| dc.description.abstract | The determination of petroleum product corrosiveness based on the ASTM D-130 standard is traditionally carried out through visual observation of the color change on copper strips after immersion in the sample. However, this process has significant limitations, such as subjective judgment and potential misclassification, which can lead to incorrect quality assessments. This study aims to develop an automatic classification system using a Convolutional Neural Network (CNN) to identify the level of corrosiveness based on images of corroded copper strips. The methodology includes collecting images of copper strips from laboratory corrosion tests, preprocessing the images through resizing and data augmentation, training a CNN model, and integrating the model into an Android application. Model optimization was performed through hyperparameter tuning, including the number of epochs, convolution blocks, and filters, to achieve the best performance. The best model was selected based on the lowest validation loss using early stopping. The trained model was then converted into TensorFlow Lite format to ensure efficient deployment on mobile devices. The evaluation results show that the developed CNN model successfully classifies 13 levels of corrosiveness with an accuracy of 98.46%, an average precision of 99.45%, an average recall of 99.28%, and an average F1-score of 99.35%. Its implementation in an Android application enables classification via camera or gallery input, offering a more objective, accurate, and practical alternative to manual evaluation methods. | en_US |
