Wednesday  |  Monongahela  |  11:30 AM–11:50 AM

#13840, Artificial Neural Network Models for the Cyclic Response of Shape Memory Alloys

The unique ability of shape memory alloys (SMAs) to remember a “set-shape” and recover large deformations makes them attractive for the design of innovative devices. In the aerospace industry, SMAs are used to design actuation components that are typically subjected to several thermal cycles. For such applications, it is important to assess the evolutionary behavior, i.e., changes in the stress-strain-temperature responses with cycles, before considering the SMA for a design component. For instance, it is critical to know the number of cycles needed to place the material in a state of near stability, where almost the same strain magnitude will be measured from cycle to cycle[1]. The appropriate number of thermal cycles needed to reach stability in a selected material is influenced by the stress magnitude and the temperature range for cycling. Mostly, such stress-strain-temperature characterization has been obtained through experiments that may be supplemented with predictive constitutive models. It is sometimes challenging to determine and directly link all the constitutive parameters, and their numerical values to specific physical observations in the material. It may be important to study the material characteristics at several cycles and test conditions.

The objective of this research is to develop and use Artificial Neural Network (ANN) models to predict the cyclic, non-linear shape memory behavior of a commercially available SMA, i.e., the binary Ni49.9 Ti50.1 (at. %) alloy, also known as 55NiTi (55 wt% Ni). ANNs are information processing systems with adaptive components that provide simpler, yet practical models that can characterize and predict complex responses. The novelty of this approach is that only the predominant factors, i.e., the stress magnitude, number of thermal cycles, and temperature state will serve as the input variables in the model, thus given a simple, yet practical approach to predicting the strain responses. The ANN model will be tested on stress magnitudes ranging between 50 and 300 MPa, temperature magnitudes between 30 and 165°C, and thermal cycles between 50 and 200 cycles. Predictions will be compared to available experimental data.

Reference:
[1] Padula II, S. A., Gaydosh, D., Saleeb, A. F., et al. (2014) Transients and Evolution in NiTi. Experimental Mechanics 54(5): 709-715.

Josiah Owusu-Danquah   Cleveland State University
Abdallah Bseiso   Cleveland State University