I spent a year and a half as a manufacturing process engineer before moving into controls. Two lessons from that stretch have stuck with me more than the rest.
Automation has a floor, and a ten-year-old is standing on it
Not every task is worth automating. Some jobs on the plant floor look tedious enough that a machine should obviously do them, and then you price out the fixture, the vision system, the robot arm, the integration time, and the math never closes. You are trying to buy a machine to do what a ten-year-old could do in an afternoon.
That is not a knock on the people doing that work. It is the opposite. No robot on the market matches the everyday combination a person brings to a station: fine dexterity, visual intuition about when something looks off, and the mobility to move around an odd-shaped part and adjust on the fly. We treat those skills as ordinary because they are common, but they are the hardest things to replicate in hardware. The jobs that resist automation are a good measure of how much human skill was actually in them.
The practical version: automate the parts of a process that are rigid, repeatable, and well defined. Leave the parts that need hands and judgment to the people who are already good at them, and stop apologizing for the line item.
A production line is an RLC circuit
The second lesson is that a manufacturing process can be written down as a circuit. I ended up formalizing this in A Circuit-Inspired Framework for Manufacturing Systems, but the core mapping is simple:
| Circuit | Manufacturing |
|---|---|
| Current source | Customer demand (a rate the line has to meet) |
| Current | Throughput |
| Resistance | Effective cycle time, per unit, adjusted for parallel stations and uptime |
| Capacitance | Buffer capacity |
| Inductance | Production inertia: how slowly output re-settles after a demand change, measurable as the time the line takes to reach 63 percent of a new rate |
Demand is a current source, not a voltage source: the customer sets a rate, and the line either keeps up or the buffer and backlog absorb the gap. Throughput is whichever is smaller, demand or capacity, so it tracks demand until it hits the wall and then clamps. The bottleneck is just the station with the largest effective cycle time. Buffers charge and discharge like capacitors when input and output rates diverge. And because output lags demand, the line obeys a second-order equation with a natural frequency and a damping ratio, which tell you whether a demand spike settles quietly or sends work-in-process oscillating up and down the line for days. You want it damped, not tuned to ring.
This is not a replacement for discrete-event simulation. It is a faster way to reason, especially if you already have the mathematical base for differential equations. Before I build a detailed model I can already say which station will starve and why, roughly how big the buffer in front of the oven needs to be to cover the night shift, and whether the line is underdamped enough that I should worry about it ringing. That intuition is worth more on the floor than a precise number I get a week later.
The throughline
Both lessons are about knowing what your tool is for. Automation is for the rigid, repeatable middle of a process. A circuit model is for fast structural reasoning, not final answers. Most of the waste I saw came from pointing a tool at the part of the problem it was worst at.
Yours Truly