A data-driven method for dissipative thermomechanics

摘要

We present a method based on quadratic programming for learning dissipative models from data. For that, we take advantage of the metriplectic structure of the system to preserve the two laws of thermodynamcis, i.e. energy conservation and non-decreased entropy. This method can be used in conjunction with EEM integrators to generate structure preserving integrators for dissipative thermomechanics. We illustrate the results by an example.