Machine Vision Moiré Fringes Level Micro-Surfaces to 0.01 Degrees

Transferring flakes of graphite a few micrometers across is an exercise in patience. In laboratories studying self superlubricity, an ultralow-friction phenomenon that could reshape micro-electro-mechanical systems, researchers must move hundreds of microscopic graphite islands from one surface to another without tilting the stamp that carries them. Even a slight tilt creates height differences of several micrometers across the contact zone, so islands on the lower side stick while those on the higher side refuse to release, and excessive tilt can compress and fracture the delicate flakes. A team reporting in the journal Discover Robotics has now built a compact, non-contact automatic leveling system that uses machine vision and Moiré fringe patterns to measure and correct tilt angles in real time, reaching a leveling precision of 0.01 degrees within 10 seconds.

The problem the authors set out to solve is one of scale. Conventional leveling approaches falter at the micro-scale for different reasons. Manual shimming depends entirely on the feel of the operator, with repeatability that varies from person to person and degrades with fatigue. Motorized commercial stages tend to be bulky; the researchers cite a layered lead-screw and worm stage standing 80 millimeters high that blocks the long-working-distance objectives of a microscope, and a ball-screw platform that delivers 5 micrometer positioning accuracy but carries 2 micrometers of reversal backlash and drifts 8 micrometers for a 5 degree Celsius temperature rise. Sensor-based active systems, meanwhile, rely on contact probes or costly encoders vulnerable to environmental noise. One electronic tilt sensor they describe has a thermal coefficient of 0.01 degrees per degree Celsius, meaning a 2 degree laboratory fluctuation consumes the entire accuracy budget, while an electromagnetic angle sensor picks up plus or minus 0.02 degrees of common-mode noise at 50 hertz mains frequency and would require impractical magnetic shielding.

The new system takes a different route, inspired by white light interferometry-based leveling techniques but built instead on the Moiré effect, a phenomenon that has underpinned precision measurement since the 1960s. Two linear gratings are used: an alpha grating mounted on the graphite flake transfer surface and a beta grating fixed to the sample stage surface. When images of the two gratings are captured by a camera and overlaid, the superposition generates Moiré fringes, and these fringes act as a powerful magnifier of angular change. According to the geometric relations derived in the paper, a given angular misalignment is amplified into a fringe period orders of magnitude larger than the grating pitch itself, which is what allows sub-microradian resolution with conventional optics.

The mathematics behind the measurement is carefully laid out. For two gratings of nominal pitch d superimposed at a small relative angle, the fringe spacing and orientation follow closed-form expressions valid under a small-angle approximation that keeps errors below 0.5 percent when the relative angle is under 5 degrees. When the transfer head tilts, its grating pitch and displacement project into the plane of the reference grating, shifting the fringe spacing, slope and, crucially, the lateral phase of the fringes. For tilts under 1 degree, the spacing and slope change by less than 0.2 percent and fall below the optical resolution limit, but the fringe phase shift remains linear with a determination coefficient above 0.999 and can be detected with sub-pixel accuracy. The team therefore adopted phase as the primary observable, deriving a closed-form tilt estimator that converts measured fringe shifts into pitch and roll angles using the rotation radius of the setup, 100 millimeters.

A single grating set, however, is blind to rotation about its own grating lines. To decouple the two tilt axes, the researchers implemented an orthogonal pair of grating sets sharing the same 10 micrometer pitch and the same 4 degree relative angle, one set sensitive to roll and the other to pitch. The 4 degree angle was chosen deliberately: in experiments sweeping grating angles from 1 to 9 degrees, the team found that too few fringes led to significant errors in identifying fringe slope, spacing and phase, while too many amplified errors in fringe spacing during displacement calculations. Five to six well-presented fringes proved optimal, and a 4 degree grating angle delivers exactly that.

Extracting the fringe parameters falls to a machine vision pipeline. Images from a camera mounted on a microscope are converted to grayscale, enhanced by histogram equalization, smoothed with Gaussian filtering, and binarized adaptively, with dilation and erosion operations cleaning up fringe morphology. A probabilistic Hough transform then detects the fringes, and line fitting to sub-pixel precision yields the spacing, slope and phase that feed the tilt angle calculation. A PID controller closes the loop, adjusting the leveling platform in real time based on the deviation between the measured tilt angle and the desired level state, with the proportional term responding to current deviation, the integral term eliminating steady-state errors, and the derivative term damping rapid changes.

One subtlety of phase-based measurement is ambiguity: because the fringe phase is periodic, the raw reading cannot distinguish between tilts separated by whole fringe orders. The authors resolved this with a clever multi-grating scheme, simulating three gratings with relative angles of 4, 5 and 7 degrees simultaneously and applying the Chinese Remainder Theorem to the three phase readings, which yields a unique tilt solution within plus or minus 3.5 degrees. Noise propagation analysis showed that camera readout noise, photon shot noise and quantization error together contribute a phase uncertainty of roughly 0.5 micrometers, comfortably below the threshold of about 20.5 micrometers required for reliable reconstruction. Across more than 5,000 Monte Carlo runs, the root-mean-square error between retrieved tilt and ground truth was 0.005 degrees, meeting the 0.1 degree accuracy requirement for graphite flake transfer leveling.

The experimental hardware was built around gratings fabricated by photolithography on quartz substrates, with a 10 micrometer grating constant and 1 by 1 millimeter dimensions, mounted orthogonally on the transfer surface and the sample stage. A scientific-grade camera with 3072 by 2048 pixel resolution, coupled to a 5 times magnification objective, provides an object-plane resolution of 0.568 micrometers per pixel. In leveling tests, the first operation reduced tilt angle error to within 0.05 degrees with an average time of 3.196 seconds, and the second operation pushed the error down to within 0.005 degrees with an average time of 6.190 seconds. With a 5 times objective, a one-pixel fringe offset corresponds to an actual displacement of 0.88 micrometers and a tilt angle change of 0.0028 degrees, which sets the system’s angular resolution. The leveling hardware itself has a minimum adjustable angle of 0.01 degrees, so the second leveling operation reaches that upper limit, comfortably exceeding the requirement of better than 0.1 degrees for graphite flake transfer.

The authors also quantified the system’s uncertainty budget through both analytical error propagation and Monte Carlo simulation with 5,000 iterations. The analytical root-sum-square estimate was 0.0025 degrees, while the Monte Carlo result including Chinese Remainder Theorem effects was 0.005 degrees, a close agreement the team says validates the uncertainty estimation approach. Major uncertainty sources include camera noise and sub-pixel detection error at roughly plus or minus 0.0012 and 0.0015 degrees respectively, grating period non-uniformity of plus or minus 0.1 micrometers contributing about 0.0010 degrees, and rotation radius uncertainty of plus or minus 0.5 millimeters contributing a similar amount. Calibration proved stable as well: ten consecutive calibrations yielded a standard deviation of 0.0008 with a maximum deviation of 0.002, and a 24-hour thermal drift test at 20 plus or minus 2 degrees Celsius showed less than 0.003 drift, though recalibration is required after optical maintenance, temperature shifts beyond 5 degrees, or every 8 hours of continuous operation.

The practical appeal of the approach lies in its footprint and simplicity. The method requires only two small gratings added to the working table surface, leaving the cramped optical path of the microscope free for other experiments, and it eliminates complex and time-consuming manual calibration. The authors note that resolution could improve further: limited by current grating manufacturing, the smallest fabricable grating constant is 4 micrometers, and if such gratings were imaged under a 20 times microscope, the theoretical angular resolution could reach 0.000112 degrees. Limitations remain, including the need to verify stability and reliability in complex industrial environments with high temperature, high humidity or strong electromagnetic interference. Still, for the micro-scale transfer tasks that underpin superlubricity research and, potentially, micro and nano fabrication more broadly, a fast, compact leveling tool that reaches hundredths of a degree in seconds is a meaningful step forward.

Subject of Research: An automatic machine vision leveling system using Moiré fringes for high-precision graphite flake transfer

Article Title: Micro-level automatic leveling system based on machine vision for graphite flakes transfer

Article References: Liu, B., Chen, Y., Wang, Y., Nie, J., Ruan, C., & Yang, J. (2026). Micro-level automatic leveling system based on machine vision for graphite flakes transfer. Discover Robotics, 2(1), Article 20. https://doi.org/10.1007/s44430-026-00034-4

Image Credits: AI Generated

DOI: 10.1007/s44430-026-00034-4

Keywords: machine vision, Moiré fringes, graphite flakes, self superlubricity, tilt angle measurement, precision leveling, micro-scale manufacturing, PID control, gratings, 2D materials, uncertainty analysis, Discover Robotics