Page 11 - IJAMD-2-2

International Journal of AI for
Materials and Design AI-driven material development for AM

cutters. The process follows two main approaches: Bond- development, as illustrated in Figure 4. Table 1 summarizes
then-form and form-then-bond. Compared to other the common AI methods employed across different stages
13
AM techniques, SHL offers cost-effective material usage of AM, including their typical applications, strengths, and
and rapid processing, making it particularly suited for limitations.
large and thick components. In addition, SHL does not
26
require support structures, enabling the fabrication of 3.1. Machine learning (ML)
complex geometries directly. SHL has also shown various ML, a subset of AI, is primarily categorized into supervised
26
applications in aerospace, automotive, medical, and learning, unsupervised learning, and reinforcement
27
28
29
bioengineering fields. 30 learning, enabling computers to learn from data without
explicit programming. Common ML methods include
32
3. Overview of AI techniques for AM k-nearest neighbor and support vector machine.
material development
ML has been increasingly applied to conventional
The advancement of AI has significantly influenced materials development. For example, Dang et al. used
33
traditional manufacturing, enhancing the productivity, the support vector regression model to predict the fatigue
efficiency, and flexibility of AM. This section explores life of titanium alloys by analyzing their microstructural
31
three major categories of AI techniques for material characteristics, including the stress intensity near the
development, analyzing their applications in materials holes and the type of holes. Besides, Ling et al. used the
34

Figure 4. Categories of AI techniques in additive manufacturing
Abbreviation: AI: Artificial intelligence.

Table 1. Summary of common AI methods, applications, strengths, and limitations across key AM steps
AM steps Common AI methods Typical applications Strengths Limitations
AM design RF, SVM, NNs, Lasso/ Feature recommendation, part RF is robust to noisy data; NNs require large datasets and
elastic net regression, mass, and cost prediction, build SVM is effective with small long training; SVMs are sensitive
GA, feed-forward NN, time estimation, topology design, datasets; GA enables global to kernel settings; GA can be
Bayesian inference, thermal compensation, shape search; NNs capture complex time-consuming; RF may bias
hierarchical clustering deviation prediction non-linear relations toward dominant features
AM process and Back propagation NN, Melt pool depth/width prediction, NNs capture complex GPR scales poorly; NNs are
performance SOM, LS-SVM, GPR, powder spreading, strength/ non-linearities; GPR provides data-hungry; GA performance
optimization Kriging, GA hardness estimation, porosity and uncertainty quantification; depends on parameter tuning
density prediction GA is good for multi-objective
optimization
In situ process NN, SVM, Naive Bayes Porosity and defect detection, Probabilistic models handle Deep models are computationally
monitoring and clustering, CNNs, deep anomaly detection, spatter uncertainty; CNNs/deep belief demanding; clustering accuracy
control belief networks (deep classification, acoustic emission networks are excellent for may be low; Naive Bayes assumes
learning), K-means analysis, multi-sensor fusion, sensor/image data; clustering feature independence
CT-aided defect identification helps early-stage pattern
recognition
Testing and Sparse representation, Point cloud-based dimensional KNN is simple and intuitive; KNN suffers in high dimensions;
validation KNN, Naïve Bayes analysis, defect classification SVMs are robust; sparse decision trees overfit easily; sparse
clustering, SVM, decision (e.g., porosity) models are suited to models require careful tuning
trees high-dimensional data
Abbreviations: CNN: Convolutional neural network; CT: Computed tomography; GA: Genetic algorithm; GPR: Gaussian process regression; KNN:
K-nearest neighbor; NNs: Neural networks; LS-SVM: Least squares support vector machine; RF: Random forest; SOM: Self-organizing map; SVM:
Support vector machine; AI: Artificial intelligence; AM: Additive manufacturing.

Volume 2 Issue 2 (2025) 5 doi: 10.36922/IJAMD025100007