Propagation of guided waves (GWs) in a welded tubular steel structure (WTSS) with rectangular cross-section is investigated using experimental analysis and finite element method (FEM) simulation with the purpose of identifying multiple welding damages on it. A sensor network made up of sixteen piezoelectric wafers is established on the WTSS, generating and collecting GWs. With the aid of a series of signal processing, a time-domain feature, termed 'time of maximal difference' (ToMD) is extracted from each captured GW signal. Base on the ToMDs, a novel concept 'damage presence probability' (DPP) is consequently defined. Finally, a probability-based damage imaging approach and a two-level image fusion scheme are adopted to predict presence and location of two different damages in welding zones of the WTSS as validation. Both simulation and experimental results demonstrate the effectiveness of the developed approach for identifying multiple damages in similar structures.