Cause-effect analysis of industrial alarm variables using transfer entropies

A method is proposed to analyze cause-effect relationships among binary-valued alarm variables. A normalized transfer entropy (NTE) and a normalized direct transfer entropy (NDTE) are formulated as basic statistical metrics, with consideration of two specific characteristics of alarm variables, namely, random delays and mutual independence among alarm occurrences. A modified statistical test is developed to determine significance thresholds of NTEs and NDTEs, based on which the cause-effect relationships can be inferred. The effectiveness of the proposed method is illustrated via numerical and industrial case studies.