**1. Introduction**

Additive manufacturing (AM), colloquially referred to as 3D printing, is defined by ASTM standard ISO/ASTM 52900 as "process of joining materials to make parts from the 3D model data, usually layer upon layer, as opposed to subtractive manufacturing and formative manufacturing methodologies" [1]. In recent years, the manufacturing industry has seen a paradigm shift due to the full range of advantages of AM, such as complexity-free design, no need for tooling, reduction in waste, and amount of production steps compared to conventional techniques. Fabrication of complex geometries without increasing the production cost is one of the capabilities of the AM. This capability of AM allows the designing of complex lightweight components without compromising the functionality. For example, in the aerospace and automotive industries, the reduction in material weight is directly linked to the reduction in fuel consumption. Therefore, the AM has matured very rapidly in terms of technical aspects and process aspects. These developments have led to convincing many industries to adopt AM for manufacturing novel parts across many sectors.

Despite many technological advantages, many problems related in particular to the control of defects (porosity rate, cracking, etc.), to the managemen<sup>t</sup> and control of heat transfer during the process, or material supply problems persist. Several operating factors, such as processing parameters, part design, support placement, powder properties, etc., can influence the quality of the finished part [2]. For example, Galy et al. [3] reviewed all the factors in the L-PBF process that influence the final quality of Al alloy parts. All these factors hinder the full adaption of the AM by many niche industries [4–6]. Many of the factors that can influence the quality of the final part (Figure 1) can be controlled in-line, and the process signatures can be directly linked to the quality [7–9].

**Figure 1.** List of process parameters, and corresponding signatures which influence the product quality [2].

This led to the development of in situ monitoring of the process and closed-loop control. There is an enormous number of in situ sensing techniques developed for all types of AM processes. However, we will focus mainly on the laser powder bed fusion process (L-PBF), also known as selective laser melting (SLM). Usually, melt pool monitoring is the most used technique to control the L-PBF process due to the dependency of melt pool dynamics on the quality metric of the part [10].

Generally, we monitor the process at three different levels, which are:


The in situ sensors for L-PBF can be divided into acoustic, optical systems, which are similar to the laser welding literature [11]. Acoustic sensors are complicated to calibrate due to the significantly high noise factor. Therefore, optical systems, such as cameras and photodiodes, are common for the L-PBF process. Clijsters et al. [12] and Craeghs et al. [13–15] have developed a co-axial melt pool monitoring system in-house for the L-PBF process in the visible-near infrared detection range. Demir et al. [16] proposed a co-axial system with a set of focused optics to control the area of focus. It is critical to control the area of focus to enhance the resolution around the melt pool. Nowadays, many commercial L-PBF machines are equipped with in situ monitoring systems. Nevertheless, the commercial systems are very closed and act as a black box. Therefore, it is essential to post-process the acquired sensors data to predict the quality of the part.

There are good critical reviews available on in situ monitoring and factors affecting the powder bed fusion (PBF) process [7–9]. Therefore, there is a lack of comprehensive review, especially focused on the L-PBF process and covers the process, in situ monitoring systems, and post-processing of in situ sensing data. This review article bridges the gap and provides a brief overview of the in situ

monitoring systems and treatment of data in the L-PBF process. Sections 3 and 4 are inspired by the work of Grasso et al. [9], which discussed commonly occurred defects in the powder bed fusion process. However, we have only reviewed the defects related to the L-PBF process.

This article is organized as follows. Firstly, we discuss the working principle of the L-PBF process and the most common defects that occur during the process, followed by the working principle of the in situ sensing devices. Finally, we review the literature available on the data treatment and automatic detection of defects in the L-PBF process using machine learning.

#### **2. Working Principle of L-PBF**

Laser powder bed fusion (L-PBF), also known as selective laser melting (SLM) and Direct melt laser sintering (DMLS), is an additive manufacturing process in which a laser source (thermal energy) selectively fuses regions of powder bed in a layer by layer fashion [17]. The schematic of L-PBF is shown in Figure 2a. The complex thermomechanical process in the form of a melt pool occurs locally due to laser–material interaction [18]. The schematic of the laser–material interaction region is shown in Figure 2b [19]. The non-uniformity of temperature distribution in the melt pool (center of melt pool is at a higher temperature) develops a surface tension gradient, which leads to the thermocapillary motion of the liquid, i.e., the liquid from the center of the melt pool is transported to the edges (colder region). When the laser source moves away from the molten pool, a negative surface tension gradient develops in the melt pool, which generates a shallow and well-distributed liquid mass. Similarly, the segregated melt at the melt pool edges will acquire sufficient surface energy, which then flows back to the hotter region, and the convection loop completes [20]. Rapid solidification with cooling rates of 10<sup>6</sup> K/s of melt pool allows for interlayer bonding [21].

**Figure 2.** (**a**) Schematic diagram of L-PBF process [17], (**b**) Laser–melt interaction region [19].

For metallic alloys, the support structures are required to anchor parts and their features to the build plate. It is necessary because the thermal gradients in the building part are high, which can lead to thermal stresses and warping if the anchors are not used. The thick build plate serves as the heat sink and also prevents the parts from warping while printing. The preheating of the build plate, which depends on the material manufactured, is done to minimize thermal stresses in the part. The dimensional accuracy and ability to print good surface finish products and fine feature details is one of the advantages of the L-PBF process.

## **3. Defects**

As we discussed, the L-PBF process is a complex thermomechanical process that is a ffected by many controllable process parameters such as laser power, scan speed, hatch distance, powder material, and powder morphology. All these parameters can strongly influence the laser–material interaction, which will influence the thermophysical mechanism, resulting in various physical phenomena such as material evaporation, microstructural evolution, melt pool instabilities, and thermal stresses. Thus, the formation of various defects during the process is inevitable. Here, we review the formation mechanism of the most prevalent defects, such as porosities, lack of fusion, balling, and cracking defects that can be detrimental to the mechanical properties of the final part.

## *3.1. Porosities*

Spherical and non-spherical porosities, which are usually smaller in size (100 μm), are the most common defects in the L-PBF process, which significantly influence the fatigue properties of the part [22–27]. The leading cause for the formation of gas porosities (generally spherical, as shown in Figure 3a) is the entrapping of gas in the melt pool due to rapid solidification. The rapid solidification of the melt pool does not allow dissolved gas to come out on the surface. As we know, at high temperatures, the gas dissolves in the melt pool very easily. Moreover, enrichment of gas in the liquid melt can be due to various factors such as low packing density of powder bed, gas inclusion during gas atomization of powders, and evaporation due to high laser power. Dissociation of oxide films, adsorbed gas, and moisture on the powder material also results in gas formation [28]. For example, Gong et al. [29] showed the presence of gas porosities due to the entrapping of gas bubbles, which originated due to the vaporization of the low melting material in the alloys. The gas bubbles could not escape to the surface due to rapid solidification. Zang et al. [30] reported the random distribution of the porosities within the printed layer, between the adjacent layers, and on the surface of the part.

**Figure 3.** Micrographs of defects observed in L-PBF, such as (**a**) gas porosities [22], (**b**) key hole porosity [29], (**c**) lack of fusion defect [22], (**d**) balling, and (**e**) thermal crack. Reused under Creative Commons Attribution License.

Another type of porosity called "keyhole porosity" is due to excessive energy input, which leads to temperatures beyond the boiling point of the material, which causes evaporation of the material and forms a plasma. The laser beam penetrates the powder bed deep through the vapor voids, which creates a large melt pool. High energy density can occur at lower scan speeds, where the longer exposure time initiates boiling, which unstabilizes the melt pool and induces small metallics balls. The intense recoil pressure creates denudation along the scan track and, together with Marangoni convection (thermocapillary convection), forms a melt pool depression into the powder layers. The breakdown of the melt pool sidewalls during solidification leads to entrapped irregular porosities in the printed layer (as shown in Figure 3b) [31].

Still, there is a lack of universal consensus among researchers on the evolution mechanism of keyhole porosities [29]. For example, according to reference [25], the keyhole porosities occur when the Marangoni e ffect outweighs the buoyancy. Similarly, Choquet et al. [32] stated that vortices, high fluid speed, and recirculation in the melt pool leads to keyhole phenomena. It is challenging to observe this type of defect experimentally and requires a sophisticated set up to observe it on-line. For instance, Cunningham et al. [33] used ultrahigh-speed X-ray imaging to capture keyhole evolution in titanium alloy while printing. They have explained the evolution of the keyhole porosity, which is as follows: As the laser turned on, the powder began to melt, and a solid–liquid interface was formed. Once the melt pool temperature reached near the boiling point of the material, localized vaporization led to recoil pressure and formed a depression. Due to recoil pressure, the melt pushed up and out of depression, which developed instability in the melt pool. Soon after, the shallow vapor depression transitioned to deep, conical depression. Then, deep vapor depression rapidly penetrated the melt pool, and displacement of the melted liquid from the center of the melt pool occurred. The displacement in the liquid introduced the liquid–vapor interface fluctuations and changed the melt pool shape from quasi-semicircular to a bimodal shape to form keyhole porosity.

#### *3.2. Lack of Fusion*

Another common defect observed in L-PBF is the lack of fusion defect (LOF), which occurs due to incomplete fusion of two adjacent layers (as shown in Figure 3c). It can happen due to (i) insu fficient energy input leading to incomplete melting of the powder layer [34], (ii) insu fficient material due to shrinkage during solidification, and (iii) poor bonding between layers due to oxidation, which influences the wetting angle between layers [29,35]. As the L-PBF process selectively melts the powder line by line and layer by layer, the LOF defects are usually present in between the layers or between the scan vectors. Sometimes, the uneven powder layer spreading in consecutive layers also leads to incomplete melting of the powder, which also results in an LOF defect in the final part. Clijsters et al. and Read et al. [12,36] showed that the rapid formation of an oxide layer on the solidified layer of AlSi10Mg leads to poor wettability in between layers, which results in an LOF defect.
