Page 71 - ESAM-1-1

Engineering Science in
Additive Manufacturing ML in additive manufacturing

Figure 2. Temporal (left) and geographic (right) distribution of review articles at the intersection of additive manufacturing and machine learning

Table 1. Outcomes of AI and ML usage in AM
Outcomes of AI and ML usage in AM Description Relevant case studies
and reviews
Surrogate modeling of process, structure, Application of AI and ML to develop surrogate models for predicting process 46-48
and property states, structural characteristics, and resulting properties
In situ monitoring and quality control Application of AI and ML for in situ defect detection and process repeatability 18,49
Digitization and Industry 4.0 Application of AI and ML to enhance productivity 30,50-52
Sustainability Application of AI and ML to enable sustainability through AM technologies 53
Abbreviations: AI: Artificial intelligence; AM: Additive manufacturing; ML: Machine learning.

12
Building on the current state of research, the following into design, process, and product characteristics. Design
sections explore the key ML themes of data quality, characteristics include the material and geometrical
data quantity, and model complexity. They examine the features of the part, which can be used to predict thermal
57
56
challenges and opportunities associated with integrating profiles, melt pool morphology, locations of potential
59
58
ML technologies into the AM industry. defects, and geometric conformity of the final part.
Process characteristics, encompassing both process
3. Data in AI-driven AM parameters and in situ sensor data, capture the real-time
As the carrier of information, data quantity and quality settings and physical phenomena occurring during the
play decisive roles in the prediction performance of printing process. For example, process parameters such
any AI algorithm, including AI-driven AM. Data of as laser power, scan speed, and printing toolpath in LPBF
numerous types and formats (e.g., tabular, graphics, dictate the dynamics of the melt pool, therefore influencing
60
three-dimensional [3D], time series, and text) have been the final part quality. To comprehensively monitor the
collected and integrated throughout the lifecycle of various intricate physical phenomena involved in the printing
AM processes. 12,54 However, AM data acquisition usually process, it is essential to deploy multiple sensors across
18
involves expensive and time-consuming experiments and various modalities. Commonly utilized sensors include
simulations, resulting in datasets that are small in scale charge-coupled device (CCD) cameras and complementary
and limited in quantity. As a result, data-driven modeling metal-oxide semiconductor cameras for recording melt
55
pool morphology; pyrometers, photodiodes, and thermal
in the AM field is hindered by data scarcity. Challenges of cameras for capturing thermal profiles; layer imaging
high dimensionality and low data quality are also major
barriers to the industrial deployment of AI-driven AM. cameras for observing layer-wise powder-spreading
conditions and part geometric conformity; acoustic
AM part quality is influenced by a multitude of sensors, such as microphones and ultrasonic transducers,
factors such as geometry, material, process parameters, for detecting noise and acoustic emission signals from the
and environmental conditions, which construct a high- melt pool and part; and vibration sensors for monitoring
dimensional problem space. Researchers train ML models machine vibrations. Product characteristics concern the
12
to detect defects and predict quality during the AM part structure and properties that reflect part quality and
process using various features, which can be categorized performance. Product structural characteristics can be
12

Volume 1 Issue 1 (2025) 4 doi: 10.36922/ESAM025040004