[Flatcircle]

Had a physics teacher in highschool, worked on a submarine. Told me that with government equipment like submarines, workers replace parts based on dates regardless of the wear and tear of the part. "If this screw is supposed to last 10 years, we're ordered to replace it in 7 years regardless of the condition."

With for profit companies, they'd try to push the parts to last 12 years instead of 10 by inspecting them and confirming they're still good.

Saving money this way isn't too big of a deal if you're McDonalds and it's an ice cream machine.

But if it's a for-profit airline it can turn into a problem.

[wenc]

It’s called predictive maintenance and it is the norm in many critical industries. Replacing equipment every fixed number of X years (no matter what the condition of the equipment is) is wasteful, costly disruptive and does not actually guarantee better reliability. (although it makes people feel better).

Most predictive maintenance is done with data — updated sensor data processed through a mathematical model derived from principles from reliability engineering. In a sense it’s actually more realistic than the X years model (which is a once off number derived from some reliability model too but doesn’t have the benefit of being updated with real data — it’s usually an overly conservative number)

https://en.m.wikipedia.org/wiki/Predictive_maintenance

There is no evidence that predictive maintenance is the cause of Boeing’s mishaps.

[schiffern]

That's true, but also beside the point.

Flatcircle's point wasn't really predictive maintenance vs. fixed lifetime, it was about government using conservative design margins vs for-profit industry's riskier design margins.

Predictive maintenance lets you achieve a better cost-vs-risk curve, but the organization still needs to select a point somewhere along that curve.

[malfist]

The other thing replacing the part can do (but not always) is guarantee you still have a source for those parts in the future. That's important for the military that might need to make more without much notice.

[lawlessone]

I've heard the word AI being used around this before... and it scares me.

[wenc]

Why?

It’s just a marketing term. Most predictive maintenance models are ML or statistical learning models and have a great track record. Some call it AI to sound hip (you can any ML model AI these days) but it’s probably just a standard model from reliability engineering.