Open Access

Junyu QiORCiD, Yannick UhlmannORCiD, Philipp Czerwenka, Jannik Maier, Zhuyun ChenORCiD, Gernot SchullerusORCiD

• Roller chain systems play a crucial role in industrial automation, yet their condition monitoring remains challenging due to the difficulty of sensor placement on moving components. This research proposes a sensorless health assessment framework MCTSDE that leverages motor driver data (torque and position), multivariate analysis, cyclic spectral coherence, and a long-term Time-series furcating Dense Encoder network for robust degradation assessment. The developed approach extracts meaningful health indicators directly from motor signals, eliminating the need for physical sensors. Experimental validation on a custom-designed roller chain testbed demonstrates that the proposed method significantly outperforms most classical indicators and several deep learning models-based indicators, in terms of the monotonicity and trendability over the degradation process. Furthermore, the proposed framework surpasses Fourier transform-based feature extraction (MFTSDE), highlighting the effectiveness of cyclostationary analysis for capturing degradation patterns. By integrating sensorless monitoring, cyclostationary analysis, and advanced multivariate time-series modeling, this research establishes an effective and cost-saving solution, paving the way for improved predictive maintenance strategies in industrial applications, especially for roller chain system.