Predicting actuation strain in quaternary shape memory alloy NiTiHfX using machine learning

Data-driven techniques are used to predict the actuation strain (AS) of NiTiHfX shape memory alloy (SMA). A Machine Learning (ML) approach is used to overcome the high dimensional dependency of NiTiHfX AS on numerous factors, as well as the lack of fully known governing physics. Detailed data extraction on available experimental studies is performed to gather any related information about the actuation strain. The elemental composition, manufacturing approaches, thermal treatments, applied stress, and post-processing steps that are commonly used to process NiTiHfX and have an impact on the material AS are used as input parameters of the ML models. Since a broad data collection is performed the information for each input factor was sufficient for the use of the majority of the accessible information in the literature on NiTiHfX AS. Considering most of the regular NiTiHfX processing factors also enables the option of tuning additional characteristics of NiTiHfX in addition to the ASs. The work is unique as is the first to fully investigate the NiTiHfX actuation strain prediction.

To forecast the NiTiHfX AS, a total of 901 data sets or 17,119 data points for eighteen inputs and one output were gathered, verified, and selected. Several machine-learning approaches were applied and joined to gather to guarantee robust modeling. The global model’s overall determination factor (R²) was 0.96, suggesting the viability of the proposed NN model. Such a model opens the possibility of intelligent material selection and processing to maximize the AS or shape memory effect of NiTiHf SMA.