Comprehensive Review and Analysis of Soiling Detection Technologies in Solar Panels

The methodology in this study follows a systematic process to collect and analyze recent research. A Scopus search was conducted to retrieve relevant studies, followed by bibliometric analysis using VOSviewer to identify key research trends, influential studies, and emerging topics. Finally, content analysis categorized the 75 selected papers into Inspection Tools, Models, and Cleaning Systems, allowing for a structured synthesis of methodologies and findings. This approach ensures a comprehensive, structured and up-to-date assessment of advancements in soiling detection research. It contains four main steps, as defined below:

2.1 Literature retrieval

This step involves the initial stage of selecting appropriate search terms and keywords to gather relevant research papers and publications. A systematic literature review was conducted using Scopus, employing various keywords such as "soiling detection solar," "soiling concentration solar," "dust detection solar," "dust concentration solar," and others. Scopus was used as our primary database because it is one of the most comprehensive and widely recognized academic databases, which covers a wide range of high-quality journals and conference proceedings in engineering, energy, and environmental sciences. Its extensive collection of peer-reviewed publications ensures that our review captures all significant research on soiling detection. Therefore, this process resulted in a total of 683 papers, covering the years 2000 to 2023.

2.2 Literature screening

The PRISMA statement [5] is utilized in the following step to provide a structured framework for identifying and selecting relevant literature. Initially, a total of 683 papers were identified. After removing duplicates, the count was reduced to 574 papers. Further screening for direct relevance to the topic of soiling detection on solar panels narrowed the selection to 75 papers, with 498 papers being excluded. Each paper was manually reviewed to ensure its relevance, with the inclusion criteria focusing on studies that specifically addressed soiling detection techniques, technological advancements, and their impact on PV performance. Papers that lacked a direct focus on soiling detection, addressed unrelated aspects of solar energy, or provided insufficient technical details were excluded. This thorough screening process enhances transparency and ensures that the selected papers contribute meaningfully to the review. Figure 1 illustrates this selection process. The remaining 75 papers were then classified into three sections: Inspection Tools, Models, and Cleaning Systems. Figure 2 presents the chronological distribution of the selected papers, highlighting the increasing interest and research activity in soiling detection over recent years. This trend underscores the growing importance of this area in the context of enhancing solar panel efficiency and performance.

1.png

Figure 1. Literature screening approach

2.png

Figure 2. The chronological span of years

2.3 Bibliometric analysis

Bibliometric analysis was conducted to systematically evaluate the structure of research in soil detection by analyzing citation patterns, keyword occurrences, and co-authorship networks. Using VOSviewer software, visual representations were generated to highlight key research themes, influential authors, and emerging topics. This approach enabled a deeper understanding of research trends and relationships among various studies by:

• Identifying influential works through citation counts and co-citation networks, determining the most referenced studies in the field.
• Mapping research trends by examining keyword co-occurrence, which helps visualize commonly studied topics and gaps in the literature.
• Analyzing collaboration patterns through co-authorship networks, providing insights into global research contributions and partnerships in soil detection.

The advantage of bibliometric analysis is that it offers quantitative insights into the research landscape, allowing for an objective identification of dominant themes, key contributors, and emerging directions. Additionally, visualized bibliometric maps provide a structured way to observe relationships between studies, guiding future investigations and fostering collaboration. This method also helps highlight emerging topics and research gaps, ensuring that ongoing studies align with industry needs and technological advancements.

2.4 Content analysis

In addition to bibliometric analysis, content analysis was conducted to gain a qualitative understanding of the reviewed studies and systematically categorize key findings. This approach enabled a structured examination of the 75 selected papers, organizing them into three main themes:

• Inspection Tools: Technologies and methods used to detect soiling accumulation on solar panels.
• Models: Statistical, empirical, and AI-based approaches for analyzing and predicting soiling impact.
• Cleaning and Mitigation Systems: Strategies designed to minimize soiling effects and optimize solar panel efficiency.

The research papers in this review were systematically categorized into three key themes: Inspection Tools, Models, and Cleaning Systems to provide a structured and comprehensive analysis of soiling detection and mitigation. These themes were chosen based on a thorough examination of existing literature, ensuring that they capture the core aspects of soiling detection research. This classification ensures a holistic approach, covering detection, analysis, and mitigation, which are the fundamental pillars of soiling research.

By classifying research into these themes, content analysis provided a clear framework for understanding various approaches and their role in soil detection and mitigation. Each study was carefully examined to extract key methodologies, findings, and limitations, enabling researchers to:

• Identify research gaps, such as the need for more cost-effective, real-time soiling detection methods.
• Compare different methodologies, evaluating their advantages, limitations, and applicability under various environmental conditions.
• Highlight technological trends, particularly the increasing role of AI, remote sensing, and automated cleaning in soiling detection.

The advantage of the content analysis is that it offers a structured qualitative synthesis, allowing for a comprehensive comparison of existing detection tools, models, and cleaning systems. This method facilitates the identification of patterns and common themes, helping researchers understand the evolution of soiling detection technologies and guiding future advancements in solar panel efficiency and performance.

Significant attention has been garnered by the detection of soiling on solar panels due to its critical impact on the efficiency and performance of PV systems. Various inspection tools and models have been developed and studied, offering promising methods to enhance the effectiveness of solar energy generation. In this section, a comprehensive content analysis of the field is provided, examining the latest advancements, tools, and models used for soil detection. Key findings are highlighted, research gaps are identified, and insights are offered to drive future developments in improving solar panel efficiency and sustainability.

4.1 Inspection tools

In this section, the various inspection tools utilized to detect different forms of soiling on solar panels are explored. The effectiveness of solar panels can be significantly compromised by soiling, making the development and utilization of precise detection tools essential. A wide array of technologies has been employed, including digital and thermal cameras, advanced sensor systems, and drones. By examining these tools, their capabilities, advantages, and limitations in addressing the challenges posed by soiling are understood. A detailed classification of the reviewed papers is provided in Table 1, highlighting the main focus and methodologies employed in each study, thereby offering a comprehensive overview of the current state of research in this critical area.

Table 1. Inspection tools

Table 1 presents an overview of various inspection tools used for detecting soiling on solar panels, emphasizing their usage, benefits, and findings from multiple studies. Digital and thermal cameras, as well as cameras mounted on unmanned aerial systems (UAS), have been extensively used. Digital cameras capture high-resolution RGB images for soiling detection and measurement through image processing techniques, while thermal cameras detect temperature differences to identify soiling, useful in low-light conditions. UAS-mounted cameras enable remote and comprehensive area coverage. Additionally, a variety of sensors, including reflectometers, dust sensors, temperature and humidity sensors, and voltage and current sensors, have been employed to collect real-time data on soiling and environmental conditions. These sensors provide high precision and accuracy but can be limited by specific environmental conditions. Other tools, such as satellite information from NASA and COPERNICUS, offer remote sensing and frequent updates but can be hindered by weather and data processing complexities. Visual inspection tools, like the Magnus stereoscopic microscope, are cost-effective but prone to human error. I-V scanners provide quantitative performance data under various soiling conditions but require regular maintenance. Together, these tools offer a comprehensive approach to soiling detection on solar panels, each contributing unique strengths and facing specific limitations.

Table 2 offers a detailed summary of the various types of cameras utilized for soiling detection on solar panels, emphasizing their respective advantages and disadvantages [49]. Digital cameras offer clear visual inspection and high-resolution images, which are essential for accurately measuring soiling accumulation. They are non-invasive but may be limited in low-light conditions and can be costly. This type of camera was used in 12 papers. Thermal cameras are known for their sensitivity to temperature differences, making them effective in low-light and nighttime conditions. However, they have reduced spatial resolution. This type of camera was used in two papers. Drones equipped with cameras, such as UAS and UAV, provide excellent accessibility and remote sensing capabilities, covering wide areas without requiring physical access to the panels [50]. Despite their advantages, drones face weather limitations and limited battery life. Cameras on drones were featured in five papers.

Table 3 summarizes the findings related to the use of various sensors for detecting different elements on solar panels. Each type of sensor has been evaluated for its advantages and disadvantages, highlighting their effectiveness in different scenarios [51]. Dust sensors, used in 23 papers, are noted for their precision and accuracy in measuring the level of dust accumulation on PV panels. However, they can be costly. Temperature and humidity sensors are employed to monitor ambient conditions, which can impact the performance and soiling rate of solar panels. Voltage and current sensors collect data on the electrical performance of PV panels, providing valuable insights into the impact of soiling on energy output. Reflectometer sensors measure the reflectivity of solar mirrors, helping to quantify the extent of soiling. TCS3200 color sensors are used for color measurement and calibration, aiding in the detection of dust and other contaminants. Light-dependent resistor (LDR) offers the ability to provide real-time data on light intensity, although they may suffer from damage and have a limited lifespan. Other sensors, such as limit switch sensors, CMP21 Pyranometers, Campbell Scientific CS215 sensors, and NRG #40C Anemometers, are used to measure various environmental and operational parameters. Spectrophotometers measure the transmittance of soiled glass, helping to estimate the impact of soiling on light absorption. The findings from these sensors contribute to a comprehensive understanding of soiling detection and its impact on solar panel performance. Each sensor type brings unique strengths and limitations, emphasizing the need for a multi-faceted approach in soiling detection strategies.

Table 2. Summary of the findings of the camera usage

Table 3. Summary of the findings of the sensor’s usage

Table 4. Summary of the findings of using satellite information, visual inspection and I-V scanner

Table 4 outlines the findings related to the use of satellite information, visual inspection, and I-V scanners for detecting soiling on solar panels [52]. Each inspection tool has been utilized by different authors, with distinct advantages and disadvantages highlighted in the table. Satellite information, particularly from NASA, offers the advantage of frequent data updates, making it highly reliable for continuous monitoring. However, its effectiveness can be limited by weather conditions. This source was referenced in three papers. COPERNICUS satellite data provides excellent remote sensing capabilities, which are advantageous for large-scale and remote monitoring, though the complexity of data processing remains a significant drawback. Visual inspection tools, such as the Magnus stereoscopic microscope, offer a low-cost solution for inspecting solar panels. However, these methods are prone to human error, which can affect accuracy. General visual inspection requires no data and can be a straightforward method for detecting soiling, but it suffers from limited accuracy, especially when detailed quantitative analysis is needed. The use of the Magnus stereoscopic microscope was noted in one paper. I-V scanners, on the other hand, provide robust quantitative data, making them effective for detailed performance analysis of solar panels under different soiling conditions. They also support remote monitoring, which enhances their utility. However, I-V scanners require regular maintenance and involve complexity in their operation. The use of I-V scanners was documented in one paper. The findings from Table 4 illustrate that each inspection tool offers unique benefits and challenges. Satellite information is valuable for its broad coverage and frequent updates, though it can be hampered by weather conditions and data processing complexities. Visual inspection methods, while cost-effective, may lack the precision needed for thorough analysis. I-V scanners provide detailed quantitative data but require maintenance and can be complex to operate. Together, these tools contribute to a comprehensive strategy for detecting and mitigating soiling on solar panels.

The inspection tools used for detecting soiling on solar panels have yielded several significant findings. Various types of cameras, including standard digital cameras, thermal cameras, and drones equipped with cameras, have been employed to gather detailed information about soiling on PV panels. Digital cameras provide clear visual inspections and high-resolution images, which are crucial for precise measurement of soiling accumulation through image processing techniques. While fixed cameras offer localized monitoring, flexible or drone-mounted cameras can cover wider areas, enhancing the scope of inspection. Thermal cameras, known for their sensitivity to temperature differences, are particularly effective in low-light and nighttime conditions. They detect soiling by capturing temperature variations on the panels. The use of drones equipped with cameras has revolutionized remote sensing capabilities, allowing comprehensive inspections without physical access to the panels, which is especially useful in large-scale solar farms.

Additionally, various sensors, such as dust sensors, temperature sensors, voltage sensors, and current sensors, have been extensively used to gather real-time data on soiling levels and panel performance. Dust sensors measure the level of dust accumulation by detecting particulate matter size and concentration, while temperature sensors monitor ambient and panel surface temperatures, aiding in the detection of soiling that affects thermal properties. Voltage and current sensors provide data on electrical performance, helping to quantify the impact of soiling on energy output. Other inspection tools, including satellite information, visual inspection tools like microscopes, and I-V scanners, have also been employed. Satellite data, combined with tools like the Google Earth Engine, is used to monitor soiling patterns over large areas, particularly in arid regions. Visual inspection tools offer detailed observations of soiling particles and their impact on panel surfaces, while I-V scanners analyze the electrical characteristics of PV panels under different soiling conditions, helping to understand performance variations.

The integration of various inspection tools provides a comprehensive approach to detecting soiling on solar panels. Cameras, with their high-resolution imaging capabilities, are highly effective but may face limitations in certain weather conditions. Sensors offer precise real-time data but can also be affected by environmental factors. Combining these tools with advanced technologies like drones and satellite data enhances inspection capabilities and provides a holistic view of soiling impacts.

Future research should focus on integrating advanced technologies such as artificial intelligence and machine learning with existing inspection tools to improve the accuracy and efficiency of soiling detection. Developing AI algorithms that can analyze data from multiple sources, including cameras, sensors, and satellites, in real-time could significantly enhance detection capabilities. Research should also aim to develop more robust and weather-resistant sensors that can operate effectively in diverse environmental conditions. Exploring the use of novel materials and technologies for sensors to improve their sensitivity and accuracy is another promising area.

Moreover, investigating cost-effective solutions for soiling detection that can be widely adopted, especially in large-scale solar farms, is essential. Developing low-cost, high-efficiency tools that can be easily maintained and deployed will facilitate broader adoption. Emphasis should also be placed on comprehensive data analysis techniques that can integrate data from various inspection tools to provide a more accurate assessment of soiling. Utilizing big data analytics to identify patterns and trends in soiling across different geographical locations and environmental conditions can further advance the field.

By addressing these future research directions, the field of soiling detection on solar panels can advance significantly, leading to improved solar panel efficiency and performance.

4.2 Models

This section examines various research papers focusing on the use of models in the field of soiling detection on solar panels. Table 5 summarizes the studies, highlighting the types of models employed, their authors, years of publication, and specific focuses. The models explored include statistical and regression models, stochastic models, empirical models, and deep learning models, each offering unique approaches to analyzing and predicting the impact of soiling on PV systems. The effectiveness, advantages, and limitations of these models are discussed, providing a comprehensive understanding of their application in optimizing solar panel performance and maintenance.

Table 5 details a comprehensive overview of various models used for soiling detection on solar panels, covering their applications, benefits, and specific focus areas. Statistical and regression models have been used to compare the effectiveness of visual inspection and infrared thermography, study the impact of surface contaminants on energy generation, and predict daily CI changes based on environmental variables. These models also investigated the impact of dust accumulation using transmittance loss, PM deposition-based, and empirical models. Stochastic models, such as dynamic models using Markov chains, accounted for dust accumulation and seasonal variations. Empirical models analyzed the effect of dust on transmittance and quantified soiling rates using a temperature-corrected performance ratio (CPR). Deep learning models utilized convolutional neural networks for dust detection and classification, while machine learning models employed various algorithms to detect cleanliness levels and estimate soiling. These studies highlight the diverse approaches and technologies used to enhance the detection and analysis of soiling on solar panels, each offering unique benefits and facing specific challenges.

Table 5. Models for soiling detection

Table 6. Summary of the findings of using statistical and regression models, stochastic models and empirical models

A concise overview of the findings from the application of statistical and regression models, stochastic models, and empirical models in detecting soiling on solar panels is presented in Table 6 [66, 67].

Statistical and regression models, utilized in nine papers, include regression and linear models, which are widely used to analyze relationships between environmental factors and soiling accumulation. These models effectively predict trends in soiling ratios and estimate power loss due to dust accumulation. Their key strength lies in their simplicity and interpretability, making them useful for initial assessments and historical data analysis. However, their performance can be limited by input data errors, and they struggle with handling complex, nonlinear relationships and large datasets with multiple interacting variables. Empirical studies have shown that while regression models provide a reasonable estimation of soiling effects, they lack the adaptability required for dynamic environmental conditions.

Stochastic models, such as the Markov chain model used in one study, introduce probabilistic approaches to capture the randomness in soiling accumulation. These models are particularly useful in scenarios where environmental variables fluctuate unpredictably, allowing for more robust predictions. The primary advantage of stochastic models is their ability to incorporate uncertainty, making them suitable for long-term soil forecasting. However, they are highly sensitive to parameter selection and require extensive historical data for calibration. While these models provide more flexibility than deterministic approaches, their computational complexity and reliance on extensive data make them challenging to implement in real-time applications.

Empirical models, discussed in eleven papers, cover a range of applications, including estimating the thickness of accumulated dust layers and assessing power loss. Examples include transmittance loss models, PM deposition-based models, and mathematical models based on Brownian motion, impaction, and sedimentation. The main advantage of empirical models is their adaptability to specific locations and conditions, as they are derived directly from experimental data. However, their performance is highly dependent on the availability of high-quality data, and they often require periodic recalibration. While these models are effective for short-term predictions, they may lack generalizability across different climatic regions.

Table 7 features a concise summary of the findings from the use of deep learning models in detecting soiling on solar panels [68]. Deep learning models, particularly CNN and ANN, have revolutionized soil detection by enabling automated feature extraction from images and sensor data. CNNs have demonstrated exceptional performance in detecting soiling patterns using aerial and terrestrial images, reducing the need for manual inspection. These models excel in feature learning and can generalize well to new datasets when trained effectively.

Table 7. Summary of the findings of using deep learning

Table 8. Summary of the findings of using machine learning

Table 9. Comparative summary of each model

CNNs, for example, analyze pixel-level variations to detect dust accumulation on solar panels, improving accuracy in classification tasks. They can outperform traditional machine learning models by learning hierarchical features directly from raw image data. ANNs, on the other hand, are well-suited for nonlinear relationships between environmental conditions and soiling accumulation, providing robust predictive capabilities. However, deep learning models are data-intensive and require high computational power, making their deployment challenging in low-resource settings.

Despite their advantages, deep learning models also have limitations. Their effectiveness heavily relies on the availability of large, high-quality labeled datasets, and they are prone to overfitting when trained on limited or unbalanced data. Computational requirements can also pose a challenge, as training complex networks requires substantial processing power and memory. Furthermore, interpretability remains a concern, as deep learning models function as "black boxes," making it difficult to understand how predictions are generated.

Empirical studies highlight that while CNNs achieve state-of-the-art accuracy in soiling detection, their dependence on large labeled datasets remains a key limitation. Some research suggests that transfer learning techniques and data augmentation methods can mitigate these issues by leveraging pre-trained models and enhancing dataset diversity. Additionally, hybrid models combining deep learning with traditional machine learning techniques have shown promise in improving accuracy and reducing computational complexity.

A concise summary of the findings from the use of machine learning models in soiling detection for solar panels is shown in Table 8 [69]. Machine learning models have gained recognition in soiling detection due to their ability to process large datasets and identify complex patterns in soiling accumulation. These models, including K-means clustering, K-Nearest Neighbor (KNN), Neural Networks, Random Forest, Support Vector Machines (SVM), Naïve Bayesian classifiers, Fisher's Linear Discriminator, Multilayer Perceptron, and XGBoost, excel in detecting soiling from image-based data and sensor measurements, offering higher accuracy compared to traditional statistical methods.

A key advantage of machine learning models is their adaptability to diverse environmental conditions and their ability to improve over time as more data becomes available. For instance, Random Forest and XGBoost, both ensemble learning techniques, demonstrate high predictive accuracy by aggregating multiple decision trees. SVMs are effective in distinguishing between clean and soiled solar panels based on feature extraction techniques, while KNN models leverage similarity-based learning to classify different levels of soiling.

Despite their strengths, machine learning models also have limitations. Their performance depends heavily on the availability of high-quality training data, and they require significant computational resources for real-time applications. Moreover, certain models, such as SVMs and neural networks, can be challenging to interpret, making it difficult to understand the decision-making process. Additionally, overfitting remains a common concern, particularly when models are trained on limited or biased datasets.

Different models have been implemented to process data for detecting soiling on solar panels, each with distinct strengths and limitations. Statistical models mathematically represent relationships between variables, making them effective for analyzing data and making predictions when patterns are clear. They are particularly useful for identifying trends in historical data, but struggle with complex and nonlinear relationships.

Stochastic models incorporate randomness or uncertainty to account for probabilistic factors such as weather patterns. These models are valuable for long-term soiling predictions but require large datasets for accuracy and are less efficient with smaller datasets. Empirical models, which rely on observed data and experiments, offer adaptability and simplicity but are heavily dependent on data availability and site-specific conditions.

Machine learning models enhance soiling detection by identifying trends and anomalies across diverse environmental conditions. They excel in tasks such as image recognition and real-time monitoring, but require extensive data for training. Deep learning models provide advanced capabilities in feature learning and predictive modeling, making them ideal for analyzing complex data structures like image-based assessments of solar panels. However, both machine learning and deep learning models demand significant computational resources and large training datasets.

A comparative analysis highlights the effectiveness of these models under different conditions. Statistical and regression models, such as linear and empirical models, are valued for their simplicity and ability to establish clear relationships between variables, but are best suited for structured data with well-defined patterns. Stochastic models, like the Markov chain model, effectively incorporate randomness and seasonal variations, making them useful for long-term soil predictions, though their accuracy depends on large datasets.

Empirical models estimate the impact of soiling on transmittance and energy loss, offering practical insights but relying heavily on high-quality experimental data. In contrast, machine learning models, including decision trees, SVM, and artificial neural networks, provide superior predictive capabilities by identifying complex patterns and trends in large datasets. These models are particularly well-suited for adaptive soiling detection and real-time monitoring. Deep learning approaches, such as CNNs and LSTM networks, achieve the highest accuracy in image-based soiling detection, making them ideal for automated inspection systems, though they require extensive computational resources and large training datasets to maintain performance.

The choice of model depends on the specific requirements of the application. Empirical and statistical models are advantageous for rapid estimations with minimal computational costs, while machine learning and deep learning models excel in automated detection and large-scale monitoring. Future research should explore hybrid approaches that integrate multiple models, leveraging their strengths to improve accuracy and efficiency in diverse environmental conditions.

To provide a clearer comparison, Table 9 summarizes the predictive capabilities, strengths, and weaknesses of each model discussed.

4.3 Cleaning and mitigation systems

This section examines various research papers focused on cleaning and mitigation systems for soiling detection on solar panels. Table 10 provides an overview of the studies, highlighting the authors, publication years, and specific focus areas. These systems employ a variety of innovative approaches, including IoT technologies, robotic systems, anti-soiling coatings, and automated cleaning mechanisms, to address the challenges posed by dust and dirt accumulation on PV panels. The effectiveness, advantages, and limitations of these cleaning and mitigation strategies are discussed, offering insights into their application for optimizing solar panel performance and maintenance.

Various publications focus on designing and proposing mitigation and cleaning systems to address dust accumulation on solar panels. Shields, such as 1D and 2D shields, have been used to mitigate dust, demonstrating effective results. Studies have shown that 1D shields perform better than 2D shields. Additionally, combining 1D shields with antistatic coatings and vibrating the panels yielded encouraging results in dust mitigation. Electrostatic cleaning devices have also been employed, providing high efficiency in cleaning performance.

Other proposed systems for dust mitigation or removal include automatic cleaning using Arduino Uno microcontrollers, automated robotic systems, and anti-soiling coatings. Automatic cleaning systems using Arduino Uno microcontrollers have maintained solar panel performance through washing mechanisms activated upon detecting dust accumulation. The use of automated robotic systems has highlighted efficiency and safety improvements, as well as enhancements in PV performance. Coated mirrors have shown a lower soiling index compared to uncoated mirrors, effectively reducing dust particles.

Future research should focus on a comprehensive analysis of cleaning systems to identify the most cost-effective and accurate solutions. Comparisons between existing systems should be made to recommend the best approach. Additionally, new systems leveraging emerging technologies and trends should be proposed, aiming to surpass the performance of existing systems. Evaluating and comparing the challenges and maintenance requirements of proposed and new systems is essential to ensure long-term effectiveness and sustainability.

Table 10. Cleaning and mitigation systems