From spectra to decisions

September 2026 · 3 min read

A vibration spectrum is a beautiful thing. Bearing frequencies, gear mesh, sidebands — a machine confessing its problems in a language you can learn to read. I spent a certification learning to read it. Then I watched what happens to all that beauty in practice: it goes on a dashboard, and the dashboard goes unwatched.

This is the standard failure of condition monitoring. We build systems that display the condition of machines, and we assume decisions will happen on their own. They don't. A dashboard has no opinion. It shows you forty trend lines and wishes you luck.

A dashboard answers "what is the machine doing?" A decision system answers "what should we do about it, and when?" These are different products that happen to use the same data.

You can tell which one you've built with a single test. Pick any chart on the screen and ask: if this line crosses that level, what specifically will someone do differently? If the answer is "they'd investigate," you have a dashboard. If the answer is "the planner pulls the work order forward to the next outage window, because the lead time on that bearing is six weeks" — you have a decision system.

The difference isn't the analytics. It's the direction you built in. Dashboards are built forward from the data: here's what we can measure, here's a chart of it. Decision systems are built backward from the action: here are the five decisions maintenance actually makes — run on, watch closer, plan the repair, do it now, shut it down — and here's the evidence each one needs to be made with confidence.

Built backward, most of the spectrum never needs to be shown. That envelope trend only matters because it moves the machine from watch closer to plan the repair. The threshold isn't a statistical artifact; it's the point where waiting starts costing more than acting. Which means setting it is not a data-science question. It's a question about lead times, spares, and outage windows — and the data scientist who won't go learn those numbers is decorating, not deciding.

So when someone shows me a condition-monitoring system, I no longer ask what it detects. I ask which decision got easier. Sometimes there's a long silence. That silence is the product gap in this whole industry.