Details

Autor(en) / Beteiligte

• Yan, Bingxin
• Ma, Xiaobing
• Yang, Li
• Wang, Han
• Wu, Tianyi

Titel

A novel degradation-rate-volatility related effect Wiener process model with its extension to accelerated ageing data analysis

Ist Teil von

• Reliability engineering & system safety, 2020-12, Vol.204, p.107138, Article 107138

Ort / Verlag

Barking: Elsevier Ltd

Erscheinungsjahr

2020

Quelle

Alma/SFX Local Collection

Beschreibungen/Notizen

• •We propose a novel degradation-rate-volatility related effect Wiener process model.•Model-RV has high accuracy in ageing process modeling and RUL prediction.•We extend Model-RV to AAT by incorporating stress-acceleration models.•A set of 3S-60 silicon rubber accelerated ageing data is used for illustration. For some materials, positive correlations between degradation rate and volatility are common observations affecting health evolution processes, referred as degradation-rate-volatility related effect. Ignoring such correlations may weaken the stability and precision of prognosis outcomes, ultimately misleading maintenance decision-makings. However, few attempts explore physical integrations of rate-volatility correlations, while quantify the correlation pattern polluted by random effect. Besides, achieving high accuracy of degradation modeling emphasizes the influence of systematic error. We address such technical challenges by constructing a Wiener-based degradation model integrating: (a) time-variant drift and diffusion coefficients linked by a linear function and (b) a systematic error term independent of process variances. Joint statistical inference employing expectation maximization (EM) is executed, based on which the analytical structure of RUL distribution is constructed. We further extend the prognosis approach to accelerated ageing test (AAT) by incorporating the stress-acceleration models. A case study on degradation data of 3S-60 silicon rubber states the superior performance of the proposed approach in reducing fitting error of degradation trajectory as well as improving the RUL prediction accuracy.