In this paper, a crack detection method is presented to detect Euler-Bernoulli beams containing arbitrary number of transverse cracks. The proposed method uses the time domain signal of the free vibration response to provide the position and depth of cracking of the Euler-Bernoulli beam that is modeled with a modified stiffness using the Spring model, with high accuracy and precision. The time history responses used in this paper are nodal computational accelerations at certain points of the beam exposed to impact load. The acceleration of the nodes is calculated with the Newmark beta method at the edge of the elastic beam`s superelements. Initially, using the computational time history of the damaged beam and the analytical model of the Euler-Bernoulli beam, the objective function of the failure detection problem, to be optimized by particle swarm algorithm, is defined and, intensity and location of transverse cracks are calculated by solving the optimization problem in Matlab environment. In order to determine the accuracy of the proposed method, three beam samples with different cracks and loadings are considered. In the first sample, the crack supposed to be in the superelements of beam and the beam considered to be with four elements as superelements. The second one has ten elements and same loading as previous. The third one has twenty elements and the loading is on the second element. All of the loadings are impulse loads. The comparison of the results of a four elements beam with the primitive conditions shows that accuracy of the deducted results were exactly matched. For the ten element beam, the results were satisfying but in the twenty element beam with asymmetric loading, obtained results indicate imprecise match.

To determine the accuracy of the developed model in real environmental conditions, different percentages of noise were added to the data of all three samples. These noise addition to data, contain 1, 3, 5 Percentages of noise.  The results show that the model presented in the presence of noise also provides accurate results and the model is not sensitive to the presence of noise in the data applied to three samples. Considering different number of elements in each sample, no convergence was observed, Also the results were not sensitive to the location of impact load applied on the samples.

The results of this study indicated that asymmetric cracking and loading variation are very effective in predicting beam failure. The results also indicate that variable reduction is very effective on the accuracy of results.

Having more cracks and therefore more elements to analyze will yield to less accurate results. To lessen these inaccuracies, it can be practical to achieve better results by assuming the location of crack is constant in the element length if the depth of crack be the matter of importance.

The number of iterations that have been executed, indicate that the pace of convergences in the developed process is less than when PSO deployed solely. This speed rate for obtaining results makes the developed method practical for solving crack detection problems in structures.