FleetTwin AI fuses CAN/ECM telemetry, vibration, oil analysis, maintenance history, technician notes, DPF/SCR events and ambient conditions into a single trained model — giving haul trucks, loaders and support vehicles a digital twin that thinks.
The Cost of Reactive Maintenance
The Foundation Model
FleetTwin ingests the full sensor and operational picture of your fleet — not just isolated alarms — training a shared representation that understands failure modes across makes, models and site conditions.
Data Modalities Ingested
System Architecture
Decision Engine
Four AI-driven outputs that turn raw sensor data into actionable maintenance intelligence.
Probabilistic fault forecasting up to 72 hours ahead. The model issues a confidence-weighted alert for each subsystem — powertrain, hydraulics, drivetrain, emissions — so maintenance teams can act before the truck parks itself.
Component-level RUL estimates with confidence intervals — not just binary alerts. Know when your engine, transmission, hydraulic pump or brake assembly will require attention, expressed in hours, cycles and calendar days.
SHAP-based feature attribution plus LLM-generated technician summaries. The model explains why it raised an alert — which sensors, which patterns, which historical events contributed — in plain language your workshop team can act on.
Risk-weighted scheduling that accounts for fleet availability, shift patterns, production targets and parts lead times. Cluster concurrent jobs across vehicles to minimise total downtime impact on the mining schedule.
Under the Hood
A time-series transformer processes CAN/ECM streams at up to 10 Hz alongside lower-frequency signals (oil analysis, vibration spectra). Cross-modal attention layers learn which sensor combinations are predictive for each fault class. The model is pre-trained contrastively on data from 40,000+ machine-years across 12 OEM platforms before site-specific fine-tuning.
Free-text maintenance notes, fault descriptions and inspection comments are encoded by a domain-adapted language model trained on mining maintenance corpora. These embeddings are fused at the same layer as sensor data, allowing the model to learn from the tacit knowledge that never appears in telemetry.
Diesel Particulate Filter loading curves and Selective Catalytic Reduction efficiency trends are modelled jointly with duty-cycle and ambient temperature. The model detects abnormal regeneration patterns, DEF quality degradation and SCR catalyst aging — avoiding the Tier IV downtime that catches most sites by surprise.
Altitude, ambient temperature, dust exposure index, road gradient and haul-road surface condition are incorporated as context tokens. Models trained at sea-level copper mines automatically adapt to high-altitude Andean sites without retraining from scratch, using learned environmental embeddings.
The inference engine runs as a containerised service on AWS/Azure or on-premises. An optimised edge variant (ONNX, INT8 quantised) runs on Nvidia Jetson or industrial PCs at the pit rim for sites with limited connectivity. All variants share the same model weights; inference results sync when connectivity is restored.
Fleet Coverage
FleetTwin ships with pre-trained adapters for the most common mining asset classes. Custom OEM integrations take days, not months.
Use Cases
Replace fixed-interval PM schedules with condition-triggered work orders. Extend component life by 15–30% while reducing unnecessary teardowns and inspection labour.
FleetTwin monitors the entire fleet in real time — alerting the on-call supervisor only when a high-confidence, time-critical event is detected. Reduce night-shift call-outs by 60%.
RUL estimates feed directly into parts ordering logic. Reduce emergency freight costs by pre-positioning components weeks before the predicted replacement date.
Feed predicted fleet availability into the short-term mine plan. Scheduler knows 48 hours in advance which trucks will be offline and adjusts haulage routes and dig-targets accordingly.
Identify hydraulic brake fade and steering-system drift patterns before they reach critical levels — protecting operators on high-grade haul roads from catastrophic mechanical failure.
Auto-generated weekly PDF and live dashboards for GMs and maintenance managers. MTBF trends, cost-per-tonne analysis, and fleet-wide reliability benchmarking across sites.
Get Started
Request a live demo with your own fleet data — or schedule a technical deep-dive with our mining AI team.