Model predictive control (MPC) has become more attractive in control engineering for the last decades because of its efficiency and robustness. In this paper, an effective control strategy is proposed for vibration reduction of mechanical flexible systems in which establishment of a global dynamic model of the controlled system is not necessary. A modified model-free adaptive predictive controller is designed by combination of MPC and model-free control theory. The novel idea of this contribution is that by using the compact-form dynamic linearization technique, the upcoming system outputs within a specified prediction horizon can be predicted in sequence. The data-based prediction model of the system only requires input/output information, and therefore the future control input increments as well as the unknown system parameters called pseudo-jacobian matrix can be estimated. To improve parameter estimation accuracy, another online estimation method namely recursive least-squares algorithm is applied instead of using the conventional projection algorithm. The control performance is verified nummerically for vibration control of a flexible ship-mounted crane represented as a multi-input multi-output (MIMO) system. Simulation results indicate that significant reduction of the crane oscillations and better control performance are observed when using the proposed controller in comparison with other traditional methods.

Model-free adaptive control (MFAC) is a data-driven control approach receiving increased attention in the last years. Different model-free-based control strategies are proposed to design adaptive controllers when mathematical models of the controlled systems should not be used or are not available. Using only measurements (I/O data) from the system, a feedback controller is generated without the need of any structural information about the controlled plant. In this contribution an improved MFAC is discussed for control of unknown multivariable flexible systems. The main improvement in control input calculation is based on the consideration of output tracking errors and its variations. A new updated control input algorithm is developed. The novel idea is firstly applied for controlling vibrations of a MIMO ship-mounted crane. The control efficiency is verified via numerical simulations. The simulation results demonstrate that vibrations of the elastic boom and the payload of the crane can be reduced significantly and better control performance is obtained when using the proposed controller compared to standard model-free adaptive and PI controllers.