Network-based control with real-time prediction of delayed/lost sensor data

We present and experimentally verify two control architectures: supervisory control and feedback control over network. With the client/server architecture we developed for supervisory control, a client can give the control commands, tune the control parameters, and receive the experimental results remotely and in real time. Real-time feedback control over network is also investigated with a ball magnetic-levitation (Maglev) system as a test bed. We devised a novel timeout scheme and derived an autoregressive (AR) prediction model for delayed/lost sensor data. By applying this timeout and prediction scheme to the cases of consecutive data loss, we showed that the test bed could survive sporadic delays of as long as three sampling periods without losing stability. The feasibility and effectiveness of the proposed network-based real-time control methodologies was demonstrated experimentally.