Predictive maintenance
Predictive maintenance flags which vehicles are worth servicing sooner, before their condition degrades into a breakdown — built from signals the system already collects, not from a separate report. The output is a planning suggestion, never an automatic action: a person still decides whether to raise a schedule or a work order.
Who it's for
The maintenance manager, fleet manager and organization administrator see the ranking and its trend, each entry carrying the resulting suggestion flag. The model's decision-threshold values themselves (below) are configured — and visible — only to the fleet manager and organization admin; the maintenance manager has no access to them at all, since it's an organization-wide policy call.
How the score is computed
Every night, before the fleet-wide anomaly rollup runs, the model rescores every in-service vehicle from three independent signal families: open fault codes (count, highest severity, how often they recur), the usage trend (distance driven in the current 30-day window against the preceding one), and service history (how many schedules are currently overdue and how many days since the last recorded service). A vehicle with no signal in any of these families never enters the ranking — nothing is fabricated where there's no data.
Each ranking entry carries an urgency score (0–100), a confidence level (0–1, rising with how many signal families actually had data) and a plain-language rationale — a short explanation of what drove the score. The ranking is fully rebuilt every night, so it always reflects the state as of the last run rather than an accumulated history of entries.
Thresholds and the service-planning suggestion
A ranking entry is flagged as suggested for planning when it clears both a score floor (70 by default) and a confidence floor (0.5 by default) at once — a single, maximally strong signal from one family isn't enough; the suggestion must be corroborated by more than one source. A separate, lower score floor (60 by default) decides which vehicles surface in the fleet-wide anomaly rollup. An organization can tune all three thresholds to its own risk tolerance — the model's underlying formula stays unchanged.
From prediction to work order
From the ranking you can open a new schedule or a new work-order form for the flagged vehicle directly — the form opens pre-filled (vehicle, a starting title, source "preventive maintenance"), but you confirm it and adjust anything needed before saving. No work order or schedule is ever created automatically from the prediction alone.
On the vehicle detail page, recent score and confidence runs show as a trend over time — the same underlying mechanism as the ranking, scoped to one vehicle (see Vehicles).
In the background
The nightly run (2:15, before the fleet-wide anomaly rollup at 2:45) rescores the whole organization and appends one trend history point per scored vehicle; points older than the organization's configured retention window are pruned in the same run.