Working notes
Field notes on industrial control, instrumentation and the small decisions that decide whether a plant runs quietly or fights you. Written down mostly so I stop re-deriving the same things every couple of years.
Updated occasionally · no tracking · no comments
-
Why the derivative term usually makes things worse
Textbooks give D equal billing with P and I. On real plants it amplifies sensor noise far more often than it damps overshoot — and the cases where it earns its place are narrower than they look.
-
4–20 mA is not obsolete, and the reason is boring
A live-zero current loop survives broken wires, voltage drop and bad grounding in ways that digital buses still don't. The elegance is in the failure modes, not the data rate.
-
Reading cavitation from the pump curve
NPSH margin is a design number; what you actually get is a system number that changes with strainer fouling, liquid temperature and how far the operator throttled the suction valve.
-
Scan time is a contract, not a statistic
Average scan time tells you almost nothing. The number that matters is the worst case, and the usual way people measure it hides exactly the events that produce it.
-
Shield grounding: one end, both ends, and when the rule flips
The single-point rule is right for low-frequency signal cable and wrong for high-frequency noise. Most arguments about it are two people describing different frequency ranges.