Previsão da rugosidade superficial de aços endurecidos através de rede neural artificial

Abstract

Surface integrity is a critical concept in the field of mechanical engineering, which refers to the quality and properties of a material's surface after it has undergone a manufacturing process. Therefore, controlling the quality of the material surface is essential since it can significantly affect its performance in aspects such as fatigue strength, wear resistance, corrosion resistance, and mechanical properties. The focus of this study was to develop and train an artificial neural network to predict the surface roughness of hardened steel after cylindrical turning, based on its hardness and machining parameters. Additionally, the behavior of the neural network was studied for different iterations and datasets to optimize the results. Experimental data previously obtained by the research group for O1 tool steel were used to validate the neural network. Finally, it was possible to develop an artificial neural network based on data from various hardened steels found in the literature, resulting in an estimated surface roughness of the O1 tool steel with an approximate error of 20%. Furthermore, it was determined that the best combination of data volume for training and validating the neural network is between 45-60%.