Just Cleaned Array but Production Still Flat: Same Day Regression Decision Tree

Why this matters

A PV array cleaning that does not move production is one of the highest-friction service moments in solar. The customer paid for cleaning and expected a measurable bump on the monitoring portal the next sunny hour. When the curve looks the same after the work as before, the natural reaction is to blame the cleaner. The truth is usually that soiling was not the limiting factor on production that day, or another loss path was already in the system and the cleaning revealed it. Working through the same-day regression decision tree resolves the call without arguing about whether the panels are actually clean.

Symptom presentation

The crew just finished a cleaning visit. The array is visibly cleaner. The customer is watching the monitoring portal and reports that production over the next clear hour looks the same as it did before the wash, or within the same daily envelope as the previous week. Inverter status shows no fault. There are no shading changes from morning to afternoon that would explain the flat curve. The cleaning was done with appropriate technique on the appropriate surface.

The customer frames it as a service failure. The data does not yet support that framing or disprove it. The job is to separate "cleaning did nothing because nothing was dirty enough to matter" from "cleaning did its job but another loss is masking the gain."

Quick checks

Confirm baseline first. Pull the production data for the same hour-of-day for the seven days before the cleaning and the same hour-of-day for the day of and the day after the cleaning. Normalize for irradiance if the site has an irradiance sensor. Without normalization, hour-to-hour comparison can swing 20 percent on cloud cover alone. The right comparison is irradiance-corrected specific yield, not raw kWh.

Walk the array post-clean. Confirm that the wash actually removed soiling. A wash done with hard water can leave a mineral film that looks clean wet and shows a haze when dry. A wash done with insufficient agitation on bird droppings or lichen leaves discrete fouling spots that drag a whole string.

Check inverter and string data. Pull per-string current at the inverter or per-module data if the system has module level power electronics. A single underperforming string or module flags the loss path immediately and the cleaning is not the issue.

Verify clipping. If the system is undersized on inverter capacity relative to DC, clipping caps the midday curve at the inverter rating regardless of how clean the modules are. A flat midday plateau that matches the inverter AC rating is normal clipping behavior, not a cleaning failure.

Decision thresholds

Use four gates.

Gate one is soiling severity. If the pre-clean soiling level was under 3 to 5 percent loss equivalent, the cleaning recovers a small change that is hard to see in noisy hour-to-hour data. Soiling-induced production loss only becomes obvious on the monitoring portal at roughly 5 percent and above. A clean of a lightly soiled array is preventative maintenance, not a recovery operation, and the customer expectation needs to be reset.

Gate two is clipping. If the system is DC-oversized to the AC inverter rating, the midday curve clips at the AC rating regardless of irradiance or soiling. A cleaning at a clipped system shows no midday gain because the inverter cannot pass the additional DC power. The shoulder hours, morning and evening, are where a clipped system shows the gain. Pull morning and evening comparison data, not midday.

Gate three is masking loss. If pre-clean production was already constrained by a per-string or per-module loss, removing soiling does not lift production until the underlying loss is addressed. A string with high resistance connections, a failing optimizer, a partial bypass diode failure, or shading from new growth caps the contribution from that string at the constraint. The cleaning gain is hidden behind the constraint.

Gate four is wash residue. If the wash left mineral film, surfactant residue, or streaks from a hard water rinse, the surface optical loss is not actually improved. Re-inspect the panels dry, in low-angle light, and look for residue.

Confirming diagnosis

Pull irradiance-corrected specific yield for the seven days before and after the visit. If the post-clean specific yield is within roughly 2 percent of pre-clean, the system was not soiling-limited and the cleaning produced the expected small change. Show the customer the normalized data, not the raw kWh.

If a per-string or per-module loss is suspected, pull current data per string at noon on a clear day. A string carrying materially less current than its peers, with the same orientation and tilt, points to a string-level fault that was already there.

If wash residue is suspected, retest by rinsing one module with distilled or RO water and comparing optical clarity after dry against an adjacent module rinsed with the original wash water.

Remediation

If soiling was not the limiting factor, communicate the diagnostic finding with normalized data. The cleaning extended panel life and prevented future loss, even if the immediate production gain was small. Schedule the next clean at a longer interval and document the soiling rate so the next visit is timed to where cleaning actually moves the curve.

If clipping is the cause, show the customer the shoulder-hour comparison. Production gain from cleaning at a clipped system lives in the morning and evening shoulders. The midday plateau is not the right window to evaluate the work.

If a string or module loss is masking the gain, schedule a separate diagnostic visit to address the underlying fault. Common causes include corroded MC4 connectors, a failing optimizer, a partial bypass diode failure, an array harness with rodent damage, or new shading from vegetation growth.

If wash residue is the cause, re-rinse with low-mineral water and inspect dry. A hard water source on the customer site is a recurring problem and either a deionized rinse step or a different water source is needed for future visits.

References

• NREL Technical Report on PV Soiling and Cleaning Effectiveness.
• IEC 61724: Photovoltaic System Performance Monitoring (data normalization and specific yield).
• IEEE 1547: Interconnection and Interoperability of Distributed Energy Resources (system-level performance reference).
• IEC 62446-1: Grid Connected PV Systems - Minimum Requirements for System Documentation, Commissioning Tests and Inspection.