ABSTRACT
In consideration of above problems in existing models, this paper adopts a novel single-layer feed-forward neural network algorithm to portray the nonlinear and hysteretic behaviours of MRE isolator, which is also called Extreme Learning Machine (ELM). ELM is a simple and effective algorithm with good generalization performance and fast learning speed compared with other algorithms such as Back Propagation (BP) algorithm (Huang et al. 2006). To best of author’s knowledge, there is no application of the ELM to predict the shear forces of MRE isolator with different external excitations. Therefore, the method used in this study is considered noble. In the proposed ELM model, the historical records of device displacement, velocity and applied current are used as the inputs of ELM while the output of ELM is the shear force, which is generated to avoid the unexpected vibration caused by hazard external loadings. The testing data from the dynamic testing
1 INTRODUCTION
Magnetorheological Elastomer (MRE) is a kind of intelligent material which has unique field-dependent properties (Gong et al. 2005; Li et al. 2010; Yu et al. 2015a). MRE isolators are semi-active control devices which have the benefits of simple structure, low power requirement and adaptive modulus ability (Berhooz et al. 2014; Yang et al. 2014). A large amount of applications of MRE isolator have been considered, such as in semi-active automotive suspension system and civil infrastructure (Li et al. 2014). Nevertheless, because of highly nonlinear feature in such a type of device, the characterisation of corresponding nonlinear behaviour is still challenging, which means that it is definitely difficult to accurately forecast the relationship between inputs and outputs of the device (Yu et al. 2015b).
