Abstract:The nonlinear character and amplitude of structural vibration will vary with the distance between load point and damage position. The structural damage information can be extracted from it. Approximate entropy can figure the complexity of a time serie. A moving load is put on the structure, calculate approximate entropy value of vibration data, and extract the nonlinear eigenvalue for the structural damage pattern recognition useing neural network. The validity of this method was reviewed by a beam under moving load. Moving load reinforces the structural nonlinear character. The results of simulation and experiment show that approximate entropy can figure the nonlinear grade of signal availably, and it is not sensitive to noise. Approximate entropy can be used for eigenvectors of pattern recognition based on neural network.