Data-Driven State Classification for Energy Modeling of Machine Tools Using Power Signals and Part-Program Information

There has been a growing interest in assessing the environmental impact and energy consumption of production facilities, resulting in the creation of various models for predicting energy usage of manufacturing equipment. These models are primarily based on empirical data gathered from field experiments, but the heterogeneity and complexity of the machines pose challenges in creating a model that can accurately represent specific machines or their components. Additionally, characterizing energy states can be costly in terms of time and experimentation, and experiments must often be repeated due to the changing nature of manufacturing equipment, such as changes in process conditions, degradation, or modifications made by operators. This study aims to address this issue by exploring a data-driven approach to classify energy models of machines by incorporating both field observations and prior knowledge. Machine learning algorithms are utilized for this purpose, using monitored signals like power requests and part-program information to guide the algorithms in a real-world application. The goal is to address the energy state classification problem as a first step toward developing autonomous energy modeling algorithms for machine tools.