Statistical analysis of a corrosion inhibitor family on three steel surfaces (duplex, super-13 and carbon) in hydrochloric acid solutions

• Published in: Electrochimica acta Vol. 53; no. 2; pp. 434 - 446
• Main Authors: Baddini, Ana Luísa de Queiroz, Cardoso, Sheila Pressentin, Hollauer, Eduardo, Gomes, José Antonio da Cunha Ponciano
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 01.12.2007; Elsevier
• Subjects: Applied sciences; Corrosion; Corrosion mechanisms; Corrosion prevention; Exact sciences and technology; Inhibitor; Metals. Metallurgy; OLS; PLS; Statistical analysis; Inhibitor; Statistical analysis; Corrosion; PLS; OLS; Alcohol; Steel; Corrosion protection; Thioureas; Carbon; Duplex; Corrosion inhibitor; Hydrochloric acid; PLS regression; Amine; Chemometrics
• ISSN: 0013-4686; 1873-3859
• DOI: 10.1016/j.electacta.2007.06.050

Previous studies have addressed the experimental and theoretical investigation of the inhibition corrosion efficiencies (ICE) of single metal surfaces. Along this line we carried out calculations concerning to 23 compounds on three different single-steel surfaces, duplex, super-13 and the carbon steel in hydrochloric acid (15% w/v) solutions. The overall experiment is composed of 69 results of weight loss ICEs at 60 °C for amines, alcohols, thiourea and its derivatives acting as corrosion inhibitors for three steel surfaces. In these studies ICEs were correlated with group and quantum AM1 descriptors through the use of three different statistical methodologies based on calibration and validation of regular and modified OLS and PLS (partial least squares) methods. All calculations have shown better results using weight isoesteric Langmuir adsorption function (WILA function), ln( θM/(1− θ)) or ln K ads, calculated from the weight loss data as the response function. The function −log( i) has been used, as well, on all comparisons. Variables describing the metal were added to the previous set of group and quantum IC variables and several models have been designed to fit the three-steel problem. Simple products of metal and IC variables with 250 (25 × 10) products were tested as model I. Selection of the best variable set was carried out for the calibration and validation procedures and these calculations indicated very few descriptors in common, i.e. each particular selection (calibration or validation) finds its own optimal descriptor set. The overall results showed excellent correlations with R 2 values between 0.80 and 0.96 and a Q 2 values from 0.75 to 0.93. We are unaware of any similar QSPR study on the steels here studied, and neither the study of such massive amount of data concerning molecular inhibitors on three different steel surfaces. Our best result for the second-order cross-validation descriptor selection employs 29 variables, Y 29. The results accurately fitted all 69 corrosion inhibitors experiments within 5% accuracy over three different steel types. A second model was designed with all 630 binary products of the metal/IC interface (((35 + 1) × 35)/2). This model uses the variables of model I plus all simple squares of the primary data. Due to the large number of composed variables we carried out calculations based on the classical partial least squares (PLS). Our best result employed nine main components that accurately fitted the 69 corrosion inhibitors experiments with obtained calibration coefficients, R 2, values of 0.95 and Q 2 values of 0.83. Both results showed excellent performance compared to previous fits found in the literature. Most of the obtained results are easily transferable to other similar many-steel studies through a simple data addition concerning the new metal surface.