Conference: Assessment of Deep Learning Algorithms for Fault Diagnosis of Solar Thermal Systems

Abstract

Solar hot water (SHW) systems are viable and sustainable devices for hot water domestic and industrial energy needs. Nevertheless, the efficient operation of these systems can be compromised if the necessary maintenance measures are not implemented. Degradation of components and malfunction in SHW systems may undergo unnoticed when coupled to traditional auxiliary energy sources. Detailed and continuous monitoring to counter this, however, elevates the overall cost of the system and thus, other methods have been explored for performance assessment and fault detection. Data-driven techniques became popular as Prognosis and Health Management approaches in mechanical components for detection, diagnostics and prognostics of complex systems. In this article, Deep Learning algorithms, such as ANN, RNN and LSTM, are analyzed as alternatives for performance prediction and anomalous behavior detection in a solar hot water system. TRNSYS simulation software is used to generate synthetic operation data for the system for nominal operational and fault-induced conditions. Similar results were obtained for the temperature predictions, with the LSTM models obtaining a lowest combined RMSE of 1.27°C, MAE of 0.55°C and variance 0.52 °C^2 , as well as the lowest relative prediction errors of 3.45%, indicating a more reliable performance. Using this model, the prediction-based anomaly detection was tested under different meteorological conditions, where overheating and heat underproduction anomalies were detected with a mean accuracy of 85% and 82%, respectively.

Keywords: Solar hot water systems, performance forecast, anomaly detection, Deep Learning.

View here

Recommended citation: Camila Correa Jullian, Jose Cardemil, Enrique López Droguett, and Masoud Behzad. Assessment of Deep Learning Algorithms for Fault Diagnosis of Solar Thermal Systems. In Proceedings of the ISES Solar World Congress 2019, pages 1–12, Freiburg, Germany, nov 2019. International Solar Energy Society. doi:10.18086/swc.2019.08.03