Optimization of surface roughness and tool life during the milling of zirconium alloy using response surface methodology and machine learning model

Abstract

Abstract Commercially, zirconium alloys are beneficial for nuclear power generation with the capacity to increase electricity output. This study employed the Response Surface Methodology (RSM) and Machine Learning (ML) model validated by physical machining experiments to investigate the surface roughness (SR) and tool life (TL) during the milling of zirconium alloys. The process parameters employed include: depth of cut (DoC: 0.1–0.35 mm), feed per tooth (FPT: 0.1–0.25 mm), and cutting speed (CS: 50–200 m/min). Milling was performed on a Computer Numerical Control (CNC) milling machine. A carbide cutter with 12 o positive rake angle, a diameter of 10 mm, and a length of 40 mm × 90 mm was employed for the end milling. A Neural Network (NN) comprising of input, hidden and output layers was trained with backpropgation using the Levenberg Marquardt (LM) optimised algorithm. The machine learning model was trained and implemented in the MATLAB 2022b environment. Results showed that the least value of SR (0.24 $$\:\mu\:$$ m) obtained via experimental measurement was achieved with the following process parameters: DoC (0.22 mm), FPT (0.15 mm), and CS (105.0 m/min). The experimental trial that produced the highest TL (108.223 min) had a combination of the following process parameters: DoC (0.10 mm); FPT (0.10 mm) and CS (75.0 m/min). For SR; the RSM optimisation gave an optimum value of 0.3845 μm at CS 106.29 m/min, DoC 0.35 mm, FPT 0.10 mm. Similarly, optimum TL value 66.3222 min at CS 112.5 m/min, DoC 0.22 mm, FPT 0.15 mm was obtained from RSM optimisation. The results from both the RSM and NN models indicate that the FPT was the most significant factor influencing SR while the combination of the CS and FPT was the most significant factor influencing tool life. This indicates that both models are suitable for predictive purpose. However, the ML model slightly outperformed the RSM model. Scanning Electron Microscopy (SEM) analysis indocated that the cracking mode of wear was suspected during machining at low CS, FPT, and DoC, whereas flaking, notching, and crater modes of wear were suspected during machining at high CS, FPT, and DoC. Hence, this study highlighted some findings that could promote the machinbility of zirconium alloys.