A Reliability Assessment Methodology for Products in Cold Environments Based on Bayesian Nonparametric Theory

Cryogenic reliability assessment for reusable liquid rocket engine components is challenged by small sample size, heterogeneous failure modes, right censoring, and systematic under-coverage of remaining useful life (RUL) prediction intervals under limited observations. This paper presents two complementary, independently deployable tools tailored for this regime: (1) a Bayesian nonparametric stratification method via a Dirichlet Process Gaussian Mixture Model (DP-GMM) for heterogeneity-aware reliability estimation under right censoring; (2) a Gaussian process (GP)-based RUL prediction pipeline with interval-based conformal calibration, which restores near-nominal interval coverage under sparse observations. We further clarify the applicable boundary of mode-aware RUL prediction in small-sample settings. A cryogenic-physics-inspired simulation study and external validation on the NASA C-MAPSS benchmark demonstrate the effectiveness and robustness of the proposed methods.

Keywords: Bayesian nonparametrics, right censoring, Gaussian processes, remaining useful life, conformal prediction.