Predictive maintenance for solar photovoltaic systems: deep learning-based anomaly identification using infrared thermography

The purpose of this Ph.D. thesis is to portray deep learning applications for the Predictive Maintenance of Solar Photovoltaic Systems to identify Infrared Thermographic anomalies. The research work carried out initiated the conceptualization, design, implementation, and evaluation of a novel Intelligent Fault Detection and Diagnosis (IFDD) System, alternatively realized as an Intelligent Remote Inspection (IRI) System for large-scale (and utility-scale) Solar Photovoltaic (SPV) Arrays having large surface areas, by deploying drone-aided Infrared Thermographic Imaging and process acquired thermal data by employing artificial intelligence (AI) techniques to precisely locate and capture SPV panels’ thermal degradation pattern for operational decision-making, diagnostic monitoring and prompt maintenance. SPV cells are fragile, and temperature is a critical environmental factor for their cascaded and cyclical thermal stresses and irreparable degradation, which are further accelerated by periodical and prolonged heatwave spells and global warming effects. The emergence of AI techniques, particularly deep learning (DL) algorithms leverage the realm of diagnostic monitoring and predictive maintenance (PdM). The thesis focuses on creating accessible, explainable end-to-end pipelines employing the DL framework, and drawing and preparing datasets from public repositories. Besides, acquiring, preparing, and labeling a quality, calibrated, and noise-free SPV panel’s thermal dataset (floating-point temperature intensity values in degrees Celsius) is a challenging task, as SPV energy-producing assets are expected to operate under healthy conditions and achieving targeted output continuously, resulting in fewer faulty labeled instances, whereas a large amount of data is required to train DL models for multiclass classification (or thermal diagnosis). The thesis work is organized into five chapters: Chapter 1 introduces the objective of Predictive Maintenance (PdM), including Fault Detection and Diagnosis (FDD), Solar Photovoltaic (SPV) energy market, global warming effect, and contribution. Chapter 2 provides a comprehensive overview of the state-of-the-art Artificial Intelligence (AI) techniques in Solar Photovoltaic (SPV) diagnostics, including their application and implementation challenges encountered while selecting and deploying AI algorithms, various Deep Learning (DL) models employed for diagnosis, explainability methods, and Infrared Radiated Thermographic (IRT) inspection for SPV power systems. Chapter 3 describes a theoretical framework for Convolutional Neural Network (CNN) comprising its architecture, layers, and training process. Chapter 4 describes the contributions proposed in Monitoring Diagnosis for Predictive Maintenance of Solar Photovoltaic Infrared Thermography. Lastly, Chapter 5 summarizes the work accomplished in this thesis and offers insights and considerations towards future works.