Quick answer
Solar fault detection flags abnormal observations; diagnostics uses data, operating context and qualified investigation to assess causes, without replacing electrical protection.
Answer: Solar fault detection identifies abnormal observations that may indicate equipment or performance problems; diagnostics investigates their cause. Monitoring data, weather-adjusted models, equipment alarms and qualified inspection can support this process. An output shortfall is not proof of a specific fault, and analytics does not replace electrical protection. Check data quality, system configuration and expected operating constraints before attributing losses or selecting corrective work.
Detection is a flag; diagnosis needs evidence
Production monitoring records measured operation. Detection can flag an unusual value or pattern; diagnosis combines that flag with additional evidence to distinguish weather, sensor problems, control limits, outages and physical faults. Automated and manual methods both have a role. Machine learning is optional, not the definition of fault detection.
DOE FEMP’s monitoring-platform guidance describes measurements, model comparisons, alarms, maintenance tickets and updated plant documentation. Platform features vary; a monitoring account does not imply every diagnostic capability is present.
Choose an investigation method for the question
| Method | Useful evidence | Important limitation |
|---|---|---|
| Output/weather comparison | Time-aligned power/energy and environmental inputs | Model or sensor bias can resemble underperformance |
| Equipment alarms and telemetry | Status, voltage/current and control records | Alarm meaning depends on exact equipment and configuration |
| I–V testing | Electrical characteristic measurements under recorded conditions | A pattern need not identify one unique defect |
| Thermal inspection | Location and temperature patterns | Conditions, reflections and operating state affect interpretation |
| Statistical/ML analysis | Repeated patterns relative to a defined baseline | Validate false alarms and performance on unseen operating conditions |
Physical inspection and electrical tests require qualified personnel and suitable procedures. Thermal drone inspection can locate anomalies but does not by itself prove cell cracking, electrical safety or a warranty claim. This glossary provides no live testing or repair instructions.
Check data and operating constraints first
Verify timestamps, units, meter boundary, missing/stale values, sensor calibration and equipment identity. Use an as-built configuration and appropriate actual-period weather when comparing modeled and measured performance. An original typical-year annual forecast alone is not a diagnostic reference for one cloudy day.
Check export curtailment, inverter clipping, commanded shutdowns, battery dispatch and known outages before labeling a shortfall a hardware failure. PVAnalytics’ clipping-level function labels patterns in AC-power data; that algorithmic indication is not an independent electrical diagnosis or proof of why the plateau occurred.
A checked shortfall and value illustration
For positive expected energy at the same boundary and period:
Relative shortfall = (expected − measured) / expected × 100%
Hypothetical: expected 1,000 kWh and measured 900 kWh give 10% shortfall and a difference of 100 kWh. At an assumed marginal value of $0.12/kWh for those units, the potential difference is $12. This calculation does not establish a fault, recoverable energy or an economic case for repair. Time-dependent tariffs and uncertain expected production can change the assessment.
Performance ratio is a different metric, based on energy yield and reference irradiance, not simply measured output divided by an arbitrary forecast. A universal PR threshold or a fixed 5%-for-three-days rule cannot establish every site’s fault status. Set investigation thresholds against measurement/model uncertainty and the operational context.
Turn an anomaly into a controlled maintenance record
Record the affected equipment, onset, data quality, known constraints, suspected causes and evidence needed to distinguish them. Prioritize safety alarms according to approved procedures; do not wait for a revenue threshold to escalate a potentially unsafe condition.
Assign an investigation owner and record diagnosis, authorized corrective action and closure evidence. Verify performance after the action under comparable conditions and retain any changed as-built inputs. Predictive maintenance concerns anticipating service needs; it should not be confused with a guaranteed diagnosis of the current event.
Inspection cadence depends on equipment instructions, contract, conditions and risk. Monitoring does not guarantee universal lost-revenue recovery, lifetime PR or fault detection within a fixed number of hours. Software suppliers should demonstrate actual supported measurements and alerts rather than promise these outcomes from a design-stage simulation alone.
Common questions
Does a 10% production shortfall prove equipment failure?
No. Check weather, data quality, operating constraints and model assumptions.
Is fault detection a substitute for protection?
No. Analytics and maintenance diagnosis do not replace required electrical protection.
Does thermal imaging uniquely identify every defect?
No. Interpret anomalies with operating conditions and corroborating evidence.
Must every system use machine learning?
No. Choose and validate methods appropriate to the data and investigation.
Related Glossary Terms
About the Contributors
CEO & Co-Founder · SurgePV
Keyur Rakholiya is identified by SurgePV as its CEO and a company co-founder. His SurgePV author page lists only role information that can be tied to the public profile below; credentials, project totals, testing claims, media appearances, and speaking engagements are not asserted without retained evidence.
Editorial contributor · SurgePV
Rainer Neumann is credited as an editorial contributor on SurgePV content. This profile does not assert engineering credentials, project totals, software-testing experience, education, speaking engagements, or media citations because independent verification evidence is not retained in the publication record.
