基因演算法應用於衝擊性撓度儀回算之研究

Backcalculation For FWD Deflections of Pavements Using Genetic Algorithm

江克培(Ke-Pei Jiang)

指導教授 ： 張德文

淡江大學/工學院/土木工程學系碩士班/碩士(2005年)

https://doi.org/10.6846/TKU.2005.00118

摘要

傳統的試驗數據回算一般多以靜態方式完成，但由實務及研究經驗顯示，靜態回算無法有效地模擬動態試驗，更無法穩定的處理觀測數據，致使撓度法於鋪面工程中之運用常止於觀摩與比較，肇於此，本研究使用格林柔度方程式解模擬鋪面動態應力波行為，並結合基因遺傳演算法發展動態回算分析，以擴充撓度法試驗的應用範圍。研究另蒐集國內落重儀現地試驗，以迭代法、資料庫法、基因遺傳算法和MODULUS回算程式所得與實驗室試驗數據比較，確定各項分析法之可信度。

研究結果顯示，以動態正算程式配合基因遺傳演算法，在理論回算或是現地實驗值的模擬中，動態程式均能較傳統回算程式更有效地掌握現地狀況。本研究針對撓度法中之衝擊試驗-落重撓度儀進行模擬，發展出配合該試驗法的動態回算程式DBFWD-GA，研究中並對基因演算因子進行影響分析，結果指出回算時間會隨族群數量和世代數增加而倍增，當初始族群為140﹝四層結構﹞，突變率為0.1時，其回算結果最佳。其回算彈性模數值和撓度值控制均方根誤差小於2%，對於判斷鋪面結構強弱應有初步成效。動力回算結果雖近似，但與靜力回算和實驗室數據仍有出入，且同一鋪面構造回算結果差異紊雜，其應用仍有必要再探究。

關鍵字

動態回算 ； 基因遺傳演算法 ； 初始族群 ； 突變率

並列摘要

Conventional data interpretation of the deflection test on pavements is mostly accomplished using the static analysis. However according to the practice and the study experiences, the static analysis could not accurately model the dynamic testing and stably handle the predictions. The deflection test needs to be analyzed carefully in order to provide reliable information of the pavements. This study uses the dynamic Green’s functions to model the stress waves propagating along the pavements. The Genetic Algorithms (GA) is incorporated with the program to develop the dynamic backcalculation analysis of the Falling Weight Deflectometer (FWD) test. In-situ deflection data of the FWD test were collected and analyzed. Comparison and validation were carried for dynamic backcalculation programs using the iterative, data bank and this method with the static backcalculation program MODULUS. The laboratory results were also used for comparisons.

Results of this study indicate that the dynamic backcalculation can reflect the pavement structure more truthfully than the static conventional analysis. A dynamic backcalculation program UTFWD-GA is thus suggested to evaluate the pavements for the FWD test. According to the parametric studies, the backcalculation time would increase with the number of populations and generations. The optimized results could be found for the GA method of the four layered flexible pavement structures, where the initialization populations is 140, and mutation is 0.1. The associated RMS errors of the moduli and the deflections were found less than 2%. Although the dynamic backcalculation results were found agreeable with each other, they would be different from static backcalculation and lab data. Nevertheless, the predictions on the same road were varied at different sections. The applications need to be restudied carefully.

並列關鍵字

dynamic backcalculation ； genetic algorithm ； initialization population ； mutation

參考文獻

3.張記恩，「撓度試驗數據處理與路面結構指標之結合運用」，淡江

交通部運輸研究所 (1992)。

Program for Rational Pavement Design , Report FAA-RA-75-

69.Kausel, E. and J. M. Roesset, “Stiffness Materices for

告」，財團法人台灣營建研究中心 (1988)。