ニューラルネットワークによるネジ締めの軸力推定

抄録

It is important to realize the suitable tightening of bolts by an impact wrench. To achieve the goal, we must know the clamping force of a bolt. But it is difficult to measure the force directly. In this paper, we estimate the clamping force by neural networks. The strength of tightening depends on the kind of works. At first, we try to classify the kind of works with a neural network based on the error back-propagation method. Then we use an angle control method for tightening. But this method requires the initial tightening snug fit. We will estimate the snug fit, then the clamping force. Finally, the control input for the desired tightening is determined based on the estimated clamping force.