TY - JOUR
T1 - Experimental investigations and prediction of WEDMed surface of nitinol SMA using SinGAN and DenseNet deep learning model
AU - Vakharia, Vinay
AU - Vora, Jay
AU - Khanna, Sakshum
AU - Chaudhari, Rakesh
AU - Shah, Milind
AU - Pimenov, Danil Yu.
AU - Giasin, Khaled
AU - Prajapati, Parth
AU - Wojciechowski, Szymon
PY - 2022/5/1
Y1 - 2022/5/1
N2 - Shape memory alloys (SMA) hold a very promising place in the field of manufacturing, especially in biomedical and aerospace applications. Owing to the unique and favorable properties such as pseudo elasticity, shape memory effect and Superelasticity, Nitinol is the most popular amongst other SMAs. However, a major challenge lies in the final surface features of the machined component. In the current study, Nitinol rods were machined using the wire electrical discharge machining (WEDM) process and subsequently, the surfaces were investigated using the Field emission scanning electron miscroscope (FESEM) technique for the features. In addition to this, Singular Generative Adversarial Network (SinGAN) and DenseNet deep learning models were prepared and applied for the prediction of surface morphology and its correlation with the process parameters. It was concluded from the study that the DenseNet model was highly effective in predicting the surface images with 100% average accuracy both with training and testing whereas the least average accuracy of 99.13% and 98.98% with training and testing respectively are observed with the MNB model. Thus, the proposed methodology can prove to be highly beneficial for prediction, specifically for manufacturing applications where the data is limited.
AB - Shape memory alloys (SMA) hold a very promising place in the field of manufacturing, especially in biomedical and aerospace applications. Owing to the unique and favorable properties such as pseudo elasticity, shape memory effect and Superelasticity, Nitinol is the most popular amongst other SMAs. However, a major challenge lies in the final surface features of the machined component. In the current study, Nitinol rods were machined using the wire electrical discharge machining (WEDM) process and subsequently, the surfaces were investigated using the Field emission scanning electron miscroscope (FESEM) technique for the features. In addition to this, Singular Generative Adversarial Network (SinGAN) and DenseNet deep learning models were prepared and applied for the prediction of surface morphology and its correlation with the process parameters. It was concluded from the study that the DenseNet model was highly effective in predicting the surface images with 100% average accuracy both with training and testing whereas the least average accuracy of 99.13% and 98.98% with training and testing respectively are observed with the MNB model. Thus, the proposed methodology can prove to be highly beneficial for prediction, specifically for manufacturing applications where the data is limited.
KW - shape memory alloy (SMA)
KW - nitinol
KW - wire electrical discharge machining (WEDM)
KW - surface roughness (SR)
KW - SinGan
KW - deep learning
U2 - 10.1016/j.jmrt.2022.02.093
DO - 10.1016/j.jmrt.2022.02.093
M3 - Article
VL - 18
SP - 325
EP - 337
JO - Journal of Materials Research and Technology
JF - Journal of Materials Research and Technology
SN - 2238-7854
ER -