Neural network modelling of copper and aluminium compatibility with biodiesel and diesel fuels

This is the authors' abstract. We don't add key points for this paper.

Artificial neural network (ANN)-based predictive model was developed to predict corrosion rate of copper and aluminium in biodiesel-diesel fuel using existing published experimental datasets. By reducing the need for additional experimental designs, this approach reduces waste of resources and accelerates optimisation of biodiesel blend formulations for fuel delivery system applications. Nine input variables were used, including metal type, biodiesel blend, fatty acid composition, total acid number, temperature and immersion time. An 8:8:1 ANN architecture trained with Bayesian regularisation algorithm achieved high predictive accuracy with an overall correlation coefficient (R) of 0.8993 and mean squared error (MSE) values of 0.006814 mpy for training datasets. Sensitivity analysis showed that unsaturated fatty acid methyl esters content, exposure duration, temperature, TAN and metal composition were influential factors affecting corrosion rate of copper and aluminium in biodiesel-diesel blend. By applying machine learning to compiled literature data, this work demonstrates an alternative approach for predicting corrosion rate without relying on lengthy immersion experiments for evaluating corrosion rate of vehicle fuel delivery system. The findings show that ANN can model and predict corrosion rate of metal and serve as a practical and efficient tool to support corrosion assessment and material selection for biodiesel applications.