Solar panel damage identification using tensorflow lite

The number of utility-scale PV installations is rising, with a power capacity of 12.5 Gigawatts installed in 2021, 10.4 in 2022, and an estimated 24 Gigawatts installed in 2023 [1]. With larger-scale installations, quicker ways of identifying and locating damaged PV arrays are needed. The solution presented in this thesis is to use drones to capture aerial photos and TensorFlow-Lite and Keras deep learning methods to determine if a panel has defects, such as debris, cracked panels, and hotspots. The model features an execution time of 0.185 seconds per picture. In addition, the model will run on an embedded system with a relatively low impact on power consumption, minimizing the reduction of flight time. The Raspberry Pi has an approx. 0.1minute effect on flight time while idling and with the worst-case scenario of affecting flight time by approximately two minutes if left running for the entire flight.

Grantor dc:publisher

University of Tennessee at Chattanooga

Author dc:creator

• Muncie, Garrick

Contributors dc:contributor

• Ofoli, Abdul
• Ahmed, Raga; Reising, Donald
• College of Engineering and Computer Science

• Deep learning (Machine learning)
• Electric power production--Measurement
• Photovoltaic power systems
• Raspberry Pi (Computer)
• Solar panels--Maintenance and repair

Licence

• http://rightsstatements.org/vocab/InC/1.0/

Language dc:language

English, eng

Repository record dc:identifier

https://scholar.utc.edu/theses/871

OAI identifier oai:identifier

oai:scholar.utc.edu:theses-2049