Neuro-fuzzy techniques for the classification of earthquake damages in buildings P.F. Alvanitopoulos a , I. Andreadis a, * , A. Elenas b a Laboratory of Electronics, Section of Electronics and Information Systems Technology, Department of Electrical and Computer Engineering, Democritus University of Thrace, GR – 67100 Xanthi, Greece b Institute of Structural Mechanics and Earthquake Engineering, Department of Civil Engineering, Democritus University of Thrace, GR – 67100 Xanthi, Greece article info Article history: Received 2 June 2008 Received in revised form 22 December 2009 Accepted 23 February 2010 Available online 1 March 2010 Keywords: Pattern classification Neuro-fuzzy systems Damage indices Seismic parameters Seismic damages abstract The identification of damages produced by severe earthquakes on constructions is impor- tant for several reasons such as public safety, economical recourses management, infra- structure and urban planning. After the manifestation of an earthquake, engineers have to evaluate the safety of existing structures and decide the actions to be taken. In this study two techniques are proposed for automatic damage classification in buildings. The inherent information contained in accelerograms is described by 20 seismic parameters. Two classification models of earthquake damages based on artificial neural networks and neuro-fuzzy systems were designed. Furthermore, they were tested for their effectiveness to classify structural, architectural, mechanical–electrical-plumbing and contents damages. The proposed systems were trained and tested with three reinforced concrete frame struc- tures. Results show correct classification rates up to 98%. According to these classification rates these techniques are proven a suitable tool for classification of earthquake damages in structures. Ó 2010 Elsevier Ltd. All rights reserved. 1. Introduction In earthquake engineering the knowledge on earth- quake damages in human constructions is of the utmost importance. The destructiveness of a seismic excitation in structures includes structural, architectural, mechanical, electrical, plumbing and contents damages. In an attempt to quantify the catastrophic results of seismic excitations on buildings, different damage indicators have been used for each type of damages. The aim of this paper is to clas- sify the aforementioned damages of seismic signals in dif- ferent reinforced concrete structures. Previous work [1], proposed classification methods con- cerning only structural and architectural damages in struc- tures. The first approach of the research was based on the shape similarity of accelerograms. However, due to the random shape of seismic signals this approach led to poor classification rates up to 38%. The second approach was based on seismic parameters which can express the dam- age potential of an earthquake [2]. The graphical represen- tation of seismic parameters is used to represent the seismic excitation instead of the more complicated accel- erograms themselves. The extracted classification results were better than the results of the first approach. Correct classification rates are in this case 79.5% for the structural and 77% for the architectural damage indicators. Neverthe- less, it was desirable to improve the results further. This has been achieved by the second work [3] that provided the best results. Fuzzy logic concepts have been incorpo- rated (fuzzyfication of the seismic parameters) and correct classification results up to 84% and 82% for the structural and architectural damages were recorded, respectively. Some preliminary results based on an artificial neural net- work classifier have been reported in [4]. In this paper two new damage classification methods based on seismic parameters are proposed. The key point of the first method is that an artificial neural network has 0263-2241/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.measurement.2010.02.011 * Corresponding author. E-mail address: iandread@ee.duth.gr (I. Andreadis). Measurement 43 (2010) 797–809 Contents lists available at ScienceDirect Measurement journal homepage: www.elsevier.com/locate/measurement