The paper addresses the problem of model-based fault detection and isolation. Synthesis techniques for the design of fault detection filters are developed. Fault sensitivity objectives and time-domain constraints are handled in the design procedure, while guaranting robustness to noise, disturbances and modeling errors. The design method involves linear matrix inequality (LMI) optimization techniques and the generalized structured singular value. The approach is applied to a 3-Tanks system and experimental results demonstrate the potential of the proposed method.