Support-vector modeling and optimization for microwave filters manufacturing using small data sets

This paper presents a support-vector modeling and optimization method to improve the electrical performance and yield rate of assembled microwave filters in the case of the scarcity of training data collected from the manufacturing process. In the method, a coupling model that reveals the effect of manufacturing precision on electrical performance of filters is developed by a multi-kernel linear programming support vector regression incorporating prior knowledge. Moreover, an expanded data strategy from a prior simulator has been introduced to solve the modeling problem of small data set. Finally, the electrical performance and mechanical structure are optimized by using the developed model, and the obtained results can assist the fabrication of the same filter in the future. Some experiments from an electrically tunable filter are carried out, and the results confirm the effectiveness of the proposed method. The method is particularly suited to an automatic tuning robot and a computer-aided manufacturing system of volume-producing filters.