Ceramic Tile Production Intelligent Decision Research Based on Reinforcement Learning Algorithm

Ceramic tile production includes a complex decision system, which involves several intelligent decision acts and might affect the product quality. In general, traditional ceramic tile production utilized many repeated empirical experiments based on their engineers to determine an appropriate production parameter and pursue the desired product quality. However, it is observed that traditional ceramic tile production mainly depends on empirical experiments and couldn’t ensure a stable product quality. Moreover, the various surrounding environments for ceramic tile production might further result in a worse product quality when the empirical production parameters determined by empirical experiments couldn’t be adjusted by the actual situation. To solve the issue that empirical production parameters determination in the traditional ceramic tile production, a ceramic tile production intelligent decision framework is firstly designed based on reinforcement learning algorithm (i.e., Deep Q-networks (DQN)) in the paper. In the framework, both environment and agent modules are built, where environment module is designed to simulate various surrounding environments for ceramic tile production and then predict the corresponding product quality in time by a self-prediction random forest (RF) model. In addition, agent module aims to rapidly adjust the production parameters adaptively based on the predicted product quality to achieve a desired final product quality. The experiment results indicate that proposed ceramic tile production intelligent decision framework could effectively solve adaptive production parameters determination issues in the practice.