Machine learning based prediction modeling of micro-EDM of Ti–29Nb–13Ta–4.6Zr (TNTZ)

Shahid Ali, Didier Talamona, Asma Perveen

Scientific Reports, 2025

Full text Open access: Gold Metadata (CSL-JSON)
DOI 10.1038/s41598-025-05118-6
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

Although Ti–6Al–4V stands out as one of the best in biomedical, automotive, and aerospace applications due to its low density and higher corrosion resistance, the toxicity associated with Al and V elements is shifting the usage of Ti-based alloys with reduced toxic content but with more biocompatible elements (Zr, Ta, and Nb). Despite the advancements in the era of micromachining, machining these alloys with traditional machining methods is highly tedious. The present research evaluates the micromachining performance of the TNTZ (Ti–29Nb–13Ta–4.6Zr) alloy employing a tungsten carbide electrode using the µ-EDM (micro-electro-discharge-machining). The primary input parameters examined are voltage (80–130 V) and capacitance (10–400nF), with a feed rate of 0.09 mm/s during the experiments. The output responses assessed include VMR, OC, CErr, and SFR. Meanwhile, due to the complexity of the µ-EDM process, it presents significant challenges in predicting performance across different machining settings. The interactions between key process parameters, such as C and V, amplify their parametric sensitivity, making conventional simulation approaches inadequate for accurately modeling these interdependencies. To address these challenges, the latter part of this study explores machine learning techniques, particularly Multiple linear regressor (MLR), decision tree regressor (DTR), and artificial neural network (ANN) for predictive accuracy. The models are evaluated using two key performance metrics: normalized root mean squared error (NRMSE) and R-squared (R 2 ). The ANN demonstrated superior capability in handling experimental variability based on the prediction results. It has the highest R 2 of 0.99, the lowest NRMSE of 0.0245, and the percentage of prediction error is less than 5%.