HeijunkaHeijunka (平準化, levelling)

Heijunka is the practice of taking variable customer demand and re-sequencing it into a steady, repeating mix that the plant floor can run to. Customer orders rarely arrive in a smooth daily pattern; demand has spikes, dips, mix swings and weekend lumps. Heijunka absorbs that variability at the scheduling layer so the line sees a smooth, repeating pattern of small mixed batches — the so-called pitch — rather than the raw, jagged order book.

Without heijunka, takt time is theatre: a line that builds three products with one piled into the first 2 hours and the other two into the last 6 is not running to takt — it is batch-running with a takt number stamped on top. Without heijunka, kanban is fragile: a pull signal arriving at random will trigger random changeovers and consume the entire SMED budget on changeover-of-the-day rather than productive output. Heijunka is the discipline that makes the rest of the Lean toolkit actually work.

Heijunka levels along two axes: volume (the total quantity per period) and mix (the product variety per period).

Volume levelling

Smoothing the per-day or per-shift total quantity. If the order book for the week is 1,000 / 1,400 / 600 / 1,000 / 1,000 = 5,000 units, volume levelling builds 1,000 per day. The total is identical; the daily commitment to the line is steady. This protects against capacity overspend on Tuesday and idle capacity on Wednesday.

Mix levelling

Smoothing the product variety within each period. If daily demand is 500 of A + 300 of B + 200 of C = 1,000 units, mix levelling does NOT build 500 A then 300 B then 200 C. It builds the smallest practical pitch — typically AAABBBC repeated, or even ABABABCABC if changeover is cheap — so that A, B and C all flow through the day. This is the dimension that has the highest payoff and the highest difficulty.

Two related concepts give heijunka its operational vocabulary: pitch and EPEI.

Pitch

The smallest unit of work that you release together. Often expressed as a time interval ("a 30-minute pitch") that contains a small mixed bundle of products. The pitch becomes the unit of progress on the pitch / hour-by-hour board — "by the end of pitch 4 we should be at 240 units".

EPEI — Every Product Every Interval

The frequency with which the full product mix cycles through the line. EPEI = 1 day means every product gets built every day; EPEI = 1 shift means every product gets built every shift; EPEI = 30 min means every product gets built within every 30-minute window. Smaller EPEI means smoother flow, smaller in-process inventory, and faster response to demand changes — but requires more changeovers (SMED matters more) and more disciplined scheduling.

The right EPEI is product- and plant-specific. Toyota famously targets EPEI = 1 vehicle (every car on the line is the next car the customer ordered). Most regulated plants land at EPEI = 1 shift or EPEI = 1 day; pharma packaging often EPEI = 1 day; sterile fill-finish often EPEI = 1 week because of the long cleaning + setup cycle.

The classical visualisation tool is the heijunka box (heijunka post box) — a physical grid with a row per product and a column per pitch interval. Each cell holds a card representing one pitch's worth of that product. The scheduler distributes the day's demand across the grid so each row's cells are spread evenly through the day rather than bunched together. The pacemaker pulls the next card every pitch interval and releases the corresponding work to the line.

Digital implementations replace the physical grid with a scheduler that publishes the next pitch to the kiosk and the pitch board; the principle and the discipline are identical. The visualisation matters because anyone walking the floor can see at a glance whether the schedule is genuinely levelled or whether one product is hogging the grid — a check that is much harder to do against a Gantt or a spreadsheet.

In a multi-step value stream, heijunka is applied at one designated point — the pacemaker process. The pacemaker is the only point in the stream that is scheduled; everything upstream pulls via kanban, and everything downstream flows in FIFO. The pacemaker is usually as close to the customer as practical — typically packaging or final assembly — so the schedule reflects actual finished-good demand, not raw-material pull.

Choosing the right pacemaker is one of the highest-leverage value-stream-mapping decisions. Too far upstream and the levelling does not propagate to finished goods; too far downstream and the changeover constraints make levelling impossible. In practice the pacemaker is where the value stream first becomes constrained by changeover — typically the packaging line in pharma, the final-assembly cell in devices, the filling line in food.

Regulated plants — pharma, supplements, devices, food — frequently dismiss heijunka as a discrete-automotive concept that does not apply to batch process. They are partially right and largely wrong.

1. Partially right — regulated batch process has constraints automotive does not: campaign cleaning windows, allergen sequencing rules, validated changeover cycles, equipment qualification dependencies. These constrain how small the pitch can be.
2. Largely wrong — within those constraints heijunka still applies. Packaging lines, blister lines, fill-finish lines, capsule lines, soft-gel encapsulation lines, snack-food packaging lines, kit-pack stations all benefit from mix levelling at shift / day EPEI. The benefits — smaller WIP, faster customer response, lower inventory, better OEE Performance, smoother quality outcomes — are unchanged.
3. Allergen sequencing is the key constraint — food / supplement plants must sequence from low-allergen to high-allergen between cleaning events to minimise cleaning burden. Heijunka has to respect the allergen graph; the scheduler computes the levelled sequence within the allergen constraints, not against them.
4. Validated changeover cycles likewise constrain EPEI but do not eliminate the discipline — a sterile line at EPEI = 1 week is still levelled in the sense that the product mix is spread evenly across the week's batches, not piled into a single product run.

Mistake 1 — levelling volume but not mix

Smoothing daily totals while building each product in one big block within the day is the half-step that captures volume benefits but leaves all the WIP, lead-time and quality wins on the table. Mix levelling is the dimension that matters.

Mistake 2 — adopting heijunka without SMED

Mix levelling at small EPEI multiplies changeover frequency by 5–20×. Without SMED the changeover cost destroys productive output and the experiment fails. Sequence: prove SMED first, then push EPEI down.

Mistake 3 — picking the wrong pacemaker

Levelling at a process step deep in the value stream does not propagate to finished goods. The pacemaker should be the last point the schedule influences before the customer-facing finished pack.

Mistake 4 — ignoring the allergen / sequencing constraint

In food / supplement plants the scheduler must respect the allergen graph: low → high between cleaning events. A naive mix-levelling algorithm that interleaves products without that constraint will trigger constant cleaning + line-clearance cycles that destroy throughput. The scheduler must encode the constraint.

Mistake 5 — heijunka box as a wall poster

The heijunka box only works if the pacemaker actually pulls the next card every pitch. Without that discipline it becomes a wall decoration. The MES has to enforce the pitch interval, not just visualise it.

Mistake 6 — levelling the schedule but not the supply

A levelled production schedule fed by lumpy raw-material deliveries trades production lumps for material lumps. Heijunka must extend to the receiving dock — supplier kanban + JIT replenishment + supplier scorecard for delivery cadence.

V5 ships heijunka as the default scheduling discipline — demand → levelled per-pitch WO releases → pitch-board → kiosk — with allergen-graph and changeover-constraint awareness.

• Heijunka-aware scheduler — demand entered at family-day / family-week granularity is levelled into per-pitch WO releases respecting takt + allergen graph + validated changeover cycles + equipment availability.
• Pitch interval configurable per line — common settings 15 min / 30 min / 1 hour / 1 shift; the scheduler computes EPEI automatically and surfaces it as a metric on the line dashboard.
• Pitch board /app/pitch — live planned-vs-actual at the pitch interval; planned units = takt × elapsed in pitch; actual = good count from kiosk; gap + andon-overlay drives intra-shift correction.
• Allergen-graph enforcement — for food / supplement tenants, the levelled sequence respects the configured allergen graph; the scheduler will not interleave high-allergen and low-allergen products in a way that triggers an unplanned line-clearance.
• Kiosk next-pitch preview — the operator sees the next pitch's product + quantity + estimated finish time before they start, so changeover preparation runs in parallel (External setup per SMED).
• WO queue = digital heijunka box — the WO queue view is rendered as a heijunka grid (product rows × pitch columns) when the workspace runs in lean mode; supervisors can rebalance by drag-and-drop within the constraints.
• OEE / 22400 link — schedule-adherence + throughput-rate KPIs are computed from the same pitch stream — no parallel data capture.
• Mobile-safe — pitch board + WO queue work on iPhone (≤390 px CSS width) with no horizontal scroll; the supervisor walking the floor sees the same heijunka grid the kiosk does.

Does heijunka apply to batch process manufacturing?

Yes — within the constraints of campaign cleaning, allergen sequencing and validated changeover cycles. The pitch will be longer (1 shift / 1 day rather than 30 min), but the discipline of spreading product mix across the planning horizon — rather than piling each product into one block — still applies and still pays off.

Heijunka vs takt time — what's the difference?

Takt is the pace per unit derived from demand; heijunka is the scheduling discipline that ensures the line actually runs to that pace by spreading the mix evenly across the period. Without heijunka, takt is a number on a dashboard the line cannot honour. Without takt, heijunka has no target to level against. They are complementary.

What's the right EPEI for my plant?

Start with the natural cleaning + changeover cycle. If a full clean takes 8 hours, EPEI < 1 shift is uneconomic — start with EPEI = 1 day or longer. As SMED progresses and changeover time drops, push EPEI down. Most discrete plants converge to EPEI = 1 shift; pharma packaging to EPEI = 1 day; sterile fill-finish to EPEI = 1 week.

Do I need a physical heijunka box?

No — digital heijunka grids in the scheduler deliver the same discipline. A physical box adds value when the floor culture benefits from a tactile, visible artefact; many plants keep a small physical board on the line for the pacemaker step even when the master schedule lives in the MES.

Doesn't heijunka conflict with the customer wanting it now?

On the contrary — small EPEI reduces customer lead time because every product is built within every pitch interval. A spiky customer order arriving on Tuesday for a heijunka-levelled plant ships from finished-goods or builds in the next-day pitch; the same order at a batch-running plant waits for the next product run, which could be 2 weeks away.

What's the SMED prerequisite for heijunka at small EPEI?

Rule of thumb: changeover time < 10% of pitch interval. If pitch = 30 min, target changeover < 3 min. If pitch = 1 shift (8 h), changeover < 50 min. If you cannot hit those numbers, the EPEI you targeted is too small and the discipline will fail.

How does V5 publish heijunka to the line?

Demand is entered or imported by family × period → scheduler levels into per-pitch WO releases respecting constraints → kiosk shows next pitch + after-next pitch → pitch board shows planned-vs-actual live → andon thresholds enforce the pace. Every change to demand or constraints triggers an automatic re-level; the floor always sees the current truth.