Abstract
Employing Laser Powder Bed Fusion (LPBF) method to manufacture NiTiHf Shape Memory Alloy (SMA) is becoming more common. The major design property for NiTiHf is the transformation temperatures (TTs) which control the activation threshold of the SMA material and enable it to create the shape change due to a microstructure phase transformation. Given the high number of fabrication factors, machine learning (ML) approaches provide a promising approach to the design of SMA to control the TTs.
The main obstacle to using ML methods is the need for an established correlation between fabrication features and material properties. The presented work develops an ML approach to enable the prediction of the TTs for additively manufacturing NiTiHf. The work uses all available experimental data on additively and conventionally manufactured NiTiHf. Selected fabrication features included in the ML models consider the elemental compositions of NiTiHf, laser power, laser speed, hatch spacing, and almost all the processing steps historically used to manufacture, or heat treat the NiTiHf for SMA.
Multiple models of Multi-Layer Perceptron (MLP) and Radial Basis Function (RBF) Neural Networks (NN) are developed to predict the TTs of LPBF-manufactured NiTiHf. The models successfully predict the TTs for various NiTiHf fabrication conditions.
