks for harsh sites, choose rated hardware, plan PoE and cabling, and set up cameras and segmentation. See Network Infrastructure and Structured Cabling, and the Omada partner page. Pricing depends on the site, the equipment and the licence tier, so request a quote and we'll scope it from there.

Sources: Omada Central and Essentials, EAP650-Outdoor, EAP650 D120-Outdoor datasheet, industrial switches, IES210GPP datasheet, VIGI AI, VIGI Camera Open API, devices supported by VIGI Open API, TP-Link.