6-5-1 Neural network results
6-5-1 Neural Network results
- The third NN configuration had structure 6-3-1:
- input layer with 6 nodes for ε, ε’, T, the logarithmic functions of ε and ε’, and the inverse function of T
- hidden layer had 5 nodes
- output layer had 1 node for flow stress s prediction
Desired and predicted flow stress vs. strain for test 28
To take into account the analytical relationship among the considered process parameters, 6-component input vectors including the logarithmic functions of ε and ε' and the inverse function of T, 1/T, were used for training and testing the 6-3-1 NN. Desired flow stress and predicted flow stress were plotted vs. strain. At least on a qualitative basis, the work hardening and the dynamic recrystallisation regions are reproduced by the predicted curve. The addition of the logarithmic functions of strain and strain-rate and the inverse function of temperature appears to provide the NN model with information critical for material behaviour modelling, at least on a qualitative basis. The 6-5-1 NN seems able to model the work hardening and work softening material behaviours, although the curve offset is still high
Who's online
There are currently 0 users and 280 guests online.