[ZengJiChuan]

For steady-state simulation, I want to assign the outlet temperature of one computing domain to the inlet temperature of another computing domain. I write the following code, but the simulation results show that the inlet temperature is always zero. I don’t know why. I hope you all Take a look at my UDF, thank you.
UDF:
#include“udf.h"
DEFINE_ADJUST(T_out_in, d)
{
real NV_VEC(A);
real sum_T_A = 0.0;
real sum_A = 0.0;
real outlet_temp;
int i;
#if !RP_HOST
Domain *domain;
Thread *threadOutlet1;
Thread *threadInlet2;
domain = Get_Domain(1);
threadOutlet1 = Lookup_Thread(domain, 44);
threadInlet2 = Lookup_Thread(domain, 43);
for(i = 0; i < THREAD_N_ELEMENTS_INT(threadOutlet1); i++)
{
F_AREA(A,i,threadOutlet1);
sum_A+=NV_MAG(A);
sum_T_A+=NV_MAG(A)*F_T(i,threadOutlet1);
}
#endif

#if RP_NODE
sum_A = PRF_GRSUM1(sum_A);
sum_T_A = PRF_GRSUM1(sum_T_A);
#endif
node_to_host_real_2(sum_A, sum_T_A);
#if !RP_NODE
outlet_temp = sum_T_A/sum_A;
Message("\n temp of outlet2 is %g\n", outlet_temp);
#endif
host_to_node_real_1(outlet_temp);
#if !RP_HOST
for(i = 0; i<THREAD_N_ELEMENTS_INT(threadInlet2); i++)
F_UDMI(i, threadInlet2, 0) = outlet_temp;
#endif
}

DEFINE_PROFILE(Inlet2_T, t, index)
{
#if !RP_HOST
int i;
for(i = 0; i < THREAD_N_ELEMENTS_INT(t); i++)
F_PROFILE(i, t, index) = F_UDMI(i, t, 0)；
#endif
}

[AlexanderZ]

instead of i which has type integer you should use type face_t
it influence F_UDMI and F_PROFILE macros

example

#include "udf.h"
DEFINE_PROFILE(pressure_profile,t,i)
{
real x[ND_ND]; /* this will hold the position vector */
real y;
face_t f;
begin_f_loop(f,t)
{
F_CENTROID(x,f,t);
y = x[1];
F_PROFILE(f,t,i) = 1.1e5 - y*y/(.0745*.0745)*0.1e5;
}
end_f_loop(f,t)
}

[ZengJiChuan]

Thanks for your reply, I feel the problem is not the use of int type, because I changed the udf to the following, the result is still that the inlet temperature is zero. I want to simulate a situation similar to a U-shaped tube flow, the middle connection part is not drawn, and the temperature is transmitted in some way.
#include"udf.h"
DEFINE_ADJUST(T_out_in, d)
{
real NV_VEC(A);
real sum_T_A = 0.0;
real sum_A = 0.0;
real outlet_temp;
face_t f;
Domain *domain;
Thread *threadOutlet;
Thread *threadInlet;
domain = Get_Domain(1);
threadOutlet = Lookup_Thread(domain, 44);
threadInlet = Lookup_Thread(domain, 43);
begin_f_loop(f,threadOutlet)
{
if(PRINCIPAL_FACE_P(f,threadOutlet))
{
F_AREA(A,i,threadOutlet);
F_UDMI(f,threadOutlet,0)=NV_MAG(A);
sum_A+=NV_MAG(A);
sum_T_A+=NV_MAG(A)*F_T(i,threadOutlet);
}
}
end_f_loop(f,threadOutlet)

#if RP_NODE
sum_A = PRF_GRSUM1(sum_A);
sum_T_A = PRF_GRSUM1(sum_T_A);
#endif
node_to_host_real_2(sum_A, sum_T_A);
#if !RP_NODE
outlet_temp = sum_T_A/sum_A;
Message("\n temp of outlet2 is %g\n", outlet_temp);
#endif
host_to_node_real_1(outlet_temp);
begin_f_loop(f,threadInlet)
{
F_UDMI(f,threadInlet,1) = outlet_temp;
}
end_f_loop(f,threadInlet)
}

DEFINE_PROFILE(Inlet2_T, t, i)
{
begin_f_loop(f,t)
{
F_PROFILE(f,t,i) = F_UDMI(f,t,1);
}
end_f_loop(f,t)
}

[vinerm]

It's not the code, but the approach that has a problem. Since you want to apply a single value, area-weighted average, as boundary condition, there are no UDMs needed, none of those. You can sum all the values in the loop and calculate average and assign it to a real scalar. Then, access this scalar in DEFINE_PROFILE and assign it to F_PROFILE.

As far as the code is concerned, though face_t and cell_t are int types, they are associated to the grid structure. Hence, you can use standard-C for loop with integer counter but that won't guarantee the access of same UDMs as face IDs. UDMs are associated with particular faces and cells.

[ZengJiChuan]

Originally Posted by vinerm

It's not the code, but the approach that has a problem. Since you want to apply a single value, area-weighted average, as boundary condition, there are no UDMs needed, none of those. You can sum all the values in the loop and calculate average and assign it to a real scalar. Then, access this scalar in DEFINE_PROFILE and assign it to F_PROFILE.

As far as the code is concerned, though face_t and cell_t are int types, they are associated to the grid structure. Hence, you can use standard-C for loop with integer counter but that won't guarantee the access of same UDMs as face IDs. UDMs are associated with particular faces and cells.

Thank you for your reply. You mean that I don't need to use UDM, so I use global variables to pass the temperature instead, and begin_f_loop function to loop. However, the calculation result is still that the inlet temperature is zero. I feel that the ADJUST function does not work, and the message function displays no information. So it's my UDF's fault, or ADJUST can't achieve the function I want. The simulation I want to do is similar to the flow of a U-shaped tube, the middle connection part is not drawn, and I want to pass the temperature through udf.
new UDF:
#include"udf.h"
real avg_T_A_penal_W01_up=0.0;
DEFINE_ADJUST(T_TRANSFER,d)
{
real total_A_penal_W01_up=0.0;
real total_T_A_penal_W01_up=0.0;

/* "Parallelized" Sections */

#if !RP_HOST /* Compile this section for computing processes only (serial and node) */
Thread *thread_penal_W01_up;
Domain=*domain;
face_t f;
real A[ND_ND];
#endif /* RP_HOST*/

#if !RP_HOST
domain=Get_Domain(1);
thread_penal_W01_up=Lookup_Thread(domain,33);
begin_f_loop(f,thread_penal_W01_up)
{F_AREA(A,f,thread_penal_W01_up);
total_A_penal_W01_up+=NV_MAG(A);
total_T_A_penal_W01_up+=NV_MAG(A)*F_T(f,thread_pen al_W01_up);
}
end_f_loop(f,thread_penal_W01_up)

/* global summation for nodes,Does nothing in Serial */
#if RP_NODE
total_A_penal_W01_up=PRF_GRSUM1(total_A_penal_W01_ up);
total_T_A_penal_W01_up=PRF_GRSUM1(total_T_A_penal_ W01_up);
#endif /* RP_NODE*/
#endif /* RP_HOST*/

/* Pass the node's total area,total temp and total vilocity to the Host for averaging */
node_to_host_real_2(total_A_penal_W01_up,total_T_A _penal_W01_up); /* Does nothing in SERIAL */
#if RP_HOST
Message("i am in host, %g, %g", total_A_penal_W01_up,total_T_A_penal_W01_up);
#endif
avg_T_A_penal_W01_up=total_T_A_penal_W01_up/total_A_penal_W01_up;
}

DEFINE_PROFILE(T_adjust_penal_W02_up,t,i)
{
#if !RP_HOST /* SERIAL or NODE */
face_t f;
#endif /* RP_HOST*/

#if !RP_HOST
begin_f_loop(f,t)
{
F_PROFILE(f,t,i)=avg_T_A_penal_W01_up;
}
end_f_loop(f,t)
#endif /* !RP_HOST */
}

[vinerm]

Yes, that's correct. You don't need a UDM and a global variable can be used. Do note that DEFINE_ADJUST needs to be hooked and it won't work until hooked at its right place. Furthermore, Domain is already provided by Fluent as second argument of DEFINE_ADJUST. Hence, you don't need commands like Get_Domain; just use d, if required. Before parallelizing, make it work in Serial. Once it works as expected, then parallelize it.

[ZengJiChuan]

Originally Posted by vinerm

Yes, that's correct. You don't need a UDM and a global variable can be used. Do note that DEFINE_ADJUST needs to be hooked and it won't work until hooked at its right place. Furthermore, Domain is already provided by Fluent as second argument of DEFINE_ADJUST. Hence, you don't need commands like Get_Domain; just use d, if required. Before parallelizing, make it work in Serial. Once it works as expected, then parallelize it.

Thank you for your reply. I hooked the ADJUST function according to your ideas, and the result of the operation is in line with my expectations. But I did the parallel operation directly without serializing it first. So how to achieve serial then parallel?

[vinerm]

Using it in Serial helps with debugging. However, if the code is already doing what you expect it to do, then you don't need to try it in serial. For these codes, serialization term does not apply, only parallelization is done; the codes are already in their serial form.