基于深度学习的3D打印混凝土蒸汽养护力学性能研究和抗压强度预测
Research on Mechanical Properties and Compressive Strength Prediction of Steam-Cured 3D Concrete Printing Based on Deep Learning
孙浚博 1王雨飞 2赵宏宇 3王翔宇4
作者信息
- 1. 重庆大学溧阳智慧城市研究院,常州 213300
- 2. 科廷大学设计与建筑环境学院,珀斯 WA6102
- 3. 重庆大学土木工程学院,重庆 401331
- 4. 华东交通大学土木建筑学院,南昌 330013
- 折叠
摘要
3D打印混凝土(3DCP)技术近年来获得广泛关注,然而,关于养护条件如何影响3DCP的力学性能的研究仍然较少.本研究主要探讨不同蒸汽养护条件(升温速率、恒温时间和恒温温度)对3D打印混凝土材料的力学性能影响规律.为了获得最佳蒸汽养护条件,通过正交试验研究了不同蒸汽养护条件下打印胶凝材料的力学各向异性.此外,基于室内试验数据,建立了条件表格生成对抗网络(CTGAN)用于扩充数据集,由291条数据扩充为1 000条数据,建立了一维残差卷积神经网络(1D-Residual CNN)用于预测3DCP的抗压强度,并建立了 5个机器学习(ML)模型用于对比,试验结果表明,CTGAN的数据增强技术可以有效提升1D-Residual CNN模型在3DCP抗压强度上的预测精度,R2最高为0.92.
Abstract
3D concrete printing(3DCP)technology has garnered extensive attention in recent years.However,few investigations focus on the effect of curing conditions on the mechanical properties of 3DCP.This study primarily investigated the influences of different steam curing conditions(temperature rise rate,sustained temperature time and sustained temperature)on the mechanical performance of 3DCP at various curing ages.To identify optimal steam curing conditions,an orthogonal experiment was conducted to study the mechanical anisotropy of printed cementitious material.Moreover,based on laboratory test data,a conditional tabular generative adversarial network(CTGAN)was established for data set augmentation,expanding from 291 to 1 000 data entries.A one-dimensional residual convolutional neural network(1D-Residual CNN)was developed to predict the compressive strength of 3DCP,accompanied by five machine learning(ML)models for comparison.Experimental results indicate that CTGAN's data augmentation technique significantly enhanced the predictive accuracy of the 1D-Residual CNN model on the compressive strength of 3DCP,with the highest R2 reaching 0.92.
关键词
3D打印混凝土/蒸汽养护/各向异性/抗压强度/深度学习/生成对抗网络Key words
3D concrete printing/steam curing/anisotropy/compressive strength/deep learning/generative adversarial network引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据