Enhanced mode recognition of tensile damage in AM 316 L stainless steel using an NPDOA-optimised VMD–CWTIS–CNN framework

A hybrid signal processing framework, termed NPDOA – VMD – CWTIS, is proposed to investigate the deformation and damage mechanisms of additively manufactured (AM) 316 L stainless steel using acoustic emission (AE)-based non-destructive testing (NDT). In this framework, the Neural Population Dynamic Optimisation Algorithm (NPDOA) adaptively optimises the key parameters of Variational Mode Decomposition (VMD), enabling accurate separation of damage-related signal components. The extracted Intrinsic Mode Functions (IMFs) are subsequently denoised and reconstructed through Continuous Wavelet Transform and image subtraction (CWTIS), which effectively mitigates signal aliasing and enhances the extraction of true AE responses, achieving a signal-to-noise ratio of 37.46 dB. The peak frequencies of the reconstructed AE signals are clustered via the K-means algorithm, revealing three distinct damage modes: dislocation motion, twin – detwinning, and stress-induced martensitic transformation. A Convolutional Neural Network (CNN) is then trained using these clustered data for rapid and accurate damage classification, achieving a prediction accuracy of 94.1%. The proposed NPDOA – VMD – CWTIS – CNN framework significantly enhances the reliability and interpretability of AE data in NDT, providing a robust and automated tool for real-time damage monitoring in metal additive manufacturing.