Application of Machine Learning for Data with an Atmospheric Corrosion Monitoring Sensor Based on Strain Measurements

Abstract

Machine learning methods were applied to data with an atmospheric corrosion monitoring sensor based on strain measurements to improve the evaluation accuracy of the thickness reduction of a low-carbon steel plate due to atmospheric corrosion. Monitoring data used in this study were taken in a previous study using active–dummy strain gauges for corrosion product experiments. Values measured by the gauges before inducing corrosion via saltwater treatment of the test piece and reference data of the thickness reduction in a reference test piece were used for training data. By using the trained machine learning methods, the errors for the outputs of the machine learning models were smaller than those for the evaluation in monitoring data of our previous study.