Optimum Rib Layout Design of Machine Tools Structures Using Neural Networks

This paper deals with an optimum design method for machine tools structures using neural networks. Machine tools structures such as a bed or a column have ribs, which layout correlates with their performance. A static and a dynamic stiffnesses, a natural frequency and a thermal displacement are hired as evaluation functions. Some of them are obtained by FEM and learned by the neural networks to recall those of unlearned rib layouts. The evaluation functions of all prospective lib layouts are estimated in short time and the optimum rib layout can be obtained by choosing that with high values. The additional learning method is proposed to decrease the estimation error especially for exclusive data. The estimation error becomes less than 4% and it leads to realize the multi-object optimum design method. Two multi-object optimum designs were carried out using the whole machining center model. The estimation error is approximately 8% and the value of evaluation function is improved as 18%. In the case of layout for many ribs, the margin of improvement decreases because the problem becomes complex. The proposed system is effective for the machine tools design in early or concept design stages.