Physics-Informed Optimization of Dross Formation inFiber-Laser Cutting: From Multi-Variable Design to aSingle Decision Rule

Laser cutting has the ability to produce detailed parts at small tolerances with minimal areas of heat impact, although theformation of dross still hinders the ability to produce repeatable and quality results. We experimented with stainless steel 201(SS201) having a five-factor design-laser power (Pu), cutting speed (V), assist-gas pressure (P), pulse frequency (F) as wellas the focus position (FP) in a 32-run Central Composite Design. On top of classical Response Surface Methodology, wepropose a physics-driven model, which consists of a dimensionless Dross Ejection Index (DEI) that balances melt-expulsionforces (assist-gas shear, recoil pressure) and retention (viscous drag, gravity, capillarity). Mapping of DEI to dross area(DA) using monotone sigmoidal law is a technique of collapsing scattered results onto one interpretable curve and providesa way of providing a practical threshold of clean cutting. Gaussian Process regression had the highest level of numericalaccuracy in cross-validated benchmarks (R2 ≈ 0.28, RMSE ≈ 22.8 mm2), but the De-sigmoid model was more interpretableand extrapolated more safely. Factor effects are in line with first principles: power and focus are leading (≈ 45% and 39%),then speed (7%), frequency (6.4%), and pressure (2.7%). The combined method transforms a tuning exercise involving fivevariables into a single-number decision rule—operate at DEI ≥ DEI∗—to provide definitive parameter indications to eliminatedross and enhance industrial quality in laser cutting.