Post-disaster damage assessment of structures by neural networks

Recent papers have investigated the contribution of structural and secondary elements in the overall damage of structures due to seismic effects. The present paper improves such methods by investigating also the marginal contribution of the geotechnical disorders and geometric regularity, in addition to the combined effect of structural and secondary elements. An adapted artificial neural networks (ANNs) method is proposed for this purpose. In this approach, three groups of parameters are considered for the quantitative evaluation of post- earthquake damage of structures: the structural group, the secondary group and a qualitatively evaluated group consisting observed geotechnical disorders and building regularity. Principal-component analysis is used in order to evaluate the effects of each input variable on the global structural damage. The ANN model is trained and validated for a collected database corresponding to 27,601 of buildings (Boumerdes earthquake, Algeria: M=6.8; May 21, 2003) and tested for 1,000 damaged buildings, located near the hypocentral zone (Bordj-El Bahri city, near Algiers), inspected during a post-quake damage survey. The assessment of the overall damage of structures based on the whole combination of three groups indicates that the developed model provides more accurate theoretical global damage predictions (98% accordance) that render it a promising tool for the inspector to decide about the final damage category.