Predictive iterative learning control with data-driven model for near-optimal laser power in selective laser sintering

Building high-quality parts is still a key challenge for Selective Laser Sintering machines today due to a lack of sufficient process control. In order to improve process control, a Predictive Iterative Learning Control (PILC) controller is introduced that minimizes the deviation of the post-sintering temperature profile of a newly scanned part from a desired temperature. The controller achieves this by finding a near-optimal laser power profile and applying it to the plant in a feedforward manner. The PILC controller leverages machine learning models that capture the process’ temperature dynamics based on simulated data while still guaranteeing low computational cost. The controller’s performance is evaluated in regards to the control objective with heat transfer simulations by comparing the PILC-controlled laser power profiles to constant laser power profiles.