Lysyi A. Cyber-physical systems for monitoring defects in photovoltaic modules of solar power plants.

Enhancing the fire safety of solar energy facility operation is an important scientific task in the field of information technologies focused on the development of cyber-physical systems (CPS) for monitoring defects of photovoltaic modules of solar power plants (PVMSPP). The dissertation analyzes the current state, methods, and tools for monitoring defects of PVMSPP. The work develops an architecture and methods of functioning of cyber-physical systems for monitoring defects of photovoltaic modules of solar power plants. The object of the research is the process of monitoring defects of PVMSPP using software and hardware tools. The subject of the research is the methods and tools of CPS for defect monitoring using software and hardware tools with distributed data processing. The purpose of the dissertation research is to ensure prompt differentiation of photovoltaic module operating modes such as fire, fire hazard, and protection activation based on the application of a CPS architecture with distributed data processing for monitoring defects of PVMSPP. Scientific novelty of the obtained results: For the first time, an architecture of CPS for monitoring defects of PVMSPP has been developed based on the concept of edge–cloud data processing distribution. The novelty of the architecture lies in its formation on the principles of distributing computations between the onboard and ground control systems of an unmanned aerial vehicle (UAV), the dispatch control system, and a cloud service. The method of data processing by software and hardware tools of the UAV onboard control system during monitoring of photovoltaic module defects has been improved. The method is distinguished by the introduction of a model for determining the optimal camera viewing direction, the application of ensembling of multi-palette thermograms and RGB images, and the transformation of pixel coordinates of detected defects into geographic coordinates. The method of ensembling multi-palette thermograms and RGB images for detecting defects of photovoltaic modules has been improved. The method differs by the development of a mathematical model for classifying defects according to their relative area with respect to the area of a single photovoltaic cell and by the introduction of post-processing to create a composite thermo-RGB image. For the first time, a method for functioning of CPS for monitoring defects of PVMSPP has been developed. The novelty of the method lies in the implementation of the edge–cloud data processing concept. Based on the application of the developed methods, the average precision indicator of defect detection of PVMSPP has been increased by 2–3%, and the value of the integral indicator of accuracy and completeness of defect detection according to the F1-score metric exceeds 90%. An algorithm for using the CPS for monitoring the PMSC has been developed. The theoretical and practical results of the research have been implemented in the development of safety systems in several organizations and in the educational process of the university. The introduction substantiates the relevance of the scientific task related to the development of methods and tools for monitoring defects of PVMSPP and presents the main scientific results and practical significance of the work. The first chapter analyzes the current state of monitoring defects of PVMSPP, their main types, and the existing methods for defect monitoring. The second chapter presents the development of the CPS architecture for monitoring defects of PVMSPP based on the edge–cloud data processing concept. The software and hardware tools of the architecture based on a convolutional neural network onboard the UAV are substantiated. The third chapter presents the improved method of data processing by software and hardware tools of the UAV onboard control system during monitoring of photovoltaic module defects, the improved method of ensembling multi-palette thermograms and RGB images for defect detection using the YOLOv12m-seg convolutional neural network model, and the developed method of CPS functioning for monitoring defects of PVMSPP. The fourth chapter presents the results of experimental studies on detecting defects of PVMSPP. The conclusions summarize the obtained scientific and practical results of the research. The appendices present implementation certificates of the research results.