Fault Detection and Identification for Maintenance Management

Abstract

Photovoltaic solar energy is increasing the energy production due to the technological advances, together to the initial investment reduction. Solar farms are being installed with larger production capacity, improving the technical challenge for developing correct and efficient maintenance management. The photovoltaic maintenance management requires to increase the reliability and reduce the operating costs. The photovoltaic panels inspection with unmanned aerial vehicles is an efficient condition monitoring technique, analyzing large areas and obtaining accurate thermographic images. Due to the large amount of data, it is necessary the use of image processing algorithms for automatic identification of faults. Despite these advances, it is required the identification of the type and the importance of the fault. This information will be used by the plant operators for developing efficient maintenance management plans. The novelty developed in this work is a robust decision system for photovoltaic maintenance management, based on the combination of image processing for fault detection and statistic techniques. The first phase of the methodology is the extraction of interest areas or possible faults with neural networks trained for this purpose. The second phase develops the statistical analysis of the radiometric data of the area detected as possible fault with neural network. The radiometry data of these areas will be analyzed with statistic models with the aim of detecting patterns for detect identification and quantification. A real case study of a solar plant is presented, and the results obtained with this methodology provide the positioning and importance of each defect, probing the strength of the method.