Page 29 - IJAMD-1-2

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

(i) Defect detection using convolutional neural networks creating standardized datasets, developing benchmarking
(CNNs) methodologies, fostering interdisciplinary collaborations,
• Case study: Researchers have applied CNNs to and addressing challenges related to data variability, model
analyze in situ monitoring images from selective interpretability, and implementation complexities.
7,8
laser melting (SLM) processes. CNNs were Through strategic investments and collaborative efforts,
5,6
trained to detect defects such as porosity and the full potential of AI-driven quality assurance in AM
surface roughness with high accuracy, enabling can be realized, paving the way for a more innovative and
real-time quality control. efficient manufacturing landscape.
• Outcome: These models significantly improved
the detection rates of defects compared to In this review article, we delve into the role of
traditional image processing techniques. AI-driven approaches in improving quality assurance
in AM processes. We explore various AI techniques and
(ii) Predictive maintenance with ML methodologies for defect detection, process monitoring,
• Application: Predictive maintenance models predictive maintenance, and design optimization,
use historical sensor data to predict equipment highlighting their potential benefits and practical
failures before they occur. Techniques such as applications. In addition, we discuss the challenges and
2,3
8
random forests and SVMs have been employed to future directions of AI-driven quality assurance in AM,
analyze patterns in the data and predict potential aiming to provide insights and guidance for researchers,
breakdowns. engineers, and industry stakeholders navigating the
• Benefit: This proactive approach minimizes evolving landscape of AM.
downtime and maintenance costs while ensuring
continuous production quality. 1.1. Past studies in the field of AM
(iii) Process optimization with reinforcement learning The integration of AI into quality assurance in AM is a
(RL) burgeoning field, promising to revolutionize manufacturing
• Implementation: RL algorithms have been used by enhancing precision, reducing errors, and optimizing
to optimize AM process parameters such as laser processes. This literature review examines current
power, scan speed, and layer thickness. By research on AI-driven quality assurance in AM, focusing
7,8
continuously learning from the process outcomes, on methodologies, applications, challenges, and future
RL agents can adjust parameters in real time to directions. The literature review was conducted using a
maintain optimal manufacturing conditions. systematic approach to ensure comprehensive coverage
• Impact: This dynamic optimization leads to of relevant studies. The databases utilized for this
9,10
improved part quality, reduced material waste, review include; IEEE Xplore, ScienceDirect, SpringerLink,
and enhanced process efficiency. Google Scholar, and Web of Science. The search queries
(iv) Anomaly detection with unsupervised learning used included combinations of the following keywords:
• Technique: Unsupervised learning methods, such “Artificial Intelligence,” “Quality Assurance,” “Additive
as clustering and principal component analysis, Manufacturing,” “3D Printing,” “Machine Learning,” “Deep
have been utilized to detect anomalies in the Learning,” and “Process Optimization.” The search was
AM process. These techniques identify patterns limited to publications from January 2020 to April 2024
4,7
that deviate from the norm, signaling potential to capture the most recent and relevant advancements in
defects. the field.
• Example: Anomaly detection models have been Bonatti et al. widely applied ML algorithms,
11
applied to monitor the melt pool in laser-based including supervised, unsupervised, and RL, to predict
AM processes, identifying irregularities that defects and optimize printing parameters. CNNs are
could compromise part quality. particularly effective in image-based defect detection. Jin
12
AI, particularly ML techniques, offers powerful tools et al. employed deep learning models, especially CNNs
for enhancing quality assurance in AM. By leveraging and recurrent neural networks (RNNs), for real-time
AI for defect detection, process monitoring, predictive monitoring and anomaly detection. These models analyze
maintenance, and process optimization, significant large datasets to identify patterns and predict potential
13
improvements in product quality, process reliability, failures. Bikas et al. used AI-driven in-situ monitoring
and operational efficiency can be achieved. However, systems sensors and cameras to collect real-time data
to fully harness these benefits, continued research during the printing process. These data were then analyzed
and development are necessary. These efforts include to detect deviations from the desired parameters, enabling

Volume 1 Issue 2 (2024) 23 doi: 10.36922/ijamd.3455