• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文期刊>Reliability engineering & system safety >Deep-learning-based inverse structural design of a battery-pack system
【24h】

Deep-learning-based inverse structural design of a battery-pack system

机译：Deep-learning-based inverse structural design of a battery-pack system

获取原文
获取原文并翻译 | 示例
           
页面导航
  • 摘要
  • 著录项
  • 相关主题

摘要

? 2023 Elsevier LtdAlong with the continuous progress of lithium-ion batteries and the automotive industry, the safety of battery-pack systems (BPSs) is gradually becoming a hot topic of concern for consumers. A number of studies have been conducted on the safety performance of BPSs from various perspectives, with the aim of designing safer and more reliable BPSs. Designing a BPS is time-consuming because it contains numerous components and is located in a complex area. Researchers must conduct extensive finite element analysis (FEA) to select the right thickness of BPS components for a safe pack system. This process is tedious, and the historical data from FEA needs to be fully utilized. This work proposes an inverse prediction method for the BPS design for enhanced reliability. This method combines FEA historical data and a deep neural network (DNN) framework to predict the thickness of components, realizing robust BPS structural design. First, an FE model of a detailed BPS is built, the FEA under vibration conditions is completed, and the basic data for building the DNN model is acquired. Next, a DNN modeling framework, including forward and backward propagations, is developed to train the historical data. A DNN model that can describe the non-linear relationship between the inputs (the three main stresses and the minimum fatigue life) and the outputs (thicknesses of critical components) is obtained. Finally, the accuracy of the DNN model is investigated in terms of error functions, and the prediction accuracy of DNN models with different numbers of hidden layers is compared. The results show that the DNN model can accurately predict the component thickness of the BPS. In addition, a third-order response surface model and a radial basis function neural network model are used for comparison and verification. The prediction accuracy of DNN models built with different amounts of data, training epochs, and Gaussian noise is also compared. A case of structural optimization is given to provide designers with ideas. The proposed inverse prediction method can lead to economical and efficient design process. It can shorten the development cycle and reduce the cost of the BPS.

著录项

获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号