How to achieve free cooling?

What I hope to do is normalize the metal material, but since I am a beginner, I will only use commands such as fix npt or fix nvt to cool down at a certain rate, which I don’t feel is very consistent with natural cooling. Is there a more consistent way to achieve free cooling? Thanks for the better advice!

You are allowed to use every command as long as you read the documentation and understand the physics of what you are simulating.
A simulation is -by definition- an imitation of a natural process and is -necessarily- based on approximations. It is perfectly fine to use a thermostat and a barostat to adjust the initial condition of a sample. If you are describing your system correctly, you should observe energy conservation upon reaching thermal equilibrium (well, a slight drift over long timescales is allowed. You get the point).
Another good attitude is to do numerical experiments and understand how your system behaves in response to different choices of numerical parameters. That will give you an appreciation of numerical noise, dependence on the initial conditions, and such.

I think what you want is fix nve, which doesn’t have a thermostat (fix nve command — LAMMPS documentation).

Well, what is “natural cooling”? Please keep in mind that a simulation is quite artificial and the time and length scales of a simulation are very different from typical experiments. As a consequence, the procedures to obtain equilibrium for a system can be artificial, too. Since only the state matters and now how you get there, this is quite valid.

I suggest you spend some time studying about equilibration and statistical mechanical ensembles in a text book on molecular dynamics and statistical mechanics, respectively.

Okay, I’ll follow your advice.

Doesn’t fix nve keep the energy of the system constant? Natural cooling, I think should exchange energy with the environment.

Do you mean I need to try every temperature control command? And then find a way to match the results through the results?

There are different thermostats implemented in LAMMPS. A velocity rescaling (which could be achieved with the Berendsen thermostat) could be helpful to quench a system badly out of equilibrium but is not suitable for getting a correct statistical sampling of the sample, especially for liquid systems (you can get away with amorphous solids, for instance).
But these topics are better discussed with an advisor than a random guy.