When screw hole position tolerance is tighter than what a fixture can locate, or when workpieces arrive with variation inside the fixture, a purely mechanical automatic screw machine will produce off-centre fastening, stripped threads and missed screws. CCD vision positioning photographs the workpiece before fastening to recognise Mark points or hole positions and compensates the coordinates sent to the motion system, which makes it a common solution for high-accuracy fastening. This article sets out the accuracy budget and the calibration methods for a vision-guided screw machine.

1. What makes up the positioning accuracy

The total error of vision-guided fastening is not a single figure but the sum of several contributions:

1. Camera resolution and field of view (FOV): pixel equivalent = FOV size ÷ number of pixels. The larger the field of view, the larger the real size represented by one pixel and the lower the theoretical accuracy. A common approach is to increase camera resolution while still covering the required field of view, or to use stitched images.

2. Calibration error: this covers camera intrinsic calibration and camera-to-mechanism hand-eye calibration. Poor calibration carries the vision error into the motion coordinates proportionally.

3. Mechanism repeatability: the repeat positioning accuracy of the X/Y/Z ball screw or timing belt modules (typically ±0.02 mm to ±0.05 mm) sets the ceiling for vision compensation — if the mechanism cannot reach the position, perfect recognition is of no use.

4. Recognition algorithm and lighting: reflections, oil contamination and coating differences all affect the stability of circle fitting.

2. How to carry out camera calibration and hand-eye calibration

Intrinsic calibration: use a chessboard or dot calibration target to determine focal length, principal point and distortion coefficients and remove the edge error caused by lens distortion. This step decides whether accuracy is consistent at the centre and at the edge of the field of view.

Hand-eye calibration: establish the mapping from image coordinates to mechanism coordinates. A common method is to drive the mechanism to nine or more known coordinates with a calibration target or feature point, capture images and record the pixel coordinates, then solve the transformation matrix by least squares. Calibration points should cover the whole working field of view rather than being concentrated in the centre.

Verification: after calibration, randomly pick 5-10 points for a reverse check and recalibrate if the deviation exceeds expectations. Recalibration is mandatory after large ambient temperature changes, an impact to the camera or lens, or a lens change.

3. Lighting and fixture considerations

Stability of vision positioning depends largely on image quality:

· Prefer coaxial or ring lighting and avoid dependence on ambient light;
· For reflective metal holes, use backlighting or low-angle light to highlight the hole edge;
· Fixtures should guarantee a repeatable workpiece attitude; use vision only to compensate incoming variation, never to make up for a loose fixture;
· For mixed-model lines, store the Mark points and recognition parameters of each model as a recipe and switch with one action at changeover.

4. When vision is worth it and when it is not

Not every application needs CCD. As a rule of thumb, if hole position tolerance matches the locating capability of the fixture and incoming parts are consistent, a precision fixture is sufficient. Vision positioning clearly pays off when: hole position tolerance must be within ±0.05 mm; workpieces are flexible or thin and easily deformed; incoming batches vary significantly; many models run in small batches on one line; or the relative position of holes and structural parts changes from batch to batch.

5. What to measure at acceptance

Fix the following tests at the acceptance stage: the hole position deviation distribution over 30 consecutive fastenings (look at mean and range, not just the best single result), the first-article confirmation time after changeover, the false recognition and missed recognition rates, and the calibration retention after a power cycle. Putting these figures into the acceptance report reflects long-term stability far better than testing only whether the screw can be driven.

Honred Technology vision-guided fastening solutions

Honred Technology offers vertical vision screw machines (CCD) and panoramic dual-CCD vision inspection machines, with camera-assisted point alignment, Mark point recognition and coordinate compensation, and recipe storage for multiple models. Process parameter advice and sample validation are available for self-tapping threads, thin plates, irregular workpieces and similar conditions. Direct supply from the manufacturer, with support for integration with robot platforms and MES systems.

0755-23170120

Service Hotline