Asoba Asoba
Predictive Fault Detection

Catch the fault.
Skip the noise.
Get the fix.

Static alarms can't tell a cloud from a fault — so teams drown in false alarms until they stop looking. We build a behavior model for each inverter, score deviations as they happen, and return a cited fix pulled from your own OEM documentation.

No site visit No new hardware Runs on-prem
Per-Inverter Behavior Model Live telemetry · predicted vs. actual state Live model 100 75 50 25 0 kW Streak: 6 · Inverter 3 00:00 06:00 12:00 18:00 24:00 Severity: Critical Predicted vs. actual: −235 kW Diagnosis: Nehanda v3

Four Steps. No Manual Triage.

From raw telemetry to a cited fix — without touching a dashboard between steps.

  1. 01 · Learn

    Model normal behavior

    Each inverter learns its expected output conditioned on local weather, tilt, and shade.

  2. 02 · Score

    Score the deviation

    Output gaps become a severity score and streak length, separating true drift from passing clouds.

  3. 03 · Ground

    Reference OEM manuals

    High-confidence faults are matched against manufacturer manuals and repair history with verified citations.

  4. 04 · Route

    Equip your technician

    Recommendations route directly into work orders. Your team reviews and approves every maintenance action.

Proof of work

Tested on a Real Fleet

Seven inverters. 2.31 MW. Six months of production data. Benchmarked directly against the site's existing rule-based alerts.

  • 216 labeled faults caught on the fleet's highest-signal inverter — 41–56% precision and recall lift over rule-based thresholds
  • Two inverters flagged as weak-signal — traced to a data-backfill gap before the signal was forwarded to the operations team
  • Same platform and security model for the pilot as for ongoing monitoring
Maintenance intelligence screenshot

Free 3-Week Diagnostic Pilot

See Your Fleet's False-Alarm Rate Before You Commit

Send at least three months of inverter or SCADA production data (15-minute intervals ideal) and basic site details. We'll tell you within the first week whether your data clears the bar to start — and if it doesn't, we'll say so up front.

  • No on-site visit, no new hardware, no engineering time from your team
  • No cost, and no obligation to continue afterward
  • Your data stays in-region and isolated for the duration of the pilot; request full deletion anytime after
Three Weeks, Three Outcomes
WEEK 1 — Find the Signal
We build a per-asset behavior model from what you send us and hold it up against actual output.
WEEK 2 — Rank the Faults
Every flagged asset is scored and ranked by severity, so you know exactly where to look first.
WEEK 3 — Show the Number
You get your false-alarm reduction rate and a ranked, cited list of what to check first — no obligation to continue.

Runs on Your Site. Your Data Never Leaves.

For teams with strict data residency or connectivity constraints, the platform deploys on-prem as a signed bundle — no cloud account, no registry login, no changes to your firewall.

Deployment

On-Prem · No Cloud Required
  • Ships as a signed bundle — installs on bare metal or VM (Ubuntu 22.04), no internet connection required after install
  • No registry login, no git clone, no cloud account — your security team doesn't need to open a new approval process
  • Optional GPU support (4+ GB VRAM) for faster on-site ML model training

Data & Governance

Local-First · Audited
  • All telemetry, model artifacts, and audit logs stay on your machine — Postgres and local filesystem, nothing external
  • RBAC, JWT auth, and a full audit trail built in — every access and override is captured
  • ISO 27001-mapped security test suite ships with the platform; run it against your own deployment

Also available as a managed cloud service — same platform, same validation pipeline, data processed in-region.

Find out how many of your alerts are real.

Three weeks. No cost. No obligation. No on-site visit required.

Get Help & Stay Updated