Page 36 - IJAMD-1-2

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

Table 3. (Continued)
Implementation Functions Explanation
methods
Data acquisition and management Step Establish robust data acquisition systems to capture high-quality data
from the AM process.
Details Ensure that data from various sources (sensors, cameras, logs) is
collected, stored, and managed efficiently. Implement data management
protocols to handle large datasets and maintain data integrity.
System integration Step Integrate AI models with existing AM systems.
Details Collaborate with AM equipment manufacturers to embed AI capabilities
into machine controllers. Develop interfaces that allow seamless
communication between AI models and machine hardware.
Pilot testing and iteration Step Conduct pilot tests to evaluate the performance of AI-driven quality
assurance systems in real-world conditions.
Details Implement AI systems in a controlled production environment, monitor
their performance, and iteratively refine the models and processes based
on feedback.
Scaling and deployment Step Scale successful AI implementations across the production line.
Details Once validated, deploy AI systems across multiple AM machines and
production lines. Ensure that support and maintenance structures are in
place to sustain AI system performance.
Continuous improvement and training Step Establish a framework for continuous improvement and operator
training.
Details Regularly update AI models with new data and insights. Train operators
and quality assurance personnel on the use of AI systems, ensuring they
can interpret AI outputs and make informed decisions.
Abbreviations: AI: Artificial intelligence; AM: Additive manufacturing; ML: Machine learning; CNN: Convolutional neural network; SVM: Support
vector machine; RL: Reinforcement learning.

However, ensuring the quality of AM parts remains a studies illustrate the practical implementation of AI-driven
significant challenge due to the complexity of the process approaches, demonstrating significant improvements in the
and inherent variability in material properties. This review accuracy of defect detection, reliability of material property
investigates the use of AI, particularly ML techniques, to classification, and efficiency of process optimization.
enhance quality assurance in AM processes. We focus on The study addresses challenges such as data
specific methodologies such as CNNs for defect detection, preprocessing, model interpretability, and integration
SVMs for the classification of material properties, and RL with existing AM systems. It also discusses the limitations
for real-time process optimization. encountered during the implementation of these AI
Concrete examples are provided to illustrate the techniques, providing insights into overcoming these
application of these AI techniques. For instance, CNNs hurdles. The findings highlight the transformative
are employed to analyze real-time data from in-situ potential of AI in quality assurance for AM and outline
monitoring systems, successfully identifying defects such future research directions for further integration and
as porosity and surface roughness in parts produced enhancement of ML techniques in AM processes.
through SLM. Case studies highlight how SVMs classify 4.1. Introduction
the mechanical properties of printed components based on
input parameters and sensor data, achieving high accuracy SLM is a prominent AM technique used to produce high-
rates in predicting tensile strength and durability. precision metal parts with intricate geometries. Despite
its advantages, maintaining the quality of parts produced
In addition, a case study on RL showcases how using SLM is challenging due to defects like porosity,
it optimizes process parameters in real time during surface roughness, and residual stresses. This case study
fused deposition modeling (FDM), leading to reduced explores the application of CNNs to enhance quality
production time and improved part quality. These case assurance in SLM by detecting defects in real time.

Volume 1 Issue 2 (2024) 30 doi: 10.36922/ijamd.3455