Skip to content

A Closed-Loop Machine Learning and Compensation Framework for Accuracy Control in 3D Printing

  • today
  • access_time 2:30 - 2:55 PM EDT
  • location_onRoom 140E
  • blur_circularConference
  • monetization_onPaid Upgrade
  • schoolOptimization

Additive manufacturing (AM) systems enable printing of 3D physical products from CAD models. Despite the many advantages of AM systems, one of their significant limitations is geometric inaccuracies, or shape deviations between the printed product and the nominal CAD model. Machine learning for shape deviations can enable accuracy control of 3D products via the generation of compensation plans, which are modifications of CAD models that reduce deviations. However, existing machine learning and compensation frameworks cannot accommodate deviations of fully 3D shapes with different geometries. The feasibility of existing frameworks for accuracy control is further limited by resource constraints that prevent printing of multiple copies of new shapes. We present a closed-loop machine learning and compensation framework that can improve accuracy control of 3D shapes in AM systems. Our framework is based on a Bayesian extreme learning machine (BELM) architecture that leverages data and deviation models from previous products to transfer deviation models — and more accurately capture deviation patterns — for new 3D products. The closed-loop nature of our framework, in which past compensated products that do not meet specifications are fed into the BELMs to relearn the deviation model, enables the identification of effective compensations and satisfies resource constraints. The power and cost-effectiveness of our framework are demonstrated with two validation experiments that involve different geometries for a Markforged Metal X AM machine printing 17-4 PH stainless steel products. Our framework can reduce shape inaccuracies by 30% to 60% (depending on the geometry) in at most two iterations. 

Learning Objectives:

  • Conduct good scanning of fully 3D printed products, and collect the point cloud data in a format for machine learning.
  • Understand machine learning algorithms can be usefully applied to data collected from additively manufactured products.
  • Derive geometric accuracy control methods for additive manufacturing systems based on machine learning models fitted to data collected from the machines.
  • Arman Sabbaghi PhD
    Associate Professor
    Purdue University, Department of Statistics
  • Wenbin Zhu
    PhD student
    Purdue University Department of Statistics