Automatic detection of deteriorated photovoltaic modules using IRT images and deep learning (CNN, LSTM) strategies. (April 2023)

• Record Type: Journal Article
• Authors: Bakır, Hale; Kuzhippallil, Francis A.; Merabet, Adel
• Is Part Of: Engineering failure analysis. Volume 146(2023)
• Publication Date: 2023-04
• Subjects: Photovoltaic system -- Convolutional neural network -- Automatic fault detection -- Image classification; System failures (Engineering) -- Periodicals; Fracture mechanics -- Periodicals; Reliability (Engineering) -- Periodicals; Pannes -- Périodiques; Rupture, Mécanique de la -- Périodiques; Fiabilité -- Périodiques; Fracture mechanics; Reliability (Engineering); System failures (Engineering); Periodicals; Electronic journals; 620.112
• DOI: 10.1016/j.engfailanal.2023.107132
• Languages: English
• ISSNs: 1350-6307

Abstract

Highlights: An automatic classification system, for faults in PV modules, is proposed based on thermographic images of working and faulty PV modules. The convolutional neural network (CNN) is used to extract features from IRT images of the working and faulty PV modules. The CNN performance is further validated by comparing its performance with long-short term memory (LSTM) under different number of images involved in the training and validation. The proposed tool enables the immediate detection of a fault or malfunction in the PV modules and its accuracy is very high. Abstract: Faults in photovoltaic systems cause a reduction of efficieny due to electricity production losses. Faults, due to overheating in photovoltaic modules, can be detected using thermographic testing that ensures a quick inetvention to correct the operation of the photovoltaic system with cost-effective tools, while maintaing a normal operation of the system. This study proposes a fault detection system for classfying the faults in the photovoltaic modules, based on the presence of hotspots, using thermographic images and a deep learing classifier based on a convolutional neural network. The thermographic images were obtained from a drone flying over the photovoltaic power farm. Manual inspection is time consuming due to the individual examination of multiple thermographic. This work presents an automotic fault detection system based on a convolutional neural network that immediately recognizes the hotspot fault in the photovoltaic module with high accuracy. The conventional neural network was trained and validated with datasets of 300, 500 and 1000 images and compared to another deep learning toool based long short-term memory neural network. It was found that the conventional neural network, with dataset of 1000 images, achieves an accuracy of 95.05 % under a computation time of 1 h and 30 min. Furthermore, it was found that its performance is better compared to the long short-term memory method with respect of different evaluation metrics.