Surface quality classification in burnished aluminum alloys based on nonlinear dynamic characteristics and machine learning

Active control of surface quality requires insight into both processing parameters and the nonlinear dynamic behavior of machining systems. However, existing studies mainly focus on microscopic surface attributes and often overlook their relationship with system-level nonlinear dynamics, limiting both predictive accuracy and mechanistic understanding. To address this gap, this study investigates the surface burnishing process (SBP) by integrating process parameters, vibration-based nonlinear dynamic analysis, and machine learning. A quantitative intrinsic mode function (IMF) screening method based on ensemble empirical mode decomposition (EEMD) and power spectral density (PSD) was proposed to enhance vibration signal denoising and feature reliability. Chaotic behavior of the SBP system was confirmed by a positive maximum Lyapunov exponent (λ max >0), and a set of recurrence quantification analysis (RQA) parameters was extracted. Three feature scenarios, SBP parameters with positional encoding, chaotic features, and their combination, were evaluated for classifying surface roughness and hardness. Results showed that surface roughness was predominantly governed by burnishing parameters, whereas hardness prediction benefited more from RQA parameters reflecting the surface deformation stability. The findings clarify the distinct roles of deterministic and dynamic factors in surface-quality formation and provide a flexible, physically interpretable framework for data-driven surface-quality prediction and adaptive manufacturing applications.