Why EU Container Glass Digitalisation Stalls at the Hot End

It's 2am on a Tuesday and the OEE dashboard is showing green. Section 4 is running at 94.2% efficiency, the MES is logging every event, and the plant's digitalisation pilot is, by every metric the vendor promised, working. What the screen doesn't show is that sections 1, 3, and 6 have been producing checks for the past 45 minutes because the blank-to-blow mould temperature delta drifted past 20°C. The only person who knows this wrote it in a paper logbook and told nobody.

That gap between what the platform captures and what actually matters on the line is why most European container glass digitalisation pilots don't scale beyond the first machine.

The integration wall that doesn't appear on the slide deck

The dominant IS machine controller across European hot ends is still the Siemens S7-400 platform, installed in more than 60% of EU hot ends. Its PLC scan cycle runs at 10–20 ms. It controls section timing, gob delivery, and machine kinematics, and it detects process drift long before the cold end generates a reject. The problem is that the S7-400 carries no native OPC-UA server interface. Connecting it to a plant historian, whether OSIsoft PI or an AVEVA System Platform, requires bespoke OPC-DA adapters built specifically for that controller generation. That is not a small integration task. It is typically the first thing that falls out of scope when the pilot budget gets squeezed.

A Glass International reader survey from 2022 found that fewer than 25% of European container glass plants had achieved full IS machine data integration into a plant-level data historian, with OPC-UA adoption at hot-end controllers sitting below 20%. Those numbers don't appear on vendor pitch decks. They appear on pilot post-mortems.

The same problem sits on the inspection side. XPAR Vision IRI systems capture up to 15 glass-distribution parameters per container at speeds up to 500 containers per minute. That is genuinely useful data. But in more than 70% of EU installations, the IRI output is not natively integrated with the IS machine PLC historian. Two rich data streams, sitting in separate silos, with no standard protocol joining them. The predictive-maintenance case that most operators were sold requires both streams synchronised to the same section reference and timestamp. Without that, you're not doing predictive maintenance. You're doing reporting.

What the pilot actually tested

In 2017 I spent three months supporting a three-furnace wine bottle plant in southern Spain during the first phase of their hot-end digitalisation programme. The pilot was scoped to one IS machine section. The numbers looked strong: gob weight variance held inside ±0.5 g at the feeder spout, section OEE improved by +3.2 points, and time-to-detect wall-thickness deviation dropped by 40%. The plant manager presented these results at a regional operations review and received full backing to scale.

Six months later, the programme hadn't moved to a second section. The OEM-affiliated advisory team had scoped the integration around their own telemetry stack, which had no data-normalisation layer for the furnace pull-rate data coming from a different vendor's SCADA. The moment the team tried to correlate gob-weight drift with forehearth temperature variance across vendors, the integration gap opened up. Two vendors. Two data formats. Zero integration pathway. The project went into "Phase 2" and stayed there until the budget cycle closed.

IIoT platform vendors routinely demonstrate ROI on a single IS section under controlled conditions. A single section running a known SKU, with an experienced operator holding the process stable, will almost always show measurable improvement. The question is what happens across eight or 12 sections on a mixed-OEM hot end during a job change.

The pilot worked because the operator made it work. That is not scalability. That is a controlled experiment with an uncommitted sample size of one.

The floor tells a different story than the dashboard

Even if you solve the integration architecture, you still have to deal with shift handover. Industry data shows that more than 60% of hot-end process deviations are communicated only verbally at shift change, with no digital record. On most lines I've audited, the IS machine operator logs three to five manual entries across an eight-hour shift. One entry every 90 to 100 minutes on a machine changing state every 20 seconds.

The gap isn't because operators are careless. It's because nobody built the data-capture step into the workflow. The 0600 handover runs for 12 minutes, covers cold-end pack rates, and moves on (and yes, I've sat in the back of that meeting watching night-shift swabbing data evaporate in real time). The hot-end superintendent picks up the rest on the floor walk. That knowledge never enters the historian. It never trains the model. When the vendor returns six months later asking why the predictive algorithm isn't firing accurately, the answer is that the training dataset reflects less than a quarter of what actually happened on the line.

Look, the data problem and the governance problem are the same problem. Shift discipline, handover structure, and role clarity around data capture are not IT questions. They are operational governance questions, and a structured management audit finds them where MES dashboards never look.

What scaling actually requires

The EU ETS Phase IV benchmark sits at 0.538 tCO2 per tonne of liquid glass produced. The Cross-Sectoral Correction Factor for 2021–2025 reduces free allocation to around 82% of that figure, exposing producers to a real carbon cost on the uncovered 18% at prevailing EUA prices of EUR 50–65/tonne. Verallia has committed to a 46% absolute CO2 reduction by 2030 against its 2019 baseline, with digital furnace twins part of the pathway. Ardagh is running EUR 150M+ in furnace rebuilds across its European estate with electric-boost integration in progress. The capital is moving. The question is whether the hot end moves with it.

Furnace digitalisation and IS machine digitalisation are not the same problem. A furnace twin improves pull-rate stability and batch-to-melt efficiency. Forming stability requires synchronised data from feeder, IS machine, lehr, and inspection, mapped to a clear ownership model for who acts on each signal and when. That is a methodology question as much as a technology question.

NNPB plunger settle-blow timing is a useful illustration. The target window is 380–420 ms. Drift of more than 25 ms from nominal produces wall-thickness asymmetry outside ±0.2 mm spec, and the defect surfaces at the cold-end ACI station. But the root cause sits only at the IS machine controller. If the MES integration doesn't reach back to the PLC with section-synchronised timestamps, the plant spends three shift cycles chasing a cold-end inspection signal that is really a forming deviation from eight hours earlier.

And this is where a digitalisation and reporting engagement built around the hot end, rather than retrofitted onto it, separates a pilot that proves a concept from a programme that changes operational outcomes across a full plant fleet.

The plants that scale past the pilot stage have one thing in common. They didn't start with the technology. They started with integration architecture, governance clarity, and role ownership, all defined before the vendor walked in the door. A vendor-neutral container glass consultant can tell you where your plant sits on that readiness curve before the next capex cycle opens. That assessment is what the Lean Glass vendor-neutral consultancy model is built to deliver, from operators who have actually run the IS machine, not just observed it.