The short answer
Machines fail predictably enough that run hours and battery health explain most of it. The value is not cheaper repairs, it is not losing a shift when a scrubber dies at a site with a two-hour access window.
Cost the failure in service terms rather than in parts, because that is where the real money is.
What data exists already
- Run hours, from the machine or from the job record
- Battery charge cycles and charging behaviour
- Fault codes, where the machine records them
- Service history, if it has been recorded consistently
- Consumable replacement intervals, such as pads and brushes
Service history is usually the weak link. Where repairs are recorded as free text or not at all, there is nothing to learn from, and fixing that recording is the first piece of work.
What actually predicts failure
| Signal | Typically precedes |
|---|---|
| Battery holding less charge | Mid-shift failure, weeks ahead |
| Rising run hours since service | General wear |
| Increasing fault code frequency | A specific component |
| Longer time to complete the same site | Degrading performance |
| Repeated resets by operators | Something nobody has reported |
The last row is worth capturing deliberately. Operators work around faults rather than reporting them, and a reset counter finds problems weeks before a fault report does.
Batteries are usually the answer
For battery-powered equipment, battery health explains a large share of unplanned failures and degrades predictably. Tracking charge cycles and capacity gives useful warning with no additional hardware in many cases.
Replacing a battery on a planned basis before it starts failing mid-shift is cheaper than the disruption, even when the battery had life left.
Keep it proportionate
This is not an aerospace condition-monitoring programme. For a fleet of machines, a simple model on run hours and battery data, reviewed weekly, captures most of the available benefit.
Start there and add complexity only if the simple version is clearly leaving value on the table.