Diagnosing Ion Beam Cleaning Ignition Failures in PVD Processes Using Interpretable Machine Learning on Kaufman-type Ion Sources for Optical Thin-Film Coatings

Weilacher A, Abb V, Schneider A, Barth M, Reichenstein T, Franke J (2026)

Publication Type: Journal article

Publication year: 2026

Journal

Original Authors: Alexander Weilacher, Valerius Abb, Alexander Schneider, Martin Barth, Tobias Reichenstein, Jörg Franke

Book Volume: 144

Pages Range: 369-374

DOI: 10.1016/j.procir.2026.02.030

Abstract

In industrial multilayer thin-film coatings production using physical vapor deposition (PVD), Kaufman-type ion sources are used for in-situ ion beam cleaning (IBC) to achieve required surface integrity under high vacuum, which is paramount for high-quality optical thin-film coatings. IBC has been demonstrated to be an effective means of removing surface contaminants; however, the process is susceptible to ignition failures, which can lead to production halts. Operational data (voltages, currents, pressures, temperatures, etc.) was collected from an industrial batch PVD system, and an XGBoost classifier was trained to predict ignition success. The classifier achieved high performance (accuracy ~ 0.91, ROC-AUC ~ 0.93, F1-score ~ 0.76) while being tuned for class imbalance. It successfully predicted failed ignition runs, enabling data-driven conclusions for process optimization. The SHAP (SHapley Additive exPlanations) framework enabled the interpretation of the model, revealing that maximum chamber pressure, cathode current/voltage and gas flow profiles during IBC, and the diffusion pump temperature are key factors influencing ignition reliability. These findings indicate that maintaining optimal vacuum pressure and stable supply conditions is imperative to minimize ignition failures. The innovative application of explainable machine learning (ML) to this ion beam-assisted PVD process yields actionable insights for process optimization and enhanced coating reliability.

Authors with CRIS profile

Involved external institutions

Ostbayerische Technische Hochschule Regensburg (OTH) Germany (DE)

How to cite

APA:

Weilacher, A., Abb, V., Schneider, A., Barth, M., Reichenstein, T., & Franke, J. (2026). Diagnosing Ion Beam Cleaning Ignition Failures in PVD Processes Using Interpretable Machine Learning on Kaufman-type Ion Sources for Optical Thin-Film Coatings. Procedia CIRP, 144, 369-374. https://doi.org/10.1016/j.procir.2026.02.030

MLA:

Weilacher, Alexander, et al. "Diagnosing Ion Beam Cleaning Ignition Failures in PVD Processes Using Interpretable Machine Learning on Kaufman-type Ion Sources for Optical Thin-Film Coatings." Procedia CIRP 144 (2026): 369-374.

BibTeX: Download