Leveraging Data-Science to Characterize Additively Manufactured Electromagnetic Components
Open Access
- Author:
- Sessions, Deanna
- Graduate Program:
- Electrical Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- October 01, 2019
- Committee Members:
- Gregory Huff, Thesis Advisor/Co-Advisor
Timothy Joseph Kane, Committee Member
Kultegin Aydin, Program Head/Chair - Keywords:
- Additive manufacturing
frequency selective surface
conformal electromagnetics
machine learning - Abstract:
- This work exposes challenges encountered when additively manufacturing electromagnetic components and proposes an implementation of machine learning techniques to relate additive process-specific geometric defects to electromagnetic response and mitigate impact. Additive approaches for radio frequency applications such as direct-write antennas and transmission lines are used to produce conformal, deployable, and low-cost solutions. However, the variation in print geometry and ink composition present a challenge in reliable device fabrication. To address these issues, a framework is developed to categorize variations in print geometry using computer vision analysis and combined with a machine learning algorithm to link geometric defects to electromagnetic response. This categorization can be implemented in additive manufacturing processes as a means of feedback for users. To test this system, a frequency selective surface of additively manufactured silver thermoplastic polyurethane ink spirals in a triangular lattice is used for analysis. This frequency selective surface acts as a spatial band-pass filter. Each spiral is imaged immediately after printing. The geometric analysis is performed using images of each printed spiral to determine geometric deviance and variation throughout the lattice. These spirals are then categorized using a t-distributed stochastic neighbor embedding machine learning algorithm to link key defects to one another (i.e. variations in arm length, width, radius, etc.). The analysis of electromagnetic response from the defects is derived from simulations of individual spirals tessellated in an infinite lattice. While defects are often process-specific, this same framework can be applied to varying geometric print jobs given a training set of geometries and simulated electromagnetic response. A discussion of these techniques and how they can be generalized and applied to additional design and fabrication problems follows.