This paper presents a new system of spring masses for crack detection in structures. For modeling cracks in arbitrary cross-sections, the Gaussian quadrature method is used to obtain the integral of the energy release rate over the cracked cross-section. In the proposed method, the system of spring masses is used to obtain the time histories of displacements, which are then combined into a relative time history by weighted summation. For detecting cracks, three springs and three equal masses with equal spacing from each other are placed on the beam, starting to move consecutively from one end. At each station, the time history of the system under excitation by an external load applied to the middle mass is extracted. The extracted data in each station shows the relative stiffness of that station compared to adjacent stations. The differences in system responses are used to estimate the position of cracks. The performance of the proposed method was evaluated using the Z24 bridge and a three-span concrete girder bridge in which a multiple-crack scenario was applied. These examples were used for evaluating the proposed system for the effect of parameters such as the weight and spacing of masses. The novelty of the presented method is to model cracks in arbitrary beam cross-sections and identify cracks in real bridge beams using a new effective mass-spring system that uses only three sensor data.