Neuro-Fuzzy Systems Applied to Structural Health Monitoring by Electromechanical Impedance

Abstract

This study investigates structural health monitoring (SHM) to identify potential damage in structures. This work focuses on applying SHM in oil and gas facilities. The main objective is to perform temperature compensation for electromechanical impedance data using two distinct neuro-fuzzy systems: the adaptive neuro-fuzzy inference system (ANFIS) and the hybrid neural fuzzy inference system (HyFIS). Both techniques combine neural networks with fuzzy sets, considering the uncertainties of the problem variables. The data was collected using PZT (lead zirconate titanate) patches installed on a steel plate subjected to five types of damage. The experiments took place in the field under varying environmental conditions. Part of the data was used to train the neuro-fuzzy networks that build the FRBS, with temperature and frequency as inputs and the real part of the impedance as output. A comparative analysis was performed by calculating the correlation coefficient deviation (CCD) between the results of the fuzzy rule-based systems (FRBS) generated by ANFIS and HyFIS and the experimental data. The results were promising, with HyFIS achieving over 90% accuracy in the experiment. The HyFIS FRBS enabled the measurement of impedance differences between baseline values and those observed under the five types of damage using three defined metrics. HyFIS was chosen due to its higher precision in validation compared to ANFIS. In summary, neuro-fuzzy networks applied to SHM have shown promising results in optimizing the training process for damage diagnosis, suggesting that the technique can be effectively used in the context of petroleum product storage tanks, which is the focus of this work.