Application of Fuzzy Pattern Recognition of Seismic Damage to Concrete Structures

There has been a long-standing desire to identify classes or groupings of buildings that exhibit similar characteristics in response to seismic events. Regional earthquake loss models, post-earthquake safety evaluations, and rapid visual screening of buildings for potential seismic hazards all utilize building damage patterns in some form. Difficulty arises in characterizing seismic building damage due to the inherent uncertainty in both the occurrence of seismic events and uncertainties related to building responses. Various trends in building performance, often based on building characteristics, are frequently listed in the literature. For example, older structures are expected to perform less well than those designed to more recent design codes; however, it remains difficult to characterize building damage due to the rarity and somewhat limited documentation of earthquake building damage. The earthquake engineering field, in its current state, is dominated by probabilistic and/or deterministic frameworks to approach this problem. Although both frameworks have yielded significant contributions to building damage assessment, they are limited in their ability to relate engineering parameters to physical building damage observations that are often expressed linguistically. Other theories under the umbrella of Generalized Information Theory offer different mathematical models and philosophical approaches to this problem. This research utilizes mathematical formulations different from probability theory for incorporating uncertain information in the context of building damage assessment. This research aims to investigate the following question: Does empirical evidence support suggested linguistic general trends of building damage from seismic events, and if so, can this empirically-based information be used to associate potentially seismic hazardous conditions to buildings in understandable terms by end users? To investigate these questions, tools within the domain of fuzzy classification and fuzzy pattern recognition are used with post-earthquake building damage data to determine the existence or nonexistence of building damage patterns. Data from the 1994 Northridge earthquake for concrete structures have been used to mathematically model linguistic building damage patterns based on important building features. A comparison of the damage patterns developed by the classification analysis included herein to linguistic descriptions listed in the literature is also investigated. Although the linguistic patterns are limited by the available data, the potential for systematically representing and processing of subjective assessments of building damage in linguistic form offers a powerful framework for modeling nonrandom uncertainties that are intrinsically different from probability theory. (C) 2015 American Society of Civil Engineers.