Visual Fault Diagnosis in Photovoltaic Modules using a Combination of Machine Learning and Deep Learning Techniques

Abstract

Photovoltaic (PV) systems usage has grown exponentially in recent times owing newlineto several factors like increasing energy demand, accelerated fossil fuel depletion and newlinerising global warming scenarios. In general, PVM are prone to fault occurrences due newlineto various reasons like continuous outdoor operations, dynamic climatic conditions, newlinethermal stresses, partial shading (irradiance), external damages, short circuits, moisture newline(corrosion) and many more. Presence of faults in PVM can significantly affect the output newlinepower production, operational lifetime, reliability and safety of the modules. In newlinerecent times, fault detection and diagnosis have evolved into an essential phenomenon newlinethat conserves the power output by enhancing the safety, operational lifetime and reliability newlineof PVM. Currently, drastic developments in deep learning techniques, sophisticated newlineequipment for imaging and trained professionals have made fault detection and newlinediagnosis in PVM more convenient, effective, minimal time consuming and accurate. newlineAdditionally, deep learning techniques have displayed superior performance in various newlinetasks like object detection and image classification. Considering the advantages stated newlineabove, the usage of deep learning techniques can be extended towards the identification newlineof visual faults in a PVM with the help of images acquired from a digital camera (RGB newlineimages) equipped in an unmanned aerial vehicle (UAV). This thesis aims at developing newlinea combined approach of deep learning and machine learning to identify visual faults in newlinea PVM accurately. The steps carried out in the research work that helped in formulating newlinethis thesis are provided below. newlineVisual Fault Identification - The major visual faults occurring in a PVM was identified newlinebased on the literature survey conducted. In the present work, five most commonly occurring newlinevisual fault conditions namely, glass breakage, delamination, snail trail, burn newlinemarks and discoloration along with healthy panel (to distinguish among faulty and newlinehealthy conditions) were considered. newlineData Acqui