Skip to content

Machine Learning for AM: A Case Study of How it is Being Used by the Department of Defense

  • today
  • access_time 3:00 - 3:25 PM CT
  • blur_circularConference
  • monetization_onPaid Upgrade

This presentation will be a briefing on the technical results of a program that recently completed that was jointly funded by The Office of Naval Research, Air Force Research Lab, NAVAIR, and NAVSEA. The program was focused on developing and demonstrating a machine learning capability known as Transfer Learning. Transfer Learning, applied to additive manufacturing (AM), is a technique whereby data from known AM experiences are used to help improve the predictions of a new (i.e. unseen) experience.

Transfer Learning may significantly reduce the time and cost of process development or qualifying a new AM machine or material if used properly. With Transfer Learning, data from prior machines and prior materials can be used to help predict performance on a new machine and/or material.

In the program, SS316L data was generated on four different AM machines: 3D Systems ProX 320, EOS M270, EOS M290, and Additive Industries MetalFAB1. Using the Transfer Learning capability, data from some of the machines was used to help predict performance on the other machines. Transfer Learning was validated to work extremely well – i.e. the resulting predictive models were very useful and significantly improved.

This presentation will explain how the Transfer Learning capability was deployed, detail the technical results of this program, and walk the audience through the steps that a user can take to rapidly develop materials and/or processes for AM. In addition, this presentation will brief the audience on where the Department of Defense, and the Air Force in particular, anticipates utilizing machine learning in order to accelerate its successful deployment of AM.

Learning Objectives:

  • Understand how Transfer Learning can drastically cut down on time and cost for parameter optimization or qualification.
  • Understand a framework with which to analyze data across different AM machines and
  • Evaluate whether or not Transfer Learning can be used by the participant to efficiently develop process parameters for their own applications.