Zoom Lens for Connector Pin Inspection: FOV, Sampling and Focus

A connector is rarely inspected at one magnification. The same camera has to frame a 20 mm housing to check for a bent latch, then resolve a 0.5 mm pitch row of pins to see whether one contact is displaced. A zoom lens is the usual answer, but the choice is not "pick the highest magnification". It is an arithmetic question: what field of view does each inspection task need, how many pixels fall on the tolerance band, and what happens to focus when the pins do not sit at the same height as the housing floor. This page works through those three questions with the published specifications of the POMEAS zoom families, and it states clearly where a 2D zoom lens stops being the right instrument.

What a connector inspection cell actually has to measure

Connector inspection splits into two families of tasks that pull the optics in opposite directions. The first family is presence and pose: is every pin present, is any pin bent, lifted or pushed back, is the latch intact, is the housing free of flash and short shots. These checks are answered from a wide view, they tolerate coarse sampling, and they benefit from seeing the whole part in one exposure. The second family is dimensional: pin width, pin pitch, row-to-row spacing, wall thickness, true position of a pin relative to a datum. These need enough pixels per tolerance band to make a measurement repeatable, and they are usually measured on a small region of interest.

A single fixed lens forces a compromise between the two. If it is chosen for the dimensional task, the housing no longer fits in the frame and the machine needs a second station or a moving axis. If it is chosen for the presence task, the dimensional measurement is sampled too coarsely to hold a tight tolerance. Zoom optics exist to remove that compromise: the magnification becomes a programmable parameter of the recipe rather than a property of the hardware.

Two limits deserve to be stated at the start, because they are the most common source of disappointment in connector projects. First, a zoom lens measures in the image plane. Coplanarity, pin height and solder-joint height are out-of-plane quantities; they need a height sensor, a laser profile or a structured-light head, and no amount of magnification substitutes for that. Second, pin material is usually plated metal. Specular reflection, not resolution, sets the practical edge quality on a bright pin tip, which is why the lighting section below matters as much as the optical one.

Why variable magnification fits connector families

Connector part numbers multiply faster than inspection stations. A single product family will run 0.4 mm board-to-board pitches next to 2.54 mm headers, and the same line will be re-tooled for a new housing every few months. A zoom lens turns that variation into a recipe change. The 12.5X series, for example, covers 0.58X to 7.5X optical magnification with a published horizontal field of view from 15.17 mm down to 1.17 mm on a 2/3 inch sensor: the wide end frames a small housing, the narrow end frames a few pins at a time. The branded 12.5X figure is the series name; the published range 0.58X to 7.5X spans about 12.9:1, which is the number that matters when you calculate how much a single lens can cover.

That range is also why zoom lenses appear in connector work as a replacement for a lens turret. A turret gives discrete magnifications and needs a mechanical move plus a re-homing sequence between them. A continuous zoom changes magnification without leaving the optical axis, and a motorized version changes it from the PLC or from the inspection recipe. For a line that inspects several connector variants, the practical gain is changeover time: recall a recipe, wait for the zoom and focus to settle, and re-run the calibration for that magnification.

The trade-off is that every magnification carries its own working point. Resolution, depth of field and distortion are not constant across the range, so a connector recipe that is validated at 2X is not automatically valid at 7.5X. The tables below list the published figures per series, and the sections after them explain how to convert those figures into a decision for a specific connector.

Published specifications of the zoom families used for connector work

Series (linked record)	Magnification	Working distance	Max sensor	Resolution	Depth of field	TV distortion
PMS-Z125D-CH12 12.5X C-mount	0.58X–7.5X	62±2 mm to 74±2 mm	2/3″	13 µm to 3.05 µm	2.75 mm to 0.05 mm	0.001% to 0.050%
LZL-12101 12.5X coaxial	0.58X–7.5X	77.4±2 mm	2/3″	down to 3.05 µm	not stated	down to 0.001%
LZ-650104 6.5X motorized	0.7X–4.5X	82±2 mm	2/3″	down to 3.95 µm	not stated	down to 0.001%
LZG-63104-A 6.5X rail, motorized	0.7X–4.5X	87±2 mm	2/3″	down to 3.95 µm	1.9 mm to 0.1 mm	0.001% to 0.035%
PMS-LZH-7505W 4K	0.68X–5.0X	80 mm	1″	10.17 µm to 2.8 µm	1.78 mm to 0.07 mm	<0.02%
PMS-LZH-14105K 14X	0.55X–7.6X	77.8±2 mm	2/3″	14.59 µm to 2.80 µm	not stated	not stated

Source: published parameter blocks of the linked product records (PMS-Z125D-CH12, LZL-12101, LZ-650104, LZG-63104-A, PMS-LZH-7505W, PMS-LZH-14105K). "Not stated" means the record does not publish the figure; it is not zero and it is not an estimate. Where a series publishes only one end of a range, only that end is quoted.

Read the table as a map of which series can serve which connector task, not as a ranking. The 4K series carries a 1 inch image circle and a wider 0.68X–5.0X range, which suits large housings and 4K cameras but stops at 5X; the 12.5X series reaches 7.5X for single-pin detail but covers a 2/3 inch sensor only; the 14X series reaches furthest and starts widest of the high-magnification group. The 6.5X group sits in the middle with a longer working distance (82 mm or 87 mm), which is often the deciding factor when the lens has to clear a fixture wall.

Sampling arithmetic: from field of view to micrometres per pixel

Every dimensional claim in a connector recipe reduces to one number: how many micrometres of the part fall on one pixel at the chosen magnification. The calculation needs only the field of view and the number of pixels across it. Using the published horizontal field of view of the 12.5X series and a 2448-pixel horizontal count — a common 5-megapixel configuration for a 2/3 inch sensor, given here as a worked example so you can substitute your own camera — the numbers fall out as follows.

Magnification	Horizontal FOV	µm per pixel	Pixels per 0.5 mm pitch	Pixels per 40 µm tolerance band
0.58X	15.17 mm	6.20	81	6.5
1.0X	8.80 mm	3.59	139	11
2.0X	4.40 mm	1.80	278	22
4.0X	2.20 mm	0.90	556	44
7.5X	1.17 mm	0.48	1044	84

Derived: FOV per step interpolated linearly from the published 15.17 mm at 0.58X and 1.17 mm at 7.5X (field of view scales as 1/magnification); µm per pixel = horizontal FOV ÷ 2448; pixel counts = 500 µm or 40 µm ÷ µm per pixel. Calculated values, not published specifications.

The last two columns are the ones that decide a project. A 0.5 mm pitch is generous: even the wide end puts 81 pixels on it, which is more than enough to count pins and see a displaced one. A 40 µm tolerance band — roughly what a ±0.02 mm pin-width or true-position callout implies — gets only 6.5 pixels at 0.58X and 11 pixels at 1.0X. That is the difference between a measurement and an opinion.

A working rule that keeps quantisation error near 10% of the band is to require at least ten pixels across the tolerance band. Dividing the band by ten gives the coarsest acceptable sampling, and the magnification follows from the table above. For a 40 µm band that means at most 4 µm per pixel, which is reached at 1.0X and comfortably met above 2X. For a 100 µm band the same rule allows 10 µm per pixel, so the wide end of the zoom is already sufficient and only the presence checks need the low magnification. This rule is an engineering guideline, not a published specification; the real limit also depends on the edge-detection algorithm, the lighting and the contrast of the feature.

Resolution, distortion and the position error at the frame edge

Optical resolution sets the smallest feature the lens can render at all; sampling sets how finely that image is divided into pixels; distortion displaces the image of a feature away from where a linear model puts it. The 12.5X series publishes 13 µm at the wide end and 3.05 µm at 7.5X, with TV distortion between 0.001% and 0.050%. For connector work the resolution figure is rarely the binding constraint at 2X and above — a pin edge is a high-contrast step more than 100 µm wide — but distortion is worth a calculation, because it acts on absolute position, which is exactly what a true-position callout measures.

Distortion is specified as a fraction of distance from the image centre, so its effect grows towards the frame corner. At 0.58X the published diagonal field of view is 18.97 mm, so the half diagonal is about 9.49 mm; 0.050% of that is roughly 4.7 µm of apparent displacement at the corner. At 7.5X the diagonal shrinks to 1.47 mm and the same percentage works out to about 0.4 µm. Both corrections are small, but the wide-end figure is the same order as a tenth of a tolerance band on a 0.05 mm callout, which is why a connector recipe that measures true position should either calibrate distortion into the recipe or keep the measured features near the frame centre.

There is a second, cheaper defence: measure the same feature set at the same frame position for every part. Distortion is a fixed pattern in the image plane, so a recipe that always places pin 1 at the same coordinates cancels most of the error without any calibration model. What it does not cancel is part-to-part variation in presentation, which is why the fixture, not the lens, usually decides whether the measurement repeats.

Depth of field versus pin height

Connector pins and housings are not coplanar. The published depth of field of the 12.5X series runs from 2.75 mm at 0.58X down to 0.05 mm at 7.5X, and the 6.5X rail series runs from 1.9 mm to 0.1 mm. At the wide end, a whole connector sits comfortably inside the depth of field. At 7.5X, a 0.05 mm window is narrower than the height difference between a pin tip and the housing floor on most connectors, so a single exposure cannot hold both in focus.

Three strategies work, and they differ in cost rather than in elegance. The first is to accept a single compromise plane at the midpoint of the height spread and to keep the measurement on features that live near that plane — pin width at mid-height, for example, rather than the pin tip. The second is to take two exposures at two focus positions, which requires a motorized focus axis and a recipe that knows which features belong to which plane. The third is to hold the part on a fixture that presents the measured features in one plane and to inspect the rest at the wide magnification, where depth of field is not a constraint. In practice most connector cells use the third strategy for the housing checks and the first for the pins.

Focus also drifts for a less obvious reason: thermal growth between the lens and the part. A cell that is validated cold and run warm will lose focus at 7.5X long before it loses it at 0.58X, because the depth of field shrinks by more than fifty times across that range. A periodic autofocus or a reference-target re-check at the production magnification is cheaper than scrapping parts.

Illumination: coaxial, low-angle and ring

Connector pins are plated metal, and plated metal is a mirror. A coaxial lens, which injects light along the imaging axis and returns only what reflects straight back, gives a flat, evenly lit top surface: good for pin tips that face the camera, poor for the side walls of a pin. A low-angle ring emphasises edges and standing features, which is what makes a lifted pin visible, but it also produces shadows inside the housing that can look like contamination. A diffuse dome removes the specular hotspots but flattens the height cues.

The published coaxial variants in the zoom families are the practical starting point when the features of interest face the camera: the 12.5X coaxial series carries a Ø25 mm coaxial light port, and the 6.5X coaxial variant shares the motorized body with the non-coaxial model. Both accept a ring or a bar light on the front thread when the recipe needs edge emphasis instead. Because the illumination has to move with the magnification — the same light that fills a 15 mm field at 0.58X will overexpose a 1.17 mm field at 7.5X — a connector recipe should store light intensity per magnification rather than per part.

One detail catches most first builds: at high magnification the working distance is measured from the front of the lens, and a coaxial illuminator or a polariser added in front of it changes both the working distance and the light path. Re-check the focus after adding any front optics, and re-run the pixel-size calibration if the magnification setting is reused with a different front configuration.

Manual, detented or motorized: choosing the control

The optical choice is only half of the decision; the other half is how the magnification gets set. A manual zoom with a locking screw is the cheapest option and is entirely adequate when a cell inspects one connector variant at one magnification for months. A detented zoom gives repeatable mechanical positions — the 12.5X detented model publishes 15 positions — which lets a recipe recall a magnification without an encoder, at the cost of being limited to those positions.

A motorized zoom with an RS-232 interface moves the choice into the recipe. The 6.5X motorized model publishes a 9-pin RS-232 interface, closed-loop control and a DC servo motor; the 12.5X family offers both single-motor and dual-motor configurations, where the dual-motor version separates zoom and focus so the two can be commanded independently. Independent focus is what makes the two-exposure strategy above practical, and it is also what allows a slow autofocus search without moving the magnification.

For a connector line that runs several part numbers, the motorized route usually pays for itself in changeover time, provided the integration handles three things: the settling delay after a move before the first image is grabbed, the position feedback that proves the move completed, and the per-magnification calibration that the recipe must carry. The PLC integration path and the single-motor versus dual-motor comparison are covered in the linked guides below.

Integration checklist for a connector cell

Before a connector recipe is released to production, eight items are worth checking in this order. Each one has a failure mode that shows up as a mysterious measurement error later.

1. Field of view per task. List every inspection task on the connector and the field of view it needs; confirm each one falls inside the published range of the chosen series, with margin for fixture tolerance.

2. Pixels per tolerance band. Convert every dimensional callout into pixels using the sampling table and confirm the ten-pixel rule, or justify a lower count explicitly.

3. Frame position of measured features. Decide where in the frame the dimensional features sit, and keep that position identical across parts so distortion acts as a constant.

4. Height spread versus depth of field. Measure the height difference between the highest and lowest feature that must be in focus at the production magnification, and compare it with the published depth of field at that magnification.

5. Illumination per magnification. Store light intensity and exposure per magnification, not per part number, and re-validate after any front optic is added.

6. Settling and feedback. For a motorized zoom, define the settle time before the first acquisition and read back the position to confirm the move completed.

7. Calibration per magnification. A single pixel-size calibration does not transfer across the zoom range; calibrate at each magnification the recipe uses, and re-calibrate after a lens swap or a front-optic change.

8. Out-of-plane features. List the features that are actually height measurements and confirm they are handled by a height sensor, not by the zoom lens. This is the item most often skipped and most often responsible for a "the vision system cannot measure coplanarity" complaint.

Where a zoom lens is the wrong instrument

Three connector tasks should be routed elsewhere. Coplanarity and pin-height measurement need a height channel: a laser displacement sensor, a confocal sensor or a structured-light head, because a 2D image has no depth information to give. Fiber-optic connector end-face inspection is a different optical problem with its own magnification and lighting conventions, and it is covered separately on this site. Electrical continuity, Hi-Pot and signal integrity are test-instrument tasks; optics can find a bent pin but cannot tell you whether the contact resistance is in specification.

Within the 2D tasks, the honest limit is the tolerance. A zoom lens on a 5-megapixel camera can hold a 0.05 mm callout with margin and a 0.02 mm callout with care; below that, the constraint moves to the sensor, the sub-pixel edge algorithm and the fixture, and a higher-magnification lens alone will not fix it. If the connector roadmap includes 0.01 mm true position, plan for a metrology-grade station rather than a zoom-lens cell.

Selection path in one page

For a housing-level presence and pose check, start at the wide end: 0.58X or 0.68X, where a 15 mm to 18 mm field covers the part and depth of field is measured in millimetres. For pin-level dimensional work, start at 2X and verify the pixel count against the tolerance band before committing. For single-pin detail, 0.5 mm pitch and below with tight callouts, use 4X to 7.5X and accept the shallow depth of field as a fixture problem rather than an optical one. If the cell must clear a tall fixture wall, prefer the 6.5X group with its 82 mm to 87 mm working distance over the shorter 12.5X group.

Two further guides shorten the rest of the work: the configuration guide covers how zoom, auxiliary lens and objective combine when the required field of view falls outside a single range, and the machined-parts page covers the same one-camera-many-magnifications argument for a different part family. For line integration, the PCB inspection and PLC integration pages cover the recipe and handshake side.

Related reading

Product records used for the figures above: PMS-Z125D-CH12 12.5X C-mount zoom lens, LZL-12101 12.5X coaxial continuous zoom lens, motorized 12.5X zoom lens, LZL-12100D15 detented 12.5X, LZ-650104 6.5X motorized, LZG-63104-A 6.5X rail, 4K 0.68X–5.0X zoom lens and 14X continuous zoom lens.

Further reading: machine vision zoom lens configuration guide, zoom lens for machined parts, PCB AOI lens upgrade: variable FOV, motorized zoom lens and PLC integration, single-motor versus dual-motor zoom lens, coaxial, low-angle and RGB lighting and auxiliary lens guide: 0.5X–2X.