Operational mode discovery and causal analysis for cascading failure detection and environmental monitoring: A case study in automatic assembly systems

Abstract

Traditional predictive maintenance approaches in assembly systems often overlook the complex interplay between operational modes and their distinct failure propagation mechanisms. This study contributes to predictive maintenance within the Industry 5.0 paradigm by introducing an unsupervised operational-mode learning and causal forecasting framework designed to jointly enhance operational reliability and environmental sustainability. The proposed methodology integrates unsupervised operational-mode discovery, supervised classification for real-time mode identification, and mode-conditioned Granger causal analysis to uncover context-dependent causal dependencies among system variables, forming the causal foundation for mode-specific forecasting models. A three-station automatic assembly line with shared resources serves as the validation case study, using over 190 days of multi-sensor operational data. The mode-specific analysis reveals three fundamentally distinct causal topologies: a sustainability-driven mode, where carbon-footprint and power metrics act as dominant causal drivers; an operator-interaction-driven mode, where human–machine delays generate multi-path cascades across subsystems; and an emergency-propagation mode, where failures propagate bidirectionally between stations, forming tightly coupled cascading loops. By analyzing mode-specific causal structures, the proposed framework demonstrates improved interpretability of failure propagation mechanisms across operational contexts and provides a principled basis for developing mode-specific causal forecasting models. Moreover, the integration of environmental indicators into the causal layer enables a dual-objective analysis that links reliability degradation with sustainability impact. The findings highlight how mode-aware causal analysis can support context-adaptive and sustainability-oriented predictive maintenance, while positioning mode-specific multivariate regression forecasting as a natural extension of the proposed framework. • Introduces an operational mode discovery and causal analysis framework for Industry 5.0 assembly systems. • Integrates unsupervised and supervised learning to identify context-dependent failure propagation regimes. • Uses Granger causality to link sustainability indicators with reliability via mode-specific topologies. • Empirically validates the framework on assembly lines, proving stability over unstable global baselines. • Identifies dominant cascading failure pathways through spectral analysis and propagation intensity metrics.