An LSTM-autoencoder based online side channel monitoring approach for cyber-physical attack detection in additive manufacturing

Additive manufacturing (AM) has gained increasing popularity in a large variety of mission-critical fields, such as aerospace, medical, and transportation. The layer-by-layer fabrication scheme of the AM significantly enhances fabrication flexibility, resulting in the expanded vulnerability space of cyber-physical AM systems. This potentially leads to altered AM parts with compromised mechanical properties and functionalities. Furthermore, those internal alterations in the AM builds are very challenging to detect using the traditional geometric dimensioning and tolerancing (GD&T) features. Therefore, how to effectively monitor and accurately detect cyber-physical attacks becomes a critical barrier for the broader adoption of AM technology. To address this issue, this paper proposes a machine learning-driven online side channel monitoring approach for AM process authentication. A data-driven feature extraction approach based on the LSTM-autoencoder is developed to detect the unintended process/product alterations caused by cyber-physical attacks. Both supervised and unsupervised monitoring schemes are implemented based on the extracted features. To validate the effectiveness of the proposed method, real-world case studies were conducted using a fused filament fabrication (FFF) platform equipped with two accelerometers. In the case study, two different types of cyber-physical attacks are implemented to mimic the potential real-world process alterations. Experimental results demonstrate that the proposed method outperforms conventional process monitoring methods, and it can effectively detect part geometry and layer thickness alterations in a real-time manner.