Downtime gets managed in two ways in most plants. You wait for something to break and fix it quickly, or you service equipment on a fixed calendar whether it needs it or not. The first is expensive when it happens. The second is expensive all the time, and still doesn’t prevent the failures that matter.
The Case Against Calendar-based Maintenance
Servicing a machine every 500 hours assumes every 500 hours of running is equivalent. It isn’t. A press running soft material at moderate speed accumulates wear very differently from the same press running hard material at maximum rate. Fixed intervals overservice the easy duty cycles and underservice the hard ones, which is the worst of both outcomes: you’re spending money on unnecessary work and still getting surprised.
What Condition-based Prediction Looks at
A model watches for the signatures that precede failure rather than counting hours. Vibration amplitude and frequency shifts. Motor current draw creeping up as bearings degrade. Temperature rising at a point in the cycle where it previously didn’t. Cycle time drifting by fractions of a second. Reject rate on a specific machine ticking upward before anyone flags a quality problem.
Individually these signals are noise. What a model does is learn the combination that has historically preceded failures on that equipment, and flag when the current pattern starts to resemble it. The output is a probability and a rough window — elevated failure risk on this asset over the next two to three weeks — which is enough to schedule intervention into a planned changeover rather than a Tuesday morning breakdown.
Failures You Can’t Predict
Be clear-eyed about scope. Predictive models handle degradation — wear, fatigue, contamination, drift. They do not handle sudden events: a foreign object, an operator error, a power surge, a component with a manufacturing defect that fails without warning. If someone promises to eliminate unplanned downtime, they’re describing a category of failure that doesn’t exist.
There’s also a data prerequisite that’s easy to skip past. Models need failure history to learn from. If your maintenance records are handwritten, inconsistent, or record only that a repair happened without what failed or why, there’s nothing to train on. Improving maintenance record quality is often the real first project.
The Wider View of Downtime
Equipment failure is only one cause, and often not the biggest. Material shortages, changeover overruns, tooling unavailability, operator absence, and waiting for quality sign-off collectively account for more lost hours than machine breakdown in most plants. These are all predictable too — from schedule, inventory, and staffing data rather than sensors.
Before investing in sensors, categorise your actual downtime by cause for a full quarter. Many manufacturers find the cheapest available gains are in materials and changeovers, not maintenance.
ticktick.ai flags downtime risk from both directions — material and schedule constraints alongside equipment condition — so the biggest cause gets attention first.
