Cluster and discriminant analysis of electrochemical noise statistical parameters

In order to identify the different pitting states, some pattern recognition (PR) procedures, including principal component analysis (PCA), hierarchical agglomerative cluster analysis (HACA) and linear discriminant analysis (LDA), were applied for analyzing electrochemical noise (EN) statistical parameters from a typical pitting system of Q235 carbon steels in NaHCO3 + NaCl solutions. Firstly, according to the PCA results, the EN mean value (E¯ and I¯) and standard deviation (σE and σI) were determined as the descriptors for clustering. Then, using the selected four statistical parameters as variables, the cases from different pitting states were classified by the HACA to three clusters, which relates to the metastable state, intermediate state and stable state, respectively. It shows a good agreement with the classification obtained from k-means cluster with E and log|I| as variables. Based on the cluster results, the pitting states of the ungrouped data points from the similar pitting processes also can be distinguished according to the established discriminant function(s).