2007 Volume 47 Issue 8 Pages 1126-1130
Hot flow stress of CrMoV steel was extracted using hot compression test. Test was carried out in strain range of 0.1–0.9, with strain rate of 0.1–5 s−1 and the temperature of 900–1200°C. Flow stress was predicted by using conventional regression method, i.e., Zenner–Holloman parameter with hyperbolic function. The results showed low accuracy of prediction with a high relative error. Several two hidden layers Artificial Neural Networks (ANN) architectures with back propagation algorithm and momentum learning process were applied using Matlab software. They used strain, strain rate and temperature as input and flow stress as output. It was found that an optimum architecture of 3–9–10–1 shows proper prediction with respect to the conventional regression method, i.e., the relative error reached −0.13% in place of 11.52%. This ANN method is also capable of generating high precision output for unseen deformation conditions if proper initial weights and biases are used.