A neural network-assisted Bayesian approach for identifying thermal properties and interfacial thermal contact resistance in thermal protection structures

Thermal protection structures (TPS) are extensively utilized in aerospace heat shielding applications, as they fuse the advantages of multiple single-layer thermal protection materials. Thermal properties and thermal contact resistance (TCR) at TPS interfaces may differ from standard test results under service conditions due to factors such as pressure, temperature, and surface roughness. This study proposes a systematic method for identifying the thermal properties and TCR of a TPS composed of carbon/carbon and carbon/phenolic composites. To simplify the problem and improve identification accuracy, the TPS is decomposed into three response models, each targeting the identification of carbon/carbon thermal properties, carbon/phenolic thermal properties, and TCR. Accordingly, three artificial neural networks were developed to rapidly predict temperature responses, sensitivity analysis was employed to reduce the number of simultaneously identified parameters, and Bayesian networks were used to identify the thermal properties and TCR. The method was validated through quartz lamp heating experiments, with results showing a maximum deviation of 2.56% between the identified and measured temperatures. Based on the identification results, the interface temperature was further investigated, revealing a significant temperature difference of up to 111.87 K (with an average interfacial temperature of 1336.81 K and nearly zero contact pressure). The impact of ignoring TCR on the identification results was also analyzed, showing that accounting for TCR improves the prediction accuracy by 6.38% (57.35 K) compared to the model neglecting it.