Vibration signal–based surface roughness prediction with hybrid kernel extreme learning machine for robotic face milling

Industrial robots, in contrast to computer numerical control (CNC) machine tools, offer expanded workspaces and superior flexibility. However, their intricate dynamic characteristics pose challenges for accurately predicting surface roughness within the context of current data-driven intelligent manufacturing requirements. To address this issue, an intelligent prediction model utilizing cutting vibration signals was adapted. Comprehensive and diverse vibration modes were gathered through robotic face milling experiments targeting envelope zone vertices, with the posture influence in the construction of the data sample set being considered. By combining discrete Fourier transformation (DFT) and successive variational mode decomposition (SVMD), a reconstruction and processing method for stable cutting vibration signals was proposed, obtaining a series of multiple intrinsic mode functions (IMFs) that were free from noise and aliasing. Then a novel hybrid kernel extreme learning machine with Gaussian and polynomial hybrid kernel function (RBF_Poly_HKELM) was proposed to establish a mapping relationship with surface roughness by using enhanced Dragonfly algorithm (EDA) to improve the prediction accuracy. Experimental results under the testing postures demonstrated that this innovative approach achieved high-precision prediction of surface milling roughness, with mean absolute error (MAE) of 0.073 μm and root mean square error (RMSE) of 0.092 μm. Additionally, orthogonal experimental analysis provided valuable insights for selecting optimal cutting parameters tailored to specific milling tasks.