A pinch roll performance degradation evaluation method and system based on digital twinning
By constructing a hybrid transfer learning model of multibody dynamics finite element model and deep belief network, the problems of sample scarcity and model fidelity in the performance degradation assessment of pinch rolls are solved, achieving high-precision pinch roll condition diagnosis and supporting predictive maintenance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUANGGANG NORMAL UNIV
- Filing Date
- 2026-05-07
- Publication Date
- 2026-07-17
AI Technical Summary
In existing technologies, the maintenance strategies for pinch rolls suffer from over-maintenance and under-maintenance, leading to high costs and equipment failures. Furthermore, digital twin technology faces challenges in assessing the performance degradation of pinch rolls, including difficulties in sensor deployment, sample scarcity, and difficulty in ensuring model fidelity, resulting in low assessment accuracy.
A multibody dynamics finite element model was constructed, and thermo-structure interaction analysis was performed. Combined with simulation-measured data hybrid transfer learning of deep belief network, simulation data similar to the measured data was generated through dynamic threshold adaptive screening and multi-cascade screening. Pre-training and parameter transfer were then performed to establish a DBN diagnostic model.
It achieves high-precision automatic diagnosis of the normal, moderate wear and severe damage states of the pinch roller, with a diagnostic accuracy of 88.23%, meeting the requirements of engineering applications and providing technical support for predictive maintenance.
Smart Images
Figure CN122413832A_ABST