Transcript Document

Near Infrared Spectroscopy
for biomass studies
OVERVIEW
•
•
•
•
1. About the Center NIRCE
2. NIR spectroscopy on biomass
3. MSPC + an example
4. Offline mixtures
OVERVIEW
•
•
•
•
1. About the Center NIRCE
2. NIR spectroscopy on biomass
3. MSPC + an example
4. Offline mixtures
NIRCE 2002-2003
Biofuels Umeå
Biofuels Vasa
Forest seeds Umeå
Calibration Umeå
Medical and Optical Vasa
Short courses
NIRCE 2004-2006
NIRCE ONLINE
NIRCE IMAGE
NIRCE CLINICAL
What do we offer?
Graduate courses and short
courses
Research projects
Advice and consulting
Method development
Instrument pool
Workshops and symposia
NIR2007
OVERVIEW
•
•
•
•
1. About the Center NIRCE
2. NIR spectroscopy on biomass
3. MSPC + an example
4. Offline mixtures
Non-food
Biomass
Food & feed
Bioenergy
Pulp and paper
Forestry
Building materials
Textiles
Consumer products
Feed and safety
Where is biomass found?
• Biotechnology
• Natural products
• Bioenergy
What is special about
biomass?
•
•
•
•
•
•
O-H
C-H
N-H
C=O
different atom sizes = good
IR+NIR energy = movements of
bonds
O
H
O
H
H
O
O

H
H
H


H
H
Near Infrared Spectra (NIR)
Cosmic Gamma Xray Ultraviolet Visible NIR Infrared Microwaves
• 780-2500nm
• Suitable for all organic and bio materials
• Robust for industrial use
• Good penetration depth
• Many modes of measuring
• Powerful multivariate results
Near Infrared Spectra
• Fast
• Simple sample preparation
• Nondestructive
• Online for process applications
• Need for calibration
• Opportunity for data analysis
OVERVIEW
•
•
•
•
1. About the Center NIRCE
2. NIR spectroscopy on biomass
3. MSPC + an example
4. Offline mixtures
NIR for Process
Monitoring in Energy
Production by Biofuels
Tom Lillhonga
Swedish Polytechnic
Vasa, Finland
[email protected]
Paul Geladi
Head of Research
NIR Center of Excellence
Umeå, Sweden
[email protected]
Alholmens Kraft
•
•
•
•
•
Worlds largest biomass-fuelled power plant
Fuels: biofuels, peat and coal
Almost 1 km2 of storage
Furnace is 15 ton sand fluidized-bed
One 20 ton truck every 5 min.
www.alholmenskraft.com
A reminder
Problem definition
• Biofuel consumption: 750-1000 m3/h
• Large variations in moisture content
• Moisture determination off-line is very
slow and not valuable for process
monitoring
 Unwanted variations in steam and
electricity production
 Reduced competitive strength
Controls
z1
zJ
x1
Inputs
y1
Output(s)
Industrial
process
xK
yM
y(t) = F[x(t),z(t)]
y(t) = F[x(t),z(t)]
• F should be known
• x(t) should be known
• z(t) set by operators
Inside
Ambient temperature -25 to +25
Dust
Humid
Steam and compressed air
Heavy equipment
Sampling and measurements
• Samples were collected manually from
a conveyor belt (at line)
• A digital photo was taken of every
sample
• NIR-spectra at-line
• Reference samples analysed off-line by
industrial standard 17h@105°
Sampling and measurements
• Measurements were done during summer of 2003
• Samples were collected manually from a conveyor
belt (at line)
• Sample temperature was measured
• A digital photo was taken of every sample
• Grinding was tried (Retsch Mill SM2000)
• NIR-spectra at-line
• Reference samples analysed off-line by industrial
standard
Foss NIRSystems 6500
grating instrument (Direct Light)
2 Si
4 PbS
71 W
13 cm
λ0
monochromator grating
5 cm ø
Det
Det
Integrating
sphere
Det
Fiberoptic
Fiberoptic
Mirror
Process NIR spectrometer based on moving
grating
Dataset
•
•
•
•
NIR-spectra, 400-2500 nm, every 2 nm
All spectra averages of 32 scans
Calibration set: 160 samples
Test set: 61 samples
Spectra of calibration set (+3 outliers)
Milled samples
PCA-model
• All calculations are done with MATLAB
6.5 and PLS_Toolbox v. 2.1 and v. 3.0
• Identification and removal of outliers
• Clustering observed
Score plot of PCA-components 1 and 2
Series start
Moisture, %
Sample moisture (replicates with red)
Sample number
Moisture histogram
PLS-model
• Pre-treatment of spectra
- noisy wavelengths removed
(2300-2500 nm)
- smoothing and second derivative
calculated with Savitzky-Golay method
• Mean-centred spectra
• NIPALS- algorithm and cross validation (venetian blinds)
used
• RMSECV = 2.6 % for 7 components
Percent Variance Captured by PLS-Model
LV #
1
2
3
4
5
6
7
8
9
10
-----X-Block----This LV
18.09
19.52
41.02
1.728
2.118
1.138
0.788
1.008
0.688
0.498
Total
18.09
37.61
78.63
0.35
2.46
3.59
4.38
5.38
6.06
6.55
-----Y-Block----This LV
Total
45.48
45.48
17.75
63.23
3.91
67.14
10.07
77.21
4.76
81.97
4.06
86.02
3.96
89.98
1.90
91.88
1.75
93.63
1.54
95.17
Loading-plot for PLS-component 1
water peaks
Diagnostics for PLS-model
Moisture,
%
5
4.5
RMSECV = 2.6 %
for 7 components
4
3.5
3
2.5
2
RMSEC
1.5
1
0.5
1
2
3
4
5
6
7
8
9
10
11
PLS Comp.
Predicted vs. measured moisture of
calibration set
65
Y Predicted (moisture-%)
60
55
50
45
r2 = 0.85
40
35
35
40
45
50
55
Y Measured (moisture-%)
60
65
PLS-predictions on test set
Moisture,
%
75
* = lab
70
o = NIR pred.
65
60
55
50
45
40
35
30
25
0
10
20
30
40
50
60
Sample number
Acknowledgements
Stig Nickull
Sari Ahava
Sten Engblom
Bo Johnsson
Johanna Backman
Morgan Grothage
Standard deviation for replicates
Replicate
sample
numbers
Standard deviation
for
five replicates, %
Standard deviation for
PLS predicted values
of replicates, %
1
0.86
0.95
2
0.99
3.52
3
1.07
3.17
4
1.14
not calculated
5
1.84
not calculated
6
2.25
not calculated
Future experiments
• Off-line measurements on fuel
mixtures (H2O, ash, energy)
• Improved sampling probe
• Seasonal effects?
• Temperature
• Time series analyses
• On-line measurements
• Model included in process monitoring
OVERVIEW
•
•
•
•
1. About the Center NIRCE
2. NIR spectroscopy on biomass
3. MSPC + an example
4. Offline mixtures
Off-line work
•
•
•
•
•
•
At SYH
CD 128l InGaAs 900-1700nm
Integrating sphere with lamp
Large glass plate
Mixtures
Linda Reuter of Wismar Polytechnic
Simplex mixture design
Coal
1/0/0
0.5/0/0.5
0.5/0.5/0
0.33/0.33/0.33
0/1/0
0/0/1
Peat
0/0.5/0.5
Biofuel
Coal
Peat
Biofuel
+H2O
H2O x 3
Mixing
(remixing)
10x
NIR spectrum
32 scans
Ash x 3
Energy x 3
110x128
11x128
33x128
Average reference values
moisture, energy, ash, spectra
all 10 replicates
Average spectra and
average reference values
Individual references values
and average spectra
Figure 10
110x128
11x128
Table 3: RMSECV results (in parentheses number of components used)
Data set
Moisture %
Energy MJ/kg Ash %
110S
0.94 (14)
0.39 (8)
2.1 (12)
11S
2.3 (5)
0.63 (4)
5.6 (5)
33S
1.8 (7)
0.83 (6)
2.6 (8)
Conclusions
• Max bias / variance
-moisture 1.8%/ 3%
-energy 0.5 / 0.75 MJ/Kg
-ash -5 / 7 %
• Reference replicates important
• Spectral replicates important
Works well
• Design repeated in score plot
• Classification possible
• Within run error smaller than between-run error
• PLS prediction H2O, ash, energy