Fault Detection based on neuro-fuzzy and mathematical models applied to a solar inverter

Abstract

This paper presents an implementation of a fault detection scheme based on the identification of neuro-fuzzy models of a photovoltaic system in the AC conversion circuit. Fast fault detection implies the ability to perform preventive rather than corrective maintenance, which represents a benefit in the economical, material and environmental fields. Therefore, a method based on fuzzy logic with fuzzy residuals is proposed for the correct and fast detection of different types of faults. A residual evaluation is performed and a decision is made to what kind of threshold for robust fault detection is better. The results obtained show the good performance of the proposed scheme, where the detection system demonstrates its robustness to different faults. The proposed method is compared with traditional methods like mathematical model based on state-space equations and detection methods like Unknown Input Observers (UIO). The methods were applied into a real solar inverter and the results obtained were similar and as good as in simulations.