ニューラルネットワークを用いた物体内発熱源の同定

概要

論文の詳細を見る
A non-iterative method for the identification of the heat source in a solid body by a neural network approach is proposed. In this method, the location of the heat source is identified by the neural network through pattern matching with the surface temperature distribution. Then the intensity of the heat source is determined by a boundary integral method utilizing the information of heat-source location predetermined by the neural network and introducing an appropriate harmonic function. Without measuring the surface heat flux, which was required in the previous boundary integral method, this combined approach enables us to estimate both the location and the intensity of heat source more rapidly than by ordinary iterative optimization methods. The simulated results show that the proposed method is effective and has the potential to deal with noisy measurement data. Moreover, it is shown that the robustness of the method is improved by training the neural network using noisy learning data.
1993-11-25