Page 40 - IJAMD-1-3

International Journal of AI for
Materials and Design
Metal AM porosity prediction using ML

1. Introduction trees (DTs) in predicting the mechanical characteristics and
behavior of 3D-printed materials. 16,17 Table 1 presents an
Additive manufacturing (AM) is the process of constructing overview of the literature on ML approaches for predicting
products by joining materials layers upon layers facilitated defects in the L-PBF process.
by a three-dimensional (3D) computer-aided design (CAD),
which differs from traditional subtractive manufacturing ML algorithms that predict the quality of the built
processes where products are fabricated by removing excess product can significantly enhance the sustainability of the
material from a larger block. Depending on the industry, manufacturing process in real time as they reduce material
1
AM equipment has the flexibility of using a wide array of and energy wastage to construct good quality products.
materials such as glass, metals (steel, titanium, and gold), and When an ML system predicts an anomalous layer (like high
2
thermoplastics (polycarbonate and polylactic acid). AM can porosity) is laid, it can alert the operator; the layer could be
decrease energy and material wastage, deliver greater design re-melted to remove the pores to improve or maintain the
flexibility, and prolong product life. AM encompasses various part’s quality and characteristics.
processes for building 3D objects layer by layer using different Previous studies have applied various ML approaches to
materials. These processes include laser powder-bed fusion detect irregularities in additively fabricated parts produced
(L-PBF), direct energy powder metal deposition, and wire arc by L-PBF processes. However, limited research has focused
AM. L-PBF is a metal AM technique that uses a laser beam on determining the porosity of these parts. Typically, porosity
3,4
to selectively melt and fuse metal powder to create dense 3D is assessed using post-processing techniques such as X-ray,
metal parts. It offers advantages such as design freedom and computed tomography (CT) scans, and synchrotron-based
3,5
is widely used to manufacture complex metal parts. 6 micro-CT. Notably, there is a gap in the literature regarding
L-PBF for AM faces several challenges. The quality and the use of in situ approaches for porosity detection.
reliability of parts produced are crucial aspects that need This paper presents the development of a new data
monitoring and improvement. The non-uniform melt analysis algorithm for the detection of defects which may
7
pool size is a common issue, affecting the printed parts’ occur naturally during the L-PBF production process.
strength and dimensional accuracy. 8-10 In this process, a laser beam scans the part CAD design

If the major challenge in AM of improved reproducibility on the powder bed, which melts and fuses the powder to
of product quality could be solved, the process could be the underlying solid material. Next, the “build chamber
automated. A recent study evaluated the process capability actuator” (Figure 1A) automatically lowers by a specified
11
and baseline variations between multiple L-PBF machine layer thickness, and a new layer of powder is laid. The process
configurations, finding a systematic error in build accuracy continues until all the designed layers are fused, which
across all features. Another study assessed the feasibility of completes the part manufacturing process. Furthermore, to
12
using L-PBF for remanufacturing and identified challenges prevent any oxidation during fusion, the build chamber is
such as misalignment in restoration, keyhole defects, and gas supplied with a continuous flow of inert gas (like argon).
30
pores in the boundary area between original and restored Although the L-PBF process shows a lot of potential, it is
parts. A round-robin test investigated the reliability of susceptible to defects occurring within the built part related
13
additive-manufactured specimens and found that optical to the set process parameters (such as layer thickness, laser
microstructure inspection was beneficial in determining power, and scan speed). Common problems arising from
true porosity and the effects of powder scatter. An inter- the L-PBF process are the occurrence of pores, voids, and
13
laboratory test on reproducing porous samples for sound lack of fusion owing to under- or over-melting. Due to the
31
absorption using different AM technologies highlighted process’s sensitivity to the build parameters, much research
discrepancies due to shape and surface imperfections has been done on in situ monitoring of the L-PBF process. 32
induced by the manufacturing process. The need for AM processes such as L-PBF generate a wealth of data,
14
33
reproducibility and benchmarking in AM is also emphasized often in the form of raw time series, and are challenging to
in surgical robotics, where a lack of systematic approaches process owing to their size. For example, in situ pyrometers
and accurate experimental descriptions hinders progress. 15 are utilized to gauge the reflected optical emissions emanating
Machine learning (ML) approaches have been increasingly from the melt-pool region which can be translated into melt
used to predict additively manufactured products’ quality and pool temperature data (as depicted in Figure 1B, details are
mechanical behavior. These approaches offer advantages over discussed in section 2). Connecting the acquired data to the
traditional statistical methods in handling complex and non- product’s quality is a formidable challenge. ML can be used
34
linear patterns in manufacturing process data. Several studies to analyze these raw time-series data to predict quantifiable
have demonstrated the effectiveness of ML algorithms such outcomes, such as porosity, surface roughness, and fatigue
as neural networks, support vector machines, and decision life. Previous studies have shown that working with
1,31
1,31
Volume 1 Issue 3 (2024) 34 doi: 10.36922/ijamd.4812