Automatic classification of polymer coating quality using artificial neural networks

Abstract

Two artificial neural networks (ANN)—one for classification of polymer coating quality based on phase angle (Φ)-log frequency ( f) data and one for classification based on log impedance modulus (/ Z/)-log f data—have been trained using three sets of theoretical impedance spectra for polymer coated steel-spectra for good, intermediate and poor coating quality. The trained ANNs have been tested using experimental impedance spectra for six different polymer coating systems on steel collected during exposure at a remote marine test site for exposure periods up to one year. In general, excellent agreement between the predictions of coating quality made by experienced operators based on general features of the impedance spectra and parameters such as breakpoint frequency f b and pore resistance R po on the one hand and the classification results obtained from the ANNs on the other hand was obtained. Evaluation of the results of these analyses was made easier by introduction of the coating quality index (CQI) which has values between 0 and 1. Occasional discrepancies observed between classification results based on Φ-log f data vs. log / Z/-log f data occurred in the transition region between two types of classification, e.g. between intermediate and poor. These discrepancies have been explained based on the experimental data for R po and f b and their time dependence.