# Effect of Prepreg Composition on the Structure and Shear Strength of PEI/CF Laminates Fabricated by Ultrasonic Additive Manufacturing

## TL;DR

This study explores how prepreg composition affects the structure and strength of PEI/CF laminates made using ultrasonic additive manufacturing.

## Contribution

The novel contribution is the use of artificial neural networks and extra resin layers to optimize and enhance the shear strength of UAM laminates.

## Key findings

- UAM laminates with CF contents above 55 wt.% had 40% lower shear strength than thermoformed ones.
- Adding a 50 µm thick extra resin layer increased shear strength by up to 50% at 70 wt.% CF content.
- Neural networks with minimal architecture effectively optimized UAM parameters under ultra-small sample conditions.

## Abstract

In this study, laminates based on polyetherimide (PEI) with contents of carbon fibers (CFs) from 55 to 70 wt.% were fabricated by thermoforming (TF) and ultrasonic additive manufacturing (UAM) methods. The UAM laminates with CF contents above 55 wt.% possessed shear strengths lower by 40% in comparison with those of the TF ones, due to insufficient amounts of the binder in the prepregs to form reliable interlaminar joints. For enhancing the shear strength of the laminates with a CF content of 70 wt.%. up to the levels of the TF ones, extra resin layers with thicknesses of 50, 100, and 150 μm were deposited. By ranking the UAM parameters using the Taguchi method, it was possible to increase the shear strengths by 30% as compared to those of the trial laminates. Further improvements were achieved by artificial neural network (ANN) modeling. As a result, the use of the 50 µm thick extra resin layer made it possible to increase the shear strengths up to 50% relative to those of the trial laminates at a CF content of 70 wt.%. This improvement was achieved via minimizing the number of defects at the interlaminar interfaces. The dependences of both mechanical and structural characteristics of the laminates on the UAM parameters were essentially nonlinear. For their analysis and optimization of the UAM parameters, the direct propagation neural networks with the minimal architecture were utilized. Under the ultra-small sample conditions, the use of a priori knowledge enabled us to predict the results rather accurately.