Cycle or Chase
Why a fixed cycle recovers & re-planning does not
When stock runs short, the natural response is to rebuild the schedule around it. Each rebuild costs changeover time & pushes other work back, which sets up next week's shortage. That is how a one-week disruption keeps a plant re-planning for a month.
This page runs one bottling line under two scheduling rules. The first is a fixed cycle that works like a train timetable: if a run is missed, the next slot for that product is already known. The second re-plans each week around whichever products are closest to running out. Both get the same orders, the same hours & the same disruption.
The line makes 100 SKU - stock-keeping unitOne specific sellable item. A 750 ml bottle & a 1 litre bottle of the same liquid count as two, because each is stocked, counted & ordered separately.SKUs with a typical Pareto profileA small share of items accounts for most of the volume, often summarised as 80/20. Because the items differ so much in size, they need to be planned differently.Pareto profile: 20 A, B & C itemsBands of the range sorted by volume. A items are the high-volume runners & C items the slow tail. A items are made often & C items rarely.A items carry 70% of volume, 30 B items carry 20% & 50 C items share the last 10%. Neither rule prevents the disruption. The comparison is about how quickly & how steadily each line recovers once it has passed.
How the model is set up
One line, 100 products
The line has 132 working hours a week & 100 SKUs to make. About 100 of those hours go on production to meet demand, which leaves roughly 32 for changeovers & for catching up after a problem.
Changeovers use line time
Switching from one product to the next means clearing down, cleaning & resetting the line. This Changeover, or setupClearing down, cleaning, changing parts & checking the first output of the new product. The line makes nothing sellable while it happens.changeover takes 45 minutes, & nothing sellable is made while it happens. Twenty changeovers use 15 hours a week.
A, B & C items
Twenty products carry 70% of sales, the next 30 carry 20% & the remaining 50 share 10%. High-volume items are made often & low-volume items rarely, so that each batch is large enough to justify its changeover.
Two scheduling rules
The wheel keeps a fixed sequence that is published weeks ahead. Reactive re-planning rebuilds each week's plan around the SKUs closest to running out. Reactive sounds more responsive, & the model tests whether that holds up.
Where the difference appears
Both rules lose output during the disruption & neither can avoid it. They separate in the weeks after it ends, when the missed volume has to be made in hours that current demand already needs.
To start, choose Supplier shortage & press Play weeks. Terms with a dotted underline open a short definition. The glossary & the maths are at the end of the page.
Why a timetable is most useful when a train is cancelled
On the top line every fifth train is cancelled, yet the sign always shows when the next one leaves. The bottom line never cancels a train. It sends one whenever the platform looks full, so trains bunch & nobody can say when the next will come. Step through the eleven points below & the animation will show each one.
The week, hour by hour
Each bar shows one week of line hours. Solid blocks are production, grey hatching is planned changeover time & red hatching is the extra time a rushed changeover takes. Press play to step through the weeks, or hover over a block to see what ran.
The two plants compared
Both plants receive the same orders, hours & disruption. The only difference is how each decides what to run next.
Service during & after the disruption
Case fill rateThe share of cases ordered that shipped on time. Because it counts cases, the high-volume A items dominate it.Case fill each week, meaning the share of ordered cases shipped on time. The red band marks the disruption. For the reactive plant, most of the lost service comes after the band ends.
Where the stock sits in week 0
Each column is one SKU, with A items at the front & C items at the back. Column height shows weeks of CoverHow long current stock would last at normal sales. Two weeks of cover means the item runs out in two weeks if nothing more is made.cover. Red columns below the floor are SKUs that have run out while customers are still waiting for orders; the deeper the column, the more orders are waiting. Drag either model to rotate both.
How each plant spends its hours
Demand fixes the production hours. The rest go on changeovers or stay spare, & only the spare hours are available to absorb a disruption.
Average week, as run
Why the last few hours matter most
Queueing delay rises with ρ / (1 − ρ), where ρ is the share of hours in use. Close to 100%, each extra hour of changeover adds far more delay than the hour before it. Appendix §7 gives the formula.
What each rule gets for its stock & its hours
Every point on this chart is a full 78-week run with one setting changed, so each line shows the range of results a rule can reach. Lines higher on the chart deliver better service for the same stock or the same hours. Click any point to load its settings into the rest of the page.
Change the assumptions
Each assumption in the model can be changed here. Available hours & the unplanned setup factor have the largest effect on the result, so set those two from your own changeover data before relying on it.
What happened in this run
Disruptions will reach the line under either rule. With a fixed cycle, the line meets them with more spare hours & a sequence it can return to.
What the model covers
Glossary
Every term used on this page.
The maths behind the model
Four relationships drive the model: the balance of hours on the line, the shortest cycle the line can sustain, how often each class of SKU should run, & what happens to stock when the gap between runs becomes unpredictable.
Each section gives the idea in plain language, then the formula, then what it means on the line. The figures in the tables are calculated from the current settings, so the formulas can be checked against the simulation.
§1 How demand is generated
Weekly demand for each SKU is drawn from a normal distribution truncated at zero. The Coefficient of variationVariability measured against the average. A CV of 0.6 means weekly demand typically moves by about 60% of its average, which is common for slow movers.coefficient of variation rises as volume falls, & each C item has an 18% chance of no orders in a given week, which gives the tail its uneven pattern.
dit = max(0, μi(1 + cvi·Zit)), Zit ~ N(0,1) iid(1)Volume is split twice: across the bands by the 70/20/10 shares, then within each band by lognormal weights. Strictly this is Pareto-like rather than a true power law, because the band shares are set in advance. Items are then re-sorted into bands by their actual volume, as a real ABC review would do. Opening stock depends on where each item sits in the cycle, so the run starts in a settled state. Both rules start from the same position & see the same orders.
Each SKU has a stable average, but smaller SKUs order more erratically. Most of the variability sits in the C items, which is also where the reactive rule does most of its damage.
§2 How line capacity is used
The hours used must fit within the hours available. This is the model's only hard constraint.
∑i∈St ( si + Qit/r ) ≤ Ht(2)On a wheel, the set of items run each week, St, comes from the schedule, so setup
load is a constant set by the run frequencies:
Λwheel = ∑c Nc·s / Tc(3)The week therefore divides into three parts:
Only the production hours make anything. Setup hours are the cost of making many different products, & the slack is what remains to absorb a disruption. On a wheel the setup hours are fixed in advance, so the slack is known. Under re-planning they depend on events, & they tend to rise when the line is already under pressure.
§3 The shortest cycle the line can sustain
If all N items ran on one common interval T, each turn of the cycle would need N setups plus the production itself:
N·s + (D/r)·T ≤ H·T ⇒ T ≥ N·s / (H − D/r) = Tmin(4)Writing ρrun = D/(rH) for the share of hours spent producing:
Tmin = N·s / ( H·(1 − ρrun) )(5)As the line gets busier, the shortest cycle it can keep gets longer, & it lengthens sharply as utilisation approaches 100%. On a busy line it is not possible to run every item more often, which is why §4 gives A, B & C items different intervals.
§4 Why high-volume SKUs run more often
Solving the ELSP - economic lot scheduling problemThe standard problem of deciding how often to make each product on a shared machine, balancing changeover cost against the cost of holding stock.economic lot scheduling problem item by item balances setup cost against holding cost, & gives an interval inversely proportional to the square root of demand:
Ti* = √( 2s / (h·μi) ) ⇒ Ti ∝ μi−1/2 ⇒ Tc/TA = √(μA/μc)(6)Doubling the interval at each step from A to C is close to the lowest-cost pattern for a 70/20/10 split, so the wheel's frequencies can be justified on cost as well as on discipline. Running C items as often as A items would spend changeover hours that the high-volume items need.
§5 The cycle stock created by making in batches
Stock made to cover an interval is used up over that interval, so average Cycle stockStock held because items are made in batches rather than continuously. A household that shops every two weeks keeps about two weeks of food.cycle stock is half the run quantity:
Icycle = ∑i μiTi/2 = (D/2)·∑c wcTc cover = Icycle/D + SS(7)A household that shops every two weeks keeps about two weeks of food in the house; shopping weekly halves the stock but doubles the trips. The wheel holds more stock for the same reason, & that extra stock is the cost of running to a schedule.
Cycle stock depends on how often we choose to run each item, while safety stock protects against variation in demand. Plants with high stock & poor service often have plenty of the first & too little of the second. Because a fixed cycle makes cycle stock predictable, the safety stock on top of it can be sized with some confidence.
§6 Why a known next slot reduces safety stock
With a fixed interval R & lead time L, safety stock covers the protection interval in the usual way:
SS = z·σd·√(R + L)(8)If the interval itself varies, the variance gains a second term:
SS = z·√( (R̄+L)·σd2 + μd2·σR2 )(9)On a wheel, σR is close to zero because the next slot is published, so the second term drops out & safety stock only has to cover demand. Under re-planning, the timing of a SKU's next run depends on shortfalls elsewhere in the range. σR is then large & hard to estimate, which makes the buffer hard to size.
The main benefit of a fixed cycle is that the gap between runs stays the same, even when that gap is fairly long. A rail timetable gives passengers the same assurance: trains still run late, but people know when the next one is due.
§7 Why high utilisation amplifies disruption
Utilisation includes setup time, so a rule that changes over more often runs a busier line on the same equipment:
ρ = ( D/r + Λ ) / H(10)Kingman’s approximation then shows how waiting time responds, where ca & cs are the coefficients of variation of arrivals & service:
W ≈ ( (ca2 + cs2) / 2 ) · ( ρ / (1 − ρ) ) · τ(11)Reactive re-planning raises both terms. ρ rises because changeovers take more of the week, & cs2 rises because changeovers now vary between short planned ones & long rushed ones. With ρ/(1−ρ) in the formula, delay grows much faster than either input. Reducing available hours from 140 to 125 changes the inputs gradually but changes the result sharply.
§8 How a backlog slows the line down
Backlog grows by the difference between demand & output, & output is whatever the hours left after setup can produce:
Bt+1 = Bt + Dt − Yt, Yt = r·(Ht − Λt)(12)Under re-planning the number of expedited setups, et, rises with the backlog. Each one takes k times the planned setup, so output falls as the backlog grows:
∂Yt/∂Bt = −r·s·(k − 1)·∂et/∂Bt < 0(13)On a wheel, setup load comes from the schedule & does not respond to the backlog:
∂Λt/∂Bt = 0 ⇒ ∂Yt/∂Bt = 0(14)Under re-planning, a growing backlog brings more rushed changeovers, which reduce output & let the backlog grow further. If that loop is strong enough the backlog keeps rising, which shows up as the sawtooth in the service chart. The bullwhip effect works through a similar loop, & in both cases the upset lasts far longer than the event that caused it.
On a wheel, output in a bad week is the same as in a good one, so the backlog falls at a steady rate until it is cleared.
§9 Case fill & line fill measure different things
Case fill rateThe share of cases ordered that shipped on time. It is weighted by volume, so the A items dominate it.Case fill weights each SKU by its volume. Line fill, or OTIFThe share of order lines delivered complete & on time. Every SKU counts the same, so a customer who orders ten items & receives nine has a failed line even though 90% of the units arrived.Line fill weights every SKU equally, so the two measures can give very different answers:
CFRt = ∑i min(xit, dit) / ∑i dit LFRt = |{ i : xit ≥ dit }| / |{ i : dit > 0 }|(15)C items make up half of the order lines but only a tenth of the cases. A plant can report good case fill while a customer ordering across the range finds half their lines short. When reactive re-planning drops C items, case fill hardly moves but OTIF falls, so judging the two rules on case fill alone understates the gap between them.
§10 What the model leaves out
Some of these omissions favour the wheel & some count against it.
- There is one line, with no parallel resources to share the load.
- Setup time does not depend on sequence. Real wheels are sequenced to shorten changeovers by grade, colour or allergen, so a real wheel would gain more than this one. Understates the wheel.
- Lead time is a flat one week, with no supply variation apart from the disruption itself.
- All unmet demand is backordered, with no lost sales or substitution. Lost sales would hurt both rules, & reactive more.
- Demand is independent across SKUs & weeks, with no seasonality, promotions or forecast bias.
- Capacity is fixed, with no overtime, extra shifts or subcontracting. A plant that is fire-fighting usually buys extra hours, which moves the cost onto the P&L.
- Inventory records are accurate, & there is no yield loss apart from the quality hold.
- The wheel is assumed to be followed. Holding to it under commercial pressure is the hardest part in practice, & the model does not test it. Favours the wheel.
The minimum economic run length & the unplanned setup factor k carry most of the weight. Both can be measured in a real plant, so set them from your own changeover studies before using the model to support a decision.