Research On Multi-System Regulation And Diagnostic Technology For Large Reflector Antenna Service Performance

With the continuous demand for deep space exploration,large aperture,high precision and high gain are the inevitable trends in the development of large reflector antennas.On the one hand,the huge reflector makes the antenna more sensitive to environmental loads,which is easy to cause pointing error and gain loss;on the other hand,the antenna pointing accuracy and surface error are required to be within a certain accuracy range when high-frequency observation.Faced with such demanding requirements,it is important to study how to effectively increase the performance and efficiency of antenna observation.In order to solve the impact of environmental loads on the observation performance of large reflector antennas,high-performance antennas often use active control technology,including active main surface and active sub-surface technology,to compensate for the gain loss caused by antenna deformation,which can effectively increase the antenna observation time.However,the active surface technology also makes the control system very complicated,it is necessary to study the controllable and adjustable relationship between the main reflector,sub-reflector and servo system,so that the antenna control systems can work with each other;the health status of up to thousands of actuators also has an impact on the antenna performance,the maintenance cost is very high.It is necessary to study fast and accurate fault diagnosis methods,quickly locate the location of faults,and improve maintenance efficiency.Therefore,Therefore,this thesis aims to improve the observation efficiency of large-scale reflector antennas from the perspectives of control and structure.The work done is as follows:(1)Large reflector antennas are easily disturbed by environmental loads,which deforms the antenna reflector structure and causes performance losses.In this thesis,ANSYS simulation method is used to simulate the effect of wind disturbance on the performance of the antenna,for the deformed reflective surface,with the antenna pointing accuracy and gain as targets,Calculate the Best-Fit paraboloid parameters using the reflector panel node displacement information;according to the adjustable and controllable relationship between the active main surface,the active sub-surface and the azimuth-pitch system on the antenna performance,with the target pointing and Best-Fit paraboloid surface as goal,rotating the parabolic plane to the target pointing,translation,rotate the sub-surface to the best matching position.This adjustment method satisfies both point and gain requirements.(2)Aiming at the difficulty in modeling the antenna control system,the controller performance is highly dependent on the antenna modeling accuracy,and the problem of insufficient interference control capability,a deep reinforcement learning antenna multisystem control method based on Deep Deterministic Policy Gradient(DDPG)is proposed,and the method takes the uncertain interference such as wind disturbance as the random error of the network species to improve the anti-interference ability of the antenna system;the DDPG network aims at establishing the above multi-system adjustment amount.Through interactive iteration,the DDPG network learns the optimal adjustment strategy for different strategies.Finally,experimental simulation based on a 26-meter antenna model proved that the control performance based on the DDPG control strategy was overall better than that of the LQR controller in the pitch control system.In the antenna actuator network control system,the trained DDPG agent,for the wind speed of 15m/s(10m/s),through 6 rounds of regulation,the actuator control accuracy can reach 0.1mm(0.02mm),and the entire surface error of the antenna is controlled at 0.0118mm(0.0076mm).(3)The antenna servo system is a system with a high probability of antenna failure,and the transmission system failure can lead to a decrease in antenna performance.In this thesis,a detailed dynamic model of the antenna servo drive system is established,and seven different status data of the drive system are obtained through MATLAB simulation.The multi-source sensor feature-level fusion deep learning fault diagnosis model is designed,taking the timefrequency image as the input,using the convolutional neural network(CNN)to extract the image high-dimensional feature maps,merging the feature maps and then performing CNN operation,and finally through the global average pooling(GAP)and softmax classifier classification to achieve the purpose of fault diagnosis.This thesis trains the deep network with 3000 sets of different status data,and uses 200 sets of data as verification,and the fault recognition rate of the CNN+GAP diagnostic network after training reaches 96.47%,which is slightly higher than that of the CNN+FC(Fully Connected)traditional network.The training and verification of different sensor combinations,the single sensor recognition rate is up to 70.14%,the two sensors are fused up to 87.89%,but the accuracy rate of some states is less than 60%,and the recognition rate after the fusion of 3 groups of sensors is as high as92%,which shows that the deep fault diagnosis model established in this thesis has a very fusion effect and diagnostic accuracy.