™ CuDDI Copper Digital Detection Imaging. Reasons for Development One of the last… Remove BIAS...

Post on 03-Jan-2016

215 views 1 download

Tags:

transcript

CuDDICopper Digital Detection Imaging

Reasons for Development

• One of the last… • Remove BIAS from Operator to Operator• Digital Recording of Images• LIMS Integration• Standardization on Light Source• Magnification of Sample Strip• Strip Size Determination• Current Chart – NOT REAL• Automatic Measurement• A better rating…

What Color Is it?

Design

*Patent Pending

Vision Algorithm• Define Strip Area –

Automatically• Inspection Area

Determination• 8x8 Pixel scan, ~96,000

points total• Accumulate Results• Calculate Percentage of

each Rating• Pass values through

Support Vector Machine and Algorithm for Categorizing

• Ability to “Train” and Adopt Model

Step 1)Locate Strip and verify dimensions:Must be > Nominal – (Tolerance * 2)e.g. Height = 75mm +/- 5mm, so minimum height must be greater than 65mm. *account for angles

Step 2) Location results are used to define the color inspection region. The region is always slightly smaller (5-20 pixels) than the actual strip, to avoid any unwanted shadows and/or round edges.Black rectangle represents the inspection region.

Step 3)Segment entire search region into small sections (8 by 8 pixels each), and run each sample through the Classifier.About 46,000 total samples per strip.Results are accumulated.

ClassifierDetermine

s which rating the

sample belongs

to.

Extract Color from Sample

Totalize results

Step 4) Calculate percentage of each and determine the most corrosive and the most dominant result using both sides of the strip, and provide analytics:• Highest Ranking• Most Corrosive Ranking• A breakdown (in percentages) of each side of the strip

Rank

Hue Saturation Intensity Image

1a 29.5 Compare 242.3

Compare 225.7

Ignore HDR Dark

1b 42.8 Compare 252.4

Compare 231.5

Ignore HDR Dark

2a 9.7 Compare 172.8

Compare 141.9

Ignore HDR Med.

2b 9.6 Compare 150.9

Compare 191.5

Ignore HDR Med.

4a 17.8 Ignore 51.7 Ignore 113.0

Compare

HDR Med.

4b 18.1 Ignore 76.5 Ignore 61.5 Compare

HDR Med.

4c 187.1

Ignore 41.2 Ignore 41.1 Compare

HDR Med.

The ClassifierGiven a set of trained models, color samples are compared against Hue, Saturation, and Intensity.

Allows multiple samples to be trained.

A portable Color Model Library (saved in xml), allows for different color libraries to be used.

Result = 1a

Satu

rati

on

255

0

0

1a

Hue 360

1b

2a2b

Color Model Library

Data Output• PDF Reports• Email• LIMS Output• Raw Images• Log to Network

Share• USB• Print

Manual 1b 1b 4c 2c 2cCuDDI 1a 1b 4b 3a 3a

Dim. 13.11 x 73.70mm

12.76 x 73.24mm

12.64 x 76.95mm

11.25 x 71.61mm

11.78 x 74.96mm

Data 0 – 2%1a - 89%1b – 7%2e – 2%

1a - 65%1b – 26%2e – 7%

3b – 1%4a – 2%

4b – 93%

0 – 1%2a – 10%2c – 20%2e – 23%3a – 34%4a – 9%

2a – 1%2b – 6%2c – 31%3b – 37%4a – 23%

Notes Over-Rated 4a – 4b Common Missed Green

Data

Sample ID: Manual  CUDDI

1 1B 1B

2 3A 2B

3 2A 2A (Highest)3B (Most Corrosive)

4 1A 1A

5 2E 2E (Highest)4A (Most Corrosive)

6 4C 4C

7 2B 4A

8 1B 1B (Highest)2A (Most Corrosive)

9 2C 2C

10 4A 4A

11 3B 3B

12 4C 4A

13 2A 2A

14 2C 2C (Highest)4A (Most Corrosive)

Manual Rating CuDDI Method

Human Eye Digital CMOS Camera

BIASED Removes BIAS

Variable Lighting Fixed Standardized Lighting

Decision Making Fixed Algorithm*

No Digital Record Logs Results with Images

Flat Viewing Tube or Hands Innovative Holder for Strip

Differences

• Lighting• Fingerprints• Dimensions• Proper Polishing• Holes in D1838• Water Spots – LPG Samples• 2a to 3a Logic Handling • 2c to 3b Logic Handling

Critical Points

• Round Robin• D130• D1838• D4048• EN & IP Methods

• Demo Units Available• ANNEX in Current Methods?• Alternate Procedure? … lighting only?• Feedback from Committee Groups• Request

Plans & Future

CuDDI

by:

Juan J Ayala+1 (312) 622-7520

juan@ayalytical.com

THANK YOU! QUESTIONS?