The 9-Stage Job Change Lifecycle for Container Glass Plants

It's 3am on a Thursday and the job card for section 4 says T+45 min to first ware. The mould shop tech is still torquing the last neck ring. The hot-end superintendent is watching gob temperature drift and doesn't like what he's seeing. Nobody on this shift can tell you which stage of the changeover they're actually in, because nobody ever agreed there were stages to begin with. That's not a training gap. That's a system gap, and it's the single biggest one I see when I walk a hot end.

Most plants treat a job change as one undifferentiated block of downtime, bounded by "line stopped" on one end and "line running good ware" on the other. Nine distinct stages sit inside that block, each with its own owner, its own exit criteria, and its own failure modes. Skip that structure and you're not running SMED for container glass, you're running a scramble with a stopwatch on it.

Nine gates, not one mad scramble

The 9-stage Job Change Lifecycle breaks a changeover into: plan, prep, line-down, mould change, recipe load, ignition, first ware, stabilise, post-mortem. Each one has a named owner and a signed exit condition before the next gate opens.

• Plan — plant manager confirms SKU, mould set and recipe against the locked specification
• Prep — mould shop stages tooling and preheats moulds ahead of line-down
• Line-down — hot-end superintendent controls the stop, not a scramble stop
• Mould change — section mechanics execute the mechanical swap
• Recipe load — operator loads set points under superintendent sign-off
• Ignition — furnace and forehearth ramp under a controlled pull-rate curve
• First ware — QA technician releases first-article ware from quarantine
• Stabilise — line holds spec long enough for the lehr and cold end to settle
• Post-mortem — the crew signs the job card and feeds data back to plan

Plan and prep are where plants either buy or burn the next seven gates. If the mould set isn't confirmed against a locked recipe, you find out at the mould change gate, which is the worst possible place to discover it. In 2020 I audited a five-line plant in Saudi Arabia's Eastern Province where the plan stage was genuinely a phone call between shifts, nothing written, nothing versioned. Two of the four job changes that week ran a mould set that had been quietly modified by the mould shop six weeks earlier and never logged.

Staffing belongs in the plan stage too, and in the Gulf that's not optional. Saudi Arabia's Nitaqat Saudisation programme sets minimum national-employment quotas by sector, which directly caps how many expatriate hot-end mechanics a plant can roster onto a night-shift job change (Saudi Ministry of Human Resources and Social Development). A plan that assumes four expat mechanics on a shift that's only cleared for two isn't a plan, it's a hope.

Line-down, mould change and recipe load are the gates that actually eat the clock

Line-down is a controlled stop with a pull-rate curve, not a button press. Mould change is where the mechanical work happens, and it's also where mould preheat discipline either holds or gets skipped under time pressure. New moulds want a preheat window of roughly 450-500°C before mounting. Mount below that and you're buying thermal-shock checks in the first minutes of production, full stop.

I still see plants running an old Emhart eight-section machine on 1980s cam timing next to a newer Heye Smart H1 line, and the mechanical swap discipline needs to match the machine, not the brochure. Baffle marks on early ware almost always point at baffle alignment drift, so check the centring tool reading first (and yes, I know your fitter says the timing chart covers it, check it anyway).

Recipe load is where the hot-end superintendent owns recipe lock and nobody else touches a set point without sign-off. Gob temperature at the feeder spout gets held to a tight ±5°C band during the trial, with gob weight CV targeted at 0.4% or under. Top-quartile plants running a systemised protocol get a mould/job changeover down to 20-30 minutes per section. Industry-average, non-optimised changeovers commonly run 90-180 minutes per section. That's a 3-6x delta, and it goes straight to OEE.

Twenty-three minutes. That's the real number a well-drilled crew hit on a straightforward SKU repeat I watched last year. Not a furnace problem. A handover problem, most of the time, when it goes the other way.

Ignition and first ware are two different gates, not one

Ignition is the furnace and forehearth ramp, and in the US it's a compliance gate as much as a thermal one. Container glass furnaces are regulated as area sources of hazardous air pollutants under 40 CFR Part 63, Subpart SSSSSS (US EPA), which constrains how far furnace conditions can swing during a ramp event. Treat that as a pure mechanical problem and you'll blow past a permit limit chasing pull rate.

First ware is a separate gate with a separate owner. Cold-end automatic inspection, linescan gauging, wall-thickness and verticality checkers, gets held in reject-all quarantine mode until a first-article dimensional check clears, typically against a ±0.3mm wall-thickness tolerance. That release belongs to the QA technician, not the operator, and it's the gate most OEM machine-timing consultants skip straight over because it sits outside the mechanical swap they were hired to optimise.

Reject rate commonly spikes from a steady baseline of 1.5-2% to 15-25% in the first 20-30 minutes after restart. Cold-mould checks, overpress, birdswing marks, double-gob and stuck-in-mould ware all show up in that window. Stones and cords are a different animal entirely and point back at batch and refractory, not the changeover, but operators lump them into the same reject bucket every time and it muddies the KPI.

A job change isn't finished when the machine is running. It's finished when the lehr agrees with the machine.

Stabilise and post-mortem decide whether the next change-over is faster

The stabilise gate is where upstream SMED thinking usually falls apart, because the annealing lehr needs a 20-40 minute stabilisation lag after any change in container mass or geometry before residual stress, birefringence, returns within spec. Mould swab cadence tightens to every 20-30 minutes during this window against 45-60 minutes at steady-state running, reflecting the thermal cycling stress on fresh mould surfaces. Call the job done at first good ware off the machine and you'll have a lehr full of ware that fails a polariscope check twenty minutes later.

But the ninth gate is the one most plants skip outright. Post-mortem isn't a meeting where everyone agrees the shift went fine. It's a data capture step, timestamped on the job card by the hot-end supervisor, two to three section mechanics, the mould-shop technician and QA, that feeds straight back into the plan stage for next time. A plant running 150-plus changeovers a year can lose 3-5% of monthly OEE to job-change events alone, and without post-mortem data you're re-litigating the same argument every single time. The 0600 handover misses the night-shift swabbing data on most lines I audit, and that gap alone accounts for a lot of repeat mistakes.

Look, the data says one thing and the floor says another most weeks, and post-mortem is where you find out which one was right. In Europe, that data discipline matters more every allocation period. EU ETS Phase IV applies a linear reduction factor of 4.3% per year through 2027, rising to 4.4% from 2028, tightening the free carbon allowance available to glass melters (European Commission, DG CLIMA). FEVE members are pushing to shorten changeover windows precisely because every minute of unstable pull rate during a job change burns disproportionate gas per good container, and CBAM's exclusion of glass so far hasn't slowed that pressure down.

Why generic SMED templates undercount the real time

OEM-affiliated consultancies tend to optimise only the mechanical swap, mould, plunger and neck-ring timing, and wave past the furnace stabilisation and cold-end quarantine gates that actually bound total changeover duration. Generic Lean and Six Sigma boutiques import automotive-style die-change templates wholesale, without adapting for molten-glass constraints like mould preheat curves and gob viscosity windows, which systematically understates what's achievable. Neither approach accounts for workforce constraints like Nitaqat quotas or unionised changeover crew rules, even though those cap achievable speed regardless of how good the mechanical work is.

That gap is exactly what a systemised, vendor-neutral Job Change Tool is built to close, versioned recipes and mould sets, a live checklist mapped to all nine gates, and KPI tracking on changeover time, first-ware quality and time-to-stable-pack, so the post-mortem gate has real numbers to argue from instead of shift-notebook memory. Lean Glass's founder, Zaid Hassoneh, won O-I's Most Improved Job Change Plant, Globally award in 2016 running exactly this kind of gated changeover on the floor, not from a slide deck.

Nine gates, nine owners, one signed job card. It's not glamorous work. It's the work that decides whether your OEE number means anything next quarter.