Prediction of Temperature Distribution in a Packed Spheres Channel Using Artificial Neural Network

• Main Authors: Tzung-Shi Lee, 李宗熹
• Other Authors: Jee-Ray Wang
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/70832895558666519359

碩士 === 建國科技大學 === 自動化工程系暨機電光系統研究所 === 96 === The Artificial Neural Network is adopted in this research to perform (an analysis) on the temperature distribution of stack copper beads flow channel under the forced convection conditions of different air flow volumes; it attempts to do an analysis calculation of the optimum temperature distribution. Artificial Neural Network is a tool for empirical model construct and it differs from the traditional statistics approach in the following ways: (1)It is able to construct a non-linear model. (2)The network could accept various types of input signals; it possesses the ability of adaptive learning. (3)It is not necessary to define relations between the input and output by mathematical model. The Artificial Neural Network is being applied in this research to calculate the temperature distribution of stack copper beads flow channel of different air flow volumes and it is able to estimate the temperatures of different locations in the flow channel; this is relative crucial to the analysis of temperature distribution. In order to verify the accuracy of this network learning data, in addition to the prediction of flow channel temperatures, that whether the sensors used to measure the temperatures are functioning normally is also a problem of comparative importance. Recording devices used in the experiment are connected through its testing segments by thermo couple; this is to avoid the partial data access errors caused by malfunction of signal wiring, the error may become the noise of Artificial Neural Network analysis and training. The network itself could perform correcting action on the sensors that could not function normally to enhance the network availability; the network is also provided with the capability of deviation compensation in addition to performing the prediction of temperature distribution. The empirical data adopted are obtained (under following conditions): by arranging the air to enter stack copper beads flow channel in vertical direction with two exits for cooling fluid and 8 types of flow variations of forced convections; in the aspects of bead-diameter parameter design, stack copper beads of 6 mm diameter was selected. It is able to acquire the analysis model with best precision for constructing the Artificial Neural Network by using the above parameters. The empirical design model could simulate the temperature distribution conditions of different flow volumes; the analysis results manifests that the Artificial Neural Network constructed in this research could effectively predict, for cases of different flow volumes parameters, the flow channel temperature distribution conditions. The research results have been verified that its average errors comparing to the experimental measurement value is within 1.2%. This certainly proves that Artificial Neural Network could effectively predict the temperature distribution of flow channel under different operating environments and provide the temperature data required by thermal transmission analysis, and, it could be used to verify the accuracy of the measurement data.