Search documents

Browse topics

Document details

Development of Predictive Maintenance Algorithms for Photovoltaic Systems Using Synthetic Datasets
E.A. Sarquis Filho, F.C. Santos, P.J. Costa Branco
PV System, Predictive Maintenance PdM, Machine Learning, Failure Detection and Diagnosis
PV Systems and Storage – Modelling, Design, Operation and Performance
Subtopic: Operation, Performance and Maintenance of PV Systems
Event: 37th European Photovoltaic Solar Energy Conference and Exhibition
Session: 5CV.3.47
1584 - 1589
ISBN: 3-936338-73-6
Paper DOI: 10.4229/EUPVSEC20202020-5CV.3.47
0,00 EUR
Document(s): paper, poster


Automatic failure detection and diagnosis (FDD) is a key tool for the application of predictive maintenance. The use of fault data synthesized with simulation models is an alternative solution to circumvent the difficulty of data for training machine learning FDD algorithms. In this study, we have built a synthetic dataset of failures for training a fully connected neural network (FCNN) for FDD. The resulting algorithms were tested with experimental data to identify the most consistent and best performing FCNN architecture, which achieved a true positive rate (TPR) above 50% for partial shading and above 90% for all other conditions. The tested technique proved to be suitable for the detection of mild failures such as cell short-circuit. However, its performance consistency depends on the hyperparameters, which should be selected by testing with real data. Moreover, achieving high TPR values depend on clear distinction between the failures’ signatures. A synthetic dataset of failures is a powerful resource to study the overlap of signatures to develop automatic FDD algorithms.