Maximizing Efficiency in Digital Twin Generation Through Hyperparameter Optimization

In recent years, digitalization has become widespread in many fields, including manufacturing systems. Following the deployment of Industry 4.0 technologies in manufacturing, significant improvements have been noted in many production areas, such as predictive maintenance, production planning, dynamic scheduling, and more. The digital twin is one of the technologies used in this area. The objective of this paper is to develop an automated and executable digital twin process within the concept of industrial artificial intelligence. For this purpose, hyperparameter optimization is used, a method to find the best external parameters that should be set by machine learning algorithm developers. It is used to automate the process and increase its accuracy by selecting the optimal parameters that are appropriate for the dataset. In this study, a digital twin was created using the random forest method on a dataset obtained from a CNC machine for a predictive maintenance application. Hyperparameter optimization is integrated into the machine learning process and the random forest hyperparameters such as bootstrap, maximum depth, maximum features, minimum sample leaf, minimum sample split, and the number of estimators is optimized. This study is a digital twin application under the concept of industrial artificial intelligence. A machine learning algorithm with integrated hyperparameter optimization is proposed, which is more efficient and faster in the field of predictive maintenance in production. For this purpose, the Hyperopt library in the Python programming language is used.