Juang 9612116 This research is aimed at improving the engineer's ability to predict earthquake-induced liquefaction hazards. (Liquefaction is a serious form of soil failure resulting from strong-ground shaking, typically associated with major earthquakes, such as the 17 January 1995 earthquake in Kobe, Japan.) It employs the fuzzy neural network methodology, an emerging technology which can learn and adapt structured and unstructured knowledge. Historical data on field liquefaction performance, in terms of causes and effects, is re-cataloged in a consistent pattern usable by fuzzy neural networks. Uncertainty associated with field performance data and the interpretation of liquefaction hazards is handled through the use of fuzzy sets, which allows for approximate and continuous classification. Membership functions of fuzzy soil and geological parameters are established through consultation with experts in soil failure and fuzzy neural network methodology. The collected data and expert opinions, presented in a form usable by fuzzy neural networks, is then used to generate trained neuro-fuzzy systems for predicting liquefaction hazards. It is expected that this research will result in a new design methodology and procedure to predict earthquake- induced liquefaction hazards using neuro-fuzzy systems.
