부산대학교 도서관

This study takes into account the multi-point thermal model's automatic adjustment mechanism. The combined heat balance test temperature is used as the genetic algorithm's optimization target value. To minimize time and resources when running simulations, a back propagation neural network-based surrogate model between temperature and parameters is used. The best parameter set is discovered when the thermal model is modified at high temperatures. The overall temperature distribution of the thermal model is significantly improved by the 50% decrease in the heat transfer coefficient X1 of the joint body and the multi-layer insulation assembly, among others. The model's root-mean-square error before and after the adjustment is decreased from 4.51 °C to 0.95 °C. The best parameter set has been effectively contrasted and validated by reintroducing it into the low temperature environment. When the model is updated, the root-mean-square error drops from 2.73 °C to 1.09 °C. Following the model update, the joint temperature distribution is more in accord with the experimental findings, and all errors are within 2 °C. Therefore, the finite element thermal model of a space manipulator joint can be altered using a combination of back propagation neural network and genetic algorithm.