Simulation and bottlenecks8 min read

How to Compare Resilience, Not Just Throughput

DBR77 TeamPublished

To compare resilience, run every layout or investment option through the same stress cases and measure what changes: how much throughput is kept, whether orders are still delivered on time, how fast the buffers empty and where the bottleneck moves. Each stress case is a fixed change of assumptions, such as a slower station, a later order release, a shorter shift or a demand peak. Throughput tells you how an option performs when everything goes to plan. The stress cases tell you how much of that result survives when one assumption turns worse. Rank options on both.

Why the fastest option is often the most fragile

Options that win on average output often win because they removed slack: shorter routes, smaller buffers, one shared forklift instead of two. When everything goes to plan, that is efficient. When one assumption turns worse, nothing absorbs the shock.

Consider two illustrative options. Option A links three stations directly, with almost no buffer between them, while Option B keeps a small buffer rack after the second station. In the base case, A produces a little more, because fewer parts sit in the rack. Now change the order mix so that for two days the orders have a longer cycle time at the second station. In A, the slow orders starve the third station and cut output at all three stations at once. In B, the third station keeps working from the rack for part of that time, and less output is lost.

A spreadsheet of averages ranks A first every time. Only a model that runs the line through time, with the changed assumption in place, shows the difference.

Conditions change often enough to plan for. McKinsey Global Institute found that, averaging across industries, companies can now expect supply chain disruptions lasting a month or longer to occur every 3.7 years. Inside the plant, order mix, shift patterns and station speeds also move from month to month.

Five resilience measures to put next to throughput

Throughput stays on the scorecard. Add five measures that describe how an option behaves when an assumption turns worse.

  1. Throughput retained: the share of base-case throughput that an option keeps in each stress case.
  2. On-time delivery: the share of orders finished by their due date in each stress case.
  3. Buffer coverage: how long the buffers in front of the bottleneck keep it running when an upstream station slows down.
  4. Bottleneck shift: which station or vehicle becomes the constraint in the stress case, and in which periods it limits the line. Often it is not the one you planned around.
  5. Work in progress and lead time: how much material piles up and how long orders take. That is floor space and cash.

Define each measure before the runs, with a clear target. "Holds up" should mean a number, for example throughput retained and on-time delivery above a level you agree in advance.

Build stress cases from your own history

The best stress cases are changes your plant has already lived through or expects. Go through the last twelve months with production planning, logistics and procurement and list what changed:

  • the stations that ran slower than nominal for a period, and by how much
  • late or partial deliveries: which materials, how many days late, so how much later the affected orders could start
  • products or weeks that took longer than the nominal times suggest
  • shifts that were shorter than planned
  • ramp-ups: after a new machine or product, how many weeks it ran below the planned rate

Turn each into a scenario with a fixed size everyone accepts, for example the cycle time of one press 15% longer for a week or the second shift shortened by one hour. Procurement signs the supplier delays, production planning the slower cycle times. If nobody agrees on the size, the plant is still guessing.

DBR77 Digital Twin simulates the week deterministically: stations, routings, buffers, transport, shifts and the production plan with fixed times, so the same input always gives the same result. Each stress case is a variant on a copy of the base model, and the variants are compared side by side. Because the model has no randomness, a difference between two options comes from the change itself, not from noise. See Averages vs simulation over time.

See how a layout behaves when a station slows down

Think of the slowdown that hurt your plant most last year. The demo shows a slower station tested on a tight layout and on one with a buffer, and which one keeps more of its output.

Talk to an expert

Model late material and ramp-up as inputs, not excuses

Two situations are often treated as one-off excuses after the fact. Both belong in the model before the decision.

For supplier delays, a static plan assumes standard lead time. A stress case sets a later release date for the affected orders, with a size from your history. It then shows where the late material hurts: which line starves, whether the buffer before the bottleneck lasts, and whether a later rush creates a queue at the bottleneck.

Ramp-up works the same way. A new line, cell or product rarely runs at full rate from day one. Model the first weeks with longer cycle times and compare them with the planned rate. The bad weeks often come together: late material and a slow ramp-up in the same month. That combination is the case to test.

Read the results side by side

Use the same scorecard for every option and every stress case, so the comparison stays fair. The numbers below are illustrative.

ScenarioMeasureOption A: tight lineOption B: line with buffer
Base caseOutput per shift410 units400 units
Station 2 cycle time 15% longer for two daysOutput lost per shift60 units20 units
Station 2 cycle time 15% longer for two daysOn-time delivery82%95%
Component released 3 days lateBottleneck during the delayAssembly, starvedAssembly, partly fed from buffer
Demand peak of 10% in week 3On-time delivery78%91%

In plain words: Option A wins the base case by a small margin. Option B loses less in every stress case and delivers more orders on time. Whether B is worth its extra floor space and inventory depends on how often those cases happen, and your history answers that.

Keep the summary that plain: "A gives slightly more in the base case. B keeps more of its output when Station 2 slows down."

What resilience costs, and how to decide

Buffers take floor space and working capital, a second forklift costs money, and a parallel station takes room and investment. The comparison shows what each option buys with that spend. If the plant rarely faces the stress case, the leaner option may be right. If it happens every month, the fragile option pays for its efficiency in late orders. If two options tie on throughput, choose the one that keeps more of its output in the stress cases.

Stress cases also interact with demand. A layout that holds up at today's volume may not hold up at a higher one. How to test capacity against several demand scenarios is covered in How to Test Capacity Before the Next Demand Shift. To check whether the ranking depends on one uncertain input, use What a Sensitivity Analysis Should Show Before Approval.

FAQ

What is resilience in a factory simulation?

It is how much of its output and on-time delivery a layout or investment option keeps when one assumption turns worse: a slower station, a later order release, a shorter shift or a demand peak. It is measured with throughput retained, on-time delivery, buffer coverage, bottleneck shift, work in progress and lead time. The simulation compares fixed assumptions; it does not model random events.

Which stress cases should we simulate?

Start with the ones from your own last twelve months: the slowest stations, the worst supplier delays, the shortest shifts and the demand peaks. Agree the size of each with production planning and procurement before the runs.

Does a more resilient layout always need more buffer?

No. Resilience can also come from a second transport route, a parallel station or a different sequence of operations. The simulation shows which source of resilience costs least for your flow.

Do we need live machine data to model stress cases?

No. Nominal times are enough to start. Enter routings, stations and times by hand, or bring them in by import (import scope to be confirmed). Measured cycle times and order records make the stress cases more precise later. Integrations with ERP, MES and other systems are individual projects.

Conclusion

A practical first step is the list of changes that would hurt most. Pick the three that cost the most output or delivery performance in the last twelve months, agree their size with production planning and procurement, and run those three cases on every option before the next review. Then put throughput retained and on-time delivery in the same table as output per shift, so the committee sees both before it ranks.

See where simulation was used to compare options

Buffers, intralogistics routes and parallel stations are all places where a plant keeps slack. The use cases show where such changes were tested in DBR77 Digital Twin before anything was built.

Browse use cases

Sources

Want to test a decision from your plant?

Describe the decision to a DBR77 expert and we will show how DBR77 Digital Twin compares the options on output, waiting time and transport load.

Talk to an expert

Questions

Have a question about this topic?

Tell us what it is about. We will reply within one business day.

We use your data only to answer your request. See our Privacy Policy.

Digital Twin newsletter

Short notes on testing layout and flow decisions on a model, and news from DBR77 Digital Twin. You can unsubscribe at any time.