Developing an NN-MPC DC Motor Platform for an Online Real-time Laboratory

In this paper, the development of an online, real-time DC Motor speed control platform is described that is meant for teaching concepts in feedback control and related areas. In particular, the challenges in implementing a Neural Network-Model Predictive Control (NN-MPC) speed control algorithm within a finite time interval, consistently, are highlighted. The two-step procedure of i. offline training of the NN using hardware data and ii. real-time implementation of the NN to calculate predicted outputs as used in the MPC, is illustrated. A novel NN implementation that generates output predictions for a chosen prediction horizon is described; this formulation does not require the NN to be implemented recursively. This formulation ensures computations, including convex optimisation, are completed within the chosen sampling interval. The NN-MPC speed controller is implemented in the LabVIEW environment and is accompanied by a web-browser based interface that permits experimentation using different settings as well as access to experimental data. Experimental results comparing the performance of the NN-MPC implementation and a model-based MPC implementation are presented; the influence of model mismatch in tracking performance is highlighted.