The demo models that ship with ReliaSim are not a loose collection of examples. They are one bottling line — Filler, Capper, Labeler, Case Packer, Palletizer — modeled eight ways, varying across two independent dimensions.
Understanding the two axes is the fastest route to answering the question most people actually have when they start: how much model do I need?
The two dimensions
Data resolution — what you know about failures. The Constraint-Level track gives each machine a small number of lumped interrupts. The LEDS-Level track replaces them with the detailed named failure modes a Line Event Data System records, so the Capper is not simply "down" but down for a specific, attributable reason.
Topology — how much structure you represent. Four steps, each adding one idea to the one before it: a single constraint, then constraints in series, then unit conversion, then inline buffers.
These vary independently. You can model a very simple line with very detailed failure data, or a complex line with lumped interrupts, and both are legitimate. Which combination you need depends on the question you are asking and the data you actually have.
The grid
| Topology | Constraint-Level | LEDS-Level | Output |
|---|---|---|---|
| 1-Constraint — one machine, interrupts only | Open | Open | bottles |
| 5-Constraint — five machines in series | Open | Open | bottles |
| Converter — bottles → cases → pallets | Open | Open | pallets |
| Buffer-Options — inline buffers added | Open | Open | pallets |
Every model runs in the browser sandbox — no install required.
The topology axis, step by step
1 · 1-Constraint
A single constraint carrying interrupts, and nothing else. The point of this model is to isolate one relationship: how availability at a machine turns into lost production, with no other machine able to confuse the picture.
- Constraint-Level — one machine, five interrupts, Weibull(30,1) time-to-failure and Weibull(5,1) downtime. Reference result: 53.4% efficiency, 6.92M bottles over 90 days.
- LEDS-Level — the same single machine carrying all 36 real failure modes instead. Reference result: 53.2% efficiency, 6.89M bottles, with
Capper_Upper Plow Areaalone accounting for about 20% of downtime.
Compare those two and you have the whole argument for detailed failure data in one line: the headline number barely moves, but only one of them tells you which failure to go and fix.
2 · 5-Constraint
Five machines in series. Adding stations introduces the behavior a single machine cannot show — blocking and starvation, and the fact that the busiest machine is rarely the constraint.
- Constraint-Level — one interrupt each. Reference result: 54.7% efficiency, 7.09M bottles, bottleneck at the Case Packer, with 36% of time blocked or starved.
- LEDS-Level — the same five stations with 36 real interrupts distributed across them. Reference result: 54.0% efficiency, 7.00M bottles, bottleneck at the Capper at 75.7% availability.
Note that the two tracks disagree about where the bottleneck is. That is not a defect in either model; it is what happens when lumped interrupts average away the failure that actually dominates.
3 · Converter
Two constraints and three converters, so material changes unit as it flows: bottles become cases at 10:1, cases become pallets at 10:1. Output is now measured in pallets rather than bottles.
- Constraint-Level — reference result: 54.1% efficiency, bottleneck at the Filler.
- LEDS-Level — reference result: 54.3% efficiency, roughly 70,000 pallets, bottleneck at the Capper at 75.4% availability, where the
Upper Plow Areafailure mode is 54% of all Capper downtime.
This is the first model that reports in the unit the plant actually ships, which matters more than it sounds — a line optimized for bottles and a line optimized for pallets are not always the same line.
4 · Buffer-Options
The Converter topology with three inline buffers added between stages, and the most instructive model in the set.
- Constraint-Level — reference result: 54.1% efficiency, 70,113 pallets.
- LEDS-Level — reference result: 54.3% efficiency, 70,361 pallets.
Efficiency does not move. Compare against the Converter models above and the buffers have bought essentially nothing. That is the intended lesson: buffers absorb short interrupts, but the dominant loss here is a long stoppage on the constraint itself, and no amount of storage fills that. Fix the machine, not the flow.
If you want to see that argument worked through with the numbers, see How big should a buffer be? on the ReliaSim site.
Which one should you start with?
- Learning the tool — start at 1-Constraint on the Constraint-Level track and work down the topology axis. Each step adds exactly one idea.
- Deciding whether your data is good enough — compare the two tracks at the same topology. That comparison, more than any other, shows what detailed failure data buys you.
- Evaluating for a real line — go to Converter or Buffer-Options. Those carry unit conversion and storage, which nearly every real packaging line has.
Reference figures above are the published results for each demo model. Open any of them in the sandbox and re-run to reproduce.