A field guide to scrap: what your SCRAP notes are trying to tell you

Every shop writes them. Almost no shop reads them.

The operator scraps a tool. They write a short note on the traveller or in the crib log: "chipped, mid-run" or "worn, normal" or "broke first cut, bad insert". The note goes into a manila folder, or gets typed into a CribMaster field, or maybe a shared spreadsheet. Six months later, the notes have accumulated but nobody has read them together.

This is the single most under-read data source on an NZ precision shop floor. What follows is a field guide to what those notes contain and what patterns they reveal when you do read them together.

The four classes every note belongs to

After several hundred reviews, we've found that SCRAP notes cluster into four operationally meaningful classes:

1. Normal wear

The tool hit end of life doing what it was supposed to do. Phrases: "worn", "normal EOL", "tool life reached", "end of run". These are the notes that feed life-prediction models; everything else is noise against that question.

2. Breakage

The tool failed before its expected life, typically from mechanical stress. Phrases: "chipped flute", "broke on first cut", "insert cracked", "snapped mid-cycle". These flag tool-quality or setup issues — never life-prediction inputs.

3. Geometry / regrind

Tools leave service not because they're worn out but because someone reground them, switched them out for a specialty geometry, or pulled them for inspection. Phrases: "regrind", "pulled for inspection", "sent for reshape". These should register as tool-out but not tool-scrapped, and the population of reground tools should come back into the crib as new tools with their own service windows.

4. Other / ambiguous

The note doesn't parse cleanly into the first three classes. Phrases: "looked odd", "supervisor asked", "see note in logbook". Operationally these should be flagged for review by the supervisor — the note wasn't definitive, which means the scrap event wasn't either.

A good classifier should achieve 80–85% accuracy on the first three classes across typical NZ shop-floor vocabulary. The remaining 15–20% is "other" — you don't try to classify harder, you flag it.

The four patterns that matter

Once every note is classified, four patterns surface from the data.

Pattern 1: Machine-specific breakage cluster

A 5-axis machine produces disproportionate breakages relative to its part mix. The cluster persists across operators, across alloys, and across tool brands. Something is wrong with the machine — most commonly a spindle calibration drift, sometimes a coolant concentration issue, occasionally a worn toolholder.

This is the single highest-value pattern Niyamis TMS surfaces because the fix is mechanical and cheap (spindle calibration is a 2-hour job) but the damage in scrap cost compounds for months before anyone looks at it.

Pattern 2: Operator-specific wear cluster

A specific operator's tools wear faster on average than their peers' — but don't break more often. The cause is usually one of two things: the operator is running hotter speeds/feeds than the program specifies, or they're running programs optimised for a different tool brand than the one currently in the crib.

Both fix with training, not discipline. The data surfaces the pattern; the supervisor holds the conversation.

Pattern 3: Brand-specific breakage on a specific alloy

One vendor's tools consistently break during titanium or stainless runs while performing fine on aluminium. This is a metallurgy-and-geometry mismatch; the vendor's coating or grade isn't the right choice for that alloy. The fix is a brand shift on that alloy, kept in the Brand Scorecard's per-alloy view.

Pattern 4: Seasonal drift in "other / ambiguous" notes

Every shop has quarters where the "other" bucket doubles. Usually those correlate with operator-training transitions, new-hire onboarding, or a month where the supervisor was off-floor. The data itself tells you when supervisor attention slipped — which is useful for the supervisor's own retrospective.

What you need to start reading the notes

Three ingredients, all lightweight:

1. A consistent place for the note to land. Logbook, CribMaster field, Niyamis TMS SCRAP event — any of these work, as long as every scrap has a note.

2. A classifier. Either a trained human (the supervisor, for the first 500 notes) or a simple pattern-matching ruleset. TMS ships with a ruleset trained on several thousand NZ-shop SCRAP notes; it runs at ingest time and flags anything it can't classify.

3. A review cadence. Weekly, fifteen minutes. Filter to the "breakage" and "other" classes only. Skim, look for clusters, hold the conversation.

That's the whole discipline. It's less work than most shops think, and it produces the single biggest improvement in scrap rate we've seen on deployment.

Where Niyamis TMS fits

TMS's Scrap Anomaly Detection runs the classification at ingest and surfaces the four patterns above continuously. Shops that run it for a full quarter typically reduce scrap rate by 0.5–1.5 percentage points and, more importantly, find the hidden mechanical issues on their floor that nobody had spotted in months of production.

If you want to see what this looks like on your data, the Diagnose week includes a scrap-note analysis on your last eighteen months of notes before any software goes in. Book a discovery call.

— Philippe, Niyamis