Predictive Fleet Maintenance with CAN/OBD (2025): A Starter Guide
Read time: ~7 min
Last updated: 30 Oct 2025
Use the engine data you already collect to prevent failures, plan workshops, and cut breakdowns. Start with simple rules, add CAN depth when available, and automate tickets.
Why predictive maintenance beats reactive
- Reduce roadside failures and tow costs.
- Shift to planned service windows to protect revenue routes.
- Extend component life with data‑driven thresholds.
The CAN/OBD signals that matter
Start with OBD‑II PIDs available on most vehicles, then enrich with CAN where supported:
| Source | Signal | Why it matters |
|---|---|---|
| OBD | Coolant temp, oil temp | Overheat/viscosity stress |
| OBD | RPM, load, throttle | Work intensity & patterns |
| OBD | Fuel rate (calc) | Injector health; MAF sanity |
| CAN | DPF Soot, SCR/NOx | Emission system failures |
| CAN | Brake wear % | Service forecasting |
| CAN | Battery SOC | 12V/EV pack health |
Starter rules before ML
Start with OBD‑II PIDs available on most vehicles, then enrich with CAN where supported:
- Overheat: Coolant > 105°C for > 30s → reduce load; schedule inspection.
- DPF risk: Soot > 80% → force regen; sensor check.
- Braking stress: Harsh decel > 0.35g repeatedly → coach; inspect pads.
- Battery drain: Pack SOC drop > Y%/h parked → 12V/pack check.
When to add ML (and how)
- Pick a horizon (7/14/30‑day failure likelihood).
- Use features: temp deltas, idling %, load profiles, regen cycles.
- Validate against workshop tickets: precision/recall, lead time.
Your first pipeline
- Inventory vehicles, ECU types, and data sources.
- Build a signal map (CSV). - download
- Stream data at 1–5s; store summaries per trip/day.
- Apply starter rules; route alerts to a “Maintenance Tickets” queue.
- Weekly review with workshop to tune thresholds.
Turn alerts into action
- Use webhooks to create tickets in your maintenance app.
- Attach clip/telemetry context to each ticket.
- Auto‑schedule service windows; notify drivers.
KPIs to track
- Mean time between failures (MTBF)
- Unscheduled vs. scheduled service %
- No‑fault‑found (NFF%)
- Parts spend vs. mileage
Copy‑paste assets
💬 FAQs
Do we need CAN for predictive maintenance?
No—start with OBD PIDs and add CAN as available for better accuracy.
How fast should we sample?
1–5 seconds is a good balance for most signals.
What about false positives?
Add persistence windows (e.g., 30s) and combine multiple signals.
Can drivers contribute?
Yes—add a symptom prompt (“noise/vibration/smell”) in the app.
How do we prove ROI?
Track unscheduled vs scheduled %, MTBF, and workshop labor hours saved.
Glossary
- Headway: Following distance measured in seconds.
- Event taxonomy: Standardized codes for coaching and analysis.
- Hash: A cryptographic signature proving a clip wasn’t altered.
Share
LinkedIn
X