A parametric study of adhesive bonded joints with composite material using black-box and grey-box machine learning methods: Deep neuron networks and genetic programming
A parametric study of adhesive bonded joints with composite material using black-box and grey-box machine learning methods: Deep neuron networks and genetic programming
复制标题
DOI：
10.1016/j.compositesb.2021.108894
复制
发表时间：
2021-04-30
影响因子：
13.1
通讯作者：
Hou, Xiaonan
中科院分区：
文献类型：
--
作者：
Gu, Zewen;Liu, Yiding;Hou, Xiaonan
由于在连接不同的和/或新型工程材料方面具有优势，航空航天、汽车和船舶工业中使用胶接接头的情况迅速增加。接头强度是评估胶接接头性能的关键特性。在本文中，提出了黑箱和灰箱机器学习（ML）模型的开发，通过考虑连续和离散设计（几何形状和材料）变量的混合，能够准确预测单搭接接头的破坏载荷。首先，通过有限元模型计算300个具有不同几何形状/材料参数的单搭接接头样本的破坏载荷，以生成一个数据集，其准确性通过实验结果进行了验证。然后，开发了一个深度神经网络（黑箱）和一个遗传编程（灰箱）模型，用于准确预测接头的破坏载荷。基于这两个ML模型，进行了一个案例研究，以探索特定设计变量与单搭接胶接接头整体力学性能之间的关系，并可获得结构和材料的优化设计。
The aerospace, automotive and marine industries have witnessed a rapid increase of using adhesive bonded joints due to their advantages in joining dissimilar and/or new engineering materials. Joint strength is the key property in evaluating the capability of the adhesive joint. In this paper, developments of black-box and grey-box machine learning (ML) models are presented to allow accurate predictions of the failure load of single lap joints by considering a mix of continuous and discrete design (geometry and material) variables. Firstly, the failure loads of 300 single lap joint samples with different geometry/material parameters are calculated by FE models to generate a data set of which accuracy is validated by experimental results. Then, a deep neuron network (blackbox) and a genetic programming (grey-box) model are developed for accurately predicting the failure load of the joint. Based on both ML models, a case study is conducted to explore the relationships between specific design variables and overall mechanical performances of the single lap adhesive joint, and optimal designs of structure and material can be obtained.