Summary

You have built a factory and a panel that runs it and measures it. All the logic of the panel is a few lines:

  • a machine's lamp — B = BUSY: lit while the machine works on a part;
  • released — counting the source's OUT pulses, two CD4029 in binary mode: the high one's CI = the low one's CO;
  • made — counting the 0 → 1 edges of the IR pair's receiver: a part has passed the beam;
  • reset — PE = R, the counters load J = 0000;
  • the shift — REL = CK AND RUN AND NOT R7: the clock reaches the source while START is on and fewer than 128 are released.
CK1 HzSTARTRUNNOT · AND · ANDCD4069 · CD4081sourceRELEASEREL2 × CD4029releasedOUTdisplay RR7 — end of shiftIR pairat the sink2 × CD4029madeOUTMdisplay MRPER — PE of both counters: both displays to 0
The whole panel: the clock and START through a NOT and two ANDs open the release; two counters count what is released and what is made; the top bit of «released» ends the shift; R clears both.

And the experiment showed the main thing: five machines, each on average keeping up exactly with what it is fed — and a chain of them makes less and piles parts up inside. Dependent events plus fluctuations.

What next

  • Put a die on only one machine — the first, the middle, the last. Where does it hurt the most?
  • Make one machine slower (2 s instead of 1): the line will run at the speed of that machine — Herbie is found.
  • Give the machines a bigger buffer (cap): the work in progress will grow, and «made» will barely change. Stock does not cure fluctuations.
  • Release less often — once every 1.2 seconds: the machines will have it easier, but fewer parts are made in a shift. Where is the sweet spot?