A1224
Title: Bayesian modeling of degradation data with dependent increments using the t copula
Authors: I-Tang Yu - Tunghai University (Taiwan) [presenting]
Raf Loreto - Tunghai University (Italy)
Abstract: Degradation measures serve as surrogates of lifetime, that is, they allow to perform inference about the lifetime of a product unit based on the increments of its degradation path. To preserve mathematical tractability, most existing stochastic-process-based degradation models assume that increments from non-overlapping time intervals are independent, which limits their ability to capture temporal correlation commonly observed in real data. A Bayesian modeling framework is introduced that explicitly accounts for dependence among degradation log-increments by employing the t copula. A key advantage of this approach is its flexibility in accommodating arbitrary marginal distributions for the log-increments, which makes the proposed model applicable across diverse degradation processes. Computations are carried out using a Markov chain Monte Carlo algorithm, and the model is validated through an application to a real-world degradation dataset. Additionally, a simulation study is conducted to evaluate the model's goodness-of-fit, robustness, and general applicability. The simulation study confirms the effectiveness of the proposed approach in capturing complex degradation dynamics.
