.png){kind=icon}
.png){kind=icon}
Return
Variable time lag based temporal causality Bayesian network for cyclic loop in root cause diagnosis
.png){kind=icon}
PRE
AI
DOI：10.1016/j.psep.2025.108035.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Conventional methods for root cause diagnosis typically assume a constant time lag between variables, an assumption that does not always hold true given the dynamic characteristics of industrial processes. Currently, probabilistic methods are more popular than deterministic methods because they can more accurately reflect the actual complete state of the evaluated structure. However, the variable time lag makes the probabilistic relationship between causal variables no longer fixed and may form cyclic loops in the causal network. To effectively handle these loops, and more accurately describe the probabilistic relationship between cause and effect, a variable time lag Bayesian network (VTL-BN) is proposed, which gets rid of the assumption of fixed lag. Specifically, the Bayesian network is first modified to accommodate variable time lag, where the dynamic change of lag is explored based on dynamic time warping. Then, by examining the trend of time lag, the inherent causal loops are converted into temporal causal relationships. Due to this transformation, cyclic loops are broken down into acyclic ones across the time horizon, addressing the causal loops. Finally, the probability changes of variables are obtained by updating the likelihood evidence online, which allows for the identification of root cause and the inference of propagation path. The effectiveness of the proposed method is shown through two industrial process case studies.
Journal
IF:
7.8
Papers:
9.5K
Citations:
3.8W