SolarCycle AI: Predict failures. Plan collections. Recover value.

2The solution

Predict the failure, then plan the smartest collection

Telemetry becomes a calibrated fault risk score, a fault type, a collection job, and a verifiable record, automatically.

◌ Not live data

The asset health data on this page is illustrative, not live. Telemetry comes from a public benchmark dataset (Lotus PV Fault Benchmark) sourced from the internet, not a real asset. No live IoT feed is connected.

01Sense

Ingest inverter health telemetry

→

02Predict

Risk score + end-of-life window

→

03Plan

Capacity-aware collection route

→

04Prove

Append verifiable passport event

Wyndham Rooftop Cluster

Wyndham · PV benchmark telemetry · demo fallback

connecting…Urgent

Stage 1 risk

88%

Fault class

degradation

Type confidence

91%

Severe anomaly detected (degradation). Inspect immediately. Normal confidence 12%.

Fault consoleSynthetic · Lotus benchmark

Degradation

Synthetic data disclaimer: "Wyndham live" is generated from a synthetic asset whose values follow the Lotus PV Fault Benchmark degradation pattern. All other scenarios are real rows from the Lotus benchmark dataset (one row per fault class). No live IoT feed is connected.

String 1 DC voltage257.84V

String 2 DC voltage272.45V

String 1 DC current7.81A

String 2 DC current5.90A

Irradiance801W/m²

Module temp46.02°C

Fault-type probabilities

short2%

degrade91%

open3%

shadow4%

Two-stage XGBoost · live prediction

Sensor input

Vdc1257.8 V

Idc17.81 A

Vdc2272.5 V

Irr801 W/m²

6 features→

Stage 1 · Binary

88%

Faulty

88% fault conf.

classify→

Stage 2 · Fault type

Degradation

91% confidence

XGBoost multiclass

Sensor input

Vdc1257.8 V

Idc17.81 A

Vdc2272.5 V

Irr801 W/m²

↓

Stage 1 · Binary

88%

Faulty

88% fault conf.

↓

Stage 2 · Fault type

Degradation

91% confidence

XGBoost multiclass

Digital passport

manufactured2013-02-01
Inverter manufactured (model MS-5K).
installed2013-04-12
Installed at Wyndham Rooftop Cluster.
health checked2026-06-06
Telemetry ingested: temp 80C, THD 8.2%, efficiency 88.5%.
fault predicted2026-06-06
Risk 0.72 -> likely_breaking (overheating). Failure window ~21 days.
Tamper-evident · hash-chained

Inverter health trend (featured asset)

Wyndham synthetic session · not real sensor data

Last 12 hours of telemetry

Failure risk (×100)Conversion efficiency (%)

Under the hoodthe algorithms that turn data into a plan

01Predict

Two-stage XGBoost

Normal/faulty risk, then fault type

P(\text{faulty}) \rightarrow \text{class confidence}

02Filter

Greedy knapsack

Pack stops under truck mass capacity

\sum \text{mass} \le 2{,}500\ \text{kg}

priority-first

03Optimize

Exact brute-force TSP

Shortest depot → stops → centre

\min_{\text{perms}} \sum d_{i,i+1}

greedy NN fallback

04Measure

Haversine distance

Great-circle km between sites

2r\,\arcsin\!\sqrt{\operatorname{hav}(\Delta\varphi,\Delta\lambda)}

● Realsolar lifespan assumptions (25-yr panels) & the end-of-life cohort from CER install-age data

◌ Illustrativethis asset's identity, telemetry adapter, model score & passport events (no public live-IoT feed exists)

Source: two-stage calibrated XGBoost model trained on the public PV fault benchmark; lifespans per manufacturer / Sustainability Victoria guidance

Predict the failure, then plan the smartest collection

Solution

Predict the failure, then plan the smartest collection