When Advanced Manufacturing Meets Machine Learning: A Study for Improved Printed RF Reliability

Additive manufacturing (AM) circuits and devices have primarily been focused on a low-count, high-customizability product space. Even though new AM technologies can successfully integrate radio frequency (RF) electronics on flexible and conformal substrates using AM technology, there are significant challenges when adapting this variable fabrication method with high-throughput manufacturing. However, with the advancement of image classification neural networks, there now exists a method to predictively analyze additively manufactured devices in-situ. This work integrates a convolutional neural network machine learning (ML) algorithm to predict the electrical and RF performance of a printed antenna structure (inverted-F) in the X-band. The antenna samples were fabricated using a Nordson EFD Pro4 micro pen dispensing printer and all simulations were performed using Ansys High Frequency Structure Simulator (HFSS). We utilized a novel ML training method using both simulation and experimental data sets for ML algorithm generation. Our algorithm has been able to generate accurate predictions of the RF performance with respect to antenna resonance frequency, resonance depth, and operational bandwidth. Additionally, the algorithm is capable of predicting new printed antenna designs using only simulated data sets. With the integration of this ML algorithm with in-situ process monitoring, we expect to usher in a new era of higher volume, and reliable, AM circuit production.