Improved neural models for roughness in honing processes

In the present work improved neural network models for average roughness Ra in rough honing processes are studied. Four different adaptive models were tested, which integrate previously obtained direct and indirect models. Such models allow defining values for process variables from required average roughness Ra values. A control parameter d is employed for determining the error of the model, and a sensitivity parameter m measures the convergence speed of the models. Models were tested for m=1, m=10, m=100 and m=1000. Best model was selected having lowest relative error between experimental and simulated values. Peer Reviewed