摘要:AbstractThree new methods are proposed in this paper to reduce the gap between the structured singular valueμand its upper bounds with a reasonable computation time. The first one is based on the classical (D,G)-scalings formulation, the second one on a more general multiplier based approach and the third one on an advanced branch-and-bound algorithm. They all use theμ-sensitivities to evaluate which uncertainties are worth considering for either partly solving some costly LMIs or intelligently cutting the uncertainty domain. They are successfully assessed on a list of realistic benchmarks.