Neuro-fuzzy Modeling of Temperature- and Strain-rate-dependent Behavior of NiTi Shape Memory Alloys for Seismic Applications

This paper proposes a neuro-fuzzy model of superelastic NiTi shape memory alloy (SMA) wires for use in seismic applications. First, in order to collect experimental data, uniaxial tensile tests are conducted on superelastic wires in the temperature range of 0 oC to 40 oC, and at the loading frequencies of 0.05 Hz to 2 Hz with five different strain amplitudes. Then, an adaptive neuro-fuzzy inference system (ANFIS) is employed to construct a model of SMAs based on experimental input-output data pairs. The fuzzy model employs strain, strain-rate, and temperature as input variables, and provides stress as single output. Gaussian membership functions (MFs) are assigned to each input variables. A total of twelve if-then rules are used to map these MFs to output characteristic. The model obtained from ANFIS training is validated by using an experimental data set that is not used during training. The developed model is capable of simulating behavior of superelastic SMAs at various temperatures and at various loading rates while it remains simple enough to realize numerical simulations. These features of the model make it attractive for numerical studies on vibration control of structures.