Page 40 - IJAMD-1-2

International Journal of AI for
Materials and Design
AI-driven quality assurance in AM

5.5. Regulatory compliance and certification • Inconsistent data quality: The quality and consistency

Ensuring compliance with regulatory standards and of the data collected can vary significantly. For
certification requirements is critical for the widespread instance, images captured under different lighting
adoption of AI-driven quality assurance in AM, conditions or thermal data affected by ambient
particularly in safety-critical industries such as aerospace temperature fluctuations can introduce noise and
33,34
and healthcare. Regulatory agencies and standards biases in the dataset. This variability can lead to
37
development organizations are still in the process poor model performance and unreliable predictions.
of establishing guidelines, validation protocols, and • Domain adaptation: Models trained on data from one
qualification procedures for AI-enabled AM technologies. specific machine or environment may not generalize
30
Bridging the gap between emerging AI capabilities and well to others. This domain shift can limit the
regulatory requirements necessitates close collaboration applicability of the AI models across different AM
between industry stakeholders, academia, and regulatory systems.
bodies to develop clear frameworks for validating and 6.2. Model interpretability
certifying AI-driven quality assurance systems. 20,33,36,37
Moreover, addressing ethical, legal, and safety ML models, especially deep learning techniques like CNNs,
considerations surrounding AI deployment in AM will be are often criticized for their lack of interpretability. This
essential for building public trust and confidence in these “black-box” nature makes it challenging to understand
technologies. how models make decisions, which can be a significant
drawback in critical applications like quality assurance.
In conclusion, addressing the challenges and advancing
the future directions of AI-driven quality assurance in AM • Lack of transparency: Operators and engineers
requires interdisciplinary collaboration, innovation, and may find it difficult to trust and validate the model’s
concerted efforts from researchers, engineers, industry predictions without a clear understanding of the
stakeholders, and regulatory agencies. By overcoming underlying decision-making process.
technical barriers, enhancing model interpretability, • Diagnostic use: Understanding the root causes of
integrating with manufacturing workflows, developing defects is crucial for continuous improvement in AM
physics-informed models, and ensuring regulatory processes. 19,29 Black-box models may identify defects
compliance, AI-driven quality assurance has the but fail to provide actionable insights on how to
potential to revolutionize AM and drive the next wave prevent them.
of industrial innovation. 36,37 Embracing these challenges
and opportunities will pave the way for realizing the full 6.3. Implementation complexities
potential of AM technologies in creating sustainable, high- Implementing AI-driven quality assurance in AM entails
performance products for diverse applications. several practical challenges, from data integration to
system scalability.
6. Addressing limitations in AI-driven • Integration with existing systems: Integrating AI
quality assurance for AM models into existing AM workflows and control
While the integration of ML techniques for quality systems can be complex and resource-intensive. 16,18,28
assurance in AM shows significant promise, it is crucial Ensuring compatibility with various hardware and
to address the associated limitations and challenges to software components is crucial for seamless operation.
provide a balanced and comprehensive analysis. 29,30 This • Scalability and real-time processing: Real-time defect
section delves into the potential drawbacks of AI-driven detection and quality assurance require significant
quality assurance, specifically focusing on data variability, computational resources, especially when processing
model interpretability, and implementation complexities. high-resolution images or large datasets.
• Maintenance and updates: AI models require regular
6.1. Data variability updates and maintenance to remain effective. 35,36 This
One of the primary challenges in employing AI for quality includes retraining models with new data, addressing
assurance in AM is the variability in data. AM processes, data drift, and adapting to new types of defects.
such as SLM and FDM, generate large amounts of data from While ML techniques offer powerful tools for enhancing
sensors, cameras, and other monitoring systems. 21,34,35 This quality assurance in AM, addressing the limitations and
data can be highly variable due to differences in machine challenges is essential for successful implementation.
calibration, material properties, environmental conditions, By tackling issues related to data variability, model
and operator skills. interpretability, and implementation complexities, the full

Volume 1 Issue 2 (2024) 34 doi: 10.36922/ijamd.3455