LMM-DPIM-based neural network model for the efficient stochastic shakedown and reliability analyses of printed circuit heat exchangers

The randomness in structural parameters of printed circuit heat exchangers (PCHEs) is unavoidable and can significantly impact shakedown limit and reliability. However, studies on the influence of random structural parameters are limited due to the expensive computation burden. Therefore, this paper aims to achieve efficient stochastic shakedown and reliability analyses of PCHEs with multiple random structural parameters. First, the Linear Matching Method (LMM) is utilized to perform the shakedown analysis. Then, to account for the influence of random structural parameters, the Direct Probability Integral Method (DPIM), as an efficient non-intrusive stochastic analysis method, is introduced. By exploiting the advantages of LMM and DPIM, a novel LMM-DPIM is proposed for the uncertainty qualification analysis of PCHEs. Furthermore, the data-driven neural network model is implemented to improve the computational efficiency. As a result, an LMM-DPIM-based neural network model is established. Finally, the high accuracy and efficiency of the proposed model are verified through comparisons with Monte Carlo simulation. Moreover, the results of uncertainty qualification analysis reveal the effects of different random structural parameters on the shakedown limit and reliability of PCHE. Particularly, the fillet radius at the corner of channels significantly influences shakedown limit, leading to a huge reduction in reliability.