VI User Manual - 第319页

Optimization Vision 2007 4.10 User Manua l Rev 01 12 - 1 Chapter 12 Optimization 12.1 Auto Fine Tune (AFT) The Auto Fine Tune ( AFT) tool is a programming assistant using Vi-Pr o and Histograms fo r models in the library…

100%1 / 412
Production
11 - 18 Vision 2007 4.10 User Manual Rev 01
Optimization
Vision 2007 4.10 User Manual Rev 01 12 - 1
Chapter 12
Optimization
12.1 Auto Fine Tune (AFT)
The Auto Fine Tune (AFT) tool is a programming assistant using Vi-Pro and Histograms for models
in the library. It automatically optimizes the library to obtain:
Increased accuracy and repeatability:
• Vi-Pro uniform scale, X and Y scale optimization.
Fewer false faults:
• Vi-Pro score and contrast threshold.
• Histograms used in presence and polarity equations.
Shorter cycle time:
• Link permutation.
• Unused link deletion.
AFT enables transfer of a Vi-3K Series library to a Vi-5000 Series and vice versa. It also enables
execution of a .tst file in the process mode (debug) on an offline station.
To optimize a library with AFT you will need:
The library to be optimized.
A .tst file with the components to be optimized activated for testing.
Several full and empty panels.
Optimization
12 - 2 Vision 2007 4.10 User Manual Rev 01
12.1.1 Optimized parameters
12.1.1.1 X/Y scale
The X/Y Scale parameters enable adaptation of the X/Y scale ratio of the trained
model to the component’s real image. The optimization process automatically cal-
culates these parameters.
12.1.1.2 Vi-Pro uniform scale
The optimization process refines the uniform scale tolerances considering the new
X and Y scale parameters.
12.1.1.3 Score
The Score is the minimum limit
used by Vi-Pro to determine
whether an object found is a
valid instance of the trained
model. The optimization pro-
cess adjusts the score so that
no components are found vali-
dated on empty boards, and all
the components are found val-
idated on the full boards (if this
is possible).
12.1.1.4 Vi-Pro contrast threshold
The optimization process adapts the contrast threshold to minimize false faults on
full panels. The optimized contrast threshold is the minimum contrast Vi-Pro re-
turned for the model on full panels.
The X/Y Scale optimization has enlarged the trained model to best adapt it to the
real image.
Real image
Trained model
Best model
with respect
to the real image
Trained model
X Scale = 100%
Y Scale = 100%
Trained model
X Scale = 102%
Y Scale = 100%
Optimization X/Y Scale
Real image
Empty
board
results
Full
board
results
Score optimization (ideal case) (%)
0
F
r
e
q
u
e
n
c
y
(
%
)
Optimal score
(threshold
value)
Auto Fine Tune (AFT)