Markov Decision Process for Image-Guided Additive Manufacturing

Abstract

Additive manufacturing (AM) is a process to produce three-dimensional parts with complex and free-form geometries layer by layer from computer-aided-design models. However, real-time quality control is the main challenge that hampers the wide adoption of AM. Advancements in sensing systems facilitate AM monitoring and control. Realizing full potentials of sensing data for AM quality control depends to a great extent on effective analytical methods and tools that will handle complicated imaging data, and extract pertinent information about defect conditions and process dynamics. This letter considers the optimal control problem for AM parts whose layerwise defect states can be monitored using advanced sensing systems. Specifically, we formulate the in situ AM control problem as a Markov decision process and utilize the layerwise imaging data to find an optimal control policy. We take into account the stochastic uncertainty in the variations of layerwise defects and aim at mitigating the defects before they reach the nonrecoverable stage. Finally, the model is used to derive an optimal control policy by utilizing the defect-state signals estimated from layerwise images in a metal AM application.