[oehpr]

It's funny you should mention that, because most PID configurations are not very stable. Once you introduce the Integral and the Derivative, if you're not careful that PID will oscillate into the stratosphere.

In fact, I'd say that PID's that are functioning most efficiently are very nearly tuned to the point of oscillating. Even worse is the kinds of adjustments you have to make to PID's when there is significant lag between their inputs and outputs.

[rcxdude]

Indeed. Control loops are generally limited by the bandwidth of the system they are controlling, and lag in the system limits the bandwidth you can achieve. Trying to get a PID loop to move a system faster than that system's bandwidth will just create oscillations, and the more tightly the PID is tuned the worse its behaviour when exposed to shocks (especially when the system becomes significantly non-linear in the process).

You can extract more performance out of a system by having a very good predictive model of it and measuring its inputs: This can really improve how you drive the system but it's much more difficult to achieve, and you're still limited by how quickly information moves through the system in terms of how you can react.

[_0ffh]

Yeah, I think that's a much more promising approach. Have a good model, use it to build an observer, now you can control based on the state of your simulated system.

[hef19898]

You described Supply Chains pretty well, I think!