Neural Network Optimizes Surface Quality and Energy Use in Steel Machining

Researchers have achieved a 52.27 percent multiresponse optimization improvement during the wire electrical discharge machining of hardened D2 tool steel by applying a hybrid framework based on Taguchi-Grey Relational Analysis methodology. The experimental investigation focused on establishing artificial neural network models and optimizing machining parameters to simultaneously reduce surface roughness, kerf width, and total energy consumption, addressing the operational requirement for rapid and accurate estimation of indicators in sustainable wire electrical discharge machining.

The machining experiments were structured around a Taguchi orthogonal array incorporating four primary operational parameters: pulse width, pulse space, wire speed, and feed rate.

Analysis of the experimental results revealed distinct operational effects across the evaluated indicators, with feed rate emerging as the primary influencing parameter for all responses based on analysis of variance and feature importance evaluations. Increases in feed rate substantially reduced both kerf width and total energy consumption, whereas increases in pulse width and wire speed led to corresponding increases in both kerf width and surface roughness. Across all evaluated metrics, the relative order of parameter importance following feed rate was identified as pulse width, wire speed, and pulse space.

To predict the machining indicators, the researchers developed artificial neural network models utilizing a feed-forward backpropagation architecture trained with the Levenberg-Marquardt algorithm. The collected experimental data were partitioned into separate training, testing, and validation subsets to ensure rigorous model calibration and verification across the multi-parameter cutting conditions.

The resulting artificial neural network models demonstrated high predictive accuracy, yielding determination coefficients exceeding 99 percent for surface roughness, kerf width, and total energy consumption. This strong agreement between the model predictions and desired experimental targets indicates that artificial neural networks provide an effective machine learning methodology for modeling indicators and optimizing parameters in sustainable wire electrical discharge machining implementations.

Subject of Research: Optimization and neural network modeling of wire electrical discharge machining for hardened tool steel

Article Title: A hybrid ANN-GRA prediction framework for optimizing the balance between surface integrity and energy demand in WEDM of tool steel

Article References: Güvercin, S., & Günay, M. (2026). A hybrid ANN-GRA prediction framework for optimizing the balance between surface integrity and energy demand in WEDM of tool steel. Neural Computing and Applications, 38(21), Article 873. https://doi.org/10.1007/s00521-026-12522-0

Image Credits: AI Generated

DOI: 10.1007/s00521-026-12522-0

Keywords: wire electrical discharge machining, artificial neural networks, hardened D2 steel, surface roughness, kerf width, energy consumption, grey relational analysis, machining optimization, hybrid, ANN-GRA, prediction, framework