Experimental research and ANN modeling on the impact of the ball burnishing process on the mechanical properties of 5083 Al-Mg material

Creative Commons License

Basak H., ÖZKAN M. T. , Toktas I.

KOVOVE MATERIALY-METALLIC MATERIALS, vol.57, no.1, pp.61-74, 2019 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 57 Issue: 1
  • Publication Date: 2019
  • Doi Number: 10.4149/km_2019_1_61
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Page Numbers: pp.61-74
  • Keywords: burnishing, surface roughness and hardness, microhardness, strength analysis, Artificial Neural Networks (ANN), ARTIFICIAL NEURAL-NETWORK, SURFACE-ROUGHNESS, ROLLER, PARAMETERS, STEEL, OPTIMIZATION, IMPROVEMENTS, MICROHARDNESS, HARDNESS, QUALITY
  • Gazi University Affiliated: Yes


5083 Al-Mg is the widely used material in food, chemistry, vehicle, machinery, and construction sectors, as well as in the aviation and space industries. The burnishing is normally used as the finishing operation for this material with the advantages such as surface roughness, reduced fracture formation, hardness, fatigue strength, and an increase of the wear resistance. These positive improvements are dependent on burnishing process parameters such as feed rate, burnishing force, ball diameter, and a number of revolutions. The study contains determination and optimization of the machining parameters and their effects on the surface roughness, microhardness, and the strength of 5083 Al-Mg material in the ball burnishing processes. Multiple regression and ANOVA analysis were performed to identify significant process parameters. A new Artificial Neural Networks (ANN) model with different neuron structures and algorithms has also been developed using experimental results to supplement the multiple regression model as the desired R-2 values could not be achieved with the latter. The ANOVA analysis indicated that both the burnishing force and the number of revolutions have a significant effect on the surface roughness and hardness with optimums 300 N and 200 rpm, respectively. Results from the two models were compared with each other. The developed ANN model is shown to estimate the surface roughness and the surface hardness with high reliability (R-2 = 0.999992) without costly experimental trials.