We respect your privacy
Mneme-AI only uses cookies strictly necessary to operate the platform. You can customise your preferences at any time. Learn more
Why predictive maintenance is a game-changer
Corrective maintenance (repairing after failure) costs three to five times more than preventive maintenance (scheduled servicing). Predictive maintenance goes one step further: anticipate failures before they happen, by reading the weak signals sensors emit continuously.
Historically reserved for heavy industry (turbines, production lines), it becomes accessible to buildings thanks to three converging forces: affordable IoT sensors, AI embedded in modern BMS platforms, and protocol standardisation. Sensia integrates four native predictive algorithms on every supervised equipment.
The four native predictive algorithms
RUL (Remaining Useful Life) estimates time-to-failure by linear regression on 90 days of health score. Typical precision ±15% at 30 days.
MTBF (Mean Time Between Failures) computes the mean time between failures per equipment class, from workflow transition history (status triggered → resolved).
Unsupervised anomaly detection flags behaviours deviating from baseline without manual labelling — a drifting temperature sensor, an overnight electrical consumption spike.
Composite health score aggregates these three metrics plus active alarms state into a 0-100 score per equipment, readable at a glance in the Pulse panel.
How these algorithms run in production
Two pipelines coexist. The polled pipeline runs daily, full org × equipment scan, asyncio Semaphore(10) parallelises up to 10 organisations. For each equipment: component computation (age, work orders 90d, alerts 30d, anomalies, utilisation), 0-100 score, history snapshot, RUL estimation, MTBF/MTTR, persist on entity_record.attributes, LLM-enriched recommendation if at-risk.
The event-driven pipeline triggers on the twin.{thing_id}.changed firehose (transitions, sync bursts, alerts), debounced 60s per equipment, sub-minute latency after change. The polled stays as a safety net.
Native integration in Pulse and ThingPanel
Predictive metrics show up directly in the « Reliability » section of the Pulse dashboard: MTBF, MTTR, Availability per equipment fleet, with drill-down on the top 5 at-risk equipment. The ThingPanel (equipment side popup) displays a dedicated RUL widget: estimated remaining time, confidence level, LLM-generated contextual recommendation (« replace », « repair », « monitor » with justification).
No separate tab to learn: predictive maintenance integrates into existing pages.
Wire to your rule engine
The rul_days variable is automatically exposed to the rule engine. A typical « End of life approaching » rule configures in two clicks: rul_days < 7 fires a critical alert, rul_days >= 14 clears it (anti-flapping hysteresis). Same for health_score, mtbf_days, availability_pct.
You combine these variables with fleet conditions (only the AC units in the Marseille building, for instance) to target your alerts.
Included in the Hypervision range
Full predictive maintenance (RUL, LLM recommendations) is included in Sensia Hypervision, priced per m² per year and published openly: €4,500/yr flat below 3,000 m², then €1.50/m²/yr, and €1.10/m²/yr above 25,000 m².
Sensia Compliance is limited to base reliability metrics (MTBF, MTTR, availability), bounded to the regulatory scope, with no predictive RUL and no AI layer.
Models trained on your historical data and a dedicated support SLA fall within the bespoke Enterprise tier (€2.50–4/m²).