---

In this study a feed forward back propagation artificial neural network (ANN) model was established to predict Vickers microhardness in aluminum-alumina nanocomposites which have been synthesized by mechanical alloying and hot pressing. Volume percent of reinforcement, size of nanoparticles, force in microhardness test; and mechanical alloying parameters, such as time, ball to powder ratio (BPR) and speed of ball mill were used as the inputs and Vickers microhardness as the output of the model. Effective parameters in training such as learning rate, hidden layers and number of neurons, were determined by trail and error due to amount and percentage of errors. Regression analysis in train, validation and test stages; and mean squared error were used to verify the performance of neural network. Average error of predicted results was 2.67% or 2.25 Vickers. Also mean squared error for validation data was 7.76. As can be expected, ANN methods reduce the expenses of experimental investigations, by predicting the optimum parameters.

---

Nanostructured (Al-8wt%Zn-3wt%Mg) alloy was synthesized by simultaneous fracture and cold welding mechanisms in mechanical alloying with initial elemental powders and subsequently, this alloy was applied as matrix to fabricate Al/Short glass fiber nanocomposite in 1, 3 and 5 percent of glass fibers. The resulting powders were consolidated under 400 MPa at 380 °C in cylindrical die to produce consolidated nanocomposites. Relative density of samples reduced with increasing the percentage of glass fibers and this trend was more intense from 3 to 5 percent. Also compressive strength and hardness were investigated for these samples in different percent of glass fiber. The results showed that strength and hardness were enhanced with increasing glass fiber but decrease of mechanical properties was observed in 5 percent due to reducing in relative density. Compressive strength was compared between nanocomposite with pure Al and Al alloy matrix, and results show more reinforcement in Al based sample.