Thermal reliability analysis of radiation-coupled heat transfer systems: An adaptive multi-layer importance sampling active learning kriging method

Structures involving radiative heat transfer are widely encountered in aerospace, advanced manufacturing, and energy systems, where thermal reliability analysis is often challenged by strong nonlinearity, irregular failure domains, and high computational cost. This study considers three representative radiation-coupled heat transfer problems, including a thin plate heat dissipation system (surface radiation), a wafer rapid thermal annealing process (surface-to-surface radiation), and a turbine blade thermal barrier coating (participating media radiation), aiming to develop a unified reliability analysis framework for different radiation mechanisms. To address these challenges, the present work concentrates on achieving adaptive exploration of failure regions while simultaneously reducing computational cost. Accordingly, an adaptive multi-layer importance sampling active learning Kriging (AK-AMLIS) method is developed. The method eliminates parameter scale disparities and constructs multiple importance sampling distributions associated with spherical layers, enabling adaptive expansion of the sampling domain and improved coverage of potential failure regions. Results from the three problems studies show that AK-AMLIS achieves high accuracy with significantly improved efficiency. Compared with Monte Carlo simulation (MCS), the computational cost is reduced by three to four orders of magnitude; compared with the classical AK method, the runtime can be reduced to 25% in strongly coupled problems; and compared with the optimised AK method also based on importance sampling, it demonstrates better stability and accuracy in capturing highly nonlinear and irregular failure domains. This work provides an efficient and robust computational framework for thermal reliability analysis of complex radiation-coupled heat transfer systems.