Analysis of time-varying cause-effect relations based on qualitative trends and change amplitudes

Determination of root causes for occurring abnormalities is a sophisticated work in process monitoring of complex process industries. Many existing methods are based on an assumption that causal-effect relations are time invariant, whereas real industrial processes may have complex dynamic characteristics, leading to time-varying casual-effect relations. In this paper, a new cause-effect detection method is proposed to analyze time-varying cause-effect relations between the key variable and influence variables. This study is motivated by a common observation in practice that if the key variable takes an increasing/decreasing amplitude change, then corresponding influence variables are with amplitude changes in the same/opposite qualitative trends, and such relations between trend changes may change with time. The proposed method consists of three main steps: i) the time series of the key variable is represented by short straight lines to formulate piecewise linear representation (PLR) segments, ii) qualitative trends of involved variables are extracted on the basis of PLR segments, after estimating time delays between influence variables and the key variable, iii) contribution factors of influence variables to changes of the key variable are calculated by building a multiple linear regression model for normalized amplitude changes of involved variables. The effectiveness of the proposed method is demonstrated by numerical and industrial case studies. (C) 2022 Elsevier Ltd. All rights reserved.

Automated summary

This paper proposes a new cause-effect detection method to analyze time-varying cause-effect relations in complex process industries. The method utilizes piecewise linear representation of the key variable's time series, extracts qualitative trends of involved variables, and calculates the contribution factors of influence variables to the changes of the key variable.