a robust technique for lumped parameter inverse boundary value problems
DESCRIPTION
A Robust Technique for Lumped Parameter Inverse Boundary Value Problems. P. Venkataraman. Todays Presentation. Introduction to Lumped Parameter Inverse BVP The Solution Procedure The Example Conclusion. Lumped Parameter Inverse BVP. An Example :. - PowerPoint PPT PresentationTRANSCRIPT
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
A Robust Technique for Lumped Parameter Inverse Boundary Value Problems
P. Venkataraman
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
Todays Presentation
1. Introduction to Lumped Parameter Inverse BVP2. The Solution Procedure3. The Example4. Conclusion
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
3
Lumped Parameter Inverse BVP
An Example :2
1 1 1 2 1 1 3
2 4 1 2 5
3 6 1 3
0;
0;
0;
y b y b y y b
y b y y b
y b y y
Fluid flow in a long vertical channel with fluid injection
Forward Problem:
Given: 1 1 2 3
1 1 2 3
(0) 0; (0) 0; (0) 0; (0) 0;
(1) 1; (1) 0; (1) 0; (1) 1;
y y y y
y y y y
1 2 3
4 5 6
100; 100; 276;
100; 1; 70;
b b b
b b b
and
Find: 1 2 3y x , y x , y x
Inverse Problem:
Given: 1 2 3y x , y x , y x and1 1 2 3
1 1 2 3
(0) 0; (0) 0; (0) 0; (0) 0;
(1) 1; (1) 0; (1) 0; (1) 1;
y y y y
y y y y
Find: 1 2 3 4 5 6b ,b ,b ,b ,b ,b
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
4
Lumped Parameter Inverse BVP
Forward Problem:Given: 1 1 2 3
1 1 2 3
(0) 0; (0) 0; (0) 0; (0) 0;
(1) 1; (1) 0; (1) 0; (1) 1;
y y y y
y y y y
1 2 3
4 5 6
100; 100; 276;
100; 1; 70;
b b b
b b b
and
Find: 1 2 3y x , y x , y x
Inverse Problem:Given: 1 2 3y x , y x , y x and
1 1 2 3
1 1 2 3
(0) 0; (0) 0; (0) 0; (0) 0;
(1) 1; (1) 0; (1) 0; (1) 1;
y y y y
y y y y
Find: 1 2 3 4 5 6b ,b ,b ,b ,b ,b
• is well posed (solution exist and unique)• assumes perfect measurement of parameters and boundary conditions• error in the solution will vanish as the perturbation in the parameters tends
to zero
• Inverse problems are considered naturally unstable, ill-posed, not unique• cannot be satisfactorily solved mathematically• no valid inverse problem based on smooth or perfect data• all current methods use some sort of regularization (artificial objective
function for minimization)• inverse problem cannot be satisfactorily solved without partial information
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
5
Lumped Parameter Inverse BVP
The solution of inverse BVP in this paper is robust:
is natural based on derivative informationprocedure can be adapted fro the forward problem too
does not require regularization
does not require dimensional control
does not require partial information
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
6
The Solution Procedure
Some Assumptions :
• we simulate discrete non smooth data to represent measurement error from smooth data of the forward problem
• each data stream is connected to a differential equation• the solution of the inverse BVP is the value of the lumped
parameters• the solution of the inverse BVP is also the trajectory based on
the parameters• quality of solution is determined by closeness of trajectory
determined using the value for parameters to the original smooth trajectory
• final trajectory is obtained using numerical integration (collocation)
• for comparison we assume that the boundary conditions are not perturbed
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
7
The Solution Procedure
Step 1: Data Smoothing using a recursive Bezier filterBezier filter determines the best order that minimizes the sum of least squared error (LSE) and the sum of the absolute error (LAE)
Step 2: The first optimization procedureObtain the first estimate for the lumped parameters by the minimization of the sum of the residuals over a set of available data points
Step 3: The second optimization procedureObtain the second estimate for the lumped parameters by minimizing the sum of the error between original data and data obtained through numerical integration (collocation) over a reduced region
Step 4: Final numerical integration to generate the trajectory based on the solution in Step 3 Boundary conditions are assumed perfect to compare the trajectoryTrajectory is compared to underlying smooth trajectory
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
8
The Solution Procedure
MATLAB was used for all calculations
A combination of symbolic and numerical processing was used to postpone round-off errors
The following MATLAB functions was used in the implementation• fminunc : unconstrained function minimization• matlabFunction : conversion of symbolic objects• bvp4c :
No special programming techniques were necessary
Computations were performed using a standard laptop
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
9
The Example
Example of fluid flow in a long vertical channel with fluid injection on one side
R is the Reynolds number and Pe is the Peclet number. A is an unknown parameter which is determined through the extra boundary condition. The example is defined for a Reynolds number of 100 for which the value for A is 2.76.
21 1 1 1
2 1 2
3 1 3
1 1 2 3
1 1 2 3
0;
1 0;
0;
(0) 0; (0) 0; (0) 0; (0) 0;
(1) 1; (1) 0; (1) 0; (1) 1;
e
y R y y y RA
y Ry y
y P y y
y y y y
y y y y
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
10
The Example – Solution to Forward Problem
Solution to forward problem using Bezier function – order 20
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
11
The Example – Inverse Problem with Smooth Data
The Bezier function technique with 18th order functions
IG : Initial GuessOpt1 : First OptimizationOpt2 : Second OptimizationES : Expected Solution
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
12
The Example – Perturbation in Boundary Conditions
Most solution to inverse BVP do not consider change in BC
This work accommodates changes in BC as it works with a clipped region
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
13
The Example – Inverse Problem with Non Smooth Data
10% perturbation – 31 points
Variable y1Variable y2
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
14
The Example – Inverse Problem with Non Smooth Data
10% perturbation – 31 points
Variable y3
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
15
The Example – Inverse Problem with Non Smooth Data
20% perturbation – 31 points
Variable y1Variable y2
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
16
The Example – Inverse Problem with Non Smooth Data
20% perturbation – 31 points
Variable y3
P. Venkataraman
Mechanical Engineering
P. Venkataraman Rochester Institute of TechnologyDETC2012 – 70343: A Robust Technique for Lumped Parameter Inverse Boundary Value Problem
32nd CIE, Chicago IL, Aug 2012
17
Conclusions
This paper presents a robust method for inverse lumped parameter BVP
The method is based on describing the data using Bezier functions
The method involves three sequential applications of unconstrained optimization
The method does not require regularization
The method does not require dimensional control
The method does not require additional information on the nature of the problem or solution
The method accommodates the perturbation of the boundary conditions