+86-159-2181-9772 | ben.liu@ulilaser.com

Calibrating CCD Vision Alignment for Laser Soldering Systems

16 11 月, 2026 · 行业动态

Calibrating CCD vision alignment on a laser soldering system means establishing four transforms in a fixed order — the camera's own lens model, the pixel-to-machine (or pixel-to-galvo) mapping, the camera-to-laser-spot o…

Calibrating CCD vision alignment on a laser soldering system means establishing four transforms in a fixed order — the camera’s own lens model, the pixel-to-machine (or pixel-to-galvo) mapping, the camera-to-laser-spot offset, and the feeder-nozzle offset — then proving all four with a traceable target and repeated measurements. Before a machine is released to production you should see a maximum field residual of about 10 µm and static repeatability of ¼ pixel or better. Skip any layer and the symptom appears later as solder balls placed off-pad, one-sided wetting on a joint, or intermittent “vision fail” alarms on fine-pitch parts.

Calibrating CCD Vision Alignment for Laser Soldering Systems

What Does CCD Vision Alignment Calibration Actually Involve?

Alignment is not a single “calibrate” button. On a laser soldering platform it is a chain of five corrections, and each one fails differently. Treating them separately is what makes a calibration reproducible instead of a lucky result.

Layer What it corrects Typical method Residual target
Intrinsic (lens model) Focal length, pixel scale, principal point, radial and tangential distortion Dot grid or chessboard captured at 5–9 poses and tilts < 0.1 px reprojection error
Pixel-to-machine (extrinsic / hand-eye) Translation, rotation, scale, skew; galvo DAC nonlinearity if a scan head is used 25–81 marks covering ≥80% of the field; affine for gantry, 2nd–3rd order polynomial for scan heads ≤ 10 µm max residual across field
Camera-to-laser-spot Where the heat actually lands relative to what the camera sees Low-energy pulse on an aim coupon, spot centre measured under a toolmaker’s microscope ≤ ¼ of the spot diameter
Feeder / nozzle offset Where the solder ball, paste dot or wire tip is deposited Vision measurement of a gauge pin or a test deposit ≤ ±15 µm
Focus / Z map Working distance and magnification change over the board Focus sweep at 5–9 board positions Sharpness within the lens depth of field

The extrinsic layer deserves the most attention, because it absorbs errors from the others. On gantry-style platforms a two-parameter-per-axis affine fit is usually enough; on scan-head platforms the field distortion is intrinsic to the F-theta lens, so a polynomial fit with marks near the corners is mandatory — a 3×3 grid will look perfect in the centre and be 40–60 µm out at the edges.

Which Target, Lens and Lighting Setup Gives You a Calibration You Can Trust?

  • Lens choice: prefer a telecentric objective. With a conventional lens, magnification changes with working distance, so a 0.2 mm height difference on a connector body converts directly into a scale error. Telecentric optics hold magnification across roughly ±0.3–0.5 mm of Z, which is usually enough for a populated automotive board.
  • Resolution: aim for 3–8 µm per pixel so the smallest pad or fiducial spans at least 10 pixels. Below that, sub-pixel centroiding supplies the rest, but only if contrast is stable.
  • Target: chrome-on-glass or ceramic with 0.5–2 mm marks on 5–20 mm spacing, covering at least 80% of the field, flatness under 10 µm. Mount it at the same height as the product, or add a Z compensation term — a plate sitting 1 mm higher than the board is a classic source of “good on the plate, off on the product”.
  • Lighting: coaxial for specular pads and fiducials, dome or ring diffuse for matte surfaces, backlight for transparent or thin substrates. Freeze exposure and gain during calibration and lock them into the recipe. Auto-exposure that drifts between calibration and production shifts the centroid and quietly degrades accuracy.
  • Environment: calibrate at 20 ± 2 °C after a 30–60 minute warm-up. A cold machine moves by tens of microns in the first hour.

How Do You Run the Calibration Step by Step?

  1. Mechanical and thermal preparation. Clean the target and the lens window, verify the fixture is seated and torqued, confirm the Z height with a gauge, then let the machine idle through its warm-up cycle.
  2. Intrinsic calibration. Capture the grid at multiple tilts. Reject any pose with a reprojection error above 0.15 px rather than averaging a bad set into the result.
  3. Pixel-to-machine fit. Teach 25–81 marks across the full field, including the four corners and the mid-edges. Run the fit, then inspect the residual map — not just the RMS number. A low RMS with a hot corner means the model is wrong, not that the machine is good.
  4. Aim check. Fire a low-energy pulse at five positions spread across the field, measure spot centre against the mark centre under a microscope, and enter the vector offset. Repeat until all five are within a quarter of the spot diameter.
  5. Feeder offset. Bring the nozzle or wire tip into the camera field, measure its centre, and store the tool offset. Recheck after any nozzle change — a replacement nozzle is rarely concentric to the old one.
  6. Save and version. Store the calibration as a named, dated file tied to the lens, nozzle type and product recipe. Record who ran it and the residual figures, so a later drift can be compared against a baseline rather than a memory.

Why Does Alignment Drift, and How Do You Troubleshoot It?

The pattern of the error tells you which layer is wrong. Measure the offset at the centre and at two opposite edges of the field before touching any setting.

Symptom Most likely cause Corrective action
Same offset everywhere in the field Translation term, camera/spot offset, or thermal drift after warm-up Re-run the aim check; confirm warm-up time; verify fixture seating
Error grows linearly from the centre outward Scale error — wrong pixel size or calibration plate at a different Z Verify plate height and magnification; re-fit with the correct Z
Error rotates around the field centre Theta or rotation term, often after a head or fixture swap Re-fit rotation with marks far from the centre
Near-zero centre, large error at corners Lens distortion or an under-fitted polynomial Add marks at the corners and edges; move to 3rd order
Error changes with component height Parallax on a non-telecentric lens Add a Z-dependent scale term or switch to telecentric optics
Scatter that changes hour to hour Thermal drift, loose target, flickering illumination Log room temperature, re-torque the target, lock light output
Intermittent “fiducial not found” Contrast, threshold, or paste/mask contamination over the mark Re-teach the pattern, adjust exposure, keep the pad keep-out clear

What Acceptance Checks Should Be Signed Off Before Production?

  • Static repeatability: measure one fiducial 30 times without moving; 3σ should be ≤ ¼ pixel or ≤ 5 µm, whichever is tighter for the application.
  • Field accuracy: residual ≤ 10 µm at every mark, and ≤ 1/5 of the smallest pad width on the actual product.
  • Thermal stability: repeat the centre measurement hourly for four hours; drift should stay inside 5 µm.
  • Gage R&R: run three operators × ten parts × three trials on a real board; under 10% is ideal, under 30% is acceptable for solder placement.
  • First-article verification: solder a golden board, cross-section two joints, and confirm the ball or paste deposit is centred on the pad with symmetric fillets on both sides.

A daily two-minute check — one mark, five repeats, logged — catches most drift before it becomes scrap. ULiLASER systems store this check as a one-button routine in the alignment recipe, so the operator records a number rather than a judgement. Recalibrate fully after any lens, nozzle, head or fixture change, and otherwise on a fixed interval your process data supports.

FAQ

How often should I recalibrate the CCD alignment? Full calibration after any hardware change and typically every 3–6 months; the daily single-mark check covers everything in between. If the daily log shows drift beyond 5 µm twice in a week, move the full calibration forward.

Can one calibration cover all products on the machine? No. The intrinsic model can be shared if the lens and Z are unchanged, but the pixel-to-machine fit and the tool offsets should be versioned per recipe, because nozzle length, board height and fixture stack-up all shift the effective transform.

Why is placement perfect on the calibration plate but off on the real board? Almost always a Z mismatch — the plate is thicker or thinner than the product, so magnification differs. Second most common is a lighting difference that moves the fiducial centroid. Check both before re-fitting the model.

Does closed-loop temperature control compensate for alignment error? No. Temperature control keeps the joint from overheating or cold-starting, but it cannot move a misplaced solder ball onto the pad. Alignment accuracy and thermal profile are independent loops and must be validated separately.

What pixel resolution do I actually need? Enough that the smallest fiducial or pad spans at least 10 pixels — 3–8 µm per pixel covers most automotive and semiconductor work. Higher resolution helps only if the mechanics and lighting are stable enough to use it.

If you share your pad pitch, component height variation and field size, our engineers can map out the specific calibration tolerances and check routines for your application.