一般化線形混合モデルを用いた逆問題による損傷同定の高精度化　（電気ポテンシャル法を用いたCFRP積層構造のはく離同定への適用）

摘要

本稿は，線型混合モデ′レを用いた逆問題による損傷rn同定問題の高精度化に関する物である．rn損傷や異常の同定における逆問題解析は推定誤差最rn小化の最適化問題であり，応答曲面, ニューラルネrnットワーク, 局所フレキシビリティ法や遺伝的アrnルゴリズムなどの様々な最適化手法を用いた手法や，rnマハラノビス距離等による判別分析, サポートベクrnターマシンや空間統学などさまざまな統計処理rn法を用いた手法が検討されている．%This paper is about improvement of diagnostic accuracy of the damage identification applying the generalized linear mixed model to the inverse problem. Generalized linear mixed model (GLMM) is the extend method of the linear regression analysis include the random effect. By the method, relation between the dependent variable and the independent variables are divided to fixed effect and random effect. The fixed effect means true relation and the random effect means the unknown fluctuation because of the individual specificity. When the damage diagnosis conducted by the inverse problem, this unknown fluctuation is caused by not only the difference of the lot of test piece, but also the unknown parameters of the damage. For example, for the damage size identification problem, location of the damage is the unknown parameter. This unknown parameter is unknown at the time of damage identification but this unknown parameter is known at the time of constructing the model of the inverse problem and the time of deciding the parameter for the inverse problem model. Then in this research, improvement of diagnostic accuracy of the damage identification using inverse problem via the GLMM is conducted. The method is applied to the delamination identification via electric potential method of CFRP laminate. FEM analyses are conducted to obtain electric potential changes due to delamination crack creations with seven-electrode type specimens. By comparisons of the estimations without the random effect and with the effect, a better diagnostic tool is discussed in detail. As a result, GLMM improve the diagnostic accuracy of size and location identification of the delamination crack.