Journal
IEEE TRANSACTIONS ON POWER ELECTRONICS
Volume 34, Issue 8, Pages 7161-7171Publisher
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/TPEL.2018.2883947
Keywords
Artificial intelligence; automated design for reliability (ADfR); power electronic systems
Categories
Funding
- VILLUM FONDEN under the VILLUM Investigators Grant
Ask authors/readers for more resources
This paper proposes a new methodology for automated design of power electronic systems realized through the use of artificial intelligence. Existing approaches do not consider the system's reliability as a performance metric or are limited to reliability evaluation for a certain fixed set of design parameters. The method proposed in this paper establishes a functional relationship between design parameters and reliability metrics, and uses them as the basis for optimal design. The first step in this new framework is to create a nonparametric surrogate model of the power converter that can quickly map the variables characterizing the operating conditions (e.g., ambient temperature and irradiation) and design parameters (e.g., switching frequency and dc link voltage) into variables characterizing the thermal stress of a converter (e.g., mean temperature and temperature variation of its devices). This step can be carried out by training a dedicated artificial neural network (ANN) either on experimental or simulation data. The resulting network is named as ANN(1) and can be deployed as an accurate surrogate converter model. This model can then be used to quickly map the yearly mission profile into a thermal stress profile of any selected device for a large set of design parameter values. The resulting data is then used to train ANN(2), which becomes an overall system representation that explicitly maps the design parameters into a yearly lifetime consumption. To verify the proposed methodology, ANN(2) is deployed in conjunction with the standard converter design tools on an exemplary grid-connected PV converter case study. This study showed how to find the optimal balance between the reliability and output filter size in the system with respect to several design constraints. This paper is also accompanied by a comprehensive dataset that was used for training the ANNs.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available