本研究是利用外徑3.5 mm、內徑3 mm之不鏽鋼管製作總長2 m、九個折彎數的震盪式熱管，文章中包含製作、真空處理、填充及封裝，並利用去離子水做為其工作流體，針對不同填充量、不同風速及不同加熱量下進行測試與分析，並利用外徑6 mm、內徑3 mm耐熱玻璃管製作可視化玻璃震盪式熱管，對其不同填充率下震盪的效果進行比較。將實驗所獲得的結果利用類神經網路進行訓練，並利用網路運算結果與其它數據進行比對及分析，探討不同網路結構及輸入參數對預測結果的影響。 結果顯示流體填充率適用範圍為30%~70%，利用填充率、加熱量、風速為輸入值，整體熱阻值為輸出值，再以不同輸入組合代入不同層數之類神經網路進行比較，結果顯示輸入組合15%、40%、60%及80%效果最佳，以及輸入單層網路的結果較雙層網路者為佳，其預測之平均誤差最佳為0.0541 K/W。 This research utilizes stainless steel tube having external and internal diameter with 3.5 mm and 3 mm to manufacture closed loop pulsating heat pipe. The study includes manufacturing process and the vacuuming management for filling and packaging. The experiment use D.I. water as the working fluid. Different filling ratio, wind velocity and heating power are used to test the thermal performance. An Artificial Neural Network (ANN) is then trained with the above available test data. Fully connected feed forward multi-layer ANN configuration is adopted. The experiment result shows that the applicable filling ratio is between 30% and 70%. The ANN consists of three input nodes corresponding to the filling ratio, the heat input and the wind velocity and a single output node corresponding to the total thermal resistance. The result shows the best series of filling ratio are 15%, 40%, 60% and 80%. And the one hidden layer is better than two hidden layer, the best mean error is 0.0541K/W. The final part of the thesis also reports on preliminary experimental results of using Pyrex glass to manufacture a visual pulsating heat pipe to compare pulsating motion at different filling ratio.
