Physics-guided explainable machine learning for multi-response modeling of electrochemical micro-machining using polymer graphite electrodes
Scientific Reports, cilt.16, sa.1, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 16 Sayı: 1
- Basım Tarihi: 2026
- Doi Numarası: 10.1038/s41598-026-46315-1
- Dergi Adı: Scientific Reports
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, BIOSIS, Chemical Abstracts Core, MEDLINE, Directory of Open Access Journals
- Anahtar Kelimeler: Electrochemical micro-machining, Explainable machine learning, Physics-guided modeling, Polymer graphite electrode, Process optimization
- Gazi Üniversitesi Adresli: Evet
Özet
Electrochemical micro-machining (ECMM) enables high-precision fabrication of micro-features in difficult-to-machine materials; however, its strongly nonlinear, multi-physics nature and the high cost of experimentation severely limit reliable data-driven modeling. This study presents a physics-guided machine learning framework for robust multi-response prediction of ECMM performance using polymer graphite electrodes. Controlled experiments were conducted with non-treated and cryogenically treated electrodes, and four critical responses were evaluated: material removal rate (MRR), overcut (Oc), surface roughness (Ra), and taper angle (Ta). Physics-guided descriptors incorporating interaction-driven and severity-based features were constructed to embed mechanistic structure associated with electrochemical excitation and tool-electrolyte-workpiece coupling. Ensemble learning models were trained and rigorously validated using repeated cross-validation. The optimal physics-guided XGBoost models achieved coefficients of determination of 0.817 (MRR), 0.914 (Oc), 0.866 (Ra), and 0.769 (Ta), with corresponding mean absolute errors of 0.013 g/min, 0.026 mm, 0.261 μm, and 0.049°, respectively, consistently outperforming polynomial and purely data-driven baselines. Ablation analysis confirmed that physics-guided feature integration improved both predictive accuracy and cross-validated stability under limited experimental data conditions. Parity and residual diagnostics demonstrated strong generalization without systematic bias. The results establish physics-guided learning as an effective strategy for bridging electrochemical process physics and data-driven modeling, enabling accurate, robust, and interpretable prediction of ECMM responses. The proposed framework provides a scalable foundation for intelligent micro-manufacturing, with potential applications in adaptive process optimization and digital manufacturing workflows.