To ensure the calibration accuracy of tri-axial vibration sensors, it is necessary for the tri-axial standard vibrator to output low-coupling and low-distortion vibration signals. However, due to the complexity and nonlinearity of the tri-axial standard vibrator structure, the output signals exhibit significant residual coupling and harmonic distortion. To address this issue, this paper proposes a neural network inverse model online iterative decoupling control method to achieve decoupling between axes and improve the accuracy of the output signals. Firstly, taking the leaf-spring-type tri-axial standard vibrator as the research object, the decoupling structure and the coupling characteristics of inter-axis motion were theoretically analyzed, and the orthogonal vibration suppression ratio and harmonic distortion were quantified. Then, the inverse model of the tri-axial standard vibrator system was constructed using the neural network, and the original system was controlled by feedforward series method, which initially improved the inter-axis coupling interference and harmonic distortion. On this basis, to more accurately characterize the dynamic coupling characteristics of the tri-axial standard vibrator, an online iterative learning mechanism was introduced. By dynamically updating the samples, the inverse model was periodically iteratively optimized to gradually improve its fitting accuracy, thereby achieving precise control of the inter-axis coupling and harmonic distortion of the tri-axial standard vibrator. Finally, a vibration acquisition and analysis system was built based on AD7606B, and an experimental test system for the tri-axial standard vibrator was integrated and constructed. The experimental analysis results show that the neural network inverse model online iterative decoupling control method can control the orthogonal vibration suppression ratio of the tri-axial standard vibrator within 2% and the harmonic distortion within 1%, verifying the effectiveness of the decoupling control method and providing a technical reference for high-precision decoupling control of multi-axis vibration test system.