Experimental data fitting from numerical solution of PDE in Matalab

Question

Jung 2011-12-16

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/24126-experimental-data-fitting-from-numerical-solution-of-pde-in-matalab

采纳的回答： Walter Roberson

Hello, I have tried to fit the experimental data by PDE equation. Can you teach me how to do it?

My parabolic PDE equations is dq/dt=(1/r^2)*d(D*(1/r^2)*(dq/dr)).

I.C. q(r,0)=0

B.C. dq(0,t)/dr=0

B.C. dq/dt=kl*(c-cs) at r=R, c=cs

I searched previous questions (optimaization toolbox-->lsqcurvefit). But, I need to have more detailed procedure.

Thank you for your help in advance.

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Sean de Wolski 2011-12-16

Can you please format your code and include the code you've tried, i.e. your setup and call to lsqcurvefit().

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Walter Roberson 2011-12-16

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/24126-experimental-data-fitting-from-numerical-solution-of-pde-in-matalab#answer_31710

In the subexpression d(D*(1/r^2)*(dq/dr)) is D a constant? Does the leading "d" indicate differentiation? If so, then with respect to which variable?

Does dq(0,t)/dr indicate dq/dr evaluated at (0,t) ?

You show the condition c=cs in your second boundary condition, but you had not mentioned c before. Is c a function of r and t ? Why mention the "r=R" if r is not involved in kl*(c-cs) ? At c=cs then (c-cs) would seem to be 0, and since no matter what finite value kl is, kl*0 is going to be 0, you seem to be indicating that dq/dt will be 0 whenever c=cs -- was that your intention ?

Some of the combinations I tried only have q(r,t) = 0 as a solution, and the other combinations didn't resolve. I will need to see your clarifications to go further.

Note to myself:

pdsolve([diff(q(r, t), t) = (diff(D1*(diff(q(r, t), r))/r^2, t))/r^2, q(r, 0) = 0, (D[1](q))(0, t) = 0], q(r, t))

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

Answer 2

Marc 2011-12-21

2
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/24126-experimental-data-fitting-from-numerical-solution-of-pde-in-matalab#answer_32132

This problem has been looked at in great detail by many. I recommend a group at Rutgers, led by Prof. Glasser. Their work looks at metal adsorption within oxide pellets and spheres.

Their research focuses on metal profiles after drying but their adsorption model is very complete. Plus the references they cite lead to more simplified approaches.

My recommendation is to transform the system of PDEs into ODEs using a finite difference approach in the spatial domain. Leaving a system of ODEs ( or DAE) easily solved in Matlab by ode15s.

You can then set up a function within your function something like this for nlinfit or lscurvefit....

function [t,y]=yourfunc(a,b,c)

tspan = [to tf];

[t,y] = ode15s(@odefunc,tspan,yo);

function dydt = odefunc(t,y)

dydt = zeros(some_value, 1); dydt(1) = a*.... your equations with parameters a,b and c you want to fit

end

I have left out ALOT of details as there are a lot of examples in the Matlab documentation on these functions.

Good luck.

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Jung 2012-3-22

Thanks alot! This is very helpful.

请先登录，再进行评论。

Answer 3

Jung 2011-12-18

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/24126-experimental-data-fitting-from-numerical-solution-of-pde-in-matalab#answer_31762

Hi Walter and Sean, thank you for your comments and teaching. I think I need to define the equations clearly to ask more. The equation is supposed to describe the concentration of profile of adsorbate inside a spherical particle. q: amount of adsorption q is the function of time and radius (function q=q(r,t)) r: radius of particle (r is from 0 to R) d: differentiation D: diffusivity of adsorbate c: concentration of adsorbate (c is from c0 (at t=0) to c(at t=time at equilibrium) k: liquid film transport coefficient cs: concentration of adsorbate at particle surface (r=R)

dq/dr=(1/r^2)*d(D*(dq/dr))is equal to dc/dr=(1/r^2))*d(D*(dc/dr). Initial condition: c=q=0 at t=0 Boundary condition: dc/dr=dq/dr=0 at the center of a particle at any time Another boundary condition: -dc/dt=k*(c-cs) at r=R at anytime

I have experimental data (c VS t) so that I have tried to fit the data to above equation to estimate diffusivity and check the validation of this euqation. I am a beginner in matlab so that I don't have matlab code right now. could you teach me how to fit the data using above equation. Thank you for your help.