Towards a Data-Driven Process Monitoring for Machining Operations Using the Example of Electric Drive Production

Abstract

The market volume of electric drives for industrial applications and electric mobility is increasing steadily. Thus, efficient ways for monitoring and optimizing the production of electric drives are gaining importance. Besides winding, joining or impregnation processes, machining operations have a high share in the production chain. However, developing a process monitoring system for machining centers can be a cost-intense matter due to the need of addressing a manifold and dynamic error-space. Therefore, this paper examines potentials of cost-efficient data-driven approaches for process monitoring of machining operations on the example of electric drive production. In this context, a flexible approach for detecting current operational states by means of supervised machine learning is proposed. Since labor-intense modeling of process models based on a priori knowledge and first principles gets dispensable, the basis for self-adapting monitoring solutions is laid. Diverse process parameters such as structure-borne sound or cutting forces are suitable to train process and behavioral models. By realizing a system for operational state detection without necessary access to control-internal data, a cost-efficient process monitoring of the often heterogeneous machinery in electric drive production is enabled.