Abstract

Porcelain tile represents the best product ever developed within the class of ceramic tiles, due to its excellent tribo-mechanical properties and its aesthetic aspect, such as glossiness, which is a fundamental criterion for product quality control, due to its great appreciation by customers. The process of glossiness gain of porcelain tile consists of the rotation of abrasive blocks in contact with the tile surface, eliminating irregularities and adding value to the product by assigning the desired glossy effect. However, the study and optimization of this process, in order to obtain an ideal gloss value and a better glossiness distribution, require a very complex analytical model with many variables, making this investigation a complicated task. In addition, a common practice in the industry is to adjust the polishing process by trial and error. A promising alternative is the use of artificial neural networks model, due to its great ability to treat nonlinear problems and with lots of variables, intrinsic characteristics to the polishing process. Therefore, the goal of this work is to analyze the influence of some kinematic parameters on the gloss value of porcelain tiles. This analysis will take place with the implementation, coded in Python, of a set of eight identical neural networks, but with different input variables. The training data comes from computer simulations and gloss measurements. Each neural network will be able to predict the gloss value of a region of the tile based on kinematic parameters provided by computer simulations. The performance of each neural network will be analyzed based on the coefficient of determination, so that it will be possible to verify which group of parameters has the greatest capacity to explain the gloss variable. The purpose of developing this tool is to verify the potential of this model for possible applications both in the field of research and in the industrial sector.