Transcript The Unscrambler® A Handy Tool for Doing Chemometrics Prof. Waltraud Kessler Prof. Dr.

The Unscrambler®
A Handy Tool for Doing Chemometrics
Prof. Waltraud Kessler
Prof. Dr. Rudolf Kessler
Hochschule Reutlingen, School of Applied Chemistry
Steinbeistransferzentrum Prozesskontrolle und Datenanalyse
Camo Process AS
Topics
• The Unscrambler® by Camo
• Many possibilities for Analysing Data
• Examples
• NIR-Spectra
• Fluorescence Exitation Emission Spectra
• Life Demonstration
• 3-way Data Handling
2
The Unscrambler Main Features
®
Exploratory Analysis
Descriptive statistics
Principal Component Analysis (PCA)
Multivariate Regression Analysis
Partial Least Squares regression (PLS)
Principal Component Regression (PCR)
Multiple Linear Regression (MLR)
Prediction
Classification
Soft Independent Modeling of Class Analogies (SIMCA)
PLS-Discriminant Analysis
Experimental Design
Fractional and full factorial designs, Placket-Burmann,
Box Behnken, Central Composite, Classical mixture
designs, D-optimal designs
ANOVA, Response Surface ANOVA, PLS-R
3
The Unscrambler Also Features…
®
•
•
•
•
•
•
Raw data checks
Data preprocessing
Over 100 pre-defined plots
Automatic outlier detection
Automatic variable selection
… and more
4
Example: Fiber Board Production
In-situ Measurements of Fibres in Blowpipe
NIR FOSS Process Spectrometer
with fibre bundle and
diffuse reflectance probe
400 - 2200 nm
Blowpipe:
~ 180°C
~ 5 bar
~ velocity of fibres ~ 20 m/s
5
Fiber Board Production
NIR-Spectra of Fibres in Blowpipe
Spectra contain the
following information:
• kind of wood
• fineness
• degradation of lignin
Information is hidden within complete wavelength range
Information overlaps – separation by PCA
6
Principal Component Analysis
Separate the Overlapping Information
mRf
ScoresmRf
mRf
PC2
mRf
PC2 = Fineness
coarse
fine
mRf
0.05
0
oRf
oRf
oRf
oRf
oRf
oRf
oRf
oRf
oRf
oRf
oRf
oRf
oRf oRf
oRf
oRf
oRf
-0.05
-0.10
RESULT 8,X-expl : 72%,24%
oRf
oRf
oRf
oRf
oRf
oRf
oRf
oRf
oRf
oRf
oRf
oRf
oRf
oRg
oRg
oRg
oRg
oRg
o Rg oRg
oRg
oRg
oRg
oRg
oRg
oRg
oRg
oRg
oRg
oRg
oRg
oRg
oRg
oRg
oRg
oRg oRg oRg
oRg
oRg
oRg
oRg
oRg
oRg
oRg
oRg
oRg
-0.05
0
PC1 = kind of wood: Spruce
mRf mRf
mRf
mRf
mRf
mRf
mRfm Rf
mRf mRf
mRf
mRf
mRf
mRf
mRf
mRf mRf
mRf
mRf
mRf
mRf
mRf
mRf
mRf
mRg
mRf
mRf
mRg
mRg
mRg mRg
mRf
mRf
mRg
mRg
mRg mRg
mRg
mRg
mRg
mRg
mRg
mRg
mRg
mRg
mRg
mRg
mRgmRg
mRg
mRg
mRg
mRg
mRg
mRg
mRg
mRg
mRg
mRg
mRg
mRg
mRg
mRgmRg
PC1
0.05
0.10
Spruce with bark
7
Principal Component Analysis
Scores and Loadings for PC1 and PC2
X-loadings
X-loadings
0.1
0.1
0
0
-0.1
X-variables
500
RESULT 9, PC(X-expl):1(72%)
1000
1500
2000
-0.1
X-variables
500
RESULT 9, PC(X-expl):2(24%)
PC1:Kind of wood
1000
1500
2000
PC2:Fineness
Scores
Spruce
with bark
Sc ores
0.05
0.05
fine
0
0
Spruce
-0.05
coarse
-0.10
-0. 05
Sampl es
06:21:49 _13.11.
11:43:52 _13.11.
RESULT 2, PC(X-e xpl ):1(95%)
07:05:02 _14.11.
14:15:56 _14.11.
Samples
06: 21:49_13. 11.
10: 19:54_13. 11.
RESU LT5, PC(X-expl):
2(24%)
14: 37:28_13. 11.
08: 47:24_14. 11.
14: 06:48_14. 11.
8
PLS Regression
Degradation of Lignin for Spruce
Scores
PC2
0.02
0.01
2.2
2.2
2.2
2.2
0
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2.5
2.5 2.5
2.5
Predicted Y
3.0
2.9 2.9
2.9
2.9
2.92.9
2.92.9
2.9
2.9 2.9
Elements:
Slope:
Offset:
Correlation:
RMSEP:
SEP:
Bias:
2.7
2.9
2.4
-0.01
30
0.910025
0.237915
0.941680
0.085069
0.086517
0.000981
2.9
-0.02
PC1
-0.06
0
-0.03
0.03
X-variance
RESULT 16, (Y-var, PC): (SFC,2)
Residual Sample Variance
0.04
0.0000015
0.03
0.0000010
0.02
0.5E-06
0.01
0
Samples
10
RESULT 16, PC:2 2
3.0
2.7
2.4
2.1
0.06
RESULT 16,X-expl: 80%,5% Y-expl: 88%,6%
0.0000020
Measured Y
2.1
20
Y-variance
Residual Sample Variance
0
Samples
10
30
30
20
RESULT 16, PC:2 2
9
Analysing Three-Way Data
od
e3
L
M ode 1
M
Two different types of modes
are distinguished:
• Sample mode - O
• Variable mode - V
I
M ode 2
K
Sample mode -usually first mode
Variable mode -usually second and/or third mode
OV2 or O2V
10
Substructures in Three-way Arrays
K vertica l slice s
L f ro nta l slice s
I hor izo ntal slice s
Three-way arrays can be divided into different slices
Decide which slices are put together to form a two-dimensional array
11
Three-way Data Example:
Fluorescence Excitation Emission Spectra
• Samples: 32 fibres from steam treated and ground woodchips
• X-Data: Fluorescence Excitation-Emission spectra
(250 - 575 nm) x (300 - 600 nm)
• Y-Data: Kind of wood (beech and spruce)
Severity of treatment (a combination of time and temperature)
Age of wood (fresh and old)
Plate gap of grinding ( fine and coarse).
12
Three-way Data Example:
Fluorescence Excitation Emission Spectra
Beech
Spruce
Treatment: low
middle
severe
13
Fluorescence Excitation Emission Spectra
Results of N-PLSR
14
Fluorescence Excitation Emission Spectra
x1 and x2 Loading Weights
15
Possibilities for Three-way Data in
The Unscrambler®
• 3D Data Import: ASCII, Excel, JCAMP-DX, Matlab
• Swapping: toggle freely between the 6 OV2 and 6 O2V
layouts of a 3D table
• Matrix plots: Contour and landscape plots of the samples
• Variable sets: Create Primary variable sets and
Secondary Variables sets
16
The Unscrambler® Benefits
•
•
•
•
Easy to make models
Easy to interpret results
High user-friendliness
Less time spent doing data analysis,
more information extracted from your data
• Faster decision making
17
Try The Unscrambler® 9.2 for 30 days
Free trial version available on www.camo.com
 Fully functioning version
 Includes the Unscrambler user manual
 Includes 7 tutorial exercises and associated files
 Includes 3 demonstration tours
For details, contact:
CAMO Software India Pvt. Ltd.,
14 -15, Krishna Reddy Colony, Domlur Layout,
Bangalore - 560071, INDIA
[email protected]
18