Intelligent surface integrity management : integrating mechanical systems with explainable AI for active resonance damping

Abstract

This study synthesizes research on two critical industrial challenges in tribological systems: wear evolution and chatter instability. The research trajectory moves from high-resolution observation to detection, and finally to active suppression. Paper I - ’In-situ precision wear characterization using Digital Holographic Microscopy and computational image processing’, addresses the limitations of ex-situ wear measurement. It introduces a novel integration of Digital Holographic Microscopy (DHM) with Acoustic Emission (AE) sensors on a pin-on-disc tribometer. By applying advanced computational image processing, specifically Laplacian operators and Canny edge detection to holographic phase data, the study demonstrates the ability to quantify wear track width in real-time. This optical ’ground truth’ is then correlated with acoustic signatures, enabling precise wear characterization without interrupting operation.Paper II - ’Chatter detection and identification based on mode decompositions’, tackles the detection of ’chatter’, a self-excited vibrational instability that plagues rolling and machining processes. Standard spectral analysis with FFT often fails to identify the early onset of chatter due to the non-stationary nature of the signals. This study rigorously evaluates Empirical Mode Decomposition (EMD), Bivariate EMD (BEMD), and Variational Mode Decomposition (VMD). The findings confirm that VMD provides superior mode separation, isolating chatter frequencies into distinct IMFs. Features extracted from these modes are used to train Machine Learning classifiers, with Random Forest models achieving 96% accuracy in distinguishing stable operation from chatter onset.Paper III - ’Damping of vibrational resonance induced chatter marks: Probabilistic control approach’, closes the loop by implementing an active control architecture. Utilizing the features identified in Paper II, an XAI model forecasts the intensity of future resonance frequencies. Two probabilistic control strategies - Bayesian Optimization (BO) and Optimal Control (OC) are employed to adjust rotational speeds in real-time. The results demonstrate that the OC strategy not only suppresses failure frequencies to negligible levels (0.035 m/s2) but also induces a ’healing’ effect on the contact surfaces through controlled elastic deformation. Collectively, this body of work establishes a validated framework for Smart Tribology and Digital Twin transforming passive machine elements into intelligent systems capable of self-diagnosis and self-correction.