Approximate Bayesian assisted inverse method for identification of parameters of variable stiffness composite laminates. (1st July 2021)

• Record Type: Journal Article
• Authors: Chen, Boxuan; Zeng, Yang; Wang, Hu; Li, Enying
• Is Part Of: Composite structures. Volume 267(2021)
• Publication Date: 2021-07-01
• Subjects: Adaptive nested sampling -- Auto-Encoder -- Neural Network -- Tikhonov regularization -- Variable Stiffness Composites; Composite construction -- Periodicals; Composites -- Périodiques; 624.18
• DOI: 10.1016/j.compstruct.2021.113853
• Languages: English
• ISSNs: 0263-8223

Abstract

Abstract: The uncertainties of parameters are important for the performance of Variable Stiffness (VS) composite laminate plates. When parameters of VS composite laminate are given, it is easy to investigate physical characteristics of VS composite laminate. However, it is difficult to identify values of parameters due to a typical ill-posed inverse problem. In order to address this problem, an innovative approximate Bayesian computation (ABC) is suggested to identify composite parameters by considering uncertainties in this study. Compared with traditional Bayesian framework, ABC can avoid the calculation of likelihood which makes inverse procedure more complex or even intractable in practice. Generally, four advanced techniques are integrated to satisfy demands of ABC method in this study. In the suggested framework, a powerful Auto-Encoder (AE) is used to reduce calculation of the response with little information loss. Sequentially, Tikhonov regularization is integrated into ABC. Furthermore, to reduce computational cost, a high sample accepted rate-adaptive nested sampling method and Neural Network (NN) used to construct the mapping between parameters and responses are utilized. Finally, the efficiency and flexibility of the proposed method are validated by three cases.