Semi-physical simulation experiment of the multi-view non-cooperative measurement method for nozzle swing angle

The swash angle of a nozzle is a critical indicator of engine control accuracy. Visual inspection technology, with its non-contact and high-precision advantages, has attracted attention for measuring nozzle swash angle under static and dynamic conditions. However, existing methods that rely on cooperative marker points limit their applicability. To address the shortcomings of existing detection methods, this paper proposes a multi-view non-cooperative nozzle swash angle measurement method based on deep learning. To address the issue of 3D reconstruction of feature points in the multi-view image of the nozzle, a nozzle feature point detection model based on SuperPoint and a nozzle feature point matching model based on SuperGlue have been developed. A methodology is proposed to estimate the camera pose from 2D points to 3D points to complete the 3D reconstruction of feature points in the nozzle image. This methodology is based on the projection vector of feature point coordinates and the estimation of rotation parameters. A semi-physical simulation platform for non-cooperative measurement of the nozzle tilt angle is constructed to verify the multi-view non-cooperative measurement method for the nozzle tilt angle based on deep learning. The experimental results show that, in a static measurement scenario, the maximum measurement error of the multi-view non-cooperative measurement method for the nozzle angle is 0.12^◦ in the horizontal direction and 0.13^◦ in the vertical direction. In the dynamic measurement scenario, the measurement error of the tilt angle in the horizontal and vertical directions is controlled within an error range of ±0.15^◦, which meets the measurement requirements of the nozzle angle. This method does not rely on the cooperation information provided by the nozzle, simplifies the measurement method of the nozzle swing angle, and is highly reliable.