Prediction of ceramic glaze properties by artificial neural networks

Analyzing the current state of Turkish ceramic sector, it is seen that there is a continuous investment in technology and modern innovations are adopted. In this context, a great number of prototypes are to be developed in ceramic production sector either to improve the current glaze or to design a new glaze. In this research and development process, labour loss, time loss and therefore an increase in cost are reflected to the product cost. In order to minimize these shortcomings, it is aimed to estimate glaze luminosity and appropriate glaze components using data from production processes.In this study, recipes were prepared using washed kaolinite from Uşak province, lead tetroxide and quartz triple diagram and the samples were fired at two different temperatures (950 0C and 1150 0C). The input variables of artificial neural networks (ANN) were temperature, surface tension and expansion coefficient data. The output variables of ANN were component values of the glaze and vitrification value. The ANN model was designed using Toolbox of MatLab 2009 package program.The training results and test results were compared to the real values in order to check the network performance. As a result of training, high performance was observed for vitrification at regression value R2=0,7856, fault value 7,02%; for glaze components between values R2=0,8588 and R2=0,9894, fault values 8,63x10-3 and 0,63%. As a result of the test, as in training, high performance was observed in the network for vitrification at regression value R2=0,8167, fault value 7,02%; for glaze components between values R2=0,9996 and 0,9986, fault values 8,63x10-3 and 0,63%.Consequently, it was determined that the artificial neural network model could be used successfully in ceramic sector in achieving harmony between the glaze and the structure. At the same time, it was shown that it could lead a faster result by decreasing cost in Research & Development activities.