FMI4cpp is a cross-platform FMI 2.0 implementation written in modern C++.
Influenced by its spiritual brother FMI4j, it aims to be an easy to install, easy to use, object-oriented and fast FMI implementation for C++.
FMI4cpp supports both Co-simulation and Model Exchange.
Refer to BUILDING.md
#include <iostream>
#include <fmi4cpp/fmi4cpp.hpp>
using namespace fmi4cpp;
const double stop = ...;
const double stepSize = ...;
int main()
{
fmi2::fmu fmu("path/to/fmu.fmu");
auto cs_fmu = fmu.as_cs_fmu();
auto me_fmu = fmu.as_me_fmu();
auto cs_md = cs_fmu->get_model_description(); //smart pointer to a cs_model_description instance
std::cout << "model_identifier=" << cs_md->model_identifier << std::endl;
auto me_md = me_fmu->get_model_description(); //smart pointer to a me_model_description instance
std::cout << "model_identifier=" << me_md->model_identifier << std::endl;
auto var = cs_md->get_variable_by_name("my_var").as_real();
std::cout << "Name=" << var.name() << ", start=" << var.start().value_or(0) << std::endl;
auto slave = cs_fmu->new_instance();
slave->setup_experiment();
slave->enter_initialization_mode();
slave->exit_initialization_mode();
double t;
double value;
auto vr = var.valueReference();
while ( (t = slave->get_simulation_time()) <= stop) {
if (!slave->step(stepSize)) {
std::cerr << "Error! step() returned with status: " << to_string(slave->last_status()) << std::endl;
break;
}
if (!slave->read_real(vr, value)) {
std::cerr << "Error! step() returned with status: " << to_string(slave->last_status()) << std::endl;
break;
}
std::cout << "t=" << t << ", " << var.name() << "=" << value << std::endl;
}
slave->terminate();
return 0;
}
Would you rather simulate FMUs in Java? Check out FMI4j!
Would you like to build JVM based FMUs? Check out FMU4j!
Perhaps you want to build FMUs using plain Python? Check out PythonFMU!
Need to distribute your FMUs? FMU-proxy to the rescue!