REVISTA DYNA NEW TECHNOLOGIES

Key words:

machine learning, convolutional neural network, image processing, wire plastic deformation processes, CNN, aprendizaje automático, red neuronal convolucional, procesamiento de imágenes, procesos de deformación plástica del alambre

Article type:

ARTICULO DE INVESTIGACION / RESEARCH ARTICLE

Section:

RESEARCH ARTICLES

ABSTRACT:
Machine learning, particularly the convolutional neural network (CNN), is a potentially competent tool for image processing. In this work, the technique is ?rst utilized to perform an analysis of the di?erent wire plastic deformation processes. In particular, a CNN is established and trained using 3200 image fractions with a resolution of 80 × 80. The relevant architecture consists of three convolutional layers in conjunction with polling layers with relu activation. By properly tuning the network, we achieve good training and validation accuracies of 97.7% and 97.1% to identify between two underlying treatments by observing only an insigni?cant cropped fraction of the material’s cross-sectional pro?le. We argue that speci?c features of the architecture, such as the augmentation process’s rescaling parameter, are essential in guaranteeing a satisfactory accuracy rate. The possible implications of the present study are also addressed.

Keywords: machine learning, convolutional neural network, image processing, wire plastic deformation processes