The work presented in this paper provides an insight into the current challenges to detect incipient damage in complex metallic structural components. The goal of this research is to improve the confidence level in diagnosis and damage localization technologies by developing a robust structural health management (SHM) framework. Improved methodologies are developed for reference-free localization of fatigue induced cracks in complex metallic structures. The methodologies for damage interrogation involve damage feature extraction using advanced signal processing tools and a probabilistic approach for damage detection and localization. Specifically, piezoelectric transducers are used in pitch-catch mode to interrogate the structure with guided Lamb waves. A novel time-frequency (TF) based signal processing technique based on the matching pursuit decomposition (MPD) algorithm is developed to extract time-of-flight damage features from dispersive guided wave sensor signals, followed by a Bayesian probabilistic approach used to optimally fuse multi-sensor information and localize the crack tip. The MPD algorithm decomposes a signal using localized TF atoms and can provide a highly concentrated TF representation. The Bayesian probabilistic framework enables the effective quantification and management of uncertainty. Experiments are conducted to validate the proposed detection and localization methods. Results presented will illustrate the usefulness of the developed approaches in detection and localization of damage in aluminum lug joints.