Container Glass OEE: The 4-8 Points You Already Own

The industry average pack-to-melt ratio sits between 85% and 90%. World class is 92–95%. On a 300 TPD furnace, lifting PTM from 87% to 91% recovers 12 additional saleable tonnes per day. No new IS machine. No furnace rebuild. Twelve tonnes, every day, sitting inside the process you already run.

Most container glass plants are leaving 4–8 OEE points on the floor right now. Not in the furnace design. Not in the IS machine generation. In the controllable losses that accumulate shift by shift, job change by job change, and colour campaign by colour campaign. This piece maps where those points actually live.

The job change drag that never gets properly costed

In 2017 I was running a job change review for a two-furnace, six-line plant in the GCC. The plant manager was confident his average changeover ran 4.5 hours. When we pulled line-down timestamps against first-stable-pack times across 90 days, the actual number was 6 hours 40 minutes. Three hours and twenty minutes. That was the average cross-shift spread on an identical SKU running the same machine with the same moulds.

That variance is the real cost. The downtime total is a known number and someone is already budget-managing it. The variance is invisible until you measure it, and it points directly at the absence of standardised execution. The best shift team and the worst shift team were separated by half a working shift on the same job.

A systemised Job Change Tool closes this gap by locking every recipe, mould set, and forming spec into a single versioned source of truth, then mapping execution against the 9-stage Job Change Lifecycle so progress is visible in real time rather than reconstructed from memory at the 0600 handover. Plants working with this structure have cut changeover time by 30–60% and, more importantly, compressed the cross-shift variance. That compression is where the compounding OEE recovery comes from. A vendor-neutral approach matters here: a methodology shaped by what one OEM sells is not a methodology built around what the plant actually needs.

Gob weight and forehearth stability are the same problem

If your forehearth is running ±5–8°C at the feeder bowl exit instead of the target ±2–3°C, you already have a gob weight problem. Temperature instability drives viscosity excursions, viscosity excursions drive gob weight scatter, and that scatter drives overweight and underweight rejects straight to the checkweigher. Those four steps happen in sequence, every cycle, all shift.

The target for gob weight variation on a 200–330 mL container is 1σ ≤ 0.5 g. Most lines I walk into are running at 1σ = 1.5–2.0 g on uncontrolled feeders. That number alone accounts for a significant share of checkweigher rejects, and those rejects hit the quality-rate numerator directly. Closed-loop feeder control (a Bucher Emhart IS-CAM or comparable servo-drive plunger system) routinely brings that σ down to ±0.7 g, adding +1.5 to +2.5 OEE quality points. But before you go near control hardware, check the PLC cascade loop tuning on the forehearth (and yes, I know your controls tech will say the settings are fine, but get the trend log anyway and look for cyclic excursions). Poorly tuned proportional-integral gains are the most common cause, and correcting them costs nothing but a planned intervention shift.

The hot-end superintendent owns this. Not the operator. The superintendent pulls the forehearth trend log, identifies whether excursions are cyclic or drifting, and makes a go/no-go call before the problem compounds into a quality event on the next shift. If that ownership is not explicit, the problem gets handed from shift to shift indefinitely.

The lehr is hiding half your check rejects

Checks are thermal cracks, typically initiating in the annealing lehr, and they are the most common defect mode I see misdiagnosed on underperforming lines. The hot end gets blamed. The gob weight gets blamed. Nobody looks at the lehr lateral temperature gradient until the rejects are in the hundreds per hour.

Zone 1 entry on a standard soda-lime line runs at 560–580°C. The target lateral gradient across the belt width at that point is ≤5°C. Plants running older Wellman-type or Sievert-type lehrs frequently show ±8–12°C gradients, and those gradients produce check-defect rates 30–45% higher than a correctly profiled lehr. Not a new lehr. Thermocouple placement verification, zone heater balance adjustment, and one planned shift to correct the profile. I have seen this completed in under eight hours with measurable check reduction inside 48 hours of restart.

Birdswings are the second defect mode worth separating out here. They point at shear blade condition or alignment, not the lehr. But they often appear in the same cold-end reject stream as checks, muddying the Pareto and misdirecting the investigation. On a 10-section double-gob machine, two or three sections routinely generate 60–70% of all defects. Shift-level monitoring masks that completely. Section-level monitoring at ≤15-minute resolution using a hot-end vision sensor (a Heye HST or ISEC-type unit works well here) reduces defect incident detection from 2–4 hours to under 15 minutes, recovering +0.3 to +0.8 OEE availability points per occurrence. That adds up fast on a line running poor glass.

The cold end is telling you what happened two to four hours ago. If your forming decisions are being made from a cold-end reject report at handover, you are not managing quality. You are archiving it.

Colour scheduling is the availability win most plants never touch

Colour change is the most under-managed availability loss in container glass. An amber-to-flint flush runs 12–20 hours. Green-to-amber can run 20–40 hours depending on pull rate and furnace residence time. Those numbers are not negotiable. They are physics. What is negotiable is the order in which colour campaigns run across the furnace campaign.

Sequencing campaigns lightest-to-darkest chronologically (flint to amber to green, not reactive and not mixed) reduces total annual colour-change downtime by 20–30%. On a furnace running three or more colours per campaign, that sequence discipline translates to +1.5 to +2.5 OEE availability points annually. No capital. No equipment. Commercial planning and production planning aligned on the same calendar, which turns out to be harder than it sounds at most plants.

And in the GCC this pressure is acute. The container glass market across the Gulf was valued at approximately USD 820 million in 2024, growing at around 4.2% per annum driven by Saudi Vision 2030 localisation mandates for food and beverage packaging. That growth is pulling plants into wider SKU mixes and broader colour portfolios faster than scheduling discipline has kept pace. A plant that cannot manage colour campaign sequence will lose availability points steadily as its range expands, and no amount of machine investment will recover them.

Shift variance is the OEE lever that equipment reviews always miss

Same IS machine. Same SKU. Same mould set. A ±3–5 OEE point spread between your best and weakest shift teams. This is the number that OEM-affiliated reviews and generic Lean/Six Sigma boutiques almost universally fail to surface, because both treat operator process as a constant.

It's not a constant. It is the largest single controllable variable on most lines.

Structured shift-comparison analytics track gob weight response time, section-alarm escalation cadence, and time-to-stable-pack by shift and by operator. That data consistently identifies 1–2 OEE points that equipment-scoped reviews leave permanently invisible. The intervention is not a training programme. It is targeted coaching on specific responses, standardised handover content, and a shift-comparison KPI the hot-end superintendent reviews weekly. A number, owned by a person, reviewed on a schedule.

Look, you can spend a year chasing furnace efficiency gains of half a point each, or you can run a structured hot end audit that treats job change execution, gob weight discipline, lehr gradient management, colour scheduling, and shift variance as a connected system. The 4–8 OEE points are already there. So what is stopping your plant from finding out which lever to pull first?