ニューラルネットワークを用いた小口径トンネルロボットの方向修正量モデル形成法

概要

論文の詳細を見る
This paper describes the modeling method of the amount of directional correction for a small tunnelling robot. We have already proposed the linear time series and the neural modeling methods. The neural network has a 4-layer construction with parameters for soil hardness for application to various soil hardnesses. In this study, we use a 3-layer neural network without parameters for soil hardness to form the model for the amount of directional correction which can be applied to various soil hafdnesses. The input of the neural network is the pitching and yawing angle difference and the head angle of pitching and yawing directions. The output of the neural network is the amount of pitching and yawing directional correction. This neural network learns from errors between experimental data and output of the neural network. We investigated a comparison of the linear time series model and the proposed method, the modeling of the amount of both pitching and yawing directional correction, and the relationship between model order and modeling error. These investigations and modeling results showed the validity of this method.
社団法人日本機械学会の論文
1992-11-25