Thermal behavior prediction of the spindle-bearing system based on the adaptive thermal network modeling method

Differences in comprehension of the heat transfer mechanism in spindle-bearing systems result in diverse thermal network modeling approaches. This leads to variations in simulation results of thermal prediction. To eliminate this epistemic uncertainty and determine the optimal thermal network model for the spindle-bearing system under various operating conditions, this study proposes an adaptive modeling method. Firstly, common thermal network models based on different mechanistic understandings are briefly introduced. Next, a convolutional neural network (CNN) surrogate model based on incremental learning is proposed to accurately approximate numerous thermal networks with limited simulations. Furthermore, considering that the spindle-bearing system comprises a multi-support structure, the grey relational analysis is utilized to mitigate the interaction effects of different support bearings on thermal characteristic analysis. Moreover, based on the CNN surrogate model, a two-step optimization method is proposed to obtain a thermal network model which balances accuracy and structural simplicity. Finally, the effectiveness and accuracy of the optimal thermal network model are validated through the thermal characteristic experiments of the spindle system under various operating conditions.