Font Size: 
An Improved Nested Sampling Method for Path Parameter Inference of Variable Stiffness Composite
Xin Wang, Hu Wang, Yong Cai, Guangyao Li

Last modified: 2020-07-20


Faced with complex design space and constraint conditions, Bayesian inference is another efficient parameter inversion method to inference a reasonable scheme of fiber laying path described by polynomial function for variable-stiffness composite other than traditional optimization algorithms. Nested Sampling (NS) is a newly developed and highly efficient sampling algorithm which has been successfully integrated into Bayesian inference for model selection. However, a constrained condition sampling step exists in each iteration step in the original NS algorithm, which generates new samples with higher likelihood from prior space to replace the sample that has the lowest likelihood evaluated in the previous iteration. The progress has an important effect on the efficiency of the algorithm. This paper presents an improved Nested Sampling (INS) to promote the proposal of effective parameter samples by introducing spatial position updating mechanism of Meta-heuristic algorithm into standard Nested Sampling instead of drawing new samples. In addition, Variational Auto Encoder (VAE), one of the unsupervised learning frameworks, is also employed to extract the characteristic value of design variables which are utilized to construct the mapping relationship with variable-stiffness composite performance index. Such an indirect network framework not only more accurately replicates the relationship between design variables and target performance, but also improve the computational efficiency. The contribution of this work is to build a reverse network integrating a modified NS algorithm, which can get the design scheme of fiber path more efficiently and accurately.


numerical methods; algorithm

An account with this site is required in order to view papers. Click here to create an account.