Abstract: The analysis object of this paper is a cylindrical pressure shell with a single orthogonally nozzle and reinforcement structures, consisting of a nozzle wall and an insert plate. Taking the non-iteration design process of the reinforcement structures of this cylindrical pressure shell as an example, this paper proposes a non-iteration design method based on the gradient boosting decision tree (GBDT). This work provides a reference for the design of structures without the need of iteration design process. The non-iteration design method proposed in this paper includes three parts: dataset generation, generation of the non-iteration design surrogate model, and development of a dedicated APP for non-iteration design. Firstly, a machine learning dataset is obtained by the parametric finite element analysis through ABAQUS-Python for the reinforcement structures of the cylindrical pressure shell. Then, based on the Gradient Boosting Decision Tree and the multi-output regressor, a non-iteration design surrogate model is generated based on the dataset. Finally, a dedicated forward design APP (HUST-FWD01) is developed to standardize the non-iteration design and verification processes for the reinforcement structures. The results show that the dimensions of the reinforcement structure can be obtained without the need of iteration by the non-iteration design surrogate model, which is generated based on the non-iteration design framework proposed in this paper. In the surrogate model, the maximum stress near the opening of the pressure shell is used as the input, and the dimensions of the reinforcement structures are the outputs. The verification examples show that the errors are all within 3% between model inputs and the FEA results corresponding to model outputs. The research results can provide a design method without the need of iteration for the reinforcement structures of cylindrical pressure shell with a single orthogonally hole as well as the structures akin to them.

Key words: non-iteration design, reinforcement structures, gradient boosting decision tree, APP development

摘要： 该文以耐压圆柱壳正交单圆孔的围壁-嵌入厚板联合加强结构为对象,提出一种基于梯度提升决策树（gradient boosting decision tree, GBDT）的正向设计方法,为构建免迭代结构设计范式提供借鉴。所提出的正向设计方法包含数据集生成、正向设计代理模型构建和正向设计专用APP开发这3个方面的研究内容。首先,基于ABAQUS-Python参数化有限元分析,获得正交单圆孔圆柱壳加强结构的机器学习数据集;然后,基于GBDT和多输出回归器,通过挖掘数据集形成了加强结构正向设计代理模型;最后,开发了专用正向设计APP（HUST-FWD01）,实现加强结构正向设计与准确性验证流程的固化。结果表明：提出的正向设计框架无需迭代,即可完成以“目标孔边应力最大值”为输入、以“加强结构设计方案”为输出的直接正向设计。验证算例表明：加强结构正向设计方案所对应的有限元计算结果与正向设计输入值之间的误差均在3%以内。研究成果可为耐压圆柱壳正交单圆孔加强结构以及类似结构的设计,提供一种无需迭代的设计方法。

关键词: 正向设计, 加强结构, 梯度提升决策树, APP开发