Addressing Heterogeneous Time-Frequency Causality: Source Consistency Exploring for Industrial Root Cause Alignment...

Area of Science:

• Process monitoring and control
• Data-driven fault diagnosis
• Causal inference in complex systems

Background:

• Process variables exhibit temporal trends and periodicities, leading to distinct time-domain and frequency-domain causalities.
• This causal heterogeneity challenges accurate root cause diagnosis (RCD).
• Existing methods struggle to reconcile differing causal perspectives.

Purpose of the Study:

• To propose a novel framework, Causal Source Consistency Analytics (CSCA), for robust RCD.
• To achieve time-frequency synergy in Granger causality (GC) analysis.
• To overcome the causal heterogeneity challenge in fault diagnosis.

Main Methods:

• Developed a nonlinear enhancement module for temporal feature extraction.
• Designed a parallel causality learning module with differentiable frequency-domain expansion for time-domain and frequency-domain GC.
• Incorporated a time-frequency entropy constraint for causal significance and sparsity.
• Proposed a root cause alignment module using eigenvalue decomposition and approximate exponential transformation for end-to-end source consistency.

Main Results:

• The CSCA framework successfully identified root causes in the Tennessee Eastman process and a gas turbine application.
• Demonstrated that CSCA overcomes causal heterogeneity by enabling consistent root cause identification across domains.
• Ablation studies validated the effectiveness of the time-frequency synergy approach.

Conclusions:

• CSCA provides a unified approach to RCD by ensuring source consistency between time-domain and frequency-domain causalities.
• The framework effectively addresses the limitations of traditional methods facing causal heterogeneity.
• CSCA offers a promising solution for reliable fault diagnosis in industrial processes.