693 / 2022-03-31 15:47:31

Artificial Intelligence Aided Design for Film Capacitors

dielectric materials,film capacitors,capacitance,machine learning

Driven by energy-related demands and the efforts of many researchers, film capacitors, with the stability of their electrical values over long durations, have become key devices in many fields, especially high current pulse loads or high AC loads in electrical systems application scenarios. With the increase in application requirements and the expansion of the application range from commercial and military, the personalized customization of film capacitors that are different from the mass production for specific application conditions becomes more and more important. At the same time, the efficiency and greening of the production process are also looking forward to the innovation of the design system of film capacitors. However, the production process of film capacitors is complex, and it is difficult to derive the functional relationship between production parameters and product performance. On the other hand, the historical accumulation of the film capacitor industry makes the relevant data resources relatively abundant. Because of the above situation, based on the back propagation neural network theory, this paper builds a film capacitors design model by learning the design and performance data of 43,684 film capacitors collected by authors, thereby establishing the relationship between design parameters and performance indicators. After that，according to the established model and the given dielectric materials, the capacitance values of produced film capacitors are predicted, and then the appropriate dielectric materials are screened out through reverse design according to the established model and the expected capacitance values of film capacitors. Furthermore, this paper analyzes the distribution characteristics of the predicted value and absolute error, mean absolute error, mean square error, and root mean square error on different forecast intervals under the two forecast directions. Moreover, this paper studies the influence of the number of hidden layer nodes and the number of computations on prediction accuracy. The results of the analysis demonstrate the great potential of the proposed method to support film capacitor industry practitioners in making more efficient and cost-effective decisions by minimizing the effort required for the design-production-test iteration process. Finally, this paper discusses the construction of a multi-dimensional data learning model considering more design parameters and performance indicators, and looks forward to the future application of physics-data dual drive and causal inference in the artificial intelligence aided design for film capacitors.