Why your commercial solar system is under-earning

A solar system that was switched on and then left alone looks healthy from the ground. The panels are up, the inverter shows a light, the meter ticks over. None of that tells a finance or facilities director whether the asset is still earning what it was bought to earn. Underperformance is quiet by nature: a dead string the inverter keeps reporting around, an inverter derating on heat, shading that crept in as a tree or a neighbouring unit grew, soiling nobody has cleaned, a monitoring feed that died a year ago. The bill drifts back up, but nothing flags as a fault.

This audit is aimed squarely at that owner's problem. It is for a live system you already operate, and its job is to separate a real, fixable loss from an ordinary weak-light period, then put a cause and a priority against anything that is genuinely wrong. The diagnostics run through Solar Tech Support, the group's independent, brand-agnostic diagnostics and repair arm, so the read is on the equipment in front of us whoever made it and whoever installed it. If you are buying, funding or disputing a system rather than running one you own, the broader independent audit is the document written for that, and this page stays out of its way.

The hardest part of judging an owned system from the bill alone is that solar output swings with the weather and the season, so a genuine fault and a run of dull months can look identical on a meter. An owner needs to know which one they are looking at before spending a penny putting it right.

That is the question we settle first. We establish what a correctly working version of your system should have produced over the period in question, built from your real site rather than a generic yield-per-kWp rule of thumb: the array size and orientation, the pitch of each roof plane, the shading as it stands today and the location, run as a half-hourly model of the year. Set against your actual generation over the same window, that tells you whether the shortfall is real or whether the system did about what the weather allowed. Where the data only covers part of the period, we say what can and cannot be concluded from it rather than filling the gap with an invented figure.

When the loss is real, it is rarely one dramatic failure. It is usually a handful of smaller losses stacking up, and the value to an owner is naming each one and ranking it, so a limited maintenance budget goes to the change that buys back the most generation. We attribute the shortfall across the usual commercial-roof causes.

• A faulty or dead string: a tripped or failed string takes a slice of the array offline while the inverter keeps reporting, so the loss hides until output is read string by string.
• An inverter at fault or throttled: clipping, derating on heat, sitting on a logged error nobody read, or failed on one channel while the rest carry on.
• Shading creep: shade that was not there at commissioning. A grown tree, a neighbouring extension, new plant on an adjacent roof, dragging down the strings they touch.
• Soiling: dust, lichen, moss and bird fouling that build gradually and shave output without ever flagging as a fault.
• Blind monitoring: a dropped data feed, which is often why nobody noticed the underperformance in the first place.
• Genuine degradation: the normal, gradual decline panels are expected to show, separated from the faults so you know what is wear and what is recoverable.

Each loss is sized as far as the evidence allows and put in priority order, so the report reads as a plan an owner can act on rather than a list of complaints.

Most of this can be settled from the system's own evidence, which keeps the audit cheap for an asset you already own. We start with the data the system holds: the metering, the inverter logs and channel-level records, and whatever monitoring history exists. For many systems that is enough to locate the loss, because a dead string or a derating inverter leaves a clear signature once you know the shape of a healthy curve.

Where the data cannot settle it, we go to the roof. Our own qualified, insured pilot can fly the array to read the shading and the panel condition from the air without a roof-access scaffold, and where a fault needs confirming at the panel or string, that hands-on work runs through Solar Tech Support. The rule is to use the cheapest accurate method first and escalate to a site visit only where the question genuinely needs one, so you are not paying for a survey to find something the data already told us.

You cannot judge an array on a single low month, because the weather varies more than any fault does. Underperformance means the system is sustaining less output than it should once the weather is accounted for, and two metrics carry that judgement.

Specific yield is the actual AC energy a system produces per unit of installed DC capacity over a period, in kWh per kWp. It lets your site be compared against its own modelled figure, season by season. National orientation figures are only a sanity check, because a number that looks low against an average can be entirely correct for a north-east-facing roof in a cloudier region. Performance ratio is the weather-normalised metric, defined formally in IEC 61724-1. It expresses actual AC output as a proportion of the theoretical output for the irradiance the site actually received, so the weather is stripped out of the comparison. A performance ratio of 0.80 means a fifth of the available solar energy was lost along the way, to temperature, soiling, shading, inverter clipping, wiring resistance and downtime combined.

The band labels the trade uses are widely repeated conventions rather than a regulatory pass or fail, and IEC 61724-1 defines how the ratio is calculated rather than what value counts as good. As rules of thumb, a performance ratio of roughly 0.75 to 0.85 is normal for a commercial array, above 0.85 reads as high-performing, and below about 0.70 points to significant losses and warrants investigation. The test that actually settles it is your site against its own model, weather-normalised and sustained over time. A single deviation is noise. A persistent gap against the model is the signal.

Monitoring tells you a string is down. It rarely tells you which module, junction box or connector is the cause, and on a commercial roof aerial infrared thermography is the most efficient way to close that gap. An inspection-grade thermal camera flown over the array reads the surface temperature of every module. A healthy cell runs cool because it is exporting its energy as current. A cell that has gone open-circuit, a bypass diode that has failed, a hot solder joint or a corroded connector all show as a temperature anomaly, because the energy that should have left as electricity is dissipating as heat instead.

The recognised thermal-imaging procedure for grid-connected PV is set out in IEC TS 62446-3, which defines how the survey is flown, the irradiance and wind conditions it needs to be valid, and the failure signatures the analyst is reading. A single hot cell, a hot module, a hot string and a hot bypass-diode segment each have a distinct thermal fingerprint, so the report says what is wrong and where rather than only that something is. We fly this in house under our own A2 Certificate of Competency and General VLOS authorisation, so the thermal pass is part of the audit rather than a separate sub-contract, and the camera is pointed at the row the monitoring data already flagged. The two techniques answer a question neither answers alone: the data says how much output is missing, the scan says which physical component is taking it.

Soiling and a real fault can both show as a slow slide in output, and the costly mistake is to clean a roof that has an electrical problem, or to chase a fault that a wash would have cured. The two separate on their signature. Soiling is broadly uniform across an array, gradual and recoverable: output drifts down over weeks, then steps back up after meaningful rain, because rain clears most loose dust even though it leaves the sticky contaminants behind. A fault is the opposite shape. It is localised to a string or a module rather than spread across the field, it does not recover after rain, and on a thermal scan it shows a hot signature a clean panel never does.

That distinction decides the remedy and the spend. Where the loss is uniform, recoverable and rain-correlated, the answer is a survey-led cleaning schedule set from the site's own soiling rate rather than an electrical investigation, and the rate is genuinely site-specific: agricultural dust, coastal salt, roadside grime and low-pitch roofs that drain poorly all soil faster. Where the loss is localised, persistent and thermally hot, no amount of cleaning will fix it and the work moves to fault diagnosis. The audit makes that call deliberately, because washing a roof that needs a connector replaced wastes money and leaves the real loss running.

Some apparent underperformance is not a fault at all, and reading it correctly matters before anyone pays for a site visit. Degradation is expected rather than failure. Analysis by the US National Renewable Energy Laboratory puts the median degradation of crystalline-silicon modules at around 0.5% a year, with premium n-type cells typically lower, so roughly 85 to 90% of the original rated output is commonly retained at 25 years, and linear performance warranties are written around that curve. It only counts as underperformance when the measured loss runs ahead of the warranted, modelled curve.

Export limitation can clip output by design. Where an EREC G100 export-limitation scheme throttles the inverter to hold the site inside its agreed export cap, the surplus solar is simply not produced. That is a legitimate gap against the unconstrained model rather than a fault, and the standard mitigation is battery storage to capture the throttled energy instead of wasting it. The trap is that a capped output curve has two causes that look alike on a chart: a genuine derating fault, where the inverter throttles itself on heat or a grid-voltage excursion, and designed clipping working exactly as intended. The inverter's own event log usually separates them, recording the derating reason and the grid parameters the connection is engineered to under EREC G99. Getting that right is the difference between replacing a healthy inverter and correctly identifying that the array was throttled to its export agreement.

Monitoring granularity decides what can be caught at all. Inverter or plant-level monitoring aggregates everything, so a single faulty string can sit in the total for weeks. String-level monitoring measures DC current and voltage on each string, catching the highest-impact fault modes first, often before they move the plant performance ratio. The standard detection technique is peer-group comparison: each string is checked against its identical siblings, and the outlier is flagged.

The deliverable is a plain written account of what your system should produce, what it does produce, and what stands between the two, with each loss attributed, sized where the data allows, and a recommended fix in priority order. If the system is healthy and the shortfall is just expected degradation and weather, we say so, and you have spent a little to confirm the asset is sound. If a single connector or one failed inverter channel is the problem, we say that rather than reaching for a wholesale replacement.

That honesty holds because the diagnostics arm has no incentive to condemn working kit. We do not quote a fabricated uptime or recovery percentage to flatter the report or alarm you into action; we report what the data supports, and where economic figures come into it they are confirmed against a model of your own site, never a published rate. From there you can act with us, with your existing maintainer, or by taking the report elsewhere. A common next step, where the audit shows the system needs ongoing attention, is to move it onto operations and maintenance so the next bit of drift is caught early rather than at the next audit.