The Line That Starved Every Time the Robot Blinked
A two-station assembly cell had no buffer between the stations. Station A picked and placed a part (cycle time 4 s). Station B assembled it (cycle time 5 s). Station A was faster; Station B was slower. Without a buffer, Station A had to wait for Station B to finish before it could place the next part. The line’s throughput was limited by Station B (5 s), and Station A sat idle 20% of the time. Worse, when Station B had a 10-second hiccup (a sensor glitch), Station A also stopped — because there was nowhere to put the finished part. The two stations were coupled: one hiccup stopped the whole cell. We added a 5-position buffer between them. Station A could keep placing parts into the buffer while Station B recovered. Throughput rose from 12 ppm to 18 ppm — a 50% gain from a $200 belt.
A buffer between stations is the cheapest throughput improvement in automation. Here is why, and how to size it.
Why a Buffer Decouples Stations
Without a buffer, two stations are “coupled”: the fast one must wait for the slow one, and any hiccup in either stops both. With a buffer (a queue of 3–5 parts), the fast station can keep producing into the buffer while the slow one catches up. The slow station’s hiccups are absorbed by the buffer, not transmitted upstream. The line runs at the average of both stations, not the worst instant.
This is the single most common design flaw in new cells. Engineers design for the “ideal” cycle time (both stations at their average). Real stations vary — a sensor takes a little longer, a part sits crooked. The buffer absorbs the variation. Without it, the line runs at the worst-case cycle, not the average.
The buffer rule: Put a 3–5 part buffer between any two stations with different cycle times. It decouples them, absorbs hiccups, and lets the fast station keep producing. The 50% throughput gain came from a $200 belt — not a faster robot.
How Big Should the Buffer Be?
The buffer size depends on the variation in the slow station. If the slow station varies by ±20% (a 5 s cycle that takes 4–6 s), a buffer of 3–5 parts absorbs the variation. If the slow station has long hiccups (a 30-second jam every 50 parts), the buffer needs to hold 30 seconds of production — about 6 parts at 2 s per part. The rule: size the buffer to cover the expected hiccup time, not the average cycle.
A buffer that is too small (1–2 parts) does not decouple anything — it just holds the part being passed. A buffer that is too large (20 parts) ties up work-in-progress and takes floor space. Five is the sweet spot for most cells.
The Infeed Buffer: Protect the Bottleneck
The most important buffer is in front of the bottleneck. If the bottleneck station ever starves (waits for a part), the whole line loses that capacity. A buffer in front of the bottleneck ensures it always has a part ready. The buffer behind the bottleneck (after it) is less critical — the downstream stations can wait. The rule: buffering protects the bottleneck, not the fast stations.
Conveyor Buffers vs Gravity Buffers
A buffer can be a powered conveyor (parts accumulate on a belt), a gravity rail (parts slide down), or a rotary accumulator. The choice depends on the part shape. A powered conveyor works for anything. A gravity rail needs the part to slide (and needs a stop at the end). A rotary accumulator holds many parts in a circle for space-constrained cells.
A Buffer Sizing Example
| Situation | Buffer Size |
|---|---|
| Two stations, similar cycle, small variation | 2–3 parts |
| Fast infeed, slow bottleneck | 5–8 parts (in front of bottleneck) |
| Bottleneck has frequent 30 s jams | Cover 30 s of production (~6 parts) |
| Long changeover at downstream station | Hold 10–15 min of WIP |
| Cosmetic parts (scratch risk) | Use soft-lane buffer, no part-on-part |
A Buffering Checklist
- Identify the bottleneck station (from the cycle time study).
- Is there a buffer in front of the bottleneck (prevents starvation)?
- Is there a buffer between any two stations with different cycle times?
- Is the buffer sized to cover expected hiccups (30 s)?
- Does the buffer signal upstream when full (prevents overproduction)?
- Does the buffer signal downstream when empty (prevents starving)?
- Is the buffer material compatible with the parts (no scratching)?
- Is the buffer accessible for jams?
- Does the line run without the fast station waiting?
- Measure throughput before and after (prove the gain).
The Bottom Line
The line that starved every time the robot blinked had no buffer. A 5-part queue decouples the stations, absorbs the hiccups, and lets the fast one keep producing. The cheapest throughput gain in automation is not a faster motor — it is a place to put the part while the next station catches up.
