Pre-earthquake fuzzy logic and neural network based rapid visual screening of buildings

Abstract

When assessing buildings that may collapse during a large earthquake, conventional rapid visual screening procedures generally provide good results when identifying buildings for further investigation. Unfortunately, their accuracy at identify buildings at risk is not so good. In addition, there appears to be little room for improvement. This paper investigates an alternative screening procedure based on fuzzy logic and artificial neural networks. Two databases of buildings damaged during the Athens earthquake of 1999 are used for training purposes. Extremely good results are obtained from one database and not so good results are obtained from the second database. This finding illustrates the importance of specifically collecting data tailored to the requirements of the fuzzy logic based rapid visual screening procedure. In general, results demonstrate that the trained fuzzy logic based rapid visual screening procedure represents a marked improvement when identifying buildings at risk. In particular, when smaller percentages of the buildings with high damage scores are extracted for further investigation, the proposed fuzzy screening procedure becomes more efficient. This paper shows that the proposed procedure has a significant optimisation potential, is worth pursuing and, to this end, a strategy that outlines the future development of the fuzzy logic based rapid visual screening procedure is proposed.