Genetic algorithms in molecular design of novel fabrics Sylvia Wower / Market Research / Philadelphia / DVIRC / Manufacturing / Philadelphia MSA

1. NTC Project: C05-PH01
Fitness Landscape
Genetic Algorithms in Molecular Design
of Novel Fibers
Les Sztandera, leader (PhilaU);
Hugh Cartwright (Oxford), Chih-Chung Chu (Cornell)
The formulation of materials which satisfy strict property
constraints is an increasingly important problem in polymer
chemistry. We are using two techniques from the field of
artificial intelligence to help design such polymers. As a
model we used a neural network whose role is to predict the
properties of a given polymer from its composition or struc-
ture, thereby solving what is known as the forward prob-
lem. We also used a genetic algorithm, which solves the
inverse problem by acting as a search procedure to find the
optimum formulation. Combined together in a collabora-
tive manner, these two techniques form a paired algorithm
in which the neural network is used in the calculation of the
genetic algorithms’ fitness function. The tool created this
way is known as a Hybrid Intelligent System. We chose
object-oriented design to develop the software tool, rather
than a functional approach, because it produces more main-
tainable and easily understood system architecture and
code.
: Tg = 299.9 K
[z-axis value = fitness function]
Initially, our work focused on the second of the two tools
required for the Hybrid System, the Genetic Algorithm
(GA). The present formulation of the GA comprises two
sub-systems: a GA engine with five fitness functions and a
graphical user interface (GUI) front end which provides the
user with an easy route into the functionality of the GA en-
gine. The design methodologies we used to produce the
GA are analogous to the
evolutionary in-
cremental software
engineering model.
Starting from a basic
genetic algorithm, we
added input and output
functionalities to yield a
checked test-bed model,
which we will use as the
basis of the polymer
modeling system. We
designed a suitable GUI sub-system so the user can enter
and modify GA parameters; this incorporates checks to
validate parameters as they are entered.
The fitness function is a quantitative measure of solution
quality and is specific for a given target glass transition
temperature (Tg) [see graph]. The fitness landscape shows
the quality of the formulation solutions across the search
space. Initial versions of the algorithm included only a
single fitness function. Recently, we added a number of
functionalities and four additional fitness functions which
we incorporated into the architectures of both the GA en-
gine and the GUI.
Case Studies
In our Case Study #
1, we used the neural network to for-
mulate the mole fractions of three constituent monomers
which will form a terpolymer with a target Tg. The system
under investigation contains as its monomers, n-octadecyl
acrylate, ethyl acrylate and acrylonitrile. The data for this
system came from Jordan.1,2
Our Case Study #
2 aimed to
solve a problem that is of considerably higher dimensional-
ity, since the copolymer system now contained nine co-
monomers. The forward problem was of the same form as
in Case Study #
1 and again was solved by a neural network.
However, the inverse problem demands another method of
solution since employing the custom algorithm used in
Case Study #
1 was only practicable due to the small-scale
nature of the problem. In Case Study #
2, the search space
becomes more complex, so that a simplistic algorithm
would be ineffective.3
Instead we chose the genetic algo-
rithm to solve the inverse problem.
Continuing Research
We are now extending our work to include data from
even larger databases, and to investigate the prediction of
properties of random sets of polymers formed from an al-
phabet of divalent molecular fragments. This problem has
previously been studied using linear correlation; early re-
sults suggest that the neural network hybrid method will be
a more accurate predictor of polymer properties.
Other Contributors: Graduate Students: Jonathan Mohr,
Sylwia Wower, Xi Chen (PhilaU); Undergraduate Stu-
dents: Andrew Regis (PhilaU), Rohan Gunatillake (Ox-
ford); Contributing Faculty: Fernando Tovia (PhilaU).
Industry Interactions: 2 [Tribology Consulting Int., ETHI-
CON Products Co]
Project Web Address:
http://www.ntcresearch.org/projectapp/?project=C05-PH01
For Further Information:
Using artificial intelligence techniques,
we are designing polymer formulations
with specified properties, such as stretch,
strength, bulk, comfort and dyeability.
1. E. F. Jordan et al., J. App. Poly. Sci., 16 :3017 (1972).
2. E. F. Jordan et al., J. App. Poly. Sci., 17:1545 (1973).
3. M. Mitchell, J. H. Holland and S. Forrest, in Advances in Neu-
ral Information Processing Systems 6, Eds. J. D. Cowan, G.
Tesauro, and J. Alspector, Morgan Kaufman, San Francisco
(1994).
4. H. M. Cartwright, Applications of Artificial Intelligence in
Chemistry, OUP, Oxford (1994)
5. Hugh M Cartwright and Les M. Sztandera (Eds.), Soft Com-
puting in Chemistry, Springer-Verlag, Heidelberg (2002).
6. Hugh Cartwright (Ed.), Intelligent Data Analysis in Science,
Oxford Chemistry Masters Series, Oxford Univ. Press (2000).
7. Hugh M. Cartwright, Investigation of Structure-Biodegrada-
bility Relationships in Polychlorinated Biphenyls using Self-
organizing Maps Neural Computing and Applications 11:30
(2002)
8. Ketan Patel and Hugh M. Cartwright, Clustering of Large
Data Sets in the Life Sciences in: Soft Computing in Chemis-
try, Hugh M. Cartwright & Les M. Sztandera (Eds.),
Springer-Verlag. Heidelberg (2002).
9. H. M. Cartwright, L. M. Sztandera and C. C. Chu, Genetic Al-
gorithms in Molecular Design of Novel Fibers, International
Journal of Intelligent Systems [submitted] (2005).
10. Rohan Gunatillake, Part II Chemistry thesis: Oxford Univ.,
Hybrid Intelligent Systems in Polymer Design (2005)
National Textile Center Research Briefs – Chemistry Competency: June 2006

2. NTC Project: C05-PH01
Les M. Sztandera, a Professor of Com-
puter Information Systems at PhilaU
served as a Distinguished Fulbright
FLAD Chair in Information Systems
Les earned a Diploma from Cambridge
(England) in 1989, an M.S. from Univ. of
Missouri in 1990 and a Ph.D. in com-
puter and engineering science from the
Univ. of Toledo in 1993. Les’ research
interests include fuzzy logic, pattern
recognition, computer vision, genetic
algorithms, neural networks, hybrid in-
telligent systems, and modeling and
management of uncertainty.
I98-P01, S01-PH10, C04-PH02s*, C05-PH1*
sztanderal@philau.edu
(215)-951-5356
http://faculty.philau.edu/sztanderal
Hugh M. Cartwright, a Lecturer and
Laboratory Officer in Chemistry at Ox-
ford Univ. (UK), joined the faculty in
1984. He earned a B.Sc. in 1969 and a
Ph.D. in 1972 in chemical sciences at
the Univ. of East Anglia (Norwich
Eng.). Hugh is the author of Applica-
tions of Artificial Intelligence in Chem-
istry. His research interests center on
the use of artificial intelligence in sci-
entific problems, such as the dispersal
of airborne pollution, optimization of
organic synthesis, industrial process
control and development, drug design,
bacterial growth, bio-informatics and
the assessment of medical data.
C04-PH02s, C05-PH01
hugh.cartwright@chem.ox.ac.uk
44 (0) 1865 275 483
http://www.chem.ox.ac.uk/researchguide/hmcartwright.html
Chih Chung Chu, a Professor in Bio-
medical Engineering at Cornell, joined
the faculty in 1978 after 3 years at Univ.
of Alabama - Birmingham. C.C. earned
a Ph.D. in polymer chemistry from Flor-
ida St in 1976 and a B.S. in chemistry
Tamkang Univ. (Taiwan) in 1968 and
served from 1986-90 as Visiting Re-
search Associate Professor of Surgery
at the Hahemann Univ. School of Medi-
cine (Philadelphia). His interests in-
clude basic research in polymer/fiber
morphology and degradation mecha-
nisms and applied research in bio-
medical polymers and fibers for human
body repair.
M01-CR01*, M03-CR04*, C04-PH02s, C05-PH01
cc62@cornell.edu
(607)-255-1938
http://www.human.cornell.edu/txa/faculty/Chu/whois_frame.html
National Textile Center Research Briefs – Chemistry Competency: June 2006