A soft sensor for property control in multi-stage hot forming based on a level set formulation of grain size evolution and machine learning

Abstract

Close-die forging usually has the goal of shaping a workpiece at reduced forming forces and to set the properties for the application at hand utilizing the microstructural changes occurring at high homologous temperatures. The evolution of the property-determining microstructure can be treated as a dynamic system during hot forming. It is controlled by the temporal evolution of the temperature and velocity fields at the material points in the workpiece. Variations in the initial microstructure, friction and heat transfer conditions lead to variations in properties. When forging on screw presses, the workpiece is brought into its final shape with several blows. Here it generally seems possible to control the dynamic system of microstructure evolution over the course of the blows. Adaption of impact energy and pause times between impacts could be used to compensate the disturbances caused by, e.g., a delayed transport of the workpiece. So far, no work is known which uses the impact energy and pause time as a control variable to directly control the microstructure and property formation during forging. Instead of trying to control the microstructure evolution at each point in the workpiece, the present paper proposes to define regions of a desired microstructure state, e.g., grain size, and to track the evolution of the domain boundary. The boundary of the domain in which the target grain size is attained is extracted as a level set. This work shows the existence and proposes a fast surrogate model to predict its location as a function of the control input. The model can thus be considered a soft sensor that estimates the microstructural state based on information that can be retrieved in the process.

Keywords

• Close-die forging
• Microstructure control
• Process control
• Machine learning