Using airborne images of heliostats to monitor glass defects


Supervisory Chair


External Member

Dr Arifur Rahman


A large fraction of the O&M expenditure  for a Concentrating Solar Thermal Power (CSP) plant is associated with cleaning the mirror field.  ANU is participating in an ARENA-funded project to carry out robotic inspection of soiling levels of heliostats in a CSP plant, by flying drones above the mirrors, gathering data about cleaniness levels and processing this data so that the plant operator knows which areas of the heliostat field to preferentially clean. 

In this project, the objective is to use this data to identify defects in the mirrors, and monitor how the defects evolve over time.  Defects may include cracks in the glass, areas of delamination and corrosion.  This project involves interpretation of real data obtained from drone flights, and implementation of computer vision methods. 

Defect diagnosis can be addressed using image texture descriptors, and standard pattern recognition models. (SVM, K-Nearest Neighbours). An alternative approach is to learn feature extractors automatically from training data without human knowledge. One of these methods, Convolutional Neural Networks (CNNs), has been used to set the state-of-the-art in many visual recognition tasks, but is not yet explored for the task of defect diagnosis.


Updated:  1 June 2019/Responsible Officer:  Dean, CECS/Page Contact:  CECS Marketing