| Summary: | 碩士 === 北台科學技術學院 === 機電整合研究所 === 94 === This thesis describes the combination of Integrating Computational Fluid Dynamics (CFD) with Back-propagation Neural Network (BNN) to predict multiple heat source chip temperature profiles.
This text presents 100 groups of training data by CFD. Every data included 70 groups of statistic, and the first 6 groups are 3 heat source seat punctuation marks within the chip as the input of BNN at the same time. On the other hand, the last 64 groups are the 64 temperature control points on the chip as the output of BNN. According to the learning ability of the BNN, the network corresponding relation between coordinates data input and temperature value output is established. It caused the mathematics model by BNN predicting that multiple heat source chips distribution. Finally, set up 16 groups of statistic database analyzed by CFD as the database testing the mathematics model. The result shows that BNN mathematics model with training finishes can be produced the temperature distribution and maximum temperature by estimating different coordinates position with 97% accuracy. The value of the heat source chip estimated by BNN is more efficient than CFD about times. It can offer industries effectively that speedy prediction and the best analytic application.
|
|---|
