The use of supervised artificial intelligence methods in quality determination in continuous production lines: a case study of ceramic industry

In the modern manufacturing environment shaped by the Fourth Industrial Revolution, the critical role of data analytics and machine learning in enhancing quality management is increasingly gaining importance. The ability to predict final product quality at the early stages of the production process can aid in making manufacturing more efficient. Study underscores the significance of accurate quality predictions, which are intricately linked to the depth and scope of training data used in ML models, and presents an approach to overcome some key challenges in the ceramic manufacturing sector. Several traditional machine learning algorithms, including deep neural network algorithms, as well as multivariate analyses, were applied to develop predictive models for determining the final product quality in ceramic industries in the context of a pilot production process. The study encompasses a detailed case study that conducts an extensive comparison among machine learning models developed to predict the quality of tiles post-firing, based on an imbalanced dataset provided by the enterprise, enabling early prediction of tile quality before entering the kiln. Due to the inability to reach the desired minority class F1-Score and Cohen's Kappa values with machine learning classification models created using resampling methods, a deep neural network model capable of operating on an imbalanced dataset was developed for the first time specifically for the industry. The developed deep neural network model successfully predicted the minority class with an F1-Score of 0.75. The Cohen's Kappa value of the model was calculated to be 0.7413.