Analysis of cold compaction for Fe-C, Fe-C-Cu powder design based on constitutive relation and artificial neural networks

• Published in: Powder technology 2019-07, Vol.353, p.330-344
• Main Authors: Shin, Da Seul, Lee, Chi Hun, Kim, Suk Hyun, Park, Dong Yong, Oh, Joo Won, Gal, Chang Woo, Koo, Jin Mo, Park, Seong Jin, Lee, Seung Chul
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Artificial neural networks (ANN); Cold compaction; Compaction; Computer simulation; Constitutive relationships; Copper; Design; Design parameters; Fe-C; Fe-C-Cu; Informatics; Iron; Leave one out cross validation (LOOCV); Mathematical models; Neural networks; Particle size; Particle size distribution; Powder; Properties (attributes); Size distribution; Volumetric strain
• ISSN: 0032-5910; 1873-328X
• DOI: 10.1016/j.powtec.2019.05.042

The constitutive relations for Fe-C and Fe-C-Cu powder compactions were investigated with the three consitituents: i) powder design parameters, ii) material related properties, and iii) final compaction properties. With the concept of materials informatics, this approach enables to predict the final compaction properties depending on the material conditions. The correlations between powder design parameters (particle size, graphite content, lubricant content, particle size distribution, copper content) and material related properties (ρTap, γ, a, b, n) in Shima-Oyane model were characterized by the compaction experiments and artificial neural network (ANN) model. The ANN model was developed to predict the effect of powder design parameters on the material related properties. The average mean absolute percentage error of predicted material related properties was 2.194%. The final properties (green density, density gradient, effective stress, hydrostatic stress, effective strain, volumetric strain) were calculated by the compaction simulation based on the experimental and predicted material related properties. [Display omitted] •Compaction behaviors of Fe-C and Fe-C-Cu were characterized with Shima-Oyane model.•A new artificial neural network model was developed for cold compaction.•Leave one out cross validation and hyper parameter tuning approaches were applied.•Sensitivities were calculated among powder design parameters and final properties.•Integrated system was developed to predict the powder compaction behavior.