Equivariant adaptive separation by independence (EASI) algorithm as an online blind structural identification is very important not only to better understand the structural response but also to conduct an efficient maintenance and management strategy. However, the traditional EASI algorithm is encountered by some drawbacks. It uses a constant step-size parameter and needs to establish a tradeoff between the misadjustment in the steady-state and the convergence rate. This paper proposes a new variable step-size equivariant adaptive source separation via independence (VS-EASI) algorithm for online blind modal identification of structures. Unlike the traditional EASI algorithm, the proposed algorithm adaptively updates its step-size based on the input signals and the un-mixing matrix, through establishing a new function between the step-size and the separating indicator. This will lead to better performance of the proposed method and the fast convergence speed is reached while the steady-state error is low. Furthermore, this algorithm mitigates the irrelevant noise making it more suitable than the EASI algorithm for practical application.Simulation results of synthetic examples and a benchmark structure verify the superior convergence and better performance of the proposed algorithm in the steady-state over conventional EASI with fixed step-size in stationary environments as well as non-stationary ones.