Reliability analysis of complex systems with long memory effects

In system prognostics and health management, reliability analysis and degradation models are crucial for predicting the lifetime of complex systems. Recent studies have highlighted the significance of Long-Term Memory (LTM) effects in degradation processes, where future behaviors are strongly coupled with historical degradation patterns. Although most existing research has focused on the LTM effect in univariate degradation processes with sufficient data, there are scenarios where degradation data is insufficient and existing degradation models are not reliable for them. Furthermore, advancements in measurement technology enable access to multiple performance characteristics (PCs) that degrade over time, yet the integration of LTM into multi-PC models has been underexplored. In this dissertation, we first propose a novel LTM-integrated Multivariate Degradation Model (MDM) based on multivariate fractional Brownian motion (MFBM) to capture both global LTM and the cross-wise correlation among multiple PCs. Additionally in second research, another MDM based on the Generalized Cauchy (GC) process is introduced to simultaneously account for global LTM, local irregularities, and cross-wise correlations. For both models, a maximum likelihood method is developed for parameter estimation. Validation through simulations and physical experiments on solar energy devices demonstrates the superiority of the proposed MDMs in life prediction, addressing the common underestimation of lifetime uncertainty in traditional approaches. Lastly, to address the issue of insufficient degradation data, an AI-based degradation model using Physics-Informed Machine Learning and Neural Networks is proposed to predict the lifetime of multiple material degradation paths with LTM effects.

Author

• Asgari, Ali

Identifier

hdl:10057/56107

OAI identifier oai:identifier

oai:soar.wichita.edu:10057/56107