การวัดปริมาณยางแห้งด้วยวงจรหกพอร์ต

Abstract

A development of a mathematical model for measuring dry rubber content (%DRC) based on reflected waves collected from a six-port reflectometer technique (SPR) is presented. The conventional model was developed based on single frequency reflected-wave datasets. Instead, this thesis proposes a model using reflected powers obtained from three frequencies (1, 1.5, and 2.16 GHz). Several dilute concentrate latex samples of 20-60 %DRC were prepared and the reflected powers were measured using the fabricated SPR. To develop the temperature-independent model, several latex samples from 20-45°C were prepared to measure reflected powers. The reflection powers are converted to DC voltages, digitized, and stored in the microcontroller embedded in the SPR. These datasets are split into three subsets for training, validating, and testing. Three neural algorithms; The Bayesian Regularization algorithm (BRA), The Levenberg-Marquardt algorithm (LMA), and The Scaled Conjugate Gradient (SCG); are applied as a learning tool for developing the models. It is found that the model uses three-frequency datasets and trained by the BRA algorithm is the best model among others. Compared with the true DRC datasets, the least mean square error (MSE) and the correlation coefficient (R) of the model are 0.1264 and 0.9997, respectively.