SE-ResNet based disturbance identification algorithm for microthrust measurement system

Micronewton thrusters play a crucial role in the aerospace field, where the accuracy of micronewton thrust measurement is significantly impacted by environmental vibrations. However, existing methods for identifying vibration disturbances often fall short in terms of accuracy, especially under non-stationary working conditions. To address this issue, this study introduces a residual neural network (ResNet) to recognize impulse interferences and step signals and experimentally evaluates its performance. The Squeeze-and-Excitation (SE) module is incorporated to optimize the network, as it can adaptively enhance important features. This results in SE-ResNet having enhanced channel attention mechanisms, which improve the accuracy of disturbance recognition. The experimental results demonstrate that SE-ResNet can accurately identify the impulse, step, and steady-state responses of cantilever beams with an accuracy of 93.54% under the real-time control system for the microthrust measurement of cantilever beams. This is a notable improvement over the 88.91% accuracy achieved by ResNet alone. The proposed method shows great potential for providing a foundation for subsequent interference suppression against impulses and steps.