Placement Vision Centering: What the Camera Sees

Placement vision exists to answer one question quickly: where is the part relative to the nozzle that is holding it? A machine that answers correctly places within a few tens of microns; one that answers approximately places acceptably on coarse parts and fails on a 0.4 mm pitch device, and the difference is usually in what the camera can see rather than in the mechanics.

Why Centering Determines Placement Accuracy

The machine knows where the nozzle is, but not where the part sits on the nozzle, and the offset between the two changes with every pickup. Centering measures that offset and applies it as a correction before the part is placed. Without it, the accuracy of the machine is the accuracy of the feeder and the pickup combined, which is far worse.

The correction has to be made in both position and rotation. A part that is picked slightly off centre is also rotated relative to the nozzle, and the resulting placement error grows with distance from the nozzle axis. For a large package the rotational error dominates, while for a small chip the positional error does. A rotational error of half a degree moves the corner of a 30 mm package by more than 0.1 mm, which is enough to push a ball off its pad.

Line Scan Versus Area Cameras

A line scan camera images the part as it passes over the sensor, building a picture from the movement of the head, and it is fast enough to keep up with high placement rates. An area camera takes a full frame in one exposure and needs the head to stop, which is slower but gives a complete image without motion blur.

The choice affects which defects can be detected. A line scan image is built from one row at a time, so a lead that is present but poorly lit may be missed, while an area camera captures the whole part at once and shows the true shape. Machines built for speed generally use line scan, and the inspection capability is part of what the speed costs. Where a machine offers both, the area camera is often reserved for parts with fine leads or for first article verification, while production runs on the line scan.

Component Recognition and Defeats

Recognition works by finding the features the algorithm was taught to look for: leads, balls, a body outline or a corner. A part defeats recognition when a feature is invisible, when something else looks like the feature, or when the tolerance expected is smaller than the variation in the part itself.

The common defeats are a dark body on a dark background, a lead frame that reflects the light like a mirror, and a nozzle that appears in the image and is mistaken for a lead. Each produces either a false correction or a rejection, and the remedy is a change in lighting, nozzle or algorithm rather than in placement pressure.

Lighting and Contrast for Leads

Lead recognition depends on contrast between the metal and the background, and the lighting is designed to create it. Backlighting through the gap between the leads gives a silhouette that is easy to threshold, while a reflective top light produces highlights that vary with the surface finish of the leads.

Where the finish changes between suppliers, the optimum lighting changes with it. A component that recognized reliably with one finish may fail with another that reflects differently, and the change is often reported as a placement problem because the machine rejects the part rather than placing it badly. Keeping a golden sample for each critical part and re-teaching from it after a supplier change is a cheap way to keep recognition stable.

Nozzle and Part Interaction

The nozzle is part of the image, and a tip that is too small allows the part to sit off-axis while a tip that is too large obscures the features the algorithm needs. A worn or contaminated tip changes the apparent edge of the part and introduces a systematic offset that appears as a placement error.

Suction also affects what the camera sees. A part that is slightly lifted by a partial vacuum leak sits at a different height, which changes the magnification and therefore the measured size. Cleaning tips and verifying vacuum levels is part of keeping centering accurate, not only part of keeping pickup reliable.

Teaching a New Component

Teaching a new part means selecting the recognition type, defining the search area, setting the tolerance and choosing the lighting, and the result should be verified by placing parts and measuring them rather than by accepting the first success on the machine display. A recognition window that only just succeeds is a source of intermittent failures.

Where a part has no suitable feature, a body-outline recognition with a generous tolerance is often better than forcing a lead-based algorithm. The decision should be recorded with the part number so a later change of supplier does not silently invalidate the setup. A recognition setup that is not documented is re-invented by the next engineer, usually in the middle of a production problem.

Verifying Centering Accuracy

Centering accuracy is verified by placing parts on a calibrated substrate or a glass plate and measuring the offset from the programmed position, in both axes and in rotation. The measurement should be taken across a batch, because a single placement can be correct by coincidence while the distribution is wide.

The results feed the placement accuracy figures for the machine and show whether centering is contributing to the error. Where the mean offset is non-zero the cause is systematic, such as a fiducial error or a nozzle offset, while a wide distribution points to centering or pickup variation.

Common Failures and Their Signatures

A systematic offset on one head points to a nozzle or camera calibration issue. An offset that appears only on one part type points to the recognition setup. An intermittent offset on all parts points to pickup, and a rotational error that increases with part size points to a centering correction that is being applied at the wrong radius.

Reading the signature is faster than adjusting settings at random. The fiducial system should also be checked, because an error there is applied to every part on the board and looks like a centering problem on all of them.

When Vision Cannot Help

Vision cannot correct a part that is damaged, a lead that is bent, or a body that is out of tolerance, and it cannot fix a placement error caused by the board moving after alignment. Where the board is not clamped and supported, the fiducial alignment is valid only at the moment of measurement, and the placement that follows inherits the movement.

It also cannot see a joint, so a placement that is visually correct may still produce a poor joint because the paste was missing or the pad was contaminated. The inspection step that follows is what covers that case, and the division of labour between placement and inspection should be explicit rather than assumed. An accurate placement over a short deposit produces the same defect as a misplaced part, and only the inspection data separates the two.

FAQ

What does vision centering actually measure? The offset between the part and the nozzle holding it, in position and rotation, which the machine then applies as a correction before placing the part.

Why does a part fail recognition when it looks fine to the eye? Lighting and contrast. A dark body, a mirror-like lead frame or a nozzle entering the image can each defeat an algorithm that depends on a specific feature.

How is centering accuracy verified? By placing parts on a calibrated substrate or glass plate and measuring the offset across a batch, in both axes as well as in rotation, rather than judging a single placement.