Automation of field inspection in large scale solar farms - Alspin

Solar energy is the fastest growing segment in the energy industry with a global installed capacity over 1000 GW. For Equinor as a market driven power producer, solar energy is an important part of its expanding renewable portfolio in selected markets as Brazil, Argentina, Poland, and Northern Europe. Plant monitoring plays a key role both ensuring safe operations and optimizing energy production. Monitoring the electrical performance of a PV plant typically allows for fault and power loss detection only at the inverter and string levels where several tens or hundreds of modules are connected. More granular faults, at the PV-module level, require additional sensing techniques for reliable detection and diagnostics. Autonomous drones equipped with infrared vision represent an attractive and rapidly evolving technology for cost-effective monitoring of large- scale solar power plants. By using automatically generated flight plans, drones can guide themselves to all parts of the plant in search of modules with thermal hotspots or other signs of a faulty solar cell. Advanced computer vision techniques and machine learning algorithms can be used to identify and classify faults. These can then be reported back to the plant operations and maintenance team for additional diagnostics or replacements. The Alspin project has addressed these topics and developed several tools and methods including (i) automated airborne PV plant monitoring with high-resolution visual and infrared sensing techniques, (ii) a flight planner for autonomous drones operating in PV plants, (iii) advanced computer vision with unsupervised machine learning methods for microscale PV module fault detection and classification, and (iv) a solution for visualizing results and assisting maintenance operations. The work has led to a prototype consisting of an automated inspection system and autonomous drone guidance that has been demonstrated in a proof-of-concept pilot at a solar plant in Brazil.

The key results of the project include: - A solution combining computer vision methods with unsupervised machine learning to automatically detect faults in PV modules using dual infrared and color images. - A solution for automatic drone path planning including a virtual testing environment. - A solution for glocalization and visualization of images and results, providing an overview of plant condition and supporting maintenance planning. - Study of additional luminescence imaging techniques relevant for field-based PV inspection - A large database of PV modules images

The goal is to realize efficient and automated system for solar module inspection and optimize energy production (i.e. high yield and warranty claims) in large scale solar installations. Solar is the fastest growing segment in the energy industry, because of the simple installation and short time to production. Solar farms now contain more than 1 million individual modules, requiring larger space and continuous maintenance for keeping high productivity. Having recently changed its name, Equinor (formerly Statoil) has a mission to invest in renewable energy and to develop its business within green energy. Through joint ventures with Scatech Solar, Equinor intends to use the global presence to expand the portfolio to become a diversified energy company where solar will be part of Equinor new businesses to put the company at the front of green energy industries. An efficient, automated and autonomous inspection and analysis system will be an important tool for both monitoring their investment and optimizing the yield. Manual inspection of solar systems in the field is common today. Interpretation of the IR thermal images is complicated by artefacts that appear in data and dirt on modules, which may not be easily distinguishable from faults. Also, correlation between visual and electrical data is often not practical. With the large installations of today, manual inspection is no longer efficient. Therefore, sophisticated approaches are necessary to determine when defects have a sufficient impact on the energy production to warrant replacement of solar modules. This project aims to realize fault models from the field arrays, high-resolution and multispectral sensing for automated airborne monitoring and inspection from remote distances. Machine learning will be established to correlate inspection with electrical data to evaluate the impact of the defects on the production. A cost-benefit analysis will be integrated to control the operation and maintenance of solar farms.