Online Sequential Extreme Learning Machine for Vibration-Based Damage Assessment Using Transmissibility Data

Abstract

Traditional vibration-based damage assessment approaches include the use of feed-forward neural networks. However, the slow learning speed of these networks and the large number of parameters that need to be tuned have been a major bottleneck in their application. This paper proposes to use an emergent learning algorithm called the online sequential extreme learning machine (OS-ELM) algorithm. This algorithm provides good generalization at fast learning speeds, allows data to be learned one by one or block by block, and the only parameter that needs to be tuned is the number of hidden nodes. A single-hidden-layer network is trained to detect, locate, and quantify structural damage using data derived from transmissibility measurements. Two experimental cases are presented to illustrate the approach: an eight-degree-of-freedom (DOF) mass-spring system and a beam under multiple damage scenarios. To demonstrate the potential of the proposed algorithm over existing ones, the obtained results are compared with those of a model updating approach based on parallel genetic algorithms.