Machine Learning, Digital Twin, and Cybersecurity Digital Twin in Additive Manufacturing

L. Wang, R. Mosher, P. Duett
Mississippi State University,
United States

Keywords: machine learning, deep learning, digital twin, cybersecurity digital twin, additive manufacturing, convolutional neural network


Additive manufacturing (AM) has been used in aerospace, automotive, medical sectors, etc. Although AM has achieved great advances in aerospace, there are still challenges for this sector such as part's intrinsic anisotropy, cracks, porosity, surface roughness, and residual stress due to the high thermal gradient. In-space AM (production outside the earth's atmosphere) has many more challenges due to its challenging environments and space conditions such as microgravity. Microgravity can lead to a dripping-shaped deposition, affecting the heat flow, causing thermal problems, and impacting the mechanical properties of products. It is necessary to perform modeling and simulation for AM processes and improve the product quality. This presentation will introduce the trends and state of the art of machine learning (ML), digital twin (DT), and cybersecurity digital twin (CDT) in AM, and highlight challenges and future work. ML has been used to optimize AM process parameters and unravel the process–microstructure–property relationships. A vision-based in situ real-time defect monitoring method has been used to acquire image data and then recognize the defect class and locality in the AM process. Deep learning (DL) has been employed to extract defect image features in continuous frame images and train the network to learn the defect class label as well as its position. A data augmentation procedure should be implemented if the training data is generated from a small dataset. A fault diagnosis method for AM has been developed based on a convolutional neural network (CNN). CNN is used to process, detect, and classify anomalies in AM with good accuracy. DT is one of concepts of Industry 4.0. It refers to a digital informational construct that mirrors the behavior of a physical or cyber-physical system in a real-time simulation environment. DT can assist AM with predicting the errors/defects and mitigating them through real-time process monitoring and simulation, especially in the aerospace industry. DT offers opportunities regarding the simulation, forecasting, optimization, and monitoring of AM processes. There is a need for DT to secure the identity and protection of its genuine twin, which requires the utilization of cryptography algorithms. CDT is a virtual replica of the system that accompanies its physical counterpart during its lifecycle, consumes real-time data if required, and has the sufficient fidelity to allow the implementation, testing, and simulation of desired security measures and business continuity plans. CDT is the application of DT to allow for security analysis and monitoring which may not be directly feasible on the physical counterpart without causing disruption. It is designed and built to simulate a variety of cyberattack scenarios with the aim of testing the potential vulnerabilities in the corresponding real system. The attacks can be denial-of-service (DoS), data injection attacks, man-in-the-middle (MITM) attacks, etc. DT and in particular CDT can prevent cyberattacks by providing modelling/prediction capability. The use of ML for cyberattack detection and mitigation as well as secure blockchain for managing intellectual property rights help lower cybersecurity risks in AM systems.