Platform

Seven modules. One asset record.

Every module reads and writes the same record of sites, blocks, strings, findings and work orders, so a fault found by one module is visible to all of them.

Pilot

Solvane Vision

Defect detection for RGB and thermal drone imagery. Vision segments every module in the frame and flags those that are statistical outliers against the rest of the array, then proposes a finding with a location, evidence and a confidence score.

  • Thermal: dead strings, bypass-diode failures, hotspots
  • RGB: cracks, soiling, vegetation encroachment
  • On-device by default; a cloud vision model adds 11 classes and written explanations
  • Every finding stays a proposal until an engineer confirms it
See it working ›
Input
RGB / rendered thermal JPEG, PNG, WebP
Output
Type, bbox, severity, confidence, modules, safety, action
Classes
6 on-device · 11 with cloud model
Speed
~0.2 s per 1,600 px frame in the browser
Benchmark
24/24 planted defects, 0 false positives (synthetic)
Pilot

Solvane Strings

Unsupervised anomaly detection on string-level telemetry. Each string's daily yield is divided by the block median, which removes weather, and scored with a robust median/MAD z-score over a rolling 7-day window.

  • Flags strings below z = −3.5, with how far off and since when
  • Links a flagged string to the matching image finding
  • Creates SCADA-only findings for faults the drone missed
Input
Daily kWh per string (CSV, or live via OPC UA / Modbus)
Method
Weather-normalised PR, median/MAD robust z
Window
7 days, 30-day history shown
Output
Strings flagged, z-score, deficit vs peers
Pilot

Solvane Value

The loss and priority model. Each finding is converted into kWh and dollars per year from modules affected, module rating, specific yield, a defect-type loss factor and your PPA price. It is then scored 0–100 from severity, confidence, safety and value.

  • P1 ≥ 75 · P2 ≥ 55 · P3 ≥ 35 · P4 below, with 2 / 7 / 30 / 90-day SLAs
  • Critical findings are always at least P1
  • Telemetry-measured losses override the default factor
Hotspot
15% of module output
Bypass diode
33% (one substring)
String outage
100% of the string
Cell crack
8%
Soiling
6%
Vegetation
12%
Pilot

Solvane Copilot

Ask questions of the whole plant: what to fix first, what changed since the last flight, which strings are underperforming. Answers cite finding IDs, and Copilot says so when the data can't answer.

  • Offline mode answers from the record with no model
  • Language-model mode is grounded in the same record
  • Separates unreviewed proposals from confirmed findings
Q · What should the crew fix first? 1. F-3 Dead string, Block B1 (P1, ~$1.5k/yr) 2. String S07 at 69% of peers, Block A2 (P2) 3. F-2 Bypass diode, Block B1 (P2, safety risk)
Pilot

Solvane Work Orders

Every confirmed finding becomes a work order with a priority, an SLA due date, an assignee and the evidence attached. Track open → in progress → done, export to your CMMS, or print a signed inspection report.

  • CSV for Maximo, SAP PM and Fiix; native sync on Portfolio
  • Overdue flags against each priority's SLA
  • Inspection report with every finding and loss figure
WO-1003
P1 Dead string · B1 · due in 2 days
WO-1002
P2 Bypass diode · B1 · due in 7 days
WO-1004
P3 String S18 IV-trace · B1
WO-1001
P3 Hotspot · B1 · in progress
Preview

Solvane Fleet

Portfolio view: revenue at risk, open P1s, median time to repair and inspection coverage, compared across every site and owner. Built for asset managers who report to investors.

Ships to pilot partners next. Fleet uses the same per-site measures as the live demo, rolled up across sites.

Preview

Solvane Forecast

Uses multi-season inspection history and string trends to show where degradation is accelerating, such as a block whose hotspot count is growing or a string that is slowly drifting, before it shows up in yield.

In development with pilot partners who have at least two seasons of inspection data.

See all five Pilot modules working today.