Operation & Maintenance Best Practices Guidelines (Version 6.0)

O&M service providers face growing pressure to increase efficiency while reducing human intervention. Key trends include the use of data-driven techniques, Industry 4.0 solutions, and robotics, such as drones, to streamline operations. Innovative practices leverage data from both field experience and monitoring, feeding into digital platforms or digital twins to support decision-making. The rise of AI and language-based models is automating analysis and reporting, saving valuable time. Additionally, with the increasing frequency of severe weather events, tailored planning and mitigation measures are essential. This chapter highlights important technologies being developed, some close to mainstream adoption, others still in early stages.

O&M service providers are under increasing pressure to do more with less. Increasing human resource efficiency through the use of data-driven and Industry 4.0 techniques are key themes for O&M as the industry works to reduce the number of human interventions and embraces digitisation. New trends include more and more the use of robotic solutions with a new wave of in field robotics to integrate the work carried out by the use of drones. Innovative O&M practices will include data-driven measures coming from both field experience and monitoring. All of the information collected along the whole value chain, must be streamlined into digital platforms (or digital twins) in order to avoid information loss that can act as decision support system for the best actions to follow in case of deviations. Processes are aided by the incredible acceleration we have seen in the use of AI introducing language-based models that will free valuable time in terms of automated analysis, actions, reporting, etc. Finally, severe weather events are increasing in frequency and bespoke planning, and dedicated mitigation measures must be put in place.

The following chapter lists important technology areas being developed by several innovative industry service providers. Many of these new technologies are becoming close to mainstream adoption, others are in early-stage development.

Drone in a box

Purpose and description

A “drone in a box” system for solar PV involves an autonomous drone that resides in a secure, weather-resistant docking station when not in use. These systems are designed for automated deployment, operation, and data collection, requiring minimal human intervention. Equipped with advanced sensors, these drones can perform various tasks like thermal imaging for fault detection, visual inspections for cleanliness or damage, and even monitoring vegetation growth near panels. The “box” serves as a charging station and a safe storage unit, enabling frequent, programmed flights to optimise operations and maintenance efficiency on large solar farms.

State of play

The “drone-in-a-box” system for solar PV operations is gaining traction as a cutting-edge solution for automated inspections and monitoring. While regulatory challenges, such as BVLOS operation restrictions persist, ongoing legislative progress and advancements in solar-powered docking stations are enhancing their feasibility. Despite high initial costs, their potential for ROI through operational efficiency makes them a promising technology for the solar industry.

In field robotic solutions

Purpose and description

Field inspection is typically carried out manually and is time consuming, preventing the deployment of novel characterisation techniques in a cost-effective way. Using aerial inspection increases throughput, however in some cases this is impeded due to restrictions and current legislation which requires the presence of a pilot. In addition to this, aerial inspections cannot visualise faults such as loose/faulty bypass diodes or connectors, or back sheet cracking/chalking, as they are located on the underside of the modules. In field robotic solutions can have a multipurpose: continuous field inspection, check as built vs. drawings, etc.

State of play

In field robotic solutions are coming to the market, starting from the inclusion of sensing solutions in cleaning robots/mowing robots. Rovers are being developed to transfer the cleaning robots from one array to another and while not used as a carrier, they can provide services such as continuous field inspection.

Photoluminescence

Purpose and description

Daylight photoluminescence (DPL) is a relatively novel imaging technique utilised in photovoltaic (PV) system inspection, using the sun as excitation source. Filtering the luminescence signal from the strong sun irradiation is its main challenge. Images acquired at two different operating points (OPs) of the module, allow the subtraction of the background radiation while maintaining the luminescence signal. To avoid strong variation in the sunlight conditions, images of these two operating points (OPs) must be taken within a short time interval, and to decrease noise, several images of each OP must be taken and averaged. Contrary to EL, PL does not require cables disconnection.

State of play

Different methods have been developed to switch between the two different OPs of a PV module. Recently, researchers have focused on inverters’ features that allow for an automatic switching. For example, in the project TRUST-PV, a DPL-ready inverter has been developed with the capability of toggling between manually selectable OPs of connected PV modules. The synchronisation of image acquisition and OP switching becomes particularly challenging if the camera is applied to unmanned aerial vehicles. To overcome this challenge, algorithms are developed to identify OP switches in a set of images taken in the field by investigating image intensities. The integration of PL in field robotics is currently ongoing, for example in the Horizon Europe project SUPERNOVA.

UV Fluorescence imaging

Purpose and description

UV-Fluorescence imaging is a non-destructive imaging technique for failure analysis of solar PVmodules. The development of the technique started around 2010 with first publications in 2012 (Köngtes et al., 2012; Schlothauer et al., 2012; Eder et al., 2017; Muehleisen et al., 2018).

UV-Fluorescence measurements must be performed in a dark environment (typically at night) by illuminating the solar PV-modules with UV-light (<400nm). Most encapsulants show fluorescence in the visible region and thus the material’s response can be captured with a photographic camera. Modules do not need to be disconnected or powered during this procedure.

The observed fluorescence of the encapsulation above the cells with respect to (i) spatial distribution, (ii) intensity and (iii) spectral shift of the fluorescent light is dependent on operation time in the field, climatic conditions, and the type of encapsulant and backsheet used. Furthermore, the fluorescence signal depends on the type of defect (micro cracks in c-Si cells, hotspots, or glass breakage).

Imaging of solar PV modules typically takes less than 60 seconds. An example of UV-fluorescence is given in Figure 15. The advantages of the technique are that no modifications are necessary to the solar PV systems and, when used in combination with EL, an evaluation of timelines for various instances of damage becomes possible as the fluorescence signal is a function of time. New cracks for instance are only visible in EL because there was no time to “bleach” the fluorescence signal.

State of play

There are several things to consider when performing drone-based UV Fluorescence (UVF) Imaging inspections. The cost of drones and trained pilots can be a prohibitive factor in using UVF technology. Similarly, conditions must be stable enough to take images in the dark with a 0.1 second exposure time and the drone needs to be powerful enough to support the extra weight of a camera and a UV lamp.

A minimum of two trained people are required for a UVF inspection, one being the pilot and the other being the photographer. The extra weight of the camera and the UV lamp on the drone means that batteries drain quicker and poses limits on inspections. These constraints are increased further by the UV lamp drawing power from the battery as well. This means that a 4.5 Ah battery can provide a flight time of 8-10 minutes. Moreover, the drone’s flight path must be relatively low to be able to capture quality images.

Estimates predict that it is possible to inspect 720 modules per hour (including time for six battery changes) if conditions are perfect. However, there are several other factors that can affect inspection time, such as project design and weather conditions. To be most effective UVF inspections must be done in the dark and in calm conditions, both of which are far from guaranteed. Working in the dark risks damage to the drone from increased operating difficulty, secondly finding staff willing to work at night comes with added costs to the project (paying overtime or taking on more staff). Moreover, new modules with UVA transparent EVA technology reduce the effectiveness of drone based UVF inspection. Despite these drawbacks, using drones to perform UVF inspections can save time, particularly when inspecting rooftop installations as staff do not need to get up onto roofs.

As this technology is still emerging, many O&M service providers lack the in-house expertise to interpret the findings of UVF inspections. This adds an extra layer of cost to the process and has prevented the technology being mainstreamed for solar PV power plant inspection.

Predictive Maintenance for optimised hardware replacement

Purpose and description

Preventive Maintenance occurs periodically according to contractually agreed schedules and based on expert knowledge. In addition, Preventive Maintenance may be scheduled when the operator identifies an unexpected deviation in performance through the monitoring system. Different maintenance optimisation models are employed to find the optimal cost to benefit balance between maintenance interventions. These models count on the probability of failure of each component of the solar PV system and the impact of that failure on the entire system. For example, the actual lifetime of solar PV inverters under different operating conditions is still uncertain. In practice, inverters will not fail in a predictable way, after a certain period of time, as usually modelled in business plans. Moreover, failure-based maintenance i.e., replacing inverters as they fail may not be the most efficient solution.

A good predictive monitoring system could help with assessing the optimal hardware replacement cycle by modelling the uncertainty in the time-to-failure with a known probability distribution function. Maintenance optimisation models use the output of root cause analyses and remaining useful lifetime analyses to predict future asset failures. This can be used to optimise planning of maintenance and related resource allocation.

Big data analytics can bring added value at any stage of O&M objectives: analysis from observation of collected information, fault detection, fault diagnosis, and optimisation through recommendations issued from the advanced monitoring system. Today different approaches are proposed. Whereas classic Artificial Intelligence (AI) proposes an advanced diagnostic through knowledge-based models, unsupervised and supervised learning methods offer different approaches (e.g. neural networks) using statistics.

The advantages of these Predictive Maintenance optimisation models are that they lower the cost of maintenance by scheduling it more effectively. The diagnostic element of the models also helps to reduce plant downtime. However, the methods are sensitive to device models and brands, making them hard to generalise.

State of play

Today, no model has been proven to be completely reliable. Big-data analysis allows easy recognition of a fault and, in some cases, provides a clear diagnosis and recommendations on the short-term actions to take to avoid probable upcoming issues. The trend is to model the behaviour of the entire system and to plan optimal maintenance and hardware replacement programmes in the medium to long-term. This will of course reduce the overall risk of a solar PV project and, increase investment attractiveness.

Augmented Reality | Smart Glasses

Purpose and Description

Virtual or augmented reality refers to digital elements of interactions using cameras on e.g. smartphones, tablets, or special devices such as smart glasses. Specifically, virtual reality is a computer-generated simulation of a three-dimensional environment that can be interacted with by a person using special electronic equipment. Augmented reality refers to an enhanced version of the real world achieved through using digital elements. For the sake of simplification, the term augmented reality is used in the following referring to the use of smart glasses in O&M.

O&M service providers and their operations teams face the recurring challenge of working with a considerable variety of hardware and software from different manufacturers at various sites (at sometimes remote locations). This heterogeneity requires broad knowledge, skill transfer, and good cross-departmental communication. New technologies based on augmented reality can support O&M service providers with these challenges by easing the collaboration between offices and field engineers.

Corresponding software applications combined with smart glasses enable users to interact visually and acoustically to support works on site. The field engineer using the smart glasses is connected to a supervising (desktop) user who will be able to guide them through working steps, using the desktop version of the respective software. The smart glasses user is connected to the supervisory user via an integrated headset. Visually, conditions on site are recorded by an integrated camera.

The recordings are then displayed live for the supervisory user who can add explanatory diagrams, screenshots, comments, etc. These additions are then displayed on the lens of the smart glasses. This ensures secure working in line with common H&S requirements (hands free) while the field engineer is guided through working procedures. Furthermore, holograms can be used to enable access to animated maintenance instructions.

State of Play

Smart glasses and corresponding software solutions are becoming more popular in the O&M segment. Decreasing price levels for O&M services require improved service/cost efficiency. Augmented reality can support O&M service providers ‘operations by easing skills and information transfer and ad hoc solutions which can positively affect service efficiency.

There are many advantages to this technology, including: increased efficiency in O&M service provision; more fluid knowledge transfer between senior and junior colleagues; and effective upskilling of O&M personnel, resulting in fewer resourcing challenges and generating savings on internal costs for O&M service providers.

However, there are still limitations on the technology’s usefulness. A stable internet connection is required to maintain contact between the field engineer and the supervisor. This can be

problematic for solar PV power plants in more remote locations. At present the technology is also expensive. However, as it becomes more mainstreamed, cost competitiveness should improve.

Internet of Things (IoT) and auto-configuration

Purpose and description

Internet of Things (IoT) in solar PV systems represents an interoperability environment where all devices in the field are connected to each other and show themselves as available to be connected to the system. This can improve integrated, secure communication and efficiency. Each connecting device should provide the following information:

•         Device parameters (brand, type, Serial Number, internal datasheet specifications)

•         Device status and conditions (operational status, temperature, etc.)

•         Connection with other devices & mapping (strings connected, inverter, sensor position, etc.)

•         Any other relevant information

Standardisation efforts (e.g. SunSpec Alliance’s Orange Button initiative) are taking place throughout the solar PV market and will help to improve on configuration costs for solar monitoring. However, the solar monitoring industry will also benefit heavily from the emerging Internet-of- things technologies that further improves plug-and-play behaviour of device communication, improves the quality and the security of the communication, and reduces the cost of hardware.

State of play

There are several advantages to this technology. Principally, it can reduce the costs of monitoring hardware and infrastructure. Similarly, it eases the configuration and maintenance of monitoring systems, whilst improving the quality and stability of data. It also provides for improved secure communications.

However, there is a risk that existing hardware and monitoring equipment will not be compatible with the new technology, resulting in expensive hybrid solutions until it becomes more mainstreamed.

Many Internet-of-Things (IoT) technologies have passed the prototype phase and are available for massive deployment. However, many different technological solutions and approaches are still available in the market and no final best practice approach has emerged.

Again, this leads to a standardisation issue for the industry-wide adoption of Internet-of-Things technology within the solar industry and, as such benefits from its advantages will be reduced when considering solar PV on a larger scale.

Solar PV Monitoring-Imagery Data fusion

Purpose and description

Current solar PV monitoring solutions track key parameters of solar PV assets (e.g. energy production, irradiance, performance ratios, etc.), with high temporal resolution (e.g. up to 1-10 minutes) and trigger alarms when deviations form expected performance occur. However, there are no specific optimisation objectives linked to the detection of underperformance. This method, which relies solely on solar PV monitoring data, presents two significant intrinsic limitations:

•         Expert-dependence: As such, a misconfiguration of (manually defined) expected performance data often leads to misdetection (or misinterpretation) of deviations from the monitored performance data (i.e. false negatives/positives)

•         Insufficient spatial granularity: solar PV monitoring data alone is insufficient for identifying the root-causes and locations of energy losses within solar PV systems, as their best spatial resolution is typically down to string level (i.e. 10-30 solar PV modules combined). As a result, several underperformance issues – especially at solar PV array, module, and submodule level – may remain undetected or unidentified

Currently, root cause analysis at higher granularity is carried out through various aerial imaging inspection techniques, some of which are described earlier in the chapter). Although these methods have impressive time-efficiency and spatial resolution of aerial imagery data analytics (inspection rates of several MW/hour; detection down to submodule/cell level), there are also considerable drawbacks:

•         Practically inexistent temporal granularity: Aerial imagery inspections/scans of solar PV power plants are carried out per-schedule (e.g. bi-annually), rather than as part of preventive maintenance. This means they can, at best, only offer a qualitative “instant picture” of the condition of a solar PV power plant and its components

•         Decoupled from solar PV monitoring: There is no real-time communication or correlation with crucial solar PV monitoring data (inverter outputs, PR, weather data, etc.), preventing precise determination of the causes of underperformance and power losses with image data (fault) signatures

From this perspective, enabling fusion (and interoperability) between heterogeneous solar PV monitoring and imagery data/sensors, will be a key functionality and differentiator for next generation “integrated” solar PV monitoring solutions. Indeed, this concept offers key advantages:

i) solar PV performance monitoring data becomes more actionable, leveraging the diagnostic capacity and accuracy of image data with high spatial granularity; ii) the solar PV imagery data gain a temporal and quantitative dimension, being coupled and correlated with real-time monitoring data and power gain/loss analytics.

Other innovation pathways towards solar PV monitoring-image data fusion solutions can include their interfacing with solar PV digital twins, for example, or the integration of BIM and GIS data, and the

replacement of (aerial) IR image data by hyperspectral or multispectral image data of solar PV power plants.

State of play

Several commercial solutions of advanced solar PV monitoring exist, offering software-driven quantification and classification of string/inverter-level failures, data analytics for soiling rates and performance degradation, and weather and energy flow analytics. On the other hand, turnkey

commercial aerial-IR imagery services offer AI-based data analytics, fault diagnostics and reporting, as well as recommendations for corrective maintenance actions. Yet, in practice, solar PV monitoring

platforms are decoupled from IR imagery diagnostics and not optimally aligned in today’s solar PV O&M.

Concepts towards aggregation and fusion of solar PV monitoring and inspection/imagery data are under development and being patented, in ongoing international R&D projects. The aim is to gain validation by 2024. Over the last 5 years, there have been efforts and patented methodologies that couple solar PV monitoring and/or IR imaging data with physics-based solar PV yield simulations and loss analytics.

Use of Generative AI in workflow’s automatisation

Purpose and description

In the coming years, large language-based AI models (LLM) will reshape engineering software for renewable energy asset operations, driven by the urgency to enhance human-computer interaction. These models, adept at comprehending instructions and configuring settings seamlessly, promise significant time savings by expediting data queries and contextual data visualisation. As the PV sector grows in complexity, these models are set to become indispensable tools, streamlining operations, and ensuring efficient asset management. Through language models the user will be able in the future to automatically generate or bypass dashboards. The use of natural language will reduce barriers created by the need of having programming skills.

State of play

The use of generative AI in PV operation and maintenance workflows is advancing, supported by EU initiatives such as Horizon Europe’s SUPERNOVA project. AI-driven solutions, including language-based models and large language models (LLMs), are automating tasks like reporting and data analysis while providing actionable insights tailored to operators. These tools reduce

manual workload, enhance decision-making, and shift the focus from traditional analytics to more advanced “insight as a service.” By integrating established industry KPIs, such as the Cost Priority Number (CPN), these innovations improve efficiency and profitability in PV asset management.