On 15 November, a team at the Institut für Strahlwerkzeuge (IFSW), University of Stuttgart, published an open-access paper in Scientific Reports: combining an ultrafast laser (600 fs, 1030 nm) with a galvanometer scanner and a coaxial optical coherence tomography (OCT) system for real-time closed-loop depth control, they volumetrically ablated AISI 304 stainless steel with a depth accuracy of 1–2 µm for target depths of 50–500 µm. Two corrections — extrapolation-based compensation of an initial depth offset and a pixel-wise moving-average termination criterion — cut the deviation of the basic closed loop, which exceeded 20 µm: a 100 µm target measured 99.7 µm, a deviation of only 0.3 µm.
01 / The problem: deeper is harder to control, and depth sensing itself drifts
Depth matters in volumetric ablation because it directly decides function: how much fluid a microchannel holds, the effective area of a cooling microcavity, the optical performance of a lens-shaped surface — all hang on depth. But getting depth right first time is hard, especially for deep features beyond 50 µm: beam divergence changes the fluence, pulse overlap and ablation efficiency along the depth axis; an unknown or non-uniform initial topography changes energy deposition and material response; and the laser-induced periodic surface structures (LIPSS) formed early in ablation alter the laser–material coupling.
The measurement side is trickier still. OCT depth sensing can run coaxially with the processing beam and track pulse by pulse, but the evolving surface morphology during ablation introduces two artefacts into the OCT signal: a depth offset in the initial ablation phase, and local fluctuations of the OCT depth reading. The paper’s reference experiment shows the two artefacts together can push the final depth more than 20 µm off target — in the Appendix B example, a 100 µm target ended at 120.1 µm under the basic closed loop. That is the gap this work closes: previous OCT closed-loop ablation reached mean form accuracies of about 10–20 µm, and about 14 µm RMS on non-uniform materials such as wood, while “first-time-right, single-digit-micrometre” accuracy had not been achieved.
02 / Method and evidence: two corrections bring deviation down to 1–2 µm
The setup is industrial-grade: a Fibercryst Femto 30 ultrafast laser, 600 fs, 1030 nm, 70 kHz, with pulse energies up to 35 µJ (2.5 W average power), focused through a galvanometer scanner and a 163-mm F-θ lens to a 50 µm spot; a Precitec CHRocodile 2 OCT system centred at 1080 nm, combined with the processing beam on the same dichroic mirror, sampling at 70 kHz in sync with the laser repetition rate so it can track the surface pulse by pulse; and an RTC 5 card with proprietary software closing the loop. The tests ablated 2-mm squares on 25 × 4 mm cold-rolled AISI 304 stainless-steel samples, with an 8 µm line spacing and scan speeds of 560–1120 mm/s.
Each correction targets one artefact: first, the approximately linear depth progression after the OCT signal stabilizes is extrapolated to determine the morphology-dependent initial OCT depth offset, which is then compensated in the control algorithm; second, a pixel-wise moving average of the last ten depth measurements serves as the termination criterion — once a pixel’s moving average reaches the target depth, processing there stops permanently — suppressing the uncertainty of individual readings. Validation against an independent laser scanning microscope: a 100 µm target measured 99.7 µm after 300 scans, a deviation of only 0.3 µm; across different initial surface states, target depths (50–500 µm) and peak fluences, the absolute deviations between measured and target depths were predominantly within 1–2 µm, with a maximum relative error of 3%, an improvement of about one order of magnitude over the basic closed loop — and OCT depth sensing stayed reliable even on rough and highly modulated surfaces. The team also machined a complex three-dimensional geometry with the same control (Figure 6), the target specified as a grayscale image and checked by OCT measurement and scanning electron microscopy.
03 / Significance and limits: a depth gauge for precision laser machining
The engineering value is direct: microfluidics, thermal management, optical surfaces and lubricating microchannels all need depth right first time, not by trial and error and rework. The point of the OCT coaxial closed loop with its two corrections is that it makes both “measuring right” and “stopping right” accurate to single-digit micrometres, and insensitive to the initial surface state and process parameters — the accuracy held on the as-received cold-rolled surface (Sa = 0.80 µm) and the polished surface (Sa = 0.03 µm) alike, and across fluences. For anyone building laser-processing software and processes, the pattern transfers directly: model the measurement artefact into the control, instead of only tuning laser parameters.
The limits should be stated plainly too: the results are verified so far only on AISI 304 stainless steel, and the paper gives no data for other materials or for multilayer work such as selective ablation layer by layer; the supporting data are not published with the article and must be requested from the corresponding author; and this is an industrial-grade laboratory setup — moving to a production line still has integration and throughput to clear. What to watch next is whether the closed loop extends to freeform surfaces and multilayer workpieces — the step that would carry it from the laboratory to the shop floor.
PRIMARY REFERENCES & LINKS
- Micron-precision depth control in automated volumetric laser ablation with ultrashort laser pulses
Paper (PDF, 14 pages in full) and publication record verified: publication date 15 November 2026, DOI 10.1038/s41598-026-79064-2, CC BY 4.0 licence, authors and affiliations (IFSW, University of Stuttgart; Lightpulse Laser Precision), and the depth accuracy, target-depth range, measured value for the 100 µm target, basic closed-loop deviation, laser and OCT parameters, the two correction methods and the complex-geometry demonstration in Figure 6 in the abstract and main text; the Figure 1 caption checked.
Open source link ↗ - DOI: 10.1038/s41598-026-79064-2 (persistent identifier)
Persistent identifier for the same paper, for citation and retrieval.
Open source link ↗
