Analysis and prediction of the tensile strength of aluminum alloy composite using statistical and artificial neural network technique

Abstract

In this research work, aluminum alloy (Al-20Fe-5Cr) matrix-based aluminum oxide (Al2O3) reinforced composites were developed through the powder metallurgy (P/M) process. Effect of compaction pressure (200, 250 & 300 MPa) and wt.% of Al2O3 (0, 10, 20 & 30 wt.) on tensile strength and percentage elongation has been analyzed through statistical and artificial neural network techniques (ANN). The mixture of Al-alloy powder particles and Al2O3 particles were synthesized in a centrifugal ball mill for 20 min. Compaction of synthesized powder was carried in the standard tensile die using a uniaxial hydraulic pressing machine. Sintering was performed at temperature 580 ± 20 °C for one hour in an argon gas environment using an electric tubular furnace. It was found that tensile strength enhanced significantly with the addition of Al2O3up to 20 wt.% and then declined sharply for the 30 wt.% of Al2O3at all compaction pressures. The highest tensile strengths were found for each wt.% of Al2O3 at compaction pressure 300 MPa compare to other compaction pressures. Tensile strength increased from 105 to 158 MPa with the addition of 20 wt.% Al2O3 and decreased to 142 MPa for 30 wt.% at 300 MPa compaction pressure. The improvement resulted from better compaction, leading to more plastic deformation, better packing, and high effective contact area. However, the percentage of elongation decreased from 23.2% to 2.2% with an increment of wt.% of Al2O3 for compaction pressure 200 MPa, while for 300 MPa, its value drops from 25.8% to 6.5%. This depreciation can be reasoned for the reduction in ductile matrix content and dilute flowability of the Al matrix, which occurred due to brittle Al2O3. The statistical analysis using ANOVA revealed that the compaction pressure is the primary control factor influencing tensile strength by 90.3%. The feed-forward network with a back-propagating gradient-descent error minimization training approach and mean squared error (MSE) as performance function was employed to model and predict tensile strength. The developed 3-layered multilayer perceptron (MLP) with 2–10–2 network architecture established a correlation between the inputs and outputs with minimum error (MSE) below 1% and maximum correlation coefficient (R) close to 1.