Page 30 - IJAMD-1-2

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

immediate corrective actions. Stavropoulos et al. products and more efficient manufacturing processes.
14
assessed machine health and predicted maintenance needs Table 1 provides an overview of literature reviews on AM
using predictive models, thereby reducing downtime and AI.
and improving reliability. Plathottam et al. used high-
5
resolution imaging combined with AI techniques, such as 2. Challenges in quality assurance
CNNs, to detect surface defects and layer inconsistencies. AM processes offer unparalleled design freedom and
Several studies have shown that AI can identify defects flexibility, but they also introduce unique challenges
with higher accuracy and speed compared to traditional to quality assurance that must be addressed to ensure
methods. Stavropoulos et al. analyzed acoustic signals the reliability and performance of manufactured parts.
16
15
generated during the printing process using ML algorithms Understanding and mitigating these challenges is crucial for
to detect internal defects that are not visible on the surface. advancing the adoption of AM across diverse industries.
9
17
Stavropoulos et al. optimized AI algorithms printing In this section, we explore the multifaceted challenges in
parameters such as temperature, speed, and layer thickness quality assurance within AM processes in detail.
to enhance the quality of the final product. Genetic Various ML techniques are being explored to enhance
algorithms and RL are particularly effective in exploring quality assurance in AM. This section provides a comparative
18
the parameter space. Talaat and Hassan controlled AI analysis of different AI-driven approaches, evaluating their
adaptive systems printing parameters in real time, adjusting effectiveness in detecting defects, optimizing processes, and
to variations in material properties and environmental ensuring material properties. The discussion presented in
conditions. Rojek et al. stated that high-quality, annotated Table 2 includes CNNs, SVMs, and RL, highlighting their
19
datasets are essential for training AI models. However, respective strengths and limitations.
obtaining sufficient data can be challenging, especially for
rare defects. Moreover, Kantaros and Ganetsos proved 2.1. Geometric complexity
20
that training sophisticated AI models requires significant One of the defining features of AM is its ability to fabricate
computational power, which can be a barrier for small complex geometries with intricate internal structures that
21
and medium-sized enterprises. Kantaros et al. presented are difficult or impossible to achieve using traditional
technical and logistical challenges in integrating AI-driven manufacturing methods. While this capability enables
quality assurance systems with existing AM workflows innovative designs and lightweight structures, it also
and machinery. Kantaros et al. combined different AI presents challenges for quality assurance. Traditional
22
25
techniques, such as integrating CNNs with RNNs or inspection techniques, such as visual inspection and
using hybrid models that combine ML with physics-based coordinate measuring machines, may struggle to
simulations to enhance predictive accuracy. Hunde and comprehensively assess the dimensional accuracy and
Woldeyohannes implemented AI models on edge devices internal features of complex AM parts. Ensuring the fidelity
23
10
to enable real-time processing and reduce latency, which of intricate geometries and maintaining tight tolerances
is crucial for in-situ monitoring and control. Zhu et al. across all dimensions require advanced metrology tools
24
developed industry standards and benchmarking datasets and sophisticated inspection methodologies tailored to the
to facilitate the comparison of different AI approaches and unique characteristics of AM.
accelerate their adoption in the industry.
2.2. Material variability
The literature indicates that AI-driven quality assurance
has the potential to significantly improve the reliability AM processes utilize a wide range of materials, including
and efficiency of AM processes. Advances in ML and deep polymers, metals, ceramics, and composites, each with
learning have enabled the development of sophisticated its own unique properties and processing requirements.
models for defect detection, process optimization, and However, these materials often exhibit inherent variability
real-time monitoring. 14,15 However, challenges such as data in composition, particle size distribution, and mechanical
availability, computational requirements, and integration properties, posing challenges for quality assurance.
with existing systems must be addressed to fully realize the Inconsistent material properties can lead to variations
benefits of AI in quality assurance for AM. Future research in part performance, dimensional accuracy, and surface
should focus on developing hybrid AI models, leveraging finish, undermining the reliability and repeatability of
edge computing, and establishing industry standards to AM processes. 11,12 Addressing material variability requires
promote widespread adoption. 9,18,21,24 The continuous robust material characterization techniques, quality
evolution of AI technologies promises to drive further control measures, and process optimization strategies to
advancements in AM, ultimately leading to higher-quality ensure consistent part quality across production batches.

Volume 1 Issue 2 (2024) 24 doi: 10.36922/ijamd.3455