Home / Insights

Warehouse layout series · Part 4 of 5

Central Store, Supermarket or Kanban? Feeding the Production Line

Contents
  1. Three ways to feed a line
  2. The plant
  3. The Layout Lab
  4. What the simulation shows
  5. What each change contributes
  6. Sizing the milk run
  7. What the waiting costs
  8. What the shift does not show
  9. How to check your own plant
  10. What the model leaves out
  11. Key points
  12. Sources

Parts 1, 2 and 3 laid out a distributor's warehouse. A manufacturer has a second customer inside the building: the production line. When a line runs out of a material, it stops, and every minute it waits is lost output. In OEE terms it is an availability loss (see what OEE really measures), and it often starts with the layout: where the materials are kept, and how they reach the line.

This part follows Ridgeline Coatings, the fictional Vancouver coatings maker from the ABC analysis for manufacturers post, through one shift with three ways of feeding its two filling lines.

Three ways to feed a line

Table 1. The three set-ups

Set-upWhere the materials waitAt the lineHow a refill is triggered
Central storeIn the central store with everything else, laid out by part numberOne load of each materialThe operator calls a material handler when a load runs out; the store issues the load
Line-side supermarketA small rack between the lines, holding a few loads of each line material, topped up from the store by forkliftOne load of each materialThe same: called when a load runs out
Kanban milk runThe same supermarketTwo loads (two bins)When the first load is used up, a card goes up; a handler on a fixed 20-minute loop delivers every carded load

A supermarket, in lean terms, is a small store owned by production and placed next to the lines it serves, from which material handlers take what the lines use. The two-bin system needs no cards at all: the empty bin is the signal, and the second bin keeps the line running while the first is refilled. A milk run is a handler who follows a standard route on a fixed timetable, for example every 20 minutes, collecting the signals and delivering the materials.

The plant

The building is the 72 by 52 m building of the earlier parts, with the same doors and staging. The west third is the central store: about 330 pallet faces, each with several levels above. In the east, two mixing tanks feed two filling lines, each filling 3 pails a minute at standard speed. Each line uses five materials: resin and pigment at the mixer, and pails, lids and labels at the filler. One load (a drum of resin, a pallet of pails, a box of label rolls) lasts 40 to 150 minutes of running, and a line stops when any one of its five materials runs out.

  • Two material handlers walk at 0.9 m/s with a pallet jack or cart: 20 seconds to start a job, 40 seconds to pick up a load, 30 seconds to set it down at the line. At the central store, booking the issue of a load adds 2 minutes.
  • One forklift puts away inbound pallets (6 trucks a day) and, with a supermarket, brings 3 more loads from the store when a supermarket face is down to 2.
  • Finished pails are palletised at the end of the lines (36 to a pallet) and taken to shipping by pallet jack; carriers collect every hour.
Central storeFilling line AFilling line BMixingSupermarketPalletisingReceiving stagingShipping stagingThe only area thatchanges: none inthe central-storeset-up; a doublerack in the other two
Figure 1. Ridgeline's plant. Mixing, the filling lines, palletising and the central store are the same in all three set-ups; only the supermarket changes.

Mixing and palletising are part of the plant in every set-up: the mixers use the resin and pigment, and palletising stacks the finished pails for shipping. They are drawn so you can follow the materials and the output, but the comparison is about one area, the supermarket between the lines, and about when a refill is asked for.

The Layout Lab

The simulator runs the three set-ups through the same shift. The supermarket area is shaded amber where it exists and outlined in dashes where it does not. Each line's outline turns red when it is waiting for material, and the loads waiting at each line are drawn next to it: one in the first two set-ups, two with kanban. Change the line speed to see how each set-up copes when the lines run faster or slower.

Figure 2. The Warehouse Layout Lab with Ridgeline's three set-ups. The animation replays the shift 12 times faster than life; the table is the whole simulated shift.

Open the Layout Lab on its own page for a larger view.

What the simulation shows

Table 2. One shift at standard line speed

MeasureCentral storeLine-side supermarketKanban milk run
Line time lost waiting for material (both lines)237 min118 min0 min
Availability, counting material losses only75.3%87.7%100.0%
Pails filled2,1692,5252,880
Material deliveries to the lines475661
Call or card to delivery (average)5.0 min2.1 min15.0 min
Material handler hours4.32.41.9
Handler walking5.0 km3.3 km1.9 km
Stock at the lines (average value)$4,400$4,500$11,900
Stock in the supermarket (average value)None$36,600$36,500
0120240360480600720840960Line minutes in the shift (two lines, 960 in all)Central store75%Line-side supermarket88%Kanban milk run100%RunningWaiting for material
Figure 3. Line minutes running and waiting for material. Only material losses are counted; breakdowns, changeovers and slow running would come on top.
  • From the central store, the lines wait 237 minutes a shift, about a quarter of the time. Every run-out stops a line for the whole refill: the call, the walk to the store, the issue, the walk back. That takes 5.0 minutes on average, and it happens 47 times a shift across the two lines.
  • A supermarket halves the waiting, but does not remove it. A refill takes 2.1 minutes and the lines wait 118 minutes. Most of the gain comes from skipping the store's issue booking and its part-number layout, not from the shorter walk (see the next section). The line still stops at every run-out, because nobody is asked for material until it has run out.
  • Two bins and a milk run remove the waiting here. The card goes up when the first load is used up, so the line runs on the second load while the milk run comes round. This assumes steady use and cards seen as soon as they go up; see sizing the milk run. Cards wait 15.0 minutes on average for delivery, and the lines never notice. Across ten other simulated days the lines waited 232 to 249 minutes with the central store, 116 to 122 with the supermarket and 0 with kanban.
  • The milk run also needs less handling. One loop delivers every carded load, so the handlers work 1.9 hours against 4.3 from the central store, and walk 1.9 km instead of 5.0 km. All three set-ups keep two handlers; the time is freed for other work, not saved as cash, and a timed milk run ties one person to its timetable.
  • The price is stock and space at the line. Two bins hold 2.6 times as much at the lines: $11,900 on average against $4,500. The supermarket holds about $36,500, but that stock moves out of the central store; it does not have to be extra if the store's reorder points are lowered to match.

What each change contributes

The three set-ups change several things at once: where the materials are, how the store is run, how many loads wait at the line and what triggers a refill. To separate them, the model was also run with one change at a time, averaged over ten simulated days.

Table 3. One change at a time (ten-day averages; rows 2, 3, 5 and 6 are model checks, not set-ups in the Lab)

Set-upLine minutes lost a shiftHandler hours
Central store, as set up2404.4
Central store, no 2-minute issue booking1593.1
Central store, no issue booking, line materials in the store faces nearest the lines1282.6
Line-side supermarket, called when a load runs out1192.4
Central store, two bins at the line, called when the first runs out05.8
Supermarket, two bins at the line, called when the first runs out02.8
Supermarket, two bins and a 20-minute milk run01.9
  • The trigger matters most. With two bins at the line and a call when the first runs out, the lines stopped waiting even when every load still came from the central store. Calling before a material runs out, not where it is kept, removes the waiting.
  • The store's way of working costs more than its distance. Booking each issue (2 minutes) and keeping line materials in part-number order account for most of the central store's extra waiting. With both removed, the central store loses 128 minutes against 119 for the supermarket.
  • The supermarket and the milk run cut the handling. Two bins fed from the central store need 5.8 handler-hours a shift; from the supermarket, 2.8; with the milk run, 1.9. Note that in the central-store set-up the handlers are idle most of the shift while the lines wait: the problem there is calling late, not a shortage of people.

Sizing the milk run

Kanban works here because every load lasts longer than the time it can take to replace it. The worst case is a card that goes up just after the milk run has left: it waits a full loop, then the handler collects and delivers. So the rule is

\[t_{\text{load}} > t_{\text{loop}} + t_{\text{deliver}}\]

where \(t_{\text{load}}\) is how long one load lasts at the line, \(t_{\text{loop}}\) the time between milk runs and \(t_{\text{deliver}}\) the time to collect and deliver. The load that runs out first sets the limit: here a drum of resin lasts 40 minutes.

060120180240300102030405060Minutes between milk runsLine minutes lost a shiftOne drum of resin lasts 40 min
Figure 4. Line minutes lost with the kanban milk run, as the time between runs grows. Once a run takes longer than a drum of resin lasts, the lines start to wait.

With loops of 10 to 30 minutes the lines lost nothing. At 40 minutes the loop is right at the limit: the lines happened to lose nothing on this day, but one unlucky card would stop a line. At 50 and 60 minutes they lost 134 and 256 minutes. Line speed works the same way: at 200% speed every load lasts half as long. The central store then loses 382 minutes a shift and the supermarket 210. Kanban loses 0, but the 20-minute loop is now at its limit: in the speed sweep it lost 21 minutes at one speed. When lines speed up or loads get smaller, shorten the loop or add a bin.

The model also assumes the milk run sees every card posted before it leaves, as with an electronic signal or a light. With paper cards picked up on the round, a card can wait a whole extra loop, so the rule becomes \(t_{\text{load}} > 2\,t_{\text{loop}} + t_{\text{deliver}}\), and a 20-minute loop is too long for a 40-minute drum. In the model, cards seen one loop late cost 2 minutes a shift at standard speed, 289 at 150% and 448 at 200%.

What the waiting costs

If Ridgeline makes up the lost pails with overtime, each line-hour lost costs a crew of 3 operators at time and a half:

\[\text{Overtime a year} = h \times c \times w \times 1.5 \times d\]

where \(h\) is line-hours lost a day, \(c\) the crew per line, \(w\) $40 an hour loaded and \(d\) 250 days a year. For the central store that is 3.95 × 3 × $40 × 1.5 × 250 = $178,000 a year.

Table 4. What each set-up costs a year (Canadian dollars)

ItemCentral storeLine-side supermarketKanban milk run
Line-hours lost a day (both lines)3.92.00.0
Overtime to make up the lost pails, a year$178,000$89,000$0
Material handler time, a year (time freed, not cash unless reused)$43,000$24,000$19,000
Extra stock at the lines against line-side, carrying cost a year––$2,000
Supermarket racking, one-offNone$5,000 to $12,000$5,000 to $12,000
  • Moving to a supermarket saves about $89,000 a year in overtime, for $5,000 to $12,000 of racking (34 faces on two levels, 68 positions, at the Part 3 price range; a material can hold up to 5 loads across its positions).
  • Adding two bins and a milk run saves about $176,000 a year in overtime against the central store, after the carrying cost of the extra stock at the lines (at 25% a year). Handler time falls too, but it only becomes cash if the freed hours replace other paid work.
  • If the lines are sold out, the cost is lost sales, not overtime. At an illustrative $8 contribution a pail, the 355 pails a day the line-side set-up loses against kanban would be worth $2,800 a day. See what OEE losses cost for how to value lost output.

What the shift does not show

  • Shelf life and lot control. Coatings materials have expiry dates and lot numbers. A supermarket must still be used first-expired, first-out, and every load moved to the line must stay traceable.
  • Inventory records. Stock that moves to a supermarket or the line without a transaction is how record errors start. Decide where the stock is booked: issued to the line when it leaves the store, or when it is used.
  • Batch changes. When a line changes colour or product, the line-side loads change too. Kanban works best for the materials every batch uses (pails, lids, labels); batch-specific materials are often kitted for each batch instead.
  • Space and safety at the lines. Two bins and a supermarket need floor space next to the lines, clear of walkways and fire exits, with forklift traffic kept away from operators.
  • Other losses. Breakdowns, changeovers and slow running are not modelled. A line that never waits for material can still have a low OEE.
  • The receiving dock. Topping up the supermarket takes forklift time ahead of inbound put-away, so pallets wait a little longer at receiving: 2.4, 4.0 and 4.7 minutes on average.

Table 5. Choosing how to feed a line (rules of thumb; the model checks inform the first three rows)

If this is trueStart with
Lines stop at every run-out while a refill is fetchedTwo bins, so the signal comes before the line runs dry, wherever the material is kept
Each refill from the store takes long because of paperwork or a part-number layoutA supermarket next to the lines for the materials every batch uses, issued in bulk to it
Handlers spend their day walking separate refillsA timed milk run, with the loop sized for the load that runs out fastest
Materials are specific to each batch or productA kit for each batch, delivered before the batch starts
Floor space at the line is very tightSmaller loads and a shorter loop, rather than more stock at the line

How to check your own plant

  1. Log every stop for a week with its cause, and count the minutes waiting for material.
  2. Time a refill from the moment a material runs out to the moment the line restarts.
  3. List the materials each line uses every batch, how long one load lasts, and where each one is stored today. Use the criticality rating to find the cheap items that can stop a line.
  4. Size the loop from the load that runs out fastest, and check that one load lasts longer than the loop plus the delivery time.
  5. Try it on one line with a temporary rack and cards before buying racking for every line.

What the model leaves out

  • One shift with fixed times, two lines and five materials a line; real lines use many more items, often in boxes rather than pallets.
  • Material is always in stock in the store; supplier shortages (see safety stock for manufacturers) are a separate problem.
  • Loads last a fixed time at a given speed; no scrap, no batch changes, no breakdowns.
  • Calls are answered in order by the first free handler; in a real plant calls are often noticed late.
  • The milk run sees every card posted before it leaves; with paper cards collected on the round, add one loop to the delivery time.

Key points

  • A line waiting for material is an availability loss, and where materials are kept decides how long each wait is.
  • Calling before a material runs out, with a second load at the line, removed the waiting in the model wherever the material was kept.
  • Here, a supermarket halved the waiting from 237 to 118 minutes, mostly by skipping the store's issue booking and part-number layout.
  • Two bins and a timed milk run kept the lines fed with the fewest handler hours, for 2.6 times the stock at the lines.
  • Size the milk run from the load that runs out fastest: the loop plus delivery must be shorter than that load lasts.

Earlier in this series: Part 1, what a bad layout costs; Part 2, flow patterns; Part 3, aisle widths. Next: Part 5, from layout to cash: inventory turns and a layout payback calculator. Related: what OEE really measures and ABC analysis for manufacturers.

Sources

Ridgeline Coatings, its plant, materials, times and costs are an illustrative example built from stated assumptions, and the simulator is a simplified model. It shows how much the way materials reach a line can matter; it does not predict your plant's losses. Measure your own lines' waiting time before you change anything.

Keep reading

Related articles

All 186 articles

Ready to find your next improvement?

Book a strategy call. We will talk through your goals, pain points and where AI, systems or process work can make the biggest difference.