NIWeek Vision Summit

jabgoldfishΤεχνίτη Νοημοσύνη και Ρομποτική

19 Οκτ 2013 (πριν από 3 χρόνια και 9 μήνες)

92 εμφανίσεις

1

NIWeek Vision Summit

August 2
-
3, 2011 Austin, Texas



www.ni.com/niweek/summit_vision

2

Breaking New Ground with
Vision Inspection Systems

Dr. Dan Milkie

Senior Developer

Coleman Technologies, Inc.

NIWeek Vision Summit

August 2

3, 2011 Austin, Texas

What if there’s
no existing solution
?


Prototyping tips



Planning guide



Implementation



Imaging algorithms


Finding small defects


3D laser triangulation



Final lessons

3

About us


16 year National
Instruments
Alliance Partner



Basler
Vision System Integrator



NI Certified
Developers &
NI Professional
Instructors



Advanced
engineering & science
degrees

4

Vision Applications

5

Industrial

Research


Dinnerware
defect detection system


Dinnerware
color pattern inspection system


Robotic
seed germination classifier


High
-
speed seed counting system


Laser
drilling inspection


Thin
film defect identification


Mirror
defect detection


PCB
contact pad inspection system


Color
tablet tracking


Particle
size analysis
(powders)


Particle
size analysis
(liquid suspension)


Glass
rod inspection


Well
plate inspection system


Bio
-
sample
thermal imaging


Wellplate

imaging system


Crystal
finder/classifier


Biaxial tissue tester


Two
-
photon microscopy

Plate Inspection System

The Challenge

Dinnerware Inspection


Many different dishes



1 dish per second



Most difficult defect:


White bumps on

white plates

6

Defects

Can we image the defects?


Prototype with what you have
(or loaners)


Area
-
scan camera (GigE)


Lighting


Directional : Desk lamp


Broad sources : Room lights, diffusers


NI Vision
Development
Module



Optimize lighting & camera
placement


Tip

: Replace camera with your eye


7

Okay,
I
see the defects, but can the
computer?

8


Start with NI Vision
Assistant


Fastest way to test
processing functions


Estimate time using the
“Performance Meter”


Quickly turn scripts into
LabVIEW VIs


Okay,
I
see the defects, but can the
computer?


Start with NI Vision
Assistant


Fastest way to test
processing functions


Estimate time using the
“Performance Meter”


Quickly turn scripts into
LabVIEW VIs


9

Proof of Principle


Are you confident in your plan yet?


If not, prototype in LabVIEW!


Combine acquisition and analysis


Tip

: Use tester with fresh data
or

saved
image
sets



Practice good coding style


Prototypes


Final version


Good code encourages trying new ideas


Documentation should be automatic

10

Going for it

Ready to build your system?


Know your
test
set, criteria



Balance goals with sliding scales

instead of all
-
or
-
nothing



Solve
most pressing
issues first


Example

: Our
1
st

generation machine
tested 1
plate type
(their most popular) and only 4
defect
types

11

Time

$$

Accuracy

Modular, Modular, Modular

Independent Stations

1.
Bottom view

2.
3D
imaging

3.
Top view




Lesson learned :


Needed to add baffles


Found interference between
stations


Changed how plates cross gaps


Vibration issues

12

NI PCIe
-
1430

Dual
CameraLink

NI PCIe
-
8235

4x GigE

PC

1

2

3

Small defect imaging


First version :




Next version:




Defects show up
as dark spots:

13

LED

light

Glancing Angle

Transmission

Small Defects are Hard to Find!


Problems :


Defects are small, low contrast


Large gradients in image


Plate
-
to
-
plate variations


Image size (8MP!)


Must process in < 0.5 second



Solution:


A custom pixel
-
by
-
pixel threshold
for each image.

14

Defect!

Step 1 : Unwrap Image

Problem : We just need rim pixels


Use image masks?


Longer process times


Still have large images


Solution : IMAQ Unwrap


8M pixel square
-
> 1.5M pixel strip


Aligns gradients, rim transitions in one
direction

15

Step 2 : Create Custom Golden Master

What should this image look like if it were perfect?



Remove
defects
using median smoothing


Tip

: Use X Size >> Y Size to preserve gradients & edges


Performance Boost
:
Reduce image resolution
, Smooth,
then
Resample back to original size
.

16

Unwrapped Original

Median Smoothed

“Golden Master”

Defect

Defect Removed!

Step 3 : Pixel
-
by
-
pixel Threshold

17

“Golden Master”

Results

Subtract a

constant to set

a “lower”

threshold.

Original

Defect Found!

Threshold
original image
using IMAQ
Compare

Benefits of Pixel
-
by
-
pixel Thresholds


Works with:


Gradients, edges, speckle, large dynamic contrasts



Every image checks against itself


Robust against image
-
image variations, changin
g

lighting conditions



Fast
(<100ms for 1.5MP)


Limited by smoothing
performance

18

3D Imaging


Simple inspection for geometric errors


Gouges, bulges, warp


19

Warp examples

Laser Triangulation

20

Laser

Line

Original

Threshold

applied

Convert pixels to height

21

With a little more
geometry, we can also
correct for perspective :

Laser

Line

q

H

dy

Height =

3D Laser Height Measurement

22


Each camera frame = 1 cross section (per laser line)


Tip

: Add multiple laser lines for more cross sections










250 um cross section resolution (100 FPS @ 1” per sec)


Tip

: Reduce ROI for fastest frame rates



100
um height
resolution



Completed system


Fast development with NI tools:


Completed in < 3 months



Reliable


Inspected over 1 million dishes



Multiple follow
-
up systems


23

Conclusions


Invest in good prototyping practices



Prepare for unknown hurdles


Use modular, flexible architectures in hardware, layout,
software



Defect finding algorithm


Pixel
-
by
-
pixel thresholds



Simple 3D laser scanner


Inexpensive, easy to setup

24

Questions?

25

26

NIWeek Vision Summit

August 2

3, 2011 Austin, Texas

Keynotes

Industry Keynote: How NI Technology Powers the Space Elevator

LaserMotive

Tuesday, August
2

1:00
-
2:00

Academic Keynote: Industry Trends and Intelligent Production Systems of the Future

Interdisciplinary Imaging and Vision
Institute, RWTH Aachen University

Wednesday, August
3

1:00
-
2:00

Tuesday, August 2

3D Vision and the Kinect

National Instruments

10:30
-
11:30

Wacky Optical Tricks for Machine Vision

GE Global Research Center

2:15
-
3:15

Panel Discussion: Latest in Camera Technologies

Cyth

Systems

Basler

Vision Technologies

Pleora

Technologies

Toshiba TELI

3:30
-
4:30

Machine Vision and Industrial Robotics: From Design Concepts to Factory Floor Deployment

ImagingLab

4:45
-
5:45

Wednesday, August 3

Precision Metrology

National Instruments

10:30
-
11:30

Is LabVIEW FPGA Right for My Vision Application?

National Instruments

2:15
-
2:45

Autofocus System for an Ellipsometer

Nanometrics Inc.

2:45
-
3:15

Developing a Quality Inspection Method for Selective Laser Melting of Metals Using a High
-
Speed NIR Camera

Katholieke Universiteit Leuven

3:30
-
4:00

Web Inspection of Optical and Medical Fibers

Adsys Controls Inc.

4:00
-
4:30

Breaking New Ground with Vision Inspection Systems

Coleman Technologies Inc.

4:45
-
5:15

Development of a Digitally Multiplexed Bioassay Reader with Magnetic Bead Technology

MoviMED

4:15
-
4:45

27