The aero ball joint is pivotal in aircraft duct systems due to its favorable properties, including displacement compensation and flexibility. In the stress assessment of air ducts, ball joints are usually simplified by using “Joints” connections to reduce the convergence problems caused by non-linearity, which requires a high degree of accuracy in the characteristic parameters of the ball joint. Accordingly, this paper builds a high temperature and pressure fatigue test platform to investigate the bending characteristics of the ball joint at different temperatures and pressures and points out the limitations of the current method. Then, a method combining finite element analysis (FEA) and the BP neural network is proposed to obtain the characteristic parameters of the ball joint. The results showed that the bending process of the ball joint tended to have two typically different stiffness properties, which were high rigidity and low rigidity. The bending characteristics were strongly influenced by pressure, but less influenced by temperature. The existing test platform increased the force reaction at the contact areas of the ball joint, resulting in errors in the measurement of characteristic parameters. The BP neural network prediction method could effectively alter the ball joint properties and reduce errors.