By: Dauda Lawal – Web Developer

Solar energy is becoming increasingly important to Nigeria’s evolving energy landscape. However, operating a large portfolio of solar photovoltaic (PV) plants introduces challenges that extend beyond generating electricity. Operators must maintain inverter reliability, detect underperforming panels, monitor environmental conditions, optimize maintenance schedules, and meet contractual power generation commitments.

When faults remain undetected for weeks, even a relatively small performance loss across thousands of solar strings can translate into substantial lost energy and revenue. Traditional inspection routines and monthly performance reports often reveal these problems too late, forcing maintenance teams to investigate faults after generation has already declined.

To address these challenges, Arthurite Integrated has proposed an AWS-powered Solar Fleet Performance Monitoring and AI-Driven Fault Diagnostics Platform for UPNEPA. The initiative is designed to replace retrospective reporting and manual site inspections with continuous monitoring, machine-learning-based performance analysis, automated fault diagnosis, and data-driven maintenance workflows.

The proposed solution covers 14 utility and commercial-scale solar sites with a combined capacity of approximately 62 megawatt-peak (MWp), 480 inverters, and 9,600 monitored solar strings.

By combining AWS IoT Greengrass, AWS IoT Core, Amazon Timestream, Amazon S3, and Amazon SageMaker, the platform aims to help UPNEPA identify underperforming equipment within hours, determine probable causes before dispatching technicians, and improve the consistency of fleet performance reporting.

This article explores the operational challenges behind the proposed initiative, the AWS architecture designed to address them, and the business outcomes the project aims to deliver.

This article describes a proposed engagement based on Arthurite Integrated’s Statement of Work. The performance improvements and financial benefits discussed are targets, not verified outcomes from a completed deployment.

The Challenge: Why Traditional Solar Fleet Monitoring Falls Short

1. Delayed Detection of Solar Underperformance

Solar PV plants depend on consistent performance across panels, strings, inverters, and supporting infrastructure. Dust accumulation, damaged modules, inverter derating, electrical faults, shading, and communication failures can all affect output.

UPNEPA’s existing monitoring process relies heavily on vendor portals, monthly performance reports, quarterly site inspections, and periodic drone surveys. These processes provide useful information, but they can leave significant gaps between the onset of a fault and its detection.

The Statement of Work estimates that underperformance currently takes approximately 21 days to detect. During that period, an affected string or inverter may continue operating below its expected output, reducing energy generation and potentially putting power purchase agreement (PPA) performance commitments at risk.

The proposed platform aims to shorten this detection window by comparing expected and actual solar production continuously at string level.

2. Limited Fault Diagnosis Before Technician Dispatch

Detecting a reduction in total plant output does not automatically reveal its cause. A decline in production may result from dirty panels, a failing inverter, a damaged string, vegetation-related shading, or a communication problem.

Without sufficient diagnostic information, maintenance teams may dispatch the wrong personnel, carry unsuitable spare parts, or require multiple visits before resolving the issue.

The baseline documented in the proposal estimates an average of nine days from detection to repair. It also indicates that technicians are typically dispatched without a confirmed probable cause.

This reactive approach can increase travel and inspection costs, prolong equipment downtime, and divert maintenance resources from higher-priority issues.

3. Manual Inspections and Reporting Workloads

UPNEPA’s solar fleet generates large volumes of operational data. The proposed scope includes approximately 13.8 million string-level DC current and voltage readings per day, 20.7 million inverter readings and events, and approximately 200,000 weather-station readings daily.

Despite this data availability, much of the information is distributed across different vendor portals and operational systems. Monthly performance reports take approximately 10 working days to prepare, while drone thermal images are reviewed manually and may take around three weeks to analyze.

The project therefore targets three related problems: late fault detection, limited visibility into root causes, and time-consuming reporting processes.

A centralized, automated analytics platform could help turn these fragmented data sources into timely operational intelligence.

The AWS Solution: An Intelligent Solar Fleet Performance Platform

Arthurite Integrated’s proposed architecture combines industrial edge computing, cloud-based data ingestion, time-series storage, machine learning, computer vision, and automated maintenance workflows.

At each solar site, AWS IoT Greengrass V2 gateways collect readings from inverters, string monitors, and weather stations. The gateways normalize data from different equipment vendors, apply local health rules, and buffer measurements when connectivity is interrupted.

AWS IoT Core securely transports data into the cloud, while Amazon Kinesis Data Streams and AWS Lambda coordinate ingestion and processing. Amazon Timestream for InfluxDB supports recent operational time-series queries, and Amazon S3 provides long-term storage for historical measurements, drone imagery, and analytical datasets.

AWS IoT SiteWise organizes the fleet into a consistent asset hierarchy, from individual solar strings and inverters to blocks, sites, and the complete fleet.

Amazon SageMaker analyzes expected versus actual output, detects anomalies, and estimates probable fault causes. It also supports computer-vision analysis of drone thermal and visual imagery to identify potential panel defects.

When the platform identifies an actionable issue, Amazon EventBridge and AWS Lambda can create prioritized maintenance work orders with diagnostic evidence attached. Amazon QuickSight provides dashboards for performance engineers, operations teams, asset managers, and commercial stakeholders.

The result is a proposed architecture that connects monitoring, diagnosis, maintenance, and performance reporting in a single workflow.

Five Key Engineering Capabilities Behind the Proposed Solution

1. Real-Time Solar Performance Monitoring with AWS IoT

The first capability is the transition from retrospective reporting to continuous monitoring.

AWS IoT Greengrass V2 is designed to collect data from multiple inverter vendors, string-monitoring systems, and weather stations. By normalizing measurements, timestamps, units, and equipment alarm codes, the platform can compare performance consistently across sites.

This is important because a solar fleet rarely operates under identical environmental conditions. Irradiance, module temperature, wind, shading, and local weather all influence expected generation. A reliable monitoring system must account for these factors before interpreting a production decline as a fault.

The proposed architecture uses AWS IoT Core and Amazon Kinesis Data Streams to ingest operational data, with Amazon Timestream supporting recent high-resolution queries and Amazon S3 retaining historical information.

The design also includes local buffering for up to seven days of data, allowing gateways to preserve measurements during connectivity interruptions and synchronize them when the connection returns.

Key capabilities include:

• Continuous ingestion of inverter, string, and weather data.
• Normalization of measurements across supported equipment vendors.
• Asset-level monitoring through AWS IoT SiteWise.
• Local detection of gateway, sensor, and communication problems.
• Store-and-forward data synchronization after network recovery.

These capabilities are intended to give UPNEPA a more complete and timely view of solar fleet performance without requiring every site to maintain uninterrupted connectivity.

2. AI-Powered Fault Detection with Amazon SageMaker

Monitoring becomes significantly more useful when the platform can distinguish expected variations in solar output from genuine equipment problems.

The proposed solution uses Amazon SageMaker to predict expected output for each monitored string based on irradiance, module temperature, solar position, panel specifications, and historical healthy operating behavior.

The platform then compares predicted performance against actual readings. Persistent deviations, peer-string comparisons, inverter alarm codes, and weather conditions help determine whether a string or inverter requires further investigation.

The proposed analytics capabilities include:

Expected-output modeling: Estimates how much power a string should generate under current environmental and operating conditions.

Underperformance detection: Identifies sustained differences between expected and actual output, assigns severity, and estimates associated energy losses.

Probable-cause classification: Uses the pattern of deviations, inverter alarms, weather, rainfall, and cleaning history to estimate the likely cause of underperformance.

Soiling analysis: Estimates production losses caused by accumulated dirt and helps prioritize cleaning based on expected energy recovery.

For example, if multiple strings within a block show a similar gradual decline without corresponding inverter alarms, the system may classify soiling as the probable cause. If all strings connected to one inverter decline together alongside an inverter fault code, the system may instead recommend an inverter inspection.

These diagnoses are intended to support maintenance planning rather than replace technical verification. Model confidence, data quality, and historical evidence will determine how recommendations are presented to operations teams.

3. Drone-Based Solar Panel Inspection Using Computer Vision

Solar panels can develop defects that are difficult to identify through electrical measurements alone. Thermal anomalies, cracked modules, disconnected strings, and overheating connectors may require targeted physical inspection.

UPNEPA’s proposed solution includes analysis of drone-captured thermal and visual imagery using computer-vision models hosted on Amazon SageMaker.

Drone images are uploaded to Amazon S3 with site, flight, and location metadata. The model analyzes the imagery to identify potential hotspots and visible defects, helping inspection teams prioritize areas requiring further investigation.

The platform is designed to support:

• Thermal hotspot detection.
• Identification of potential cracked or damaged panels.
• Detection of disconnected strings and overheating connectors.
• Location-linked defect records and image evidence.
• Comparison of inspection results across survey dates.

The proposed workflow also connects detected defects to maintenance activities. For example, a critical thermal anomaly could generate a prioritized work order containing the affected location, image evidence, severity, and recommended inspection action.

The Statement of Work estimates approximately 98,000 drone images annually across the 14 sites. Automating the initial image review is intended to reduce the analysis turnaround time from approximately three weeks to 24 hours within 12 months.

Computer-vision detections will still require validation, and physical inspection may be necessary before confirming a defect or authorizing repairs.

4. Automated Maintenance Work Orders and Fault-Based Dispatch

Fault detection delivers its greatest operational value when the right maintenance action follows.

Arthurite’s proposed architecture connects diagnostic events to maintenance workflows through Amazon EventBridge and AWS Lambda. Each actionable event can be enriched with the affected site, inverter or string, probable cause, estimated energy loss, supporting evidence, and recommended maintenance response.

The platform is designed to integrate with UPNEPA’s existing computerized maintenance management system (CMMS) through an available API. If the current system cannot support the required integration, the proposal allows for a lightweight work-order list within the dashboards.

Examples of the intended workflow include:

• Inverter fault: Create a high-priority work order with relevant fault codes, estimated energy loss, and suggested spare parts.
• Persistent string underperformance: Provide the string location, deviation history, peer comparison, and a proposed electrical test plan.
• Excessive soiling: Identify affected blocks and recommend a cleaning window based on estimated energy losses and relevant weather information.
• Critical thermal defect: Attach drone evidence and location details to a priority maintenance request.
• Communication failure: Recommend a remote gateway or network investigation when missing data does not correspond to a genuine decline in plant output.

This distinction is important. A missing sensor reading should not automatically trigger a field visit if the plant’s generation remains normal. By differentiating communication problems from actual production losses, the platform aims to reduce unnecessary dispatches.

The project targets work-order creation in under two minutes and aims for at least 85% of work orders to have a probable cause identified before dispatch within 12 months.

5. Centralized Fleet Analytics, PPA Reporting, and Operational Visibility

A multi-site solar operation needs more than individual equipment alerts. Management also requires a consistent view of generation, availability, contractual performance, maintenance status, and revenue exposure.

AWS IoT SiteWise provides the proposed asset hierarchy, while Amazon QuickSight supports dashboards for different operational and business users.

The planned dashboard suite includes:

Fleet Overview: Displays generation, performance ratio, availability, open alerts, and PPA status across sites.

Site and String Performance: Highlights underperforming strings, inverter efficiency, estimated energy losses, and performance deviations.

Diagnostics and Dispatch: Shows probable fault causes, recommended maintenance crews, work-order priorities, and closure outcomes.

Drone Inspection Results: Maps detected defects, their severity, repair status, and changes between surveys.

PPA Compliance Reporting: Compares actual and expected energy generation against contractual performance requirements.

Amazon Athena and AWS Glue can support historical analysis and reporting from the S3 data lake. Maintenance outcomes can also feed back into the diagnostic process, allowing future models to be evaluated against confirmed findings and completed repairs.

This integrated view is intended to help engineers prioritize the faults with the greatest operational impact while giving asset managers and commercial teams more timely performance information.

Expected Business Impact: What UPNEPA Aims to Achieve

The proposed platform establishes measurable objectives for evaluating the effectiveness of continuous monitoring, automated diagnostics, and maintenance optimization.

Performance indicator	Current baseline	Proposed 12-month target
Time to detect underperformance	Approximately 21 days	Under 4 hours
Work orders with probable cause identified before dispatch	Approximately 0%	At least 85%
Dispatches without a known cause	Approximately 100%	15% or fewer
Average repair time after detection	Approximately 9 days	4.5 days
Fleet performance ratio	Approximately 78.9%	80.7%
Inspection costs	Estimated ₦70M annually	40% reduction
Drone image analysis turnaround	Approximately 3 weeks	24 hours
Data completeness after reconnection	Not measured	At least 99.5%
Monthly performance reporting	Approximately 10 working days	Automated daily reporting
New site onboarding	More than 8 weeks, estimated	5 working days or fewer
Platform availability	Not measured	99.9%

These are proposed performance targets, not achieved results. Baselines are estimates to be validated during discovery, and actual outcomes will depend on data quality, model accuracy, site connectivity, equipment compatibility, and operational adoption.

If achieved, these improvements could help UPNEPA identify faults earlier, reduce unnecessary site visits, improve maintenance planning, and strengthen visibility into contractual solar performance.

The project also aims to improve the fleet performance ratio from approximately 78.9% to 80.7%. Realizing that improvement would require the platform to identify actionable losses and for maintenance teams to resolve the underlying issues; analytics alone cannot guarantee additional generation.

An 18-Week Implementation Roadmap

Arthurite Integrated proposes an 18-week implementation program, beginning with four pilot sites before expanding to the complete 14-site fleet.

Phase 1: Discovery and Design — Weeks 1–3

The team will assess data sources, inverter interfaces, site connectivity, historical production records, maintenance workflows, and PPA reporting requirements. The phase will establish the asset hierarchy, security architecture, and agreed measurement baselines.

Phase 2: Edge and Data Foundation — Weeks 4–8

The project will deploy the initial edge gateways, configure AWS IoT connectivity, establish the data-ingestion pipeline, and build the asset model. Four pilot sites will be used to validate ingestion, data quality, local buffering, and synchronization after connectivity interruptions.

Phase 3: Performance Analytics and Fault Diagnosis — Weeks 9–13

The team will develop expected-output models, underperformance detection, probable-cause classification, and initial automated work-order workflows. Drone-image labeling and model development will progress alongside the analytics work.

Phase 4: Drone Analytics, Dashboards, and Fleet Rollout — Weeks 14–18

The final phase will expand the platform to all 14 sites, activate drone analytics, deliver QuickSight dashboards, test failover and recovery, and conduct user acceptance testing and staff training.

The proposed rollout is designed to validate the technology in stages before relying on automated diagnoses in routine maintenance workflows.

Advancing AWS Solutions for Renewable Energy

The UPNEPA initiative demonstrates how AWS cloud services, industrial IoT, machine learning, and computer vision can support renewable-energy operations.

AWS IoT Greengrass provides local processing at remote solar sites. AWS IoT Core and Amazon Kinesis support secure data ingestion, while Amazon Timestream and Amazon S3 provide complementary capabilities for recent time-series analysis and historical data storage.

Amazon SageMaker brings together performance prediction, anomaly detection, probable-cause classification, and drone-image analysis. AWS event-driven services connect the results to maintenance workflows, while Amazon QuickSight provides fleet-wide operational and commercial visibility.

This use case complements Arthurite Integrated’s broader energy technology portfolio, including proposed solutions for smart meter analytics and load forecasting, gas compressor predictive maintenance, and AI-powered HSE monitoring.

Together, these initiatives illustrate how cloud and AI technologies can be applied across renewable generation, electricity distribution, and industrial energy operations.

Conclusion: Turning Solar Performance Data into Actionable Intelligence

Solar fleet optimization requires more than tracking total electricity generation. Operators need to understand which assets are underperforming, why their output has declined, and what action will restore performance efficiently.

Arthurite Integrated’s proposed AWS-based platform for UPNEPA aims to connect those decisions through continuous string-level monitoring, machine-learning-based fault diagnosis, drone image analytics, automated work orders, and centralized performance dashboards.

The proposed architecture is designed to reduce the time between the onset of a fault and its detection, provide technicians with better diagnostic information before dispatch, and automate reporting across a distributed solar fleet.

Delivering these benefits will depend on reliable data access, representative model training, validated diagnostic performance, suitable site connectivity, and effective collaboration between analytics and maintenance teams.

With those foundations in place, intelligent solar monitoring could help UPNEPA move toward more proactive maintenance, improved asset utilization, and stronger performance management across its renewable-energy portfolio.

Looking to optimize renewable-energy operations with AWS and AI?

Arthurite Integrated develops cloud, industrial IoT, data analytics, and AI-enabled solutions designed around real operational requirements. Contact Arthurite Integrated to discuss how a tailored digital strategy could help your organization improve asset performance, streamline maintenance, and make better use of operational data.