Aplicación de redes neuronales en operaciones industriales de manufactura para productos prismáticos circulares en materiales metálicos no ferrosos

Abstract

Currently in Ecuador, the market has welcomed the machining of various metallic materials, especially non-ferrous ones such as aluminum alloys with heat treatment of the 6xxx and 7xxx families, to make circular primary products. The present research aims to study the surface integrity and the tabulation of the flow of removed material applying unsupervised learning (association and grouping rules) and supervised (artificial neural network) in the manufacturing process on a CNC lathe to machine the aluminum shafts. Two constant cutting parameters were used, such as the cutting speed of 420 m / min and the volume of material removed of 22.2 cm3. Before the machining process, simulations were made using computer-aided design and manufacturing software to make a comparative analysis with real time in the processing of each test, determining the correlation of the census data. In the development of unsupervised learning, the correlation of the cutting parameters was studied and the clustering algorithm was designed based on the maximum value of the Elbow analysis, concluding that to study the results of the aluminum alloy AA 7075 T6, it is necessary an orthogonal arrangement of twenty-seven levels. For the neuronal analysis, the results of the surface roughness were classified using the Likert scale of five levels (low, regular, good, very good and excellent) and in the structure of the neuron it was formulated according to the depth, advance and speed of court; presenting 75% efficiency in the architecture of the cutting feed and depth in the AA 6061 T6 and AA 7075 T6 material table and 100% efficiency in the architecture of the cutting feed and depth in the AA 6061 T6 material table.