In this paper, a non-linear dynamic equation of a rotor-bearing system with looseness between the pedestal and the casing is established. The non-linear parameters of the pedestal looseness fault are identified by means of the method of parameter optimization of genetic algorithm (GA), based on the displacement response signal in the vertical direction of the shaft end which has looseness. In view of the traditional genetic algorithm having problems of slow evolution and premature convergence, an improved method based on the genetic process of the traditional genetic algorithm is proposed. The method can take a greater mutation rate, restrain the premature convergence problem, and to speed up the evolution. Studies show that the improved method can effectively improve the efficiency of loosening parameters identification. The non-linear parameters of pedestal looseness fault are identified by the improved genetic algorithm, and the influence of mutation rate on recognition results is studied. Finally，the looseness parameters identification based on true measured signal is performed.