Cyber acoustic analysis of additively manufactured objects

Abstract

The potential for intellectual property theft has been shown in the additive manufacturing industry using acoustic side-channel attacks lately. This paper aims to discuss the rate of success for recreating the G-Code of an object from the acoustic features and further elaborates on regression model analysis that provides the G-Code. Acoustic and G-Code data was analyzed in a training phase and an attack phase. In the training phase, a supervised machine learning algorithm was trained using Python, which is an interpreted, object-oriented, high-level programming language. During the attack phase, the created algorithm was used to process new acoustic data and to reconstruct the G-Code. The accuracy of the classification models and the regression models were determined. The classification accuracy was determined with k-fold cross validation, and the regression model accuracy was determined by scoring the regression models within the algorithm. Although classification and regression algorithms developed showed promising results, lower model accuracy was observed when the X and Y motors moved together. In the future, the team hopes to further increase the model accuracy so that an unknown shape can be replicated successfully. While security measures for cyber-security have previously been investigated, very little research has considered acoustic side-channel attacks on their ability to reconstruct G-Code and steal intellectual property. The findings of this novel research project showed some promising preliminary results on a sample case study.