Combined Use of Acoustic Measurement Techniques with X-ray Imaging for Real-Time Observation of Laser-Based Manufacturing

Abstract

Ensuring high-quality control in laser additive manufacturing and laser welding relies on the implementation of reliable and cost-effective real-time observation techniques. Real-time monitoring techniques play an important role in understanding critical physical phenomena, namely, melt pool dynamics and defect formation, during the manufacturing of components. This review aims to explore the integration of acoustic measurement techniques with X-ray imaging for studying these physical phenomena in laser manufacturing. A key aspect emphasized in this work is the importance of time synchronization for real-time observation using multiple sensors. X-ray imaging has proven to be a powerful tool for observing the dynamics of the melt pools and the formation of defects in real time. However, X-ray imaging has limitations in terms of accessibility which can be overcome through combination with other more-accessible measurement methods, such as acoustic emission spectroscopy. Furthermore, this combination simplifies the interpretation of acoustic data, which can be complex in its own right. This combined approach, which has evolved in recent years, presents a promising strategy for understanding acoustic emission signals during laser processing. This work provides a comprehensive review of existing research efforts in this area.

1. Introduction

Laser-based manufacturing is a rapidly growing industry with a wide range of applications in manufacturing processes [1]. One example of this technology is laser metal welding, which offers non-contact, high-speed processing compared to traditional methods [2]. Laser metal welding is utilized in various fields, including electronics, automotive, aerospace, aircraft, and other industries [3]. Another significant aspect of laser-based manufacturing is metal additive manufacturing (AM), which has the potential to revolutionize global parts manufacturing and logistics [4]. For instance, directed energy deposition (DED) additive manufacturing shows promise in fabricating multi-material parts with controlled microstructures by locally solidifying the deposited powders of different metallic alloys [5]. DED has proven to be effective in repairing damaged automotive and aerospace components, surpassing traditional welding with enhanced precision and automation suitability [6]. Powder bed fusion by laser beam (PBF-LB), another type of AM [7], efficiently utilizes a high-energy laser to selectively melt metal powder layer by layer, solidifying sequentially to produce final lightweight parts. Major advantages of PBF-LB include cavity fabrication, three-dimensional mesh or lattice parts, and geometrical freedom in design [8,9].

In laser-based manufacturing techniques, such as PBF-LB, a keyhole (also known as a vapor depression [10]) is a deep cavity filled with vapor that forms within the molten metal pool when the laser beam interacts with the metal powder. The metal surrounding the keyhole melts due to heat conduction, forming a molten pool [11]. Unstable collapse of the keyhole traps gas bubbles, resulting in voids or pores in the solidified material [12]. These porosities caused by the keyhole can lead to defects in fusion, where layers are not properly bonded. Adjusting the laser energy to the keyhole regime can help mitigate this issue [13]. Additionally, laser manufacturing can also cause cracks and delamination, affecting stress concentration and fatigue resistance [14]. Therefore, a significant obstacle to the widespread industrial adoption of laser-based additive manufacturing is the possibility of keyholes and other various defects (such as balling [15], rippling [16], and spattering [17]). These defects occur during the manufacturing process and impact the quality of the final product [18,19]. For instance, during laser welding, pores, cracks, lack of fusion, and spattering defects inevitably occur, weakening joints and potentially affecting performance [20]. Spattering refers to the ejection of molten metal droplets from the melt pool during the laser–material interaction [21]. However, analyzing the keyhole and its behavior is challenging because the keyhole is concealed within non-transparent workpieces and inaccessible to conventional imaging and non-imaging diagnostics [22]. Therefore, real-time observation of dynamic phenomena during this process and detecting defects in real time can be valuable for industry and manufacturing.

To study the complete volume of material where pores and keyholes form, real-time measurement techniques are necessary. X-ray imaging is a significant technique for monitoring unknown laser–material interactions and the underlying physics. Synchrotron X-ray imaging is a valuable tool for gaining real-time insights into subsurface phenomena, including the dynamic behaviors of melt pools and keyholes. Therefore, real-time imaging of dynamic phenomena includes the motion of spatter and pore formation during laser–material interactions, allowing researchers to understand the origins of defects. For instance, researchers at the Paul Scherrer Institute [23] have miniaturized and specially designed PBF-LB devices for in situ X-ray measurements of PBF-LB. The designed PBF-LB devices mimic the commercial PBF-LB process (shown in Figure 1) and exemplify the apparatus developed. The apparatus enables the use of X-ray measurements to study the laser melting process of PBF-LB under controlled conditions [24].

In addition to the advantages of synchrotron X-ray imaging, there is a notable absence of sufficient details regarding the thermal history and chemical composition of the processed material [26]. Furthermore, the practical utility of X-ray imaging is constrained by issues related to portability and usage outside of laboratory settings [27]. Therefore, alternative measurement techniques are imperative to address these inherent limitations.

Acoustic emission (AE) sensors can address issues of portability and usage outside of laboratory settings [27,28] in X-ray utilities while also being applicable for real-time measurement of dynamic processes in laser advanced manufacturing. AE analysis, as a low-cost and robust monitoring technique, is widely used in laser manufacturing and extends across various manufacturing processes. In laser welding, AE signals are utilized to assess penetration depth [29,30]. In additive manufacturing, they also play a crucial role in defect detection [31,32] and distinguishing between different materials and process regimes [33]. For example, AE signals are employed in DED to differentiate between stable and unstable processes (such as geometric fluctuations, oxidation effects, or other process anomalies) by measuring the mean intensity of the acoustic emissions [34].

Real-time monitoring of industrial processes with AE involves collecting and analyzing AE signals to identify and assess defects throughout the structure [35]. AE signals are valuable tools due to their cost-effectiveness and ability to achieve high temporal resolution (attributed to their 1D signal nature) [36,37]. This high temporal resolution allows AE sensors to detect and record rapid changes in the material during the laser melting process. However, it is important to note that AE signals are highly sensitive to environmental factors, making them susceptible to interference and variations based on surrounding conditions [38].

Interpreting AE signals in laser metal manufacturing is complex due to various factors [39], including variability in process parameters like power settings and scanning speed [40] and material composition [41]. Discriminating between signals associated with regular processing and those indicating defects also requires sophisticated signal processing algorithms. These complexities of interpreting AE signals arise from the intricate nature of AE signals and the need for accurate analysis to ensure quality control [42]. Therefore, integrating AE signals with X-ray imaging, which serves as a ground truth, can enhance the overall analysis.

This paper comprehensively explores AE measuring techniques when coupled with X-ray imaging. X-ray imaging serves as a ground truth, facilitating the validation of other measurement techniques [43]. Precise synchronization of signals captured through various methods with X-ray imaging is critical for achieving this validation. This synchronization allows for subsequent validation and establishes a foundation for robust and accurate results. The present study conducts a comprehensive review of synchronization methods employed in AE signals in conjunction with X-ray imaging, aiming to ensure the validation of obtained results.

The rest of this review is organized as follows: Section 2 discusses the X-ray measurement techniques and their usage for studying the dynamics of the melt pool. Section 3 explores two types (contact and non-contact) of AE technology synchronized with X-ray imaging to investigate the dynamic process in laser-based manufacturing. Finally, Section 4 proposes some conclusions regarding the synchronization of AE testing technology with X-ray imaging, including the advantages, challenges, and development of this synchronization.

2. Synchrotron X-ray Applications in Monitoring Dynamic Phenomena for Laser-Based Manufacturing

This section provides an overview of the application of synchrotron X-ray imaging in monitoring the dynamics of the melt pool [44,45,46,47,48] and powder motion [49,50,51]. These dynamic phenomena are crucial for understanding laser material interactions in laser-based manufacturing.

2.1. Melt Pool Dynamics

The primary result of the interaction between the laser and materials is the formation of a melt pool. Understanding the characteristics of the melt pool is crucial for achieving high-quality products [52]. Cunningham et al. [53] utilized synchrotron X-ray imaging to reveal the sequence of keyhole formation in PBF-LB for metal additive manufacturing, based on laser power density. The study identified five key stages at an approximate frame rate of 200 kHz: (1) melting; (2) formation and growth of a vapor depression; (3) instability of the vapor depression; (4) formation and growth of a keyhole; and (5) change in the shape of the melt pool (Figure 2). The study also discovered two distinct transitions: The first occurred when there were fluctuations in the vapor depression, resulting in an accelerated increase in aspect ratio (depth to width) as seen in D to G in Figure 2 The second transition occurred when the aspect ratio [48,54,55] approached a value close to 0.5 [53]. This evolution from a hemispherical melt pool (conduction mode) to a well-defined sharp keyhole provides more direct insights than other methods that rely on cross-sections of solidified weld pools.

A challenge in X-ray imaging is to display clear boundaries of the melt pool and vapor depression during laser–material interactions. Parab et al. conducted a study on the impact of exposure time on hindering the identification of features with slight density differences, as shown in Figure 3a, where the melt pool lacks sufficient contrast. The laser scanned from left to right, with a sample thickness of 380 µm, a scanning speed of 500 mm/s, and a laser power of 260 W. Increasing the exposure time from 1 to 20 µs (Figure 3a–d) enhances the visibility of the melt pool due to a higher photon flux, which improves the signal-to-noise ratio. These images, with stable interfaces between solid–liquid and liquid–gas phases, are suitable for reliable quantitative analysis of melt pool dynamics. Additionally, further improvements for heavier materials like Ni-based superalloys are expected by utilizing higher-energy photon sources and making beamline adjustments [56].

Parab et al. [56] utilized synchrotron X-ray imaging to observe the dynamics of the PBF-LB process in high resolution, providing insights into vapor depression and melt pool phenomena in both temporal and spatial dimensions. Figure 4 displays high-speed X-ray images of laser AM for an Al-Si10-Mg plate and powder bed fusion for Ti-6Al-4V, capturing the intricacies of vapor depression. The experimental parameters for each case are detailed in the caption of the figure. The images reveal dynamic oscillations in the liquid–vapor interface during the processes. To study dynamic phenomena in the PBF-LB process, ultrafast laser-melting experiments were conducted using a high-speed camera at 1.08 million (Figure 4a,c) and 10 million (Figure 4b) frames per second (fps). The first image sequence (t = 0.8 to 1.1 µs) showed minor changes in vapor depression geometry, while the second sequence (t = 8.3 to 8.6 µs) revealed significant differences. Slower recording speeds are suitable for tracking major vapor depression changes. However, ultrafast recording is essential for capturing rapid oscillations in the liquid–vapor interface, as slower speeds may result in missed phenomena [56]. It is important to note that the 10 MHz camera recording speed represents the maximum continuous recording rate achievable with currently available high-speed CMOS cameras [57,58].

Understanding the flow of the melt pool is crucial for controlling both the microstructure and the process itself [59]. However, evaluating this flow is challenging due to the opacity of liquid metals, which requires in situ assessment. To address this issue, researchers introduced highly attenuating tungsten carbide particles into a blend of aluminium and copper powders with varying compositions. This allows for the visualization of Marangoni flow traces in the melt pool through high-temporal-resolution synchrotron X-ray radiography experiments. The high-magnification images in Figure 5a–e display the movement of a single tungsten tracer particle around the keyhole. Figure 5f presents a composite image created by extracting the pixel-wise minimum intensity from a stack of 250 images, showing a section of the visible melt pool. This method enables us to observe the direction and speed of flow during the process [60].

2.2. Powder Motion

In PBF-LB processes, the ejection of powder spatter is a common problem that can negatively affect the building process, leading to structural defects and reduced construction accuracy [61]. To study this phenomenon, high-speed X-ray imaging is employed [62,63]. When a high-power laser hits the powder bed, the particles rapidly melt and evaporate, resulting in the release of metal vapor that pushes adjacent particles out of the bed. The liquid metal is subsequently propelled outward due to Marangoni flow and recoil pressure, causing the liquid metal to wet and melt the particles [64].

Ultrafast X-ray imaging, recorded at a speed of 1.08 million fps, allows for precise tracking of high-velocity particle ejections in the PBF-LB process. This imaging technique also provides valuable insights into particle rotation. Slower recording speeds, such as 50 kHz, can still capture the trajectories of particles moving in straight lines, but they may not capture inter-particle collisions and complex particle motion. The velocities presented in this study were calculated without considering out-of-plane motion, so these numbers represent minimum ejection velocities [56]. Zhao et al. utilized X-ray imaging to demonstrate the direction and trajectories of particles in the PBF-LB process in Figure 6a,b, respectively. Figure 6c and d display statistics on particle ejection speed and direction under two laser power conditions in PBF-LB processes. When comparing laser powers of 340 W and 520 W, the former ejects fewer particles from the powder bed and exhibits a lower maximum particle speed (12.3 m/s for the lower power case vs. 14.9 m/s for the higher power case), indicating less intense particle motion at lower laser power. The angular distribution of ejection (φ) around φ = 90° (vertical direction) is expected to be symmetrical if a sufficient number of melting events are counted. However, the dynamic flow of the melt pool affects the distribution, resulting in an asymmetric distribution of ejection angles. A subtle difference between the two laser power cases is observed, with 340 W displaying a more symmetrical angular distribution, suggesting a less turbulent melt pool flow [65].

In another study, Wolff et al. utilized X-ray imaging to gain insights into the complex dynamics of the melt pool in directed energy deposition (DED) additive manufacturing. Figure 7 presents raw and differential X-ray sequences from the DED process. The parameters used were a laser power of 200 W, scan speed of 100 mm/s, and powder flow rate of 320 mg/s. The differential images enhance the contrast of the melt pool, revealing dynamic surface variations indicated by red arrows in the raw photos. The study demonstrates that the stability of the melt pool surface is affected by fast-moving particles, particularly when the laser beam spot size, and thus the melt pool dimension, is smaller compared to the particle size. This leads to deformations and surface roughness. These findings emphasize the significance of considering powder characteristics and process control in order to achieve the desired outcomes in DED builds [5].

3. Combined Use of In Situ Acoustic Measurement Techniques with X-ray Imaging

In Section 2, it is noted that various physical phenomena can occur at different speeds, extending up to 10 MHz (or 100 ns). Therefore, it is crucial to consider time synchronization when using multiple measurement techniques simultaneously to gain a comprehensive understanding of the underlying physical processes. Acoustic emission during laser material interactions can provide a unique acoustic signature of different melting regimes (conduction, stable keyhole, unstable keyhole) [66,67,68]. At the same time, X-ray imaging allows for real-time observation of the melt pool and keyhole during the PBF-LB process. However, there are two main reasons why acoustic emission is not widely used in industrial laser processing: First, the signals are highly susceptible to background noise; secondly, establishing a correlation between the AE signal and the actual occurrences is extremely challenging [38]. Nevertheless, researchers have been able to visualize the entire dynamic process of laser material interactions by using high-speed X-ray imaging combined with AE measurements. This allows them to simultaneously acquire AE signals and determine different regimes by applying the softmax function [69] for discrimination [24]. This approach is helpful for defect detection, such as pore formation, and offers valuable in situ monitoring for quality control and efficiency improvement in PBF-LB [24]. Some studies have used AE signals to detect anomalies and defects in additive manufacturing processes [66,67,68]. For example, Tempelman et al.’s study focused on identifying keyhole pores by synchronizing the acoustic signal with the laser position at each moment, which facilitated accurate labeling of defect locations [70].

Typically, two categories of emissions arise within the laser processing zone: structure-borne acoustic emission (SBAE) and air-borne acoustic emission (ABAE) [71]. They can be considered contact and non-contact types, respectively. SBAE sensors require direct contact with the material, while ABAE measurements can be performed remotely without physical contact. Therefore, SBAE and ABAE sensors are referred to as contact and non-contact types of AE sensors, in that order. In ABAE recording, standard membrane microphones are commonly used, while in SBAE, piezoceramic transducers are mostly utilized. There are different detection principles between these two types of AE, including the nonlinear behavior of acoustic waves in solids and the linear propagation of acoustic signals in air. Due to higher damping in air than in solids, SBAE supports a broader frequency spectrum [72]. The widespread use of SBAE in laser welding for quality control is hindered by the challenging reproducibility of sensor coupling to the workpiece, which requires quick sensor reattachment when the workpiece changes [73].

AE relies on the detection and analysis of acoustic emissions that occur when internal stress is released within a material. Contact-type acoustic techniques rely on the nonlinear behavior of acoustic waves to detect and characterize defects, while non-contact techniques are based on the linear propagation of acoustic signals and are suitable for remote sensing applications. In contact AE sensing, nonlinear parameters such as wave distortion and harmonic generation are often used to detect and characterize defects or structural changes within the material [74,75]. These nonlinear effects occur due to the interaction between stress waves and material properties [76]. The detection of SBAE signals is also based on the nonlinear behavior of acoustic waves, allowing for the identification of specific phenomena related to material damage or deformation [77]. SBAE techniques typically require direct contact with the material or structure being investigated to detect mechanical vibrations or acoustic waves traveling through solids. This enables SBAE techniques to provide insights into the dynamic behavior of materials, including the initiation and propagation of defects such as cracks or microfractures [78].

In this section, we will review acoustic measuring techniques combined with X-ray imaging. We will discuss two types of acoustic sensors that are used for acoustic emission spectroscopy with X-ray imaging. Furthermore, we will review the progress of multi-sensors used in combination with other measurement techniques for these two methods (AE sensor and X-ray imaging).

3.1. Non-Contact (ABAE) Type Acoustic Emission Sensor Combined with X-ray Imaging

Wamster et al. [38] were pioneers in combining X-ray imaging with AE signal analysis for laser manufacturing processes. They used a piezo acoustic sensor, specifically an SBAE sensor, that was not in direct contact with the sample but rather with the sample holder. Therefore, this approach falls under the category of ABAE type, which does not directly capture acoustic waves from the sample during the laser process. Their study utilized a signal processing technique called wavelet packet transform (WPT) to analyze non-stationary AE signals. WPT divides the time-frequency space of the acquired AE signals into a finite number of narrow frequency bands, allowing for a more detailed analysis of how the signal’s frequency content changes over time. By selecting only the relative energies of specific frequency bands (from the third to the seventh decomposition levels), they were able to exclude noise from the experimental setup and surrounding environment. After pre-processing the data, the gradient boost technique is applied for machine learning classification, completing both pre-processing and classification within 30 ms. Nasab et al. [24] successfully synchronized X-ray imaging with an optical microphone positioned 100 mm away from the processing area of the build plate. Figure 8a illustrates stochastic regime transitions during the laser processing of 316 L stainless steel, highlighting variations in the melt pool. Figure 8b displays time-evolving AE signals and a spectrogram for keyhole phases, aiding in the identification of different regimes. This integrated setup efficiently captured air-borne acoustic emissions ranging from 10 Hz to 1 MHz during the manufacturing process. X-ray imaging provided real-time insights at a rapid frame rate of 100 microseconds (10 kHz). They optimized regime discrimination in PBF-LB monitoring, which is crucial for real-time PBF-LB/laser melting monitoring, and enhanced the synchronization of X-ray imaging with AE signals using image processing techniques. These techniques include time-windowing for laser tracking, denoising with median filters (with a 7 × 3 × 3 footprint where the first dimension is time and the other two are pixels’ 2D spatial position), and a strip removal algorithm [79] to eliminate diagonal lines caused by the constant speed of texture. Additionally, edge detection using polar coordinates is employed. These techniques improve data quality, aiding in real-time PBF-LB monitoring and control.

The sudden increase in amplitude of the AE signal is associated with the phenomenon of evaporation and the creation of a microjet. This microjet is the main factor in generating acoustic waves during the keyhole melting process. Acoustic waves in laser-based manufacturing detect changes in gas momentum that result from various phenomena, such as metal vaporization, microjet formation, capillary oscillation, surface tension gradients, liquid collisions, microjet penetration, pore formation, and periodic overheating explosions. These events collectively lead to periodic oscillations of the keyhole within a frequency range of 20–50 kHz. These oscillations can be detected using AE sensors (see Figure 8b) [24].

De Foranoir et al. (Nasab’s co-worker) [80] developed a method for in situ pore healing using operando X-ray imaging to understand the mechanisms of pore elimination. They also demonstrated the potential of AE as a monitoring technique for porosity healing during laser remelting by aligning AE signals with X-ray images without relying on machine learning algorithms. Figure 9 provides an overview of time-domain AE signal analysis during laser irradiation. This includes visualizing raw signals, shifting the time axis, superposing AE signals with video events, and highlighting a specific pore removal event. To synchronize X-ray imaging with the corresponding AE signal in the time domain, the start of the laser processing, denoted as “${\mathrm{t}}_{\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{r}\mathrm{t}}$”, was determined by identifying the first significant increase in signal intensity. The time shift $\mathrm{t}-{\mathrm{t}}_{\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{r}\mathrm{t}}$ was then applied to the time axis of the AE signal (Figure 9), allowing for the AE signal to be aligned with the X-ray video (which started with laser triggered control). Similar to determining ${\mathrm{t}}_{\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{r}\mathrm{t}}$, the end of the laser process (${\mathrm{t}}_{\mathrm{e}\mathrm{n}\mathrm{d}}$) can be determined based on signal characteristics. It is important to note that the estimation of the start time can be prone to error. However, due to the different frequencies (2 MHz for acoustic, 10 kHz for X-ray), a 100-point error (for example) in identifying the laser start (${\mathrm{t}}_{\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{r}\mathrm{t}}$) has minimal impact on aligning acoustic signals with X-ray images. This error is only half the duration of one X-ray frame.

The synchronization accuracy can be affected by reverberation time and noise interference in AE sensors. Conventional capacitive and piezoelectric microphones are susceptible to electromagnetic interference from surrounding industrial noise, and their bandwidth is limited to 100 kHz. These limitations may explain why acoustic process monitoring is not widely used in industrial production. In contrast, an optical microphone (shown in Figure 10) is immune to electromagnetic interference and can have a bandwidth of up to 1 MHz in air. Figure 10 illustrates the optical microphone’s Fabry–Pérot cavity, which uses interference to detect density changes, ensuring high sensitivity and compact size. Furthermore, reverberation times in industrial environments are typically longer for audio frequencies, often in the hundreds of milliseconds, making it challenging to accurately monitor rapid acoustic events. It is virtually impossible to achieve reverberation times shorter than 10 ms up to 100 kHz. To address this issue and improve temporal precision, it is imperative to monitor ultra-high acoustic frequencies beyond 100 kHz, where shorter reverberation times are possible, especially in environments with acoustically reflective surfaces [81].

In studies that use X-ray imaging in synchronization with AE signals, there is a lack of attention given to the significance of accounting for the time-lag effect in acoustic signals. This observation holds true even when AE is combined with other measurement techniques, such as CCD cameras, IR cameras, and pyrometers, as discussed in references [82,83,84]. The time-lag effect is caused by various factors, including the distance between the microphone and the laser–material interaction zone, as well as the characteristics of the medium through which the signals travel. It is crucial to carefully consider this aspect to ensure the accurate interpretation and practical application of AE data in laser manufacturing processes, while also avoiding misinterpretations when connecting underlying process phenomena to AE signal signatures.

3.2. Contact (SBAE) Type Acoustic Transmission Sensor Combined with X-ray Imaging

The application of contact or SBAE-type ultrasound sensors in monitoring laser manufacturing processes can be affected by the geometry of the melt pool, including width and depth. However, these sensors are unable to provide precise information about the dynamic behavior of the melt pool. Therefore, synchronized measurements are necessary to determine how the dynamics of the melt pool impact the ultrasonic response [85].

Gillespie et al. conducted a study on laser–material interaction using synchrotron X-ray imaging and two ultrasound transducers immersed in water (see Figure 11a). They adjusted the angle of the ultrasound transducer so that the refracted shear wave in the aluminium sample was directed towards the heat-affected zone (HAZ) (Figure 11b). It is known that shear waves passing through solid-to-liquid media exhibit highly scattering behavior, which helps distinguish the signal from different melt pool states. By utilizing a pair of transducers, the researchers had the option to perform either pulse-echo (PE) assessments, where a single transducer functions as both the transmitter and receiver, or pitch-catch (PC) evaluations, where one transducer sends the signal while the other captures it [27].

In order to achieve a high time resolution in ultrasound signals (and therefore a high time resolution of corresponding physical events), a higher pulse rate frequency (PRF) is required. However, it is essential to note that the PRF is restricted by the time it takes for the ultrasound to travel from the melt pool, scatter, and then be received by the transducer. This is referred to as the time-of-flight [27]. A dual-channel ultrasonic pulser-receiver (JSR DPR-500) controls transducer excitation and reception with a 10 ns duration pulse at a user-set PRF to ensure precise synchronization between ultrasound and X-ray imaging. The received ultrasound data is filtered and sent to a digitizer card (NI PXIe-5162), initiated by another trigger pulse from the pulser-receiver, ensuring exact timing. Simultaneously, a high-speed camera (Photron FastCam SA-Z) captures X-ray images at 28.503 kHz, perfectly synchronized with the ultrasonic PRF. This allows for the precise capture of an image when the ultrasound scatters from the melt pool at a specific time interval, providing accurate alignment of physical events and ultrasound signatures for analysis [27]. Figure 12 displays the synchronized X-ray imaging alongside the ultrasound signal, capturing the events before, during, and after laser activation using two pitch-catch (PC) and pulse-echo (PE) configurations. As the melt pool deepens, the ultrasound’s time-of-flight decreases. This can be seen around time slices 4.25, 5.09, and 5.65 ms in the PC configuration (Figure 12a) and 2.21, 3.89, and 4.84 ms in the PE configuration (Figure 12b) [27].

3.3. Multi Sensors (Three Sensors or More) Combined with Acoustic Emission Sensor and X-ray Imaging

Some studies have utilized multiple sensors, including XRI and AE sensors, along with other measuring techniques, to provide information from various and complementary aspects.

Shevchik et al. [43] combined laser back-reflection (LBR) and AE signals with X-ray imaging to monitor laser welding in real time. Figure 13a displays X-ray images of aluminum during a 10 ms laser pulse, showing different phases and events. Figure 13b presents the pixel intensity of the keyhole channel. Figure 13c and d show the LBR and AE signals, along with their wavelet spectrograms, for laser welding. LBR detects surface behavior, while AE captures dynamic changes in the melt pool and keyhole, and X-ray imaging provides a visual representation of the welding process. LBR and AE signals respond differently during welding, reflecting distinct physical phenomena. LBR signals exhibit sharp changes at the start and end of the welding process, measuring the reflected laser energy. In contrast, AE signals show a slight increase in amplitude at the beginning, likely related to the size of the melt pool, and gradually decrease as the melt pool shrinks towards the end. The combination of LBR, AE, and X-ray imaging provides a comprehensive understanding of laser welding, enabling defect detection and real-time control (Figure 13).

The study examines two types of defects, blowout and pore, along with three common categories, namely, conduction, stable keyhole, and unstable keyhole. While LBR can detect blowout by measuring frequency attenuation, AE cannot. This approach is also effective in distinguishing between stable and unstable keyhole states, which is crucial for preventing defects in laser welding. The method can detect pore formation and removal and improve classification accuracy by combining LBR and AE features in real-time welding defect reduction [43]. X-ray phase contrast imaging is used to enhance sensitivity, particularly at phase boundaries (solid, gas, and vapor), where X-ray beam refraction is most noticeable. Both the AE sensor and LBR are triggered by laser control, and wavelet packet transform (WPT) is employed to generate unique signatures for physical events. WPT offers several advantages, including reducing data size, denoising signals by selecting specific frequency bands, and enabling the interpretation of signals over time based on chosen frequency bands. The wavelet spectrogram enhances precision by revealing specific frequency patterns in both signals during different welding phases.

In order to achieve highly accurate classification of various regimes within metal-based PBF-LB processes, Pandiyan et al. [25] have introduced an innovative hybrid deep learning model that merges convolutional neural networks (CNNs) with long short-term memory (LSTM). This achievement is based on data gathered from four sensors, including three photodiodes (back reflection, visible, and infra-red) along with a structure-borne AE sensor. The efficacy of in situ monitoring for PBF-LB processes is further substantiated through X-ray imaging. Figure 14a depicts the mini-SLM configuration with an AE sensor and optical detectors, while Figure 14b shows schematics of the heterogeneous sensing system featuring an AE sensor and photodiode detectors. During the experiment, a fixed focus collimator is positioned at a 30-degree angle, 5 cm from the HAZ (as seen in Figure 14a), to collect optical emissions from the melt pool. These emissions are divided into three channels using various optical filters to isolate specific wavelengths. The BR Si photodiode primarily captures reflected laser light at 1070 nm, providing insights into the absorption and utilization of laser energy in the process [86]. The InGaAs photodiode (PDA20CS2) can detect light in the 800–1700 nm range, utilizing a low-pass optical filter for wavelengths exceeding the laser radiation. The Si photodiode captures visible and NIR light using a short-pass optical filter. The AE sensor (PICO HF-1.2, Physical Instruments, Physical Acoustics, MISTRAS Group, NJ 08550, USA) with a frequency range of 500–1850 kHz serves as a structure-borne sensor, mounted at the base’s bottom and recorded at a sampling rate of 3 MHz to adhere to the Nyquist Shannon theorem. Signal acquisition automatically begins for all channels via a photodiode detector, triggered by a substantial increase in light intensity near the laser’s wavelength upon interaction with the powder bed, resulting in an analog voltage. Crucially, all four channels simultaneously initiate acquisition, ensuring default synchronization [25]. In this way, the authors comprehensively assess various aspects of the PBF-LB process, enabling more informed decision-making within the process regime by receiving a wider range of radiated wavelengths.

To detect defects and correlate them with the exact time of formation, Martin et al. used X-ray imaging in combination with a diagnostic sensor suit. This improved the understanding of the process and quality control in PBF-LB. Figure 15a displays the layout of the in situ diagnostic system, which records thermionic, optical, and acoustic signals during laser exposure. Figure 15b and c show the schematics and information exchange for the laser, X-ray imaging, and diagnostic system control. The sensor suit included a photodiode, aligned with the laser beam, as well as a second dielectric mirror and a microphone positioned in the chamber (Figure 15b). All of these sensor signals were recorded synchronously at 470 kHz using an FPGA-based laser interface module. The synchronization process for X-ray imaging with diagnostic sensors begins with a command from the RTC5 control card. This command initializes the position of the laser scan head and opens the high-speed shutter. However, there is a predetermined delay of approximately 5 ms before the shutter opens. As a result, the laser control signal, which includes the power signal enabling laser irradiation and the gate signal controlling camera recording and sensor acquisition timing, is passed through an FPGA. After applying the necessary programmed delay, this signal is transmitted to the laser (see Figure 15c). This process ensures that the high-speed camera captures images in sync with the diagnostic sensor data, achieving remarkable synchronization with a precision of 2.1 microseconds (at a frequency of 470 kHz) [87].

4. Conclusions and Future Directions

The development of AE measurement synchronized with X-ray imaging over time is summarized in

Table 1. Setup characteristics of synchronized AE and X-ray imaging related to research works over the years.

	Research Works	K. Wasmer et al., 2018 [38]	C. Kube, et al., 2018 [85]	S. Shevchik et al., 2020 [43]	Gillespie, J. et al., 2021 [27]	A.A. Martin et al., 2022 [87]	De Formanoir et al., 2023 [80]
Applied Facilities
X-ray source setup characteristics	Synchrotron facility	Beamline ID19 of the ESRF	European Synchrotron (ESRF)	European Synchrotron (ESRF)	European Synchrotron (ESRF)	Stanford Synchrotron Radiation Light Source (SSRL)	(TOMCAT) beamline of the Swiss Light Source (SLS)
	Laser and X-ray acquisition synchronization	-	Laser control triggered	Laser control triggered	Laser control triggered	-	TTL triggered, provided by laser control card
	X-ray spectrum	Pink beam mean energy was 26.3 keV	Pink beam with average power around 26 keV	Pink beam with average power around 26 keV	Pink beam with average power around 26 keV	7.4 keV	Filtered within 10 to 55 keV and peak around 20 keV
X-ray detector characteristics	Scintillator	250 µm-thick Ce-doped Lu3Al5O12	-	250 µm-thick Ce-doped LuAG (Lu3Al5O12)	European Synchrotron (ESRF)	100 µm-thick YAG:Ce (Crytur)	150 µm-thick LuAg:Ce
	Pixel	-	896 × 776 pixels	-	896 × 776 pixels	1024 × 672 pixels	2016 × 2016 CMOS Imaging
	Pixel size	11 µm	1.923 µm	11 µm	1.923 µm	2 µm	11 µm
	Frame rate	28,000 to 36,000 fps	28.503 kHz	28,762 fps	28.503 kHz	20 kHz	10 kHz
	Exposure time	-	-	30 µm	-	45 µm	95 µm
Laser characteristics	Type	Single-mode fiber laser source star fiber 150 P (fiber laser star fiber 150 P/300 P)	fiber laser	Single-mode fiber laser source star fiber 150 P (Coherent Inc., Belp, Switzerland)	Fiber laser	Single-mode CW Yb-fiber laser (IPG Photonics, YLR-500-AC-Y14)	Pulsed mode
	Wavelength	1070 nm	1070 nm	1070 nm	1070 nm	1070 nm	1070 nm
	Spot diameter	30 µm	120 µm	30 µm	120 µm	73–135 µm	13.75 µm
	Maximum Power	-	477 W	1.5 kW	477 W	500 W	500 W
Acoustic sensor characteristics	Type	Piezo acoustic sensor PICO	Ultrasound transducer	Piezo sensor PICO HF-1.2 (Physical Acoustics, MISTRAS Group, Germany)	Piezo ultrasonic transducers	PCB Piezotronics	Membrane-free optical microphone (Eta250 Ultra, XARION Laser Products)
	Frequency range	500–1850 kHz [88]	Resonance frequency: 25 MHz	Sample rate of oscilloscope: 50–2000 kHz	Center frequency of 25 MHz	4 to 100,000 Hz (+2/−3 dB)	10 Hz to 1 MHz
	Place	Contact to holder	Contact	Contact	Contact	Non-contact	Non-contact
	Sampling rate	10 MHz	500 MHz	10 MHz	-	470 kHz	2 MHz
Detected phenomena		Conduction welding, stable keyhole, unstable keyhole, and spatter	Melt pool characterization	Process instabilities: conduction welding, stable keyhole, unstable keyhole, blowout, and pores	Melt pool characterization	Fluctuations in the melt pool, pore formation, defect formation	Pore removal, remelting, defect healing

.

It can be observed in Table 1 that the development of using a combination of X-ray and AE analysis has progressed gradually, but it still requires improvement in order to achieve higher resolution, a deeper understanding of phenomena, and faster recognition. For instance, the frequency range of the AE sensor has been expanded (10 Hz–1 MHz), which has facilitated the study of remelting and pore removal (refer to Table 1). The smaller melt pool resulting from the use of a smaller beam diameter (13.75 µm) has been examined using a facility with higher resolution. By enhancing the exposure time, it becomes possible to investigate time-related phenomena, such as the formation of melt pools and defects. These studies are valuable for process control and enhancing the quality of components during laser processing. However, considering the inherent specificity of acoustic signals, such as the time-lag effect, would allow for more-accurate synchronization of AE signals with X-ray images in future studies. This would aid in interpreting the AE signal in greater detail, which can be applied in machine learning and artificial intelligence for automatic detection. Furthermore, it is evident that multi-sensor systems provide more comprehensive information on defect formation and enable real-time monitoring.