yoed tsur
DESCRIPTION
renewable energyTRANSCRIPT
French-Israeli Workshop on Renewable Energies
Solid oxide fuel cells research Solid oxide fuel cells research
using impedance spectroscopy using impedance spectroscopy Yoed Tsur, Shany Hershkovitz,
and Sioma Baltianski
Department of Chemical Engineering
Technion- Israel institute of Technology, 32000
Haifa, Israel
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
Outline• Appetizer• SOFCs• Impedance spectroscopy• Genetic programming
– A very short introduction– Application to our problem
• ISGP
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
Why are we addicted to fossil fuels?
November 2010, Tel Aviv
X one week(~50 work hours)=
(Concept adapted from David Cahen to fit my own family)
French-Israeli Workshop on Renewable Energies
How Fuel Cells can help?-High efficiency
-Low ‘local’ pollutants
•4November 2010, Tel Aviv
To the grid
French-Israeli Workshop on Renewable Energies
Fuel Cells:Stacks and complexity
•5November 2010, Tel Aviv
Fuel in
French-Israeli Workshop on Renewable Energies
A reminder of linear IS
SignalSupply
Z
V
If
( )( ) where 2( )
,
VZ Z iZ fI
and Z Z comply with KK relations
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
1 2 3 4
3 1 2 4 2 2 1
22 1
( 1, , , ) ( 2, , , ) if:; (1 / )
and 2 1(1 / )
Z Z R R Z Z R RR R R R R R R
Z Z R R
1
0 1
1( ) a iZb b i
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
Distribution of time constants
A system with a finite number of time constants can be put into the following form:
01 1
( ) ( ) where 11
n nk
kk kk
gZ Z R gi
Distribution of time constants is the extension: n → ∞. Then we have:
00 0
( )( ) ( ) where ( ) 11g dZ Z R g di
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
Or in log scaleTaking an arbitrary reference frequency, and
defining:
We get:
0 ( ) ( )
( ) ( )( ) ( )10 10
( , ) ( )
L L L L
L d LZ L Z L R
K L L L
00
log ; log ; ( ) ( )L L L g
0( )( ) ( )
1 10 L L
L dLZ L Z L Ri
And in particular:
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
Equivalent circuits and DFRT
• DFRT=distribution function of relaxation times. (We also call it “the function”).
• In many cases one can find the DFRT from a given equivalent circuit.– An equivalent circuit like this:has a DFRT of two delta functions.– In most real cases: “distributed elements”
should be used -> DFRT with peaks.• Equivalent circuits are not unique.
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
Discrepancy-complexity plot
[Baltianski and Tsur, J. Electroceramis 2003]We take the discrepancy between the prediction of the model and the data (e.g., ) on a log scale vs. the model’s complexity (# of adjustable parameters). Each point represents a solution. Look for the “knee” in this plot.
2
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
Discrepancy-complexity plot
[Baltianski and Tsur, J. Electroceramis 2003]We take the discrepancy between the prediction of the model and the data (e.g., ) on a log scale vs. the model’s complexity (# of adjustable parameters). Each point represents a solution. Look for the “knee” in this plot.
2
Surprisingly amount of info can be inferred from that!
-4
-2
0
0 2 4 6 8 10
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
Avoid Over-fitting• This is one of the most common
mistakes. – It could be done ad absurdum: data set
of n point can be fitted with no discrepancy by a function with n free parameters.
• Our remedy: we use two data sets and give “merit” according to:
1 21f f f
4
3
2
1
0 x
y
1 2 3 4 5 6
4
3
2
1
0 x
y
1 2 3 4 5 6
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
Outline• Appetizer• SOFCs• Impedance spectroscopy• Genetic programming
– A very short introduction– Application to our problem
• ISGP
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
A D CA CA B
How Genetic Programming works
•Mutation •Permutation•Crossover
A B A C
Population Population
11
Population Population
22
Population Population
doublesdoubles
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
Adopting to our case• Pre-knowledge:
– We have decided to look for linear combinations of known peak functions only.
– Simple, no need to check feasibility.• Additional input from the user
– “Expected” and “too high” complexity– What type of peaks to include– Population size and total number of
generations– Normalization, etc.
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
The main loop:• A generation contains N offsprings.• They “breed” and the population is now 2N
– The population is doubled using a pre-determined reservoir of genes.
– Add a peak/ change a peak/ eliminate a peak• The program finds parameters for each new
offspring, and gives it a figure of merit• The best N-1 offsprings plus a randomly
selected one survive, and become the next generation.
• Plotting discrepancy-complexity, Nyquist &
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
Let’s see this again
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
The adaptive pressure
Is achieved by the figure of merit:Compatibility (between 0 and 1) with 2 sets timesPenalty for complexity timesPenalty for not being properly normalized.
0
0 expect
1 2 2
5( )1 exp
0.8 0.2 1 exp (1 )bC n
C
f f
C
f
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
IS measurements of a system contained a IS measurements of a system contained a MIEC and electrodes in air at the MIEC and electrodes in air at the temperature range of 500-600 °C temperature range of 500-600 °C
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
Same sample at 550 °C and varying Same sample at 550 °C and varying oxygen partial pressureoxygen partial pressure
November 2010, Tel Aviv
French-Israeli Workshop on Renewable Energies
Summary• Fuel Cells should be a part of any energy
portfolio• IS – enessential tool to improve them• Discrepancy-complexity plot• The ISGP free program
[electroceramics.technion.ac.il] – Inherently avoid most of the common mistakes that
you can find in literature (over-fitting; what is a “good fit”; “generating” information)
– Can be used both for exploration (new problems) and routinely for systems with a known DFRT shape
– Can also solve Fredholm equations of the 2nd kind
November 2010, Tel Aviv