MES/SCADA Integration in Container Glass, Vendor-Neutral

Everyone in this industry wants to talk about MES and SCADA integration like it's a software problem. Buy Ignition, point it at the PLCs, wire in Wonderware or AVEVA, and the plant is suddenly "digital". I've watched that project get funded, delivered, and quietly ignored by the hot end within six months, three separate times. The tag list was never the problem. Nobody had agreed which stream was actually the truth.

Container glass plants don't have one data source. They have six, and most of them don't trust each other.

The hot end doesn't run on one truth, it runs on six

An IS machine built on an old Emhart 8-section control cabinet from the 1980s reports timing and mould-open events one way. A newer Bottero or a Heye Smart H1 reports gob weight, plunger position and cavity identity a completely different way, often on a proprietary protocol the OEM would rather you didn't touch. Add a forehearth controller from a third vendor, a separate SCADA layer for the batch house, and a QA system tracking reject codes on a fourth database, and you understand why plant managers end up with four versions of "yesterday's OEE" in four different meetings.

This is where MES/SCADA integration either earns its keep or becomes shelfware. The job isn't collecting more tags. It's deciding which stream is authoritative for which number, and building the join logic so section-level data actually lines up with job and shift context.

OPC-UA is the plumbing, not the strategy

OPC-UA exists precisely because glass plants got tired of being locked to one vendor's historian to read their own PLC data. It's an open, vendor-neutral communication standard, and any integration built on it can, in theory, sit between an Emhart controller, a Bottero controller and whatever MES layer you choose without a forklift upgrade every time you change forming OEM. In practice I still see plants running Vertech SIL as the SCADA layer on the IS machines (a system a lot of Australian and GCC plants know well) feeding into Wonderware historians that were configured a decade ago and never touched since. The standard is vendor-neutral. The configuration usually isn't.

That's the trap. A control systems integrator sells you OPC-UA as the answer, then quietly builds the mapping so it only really works with their preferred MES. Six months later you're locked in again, just with a fancier diagram.

The protocol was never the lock-in. The undocumented mapping table nobody handed over at commissioning was.

What has to join before you add another dashboard

Before any MES rollout, I want five streams reconciled at the section level, not the line level. Section-level variance is where the money actually leaks.

• Gob weight and gob weight CV, target ≤0.4%, tied to job ID and mould set, not just machine number
• Mould preheat curve, target 480°C ±10°C, timestamped against the actual job change window
• Forehearth zone temperatures, ±2°C across five zones, not a single blended average
• Section timing, within 10ms tolerance, cross-referenced against IS machine vendor fault codes
• Swabbing timestamps and downtime codes, tagged to the operator who logged them

Get those five joined cleanly and you can actually trace a baffle mark back to a baffle alignment drift instead of guessing. Baffle marks point at baffle centring first, every time, and if your MES can't show you the centring tool reading against the timestamp of the defect, the dashboard is decoration.

The 0600 handover is where good data goes to die

In 2019 I audited a mixed-OEM plant in the Gulf, Bottero and Emhart lines running side by side under one Vertech SIL SCADA layer, chasing Vision 2030 capacity targets on a tight commissioning schedule. The forming data was fine. The MES dashboards looked sharp. But the 0600 handover missed the night-shift swabbing data on roughly seven changeovers out of ten, because the paper log and the digital system were two separate rituals nobody had merged.

Twenty-three minutes. That was the average extra job change time on that plant, purely from the day shift re-discovering what the night shift already knew and hadn't logged anywhere the MES could see. Not a furnace problem. A handover problem, wearing a digitalisation costume.

And this is the bit that gets skipped in every vendor pitch deck: the hot-end superintendent has to own recipe lock inside the MES, not just on paper. If an operator can change a set point without that sign-off showing up against the job record, your stones-and-cords traceability is fiction the moment someone tweaks a plunger setting at 2am to make a shift number look better.

Vendor-neutral is a decision, not a slogan

I've sat across the table from OEM-affiliated consultancies selling MES integration bundled with their own forming equipment, and the pitch always sounds the same: buy our machines, get our software, everything talks to everything. It does, right up until you want to add a competitor's IS machine to the same line, at which point the "integration" turns out to be one-way. A genuinely vendor-neutral container glass consultant has no stake in which OEM's controller wins that argument, and that's the only position from which the mapping table gets built honestly.

This is exactly the ground our digitalisation and reporting work sits on. We don't sell a historian or a SCADA licence. We map what your plant already owns, decide which stream is authoritative, and build the join so the numbers your hot-end superintendent trusts on the floor are the same numbers your VP of Operations defends at board level. That reconciled, section-level data is also what makes the Job Change Tool worth running, because KPI tracking is only as good as the streams feeding it.

Zaid Hassoneh built Lean Glass off two decades on the floor, starting at O-I Brisbane in 2005, and one lesson repeats on every audit: the plants leaking EBITDA aren't short on data. They're short on someone willing to say which version of the data is actually true.

If your MES and SCADA layers already exist but nobody trusts the numbers coming out of them, that's a mapping problem, not a software problem, and it's worth an outside pair of eyes before you buy another platform.