SMEDSingle Minute Exchange of Die

SMED is a structured method for reducing changeover time — the elapsed time from the last good unit of product A to the first good unit of product B. "Single minute" is the original aspiration: changeover under 10 minutes (single-digit minutes). Modern shops routinely hit it, and OTED (One-Touch Exchange of Die, under 1 minute) is achievable for many discrete operations.

The technique is mechanical, not mystical. Every changeover step is observed, timed and classified. Internal steps (machine must be stopped) are the cost; External steps (machine still running) are free. The work is to make as much of the changeover External as possible, then make the remaining Internal as fast and standardised as humanly possible.

Shingo's original method moves through four stages, each delivering a substantial percentage reduction.

Regulated environments — pharma, medical devices, supplements, food, cosmetics — add Internal work that cannot be eliminated: line clearance, cleaning validation, allergen changeover, calibration verification, two-person sign-offs. Naive SMED treats these as immovable; correct SMED treats them as the bottleneck and engineers around them.

1. Parallelise line clearance with External preparation — one operator clears the line + does the witnessed sign-off (Internal); another stages the next batch's components, labels, line-side materials and pre-stages the next-product changeover kit (External).
2. Convert cleaning from sequential to parallel — quick-disconnect product-contact parts moved offline to a wash bay; clean SIP-validated change parts pre-staged at line-side; the line itself only needs minimal CIP between batches.
3. Modular tooling validated as a family — instead of revalidating every die / mould / change part, qualify the family of interchangeable parts so the changeover is a part-swap, not a re-qualification event.
4. Pre-staged calibrated instruments — verification weights, reference standards, IPQ test materials staged External; on changeover the operator scans them in instead of fetching + checking calibration mid-stoppage.
5. Two-person sign-offs scheduled, not sequenced — the second-person witness is paged to the line before the changeover begins so they're physically present when the Internal step requiring witness occurs.
6. Validated cleaning recipes by sequence — campaign sequencing (low-to-high allergen, low-to-high colour, low-to-high potency) reduces the cleaning requirement at every transition; the worst-case cleaning happens once per campaign, not once per batch.

The single most common SMED failure is sloppy classification. "It feels like Internal" or "it's always been done during the stoppage" are not classifications; they are habits. The test is mechanical:

• Could this step be done while the previous batch is still running on the line? If yes → External, full stop.
• Could this step be done after the next batch has started running? If yes → External (post-changeover finish-up).
• Does this step physically require the machine to be stopped + de-energised + safely accessible? If yes → Internal.
• Does this step require a regulated act (witnessed line clearance, two-person e-sig at the equipment) that can only happen at the cross-over moment? If yes → Internal (and unavoidable).

Baseline videos routinely reveal 30-50% of "Internal" work failing the first test — operators fetching tools, walking to the warehouse, looking for paperwork, hunting for the next die. None of this requires the machine to be stopped. Pre-staging it as External (stage 1) typically buys back 30-50% of changeover time for almost zero capital cost.

Once you've moved everything possible to External, the remaining Internal time is engineered down through specific techniques — most of them cheap and reversible.

SMED progress is measured with specific, comparable KPIs that feed directly into OEE and ISO 22400.

• Mean changeover time per product transition (minutes) — the headline number; segment by product family + line.
• Internal vs External time split — track the ratio, not just total; a 30-minute changeover with 5 min Internal + 25 min External is dramatically more flexible than 30 min all-Internal.
• Changeover time variance — small mean with low variance is a real SMED win; small mean with high variance means tribal knowledge, not a standardised method.
• Changeovers per shift — frequency rises as SMED matures (because small batches become viable); rising frequency at flat-or-declining total changeover time is the strongest SMED signal.
• Right-First-Time-After-Changeover — fraction of post-changeover runs that produce in-spec product on the first unit; SMED can sacrifice this if poka-yoke is neglected.
• Setup-related deviation / NCR rate — regulated KPI; SMED done badly causes a spike in setup-attributable deviations.
• OEE Availability lift — SMED reduces planned + unplanned changeover-driven downtime; track Availability before vs after each kaizen event.

Mistake 1 — engineering without measuring

Buying quick-release tooling for a step that's already External, or pre-heating moulds that were never the bottleneck. Always video + classify first; engineer the actual bottleneck.

Mistake 2 — celebrating External-only wins

Cutting total changeover from 60 min to 30 min by moving 30 min to External feels great but doesn't help if the Internal portion is still 30 min — minimum batch size is unchanged. Internal reduction is the real win.

Mistake 3 — SMED at the expense of right-first-time

Skipping verification steps, rushing calibration, omitting poka-yoke to hit a changeover-time target produces a spike in post-changeover defects. The full picture is changeover time × (1 / post-changeover RFT).

Mistake 4 — frozen standard with no re-kaizen

First SMED event delivers 60% improvement; teams celebrate + walk away. The standard then drifts as operators retire, equipment changes, new products are introduced. SMED is a continuous discipline — annual re-baseline at minimum.

Mistake 5 — applying discrete-SMED logic to process changeovers without cleaning validation

Pharma + food cleaning between products is regulatorily validated. Compressing it by 30% without revalidation is a CAPA + 483 in one. Cleaning-cycle SMED requires its own validation pack.

Mistake 6 — single-operator changeover when parallel is feasible

Two operators choreographed often halve Internal time. The cost is labour reallocation, not capital; the constraint is usually mindset, not physics.

V5's changeover module instruments SMED end-to-end: video-baseline ingestion, Internal/External classification, kaizen-event tracking, line-clearance parallelisation, and tight wiring into OEE + CAPA so SMED gains stick instead of decaying.

• Changeover module on every WO — when a WO completes + the next WO is queued, the changeover is a tracked event with its own start/end, operator(s), Internal-vs-External split, deviation log + RFT-after-changeover metric.
• Video + classification workflow — supervisor uploads a baseline changeover video, walks through the timeline tagging each step Internal/External; system produces the SMED scorecard + identifies the largest-bottleneck Internal step automatically.
• EWI for changeovers — standardised work-instructions render on the kiosk at changeover-start; two-operator choreography supported with role-split step assignment; line clearance pre-staged the moment the previous WO closes.
• Pre-staging hooks — scheduler fires External-preparation tasks (warehouse pick of next batch components, labels, line-side materials, pre-heat/pre-calibrate of change parts) the moment the current WO crosses a configurable preparation threshold (typically 80% complete).
• Regulated parallelisation — line clearance + two-person e-sig run with full Annex 11 §12 audit trail in parallel with External staging; the kiosk surfaces which steps are Internal-blocking vs External-still-OK live.
• KPI feed — every changeover event publishes total time / Internal time / External time / Internal-External ratio / RFT-after-changeover into the OEE Availability + ISO 22400 changeover-time KPIs; trends visible on supervisor dashboards.
• Kaizen-event landing — SMED improvement projects are first-class records with before/after baseline, hypothesis, intervention, validation period + retained gain; they auto-link to the CAPA stream when they originate from a deviation cluster.
• Per-product-pair changeover library — system learns the typical changeover time + Internal split for every product-A→product-B transition + uses it to drive scheduler-feasible sequencing (campaign sequencing for cleaning, allergen-aware ordering).
• Mobile-safe — operator + supervisor changeover views work at iPhone (≤390 px CSS width) with no horizontal scroll; floor + office see the same live Internal/External clock.

Is the "single minute" literal?

Originally aspirational. Shingo's target was single-digit minutes (under 10), not exactly 60 seconds. OTED (One-Touch Exchange of Die, under 1 min) is achievable for many discrete operations; full sub-minute changeover in regulated batch processes is rare because of unavoidable Internal validation steps.

Does SMED apply to pharma + food + supplements?

Yes — and arguably the ROI is higher than in discrete because every minute of changeover in a validated cleanroom is more expensive. The technique is the same; the regulated overlay (validated cleaning, witnessed line clearance, two-person e-sig) constrains how much Internal time can be removed but does not remove the SMED logic.

Can we SMED-down validated cleaning between batches?

Only via a revalidation campaign. The cleaning recipe + cycle time are validated; compressing them without revalidation is a CAPA finding. The right SMED move is to parallelise (offline quick-disconnect parts cleaned in a wash bay while the line runs the next batch) and to sequence campaigns to minimise worst-case cleaning frequency.

What's a realistic SMED target for a regulated batch line?

30-60% changeover reduction in 6-12 months from a first kaizen event is common; 70%+ requires structural changes (modular tooling, parallel setup stations, validated quick-disconnects). The diminishing-returns curve flattens around 70-80% of the original baseline.

How does SMED interact with kanban + heijunka?

SMED is the prerequisite that makes them work. Small kanban quantities + small heijunka pitches require frequent changeovers; if changeover is expensive, the productivity gain is destroyed and the experiment fails. SMED first, kanban + heijunka second.

Does SMED need new equipment?

Stage 1 (External separation) is usually free — pure work redesign. Stages 2 + 3 often need modest capital (quick-release clamps, modular tooling, pre-staging carts, duplicate change parts) — typically a few thousand dollars per line for substantial returns. Full structural redesign (modular machines designed for changeover) is a major-project decision.

How does V5 instrument SMED?

Every changeover is a tracked event with start/end, Internal/External split, RFT-after-changeover and deviation log; video-baseline classification workflow is built in; KPIs feed OEE Availability + ISO 22400 changeover-time; kaizen events are first-class records with retained-gain tracking; the changeover library drives scheduler-feasible campaign sequencing.