Multiobjective Optimization of an Observer-Based Controller: Theory and Experiments on an Underwater Grinding Robot

Abstract

Hydro-Québec operates hundreds of dikes and dams built many decades ago, and is committed to the long-term sustainability of these facilities. The researchers and engineers at its research institute have designed and manufactured a submersible grinding robot prototype capable of performing the important task of grinding underwater metallic structures. This robot, with direct drive linear motors, has an excellent dynamic performance, but lacks intrinsic stiffness. The design of a control system for such a system is a major challenge for control engineers, as the algorithm must achieve sufficient dynamic stiffness to withstand the significant external perturbations generated by the grinding process and the underwater environment, while at the same time minimizing the sensitivity of the control effort to measurement noise. A discrete-time observer-based control structure has already been proposed in a previous work for this purpose, but the empirical design procedure for this structure did not consider the two contradictory objectives. Using the same controller structure, we propose a new design methodology in this brief, based on a multiobjective genetic algorithm, for these mechatronic systems, which are highly prone to the effects of disturbances. The results obtained through optimization are compared with those obtained using the empirical method. The effectiveness of the proposed design is demonstrated through underwater grinding experiments using the robot test bench we have developed.