A Defect Taxonomy Built for Container Glass, Not Generic SPC

Most defect taxonomies in container glass plants are borrowed, not built

Walk into ten container glass plants and you'll find nine different defect boards, and eight of them are dressed-up automotive SPC charts with glass words stapled on. Cp/Cpk, control limits, root-cause fishbones built for a stamping press. None of it was written for a gob of glass at 1100°C dropping into a blank mould every 0.4 seconds.

Generic quality systems treat every defect as a statistical event first and a physical one second. Container glass doesn't work that way. A stone tells you something specific about refractory wear or batch contamination. A cord tells you something else entirely, usually furnace temperature stratification or a bad cullet campaign. Lump them into one Pareto chart labelled "surface defects" and you've thrown away the diagnostic value before the shift even starts.

Where the defect actually starts tells you more than what it looks like

The taxonomy that works on the floor isn't organised by appearance. It's organised by origin: melt-end, forming, or cold-end. Stones and cords are melt-end, born in the furnace or forehearth long before the gob ever sees a blank mould. A forehearth running outside ±2°C across five zones will cord a line for a week before anyone traces it back correctly. Blisters can be either, depending on whether the gas is trapped in the melt or generated at the mould-glass interface during forming.

Cold-end problems are the third bucket, and the one most plants forget to separate out. A lehr drop from an annealing temperature swing looks like a packing problem until someone checks the lehr profile log. It isn't a forming problem at all, and treating it as one wastes a shift chasing the wrong machine.

Baffle marks, leaners, washboards and choked necks are forming defects, full stop. No amount of furnace tuning will fix them.

I audited a two-furnace, six-line plant in the US Midwest in 2017 running a mix of 1990s-era Emhart 8-section machines and one newer NNPB conversion line. The quality team kept a single defect log for all six lines, sorted alphabetically by defect name. Stones from a refractory problem on furnace two sat in the same column as leaners from a bad section timing on line five. Nobody could see the pattern, because the taxonomy hid it.

A defect name tells you what happened. Where it started tells you what to fix. Most plants only track the first one.

Eight defects, eight different diagnostic starting points

These are the defects every hot-end and forming team should be able to name and place without checking a manual:

• Stones — melt-end contamination or refractory attack, check the batch house and furnace crown first
• Cords — temperature stratification or striae in the melt, check forehearth zone profile
• Blisters — reboil gas trapped in the melt or generated at the mould wall, check swab condition and cycle
• Seeds — fine gas bubbles from incomplete fining, check furnace residence time and batch chemistry
• Baffle marks — baffle alignment drift against the blank mould, check the centring tool reading first
• Leaners — uneven cooling or a settle wave problem on takeout, check plunger timing and mould temperature balance
• Washboards — parison stretched too fast for the glass viscosity at that point in the cycle, check gob temperature and delivery timing
• Choked necks — neck ring wear or a finish mould mismatch, check finish dimensions against the SKU Library spec before blaming the machine

Notice none of those root causes are "operator error." That's deliberate. Ninety per cent of the defect calls logged as operator error on a shift report are actually equipment drift nobody measured. Not laziness. Not carelessness. Drift.

On a properly run line, the hot-end superintendent owns recipe lock. An operator doesn't touch a gob temperature set point without sign-off.

Why the taxonomy falls apart at 0600

Here's where it actually breaks on real plants. The 0600 handover misses the night-shift swabbing data on most lines I've audited, easily 70% of the time. The outgoing operator knows which sections ran hot, which mould got a fresh swab at 0400, and which cavity has been throwing seeds since midnight. None of that survives the handover if the defect log is a generic form with a checkbox for "surface defect, yes/no" (and yes, I've seen that literal checkbox more than once).

And that's the real cost. Not the rejects on the shift where the defect happens. It's the four hours on dayshift spent rediscovering what the night crew already knew, because the taxonomy gave them nowhere to write it down properly.

On a five-line plant I worked with in the Gulf region, gob weight CV was running at 0.6% against a target of 0.4%, and the defect log showed intermittent washboards on two of the five lines with nobody connecting it to gob temperature variance. Splitting the log by origin instead of appearance found the pattern in 90 minutes. It had been sitting in six weeks of data the whole time. Once mould temperature swing tightened back to target, washboard rejects fell -2.1 percent-pack over the following month.

Build it into the job change, not a wall poster

A defect taxonomy only earns its keep if it's live where decisions get made, not laminated on a wall near the break room. That means baking origin-based defect categories into the recipe notes, into the changeover checklist, and into whatever the operator sees on first ware at ignition. Our forming audit starts by mapping exactly this, plant by plant, line by line, before we touch a single set point.

Zaid Hassoneh, who founded Lean Glass after two decades on the hot end including the $220M USD Arglass Yamamura greenfield build in Georgia, puts it bluntly: most plants can name a defect but can't place it. That gap is what a vendor-neutral container glass consultant gets paid to close, and it's cheaper to fix on paper than in a Pareto chart nobody trusts.

Look, a taxonomy won't fix the furnace. It just tells you faster which fix to make. If your defect board still reads like it was built for a car door, it's worth an hour with our glass glossary to see how far off the naming convention actually is.