Page 34 - IJAMD-2-2

International Journal of AI for
Materials and Design ML-driven optimization in additive manufacturing

1. Introduction the potential of data-driven approaches in understanding
complex process-property relationships, their impact
Additive manufacturing (AM), commonly known as 3D was limited by small datasets and issues related to model
printing, has transformed modern manufacturing by generalization. Subsequent developments have since
22
enabling the production of complex geometries with high advanced beyond these early efforts, leveraging more
design flexibility and material efficiency. Unlike traditional sophisticated algorithms and larger datasets to achieve
subtractive manufacturing, which removes material to robust and scalable ML solutions in AM.
shape a final product, AM constructs objects layer by layer
from digital models, significantly reducing material waste Key ML applications in AM include process parameter
and enabling the fabrication of intricate structures that optimization, where ML models identify optimal printing
would be challenging or impossible to achieve through conditions to minimize defects and enhance part
conventional methods. Beyond material efficiency, AM performance; in situ process monitoring and real-time
1,2
eliminates many constraints associated with traditional feedback, which utilize sensor data to detect anomalies
manufacturing, such as the need for assembly and and adjust parameters dynamically; and quality assessment
joining, and accelerates the prototyping process, reducing and defect prediction, allowing for early identification
development time and costs. In addition, AM facilitates of suboptimal parts before post-processing. 23,24 These
3,4
the integration of multifunctional structures and multi- ML-driven strategies not only minimize reliance on
material designs, expanding its applicability across various empirical trial-and-error experimentation but also reduce
industries, including aerospace, automotive, healthcare, production time, material consumption, and overall costs
and consumer goods. 5-11,12 while improving repeatability and part quality. Moreover,
ML has facilitated advancements that were previously
Despite these advantages, AM still faces critical unattainable using conventional process control methods,
challenges that hinder its scalability and widespread further demonstrating its potential to revolutionize AM. 25-27
industrial adoption. The manufacturing process is highly
sensitive to variations in processing parameters, material Given the rapid evolution of ML applications in AM,
properties, and environmental conditions, leading to a comprehensive and up-to-date review is needed to
inconsistencies in part quality. 13,14 Defects such as porosity, consolidate existing knowledge and identify emerging
residual stresses, warping, and poor interlayer bonding trends. Although several related reviews have been
frequently occur, necessitating rigorous post-processing published, many are limited in scope–often focusing on
to ensure structural integrity. Moreover, optimizing AM specific materials, algorithms, or AM techniques–leaving
15
processes for different materials and geometries remains a gap in the broader understanding of ML integration
a complex and computationally intensive task, as each across diverse AM platforms. 24,28-34 In contrast, this review
combination requires precise tuning of parameters such as offers a holistic examination of ML-driven strategies
laser power, scanning speed, layer thickness, and cooling for optimizing AM processes, encompassing all major
rates to achieve the desired mechanical properties and material classes, including polymers, metals, ceramics,
surface finish. 16-19 These challenges underscore the need and carbon-based composites. By analyzing recent
for advanced methodologies to enhance the reliability, advancements across these material systems, the review
efficiency, and precision of AM processes. aims to provide a structured overview of how ML is being
applied to enhance printing efficiency, material utilization,
To address these issues, data-driven approaches and product quality. Furthermore, it addresses critical
leveraging machine learning (ML) have gained significant challenges such as data scarcity, model generalization, and
attention as a means to optimize AM workflows. ML the integration of real-time control mechanisms. Emerging
techniques enable predictive modeling, real-time process directions, including physics-informed ML models and
monitoring, and adaptive control, providing a systematic multimodal in situ monitoring, are also highlighted to offer
approach to improving manufacturing consistency and a forward-looking perspective. The insights presented in
efficiency. Early research in this field, initiated over a this work are intended to support researchers, engineers,
decade ago, demonstrated the feasibility of applying basic and industry professionals in leveraging ML to advance the
ML models, such as support vector machines (SVMs) and reliability, scalability, and intelligence of AM technologies.
Gaussian process (GP) regression, to process AM data.
For instance, GP-based surrogate models were employed 2. AM and ML method
to predict defect formation and porosity from process
parameters in metal AM, while SVM classifiers were used 2.1. AM
to construct process maps and identify stable fabrication Unlike conventional subtractive methods, which
regimes. 20,21 Although these pioneering studies validated remove material to create a final product, this advanced

Volume 2 Issue 2 (2025) 28 doi: 10.36922/IJAMD025130010