The PLC is not going anywhere

I am writing this in the office of a machine shop in Erie, Pennsylvania. Through the window is a controller installed in 1991. It is beige, it has run every shift since, and nobody in the building has a password for it, because it has never asked for one.

Once a year, someone announces its death. The announcement is always well produced. The controller does not respond, since it has a scan cycle to finish.

What the box actually does

A programmable logic controller reads its inputs, runs its logic, and writes its outputs. It does this every few milliseconds, forever, and it does it the same way on a Tuesday in July as on the day it was commissioned. That is the entire product. Everything we call reliability in a plant is a long series of boxes doing that one thing without being asked for an opinion.

The shop in Erie runs forty-one machines. Thirty-three are controlled by something older than the newest operator. The owner, who is sixty-three, has been pitched a replacement platform nine times. He keeps the decks in a drawer, because they make a good case for the drawer.

The box is not the problem. The problem is that nobody has been allowed to talk to it.

Where the software goes

We have now backed three companies that start from this view. Torquewell listens to a spindle and tells you, three weeks early, that it is going to fail. Routecard replaces the paper traveler that follows a job from saw to ship. Quench, before it was acquired, sat beside the heat-treat furnace and read the same signals the furnace already had.

None of them replaced a controller. Each one reads what the controller already says, adds a layer that is allowed to be wrong, and leaves the beige box alone to be right. The test we use is simple. If the software crashes, does the line stop? If the answer is yes, we pass, and we write a paragraph about why.

Why replacement keeps failing

A replacement has to match the old box on every edge case, and nobody has written the edge cases down. They live in the ladder logic, in comments left by a man who retired in 2006, and in the habits of the operator who knows to tap the side panel twice. A new platform inherits none of this. It starts with a list of promises and learns the rest on the floor, one stopped line at a time.

The cost is not only money. A plant that stops for a day to test a new controller loses the day, and a plant manager remembers it for a decade. This is why the best pitch against the PLC is the one that never mentions it.

What to ask a founder

Ask how many minutes an install takes, and whether a maintenance tech can do it alone. Ask what happens when the network goes down on second shift. Ask who gets the call. If the answer involves a new cabinet, a new vendor and a six-month project plan, the product is competing with the PLC, and the PLC has a thirty-year head start.

The better answer is a clamp, a cable and a Tuesday afternoon.

We expect the controller to outlast most of the platforms announced against it. We intend to fund the companies that were polite to it.