Page 111 - MSAM-4-3

Materials Science in Additive Manufacturing Sustainable manufacturing composite material optimization

AM, highlighting how ML algorithms can simplify FDM For instance, FDM machines equipped with real-time
processes by offering data-driven insights and automation. temperature, vibration, and humidity sensors can feed
Interestingly, ML models trained on extensive process and process data to AI models for predicting potential failures
mechanical testing data have achieved high accuracy in and recommending corrective actions before defects
predicting material behavior and performance outcomes. occur. 25
ANNs have been widely employed to develop surrogate Predictive maintenance has been most advantageous
models capable of replacing computationally intensive in industrial manufacturing, where machine reliability
finite element analysis (FEA), thus accelerating design directly impacts the success of mass production operations.
iterations. Gaussian process regression (GPR) and deep Hybrid AI approaches, combining ML with physics-
neural networks (DNNs) have been employed to optimize based simulation, have proven to enhance the accuracy of
input parameters for desired output qualities, including forecasting thermal distortion and residual stress in FDM-
tensile strength, impact resistance, and dimensional printed components, addressing a significant gap in AM
accuracy. CNNs have demonstrated potential in
15
detecting surface defects in FDM-printed parts through process validation. Another vital research area is AI-driven
image-based inspection, enabling real-time feedback toolpath optimization. AI models analyze and generate
and automatic quality control. AI-powered computer optimal extrusion paths that minimize travel time, material
16
vision systems have been employed to develop closed- consumption, and stress formation in printed components.
loop control systems that dynamically adjust FDM The technique is particularly useful in impeller production,
parameters during printing, effectively minimizing defects where highly curved geometries demand careful control
such as warping, delamination, and stringing. Another of extrusion direction and layer adhesion to provide
significant use of AI-enhanced FDM optimization involves fluid dynamic performance and structural integrity.
integrating MCDM methods, such as Fuzzy AHP and Finally, AI-driven part orientation algorithms have been
TOPSIS. These methods systematically examine trade-offs developed to determine the optimal positioning of FDM-
among mechanical performance, energy consumption, printed parts on the build platform. These algorithms
and material usage, supporting the development of more reduce the need for support structures and post-processing
sustainable printing setups. 17-19 while simultaneously enhancing surface finish quality. In
biomedical applications, AI-assisted bioinspired design
One of the critical challenges related to FDM is its optimization enables the production of patient-specific
variable energy consumption, which is heavily influenced prosthetics and implants. ML algorithms process patient
by print parameters, material types, and machine scan data to generate personalized FDM-printable models
efficiency. To address this issue, researchers have applied with optimized mechanics. In aerospace and automotive
26
real-time power monitoring using power meters and AI applications, AI-facilitated design of lightweight structures
algorithms to forecast and minimize energy consumption has resulted in significant material savings by leveraging
without compromising the structural integrity of the lattice and honeycomb infill patterns optimized by deep
printed parts. 20-22 Empirical studies have demonstrated learning algorithms. 27
that AI-augmented energy modeling can reduce FDM
power consumption by dynamically adjusting nozzle Despite these advancements, challenges remain
temperature and print speed, thereby maximizing energy regarding the generalizability and interpretability of AI
efficiency without compromising part quality. In addition, models due to the inherent variability in FDM processes,
RL methodologies have also been explored for self- which arise from different material properties, machine
adaptive parameter tuning, where AI agents learn optimal inconsistencies, and environmental factors. Existing work
print settings iteratively by trial and error, improving has focused on enhancing AI robustness by employing
decision-making capabilities with real-time feedback. transfer learning schemes, whereby pre-trained models
23
The application of generative adversarial networks (GANs) from similar AM processes are adapted to new materials
in topology optimization represents a novel approach, and machine setups with minimal retraining sample sizes.
enabling the generation of mechanically efficient yet Another promising direction is the integration of AI with
lightweight FDM-printed structures that support material digital twin technologies, which synchronizes real-time
conservation and sustainability. Furthermore, AI-driven data from FDM printers with digital simulations to enable
24
defect detection systems have advanced significantly, predictive process optimization and adaptive process
allowing early detection of print failures. Consequently, control. Further evolution of AI in FDM is expected to
this approach reduces material waste and production be driven by edge computing and cloud-based facilities,
downtime. The integration of AI with Internet of Things enabling real-time, multi-party optimization of AM
(IoT) sensors has also assisted in smart manufacturing. processes within distributed manufacturing networks.

Volume 4 Issue 3 (2025) 3 doi: 10.36922/MSAM025200033