Change material properties of several elements in PyMapdl

[derne058]

Hi everyone,
I'm working with PyMapdl and I'm trying to change material properties of each elements of my mesh. When the number of elements is small the changing is pretty fast, on the contrary, when I increase the number of elements the process is very slow.
Here are the commands that I'm using:

      x_filtered = (H @ x)    # filter element density with neighbour elements
      for i in range(num_elem_dd):
          mp_num = i + 1
          E = E_0 * x_filtered[i,0]**p    # change Young's modulus using filtered density
          mapdl.mp("EX",mp_num,E)
          mapdl.mp("PRXY",mp_num,ni)
          mapdl.mp("DENS",mp_num,7850)
          mapdl.mpchg(mat = mp_num, elem = int(ID_elem_dd[i]))

It is basically a for loop that ranges from the 1st element up to the last.
Is there something faster that can be used instead of using a for loop?

[mikerife]

Hi @derne058
Sorry my brain is being Friday slow. What is H @ x ??
Is there a expression for EX(x)? If so then we could probably define a field variable and then EX as a function of the field variable. See the publeic help here for more info.

[derne058]

Hi @mikerife,
H@x is the product between a filter matrix (H) and a design variables vector (x). The code I shared is part of larger topology optimization algorithm that I'm writing. Basically the number of material properties is equal to the number of design variables (so the number of values within vector x).
For each element, I define material property as:

E = E_0 * x_filtered[i,0]**p

where E_0 is the Young's modulus of solid element (E_0=200 GPa), x_filtered the vector of design variables filtered considering neighboring elements, and p is the penalization parameter (p=3). By calculating the new E, I modify the element material properties using a for loop.
I hope what I have written clarifies the code I shared.

[germa89]

Hi @derne058

In the loop you mentioned, you are issuing several gRPC calls (mapdl.mp) per each element. That can be quite a lot. To speed up the process, avoid doing a loop per element, and try to do a loop per material.

For instance, if you are dealing with a material with variable porosity, you can select all the elements which will have similar porosity (within a range), and apply the material change in all the elements at once. You can try the following pseudocode:

grading_steps = [(0, 0.1), (0.1, 0.2), (0.2, 0.3), ...] # surely this can be more elegant.

for lower_bound, upper_bound in grading_steps:
    x_filtered__ = lower_bound < x_filtered < upper_bound
    # select elements
    mapdl.esel( ...)

    # Change material properties
    mp_num = i + 1
    E = E_0 * x_filtered[i,0]**p    # change Young's modulus using filtered density
    mapdl.mp("EX",mp_num,E)
    mapdl.mp("PRXY",mp_num, ni)
    mapdl.mp("DENS",mp_num,7850)
    mapdl.mpchg(mat = mp_num, elem = "all")

The approach should speed up the process.

Additionally, if you can do all the python operations outside the loop, then you can put as much gRPC calls inside a mapdl.non_interactive context so all the commands are sent at once, and run without much gRPC overhead.