[ The Science ]

No Magic. No Black Box. Applied Photovoltaic Physics.

This page tells you exactly how SunSniffer's platform works: the measurements behind it, the physical models, the open standards, the references.

If you are an engineer, a scientist, an asset manager, or a technically inclined investor — this page is written for you.

ACTUAL SUNSNIFFER WEBPORTAL SCREEN — NOT A MOCKUP
app.sunsniffer.io/plants/suedstadt-forum/iv-curve
Site analysis · Südstadt-Forum
IV Curve
Charts & StatisticsHeat MapsIV Curve
Module 1.1_1Below warranty curve
installed 05/2013 · 13.1 yr · calculated on 12.6.2026
Degradation
0.89 %/yr
Warranty floor · 0.3 %/yr
182.5 W
Below floor
−14.5 W
I–V characteristic
Module level · AI-fitted from 15 s telemetry
MeasuredFlashlist
MPP 190 WIsc 5.17 AMPP 168 WVoc 44.6 V
Max power P
168.0 Wvs 190.0 W
Vmpp
35.0 Vvs 36.3 V
Isc
5.17 Avs 5.44 A
Voc
44.6 Vvs 45.4 V
00
The foundation · what we measure

We Measure What Physics Tells Us Matters.

Everything starts with knowing what to measure, how precisely, and how often.

Parameter
Where
Frequency
Purpose
Module voltage (U)
Every module
15 seconds
Identifies degradation patterns, shading, faults
Module temperature (T)
Every module
15 seconds
Enables STC normalisation · detects hot cells
String current (I)
Every string
16 Hz (16× / sec)
Captures MPP tracking behaviour · arc events
String voltage (U)
Every string
16 Hz
Cross-validates module sum · detects disconnects
Bypass diode events
Every junction box
Every switching cycle
DiodeVital fire risk + shading diagnosis
Checked before it's used

Every reading is graded on arrival: plausibility bounds, cross-validation against string and plant sums, and communication-quality flags (RS485 · PLC) catch bad data at the source.

Handled automatically

Suspect readings are flagged and excluded from calculations automatically — never silently blended in. Data gaps stay visible, not hidden behind an interpolated line.

Graded, not assumed

Every downstream analysis ships with its own confidence grade — how trustworthy the underlying data was, and how sure the result is. Shown in full further down this page.

This resolution is what laboratory-grade performance analysis requires. Most field systems approximate it at the inverter level. SunSniffer delivers it at every module, every string, continuously — checked, graded, and only then handed to the four engines below.

01
TrueSTC™ · continuously-fitted physical model

How Live Field Data Becomes a Repeatable STC-Equivalent Comparison Value.

Two identical modules measured on different days — different irradiance, different temperature — can't be compared directly. A raw kWh reading tells you almost nothing about whether a module is actually healthy.

TrueSTC evaluates thousands of suitable voltage and temperature measurements from each monitored module. Data quality rules exclude unsuitable operating phases, cloud effects, shading, gaps and outliers before a repeatable STC-equivalent comparison value is produced. The proprietary calculation remains part of the SunSniffer analytics engine.

The result can be compared with a datasheet starting point, an individual laboratory reference, technically similar modules or an agreed performance threshold. Repeated values create the module's performance history.

Compare that to the alternative: a laboratory flash is a high-quality reference — but a snapshot, gone the moment the module is back in the rack. TrueSTC accumulates instead of snapshotting: thousands of suitable readings pile up over days and weeks, building a repeatable comparison value and a living performance history. See how it was validated ↓

When you need the number, it is already there.

The model, not a snapshot
Model
Physical PV cell model · proprietary
Data quality
excludes clouds, shading, gaps, outliers
Input
V + T · 15 s · every module
Self-checking
flags physically-impossible readings
continuously fitted, not a single snapshotflags its own bad inputs1000 W/m² · 25°C equivalent output

TrueSTC is available as a subscription add-on within the WebPortal.

02
DiodeVital™ · Coffin-Manson thermal-fatigue modelling

Catching a Failing Bypass Diode Before It Becomes a Fire.

Bypass diodes are the single most failure-prone electrical component in a PV module. A short-circuit failure is invisible to most monitoring systems until measurable yield loss occurs — by which point the diode may have been under chronic thermal stress for months. A failed diode under reverse current flow is a primary ignition source in solar fires.

DiodeVital continuously analyses the switching cycles, duration, and thermal behaviour of every bypass diode in every junction box, applying Coffin-Manson thermal-fatigue modelling to the accumulated stress history.

It identifies elevated stress before failure — with a number attached, not just a binary fault flag — and flags actual failures in near real time.

What it watches
Coverage
every junction box
Signal
every switching cycle, continuously
Correlation
diode events ↔ module degradation · published research
Output
estimated remaining life + real-time fault flag
junction-box switching studieshotspot precursor detectionreal-time failure flagging
03
PVLib · DOE Sandia National Laboratories

Judged Against the Same Physics the Industry Already Trusts.

Any anomaly-detection system needs an expected-performance baseline to compare against. If that baseline is arbitrary, so are the alerts it produces.

SunSniffer integrates PVLib directly into the analytics platform — the de facto open-source PV performance modelling library, maintained by Sandia National Laboratories with the global PV research community, and used industry-wide to model expected plant performance.

Expected output is calculated for the exact irradiance, temperature and spectral conditions observed at your site. Deviations are flagged as physics-based anomalies — not arbitrary thresholds.

Why it's trusted
Maintained by
DOE Sandia National Labs
License
fully open source
Used by
utilities · engineers · asset managers, worldwide
Role here
expected-yield baseline for every deviation flag
github.com/pvlibDOE Sandia National Laboratoriesindustry-standard expected-yield baseline
04
SunSniffer Engine · open workflow · REST + MCP

None of This Runs Inside a Sealed Black Box.

TrueSTC and DiodeVital are proprietary engines — but a number you cannot interrogate is a number you cannot fully trust, and one you definitely cannot adapt to a plant-specific question.

So the workflow around them is open. Inside the WebPortal, a visual workflow editor lets you see every input and result, adjust thresholds, or compose an entirely new calculation from module, string and weather data — a plant-specific threshold, a custom quality metric, a bespoke model.

Saved results become first-class data, traceable back to the raw measurements — feeding reports, or Ray, directly. No exports, no separate spreadsheet, no mystery step in between.

What's open
Inspect
every input and every result
Adjust or extend
no code required
Access
REST + MCP · no vendor lock-in
Output
first-class data, usable by reports and Ray
visual workflow editorinspect · adjust · extendresults traceable to raw data
05
Ray · grounded, not generative

Ray Doesn't Guess. It Runs the Engines Above and Reads the Numbers Back.

How one question gets answered
“Why is Module 1.7_18 underperforming?”
Ray calls the real engines — not its own memory
true_stc(module, day)diode_vital(module)pvlib.run_model(plant)
Every number traces to a raw 15-second reading
TrueSTC: 94.1% of nameplateDiodeVital: normalPVLib expected: 412 kWh → actual 379 kWh
Ray's answer — built only from the above
8% shortfall vs. expected, diode health normal — soiling is the likely cause. Cleaning ROI: 11 days.

Ray is not a general chatbot answering from what it remembers. Every question is answered by calling TrueSTC, DiodeVital, PVLib or your own SunSniffer Engine workflows — live, over MCP — and narrating only what those calls return.

This is the same architectural choice behind everything else on this page: no step between measurement and answer is hidden. Ray's job is narration and synthesis of real tool output — not invention.

That is also why every answer can be traced back to the raw 15-second measurements behind it, the same way every engine result can — shown next.

Ray works from approved measurements, documented analytics and defined economic or contractual rules. Missing information is shown as a gap; the responsible professional remains in control of the decision.

If the data behind a question is too thin — too few clean readings, no clear-sky day yet, a gap in the string — Ray says so, and tells you what's missing. It does not fill the gap with a plausible-sounding guess.

The same confidence grade ships with every answer
Data Quality
93.4/100 avg
32 days with low-trust readings · 8 modules affected
Peer-Relative Performance
0.8%avg loss
272/287 modules with full MPPT context (94.8%)
How It Was Analysed
87.8%combined methods
11.9% string-peers only · confidence graded per finding
Diagnostics
253modules
showed expected-output loss with strong-enough evidence to flag
TrueSTC™ Validation

Three Measurement Campaigns. Three Different Questions. One Consistent Result.

The available evidence does not compress different measurement tasks into one artificial accuracy number. It separately tests absolute laboratory agreement, transfer of an individual reference to a changed state and detection of soiling losses against an independent field reference.

Campaign
Evidence
Result
Correct interpretation
Direct ZAE laboratory comparison
6 normally operating modules, 12 comparisons
0.82% MAE, max. 1.94%, all within ±2%
Observed agreement with laboratory flasher values
One-point reference transfer
20 modules, no exclusions
+0.60% campaign mean, 2.27% module MAE, 19/20 correct direction
Reference transfer and change reconstruction
External anonymized customer study
84 sensorized modules, 6 strings, independent Dust Detection System
0.89 percentage-point RMSE, approximately 0.9%
High-accuracy detection of soiling losses

Across three measurement campaigns, TrueSTC demonstrates very good typical agreement with laboratory flasher values, robust transfer of individual reference values to changed operating states and high-accuracy detection of soiling losses.

Different campaigns, different measurement tasks — no pooled product-accuracy figure.

The full source list behind these figures
Sockelblindheit · the field-recognition limit

You Can Measure the Loss. You Cannot Tell Anyone Where to Go.

German engineering has a word for what happens after a loss is confirmed and a crew is sent into the plant: Sockelblindheit — standing in front of a hundred modules with nothing to point at.

01 · What the inverter shows
One number per MPP input. A typical string inverter combines four strings of 28 modules — 112 modules behind a single figure; many combine eight strings or more, and a central inverter runs one MPP for every string it owns. The figure is never grounded against irradiation and temperature, so a real loss and a dull afternoon read the same.
02 · What software adds
Calculated properly against irradiation and temperature, monitoring can establish that a loss exists — and how large it is. It still cannot say where it sits.
03 · What the field can do
On site, a module losing less than 5–8% cannot be recognised at all. The range depends on the measurement method and the skill of the inspector — ours was established with physicists from the Helmholtz Institute doing the checking.
Why the loss hides in the average — and shows at the module

Modules in a string are in series: one current flows through all of them, and the strings on one MPP input of the inverter sit at one voltage. When a module weakens, the string current is forced through it anyway — and the loss appears as a drop in that module's voltage. The power loss is real, measurable at the module, and unattributable in the MPP tracker's number.

With many weak modules in a string the effect slides: at some point the accumulated loss pulls the string current itself down. The signature changes — the missing address does not.

That is why SunSniffer measures voltage and temperature at every module: the loss surfaces exactly where it occurs.

What locating the loss recovers
Thermography · Indication and Blind Floor

Thermal Imaging Can Indicate a Problem. It Cannot Continuously Measure the Electrical Loss.

Thermographic results depend on the operating and viewing conditions of the flight. A different angle, irradiance level, temperature state, wind condition or reflection can change the image. Each campaign also requires image processing and reliable mapping of findings to physical module positions.

8.3%
uncertainty in thermography-derived power-loss estimates
< 5%
electrical losses below this were not reliably identified

In a published Helmholtz comparison, power-loss estimates derived from thermography carried 8.3% uncertainty. Electrical losses below 5% were not reliably identified by the thermal assessment criteria.

Source: Teubner et al. (2017), Comparison of Drone-based IR-imaging with Module Resolved Monitoring Power Data, Energy Procedia 124, pp. 560–566 · in our reference list

Below the thermal threshold, the loss is not underestimated. It is not seen.

See the full drone-vs-continuous comparison
Independent validation

Validated by People Who Had No Reason to Be Kind.

Independent verification from institutions and operators with rigorous testing standards.

ZAE Bayern
The laboratory flasher benchmark for TrueSTC
Module flasher measurements at ZAE Bayern provided the laboratory reference for the twelve-comparison TrueSTC campaign — all results within ±2%.
Helmholtz HI-ERN
0.82% mean absolute deviation — and the field-recognition limit
Independent flasher measurements at the Helmholtz Institute Erlangen-Nürnberg validated TrueSTC against a laboratory reference — an external result produced independently of SunSniffer.
Standards context

Built on the Frameworks the Industry Already Uses.

These frameworks provide technical context for PV performance measurement, commissioning and rapid-shutdown interfaces. Product-specific conformity depends on the applicable documentation and project configuration.

Standard
What it covers
Relevance to the workflow
IEC 62446-1
Grid-connected PV systems — commissioning requirements
Defines the documentation and verification context for commissioning.
IEC 60891
Procedures for temperature and irradiance corrections
Provides procedures for correcting PV measurements for temperature and irradiance.
IEC 61215
Terrestrial PV modules — design qualification
Defines the module qualification baseline against which later field performance can be compared.
IEA PVPS Task 13
Performance and reliability of PV systems
Publishes methods and reliability research relevant to PV performance analytics.
NEC 690.12
Rapid shutdown of PV systems on buildings
Defines rapid-shutdown requirements for applicable US building installations.
SunSpec TX
Interoperable rapid shutdown communications standard
Provides an interoperable communication framework for compatible rapid-shutdown equipment.
Data sovereignty

Your Data Stays in Europe. Period.

All SunSniffer data processing and storage takes place on European servers.

This is not a marketing claim — it is an architectural choice, reflecting the platform's foundation in a country where data sovereignty and engineering precision are taken seriously. GDPR-compliant. Designed for the bankability and audit requirements that institutional solar operators face.

Architecture principles
EU-based infrastructure
All processing + storage
GDPR-compliant
By design, not retrofit
Open standards
IEC · IEA · SunSpec · PVLib
REST + MCP APIs
No vendor lock-in
SCADA interoperability
Modbus · OPC UA · REST
For the technically curious

References, Documentation, and Deeper Reading.

IEA PVPS
  • Task 13 reports on operational performance
  • PV module classification framework
  • Clear-sky methodology documentation
Published evidence
  • Teubner et al. (2017) — thermography comparison
  • TrueSTC — three documented measurement campaigns
Open Standards
  • PVLib — github.com/pvlib
  • SunSpec Alliance specifications
  • NEC / IEC / IEEE published standards

Full technical documentation, whitepapers, and peer-reviewed references are available on request.

Engineering conversation

If It Works on Paper, It Works on Your Site.

We are happy to talk to engineers, scientists, asset managers and consultants at any depth of detail. The science is open. The measurements are real.

Talk to Our Engineering Team Request Technical Documentation