Xpar Vision Deployment: Why Container Glass Plants Stall

Buying the system isn't the hard part. Every plant manager I've sat across from in the last five years has told me the same thing after a hot-end intelligence platform goes in: the cameras are up, the dashboards look sharp on the big screen in the control room, and six months later the reject rate has barely moved. That's not a software problem. It's a deployment problem, and it's the same one whether the badge on the cabinet says Xpar Vision or anything else in that category.

I'm not knocking the technology. Thermal imaging across the gob, the blank side and the ware handling has genuinely changed what's visible on a forming line that used to run on gut feel and a torch at 2am. But a camera that only alarms after the defect has already shipped down the lehr isn't a hot-end intelligence system. It's an expensive rear-view mirror.

Where Xpar Vision deployments actually stall on the floor

In 2019 I audited a two-furnace plant in the Gulf running thermal imaging across five IS lines, cameras covering gob delivery through to ware inspection. Good hardware, properly commissioned, OEM training signed off. And the section-to-section variance on identical SKUs was still running north of 40% job to job, which is exactly the number the platform was supposed to kill.

What we found wasn't a camera fault. The alarm thresholds were still set to factory defaults across every mould set, regardless of container weight, wall thickness or NNPB versus blow-and-blow process. Nobody owned recalibrating them per job. The hot-end superintendent trusted the system to flag drift, the system was flagging drift against the wrong baseline, and the operators had learned to mute alarms that fired constantly and meant nothing. Classic alarm fatigue. By the time anyone looked at the trend data properly, three shifts of baffle marks had already gone to the warehouse.

A camera that only alarms after the defect ships isn't a hot-end intelligence system. It's an expensive rear-view mirror.

The data needs an owner, not just a screen

Here's the pattern I see in plant after plant. The vendor sells the platform on OEE and defect reduction. The commissioning team gets the feeds live, sets up the standard dashboards, and leaves. Nobody at the plant has been given clear ownership of what happens between an alarm firing and a set point changing. The operator isn't authorised to touch the recipe. The hot-end superintendent is on the floor chasing something else. The data sits there, technically correct, operationally useless.

On one retrofit I worked, an older Emhart 8-section line running 1990s discrete controls had thermal cameras bolted on as an add-on package, feeding a modern dashboard that the control system underneath had no way to act on automatically. Every correction still had to be keyed in by hand. That's not a criticism of the retrofit. It's the reality on a huge share of the installed base globally, and it means the human process around the data matters more than the resolution of the camera (and yes, I know the commissioning engineer swore the default thresholds were fine for every job on that line — check them against your own mould set anyway).

The fix isn't more cameras. It's naming who owns the thermal baseline for each SKU, who's authorised to adjust it, and where that decision gets logged so the next shift isn't guessing. Twenty-three minutes. That's the average time I've seen crews burn re-discovering a thermal profile that was already known and already correct, because nobody wrote it down against the job. Not a furnace problem. A handover problem.

Feeding hot-end intelligence into the job change discipline

This is where the platform earns its keep or doesn't. A thermal imaging system is only as good as the job change discipline sitting underneath it. If mould preheat curves, gob weight CV targets and forehearth profile setpoints aren't locked per SKU, the camera is comparing today's run against yesterday's guesswork, not against a validated standard.

We build this into our Job Change Tool for exactly this reason. Every SKU carries a locked forming spec, mould preheat curve target of 480°C ±10°C for example, gob weight CV held under 0.4%, forehearth profile within ±2°C across five zones, so that when the hot-end intelligence platform flags a deviation, it's flagging against a real baseline instead of a factory default. The 9-stage Job Change Lifecycle gives the thermal data a home: at ignition and first ware, the camera output gets checked against the locked spec before the line is called stable, not after the pack has already drifted.

• Lock the thermal baseline to the SKU, not to a generic factory default
• Name the owner who signs off any threshold change mid-run
• Log the correction against the job so the next shift isn't rediscovering it

European plants running under margin pressure from ETS Phase IV compliance costs are asking us for exactly this kind of integration work, because a new furnace build isn't on the table and the controllable win has to come from inside the existing fleet. The digitalisation and reporting side of an engagement is usually where we start, mapping what data already exists before anyone talks about buying more of it.

What good actually looks like

A plant I worked with in 2022 had Xpar-class thermal coverage on three lines and treated it as a job change instrument, not a monitoring toy. Section timing held within 10ms of spec, baffle-related rejects down 45% year on year, and the 0600 handover included the night shift's swabbing data as a matter of course instead of it getting lost in a paper log. That's not a technology outcome. That's an operating discipline outcome, and the cameras just made it visible faster.

Zaid Hassoneh built Lean Glass on that distinction after running hot ends himself, including the $220M USD Arglass Yamamura greenfield build, and it shows up in every audit we run. You can read more about that global consultant experience on our site. The point isn't which camera brand you bought. It's whether your plant has the discipline to use what it's already paying for.

If your hot-end intelligence platform has been live for over a year and the variance numbers haven't moved, the camera isn't the thing to interrogate first. A vendor-neutral container glass consultant looking at how the data actually flows through your shift handover usually finds the answer inside a week.