Transaction on Machine Design Hajar Ait Lamkademe 14 2 2026 https://doi.org/10.6025/tmd/2026/14/2/85-97 https://www.dline.info/tmd/fulltext/v14n2/tmdv14n2_3.pdf Traditional predictive maintenance often lacks interpretability and fails to uncover the underlying physical mechanisms of equipment degradation. To address this critical gap, this study proposes a dynamic causal network framework that transitions maintenance strategies from purely predictive to fully prescriptive. Utilizing a comprehensive industrial machine sensor dataset, the research employs a multi stage analytical methodology integrating correlation analysis, mutual information, Granger causality testing, and rolling window dynamic network analysis. The empirical results reveal exceptionally strong linear and nonlinear dependencies among core operational variables, particularly temperature, vibration, and pressure. Furthermore, Granger causality analysis establishes directional information flows, successfully identifying vibration as a critical precursor to defect generation. Crucially, the rolling window analysis demonstrates that causal connectivity is non stationary and fluctuates significantly over time. Periods of elevated causal density correspond to degradation accumulation or abnormal system behaviors, while reduced connectivity indicates post maintenance recovery or stable operation. These findings indicate that machine health is reflected not only in individual sensor magnitudes but also in the evolving structural interactions among variables. Consequently, dynamic causal connectivity serves as a novel, highly interpretable health indicator. By uncovering the directional propagation of degradation, the proposed framework provides actionable insights for root cause diagnosis and early fault detection, ultimately supporting more proactive, targeted, and efficient maintenance decision making in smart manufacturing environments.