A new data normalization approach applied to the electromechanical impedance method using adaptive neuro-fuzzy inference system

Impedance-based structural health monitoring (ISHM) has been shown as a promising technique to detect failures at an early stage. As structural changes occur, the measured impedance signature will reflect such changes which allows damage diagnosis to be performed. However, environmental or operational changes can also cause changes in the impedance signature. Hence, in order to prevent false diagnosis, a data normalization is required. The aim of this work is to propose a new data normalization technique by determining fuzzy rule-based system (FRBS) through the adaptive neuro-fuzzy inference system (ANFIS). The training was carried out with the input variables temperature and frequency, and the output data are signature impedance values from baseline states. Temperature changes were monitored in order to implement the data normalization. For this aim, it is necessary to compensate for the effect of this variable for later prediction of impedance signatures without damage, at temperatures that were not necessarily observed in the data collection. Results obtained in the validation indicate a good accuracy of the predicted signatures since the highest correlation coefficient deviation (CCD) damage index obtained was 0.0038. For the validation phase, part of the baseline data was used for training the FRBSs and another part of the baseline data was used for the validation itself. Next, all baseline data were used in the training in order to obtain the FRBS. The CCD values between the baseline signatures from the experiment and the reference predicted for the respective temperature were close to zero, indicating good agreement of the models. Finally, the methodology proposed in this work was used for damage detection in an experiment to detect corrosion related damage in metallic structures.