Descripción
A brief introduction to the project of implementation of deep learning models in rubber manufacturing processes is presented. The main objective of this project is to use machine learning models, specifically neural networks, with existing information and data from rubber manufacturing processes, solving, in particular, the quality control of products for sale.
Raw material and process data from PTE Compounding of Mexico were used to elaborate several neural network models, presenting the best iteration obtained from hundreds of tests and variations for the given conditions and configurations. With this, models 6 and 8 had the best result obtained by minimizing its cost functions, MSE (mean squared error), and MAE (mean absolute error).
In addition to the model metrics, it was also sought to predict new quality attribute values with new product data obtained during the development of this project. These model predictions were compared with actual results.
Finally, with the results obtained, it is concluded that the proposed approach of neural network models for rubber manufacturing seems to be heading in the right direction, considering the complexity of the interactions in daily practice.
