1 Introduction

Hybrid laser-arc welding was first proposed by Steen in the late 1970s [1]. Hybrid laser-arc welding combines a laser beam with high energy density and an electric arc with large heat input. Their high efficiency, high stability, and low distortion make them far superior to pure laser welding and pure arc welding [2,3,4]. Hybrid laser-arc welding has been widely applied in machinery manufacturing, aerospace, rail transit, and many other fields [5]. The full penetration of the weld during the welding process is fundamental to ensuring the quality of the joint. Accordingly, penetration level is critical to the quality of the weld [6]. In fact, in the process of hybrid laser-arc welding, the factors that determine the occurrence of incomplete-penetration defects are not only the process parameters related to welding heat input. The stability of wire/powder feeding, workpiece assembly, position matching of laser and arc heat sources, and other factors will affect the penetration state of weld bead. Non-destructive testing and destructive testing are the most commonly used methods for inspecting weld quality. Generally, post-welding non-destructive testing is used to check for weld penetration defects. As modern industry strives towards more efficient forms of production, on-line identification of weld penetration in hybrid laser-arc welding has become a key technology and major research direction in welding automation. As such, this field has drawn the attention of many scholars.

Previous studies have proposed some on-line methods for measuring penetration levels. Acoustic wave signals and visual images are the most common means of weld quality assessment. Lu et al. [7] built a linear model that relates arc acoustic signal to arc length with respect to pulsed gas tungsten arc welding (GTAW), allowing precise prediction of the arc length to more accurately measure penetration levels and weld quality. Wang and Zhao [8] collected acoustic wave signals during plasma arc welding using a condenser microphone. Their experimental results showed that changes of the signal power spectrum can adequately reflect weld penetration levels. Liu et al. [9] proposed the use of trained neural networks to process information on the properties of arc acoustic signals during the welding process. This technique can efficiently achieve penetration recognition. Wu et al. [10] introduced signal acquisition methods that are based on visual and acoustic sensors. Their hybrid processing method successfully identified penetration levels during the plasma arc welding process. Liang et al. [11] used 3D vision sensors to collect images of the surfaces of weld pools in GTAW welding and proposed a support vector regression model to overcome the shortcomings of traditional neural networks, which require large amounts of training data. This allowed their method to more effectively measure weld penetration levels. Song et al. [12] analyzed acoustic signal characteristics in non-keyhole, keyhole, and cutting modes during plasma arc welding and built a system for measuring penetration levels to an accuracy of 93.23%. Lv et al. [13] trained a typical back propagation artificial neural network prediction model based on the characteristics of acoustic signals generated during pulsed GTAW welding. This method was able to predict penetration levels to an accuracy of 80–90%. Wu et al. [14] used images of the back keyhole of the workpiece to build an adaptive network fuzzy inference system for monitoring weld penetration during plasma arc welding. These methods have been successful in identifying penetration levels.

The introduction of classification algorithms in the field of penetration recognition is also a research hotspot in recent years. Huang and Kovacevic [15] analyzed the time domain and frequency domain of the acoustic emission signal in the laser welding process and used multiple regression analysis and neural network to achieve high-precision recognition of the weld penetration depth. Chen et al. [16] designed a multi-information fusion analysis system including welding current, voltage, vision, and sound pressure signals and used backpropagation neural networks to predict the penetration state of the joint. The results showed that the prediction accuracy of a multi-information fusion analysis system was much greater than that of a single signal analysis system. Sumesh et al. [17] used the acoustic emission signal characteristics to identify three states of welded joints, including good weld, lack of fusion, and burn through, and introduced a random forest classification algorithm to classify the joint states. Yusof et al. [18] extracted the characteristics of the collected acoustic emission signals during laser welding experiment and used the support vector machine (SVM) classification algorithm to identify two states of incomplete penetration and complete penetration. Zhu et al. [19] analyzed the fusion of the acoustic and visual signals of the GTAW process and used the particle swarm optimization SVM algorithm to identify the penetration state.

A literature review shows that many studies have used airborne acoustic signals to measure penetration levels. However, airborne sensors for acoustic signals are relatively inefficient. The noise signal in the environment has a strong interference on the penetration acoustic signal. Structure-borne sensors are able to obtain more information from acoustic signals [20]. The structure-borne acoustic sensing method can greatly improve the anti-interference ability to environmental noise and the detection ability to useful acoustic signal. It can meet the conditions for large-scale use in the welding industry because of the advantages of low cost, high efficiency, and strong versatility. In order to obtain detailed information on penetration levels during pulsed laser and plasma transferred arc (laser-PTA) hybrid welding, the present paper used structure-borne acoustic sensors to measure weld penetration. We examine how laser-PTA hybrid welding arising from incomplete penetration can be identified through an analysis of characteristic acoustic signals, thereby promoting the development of systems for measuring penetration levels.

2 Experiment details

A laser-PTA hybrid welding system was used in the experiment, as illustrated in Fig. 1a. The welding heat source is composed of a continuous plasma arc and a pulsed Nd:YAG laser beam that act in tandem on the molten weld pool. A plasma plume is generated during the interactions between the laser beam, plasma arc, and molten weld pool. A system for monitoring acoustic signals in real time was used during the laser-arc hybrid welding experiment. Figure 1b shows the schematic of this laser-PTA hybrid welding experiment. The experiment used structure-borne sensors to reduce the attenuation of acoustic energy during the transmission process and prevent interference from environmental noise. In order to reduce reflection and attenuation of structure-borne acoustic signal at the interface, a vacuum silicone grease couplant was applied evenly on the contact surface between the sensor and the workpiece. A structure-borne acoustic sensor was installed on the workpiece surface more than 50 mm away from the molten weld pool to ensure the sensing signal strength and avoid the damage caused by spatter. The propagation velocity of sound wave in solid steel is up to 3000–6000 m/s. The signal delay caused by the distance between the sensor installation position and the molten pool is only a few microseconds. Therefore, regardless of the attenuation and the time delay of the acoustic emission signal in stainless steel, it is considered that the relative position change of the sensor and the molten pool does not affect the characteristics of the detected acoustic signal.

Fig. 1
figure 1

Schematic of (a) welding process and (b) acoustic signal monitoring system in laser-arc hybrid welding

Full size image

The sensors used in experiment are structure-borne; there is little attenuation of acoustic energy transmitted through solids. Due to the high propagation speed of structure-borne acoustic wave in solid, the position change caused by laser heat source movement will not bring unacceptable time delay of acoustic signals. Therefore, the sensors are not affected by changes in installation position. During the experiments, all data on laser-induced acoustic excitations were extracted and analyzed. Acoustic signals collected during the welding process are amplified by the preamplifier, filtered by the signal processor, and sent to the computer through the data acquisition unit (ART PCI-8757) for further processing and analysis. Sampling rates in the experiments were set to 150 kHz based on the Nyquist criterion [21]. LabVIEW, a sampling and analysis software developed by National Instruments, is used to acquire data and further analyze the acoustic signals. Table 1 shows the welding parameters used in the experiments, which used 304 stainless steel plates with a thickness of 2 mm, 3 mm, and 6 mm, respectively. During the welding process, the fixture is used to keep the assembly gap of the workpiece at 0.3 mm, so that the research conditions are consistent with the assembly accuracy of the production conditions. Gaseous argon was used as both the plasma gas and the shielding gas. The same peak laser power levels and other pulse parameters are used in each experiment. Penetration levels were measured based on changes in welding speed and arc current. In particular, arc current was lowered for 6-mm plates as it was necessary to achieve incomplete penetration. However, arc current applied to 3-mm workpieces were increased to 80 A to ensure their complete penetration under the experimental conditions.

Table 1 Welding parameters used in experiment
Full size table

3 Results and discussion

3.1 Acoustic signal characteristics

In order to obtain different weld penetrations, the welding currents for 2-mm plates and 6-mm plates were respectively 40 A and 60 A, while other parameters were maintained at a constant level (welding speed 4 mm/s, laser frequency 100 Hz, peak power 3 kW, pulse duration 2 ms). Figure 2 shows the weld cross-section morphology and the acoustic signals detected during the welding process. It can be observed that the 2-mm plate was fully penetrated, while the 6-mm plate was only partially penetrated. In other words, there is no penetration defect in the 2-mm plate, while there are significant incomplete-penetration defects in the 6-mm plate. However, an ocular observation of the time-domain characteristics of the acoustic signals in Fig. 2 yields no significant features that indicate the presence of defects. Therefore, it is not enough to merely analyze acoustic signals in the time domain.

Fig. 2
figure 2

Weld cross-section morphology and acoustic signals of (a) 2 mm and (b) 6 mm thickness plate

Full size image

In order to study the differences in acoustic signals for different penetration levels, we welded 2-mm-thick and 6-mm-thick plates at speeds of 4 mm/s, 5 mm/s, and 7 mm/s, respectively. Acoustic signals were collected in real time during the welding process. Weld cross-section morphologies and signals in the time domain are shown in Fig. 3. In addition, the components of these acoustic signals are displayed in the frequency domain (0–60 kHz) through an analysis of the self-power spectrum. The results for 2-mm-thick plates show that welds are fully penetrated when welding speed is 4 mm/s or 5 mm/s. The weld reaches a critical penetration state when the welding speed is 7 mm/s. For 6-mm-thick plates, results show that all welds are in a state of incomplete penetration. By observing the characteristic spectral lines in the frequency domain, we can see that there are four frequency components that should be noted. These spectral lines occur near frequencies of 57 kHz, 37 kHz, 20 kHz, as well as at frequencies below 10 kHz, as shown in Fig. 3. The characteristic spectral lines near frequencies of 57 kHz and 37 kHz are not closely related to incomplete-penetration defects. In contrast, the spectral lines near frequencies of 20 kHz and below 10 kHz are suspected to be related to incomplete-penetration defects. This is especially the case for signal compositions with frequencies below 10 kHz. In the case of full-penetration welds, such as in the 2-mm plates, there are almost no signal components with frequencies below 10 kHz. For incomplete-penetration welds, such as in the 6-mm plates, we observe many signal components and a strong signal intensity at frequencies below 10 kHz. Figure 4 shows a clearer analysis of the results with respect to the frequency domain. This indicates the need for special analysis of acoustic signals in these two frequency bands.

Fig. 3
figure 3

Characteristics of acoustic signals detected under different welding speeds for (a) 2-mm-thick plate and (b) 6-mm-thick plate

Full size image
Fig. 4
figure 4

Characteristic frequency domain of acoustic signals detected under different welding speeds for (a) 2-mm-thick plate and (b) 6-mm-thick plate

Full size image

Characteristic frequency domains of these two frequency bands are subdivided as shown in Fig. 4. According to the distribution of acoustic signals in the frequency domain below 10 kHz, the lower cut-off point is set at 3 kHz, and the higher cut-off point is set at 7 kHz. For signals near the characteristic spectral line at 20 kHz, a narrower characteristic frequency band is set so that signal characteristics for different welding speeds can be analyzed more accurately.

Next, a Butterworth band-pass filter was used to extract these characteristic signals. For the full-penetration welds in 2-mm plates, very little information on their features could be extracted from the acoustic signals. It can be seen from Fig. 5a that, when the welding speed is 4 mm/s, almost no characteristic signal information of value can be extracted. This holds true for frequency bands from 3 to 7 kHz and around 20 kHz. An analysis of self-power spectrums, illustrated in Fig. 3a, shows no distribution of signal components in these two frequency bands. At welding speeds of 5 mm/s or 7 mm/s, increases in welding speed affect weld penetration levels, but the welds remain at full or critical penetration. Although some acoustic signal components were extracted, these extracted signals have fewer event characteristics.

Fig. 5
figure 5

Extraction of acoustic signals in characteristic frequency domain for (a) 2-mm-thick plate and (b) 6-mm-thick plate

Full size image

Characteristic signals extracted from welds in 6-mm plates are also different from those from 2-mm plates due to significant differences in their penetration levels. Figure 5 shows the acoustic signals extracted in characteristic frequency domains. The acoustic signals extracted by filtering in the frequency band near 20 kHz also have fewer event characteristics. However, acoustic signals extracted in the 3 to 7 kHz frequency band have characteristics that are different from each of the extracted signals mentioned above, the difference being that these extracted acoustic signals have obvious event characteristics: signal pulses appear in clusters to form an “acoustic event.” Fig. 6 highlights the features of these acoustic events in the time domain. It is clear that the frequencies of these acoustic events are analogous to those of laser pulses, indicating that these extracted acoustic events were generated by pulse laser excitation. This phenomenon is present only in acoustic signals from incomplete-penetration welds.

Fig. 6
figure 6

Acoustic events in acoustic signals

Full size image

From the above analysis, it can be determined that signal frequencies lower than 10 kHz constitute a signal frequency band characteristic of incomplete-penetration defects. The following section will analyze the mechanisms underlying the generation of these characteristic signals.

3.2 Physical mechanism of laser-excited acoustic wave

Figure 7 shows a schematic of phonation sources in a molten pool. Sources of phonation are constituted by certain physical processes during the course of welding, such as the electromagnetic effects of the plasma arc, energy coupling between the laser beam and the plasma arc, liquid-phase convection in the molten pool, plasma gas flow or recoil force in the keyhole, and so on. These physical phonation sources produce acoustic signals with different frequency characteristics, as well as different characteristic spectral lines in the self-power spectrum (as shown in Fig. 3). An intense plasma eruption effect is generated in the keyhole when a laser beam with a high energy density acts on the molten pool. The plasma then generates a downward recoil force on the bottom of the molten pool, causing it to vibrate at a unique frequency. Weld penetration level can be determined by this unique vibration frequency, which is closely related to the action process of the laser beam. Therefore, sources of phonation for incomplete-penetration defects are caused by interaction between laser energy and the molten pool. Accordingly, the frequency of acoustic events extracted in the frequency band below 10 kHz (characteristic frequency band of incomplete-penetration defect) will be the same as that of laser pulses as shown in Fig. 6.

Fig. 7
figure 7

Schematic of phonation source in molten pool

Full size image

The incomplete-penetration defect is at the bottom of the molten pool. The propagation of acoustic signals from incomplete-penetration defects is closely related to the acoustic transmission structure of defective welds. An incomplete-penetration defect is an acoustic transmission structure—an acoustic cavity. Figure 8 shows the physical structure and acoustic transmission characteristics of an acoustic cavity.

Fig. 8
figure 8

Physical structure and acoustic transmission characteristics of acoustic cavity under (a) low-penetration, (b) deep-penetration, and (c) full-penetration status

Full size image

A low penetration depth implies that the molten pool has a lower volume, while the weld structure has a larger incomplete-penetration defect, which creates a large acoustic cavity, as shown in Fig. 8a. In this structure, the acoustic cavity acts like a loudspeaker, amplifying the acoustic signals from various phonation sources in the molten pool. At the same time, the acoustic wave resonates at a relatively low frequency in the large acoustic cavity. This creates acoustic signals characteristic of incomplete-penetration defects. Because of the small volume of the molten pool, the recoil force generated by laser pulse energy in the molten pool can almost directly impact its bottom and generate a significant vibration effect. On this condition, the frequencies of these acoustic vibration events are analogous to those of laser pulses as shown in Fig. 6. The acoustic cavity of an incomplete-penetration defect is obviously more conducive to the amplification of these types of oscillating wave. In Fig. 8a, this characteristic oscillation wave is denoted “AW,” while acoustic waves from other phonation sources are denoted “AW’.” In other words, when an incomplete-penetration defect is present, acoustic waves characteristic of defects are more readily amplified by the acoustic cavity, thus displaying a relatively low signal frequency.

A deeper penetration level means that the molten pool has a larger volume, while the incomplete-penetration defect in the weld structure is smaller, which creates a relatively small acoustic cavity, as shown in Fig. 8b. Accordingly, the acoustic cavity will have a weaker resonance effect and a weaker amplification effect on acoustic waves from various phonation sources. At the same time, the larger molten pool will have a buffering effect on acoustic oscillation generated by interaction between laser energy and the molten pool. These factors weaken the acoustic wave AW correlating to incomplete-penetration defects.

For full-penetration welds, the molten pool has the largest volume. Obviously, because the “size” of an incomplete-penetration weld defect is zero, the acoustic cavity does not exist, as shown in Fig. 8c. Because of the large volume of the molten pool, recoil force generated by laser pulse energy in the keyhole can barely directly impact the bottom of molten pool. The laser energy generates an oscillation in the molten pool. At the same time, the large volume of the molten pool has a significant buffering effect on acoustic oscillation generated by interaction between the laser energy and molten pool, which greatly compresses the volume of the acoustic cavity. Thus, acoustic waves are not amplified by the acoustic cavity. Thus, there is barely any sign of the acoustic wave AW correlating to incomplete-penetration defects. The small assembly clearance remaining at the bottom of the weld is similar to a small acoustic cavity, which tends to produce high frequency acoustic waves. Therefore, there is no sign of low frequency components below 10 kHz in the acoustic signal self-power spectrum that correspond to full-penetration welds (as shown in Fig. 3a).

3.3 Verification experiment

The above analysis of the physical mechanisms of laser-excited acoustic wave has been experimentally verified through welding experiments on 2-mm and 6-mm plates (as shown in Figs. 3−5). The further validation design is a variable speed welding experiment for 3-mm plates. The four welding speeds of 3 mm/s, 5 mm/s, 7 mm/s, and 9 mm/s are used to weld the same bead with a length of 100 mm. The processing length of each welding speed is 25 mm, and the welding speed increases continuously in turn. The experiment results are shown in Figs. 9–11.

Fig. 9
figure 9

Weld morphology and acoustic signals in variable speed welding experiment

Full size image

As shown in Figs. 9 and 10, welds attain full penetration when welding speed is 3 mm/s. There are almost no signal components with frequencies below 10 kHz in the corresponding self-power spectrum. Depth of weld penetration gradually decreases as welding speed increases from 3 to 9 mm/s. As penetration depth decreases, the depth of an incomplete-penetration defect (h’) increases from 0 to 2.20 mm, as shown in Fig. 11. Increases in incomplete-penetration depth indicate larger acoustic cavities in the weld structure. Accordingly, as per our previous analysis, there will be more significant acoustic signal components with frequencies below 10 kHz (as shown in Fig. 10), which are then extracted as shown in Fig. 11. It can be seen that the frequencies of acoustic events in the extracted signal are entirely consistent with the frequencies of laser pulses. At the same time, the amplitude intensity of acoustic events increases gradually with increases in the depth of incomplete penetration. The experimental results in Figs. 10 and 11 support our hypothesis of the physical mechanisms of vocalization during the occurrence of incomplete-penetration defects.

Fig. 10
figure 10

Penetration state, original acoustic signal, and its characteristic frequency domain for 3-mm-thick plate under different welding speed

Full size image
Fig. 11
figure 11

Relationship of incomplete-penetration depth and characteristic acoustic signal intensity

Full size image

Another validation experiment is designed as a powder-filled welding experiment for 3-mm plates. The stainless steel metal powder is fed into the molten pool coaxially with the plasma arc. The welding parameters used are described in No. 7 process in Table 1. The powder feeding rate increased continuously from 32 to 52 g/min. As we know, melting powder consumes heat source energy. When the powder feeding rate is relatively small, the laser-PTA heat source can ensure the penetration state of the weld bead. When the feeding rate is too high, it will consume or shield more heat source energy, resulting in incomplete-penetration defects of weld bead. The experimental results are shown in Fig. 12. With the increase of powder feeding rate and the thickening of cladding metal on the weld surface, continuous incomplete-penetration defects gradually appear on the back of the weld bead. The extracted acoustic signal and the self-power spectrum distribution characteristics of acoustic signal corresponding to weld beads in different penetration states are shown in Fig. 12. These results support our hypothesis of the physical mechanisms of vocalization during the occurrence of incomplete-penetration defects. In general, the physical mechanisms of vocalization lay a foundation for on-line identification of incomplete-penetration defects in laser-PTA hybrid welding.

Fig. 12
figure 12

Weld morphology and acoustic signals in variable powder feeding welding experiment. (a) Characteristic frequency domain of full-penetration signals and (b, c) characteristic frequency domain of incomplete-penetration signals

Full size image

4 Conclusion

  1. 1.

    Acoustic signals correlating to incomplete-penetration defects have a characteristic frequency domain of 0 to 10 kHz. Acoustic signal components appearing in the characteristic frequency domain indicate the occurrence of incomplete-penetration defects in laser-PTA hybrid welding. When an incomplete-penetration defect occurs during the welding process, characteristic acoustic signals are generated interaction between the pulsed laser beam and molten pool, which are composed of acoustic events with the exact same frequencies as those of the laser pulse.

  2. 2.

    The interaction between the laser beam and molten pool produces a downward recoil force at the bottom of the molten pool, which causes it to vibrate at a unique frequency. This vibration effect produces a phonation source associated with incomplete-penetration defects, which are acoustic transmission structures that this paper describes as acoustic cavities. An acoustic cavity is similar to a loudspeaker, with an acoustic transmission function of amplifying acoustic waves and causing them to generate a resonance effect.

  3. 3.

    The structure of phonation sources and acoustic cavities are affected by different states of penetration. Large incomplete-penetration defects are correlated with large acoustic cavities, which create a more significant amplification effect on acoustic signals correlating to such defects. In relatively small molten pools, it is easier for laser pulse energy to generate vibrations that in turn produce acoustic signals correlating to defects. When incomplete-penetration defects are relatively small, acoustic cavities have a weaker amplification effect on acoustic signals correlating to defects. In the case of full penetration, there is no acoustic cavity, while the large molten pool has a significant buffering effect on laser-generated acoustic oscillations. Accordingly, acoustic signals correlating to defects are generally non-existent. On-line identification of incomplete-penetration defects in laser-PTA hybrid welding is made possible by these acoustic transmission characteristics.