Improving Accelerated Corrosion Test Predictions using Linear Polarization Resistance Sensors With Neural Networks

This work addresses the interpretation of accelerated corrosion data with specific application towards the implementation of a successful Condition Based Maintenance (CBM) program. The objectives of this work are twofold, to investigate the efficacy of specific Linear Polarization Resistance sensors, and to develop a greater understanding of interpretation of accelerated corrosion testing. Linear Polarization Resistance sensors are used to retrieve corrosion data of 1100 Aluminum and 1010 steel using accelerated corrosion tests. The accelerated tests are performed for which the times of exposure are 48 hours and 96 hours. Data from Linear Polarization Resistance sensors is compared with physical observations of metal corrosion. Preliminary tests indicate that the LPR sensor resistance varies appropriately with the corrosion rating and humidity. Quantitative standards are achieved using the ASTM G34 corrosion rating. An Artificial Neural Network (ANN) algorithm is developed in MATLAB to test the consistency and efficacy of the LPR sensors. The Neural Network proposed for this study uses twelve inputs, one hidden layer consisting of 32 neurons, and two outputs; it adapts it weight functions using the back propagation algorithm. The output parameters are the ASTM G34 corrosion rating and material loss rate. The network provides results within 13% of experimental values for corrosion rate and accurately predicts the corrosion rating. The corrosion rate predicted by LPR data is higher than experimental results as well as the ANN values, however the difference can be accounted for by the inherent assumptions in the electrochemical methods adopted.