Blank Loading: Locate, Read the Code, Then Clamp in 20 s
Cylinder head, crankshaft and block blanks are heavy, arrive loose in bins and are handled repeatedly through machining. If a blank is the wrong model or clamped the wrong way round, the loss is settled downstream after machining: one blank scrapped, possibly a tool as well, and a stretch of downtime. This cell puts error proofing before clamping. A 2D vision system absorbs incoming position variation, then code reading verifies model and orientation on the spot, so a wrong or reversed blank cannot enter the next process. The cycle is 20 seconds.
A wrong blank is paid for downstream
For error proofing, the earlier the interception, the cheaper it is. Find the wrong model after machining and the workpiece already carries labour and tool cost. Find the wrong orientation at assembly and the rework chain is longer still. This cell therefore verifies neither after machining nor before assembly, but before the robot clamp closes, and the verification result directly decides whether the clamp action executes.
Position and identity are separate problems
Bulk arrival creates two independent difficulties. First, attitude is arbitrary, so a robot on a fixed program grabs air or off-centre. Second, similar models look alike. One is a spatial problem, the other identity and orientation.
Using vision to solve identity asks an algorithm to separate near-identical blanks, and boundary samples guarantee a steady misjudgment rate in volume. Using code reading to solve position is worse: the code is read, but where the part goes is unknown. One method per problem keeps this cell light.
Localization turns offset into a robot correction
A 2D camera images the blank, locates its actual position in space and converts that through coordinate transformation into a correction for the robot clamping path. The robot clamps from the actual pose rather than relying on manual placement. The criterion is direct: how far and in which direction the position is off are geometric quantities, with no shape recognition involved, so reliability and maintainability are both good.
Why model and orientation are verified by code
Similar blank models look alike, and image comparison cannot avoid boundary samples. Code reading is discrete and definite: the code content is compared with the current machining task, pass if it matches, block if it does not. The cost is that a readable code must exist on the blank, and that precondition must be confirmed at project start rather than assumed.
Locate first, verify second
Order matters. Reverse it and the blank is still skewed when the code is read, so the read fails or misreads and verification loses meaning. Localization first brings the blank into an operable state, then identity and orientation are verified, and the result drives the clamp. The robot cell schedules the whole sequence, automatically linking recognition to clamping.
Unattended means the verification needs no human patch
The inspection cycle is 20 seconds, matching the loading and clamping rhythm, subject to site acceptance results. Vision localization and code verification replace manual handling and visual confirmation, and error proofing runs automatically, so loading no longer depends on shift patterns or human fatigue and can run unattended around the clock. A 2D camera and a code reader are mature components, so retrofit demands on an existing robot loading cell are low and payback is quick.
Questions about moving the check forward
What if the blanks carry no code?
Code verification assumes a readable code. For parts without one, either add marking upstream or use another identity method; that step cannot be assumed away.
How accurate must localization be?
Accurate enough for the robot to clamp, not metrology grade. The task is to produce a correction and absorb position error, judged by clamping success rather than coordinate error.
Can this prevent missing parts?
It prevents wrong model, wrong orientation and position error. Whether a part is present or how many were fitted is a different judgment and needs separate inspection items.
Why intercept rather than warn?
A warning depends on a human response, and people tire. Interception is enforced by the process, and the value of error proofing is that the error never reaches the next operation.