Current predictive maintenance (PdM) systems face significant challenges in detecting soft faults, such as control parameter drift and pneumatic component hysteresis. These faults are often not immediately detectable by traditional physical signal monitoring, which leads to a critical disconnect between microscopic component health and macroscopic production efficiency. To overcome this limitation, this paper proposes a Cycle Time Deviation (CTD) based predictive maintenance method, establishing it as a comprehensive equipment health indicator. The proposed methodology first statistically validates a significant correlation between CTD and equipment degradation. Subsequently, a hybrid deep learning model is employed to perform high-precision multi-step prediction of CTD trends, enabling the timely identification of early warning signals for maintenance intervention. Furthermore, a data-driven fault attribution model quantitatively analyzes the contribution of underlying physical parameters (such as servo operating power and vibration levels) to CTD anomalies, providing interpretable diagnostic results and successfully verifying field maintenance log records. This CTD framework has been validated on a critical automated production line at a motor manufacturing company in Shanghai, providing a unified, reliable, and proactive strategy for soft fault detection and targeted maintenance decision-making in complex automated manufacturing systems.
Zhang et al. (Sun,) studied this question.