Minimization of Cable-Net Reflector Shape Error by Machine Learning

摘要

During assembly of a cable-net reflector, surface adjustments must be made to meet the stick accuracy requirement of a space mission by changing the lengths of adjustable cables. It is desirable to find the optimum adjustment amount based on the current state of the cable-net reflector. However, all data needed to establish an accurate simulation model for the cable-net reflector cannot be measured. Therefore, the cable-net reflector is treated as a black box, and a machine learning algorithm (support vector machine) is used to develop a prediction model to establish the relationship between cable length adjustments and surface accuracy. The k-fold cross-validation method is used to estimate the generalization error. The simulated annealing and grid search are combined to get the optimum hyperparameters for the prediction model. Next, a surface adjustment method is developed to calculate the optimal cable lengths. A tension truss reflector example is used to demonstrate the support vector machine prediction model and the adjustment method. The machine learning algorithm identified the relationships between cable lengths and surface accuracy. Both surface accuracy and tension uniformity can be improved by adjusting the boundary cables and tension ties with this method using only information from the front-side nodes.