本文旨在建立一程式溝通解析模型與實驗數據，並剔除可能有錯誤之量測模態。對於大型結構體的動態分析，都先經由有限元素等之分析方法建立所謂的解析模型（Analytical Model），後輔以振動實驗測試。解析值與實驗數據之間的驗證存在著差異。因此解析模型必須加以修正，使之更正確地代表實際構體。首先吾人利用實驗所量測到模態的部分點數值去估算未量測點的數值，以得到完整的模態，然後由特微方程與模態之正交性檢驗其正確性，並挑出可能有錯誤之模態，以免因錯誤之實驗值反而使解析值偏離。本文提供一種修正慣性矩陣之方法與兩種勁度矩陣法，並探討修正前後的改變率。從結果顯示：當測試數據中含有錯誤數值時，若不將錯誤之模態剔除，則修正後之模型與解析模型差異甚大，而偏離實際系統。在矩陣修正方面，慣性矩陣與使用方法一所得之勁度矩陣完全滿足了其特微方程式，但是原有矩陣之帶狀（banded）特性已遭破壞；而方法二所得勁度矩陣保留了原有矩陣之帶狀特性，但是卻不完全滿足原特微方程式，且運算所需之記憶體空間較大，運算時間也較長。

The purpose of this research is to establish a computer code as an interface between the analytical model and the experimental data for dynamic systems. The analytical model are updated according to the provided experimental data. The programs also provide the ability of picking out possible incorrect modes in measurement. In addition, two aspects are considered in developing the computer code. One is that in reality the experimental data are only the partial points of the system, and the other is that errors might exist in the experimental data. If parts of the experimental data are incorrect and not eliminated from the modification, the updated analytical model will deviate from the physical one. Both of the inertia matrix and the stiffness matrix are modified in this study. In modifing the stiffness matrix, two algorithms are developed, the first one provides an effective and quick way, howeyer, completely distorts the original banded feature. The second method preserves the banded feature in matrix, but the characteristic equation is not always satisfied. Moreover, the second method takes much more computing time and memory space than the first one.

none