untetheredatom.com › EBSD/TKD › FAQ
The colors did something weird and we both know it.
Why the indexing, the colors, or the cleanup did what they did, each answer linked to the interactive part that shows it happening.
Because optical brightness and EBSD readiness are measured at different depths. An optical microscope resolves features of a few hundred nanometres, so a surface can look flawless while carrying a deformed or smeared layer that thick. EBSD collects its signal from roughly the top 10 to 40 nm at 20 kV, with the shallow end for denser materials, so that same layer is all the detector ever sees and you get a grey blur rather than bands. The numbers are the whole argument: SiC grinding leaves 0.5 to 2 µm of damage, and 1 µm diamond still leaves measurable deformation on tungsten after ten minutes, which is one to four orders of magnitude deeper than the window. That is why no amount of finer diamond fixes it and why the last step has to be a different kind of process: colloidal silica, electropolishing, or a low-voltage ion beam. Before blaming preparation, rule out the cheap causes in one minute: a tilt that is not 70 degrees, a wrong working distance, the wrong phase file, and a charging or contamination layer. Put your method on the depth axis →
The shape of every recipe is the same. Section with as little damage as you can afford, grind to flat, take diamond down far enough that the final step can finish in a sensible time, then finish with something that is not grinding or diamond. That last step is what makes the sample work. It is almost always colloidal silica on a napless cloth, often on a vibratory polisher, for far longer than feels reasonable. Twenty minutes is a minimum, an hour is common for superalloys, and commercially pure aluminium has been published at fourteen hours. Electropolishing replaces it for single-phase conductors, and a low-voltage ion beam replaces it when nothing mechanical will do. What changes between materials is not the shape but the details, and they change a lot: titanium gets worse if you walk diamond down to 1 µm, magnesium cannot touch water, and a thin film should not be polished at all. Open the full recipe for your material →
You need something after diamond, and colloidal silica is usually the cheapest way to get it. Diamond cuts mechanically and leaves deformation behind whatever the particle size: on titanium, diamond slurry left up to 300 MPa of residual stress measured by focused ion beam ring-core milling, while colloidal silica, vibratory polishing and electropolishing each left nothing detectable. Colloidal silica works chemically as well as mechanically, forming a soft reaction layer that the silica then removes, so it takes material away without pushing damage in. The alternatives to it are electropolishing, which is faster but needs a conductor and rounds edges and exaggerates voids, and ion polishing, which handles what nothing else will. Stopping at diamond is only reasonable if you are indexing a soft, coarse-grained material and will accept a poor hit rate. Compare what each method leaves behind →
Start from the published method for your alloy rather than from first principles, then tune the time. The usual starting points are Struers A2 at 39 to 48 V for aluminium and austenitic stainless, and A3 at 35 to 45 V for titanium at 18 to 20 °C. A2 at about 35 V suits nickel alloys, D2 suits copper and brass, and sodium hydroxide solutions suit tungsten. Whatever the electrolyte, the physics is the same: a current against voltage curve has a flat plateau where a viscous film controls the rate, and only inside that plateau do you polish. Below it you etch and reveal grain boundaries, above it you pit and gas evolution scars the surface. Find the plateau on your own cell by sweeping voltage and watching current, because cell geometry, flow and temperature all move it. Three cautions: electropolishing attacks phases at different rates, so avoid it for multiphase quantitative work; it rounds edges and exaggerates porosity and inclusions, as ASTM E1558 states; and many classic recipes are perchloric acid in alcohol, which is an explosion hazard with real rules attached. Read the electropolishing and perchloric acid rules →
Yes, and the voltage is the whole question, because damage depth falls steeply as you drop it. On silicon, gallium leaves about 22.5 nm of amorphous material at 30 kV, about 2.5 nm at 5 kV and about 1 nm at 2 kV. An argon broad beam gives about 8 nm at 5 kV, 3 nm at 2 kV and under 1 nm below 1 kV at a 10 degree glancing angle. Set against an information depth of tens of nanometres, a 30 kV finish is inside or beyond the window and a 2 kV finish is comfortably below it, so always end with a low-kV clean-up pass rather than stopping where the milling was fast. Two material cautions: gallium attacks aluminium grain boundaries, so use a xenon plasma source or an argon broad beam there, and those silicon numbers are the best measured set in the literature, with no equivalent published for copper, titanium or nickel superalloys. See the ion damage depths on one axis →
Because the frame the map positions are written in and the frame the Euler angles are written in do not agree, and one of them is left-handed relative to the other. A reflection is not a rotation, so no import rotation puts it back: the data has to be reflected. Everything still looks normal, which is the difficulty. The two things that catch it are a fiducial with a handedness, an L-shaped scratch or an asymmetric set of indents rather than a cross or a single line, and a specimen with a texture you already know, because a rolling texture pole figure is chiral and its mirror image is visibly the wrong one. Once you know which way your instrument and software pair goes, fix it at import rather than at plotting, so every later figure inherits the correction. Flip it and watch on the interactive page →
Ask the software, not the internet: the defaults differ between packages and change between versions, so a table copied from somewhere else will eventually be wrong for you. MTEX documents the problem directly, offers numbered settings for EDAX files and notes that Oxford and Bruker present the same problem with different alignments, and it enforces one invariant after import: the Euler angles refer to the map frame, so the map x and z axes are the axes the Bunge rotations are defined about. What settles it in practice is an afternoon: map a specimen carrying a handed fiducial, check a known texture including its handedness, verify that coherent twins come out at 60 degrees about a <111> axis and not just at 60 degrees, and import the same file in a second package to see whether the two agree. Then write the answer in the folder with the data. See the checks worked out →
No, and this is the trap. Misorientation angles survive nearly every frame error: a mirrored frame, a 90 degree swap of RD and TD, and a transposed orientation matrix all leave the angle between two grains exactly unchanged, because the angle of a rotation equals the angle of its inverse and a reflection applied to both grains cancels in the difference. So a clean boundary-misorientation distribution, a plausible grain size, a sensible KAM map and a believable sigma-3 fraction are all compatible with completely mis-framed data. Anything computed from angles alone is safe; anything computed from directions, a texture component, a Schmid factor, a slip trace, a boundary plane, a lift-out direction, is wrong and will not announce it. The one angle-based check that does bite is the twin axis: 60 degrees is not enough, it has to be 60 degrees about <111>. See what survives each mistake →
A Hough transform converts the pattern's bands into peaks (exploiting the line-point duality between real and Hough space), then a triplet-vote algorithm matches band intersections against the crystal's known interplanar angles to solve for Euler angles. This page runs the real algorithm on a live simulated pattern. See it worked out on From Kikuchi Bands to an Orientation →
Confidence index (CI) measures how much better the best solution votes than the next-best, while mean angular deviation (MAD) measures how well the matched bands actually agree with the solution geometrically, they catch different failure modes. This page checks both against known ground truth. See it worked out on From Kikuchi Bands to an Orientation →
IPF coloring maps a crystal direction onto a color wedge, and that mapping isn't one-to-one across the full orientation space, so orientations tens of degrees apart can land on nearly identical colors. This page's impostor-pair finder locates exactly these near-collisions. See it worked out on Reading an IPF Map →
Misorientation is the disorientation axis and angle that rotates one grain's orientation into the other's, minimized over crystal symmetry so you get the smallest equivalent angle. This page lets you click two grains and computes it directly. See it worked out on Misorientation, Boundaries & CSL →
Grain count depends on where you draw the line between "a real boundary" and "just orientation noise," so a stricter or looser threshold reclassifies boundaries and can merge or split grains without the underlying data changing at all. This page's threshold slider shows the count shift live. See it worked out on Misorientation, Boundaries & CSL →
It depends on the question. If you care about Hall-Petch-style mean free path, twins are real obstacles and should count as separate grains. If you care about processing history or recrystallization, annealing twins are inherited by the parent grain and merging them gives a more meaningful size. The EBSD-3 page's "merge Sigma-3 twins" checkbox shows how the grain count and mean size shift between the two conventions. See it worked out on Misorientation, Boundaries & CSL →
Pole figures plot where a specific crystal plane's normal points across all grains; ODF sections (by φ₂) unpack that into the full orientation distribution, with named texture components (cube, rolled-fcc, fibre textures) sitting at specific, recognizable addresses. This page builds textures grain by grain and marks those addresses. See it worked out on Texture: Pole Figures & ODFs →
Cleanup algorithms reassign or remove low-confidence pixels, which necessarily changes grain boundaries and small-grain/twin detection, sometimes toward the truth, sometimes away from it depending on how aggressive the settings are. This page corrupts a known-truth microstructure and runs real cleanup on it so you can see the error curve directly, it's U-shaped, not monotonic. See it worked out on Cleaning Data Honestly →
TKD (transmission Kikuchi diffraction) trades bulk EBSD's ~70° tilted-surface interaction volume for a thin-foil transmission geometry, which gives much better spatial resolution for nanocrystalline material, at the cost of needing a TEM-thin specimen. This page's interaction-volume sandbox compares both directly on the same nanocrystalline film. See it worked out on Why TKD Beats EBSD at the Nanoscale →
TKD needs enough crystal thickness for the backscattered electrons to form Kikuchi bands; below about 40 to 60 nm (material-dependent), there are too few pattern-carrying exits and the pattern washes out. The EBSD-6 page's thickness-window model shows the sweet spot directly: too thick and the spatial resolution degrades, too thin and there is no pattern to index. See it worked out on Why TKD Beats EBSD at the Nanoscale →
Lattice damage degrades Kikuchi band sharpness gradually, so IQ falls continuously as strain or damage increases, well before it gets bad enough to actually break indexing. This page damages a lattice progressively and shows exactly where IQ starts falling versus where indexing fails. See it worked out on Pattern Quality as Data →
Not safely: several different microstructural features produce IQ maps that visually resemble strain contrast, but only one of them actually is strain. This page walks through three look-alike IQ maps and which one is the real thing. See it worked out on Pattern Quality as Data →
The electron interaction volume straddles two crystals near a boundary, producing a blurred pattern from two overlapping Kikuchi patterns. The IQ drop width depends on the interaction volume diameter (beam energy, material, and tilt geometry) and the boundary inclination through the foil. The EBSD-7 page covers what IQ is and is not; the EBSD-6 page's interaction-volume model shows the geometry. See it worked out on Pattern Quality as Data →
KAM averages the misorientation to each pixel's neighbours over a kernel whose physical size is the step size times the kernel radius, so a larger step returns a larger KAM. That part is supposed to happen: the density formula divides by the step, and on a noiseless map the two cancel exactly, leaving a density that does not depend on the step at all. Measurement noise is what breaks the cancellation, because noise does not shrink when the step does, so dividing by a smaller step multiplies it up. On a simulated fine-grained bcc iron with 500 nm grains each bending at 2° per micron, Hough-grade indexing at 0.5° reports about twenty times the true dislocation density at a 45 nm step. Coarsening the step helps, but only up to a point, because a second limit closes in from the other side: once the kernel is wider than the grains, the threshold discards nearly every neighbour pair it reaches for. At a 360 nm step 79% of pairs are thrown away and the density reads about half the truth; at 720 nm every pair is thrown away and the map returns nothing at all. So the useful step is a window, not a direction, and on a fine-grained material the two conditions can fail to overlap, at which point better angular precision is the only thing that opens it. Separately, and not fixable by any step size, a surface map gives orientation gradients in two directions only, so it can fill five of the nine components of Nye's dislocation density tensor; the four that need a gradient normal to the surface stay unknown. On a realistic mixed curvature that costs about a third of the dislocation content. See it worked out on KAM, GNDs & the Step Size →
Three steps, and the page runs all three. A Monte Carlo of the beam in the tilted sample says how many electrons come back out, from what depth and with what energy: that is the source. A many-beam Bloch-wave calculation over directions on a sphere says how strongly the crystal channels electrons leaving in each direction: that is the master pattern, a property of the crystal and the energy alone. The detector geometry and the orientation then decide which part of that sphere lands on the phosphor. Use it to see what a phase should look like before you index it, to check a pattern centre, or to see how much of the pattern a given detector actually catches. Simulate a pattern →
Because the simulation leaves out everything that blurs a real pattern: the detector’s point spread and binning, the finite probe, the background from electrons that scattered inelastically on the way out, and any damage or contamination on the surface. The band positions and the band widths should match; the fine detail between bands, and the contrast, will not. That is the useful comparison to make: if the bands sit where the simulation puts them, your pattern centre and phase are right, whatever the contrast looks like. Compare band positions →
