Computational Performance of a Remote Nonlinear Predictive Controller on a Single Board Computer Applied to a Distillation Process

The computational time in a nonlinear model-based predictive control (NMPC) depends on several factors. For its implementation, very short execution times are required, which translates into specialized and robust computational capabilities. These capabilities can be evaluated using external servers and Single Board Computers (SBCs), small in size, but with important computational
features. The present work evaluated the computational time required in the implementation of a remote NMPC for three different models of a distillation column using the Python Gekko library on a Raspberry PI 3B+ board, and using an Aspen Plus Dynamic simulation as a plant through Open Platform Communication (OPC). In total, 9 cases involving between 2229 and 5109 state
variables were evaluated, finding that, in all cases, the time needed to solve the NMPC was less than 30 s, with a CPU consumption of less than 50%.