Multilayer Perceptron Model for the prediction of corrosion rate of Aluminium Alloy 5083 in seawater via different training algorithms

Abstract Corrosion inhibitor is often opted as a corrosion protection method for various industries worldwide. The development of eco-friendly corrosion inhibitor has become a trending concern due to the various environmental regulations impose by several countries. However, a laboratory testing wou...

Full description

Saved in:
Bibliographic Details
Main Authors: Zulkifli, F, Abdullah, S, Suriani, M J, Kamaludin, M I A, Wan Nik, W B
Format: Conference Proceeding
Language:English
Subjects:
Algorithms
Aluminum
Aluminum base alloys
Artificial neural networks
Bayesian analysis
Chemical analysis
Correlation coefficient
Correlation coefficients
Corrosion
Corrosion inhibitors
Corrosion prevention
Corrosion rate
Electrochemical impedance spectroscopy
Electrochemistry
Environmental regulations
Laboratory tests
Maintenance management
Mathematical models
Model testing
Multilayer perceptrons
Neural networks
Regularization
Schedules
Seawater
Spectroscopy
Water analysis
Weight loss
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abstract Corrosion inhibitor is often opted as a corrosion protection method for various industries worldwide. The development of eco-friendly corrosion inhibitor has become a trending concern due to the various environmental regulations impose by several countries. However, a laboratory testing would be such a tedious, costly and time-consuming process. Therefore, artificial neural network (ANN) has been used extensively to predict the verdict based on the experimental values. In this study, 3-layered Multilayer Perceptron (MLP) models were developed with 3 inputs (Electrochemical Impedance Spectroscopy, Ω.cm 2 ), (Potentiodynamic polarization, A/cm 2 ), (weight loss, %), and one output (corrosion rate, mm.yr −1 ). The data were divided into three parts; 70%, 15%, and 15% for model development, model validation and model testing, respectively. Three training algorithms were tested during the model development, including the Levenberg-Marquadt (LM), Bayesian Regularization (BR), and Scale Conjugate Gradient (SCG). Results revealed that the best MLP models during the development were using neuron number 4 (r = 0.99272), 6 (r = 0.99155), and 2 (r = 0.98624) for LM, BR and SCG, respectively. Among the three training algorithms, LM is opted as the best training algorithm for the corrosion rate prediction which executed high correlation coefficient (R) values during development (R = 0.99272), validation (R = 0.99905), and testing (R = 0.97908). These findings will be an essential tool for the model development with the sole purposes of predicting the corrosion rate in line to ensure the exact time for repair and maintenance schedule.
ISSN:1755-1307
1755-1315
DOI:10.1088/1755-1315/646/1/012058