The float glass process has been developed to allow the thickness to be controlled between 0.4 mm and 25 mm. The flow of glass into the tin bath is controlled via a device called tweel, as shown in Fig. 1, which is a refractory piece that penetrates into the glass melt and thereby increases or decreases the flow. This device also separates the refining atmosphere from the tin bath atmosphere. The glass temperature is around 1100°C in this section. The melt glass flows through an internal piece called the spout lip, shown in Fig. 1, into the bath of molten tin. Desired thickness of glass can be obtained by controlling the flow via the tweel. In this section, Table 1 List of input output parameters for Tin Bath Section Fig.3 A Radial Basis Function (RBF) Neural Network [10] International J ournal of Engineering Science and Innovative Technology (IJ ESIT) Volume 2, Issue 6, November 2013 Let us consider a case where we have n input variables. As all inputs are connected to each hidden node, each hidden node has an ‘n-dimensional’ centre. The selection, of the values of these centers and widths, is discussed in the subsequent paragraphs. Let I be the incoming vector for the hidden node with components, I, I)...1,. Then the output of the j" node, y,(1), in the hidden layer for the above input vector is given by: The performance of the network is defined by two parameters: Normalized Mean Square Error (NMSE), which should be minimum and the correlation coefficient(r), which should have a value near unity. These two parameters are defined as [10]: The method adopted for model development is given in Fig. 4. Table 2 Important Input-O utput parameters from point of view of operation Table 3 Input and Output parameters dropped by co-linearity analysis The results of co-linearity, sensitivity analysis and RBF NN models are discussed hereunder. C. Sensitivity Analysis are accounted. The number of nodes in the hidden layer is found by trial and error procedure such that the minimum MSE in the cross validation set is found as 0.038 during the training phase. It is required in order to obtain the best possible ‘NMSE’ and ‘r’ values for the output parameters during the testing phase of the above network as given in Table 4. It shows that the range of NMSE and R values are 0.3095 - 4.0095 and -0.0667 - 0.9209, racnactivaly Table 6 Range of the input-output parameters for RBF NN-2 A final RBF NN-Z model is developed with the topology (!QPQO-z) where 5!lUU iterations ana Conjugate Gradient as learning rule are accounted. The ranges of the values of the input-output parameters used for model development are summarized in Table 6. The number of nodes in the hidden layer is found by trial and error procedure such that the minimum MSE in the cross validation set is 0.053 during the training phase. It is carried out in order to obtain the best possible ‘NMSE’ and ‘r’ values for the output parameters during the testing phase of the above network as given in Table 7. The range of NMSE and R vary from 0.1282 to 1.8516 and from -0.0824 to 0.9443. Similarly, the developed model can further be used to carry out parametric study to find out the most effective inputs for a given output. Thus, the present work can be effectively used to screen best input which can control a mven aoumit International J ournal of Engineering Science and Innovative Technology (IJ ESIT) Volume 2, Issue 6, November 2013 Table 7 NMSE and r values of the testing data for RBF NN-2
![Fig.3 A Radial Basis Function (RBF) Neural Network [10] International J ournal of Engineering Science and Innovative Technology (IJ ESIT) Volume 2, Issue 6, November 2013 Let us consider a case where we have n input variables. As all inputs are connected to each hidden node, each hidden node has an ‘n-dimensional’ centre. The selection, of the values of these centers and widths, is discussed in the subsequent paragraphs. Let I be the incoming vector for the hidden node with components, I, I)...1,. Then the output of the j" node, y,(1), in the hidden layer for the above input vector is given by:](https://figures.academia-assets.com/34009112/figure_003.jpg)
![The performance of the network is defined by two parameters: Normalized Mean Square Error (NMSE), which should be minimum and the correlation coefficient(r), which should have a value near unity. These two parameters are defined as [10]: The method adopted for model development is given in Fig. 4.](https://figures.academia-assets.com/34009112/figure_004.jpg)
