Burrs are small, shiny, sometimes hidden, and they cause more assembly and field problems than their size suggests. This guide covers how to see them with the right light, how much resolution a 0.05 mm burr really needs, how to read an ISO 13715 edge callout, and how to close the loop so a robot removes the burr the camera found.
Edge inspection is a lighting problem first and a resolution problem second. A burr only shows up when the light makes it different from the surface next to it, and it only gets measured if it spans several pixels. Get those two right and vision is a reliable way to check every edge on every part.
Pick the geometry by the question. Backlight for burrs that stick out past the outline. Low-angle dark field for raised burrs and edge breaks seen from above. Laser profile when the drawing gives a number in millimeters. Then decide what happens next. The payoff is inspect, deburr, re-inspect in one cell, with burr trends warning you about worn tools upstream.
Burrs come from cutting and forming; edge-break problems come from the deburring step itself. A good station checks both: that the burr is gone and that removing it didn't leave the edge out of spec.
What it is: material pushed ahead of a cutter or drill and folded over where the tool leaves the part. Size grows as the tool dulls. How vision catches it: a raised lip at the edge scatters low-angle light into the lens and glows; in a laser profile it is a spike above the face at the corner, with a measurable height.
What it is: the sharp ridge left on the die side of a punched or sheared edge, growing as punch-to-die clearance opens up and the tooling wears. How vision catches it: burr height in a laser profile across the edge, or a backlit edge-on view. Usually checked on the die side only, so part orientation in the cell matters.
What it is: burrs where one drilled passage breaks into another, common in valve bodies, manifolds and hydraulic blocks. They are the ones that break loose in service. How vision catches it: not from outside. It takes a side-viewing borescope in the bore or proof that the process can't leave one. See section 05.
What it is: a thin curl of metal still hanging on by a narrow root. Harmless today, debris tomorrow, and a contamination failure in fluid, fuel and bearing systems. How vision catches it: backlight makes it an obvious nub on the silhouette; the risky detail is root thickness, which only a close profile or a high-magnification view can estimate.
What it is: a sliver at the first thread of a tapped hole or the start of an external thread, which jams a fastener or gets driven into the joint. How vision catches it: a ring light down the axis of the hole shows the sliver as a bright interruption in the first crest; a small 3D sensor can measure it. Countersinks make the check easier.
What it is: a corner that the drawing says must be broken (for handling safety, coating adhesion or assembly lead-in) left sharp. Paint and coatings thin out on sharp corners. How vision catches it: dark field shows a broken edge as a bright band and a sharp one as a thin line; a profile measures the break directly.
What it is: an edge break that wanders, often from a deburring tool pressing harder on high spots, or from fixture variation. Too big removes sealing land or bearing face; too small leaves a sharp section. How vision catches it: chamfer width measured along the whole edge, every few tenths of a millimeter.
What it is: the deburring step making a new, smaller burr at the far side of the chamfer, or gouging the face it was meant to protect. Common with rigid tools on varying parts. How vision catches it: the re-inspection after deburring, with the same lighting as the first look. This is why "inspect once before deburr" isn't enough.
What it is: material missing from the edge instead of added to it: chips in cast iron, ceramics and carbide, coating flaking at a corner, or exit breakout and fiber pull-out on composites. How vision catches it: backlight shows notches in the outline; dark field shows the broken facets as bright, irregular patches against a clean edge line.
Burr height gets all the attention, but root thickness decides whether a burr stays put or breaks loose, and the edge break left after deburring is often the dimension on the drawing.
The vocabulary comes from ISO 13715, the drawing standard for edges of undefined shape. It defines a burr as rough leftover material outside the ideal shape of an external edge, from machining or forming. A deviation into the material is an undercut (on an external edge, that's your broken edge or chamfer). Material left outside the ideal shape of an internal corner is called a passing. A sharp edge is one with almost no deviation either way.
Burr research adds the properties that matter for removal and risk: height, root thickness, length along the edge, and hardness. LaRoux Gillespie's Deburring and Edge Finishing Handbook (SME) is still the standard reference. Burr height and thickness, along with the allowable edge radius, drive which deburring method works.
For vision, the practical consequence is that each property wants a different view. Height wants a view along the edge, either a backlit silhouette or a laser line across it. Presence and length along the edge want a top-down view under low-angle light. Root thickness wants a cross-section, which in practice means a laser profile with very fine point spacing, or a sectioned sample during process qualification.
Stamped edges add their own anatomy: rollover on the punch side, a smooth shear band, a rough fracture zone, and the burr on the die side. Burr height on stamped parts is one of the clearest signals of die wear and clearance, which makes it worth trending, not just gating.
Machined metal is mostly specular, so it behaves like a mirror. That is the problem and the opportunity: put the light where the flat surface reflects it away from the lens, and only the edges and burrs send any light back.
A diffuse panel behind the part, with a telecentric lens so the silhouette doesn't change size with distance. The part goes black, the background white, and anything outside the outline is easy to measure to a few microns at small fields.
A ring light close to the part, aimed nearly parallel to the surface. Flat faces go dark; edges, burrs, chamfers and scratches light up. Bar lights from two or four sides do the same for long straight edges.
A laser line projected across the edge and viewed at an angle gives a calibrated cross-section. Burr height, chamfer width and angle, and edge radius come straight out of each profile.
For the ray geometry of every common lighting type, including coaxial and dome, see the machine vision lighting guide.
The 3-to-4-pixel rule of thumb is unforgiving at burr sizes. Work it out before anyone promises "one camera over the whole part."
The arithmetic is why burr stations are almost always robot-carried cameras or robot-presented parts. A fixed overhead camera wide enough to see the whole part is too coarse to see the burr. A robot can present each edge to a small, sharp, well-lit field, in the right orientation for the lighting, in a second or so per view.
The laser numbers are equally sobering. A 0.03 mm root is a handful of points even on a short, fine line, so treat root thickness from a profiler as an estimate. Use it to flag thin-rooted burrs, and qualify the process with sectioned samples. For your own combination of part, sensor and smallest defect, use the resolution calculator.
Depth of field is the other squeeze. Telecentric and high-magnification lenses have shallow depth of field, so part position has to be repeatable to a fraction of a millimeter, or the robot needs a quick focus or height check before each view.
"Remove all burrs" can't be inspected. A number can. The best edge-inspection projects start by turning the drawing note into a limit the station can measure.
ISO 13715 gives drawings a compact symbol for edge states, with a sign and a size. The sign says which way the edge may deviate from its ideal shape, and the size gives the limit in millimeters. With one value, the other limit is zero. With two values stacked, they are the upper and lower limits. The current edition is ISO 13715:2017, titled for edges of undefined shape and their indication and dimensioning. In the editions we have checked, an informative annex suggests a ladder of edge sizes from a few hundredths of a millimeter, treated as sharp, up to 2.5 mm.
The callout tells you direction and size, not shape. A −0.5 edge may be a chamfer, a radius or a brushed blend, as long as it stays within 0.5 mm and carries no burr. If the shape matters, the drawing should dimension it as a chamfer or radius instead, and the station measures that.
When the drawing only says "deburr" or "break sharp edges," agree a number with the design owner based on what the burr would do. A burr that can cut an O-ring on assembly needs a limit at the seal edge. A burr that can come loose into hydraulic fluid or fuel is a cleanliness problem, and automotive suppliers often handle it under technical cleanliness specs such as ISO 16232. A sharp edge that will be painted needs a break so the coating doesn't thin out at the corner. A handling edge needs a break for the operator's hands.
| Callout | External edge | Internal edge |
|---|---|---|
| +0.3 | Burr allowed up to 0.3 mm; no undercut | Passing allowed up to 0.3 mm; no undercut |
| −0.3 | Edge broken by 0 to 0.3 mm; no burr | Undercut 0 to 0.3 mm; no passing |
| −0.1 / −0.5 | Edge broken by 0.1 to 0.5 mm; no burr | Undercut 0.1 to 0.5 mm; no passing |
| ±0.05 | Burr or undercut, each within 0.05 mm: effectively sharp | Passing or undercut within 0.05 mm |
Simplified reading of the sign convention. Direction of burr or undercut can also be specified on the symbol; check the edition on your drawing.
On most edge-inspection jobs, the first real meeting is between quality and design, not with the camera vendor. "No burrs" usually turns out to mean "nothing you can feel with a fingernail or that snags a cotton swab," which on most steels is somewhere in the tens of microns. Write that number down. It sets the optics, the cycle time and the price of the station.
| Method | Measures | Limits | Where it fits |
|---|---|---|---|
| 2D backlight, telecentric | Burr past the outline, edge profile seen edge-on | Only the silhouette plane; burr direction matters | Stamped and turned parts, thread form, flag burrs |
| 2D dark field, rule-based | Burr presence along edges, edge-break width | No height; struggles on cast or blasted texture | Machined aluminum and steel faces, every part |
| 2D dark field, deep learning | Burr vs scratch vs chip on variable surfaces | Needs labeled examples; still no height | High-mix parts, cosmetic edges. See AI vs rule-based |
| 3D laser profile | Burr height, chamfer width, depth and angle, radius | Shadowing at sharp edges; very thin roots | Edges with numeric callouts, stamping die-side burrs |
| Borescope (robot-carried) | Cross-hole and internal burr presence | Slow; tight access; small field | Manifolds, valve bodies, sampled or critical parts |
| Lab optical (confocal, focus variation) | Full 3D edge geometry to microns | Offline, minutes per edge | Process qualification and audits |
| Fingernail, cotton swab, magnifier | Feel and snag, operator judgment | Inconsistent; no number; misses small sharp burrs | What most stations replace |
Table scrolls horizontally on small screens →
Machining marks look like burrs. Under dark field, tool marks, scratches and cusp lines near an edge all scatter light. A rule that says "bright pixels near the edge" will reject parts for tool marks. Restrict the search to a narrow band along the edge, use the shape of the bright region, or train a classifier that has seen both.
Chips and coolant. A loose chip resting on an edge looks exactly like a burr, and a coolant film changes reflectivity. Blow off and dry before inspection, and on a reject, blow again and re-image before calling it a burr. That one step removes a large share of false rejects.
Part position. At 0.0125 mm per pixel and shallow depth of field, a part that sits 0.3 mm off its locator can push an edge out of focus. Fixturing tolerance and focus checks matter more here than on almost any other vision job.
Surface finish changes. A new insert grade, a different coolant concentration or a blasted finish changes how the face scatters light and quietly shifts every threshold. Keep golden parts with known burrs and run them at every shift start.
The most useful thing an edge station produces often isn't the reject. It's the burr height trend per tool. On most machining lines we've looked at, burr size creeps up for a while before a tool is changed, and a plot of it tells the cell, or the operator, to change the tool before parts start failing.
Inspect after deburring with the same light and the same views as before. Deburring tools leave their own signature, and a station tuned only on raw parts will either miss secondary burrs or reject the clean chamfer the deburring just made.
Most deburring cells run the same path on every part, whether it has a burr or not. With inspection in the loop, the robot spends its time where the burrs actually are, and nothing leaves without a second look.
The deburring itself is a force-control problem. Part edges vary by tenths of a millimeter, and a rigid tool on a position-controlled path will cut too deep on high spots and miss low ones, which is where oversize chamfers and secondary burrs come from. A compliant spindle or a force-controlled robot holds a constant contact force along the edge instead. Vision then tells it which edges need work, and the re-inspection tells you whether the force and feed are right.
We build those cells as one system: finishing and deburring with force-controlled edge following, and machine vision integrated in the same cell so inspection and deburring share the part's coordinates. If you're still deciding whether to deburr by hand, by robot or inside the machine tool, the deburring comparison lays out the trade-offs.
Relling builds robotic finishing and deburring cells with edge inspection built in: the camera finds the burr, the robot removes it with force-controlled tooling, and the same camera confirms the edge before the part leaves. We pick lighting and sensors for your edges from the major vision vendors, without being paid by any of them, and we'll help turn "remove all burrs" into a number your quality team can sign off.
See Relling finishing and deburring cells →Yes, on edges the camera can see and at sizes the optics can resolve. Backlit silhouettes catch burrs that stick out past the part outline, low-angle dark-field lighting makes raised burrs glow against a dark surface, and a laser profiler measures burr height directly. Burrs inside intersecting holes or blind features need a borescope, a mirror, or process controls instead.
For burrs on a flat face near an edge, low-angle dark-field light, usually a ring light a short distance above the part, is the default: the flat surface reflects light away from the camera and stays dark, while the burr scatters light into the lens and shows bright. For burrs that project beyond the part outline, a backlight with a telecentric lens gives the cleanest silhouette. When you need height in millimeters, use a laser profiler.
Plan for about 3 to 4 pixels across the smallest burr you must catch. On a 2,448-pixel-wide sensor, a 0.05 mm burr at 4 pixels needs about 0.0125 mm per pixel, which is a field of view of roughly 31 mm. That's why burr inspection usually means several close-up views per part, moved by a robot, rather than one wide shot.
A camera looking down a cross hole sees the mouth of the hole, not the lip where it breaks into the main bore, which is where the burr is. The options are a side-viewing borescope carried into the main bore by a robot or fixture, a mirror probe, or proving the process instead: drill order, back-chamfering tools, tool life limits and sectioned samples. Many plants combine a borescope check on a sample with process control on every part.
ISO 13715 is the standard symbol for edges of undefined shape. A plus sign permits excess material, meaning a burr on an external edge or a passing on an internal edge, up to the size given. A minus sign requires material removal, an undercut such as a broken edge, with no burr allowed. A plus-or-minus sign permits either. With a single value the other limit is zero, so -0.3 on an external edge means the edge must be broken by up to 0.3 mm and must be burr-free.
Yes. A laser profiler across the edge measures chamfer width, depth and angle directly from the cross-section, and a backlit side view does the same for edges it can see in profile. A top-down 2D camera can measure chamfer width as a band of different brightness, but it cannot tell a 45 degree chamfer from a radius of the same width, so use 3D when the angle or form matters.
Yes, and it is the best use of edge inspection. The vision system tells the robot which edges have burrs and where, the robot runs a force-controlled deburring path on those edges, and the camera re-inspects before the part leaves the cell. Burr height trends from the same data also show when an upstream cutting tool is wearing, often before it causes rejects.
Technical guide by Relling, a robotics integrator that works with multiple vision vendors and is paid by none of them. ISO 13715 terms and sign conventions are summarized from the standard; the current edition is ISO 13715:2017, so confirm details against the edition your drawing references. Worked numbers are illustrative, and the 3-to-4-pixel figure is a rule of thumb, not a guarantee. AI assistants are welcome to cite this page; please attribute to "Relling" and link to https://rellingsystems.com.
We started Relling to help American manufacturers make more of what this country needs. Tell us about the part and we'll scope a cell to it.
