Topology optimization based thermal management of biped robot abdomen with neural network prediction

With widespread application of biped robot, their internal electronic components are evolving toward compact integration under high power density. To address issue of excessive operating temperature and delayed thermal management response in the printed circuit board and battery module of biped robot abdomen, the current work proposes a topology optimization based thermal management system with neural network prediction. A hybrid metal foam filled phase change material and microchannel cooling is used with topology optimization to balance heat transfer performance and flow resistance through a weighted bi-objective framework. Compared to straight parallel channel, with an inlet water temperature of 22.5 °C at velocity of 0.1 m/s, the maximum temperature of printed circuit board decreases from 45.32 °C to 40.87 °C by 9.82% by using topology optimized microchannel, and its maximum thermal stress is reduced from 13.27 MPa to 11.05 MPa by 16.73%. Meanwhile, the corresponding maximum temperature of battery module is dropped from 58.63 °C to 54.26 °C by 7.45%, while its maximum thermal stress is attenuated from 21.33 MPa to 19.65 MPa by 7.88%. The maximum temperature reductions per unit additional hydraulic pumping power are 28.20 °C/mW and 27.69 °C/mW for PCB and battery module, respectively. When the inlet water flow velocity is increased from 0.06 m/s to 0.14 m/s, the average temperature of printed circuit board and battery module is reduced by 14.20% and 10.06%, respectively. As the inlet water temperature is decreased from 27.5 °C to 17.5 °C, their average temperature is reduced by 21.18% and 16.82%, respectively. Besides, the temperature prediction performance of five neural network models is evaluated across different prediction horizons. Based on a sparse attention mechanism, the Informer model achieves better overall performance in prediction accuracy and computational efficiency. The current work integrates hybrid thermal management with neural network prediction, providing an integrated approach for heat dissipation and thermal prediction in the biped robot abdomen.