Você está na página 1de 57

Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod

Parameter Estimation in a Mathematical Model of a


Heat-Conducting Rod

Ma. Cristina R. Bargo


Ricardo C.H. del Rosario
Jose Ernie C. Lope

Department of Mathematics
University of the Philippines Diliman

MSP 2007 Convention


Bohol Tropics Resort, Tagbilaran City, Bohol
May 18, 2007
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod

Objectives

Prepare the setup for the heat conduction experiment, data


gathering
Create a time-dependent model for heat conduction on a
metal rod
Find the solution to the model numerically
Estimate the parameters by minimizing the difference between
the actual temperature and computed temperature values
Remark: Extension of the work presented by Ms. Margie De Paz
last year
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod

Objectives

Prepare the setup for the heat conduction experiment, data


gathering
Create a time-dependent model for heat conduction on a
metal rod
Find the solution to the model numerically
Estimate the parameters by minimizing the difference between
the actual temperature and computed temperature values
Remark: Extension of the work presented by Ms. Margie De Paz
last year
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod

Outline

1 Experimental Setup
2 Model and Parameter Estimation Problem
The Model
Solution to the Model
Galerkin Method
Parameter Estimation
3 Optimization Algorithms
Least Squares
Genetic Algorithm
4 Results
5 Conclusions
6 Future Work
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Experimental Setup

Outline

1 Experimental Setup
2 Model and Parameter Estimation Problem
The Model
Solution to the Model
Galerkin Method
Parameter Estimation
3 Optimization Algorithms
Least Squares
Genetic Algorithm
4 Results
5 Conclusions
6 Future Work
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Experimental Setup

Some Data

Copper Rod Specifics Aluminum Rod Specifics


Mass Radius Length Mass Radius Length
1.14 kg 0.6375 cm 0.9980 m 0.35 kg 0.6250 cm 1.0030 m

Thermocouple Locations (m)


x1 x2 x3 x4 x5 x6 x7
copper 0.0500 0.1800 0.3100 0.5640 0.6940 0.8245 0.9980
aluminum 0.0420 0.1720 0.3020 0.4320 0.6930 0.8220 0.9520
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Experimental Setup

Experimental Setup

Figure: The experimental setup: metal rod, heat source, thermocouples


and data acquisition instrument.
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem

Outline

1 Experimental Setup
2 Model and Parameter Estimation Problem
The Model
Solution to the Model
Galerkin Method
Parameter Estimation
3 Optimization Algorithms
Least Squares
Genetic Algorithm
4 Results
5 Conclusions
6 Future Work
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
The Model

Outline

1 Experimental Setup
2 Model and Parameter Estimation Problem
The Model
Solution to the Model
Galerkin Method
Parameter Estimation
3 Optimization Algorithms
Least Squares
Genetic Algorithm
4 Results
5 Conclusions
6 Future Work
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
The Model

Model Formulation

Assumptions:
heat is transferred on one direction, and temperature is
uniform over a cross-section
constant ambient temperature
constant heat flux at one end of the rod (due to the heat
source)
heat loss along the sides of the rod
heat loss at the end away from the heat source
Given parameters: radius r, length `, ambient temperature ua ,
density ρ
Parameters to be estimated: thermal conductivity k, heat
transfer coefficient h, flux Q, specific heat capacity cp
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
The Model

Model Formulation

Assumptions:
heat is transferred on one direction, and temperature is
uniform over a cross-section
constant ambient temperature
constant heat flux at one end of the rod (due to the heat
source)
heat loss along the sides of the rod
heat loss at the end away from the heat source
Given parameters: radius r, length `, ambient temperature ua ,
density ρ
Parameters to be estimated: thermal conductivity k, heat
transfer coefficient h, flux Q, specific heat capacity cp
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
The Model

Model Formulation

Assumptions:
heat is transferred on one direction, and temperature is
uniform over a cross-section
constant ambient temperature
constant heat flux at one end of the rod (due to the heat
source)
heat loss along the sides of the rod
heat loss at the end away from the heat source
Given parameters: radius r, length `, ambient temperature ua ,
density ρ
Parameters to be estimated: thermal conductivity k, heat
transfer coefficient h, flux Q, specific heat capacity cp
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
The Model

The Model

The model is given by


∂ 2 u (x, t) 2h

∂u (x, t)
ρc = k − (u (x, t) − ua )

p

∂x2



 ∂t r
u (x, 0) = u0 (x) (1)
∂u(0,t)



 ∂x = −Q k
 ∂u(`,t) = − h (u (`, t) − u )

∂x k a
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
The Model

Remarks

Used Fourier’s Law of Conduction and Newton’s Law of


Cooling to formulate the model
Solution of (1) has no analytical form ⇒ use Galerkin method
to find the approximate solution (if the solution exists)
Changes in the model:
time-dependent (before, we considered the steady-state model)
heat loss at x = ` (before, we have ∂u(`,t)
∂x = 0)
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Solution to the Model

Outline

1 Experimental Setup
2 Model and Parameter Estimation Problem
The Model
Solution to the Model
Galerkin Method
Parameter Estimation
3 Optimization Algorithms
Least Squares
Genetic Algorithm
4 Results
5 Conclusions
6 Future Work
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Solution to the Model

Variational Formulation

Multiply PDE in (1) by φ ∈ H 1 (0, `), integrate over (0, `), use
integration by parts, and impose boundary conditions

The PDE becomes:

Z ` Z `
∂u (x, t) k
φ (x) dx = − u0 (x, t) φ0 (x) dx
0 ∂t ρc p 0
Z `
2h h
− u (x, t) φ (x) dx − u (`, t) φ (`)
rρcp 0 ρcp
2hua `
Z
hua Q
+ φ (x) dx + φ (`) + φ (0)
rρcp 0 ρcp ρcp
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Solution to the Model

Variational Formulation

Multiply PDE in (1) by φ ∈ H 1 (0, `), integrate over (0, `), use
integration by parts, and impose boundary conditions

The PDE becomes:

Z ` Z `
∂u (x, t) k
φ (x) dx = − u0 (x, t) φ0 (x) dx
0 ∂t ρc p 0
Z `
2h h
− u (x, t) φ (x) dx − u (`, t) φ (`)
rρcp 0 ρcp
2hua `
Z
hua Q
+ φ (x) dx + φ (`) + φ (0)
rρcp 0 ρcp ρcp
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Solution to the Model

Some Definitions

Let V = H 1 (0, `) and H = L2 (0, `) with the following inner


products:
Z `
hψ, φiH = ψ (x) φ (x) dx
0
Z ` Z `
0 0 2h
hψ, φiV = ψ (x) φ (x) dx + ψ (x) φ (x) dx
0 rk 0

Define the operators σ : V × V → R and F : V → R


k h
σ (ψ, φ) = hψ, φiV + ψ (`) φ (`)
ρcp ρcp
2hua `
Z
hua Q
F (φ) = φ (x) dx + φ (`) + φ (0)
rρcp 0 ρcp ρcp
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Solution to the Model

Some Definitions

Let V = H 1 (0, `) and H = L2 (0, `) with the following inner


products:
Z `
hψ, φiH = ψ (x) φ (x) dx
0
Z ` Z `
0 0 2h
hψ, φiV = ψ (x) φ (x) dx + ψ (x) φ (x) dx
0 rk 0

Define the operators σ : V × V → R and F : V → R


k h
σ (ψ, φ) = hψ, φiV + ψ (`) φ (`)
ρcp ρcp
2hua `
Z
hua Q
F (φ) = φ (x) dx + φ (`) + φ (0)
rρcp 0 ρcp ρcp
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Solution to the Model

The Weak Form

Weak form of (1)


Find u ∈ L2 (0, T ; V ) such that
(
d
dt hu, φiH = −σ (u, φ) + F (φ) , ∀φ ∈ V (2)
u (0) = u0

To show well-posedness:
show existence, uniqueness and continuous dependence on the
data of the solution to (2)
equivalence of the solution to (1) and (2)
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Solution to the Model

The Weak Form

Weak form of (1)


Find u ∈ L2 (0, T ; V ) such that
(
d
dt hu, φiH = −σ (u, φ) + F (φ) , ∀φ ∈ V (2)
u (0) = u0

To show well-posedness:
show existence, uniqueness and continuous dependence on the
data of the solution to (2)
equivalence of the solution to (1) and (2)
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Galerkin Method

Outline

1 Experimental Setup
2 Model and Parameter Estimation Problem
The Model
Solution to the Model
Galerkin Method
Parameter Estimation
3 Optimization Algorithms
Least Squares
Genetic Algorithm
4 Results
5 Conclusions
6 Future Work
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Galerkin Method

Galerkin Method

Project the solution of (2) in a finite dimensional space


V n = span {φ1 , . . . , φn }
Finite-dimensional problem: Find un ∈ L2 (0, T ; V n ) such that
(
d n n
dt hu , φi iH = −σ (u , φi ) + F (φi ) , i = 1, . . . , n (3)
un (0) = PV n u0
n
X
Write un (x, t) = αj (t) φj (x) and substitute to (3):
j=1

M α̇ (t) = Aα (t) + F (t)


n
X
Initial condition: un (x, 0) = αj (0) φj (x) = PV n u0 (x)
j=1
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Galerkin Method

Galerkin Method

Project the solution of (2) in a finite dimensional space


V n = span {φ1 , . . . , φn }
Finite-dimensional problem: Find un ∈ L2 (0, T ; V n ) such that
(
d n n
dt hu , φi iH = −σ (u , φi ) + F (φi ) , i = 1, . . . , n (3)
un (0) = PV n u0
n
X
Write un (x, t) = αj (t) φj (x) and substitute to (3):
j=1

M α̇ (t) = Aα (t) + F (t)


n
X
Initial condition: un (x, 0) = αj (0) φj (x) = PV n u0 (x)
j=1
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Galerkin Method

Galerkin Method

Project the solution of (2) in a finite dimensional space


V n = span {φ1 , . . . , φn }
Finite-dimensional problem: Find un ∈ L2 (0, T ; V n ) such that
(
d n n
dt hu , φi iH = −σ (u , φi ) + F (φi ) , i = 1, . . . , n (3)
un (0) = PV n u0
n
X
Write un (x, t) = αj (t) φj (x) and substitute to (3):
j=1

M α̇ (t) = Aα (t) + F (t)


n
X
Initial condition: un (x, 0) = αj (0) φj (x) = PV n u0 (x)
j=1
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Galerkin Method

Galerkin Method

Project the solution of (2) in a finite dimensional space


V n = span {φ1 , . . . , φn }
Finite-dimensional problem: Find un ∈ L2 (0, T ; V n ) such that
(
d n n
dt hu , φi iH = −σ (u , φi ) + F (φi ) , i = 1, . . . , n (3)
un (0) = PV n u0
n
X
Write un (x, t) = αj (t) φj (x) and substitute to (3):
j=1

M α̇ (t) = Aα (t) + F (t)


n
X
Initial condition: un (x, 0) = αj (0) φj (x) = PV n u0 (x)
j=1
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Galerkin Method

Galerkin Method
- cont -

α̇ (t) = M −1 (Aα (t) + F (t)) (4)

The variables in (4) are defined as follows:

α (t) = [α1 (t) , α2 (t) , . . . , αn (t)]T


k h
[A]ij = − hφj , φi iV − φj (`) φi (`)
ρcp ρcp
Z `
[M ]ij = φj (x) φi (x) dx
0
2hua `
Z
hua Q
[F (t)]i = φi (x) dx + φi (`) + φi (0)
rρcp 0 ρcp ρcp

Goal: Find α (t) that satisfies (4) using an ODE solver


Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Galerkin Method

Galerkin Method
- cont -

α̇ (t) = M −1 (Aα (t) + F (t)) (4)

The variables in (4) are defined as follows:

α (t) = [α1 (t) , α2 (t) , . . . , αn (t)]T


k h
[A]ij = − hφj , φi iV − φj (`) φi (`)
ρcp ρcp
Z `
[M ]ij = φj (x) φi (x) dx
0
2hua `
Z
hua Q
[F (t)]i = φi (x) dx + φi (`) + φi (0)
rρcp 0 ρcp ρcp

Goal: Find α (t) that satisfies (4) using an ODE solver


Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Galerkin Method

Galerkin Method
- cont -

α̇ (t) = M −1 (Aα (t) + F (t)) (4)

The variables in (4) are defined as follows:

α (t) = [α1 (t) , α2 (t) , . . . , αn (t)]T


k h
[A]ij = − hφj , φi iV − φj (`) φi (`)
ρcp ρcp
Z `
[M ]ij = φj (x) φi (x) dx
0
2hua `
Z
hua Q
[F (t)]i = φi (x) dx + φi (`) + φi (0)
rρcp 0 ρcp ρcp

Goal: Find α (t) that satisfies (4) using an ODE solver


Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Parameter Estimation

Outline

1 Experimental Setup
2 Model and Parameter Estimation Problem
The Model
Solution to the Model
Galerkin Method
Parameter Estimation
3 Optimization Algorithms
Least Squares
Genetic Algorithm
4 Results
5 Conclusions
6 Future Work
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Parameter Estimation

Parameter Estimation

h iT
Q h cp
Vector of unknown parameters: q = , ,
k k k
Set of data points: {û (xi , tj ) | i = 1, . . . N, j = 1, . . . , N t}
(may contain errors)
un (xi , tj ; q) is the solution of the finite-dimensional problem
(3) using the parameter q and evaluated at xi at time tj

Find:

Nt N
1 XX n
min J (q) = min |u (xi , tj ; q) − û (xi , tj )|2
q∈R3 q∈R3 N · Nt
j=1 i=1
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Parameter Estimation

Parameter Estimation

h iT
Q h cp
Vector of unknown parameters: q = , ,
k k k
Set of data points: {û (xi , tj ) | i = 1, . . . N, j = 1, . . . , N t}
(may contain errors)
un (xi , tj ; q) is the solution of the finite-dimensional problem
(3) using the parameter q and evaluated at xi at time tj

Find:

Nt N
1 XX n
min J (q) = min |u (xi , tj ; q) − û (xi , tj )|2
q∈R3 q∈R3 N · Nt
j=1 i=1
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Parameter Estimation

Parameter Estimation

h iT
Q h cp
Vector of unknown parameters: q = , ,
k k k
Set of data points: {û (xi , tj ) | i = 1, . . . N, j = 1, . . . , N t}
(may contain errors)
un (xi , tj ; q) is the solution of the finite-dimensional problem
(3) using the parameter q and evaluated at xi at time tj

Find:

Nt N
1 XX n
min J (q) = min |u (xi , tj ; q) − û (xi , tj )|2
q∈R3 q∈R3 N · Nt
j=1 i=1
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Model and Parameter Estimation Problem
Parameter Estimation

Parameter Estimation

h iT
Q h cp
Vector of unknown parameters: q = , ,
k k k
Set of data points: {û (xi , tj ) | i = 1, . . . N, j = 1, . . . , N t}
(may contain errors)
un (xi , tj ; q) is the solution of the finite-dimensional problem
(3) using the parameter q and evaluated at xi at time tj

Find:

Nt N
1 XX n
min J (q) = min |u (xi , tj ; q) − û (xi , tj )|2
q∈R3 q∈R3 N · Nt
j=1 i=1
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Optimization Algorithms

Outline

1 Experimental Setup
2 Model and Parameter Estimation Problem
The Model
Solution to the Model
Galerkin Method
Parameter Estimation
3 Optimization Algorithms
Least Squares
Genetic Algorithm
4 Results
5 Conclusions
6 Future Work
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Optimization Algorithms
Least Squares

Outline

1 Experimental Setup
2 Model and Parameter Estimation Problem
The Model
Solution to the Model
Galerkin Method
Parameter Estimation
3 Optimization Algorithms
Least Squares
Genetic Algorithm
4 Results
5 Conclusions
6 Future Work
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Optimization Algorithms
Least Squares

Least Squares

Built-in optimization algorithm in Scilab (leastsq)


Solves nonlinear least squares problems
Quasi-Newton algorithm (the Jacobian of the cost function
was not supplied)
Can be used if the errors in the measurements are randomly
distributed
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Optimization Algorithms
Genetic Algorithm

Outline

1 Experimental Setup
2 Model and Parameter Estimation Problem
The Model
Solution to the Model
Galerkin Method
Parameter Estimation
3 Optimization Algorithms
Least Squares
Genetic Algorithm
4 Results
5 Conclusions
6 Future Work
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Optimization Algorithms
Genetic Algorithm

Genetic Algorithm
Overview

Optimization algorithm inspired by the concept of evolution


(“survival of the fittest”)
Suppose that we want to find q = (q1 , q2 , . . . , qn ) that
minimizes the function F (q):
“individual”: described by q
“fitness” of an “individual”: determined by F (q)
“initial population”: randomly selected
main processes of GA: selection, recombination, mutation
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Optimization Algorithms
Genetic Algorithm

Genetic Algorithm
Overview

Optimization algorithm inspired by the concept of evolution


(“survival of the fittest”)
Suppose that we want to find q = (q1 , q2 , . . . , qn ) that
minimizes the function F (q):
“individual”: described by q
“fitness” of an “individual”: determined by F (q)
“initial population”: randomly selected
main processes of GA: selection, recombination, mutation
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Optimization Algorithms
Genetic Algorithm

Genetic Algorithm
Overview

Optimization algorithm inspired by the concept of evolution


(“survival of the fittest”)
Suppose that we want to find q = (q1 , q2 , . . . , qn ) that
minimizes the function F (q):
“individual”: described by q
“fitness” of an “individual”: determined by F (q)
“initial population”: randomly selected
main processes of GA: selection, recombination, mutation
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Optimization Algorithms
Genetic Algorithm

Genetic Algorithm
Processes

Selection
compute the fitness of each individual
form the “roulette”
create the “mating pool”
Recombination
randomly select two individuals from the mating pool
“mating” (to produce population of “offsprings”)
probability of mating is determined by a fixed probability
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Optimization Algorithms
Genetic Algorithm

Genetic Algorithm
Processes

Selection
compute the fitness of each individual
form the “roulette”
create the “mating pool”
Recombination
randomly select two individuals from the mating pool
“mating” (to produce population of “offsprings”)
probability of mating is determined by a fixed probability
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Optimization Algorithms
Genetic Algorithm

Genetic Algorithm
Processes

Mutation
introduce small changes in the parameters of an individual
(“mutate”)
probability of mutation is determined by a fixed probability
the resulting (mutated) pool will make up the initial
population for the “next generation”
Optional Process: Elitism
the fitness of the next generation doesn’t necessarily improve
(because of recombination and mutation)
if the best “parent” has disappeared, introduce it again by
removing 1 arbitrary offspring
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Optimization Algorithms
Genetic Algorithm

Genetic Algorithm
Processes

Mutation
introduce small changes in the parameters of an individual
(“mutate”)
probability of mutation is determined by a fixed probability
the resulting (mutated) pool will make up the initial
population for the “next generation”
Optional Process: Elitism
the fitness of the next generation doesn’t necessarily improve
(because of recombination and mutation)
if the best “parent” has disappeared, introduce it again by
removing 1 arbitrary offspring
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Results

Outline

1 Experimental Setup
2 Model and Parameter Estimation Problem
The Model
Solution to the Model
Galerkin Method
Parameter Estimation
3 Optimization Algorithms
Least Squares
Genetic Algorithm
4 Results
5 Conclusions
6 Future Work
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Results

Results

Compare the results (using Least Squares and Genetic


Algorithm) when:
end of the rod away from the heat source is insulated (BC1)
there is heat loss at the end of the rod away from the heat
source (BC2)

Error Values
leastsq GA
BC1 BC2 BC1 BC2
Objective Function
copper 0.1645 0.1648 0.1645 0.1648
aluminum 0.0768 0.0762 0.0758 0.0762
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Results

Results

Compare the results (using Least Squares and Genetic


Algorithm) when:
end of the rod away from the heat source is insulated (BC1)
there is heat loss at the end of the rod away from the heat
source (BC2)

Error Values
leastsq GA
BC1 BC2 BC1 BC2
Objective Function
copper 0.1645 0.1648 0.1645 0.1648
aluminum 0.0768 0.0762 0.0758 0.0762
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Results

Results
- cont -

Optimal Parameters
leastsq GA
BC1 BC2 BC1 BC2
Estimated Q/k
copper 66.1440 66.1372 66.2355 66.2739
aluminum 60.9651 60.9546 60.9729 60.9583
Estimated h/k
copper 0.0312 0.0312 0.0313 0.0313
aluminum 0.0569 0.0569 0.0570 0.0569
Estimated cp /k
copper 1.1840 1.1844 1.1848 1.1864
aluminum 3.8982 3.8986 3.9000 3.8989
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Results

Results
Copper Rod: actual temperature and computed temperature

Copper Rod: Data and Computed Values

325
data
TC1
TC2
320 TC3
TC4
TC5
TC6
temperature (K)

315
TC7

310

305

300
0 500 1000 1500 2000 2500 3000
time (s)

Figure: Plot of temperature vs. position (copper rod)


Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Results

Results
Aluminum Rod: actual temperature and computed temperature

Aluminum Rod: Data and Computed Values

322
data
TC1
320 TC2
TC3
318 TC4
TC5

316 TC6
temperature (K)

TC7

314

312

310

308

306
0 1000 2000 3000 4000 5000 6000 7000 8000
time (s)

Figure: Plot of temperature vs. position (aluminum rod)


Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Conclusions

Outline

1 Experimental Setup
2 Model and Parameter Estimation Problem
The Model
Solution to the Model
Galerkin Method
Parameter Estimation
3 Optimization Algorithms
Least Squares
Genetic Algorithm
4 Results
5 Conclusions
6 Future Work
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Conclusions

Conclusions

Obtained the data using thermocouples attached to a data


acquisition instrument.
Formulated the model for heat conduction on a metal rod,
assuming that heat is lost along its entire length and at the
boundary.
Obtained the solution to the model using the Galerkin method.
Obtained estimates for the parameters Q/k, h/k, cp /k using
leastsq and genetic algorithm.
Modeling the heat loss at the end of the rod away from the
heat source produces the same output as the model without
heat loss.
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Future Work

Outline

1 Experimental Setup
2 Model and Parameter Estimation Problem
The Model
Solution to the Model
Galerkin Method
Parameter Estimation
3 Optimization Algorithms
Least Squares
Genetic Algorithm
4 Results
5 Conclusions
6 Future Work
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod
Future Work

Future Work

Create a “better” heat source (oven, in cooperation with NIP)


Reformulate the model to incorporate “realistic” assumptions
(ambient temperature, flux)
Use other optimization algorithms (gradient-based, heuristic,
neural networks, hierarchical Bayesian methods) for parameter
estimation
Implement a faster numerical method for solving the PDE
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod

References

”Instructional and Research Laboratory, Center for Research in Scientific


Computation, North Carolina State University,”
http://www.ncsu.edu/crsc/ilfum.htm.
P. Laguitao, “Estimation of Copper Rod Parameters Using Data from
Heat Conduction Experiment,” Undergraduate Research Paper, College of
Science, University of the Philippines Diliman, 2001.
H.T. Banks, R.C. Smith and Y. Wang, “Smart Material Structures:
Modeling, Estimation and Control,” John Wiley & Sons, 1996.
R.R. Briones, “Numerical Computations for Parameter Estimation in a
Smart Beam Structure,” Master’s Thesis, College of Science, University of
the Philippines Diliman, 2002.
R. Temam, “Infinite-Dimensional Dynamical Systems in Mechanics and
Physics,” Springer-Verlag, 1997.
J. Skaar and K.M. Risvik, “A Genetic Algorithm for the Inverse Problem in
Synthesis of Fiber Gratings,” J. Lightwave Technol., 16, 1928-1932 (1998).
Parameter Estimation in a Mathematical Model of a Heat-Conducting Rod

Thank you!

Você também pode gostar