Troubleshooting an underperforming PV array: a diagnostic workflow

An underperforming PV array rarely announces its cause. The symptom is almost always the same — lower-than-expected kWh on a monthly bill or monitoring dashboard — while the underlying fault could be anything from a shading tree that grew two meters in three years to a single failed bypass diode to a grid voltage that has drifted high enough to trigger inverter power curtailment. Diagnosing the actual cause requires a structured process, not a guess, because the corrective actions for these faults range from a five-minute cleaning to a full string replacement, and misdiagnosis wastes both time and money.

This article sets out a diagnostic workflow that moves from cheap, non-invasive checks toward more specific electrical tests, along with a symptom/cause/fix matrix built from the failure modes most commonly reported in field studies of underperforming residential and commercial arrays. The goal is to give an owner or technician a repeatable sequence: rule out the common, low-cost explanations first, then narrow toward instrumented diagnostics only where needed.

Two things make PV underperformance diagnostics harder than most electrical fault-finding: the DC side is live any time light falls on it, which limits what can be safely tested without proper isolation procedure, and the "expected" output against which actual output is judged is itself a modeled estimate sensitive to irradiance, temperature, and soiling assumptions. Getting the baseline right is half the diagnostic work.

Step 1: establish what "expected" actually means

Before troubleshooting a shortfall, quantify it. Compare actual generation over a period (a week or a month, not a single day, to average out weather noise) against a weather-corrected model — either the installer's original yield projection scaled by measured irradiance for the period, or a simple ratio using a nearby reference system or satellite irradiance dataset. A system reporting 8 percent below its nameplate-rated annual estimate without weather correction may be performing exactly as expected in a below-average irradiance year; the same system 8 percent below a properly weather-corrected model is a genuine shortfall worth investigating.

The performance ratio (PR) — actual yield divided by the theoretical yield at measured irradiance and standard test condition efficiency — is the standard metric for this. Well-designed European residential systems typically run at a PR of 0.80–0.87 once inverter and temperature losses are accounted for; a PR below about 0.75 sustained over a season is a reasonable trigger for investigation.

Step 2: rule out the whole-array causes first

Whole-array underperformance (all strings affected roughly equally) points toward causes that affect the system as a whole: inverter clipping or curtailment, grid voltage rise causing inverter power reduction, soiling across the entire array, or a firmware/configuration issue such as an incorrectly set power factor or export limit. String-specific underperformance (one string clearly worse than its peers under otherwise identical conditions) points toward localized causes: shading, a failed component, or connector degradation in that specific string.

This distinction is the single most useful triage step, because it immediately halves the list of plausible causes.

Symptom, cause, and fix matrix

Step 3: instrumented diagnostics

Where the visual and monitoring-based triage does not resolve the cause, move to instrumented tests. An I-V curve tracer applied at string level under stable irradiance reveals characteristic signatures: a stepped curve indicates one or more bypassed cells (often localized shading or a cracked cell); a reduced slope in the linear region indicates elevated series resistance (a common connector or solder-bond fault); a reduced Voc across the whole string suggests either a missing/bypassed module or cell-level degradation distributed across the string.

Thermal imaging under load (irradiance above roughly 500 W/m²) identifies hot spots directly: a single cell running significantly hotter than its neighbors is consistent with a cracked cell or a shaded cell forced into reverse bias; a hot connector or junction box indicates a resistive fault at that connection point.

Worked example: quantifying a shading loss

Consider a 10-panel string, each panel rated 400 Wp, expected to produce 3,400 Wp combined at standard test conditions once system losses of 15 percent (temperature, wiring, inverter) are applied: 4,000 Wp × 0.85 = 3,400 W. Field measurement under near-STC irradiance shows the string producing only 2,850 W — a shortfall of 550 W, or 16 percent.

Visual inspection shows a chimney casting a shadow across the bottom third of two panels for roughly two hours around midday. If those two panels are wired in series within the string and lack module-level optimizers, the bypass diodes activate and each affected panel's contribution drops sharply for that period — consistent with the magnitude of shortfall observed, and consistent with a localized rather than array-wide cause. The fix in this case is not a system fault at all but a design issue: module-level power optimizers or microinverters, or repositioning the affected panels outside the shadow path, would recover most of that loss. See roof mounting done right for shading assessment during the design phase, and the inverter sizing calculator for how string topology interacts with shading tolerance.

When the cause is outside the array

Not every shortfall originates at the module or string level. Grid conditions — sustained voltage above the inverter's configured threshold — cause modern grid-tied inverters to curtail output under volt-watt response rules required by EN 50549 and similar national grid codes. This shows up as a whole-array reduction correlated with grid voltage rather than irradiance, and the fix lies with the distribution network operator or with a voltage-rise mitigation measure at the property, not with the PV hardware. See grid connection and export rules explained for how this interacts with export limits.

Building a diagnostic log

Every diagnostic session should record: date, weather/irradiance conditions, monitoring data for the affected period, visual findings, any instrumented test results, and the corrective action taken. This log does two things: it prevents re-diagnosing the same fault repeatedly, and it builds the evidence base needed for a warranty claim if the root cause turns out to be a manufacturing defect rather than an installation or environmental issue.

Standards and references

• IEC 61829 — On-site measurement of I-V characteristics of PV arrays, the basis for field I-V curve testing methodology.
• IEC 62446-1 — Documentation, commissioning tests, and inspection criteria used as the baseline for comparison during troubleshooting.
• IEC 60364-7-712 — Electrical safety requirements for PV installations, relevant to safe isolation procedure before any live testing.
• IEEE 1547 — Interconnection and interoperability requirements for distributed energy resources, relevant to grid-interaction faults such as voltage-rise curtailment.
• EN 50549 — Requirements for generating plants connected in parallel with distribution networks, the European counterpart governing volt-watt and frequency response behavior.

Sources

• NREL — PV system reliability and field performance research
• PVEducation — module and array performance fundamentals
• Fraunhofer ISE — Photovoltaics Report
• IEA PVPS — PV system performance and reliability task reports

Frequently asked questions

How much output loss is normal versus a sign of a fault?

Year-on-year degradation of 0.3–0.5 percent for crystalline silicon modules is normal and within warranty expectations. A sudden step-change drop, a single string well below its peers, or a sustained performance ratio below roughly 0.75 after weather correction are signs of a fault rather than normal aging, and warrant the diagnostic workflow described here rather than being written off as expected decline.

Can weather alone explain a bad month of generation?

Yes, and it is the first thing to rule out. A cloudy month with irradiance 15 percent below the long-term average will produce roughly 15 percent less energy even from a fault-free system. Comparing to a weather-corrected model, not to the same calendar month in a previous year, is the only reliable way to separate weather variance from an actual fault.

Is thermal imaging something a homeowner can do themselves?

Consumer-grade thermal cameras (including smartphone attachments) can reveal obvious hot spots and are a reasonable first screening tool, but they lack the resolution and calibration of professional equipment used to distinguish a genuinely faulty cell from normal reflective or angular thermal artifacts. Confirmed findings should still be verified by a qualified technician before any module is condemned as faulty.

Does module-level monitoring make troubleshooting significantly easier?

Considerably. String-level monitoring narrows a fault to roughly 10–20 modules; module-level monitoring (via optimizers or microinverters) narrows it to one. For systems with complex roof geometry or partial shading risk, the diagnostic time saved over the system's life often justifies the modest added cost of module-level monitoring at installation.

What is the most commonly overlooked cause of underperformance?

Grid voltage rise causing inverter curtailment. It produces no visible fault, no error code beyond a benign-looking power reduction, and is easy to misattribute to the PV hardware itself, when the actual issue sits on the utility side of the meter and requires a different remediation path entirely.