A spatial-temporal fusion based nonlinear causality analysis framework for root cause diagnosis of faults in...

In order to ensure the product quality and safe operation of automation systems, it is very important to perform efficient and accurate root cause diagnosis of faults in industrial processes. However, some traditional methods can only be used to analyze the linear causalities, and assume that the time series meet linear Gaussian assumption and are stationary after faults occur. Due to the fault information may propagated along with the causalities among process variables, the nonstationary and asymmetric characteristics make the time series regression models poorly fit, and the accuracy of causality analysis may be affected. Inspired by the above issues, in this paper, a new spatial-temporal fusion based nonlinear causality analysis framework is proposed for root cause diagnosis of faults in nonstationary industrial processes with asymmetric distribution. In particular, the problem of causality analysis for the coexistence of stationary and nonstationary time series, as well as the coexistence of symmetrical and asymmetrical distribution time series is given more attention. Firstly, a beta distribution based variational autoencoder is constructed to extract the asymmetric features of time series in industrial processes. Subsequently, a spatial-temporal fusion adjacency matrix is introduced by fast dynamic time warping, and the spatial-temporal fusion nonlinear Granger causality analysis is performed for diagnosing the root causes of faults. Finally, two datasets from the hot rolling process are used to verify the effectiveness and performance of the proposed framework.