Abstract
Accurate real-time estimation of degradation parameters (DPs) in induction motors is critical for optimizing efficiency in coal mining operations. In this paper, a hybrid-driven method is proposed to address the challenges posed by the limited sensor data available for the motor. This method includes two models, mathematical response surrogate (MRS) and dual-graph time-frequency attention network (DGTFAN) model. MRS reduces motor’s mathematical model to a dimension-reduced information propagation model based on an improved encoder-decoder architecture. DGTFAN converts node time series data into a dimension-matched representation using dual-graph attention network with time-frequency processing. The hybrid-driven method addresses the challenges associated with reactive maintenance after motor failure and facilitates predictive maintenance of the motor. The effectiveness of the hybrid-driven method was extensively validated through structural ablation experiments. In addition, publicly available datasets of degraded motors, along with measured data from the shearer’s motor, were utilized to verify its practicality. The results demonstrate that the hybrid-driven method achieves high estimation accuracy and exhibits broad applicability.
Funding Information
  • Projects of the Ministry of Industry and Information Technology (TC220A04W-1/167)
  • National Natural Science Foundation of China (52474182)
  • Significant Science and Technology Project of Shanxi (202301010101002)
  • Priority Academic Program Development of Jiangsu Higher Education Institutions

This publication has 34 references indexed in Scilit: