以機器學習結合排液容器法量測液體密度與黏度

NCU Institutional Repository > 工學院 > 機械工程研究所 > 博碩士論文 > Item 987654321/90191

請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/90191

題名:	以機器學習結合排液容器法量測液體密度與黏度
作者:	李漢祥;Li, Han-Xiang
貢獻者:	機械工程學系
關鍵詞:	排液容器法;計算流體力學;機器學習;密度;黏度;表面張力;draining vessel method;computational fluid dynamics;Machine Learning;density;viscosity;surface tension
日期:	2022-08-30
上傳時間:	2022-10-04 12:15:34 (UTC+8)
出版者:	國立中央大學
摘要:	現今量測物性之方法中，當需量測之物性眾多時往往需使用不同的精密儀器，其中少有同時量測數種物性之儀器，且需花費額外的時間以及費用，故本研究建立一方法學，以較低之成本、時間並可彈性因應不同操作條件的多物性量測方法。以排液容器法為基礎，建置與其相仿之實驗系統，不採用文獻以迭代較為耗時之方式得到物性，而是結合計算流體力學與機器學習，以機器學習XGBoost做為流體密度、黏度表面張力的物性回歸模型，使用計算流體力學計算排液速度做為物性回歸模型的訓練及測試資料，最後再以實驗所得的排液資料驗證物性回歸模型的準確度。本文以40℃不同濃度之甘油水溶液以及35℃不同濃度之丙二醇水溶液進行實驗，再將實驗資料預處理後輸入物性回歸模型量測其密度、黏度、表面張力。密度量測之平均相對誤差均達1.3%以下，黏度量測之平均相對誤差均達6.3%以下，因實驗流體均為水溶液，表面張力對流率的影響易被黏度消弭，故於甘油水溶液與丙二醇水溶液中表面張力量測之平均相對誤差分別為17.16%與33.32%。;In the current methods of measuring physical properties, when there are many physical properties to be measured, different precision instruments are often used. Among them, there are few instruments that measure several physical properties at the same time, and it takes extra time and cost. This research aims to establish a methodology, a multi-physical property measurement method with lower cost, time and flexibility to respond to different operating conditions. Based on the draining vessel method, a similar experimental system is built. Instead of using literature to obtain physical properties in a time-consuming iterative way, it combines computational fluid dynamics and machine learning to use machine learning XGBoost as the surface of fluid density and viscosity. For the physical property regression model of tension, computational fluid dynamics is used to calculate the fluid discharge rate as the training and test data of the physical property regression model, and finally the accuracy of the physical property regression model is verified by the fluid discharge data obtained from the experiment. In this paper, different concentrations of glycerol aqueous solution at 40°C and propylene glycol aqueous solution of different concentrations at 35°C were used for experiments, and the experimental data were preprocessed and input into a physical property regression model to measure the density, viscosity, and surface tension. The average relative error of density measurement is below 1.3%, and the average relative error of viscosity measurement is below 6.3%. Since the experimental fluids are all aqueous solutions, the influence of surface tension on the flow rate is easily eliminated by the viscosity. The average relative errors of surface tension measurement in aqueous solution are 17.16% and 33.32%, respectively.
顯示於類別:	[機械工程研究所] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	33	檢視/開啟

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....