Where to get open solar and battery fault data
A working index of the public datasets behind fault detection and degradation modeling in solar and storage: what each covers, how to access it, and which we have benchmarked.
The most useful open datasets for PV and battery fault work are Lazzaretti (5-class PV faults, CC BY 4.0), GPVS-Faults (8-class grid-connected PV), Severson 2019 (LFP battery cycling to end of life) and NASA PCoE (NMC battery aging). We benchmarked all four. Other open sources cover PV production time series and infrared module imagery but are not yet benchmarked here.
The datasets we benchmarked
These four are openly licensed, widely used, and reproducible. The benchmark column links to our own held-out results in the companion report, so you can see what a standard model achieves before you invest in one.
| Dataset | Domain | Scope | License / access | Our result |
|---|---|---|---|---|
| Lazzaretti UTFPR | PV faults | 5 classes, ~497K labeled rows, 5 kW site | CC BY 4.0 | 0.998 macro-F1 |
| GPVS-Faults | PV faults | 8 classes across 2 modes, ~320K rows, grid-connected | Open (Mendeley Data) | 0.766 macro-F1 |
| Severson 2019 | LFP battery | 124 A123 cells cycled to end of life | Open (data.matr.io) | 17.2% MAPE |
| NASA PCoE | NMC battery | 13 well-formed cells (of 42) | Public domain (NASA) | 29.5% MAPE |
Other open datasets worth knowing
These are pointers, not benchmarks. We have not independently validated the specifics below, so treat the descriptions as a starting point and confirm details at the source before you rely on them.
| Dataset | Domain | What it offers | Access |
|---|---|---|---|
| NREL PVDAQ | PV production | Multi-site inverter and string time series | Open (NREL) |
| DKASC Alice Springs | PV performance | Multiple module technologies, long time series | Open |
| Raptor Maps IR modules | PV thermal | Labeled infrared module images across anomaly types | Open (GitHub) |
| Battery Archive | Battery cycling | Aggregated public cell datasets across chemistries | Open |
What is missing
Open data skews to lab conditions or single sites. There is very little open, labeled, multi-site utility-scale fault data, and almost no open desert or MENA soiling ground truth. That gap is exactly why cross-dataset generalization matters: a model that only works on the site it was trained on is not much use to an operator with a portfolio the training set never saw.
Get the raw dataset
The full index as a CSV: dataset, domain, scope, license and access, and whether we benchmarked it. Free, in exchange for an email so we can tell you when we add datasets.
Frequently asked questions
See also
See this on your own plants
NuraVolt turns your SCADA and BMS data into early fault detection, degradation-aware BESS analytics, and audit-ready reporting. A fixed-scope audit shows you what we’d find on your portfolio.