| dc.description.abstract | 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. | en_US |