Machine Learning-Enhanced Optical Monitoring for Identifying Pitting-Susceptible Zones in 316l Stainless Steel

Abstract

Forecasting stainless steel (SS) pitting corrosion remains challenging due to the need to identify nanometer-scale imperfections in surface passive films. Traditional analytical methods are costly, time-consuming, and limited to model systems with adequate signal-to-noise ratios. We propose an alternative approach that leverages optical signatures of passive layer properties which, when enhanced with unsupervised machine learning (ML) to extract signals even at noise level, successfully identifies pitting-susceptible zones (PSZ) in-situ on industrial SS 316L substrates. Complementary optical modeling and X-ray Photoelectron Spectroscopy (XPS) reveal chromium oxide deficiency in surface films over PSZ, consistent with established pitting mechanisms. This proof-of-concept demonstrates that ML-enhanced optical methods can serve as accessible, precise tools for PSZ identification, advancing the development of predictive corrosion monitoring systems.

Keywords: Pitting corrosion, machine learning, reflective microscopy, stainless steel