The observed vibration signal is always a mixture vibration signal because the operation condition is complex when the machine runs. This paper adopts the independent component analysis (ICA) to obtain the single signals from the mixture signal.The algorithm is performed by the fast fixedpoint iteration which need not choose the learning rate based on the implied target function of maximum likelihood estimation, and the subguassian and supe rguassian sources can be estimated at the same time by the chosen nonlinearity function applied to estimate the probability density function. The computer simu lations and the tests under especial conditions in engineering practice show that the algorithm is effective with simple, accuracy and faster convergence .