A method for predicting mechanical properties of composite microstructure with reduced dataset based on transfer learning. (1st November 2021)

• Record Type: Journal Article
• Authors: Xu, Yangjian; Weng, Huanbo; Ju, Xiaozhe; Ruan, Hongshi; Chen, Junjun; Nan, Chenyu; Guo, Jing; Liang, Lihua
• Is Part Of: Composite structures. Volume 275(2021)
• Publication Date: 2021-11-01
• Subjects: Composites -- Transfer learning -- Active learning -- Mechanical properties -- Reduced dataset; Composite construction -- Periodicals; Composites -- Périodiques; 624.18
• DOI: 10.1016/j.compstruct.2021.114444
• Languages: English
• ISSNs: 0263-8223

Abstract

Abstract: Establishing structure-property linkages of composites by machine learning with limited labeled samples remains an ongoing challenge. A method for predicting mechanical properties of two-phase composite with reduced dataset based on transfer learning is proposed. In this method, an analytical presentation regarding composite microstructure characteristics is proposed and used to acquire the sample label, by which sufficient dataset can be established effortlessly and used to pre-train the initial convolutional neural network (CNN) in the source domain during transfer learning. Further, more training-value samples in the target domain are generated using a cluster analysis method, and their mechanical properties, acquired by high-cost finite element analysis or experimental test, are used to label samples in the target domain, with which the network parameters are further fine-tuned for ultimately forming a surrogate prediction model. The verification example shows that the present method can use about 1/2 training samples in the target domain to achieve a similar accuracy of the conventional CNN method. In addition, a novel method for microstructure optimization of composites is further developed based on the transfer learning method and verified efficient and accurate. The strong universality of the present method allows for an extension to multiphase composites or other complicated situation.