INDUSTRY

Machinery / Equipment

Machined parts, castings and forgings, welded assemblies and assembly lines: DaoAI ACI — auto cognitive inspection — covers surface defects, dimensional and geometric tolerance, weld formation and assembly errors. High-mix low-volume lines model from one good part, so a changeover does not mean redrawing inspection boxes.

Surface and dimensional inspection of machinery parts
Back to industries

Machining, casting and forging, welding, assembly — one vision system across all four.

  • Dimensional and geometric tolerance: key dimensions and position measured in line, without taking the part off the line.

  • Surface defects: burrs, scratches, dents, blowholes and shrinkage — judged as departures from normal, so the list of shapes need not be complete.

  • Weld formation: undercut, lack of fusion, spatter and misalignment, judged on 3D topography together with 2D appearance.

  • Loading and sorting: featureless parts in a random bin, with 6D pose driving the pick directly.

Machinery & equipment, in detail

1. What is actually hard on this line

Machinery manufacturing is normally high-mix, low-volume with frequent changeovers: one part today, another tomorrow. Last-generation AOI indexes its rules to coordinates — boxes, thresholds and templates all sit on this one part's geometry — so every changeover means doing it again. The list of defect shapes never ends either: burrs, scratches, dents, blowholes, shrinkage, cold shuts, folds, undercut, lack of fusion. A system trained on defects it has seen is blind by construction to the one it has not.

2. How the solution is built

What ACI changes is the basis of the decision. DaoAI models what "normal" looks like from one good part first, then judge where the part in front of it departs — no CAD, no defect library. 2D appearance covers scratches, dents and characters; structured-light 3D topography covers depressions, missing material, weld reinforcement and flatness. When it calls something wrong, the review goes back in — see feedback learning. Thresholds and rules stay yours throughout, and an update never overwrites what you tuned by hand.

3. How to choose a machine

Sampling, incoming re-judgement and high-mix low-volume go to offline models; inspection at line takt goes inline — see 2D ACI equipment and 3D ACI equipment. If you already have cameras and a line, connect ACI OS directly and embed it through SDK / API / Docker. Stations that need a robot to pick and place are covered in robot vision.

4. How a rollout is paced

Start with one station — whichever changes over most often, or whichever has the highest cost of escape. Model from good parts, run a batch of real incoming material, feed one round of review back, and watch whether false calls fall as review accumulates. None of this requires changing the line, and none of it requires deciding what to buy first.

5. How the result is measured

Three quantities you already hold: rework and claims from escapes, engineering hours spent re-judging false calls, and hours spent reprogramming at changeover. You do not have to wait for our numbers — the arithmetic and the variables are at AOI machine cost and ROI.

Send us the part and we will tell you whether it can be done.

No purchase decision needed first. Send a set of good parts and a few typical rejects; we will model from them and show you the calls together with the rules behind them.