International Journal on Advanced Science, Engineering and Information Technology, Vol. 8 (2018) No. 6, DOI:10.18517/ijaseit.8.6.7096

Damage Level Prediction of Pier using Neuro-Genetic Hybrid

Reni Suryanita, Mardiyono Mardiyono, Harnedi Maizir

Abstract

Generally, long span bridges have multiple columns as known as piers to support the stability of the bridge. The pier is the most vulnerable part of the deck against the earthquake load. The study aims to predict the performance of the pier on the bridge structure subject to earthquake loads using a Neuro-Genetic Hybrid. The mix design of the Back Propagation Neural Networks (BPNN) and Genetic Algorithm (GA) method obtained the optimum-weight factors to predict the damage level of a pier. The input of Neuro-Genetic hybrid consists of 17750 acceleration-data of bridge responses. The outputs are the bridge-damage levels based on FEMA 356. The categorize of a damage level was divided into four performance levels of the structure such as safe, immediate occupancy, life safety, and collapse prevention. Bridge responses and performances have resulted through analysis of Nonlinear Time History. The best of Mean Squared Error and Regression value for the Neuro-Genetic hybrids method are 0.0041 and 0.9496 respectively at 50000 epochs for the testing process.  The Regression value denotes the predicted damage values more than 90% closer to the actual damage values. Thus, the damage level prediction of the pier in this study offers as an alternative to structural control and monitor of bridges.
 

Keywords:

Acceleration; Damage level; Neuro Genetic; Mean Squared Error; Regression

Viewed: 125 times (since Sept 4, 2017)

cite this paper