Page 83 - ESAM-1-1

Engineering Science in
Additive Manufacturing ML in additive manufacturing

7.4. Data privacy and security making processes, enabling users to understand how

Data privacy and security involve ensuring the legal and and why specific predictions or classifications are made.
ethical use of data by employing techniques to prevent In the context of AI-driven AM, explainability is crucial
data corruption, unauthorized access, and both external for fostering trust, as it allows engineers and operators
and internal attacks. 55,149 Data privacy and security are to verify whether model predictions align with domain
31
vital for industrial deployability because they ensure knowledge and manufacturing principles. Furthermore,
the confidentiality of proprietary designs, process explainability supports debugging by identifying potential
parameters, and customer data, protecting businesses errors or biases in the model, leading to improved
from intellectual property theft and competitive risks. performance and reliability. It also facilitates compliance
Strong privacy and security measures foster trust among with industrial standards and regulatory requirements, as
stakeholders by demonstrating compliance with data transparent AI systems are often more readily accepted
protection regulations and ethical standards, crucial for in critical applications. By empowering users to interpret
global adoption. Moreover, safeguarding data integrity and model outputs and gain actionable insights, explainability
preventing unauthorized access is essential for maintaining enhances the usability of AI in AM, making it a vital
the reliability of AI models, which rely on high-quality, component for advancing industrial deployability.
secure data for accurate predictions and decision-making 7.7. Human-AI teaming
in AM processes. One of the recent advancements to ensure
data privacy and security in AI-driven AM is federated Human-AI teaming is a vital component of industrial
learning. It enables multiple parties to collaboratively deployability, as AI-driven systems inherently involve
55
train an AI model without sharing raw data by keeping the human factors and are designed to aid human labor. 52,108 To
data decentralized and transmitting only model updates or facilitate effective collaboration between humans and AI,
parameters to a central server. This approach ensures data user-friendly interfaces must be developed to ensure clear
privacy by maintaining data ownership with the individual communication. These interfaces should present decision-
parties while enhancing security by reducing the risk of making information in a transparent and interpretable
sensitive information exposure during training. manner for human operators to understand and trust
the system. In addition, they should enable input from
7.5. Compliance with standards humans, empowering them to provide guidance, override
Compliance with existing AM qualification and operation decisions, or adjust parameters as needed. Beyond
standards, as well as AI privacy and safety standards, is a interface design, the successful deployment of AI-driven
cornerstone of industrial deployability. Adhering to these AM systems requires equipping human operators with
standards ensures that AI models meet critical requirements the necessary understanding and skills to work alongside
for process reliability, product quality, and user safety. these technologies. Comprehensive training programs
AM qualification standards, such as those governing are essential to help laborers comprehend the capabilities,
material properties, process parameters, and inspection limitations, and optimal usage of AI tools, fostering
protocols, provide a framework for ensuring consistency confidence and competence in their interaction with
and repeatability in manufacturing processes (e.g., ISO/ the system. This synergy between human expertise and
ASTM52941-20 and F3704/F3704M-24). Similarly, AI AI intelligence can significantly enhance productivity,
privacy and safety standards address ethical considerations, innovation, and operational efficiency in industrial
data protection, and algorithmic transparency, which are settings.
vital for trust and acceptance in industrial environments
(e.g., ISO/IEC FDIS23894). However, the lack of specific 8. Conclusion
standards and benchmarks tailored to AI-driven AM poses AI and ML applications in AM enable unique outcomes
significant challenges. This gap hinders consistent evaluation, such as surrogate modeling in the absence of first principles-
validation, and comparison of AI models within the AM based methods, in situ monitoring for adaptive quality
domain, limiting their scalability and adoption. Developing control, digitization, and enhanced productivity, as well
comprehensive standards that integrate AI-specific concerns as achieving sustainability. Through a review of existing
into AM practices is essential to bridge this gap and unlock research and applications of ML in AM, this perspective
the full potential of AI-driven innovations in manufacturing. highlights the challenges data scarcity and model complexity
pose to integrate ML in the AM industry. The status of
7.6. Explainability AM data and applied AI models is briefly reviewed and
Explainability in AI refers to the ability of a model to linked to these challenges. Issues of data quality, quantity,
provide clear and interpretable insights into its decision- and management are discussed alongside representative

Volume 1 Issue 1 (2025) 16 doi: 10.36922/ESAM025040004