Learning Feedforward Control for Laser Powder Bed Fusion

Abstract

Additive manufacturing is becoming increasingly important for industrial applications. Especially laser powder bed fusion processes play a decisive role. However, the technology often lacks in reliability and part quality due to complex process interactions and disturbing influences. To counteract these limitations, a flexible and high-dynamic control of relevant process parameters, taking into account the transient process conditions, is indispensable. Currently, the most process parameters are often kept constant, since the used control systems do not offer real-time capable interfaces for intervention. To enable the adjustment of parameters during the process, an open control architecture for galvanometer scanners and laser sources was developed. The architecture is based on an industrial control system in combination with a field-programmable gate array (FPGA) communicating via a real-time fieldbus. The novel control architecture is used to integrate a model-based feedforward control, adjusting the laser power, scan speed and scan strategy based on the current powder bed temperature calculated by a two-dimensional thermal model during run time. An integrated learning algorithm autonomously tunes the process parameters during the manufacturing process and determines the optimal combination of laser power, scan speed and powder bed temperature to achieve a homogeneous melt pool. On average, the standard deviation of the powder bed temperature along the scan path was reduced by 48% for the simulated test cases.