Running a 2-Furnace, 5-Line Brisbane Container Glass Plant

It's 3am on a Thursday, T-45 min on the board, and Furnace 2 is coming off Flint onto Amber for the third time that week. Section 6 is running a leaner nobody's flagged yet. The night-shift operator is guessing at the mould preheat curve because the day-shift note is somewhere in a drawer, not in a system. That plant was mine, in Brisbane, and that scene played out often enough that I stopped being surprised by it and started building a fix.

Why 180 job changes a year exposes everything a plant tries to hide

Two furnaces, five lines, three colours in rotation, Amber, Flint and EM Green. Across a year that plant ran roughly 180 job changes. That's not a big number until you sit through enough of them to see the pattern: the furnace never changes, the moulds never really change, but the outcome on section 4 versus section 9 is wildly different depending on who's on shift and whether the SKU Library, or whatever notebook was standing in for one, actually got read.

A job change on a ten-section IS machine typically costs 45-90 minutes of downtime plus a two-to-four percentage point reject spike in the first half hour of re-stabilisation. Multiply that by 180 changeovers and you're staring at a controllable loss most plants never quantify, because nobody's tracking changeover time and first-ware quality as a trended KPI. They're tracking tonnes out the door and calling it a day.

The furnace wasn't the constraint. The mould shop was.

Everyone wants to talk about pull rate. Pull rate was never our problem. Our problem was that Furnace 1's old Emhart, ten-section, running control logic that hadn't been touched since the 1990s, could be fed a new job faster than the mould shop could turn a set around. A mould swap ran 48-72 hours end to end, and job allocation between the two furnaces got rebalanced weekly against order book and mould inventory, not against what the furnace could theoretically pull.

Furnace 2 had a newer forming line and a tighter recipe lock, and it still choked on the same bottleneck. Nobody upgrades a mould shop's scheduling discipline by buying a new IS machine. I've watched OEM-affiliated consultancies walk a plant floor and recommend capex before they've asked who owns the mould inventory spreadsheet.

The furnace doesn't create your variance. Your handover does.

What a 3-5 point OEE swing between shifts is actually telling you

World-class hot-end OEE on a multi-line container plant sits around 85%+. Most plants I've audited since, across the Middle East and North America, run somewhere in the 70-78% band and call that normal. When you see a 3-5 point swing between the same line on different shifts, running the same SKU, that's not a machine fault. That's a changeover discipline gap or a handover gap, full stop.

We proved it the hard way. The 0600 handover missed the night shift's swabbing data more often than it caught it, my own rough count put it near 70% of the time on the lines I was watching closest. Swabbing cycle drift is one of the fastest routes to a hot-end reject spike above the 1.5-3.0% target band on an NNPB line, and if the incoming shift doesn't know the last swab time, they're flying blind on lubrication-related surface defects until the QA sample catches it. By then you've made a few thousand pieces of scrap you didn't need to.

Checks at the finish were our other recurring warehouse-stage reject, and they nearly always traced back to a five-to-eight degree drift in the lehr or cooling-fan profile, not a batch problem. Cord defects, when we saw them, meant we needed to look at AZS block wear in the throat before we blamed the batch house. Stones, seeds, blisters, we had all of them at different points, and every single one had a named cause once you stopped treating it as bad luck.

Colour changes taught me more than any capex review ever did

Amber to Flint to EM Green and back, over and over, is where a plant's tribal knowledge gets tested hardest. Every colour change is a full recipe swap, a mould preheat curve to hit (we ran ours to a 480°C ±10°C target), and a forehearth profile that needs to sit within about ±2°C across five zones before first ware comes off clean. Miss any one of those and you're not looking at a bad batch, you're looking at a lost shift.

In 2016 that plant won O-I's Most Improved Job Change Plant Globally award, and it wasn't because we bought new equipment. It was because we mapped every one of those 180 changeovers against a defined stage lifecycle, named an owner for each stage, plant manager for plan and post-mortem, hot-end superintendent for recipe lock, mould shop for the physical swap, operator for line-down and first ware, and stopped treating changeover as something that just happened to the shift on duty. That's the same nine-stage logic, plan, prep, line-down, mould change, recipe load, ignition, first ware, stabilise, post-mortem, that eventually became the backbone of what we now run as a systemised Job Change Tool.

And it travels. When I left Australia to lead the Arglass Yamamura greenfield build in Georgia, the US market was going through its own reckoning, O-I's Fit to Win programme was targeting roughly two million tons of excess North American melting capacity for closure, and every surviving plant needed every controllable point of OEE it could find because there was no more room for waste to hide in a shrinking network. The lesson from Brisbane held: the fastest EBITDA wins were never in the furnace. They were in how disciplined the changeover was.

What this means if you're not tracking it yet

If your plant runs anything close to 180 job changes a year and you can't tell me, by section, what your average changeover time and first-ware reject rate were last month, you're not managing that variance. You're absorbing it. Most plants I audit are leaving four to eight OEE points on the table this way, and it's rarely a single dramatic failure. It's 180 small ones, compounding.

• Lock the recipe and mould spec in one versioned library, not a notebook
• Name an owner for every stage of the changeover, not just the operator running it
• Trend changeover time and first-ware quality by section, not just by plant average

None of that requires new capital equipment. It requires someone independent enough to walk the floor without an OEM logo on their shirt and tell you where the real number is hiding. That's the whole case for bringing in a vendor-neutral container glass consultant instead of the vendor who sold you the machine, they've got no reason to point you at the mould shop when a new forming line is the bigger invoice.

I still think about that 3am changeover on Furnace 2 sometimes. Not because it went badly. Because it was ordinary, and ordinary is exactly what was costing us the most. If you want to know what a proper stage-by-stage teardown of your own changeover looks like, our Brisbane plant manager background is the reason we built the tool the way we did, not from a spreadsheet, from the floor.