The mobility sector is considered a major contributor to global greenhouse gas emissions and air pollution. As a result, many countries have initiated the transition from fossil fuel-powered to electrified powertrains. This transformation of the powertrain concept will lead to a rapid increase in the production of electric vehicles and, therefore, to a high demand for so-called traction batteries. As a production step of the traction batteries, a connection between the cell connector and the terminal of the battery cell has to be manufactured. For this purpose, laser beam welding is a reliable and efficient joining technique. In order to ensure continuous quality of the welding process during production and to detect defects in real time, reliable process monitoring is required. In this study, spectral and acoustic emissions during laser beam welding were recorded using a laser welding monitor and an optical microphone. For determining possible correlations between the signals and weld defects, various failure cases were generated by the systematic placement of disturbance elements. These elements included a contaminated surface, a gap between the cell connector and the battery cell, and a misalignment of the cell connector. Based on the recorded signals, statistical metrics were calculated. Finally, weld seams with and without defects were compared to assess the capability of both sensor systems for detecting the weld defects.

Laser beam welding is the most frequently used joining process for manufacturing the connection between the cell connector and the battery cell due to the advantages of its low energy input as well as precision and flexibility. However, ensuring uniform weld quality remains a challenge when laser beam welding is used on a large industrial scale. Moreover, a constant quality is vital because, in addition to its safety-related function, the welded joint directly affects the aging behavior of the battery cell and, thus, the battery pack in case of fast charging.1 In order to address this challenge, process monitoring systems are installed. For a reliable and entire quality assessment, sensor data fusion is proposed as an approach to relate the output data from multiple sensors.2 Using spectral and airborne process emissions data during battery cell contacting, a more comprehensive evaluation of weld seam quality can be achieved with the aim of improving the reliability and efficiency of battery systems.

The identification of defects during laser beam welding necessitates the application of two fundamental components. On the one hand, process monitoring systems are installed to capture relevant information during the process. On the other hand, evaluation procedures and methods are required to be able to correlate the measured signals and the resulting weld seam quality. These methods involve statistical analyses and data processing techniques to interpret the signals and to detect anomalies or deviations from the expected behavior. Hence, Secs. II A and II B provide a brief overview.

Process monitoring systems are crucial in quality assurance and can be classified based on various features. A criterion is the temporal aspect, distinguishing between offline and online procedures. The latter can be further categorized as pre-process, in-process, or post-process depending on the measurement position relative to the process zone. In-process monitoring uses measurement technologies, including image processing, optical, acoustic, x rays, and combined methods.3 Optical methods encompass one-dimensional, two-dimensional, and three-dimensional sensor systems, each with distinct detection mechanisms. One-dimensional systems, such as photodiodes, provide averaged signals without spatial information. Two-dimensional systems offer spatial resolution, while three-dimensional systems detect emissions based on angles.4 

Microphones, categorized under acoustic technologies, convert sound waves into electrical signals using different principles. Dynamic and condenser microphones utilize mechanical components, while optical microphones rely on optical interference patterns, which are induced by the sound pressure.5,6

In metrology, a signal is associated with a physical quantity at a given time. This relationship can be expressed in the time domain as
{ x ( k ) } 0 k N 1 = { x ( 0 ) , x ( 1 ) , , x ( N 1 ) } ,
(1)
where x ( k ) represents the amplitude at a given time and N the number of sample points.7 The statistical analysis of signals involves descriptive statistics, which includes measures of central tendency (such as mean, median, and mode), measures of variability (range, quartile distance, mean absolute deviation, and standard deviation, for instance), and measures of correlation,
λ 1 = E ( X ) ,
(2)
λ 2 = 1 2 E ( X 2 : 2 X 1 : 2 ) ,
(3)
λ 3 = 1 3 E ( X 3 : 3 2 X 2 : 3 + X 1 : 3 ) ,
(4)
λ 4 = 1 4 E ( X 4 : 4 3 X 3 : 4 + 3 X 2 : 4 X 1 : 4 ) .
(5)

Other parameters in descriptive statistics that can be applied to characterize a distribution are the so-called moments. In this study, the skewness (Skew) γ 1 and the kurtosis (Kurt) γ 2 are considered. The former quantifies the asymmetry of a distribution around its vertex.8 The kurtosis γ 2 indicates the steepness or flatness of a distribution.8 Additionally, linear (L)-moments exist alongside conventional moments. These L-moments are expected values E of linear combinations of order statistics and serve a similar purpose as moments in descriptive statistics.9 A key advantage of L-moments is their robustness to outliers as they are less influenced by sample fluctuations than conventional moments. The first four L-moments are defined as follows, where E ( X ) is the expected value of a discrete random variable X:9 

λ 1, λ 2, λ 3, and λ 4 are the first, second, third, and fourth order L-moment, respectively, and X j : m is the jth smallest variable in a sample of size m. The second-order moment is also referred to as the L-Scale. A statistical quantity analogous to skewness or kurtosis based on L-moments is L-skewness τ 3 or L-kurtosis τ 4. These are given by the L-moment ratios,10 
τ 3 = λ 3 λ 2 ,
(6)
τ 4 = λ 4 λ 2 ,
(7)
The coefficient of L-variation L-CV represents an additional statistical parameter,11 
τ 2 = λ 2 λ 1 .
(8)

Another metric considered is the bandpower B P, which calculates the average power within a frequency band. Within this study, the multitaper technique is utilized to approximate the power spectral density required for the bandpower calculation.12 

Process monitoring is essential for ensuring the weld seam’s quality. Emissions generated during laser beam welding contain valuable information that can be accessed using suitable methods, such as the statistical signal evaluation, enabling assessment of process quality.

In a study conducted by Müller,13 bead on plate weld seams were manufactured on aluminum and steel materials using a Nd:YAG laser. The results indicated that an increase in the laser power or a decrease in the feed rate reduced the reflected laser radiation. In contrast, a changed focal position led to a higher backreflection signal since the cross section of the laser beam is enlarged and, thus, caused more direct reflection from the workpiece surface. Furthermore, the occurrence of a through weld was identified by a dip in the signal. Bardin et al. also addressed the detection of a full penetration weld using an ND:YAG laser for bead on plate welds for titanium and Inconel.14 Through a spectral analysis, the researchers identified an increase in the amplitude in the ultraviolet and infrared signals, indicating the initiation of deep penetration welding. Additionally, they observed abrupt signal drops for a partial penetration during the welding process. In addition to Müller,13 Kang et al.15 also detected a change in the focal position in overlap joints of stainless steel, whereas the focus shift was related to an increase in signal fluctuations in the plasma signal.

Apart from detecting changes in the focal position, Bono et al. demonstrated that spectral emissions can be used for detecting gaps between the joining partners during laser beam welding of a nickel alloy and a mild steel.16 By analyzing optical emissions in the range of 200–850 nm, they derived correlations between the signal fluctuations and the penetration depth. An increased local signal noise indicated surface impurities, while a decrease in the average signal and higher noise indicated the presence of a gap. Chianese et al. recorded process emissions during the laser beam welding of lap joints using the photodiode-based LWM system of Precitec GmbH & Co. KG, Gaggenau, Germany.17 They observed that the energy intensity as well as the variance of the temperature and the plasma signals increased for a higher weld depth. A gap between the joining partners was detected by an intermittent change in the temperature and plasma signal.

In addition to changes in focal position and gap detection, the localization of weld defects, such as spatter formation, can also be achieved. Gedicke et al. detected spatter during laser beam welding of copper using a multi-detector system.18 Increased sensor values for plasma and temperature radiation correlated with spatter formation, while low-intensity splashes had a minimal impact on the signal levels. In addition to spatter formation, the weld defect humping, which is manifested by an elevation of the molten pool or a seam dip at periodic intervals, can also be detected with the aid of spectral emissions. Weiss et al. detected humping weld defects using online spectral process emission measurements after laser beam welding of AISI 316L metal foils.2 Signal excursions in the three curves of the LWM system indicated the presence of the weld defect.

Besides optical emissions, the acoustic process emissions (AEs) serve as a valuable source of information pertaining to the monitoring of the laser beam welding process. In order to mitigate the influence of noise, signal preprocessing was conducted in several studies before extracting signal features.19,20 After the preprocessing of the acoustic signals, the next step is to obtain signal features from the signals. For signal analysis, it can be differentiated between the time and the frequency domain.

Luo et al. monitored acoustic emissions generated during pulsed laser beam welding of stainless steel through a resonance sensor.21 In order to characterize the plasma plume based on the AE, Luo et al. introduced the parameter AE count which represented the total number of signal pulses with a peak value exceeding a threshold of 0.5.21 Another statistical parameter investigated was the root mean square which exhibited a positive correlation with the intensity of the plasma plume ejection. The findings indicated that a higher AE count value and higher amplitudes in the power spectrum exhibited a higher energy content within the plasma plume.

Yusof et al. conducted a study to investigate the relationship between statistical signal characteristics and changes in weld geometry during pulsed laser beam welding of steel.22 The analysis included parameters such as standard deviation (SD), interdecile range (IQR), mean absolute deviation (MAD), L-CV, and L-kurtosis. The coefficient of determination was used to assess the correlation between these features and the weld geometry. The results indicated a strong correlation between the standard deviation and the L-kurtosis for the weld depth and the width, with L-kurtosis exhibiting the highest correlation. Additionally, Yusof et al. examined the behavior of statistical characteristics in the presence of weld irregularities.23 The analysis revealed that a good weld surface showed a low variance in the L-CV value, while a poor or rough surface exhibited a high variance. Similar trends were observed for the L-skewness values.

The final phase of these studies aimed to draw conclusions regarding weld quality based on the identified signal characteristics. Apart from detecting surface irregularities, the research findings also showed that pores20 and cracks in the weld joint24,25 can be detected. Several studies focused on the detection of through-welds26,27 or the classification of welds based on the presence of a through-welded joint.28 Furthermore, a significant area of investigations involved the prediction of the weld penetration depth by leveraging the identified signal characteristics.29–31 

The state of the art has shown that many process defects occurring in laser beam welding processes can be detected based on the optical or acoustic emissions. However, ensuring 100% defect detection for the two emission types is a challenge. For photodiode-based systems, the strong fluctuations of the plasma plume are challenging, as they reduce the ability to detect short-time fluctuations in the molten pool or in the vapor capillary.32,33 In contrast, systems that detect acoustic emissions using microphones are affected by environmental noise.24 In addition, the signal strength of the microphone strongly depends on the distance to the melt pool due to the frequency-dependent power dissipation. This research paper proposes a novel approach for real-time defect detection during laser welding of battery cells by analyzing the acoustic and spectral process emissions. The main objective is to explore the correlations between the acoustic and spectral signatures and the related weld defects, enabling an accurate evaluation of the weld quality.

In this study, the spectral and acoustic process emissions during laser beam welding of real batteries were investigated for a 0.2 mm thick copper ribbon and a 0.2 mm thick nickel-plated steel (Hilumin®) sample, which is a diffusion annealed steel with an electrolytic nickel coating. The samples with a width of 4 mm were positioned in an overlap configuration using a fixture to ensure a reproducible positioning (cf. Fig. 1). In order to approximate the thermal conditions observed within a battery cell during the welding process, a polypropylene high-density (PH-HD) plastic plate was placed underneath the Hilumin sample. The welding process of the lap joint was carried out using an M17LSB laser bonder from F&K Delvotec Bondtechnik GmbH. This welding system uses a TruDisk Pulse 421 disk laser from TRUMPF Laser GmbH as a laser beam source, which emits laser radiation at a wavelength of 515 nm. The laser radiation was guided to the scanning optics through a fiber-optic cable with a core diameter of 100 μm. The focusing optics embedded in the laser bonder had a magnification ratio of 2:1, which resulted in a focal diameter of 200 μm on the surface of the workpiece. The welding experiments were performed in the tape-automated bonding mode of the laser bonder, where the two joining partners were clamped using a tool to ensure a zero gap throughout the welding process. The first step in this operation mode involved the clamping of the two components to ensure a zero gap using a specially designed tool, applying force perpendicular to the component surface. Subsequently, the laser beam welding process took place within the tool. After welding, the tool was retracted, completing the process step for that joint.

FIG. 1.

Experimental setup of the laser beam welding system (M17LSB laser bonder) and the sensor systems consisting of a Laser Welding Monitor (LWM) 4.0 and an optical microphone Eta250 Ultra.

FIG. 1.

Experimental setup of the laser beam welding system (M17LSB laser bonder) and the sensor systems consisting of a Laser Welding Monitor (LWM) 4.0 and an optical microphone Eta250 Ultra.

Close modal

The sensor system installed to capture the process emissions consisted of two distinct systems. The first system used for recording spectral emissions was the Laser Welding Monitor (LWM) 4.0 manufactured by Precitec GmbH & Co. KG, which is adapted for the applied laser radiation (515 nm). This system was integrated into the optical setup of the laser bonder using a beam splitter, and it featured a sensor comprising three photodiodes for detecting optical process emissions. The temperature (T) sensor was responsible for recording temperature radiation within the wavelength range of 1100–1800 nm. The backreflection (R) sensor measured the backreflected laser radiation (515 nm). Simultaneously, the plasma (P) sensor captured plasma radiation within the wavelength range of 450–780 nm, with the green wavelength range being filtered out. The measurement signals from the photodiodes were acquired through the programmable logic controller (PLC) of the LWM system at a sampling rate of 50 kHz.

For the recording of acoustic emissions in the frequency range from 10 Hz to 1 MHz, the XARION Eta250 Ultra optical microphone from XARION Laser Acoustics GmbH was used. The analog output signal from the microphone was recorded with the HDO6000A oscilloscope from Teledyne LeCroy at a sampling rate of 25 MHz. In order to achieve a synchronized measurement of the acoustic emissions, a trigger signal was generated by the laser bonder PLC at the initiation of laser emission, serving as the starting point for the measurement process. The sensor head of the microphone was positioned parallel to the weld seam at a distance of 10 mm, a height of 10 mm, and an orientation of 45° to the working plane, using an articulated arm to ensure consistency in microphone positioning regardless of the welding position. The evaluation of weld seam quality involved visual inspection, identifying surface irregularities and spatter-induced holes, and a quantitative assessment through a shear tensile test. In order to conduct this test, samples were secured with clamping jaws at a distance of 30 mm from the weld seam. The primary measured parameter in this test was the maximum force (F) that led to the failure of the welded joint.

In this study, the laser power P = 1000 W, the welding speed v = 380 mm/s, and the weld seam length of 1.5 mm were used for the reference welding process and the investigation of failures. In order to assess the behavior of the plasma signal concerning the weld depth, weld seams were manufactured with P = 1000 W using different welding speeds.

The fundamental operation of a process monitoring system that assesses the quality of weld seams based on spectral emissions involves comparing detected emission signals to a previously recorded reference signal. In order to validate the stability and the reproducibility of the reference welding process, it was repeated 34 times, resulting in visually homogeneous weld surfaces with minimal ejections. This uniform quality holds also true for the weld depth shown in Fig. 2. In order to further confirm these observations, a shear tensile test was conducted, yielding an average shear tensile strength of 95.06 N. The relative standard deviation of the failure forces was 6.5%, which was low since there is a certain measurement uncertainty when performing the shear tensile test. These results prove the reproducibility of the welding process and, thus, can be assumed to be a reliable reference.

FIG. 2.

Signal characteristics of the plasma (P), temperature (T), and backreflection (R) sensors as well as the microphone (M) for the reference laser beam welding process (above) and cross sections of the reference parameter combination P = 1000 W and v = 380 mm/s (below).

FIG. 2.

Signal characteristics of the plasma (P), temperature (T), and backreflection (R) sensors as well as the microphone (M) for the reference laser beam welding process (above) and cross sections of the reference parameter combination P = 1000 W and v = 380 mm/s (below).

Close modal

During all repetitions of the reference welding process, the sensor system already described was used for recording the spectral and acoustic emissions. Figure 1 shows an example of the emission signals of a reference weld. The analysis of the plasma signal and the temperature signal showed an initial rise to a mean value during the start of the welding process. Subsequently, these signals had low amplitude fluctuations around this mean value. Toward the end of the welding process, the signals dropped rapidly to 0 V, whereas the plasma signal reached this value earlier. A direct comparison between the two signals showed that they are approximately synchronous in time. In particular, a decrease in the plasma signal often coincides with a reduction of the temperature signal. The start of the backreflection signal is characterized by values that exceed the value range of the sensor. After this initial phase, the signal values gradually diminished and stabilized at a constant level. In the acoustic signal, small signal fluctuations around 0 V were detected at the beginning of the welding process. At a duration of 0.0006 s, in the case of the reference shown, the amplitude of these fluctuations increased sharply for a short period of time. During the welding process, the signal fluctuations are smaller with a few exceptions.

These observed characteristics of the emission signals pertaining to the reference process can be directly linked to the phenomena taking place during the welding process, while considering established research findings: During the initial stage of welding, the copper surface is exposed to laser radiation and approximately 37% of the radiation is absorbed in the solid phase.34 Consequently, nearly two-thirds of the laser radiation is reflected by the surface, resulting in a significant signal level from the R sensor. The absorbed radiation led to heating and to a transition of the copper from the solid to the liquid state. This formation of the molten pool enhanced the amount of absorbed laser radiation so that the R sensor detected less radiation. At the same time, the signal characteristics of the T and P sensors increased due to the thermal radiation of the melt. After reaching the vaporization temperature, the vapor capillary formed in the interaction zone and the T and P sensors reached their average signal level. The signal deflection of the microphone at the beginning of the welding process was related to the abrupt transition from heat conduction to deep penetration welding. During this process, the mass flow rate from the capillary suddenly increased as a result of the evaporation process so that a rise in acoustic emissions occurred.35 This interpretation aligns with the results reported by Bastuck et al.,26 who identified higher acoustic emissions when reaching the threshold for deep penetration welding. The fluctuations of the four emission signals during deep penetration welding occur due to the capillary oscillations, which lead to a variation of the metal vapor density.35,36

The observed correlation between the signal characteristics of the P and T sensors can be attributed to the vapor plume, which emits radiation in a broad wavelength range so that the vapor plume contributed to the P and the T signal.33 The contribution of the vapor plume to the plasma signal was found to be about 50%, whereas 70% of the T signal originated from the vapor plume and capillary.32 In order to verify the correlation, a Pearson’s correlation analysis was performed for all welding trials, also considering the correlations between the plasma or the temperature signal and the backreflection signal. The analysis of average probabilities (p-value) for testing the null hypothesis indicated that all correlations were statistically significant, with p-values below the significance level of 0.05. The calculated correlation coefficients (r-value) revealed a low degree of linear correlation between the temperature and the plasma signals with the backreflection signal, yielding values of −0.56 and −0.51.37 In addition, a strong linear correlation (r-value of 0.91) was observed between the temperature and plasma signals, which mathematically confirms the dependence of the signal variations observable in Fig. 2. This r-value of 0.91 almost agrees with the correlation coefficient of 0.92 calculated in the study of Eriksson et al.33 

After recording the process emissions of the reference signal and analyzing the typical signal characteristics, welding processes with defects were considered. The weld defects were intentionally introduced by adding disturbance variables, while keeping the laser power and feed rate identical to the reference weld settings. Table I presents the investigated defects and the corresponding procedures for generating them. Figure 3 provides a summary of the signals for the examined defects, with the signal of the reference process shown in black.

FIG. 3.

Signal characteristics (p, t, r, m) in the case of the different fault cases F1–F5 compared to the reference signal (P, T, R, M).

FIG. 3.

Signal characteristics (p, t, r, m) in the case of the different fault cases F1–F5 compared to the reference signal (P, T, R, M).

Close modal
TABLE I.

Overview of the test series with the artificially introduced weld defects.

Error generation method
Defect case F1 Incorrect positioning—start of welding next to the upper joining partner 
F2 Contamination of the copper surface by GLEITMO 588M 
F3 Contamination of the copper surface by machine oil 
F4 Contamination of the Hilumin surface by GLEITMO 588M 
F5 Incorrect positioning—welding end next to the upper joining partner 
Error generation method
Defect case F1 Incorrect positioning—start of welding next to the upper joining partner 
F2 Contamination of the copper surface by GLEITMO 588M 
F3 Contamination of the copper surface by machine oil 
F4 Contamination of the Hilumin surface by GLEITMO 588M 
F5 Incorrect positioning—welding end next to the upper joining partner 

Defect case 1 in Fig. 3 resulted from a misalignment of the weld seam, where the weld start was positioned before the copper sample. This scenario, which can occur in real production environments due to improper orientation or positioning of the cell connector, is particularly critical for battery contacts. A comparison with the reference signal revealed deviations in all four signal characteristics. The plasma and temperature signals exhibited a steep increase at the beginning of the welding process, while the backreflection signal decreased faster than the reference signal. The acoustic signal displayed short, high signal amplitudes, indicating the misalignment of the weld.

In addition to this misalignment, various cases of component contamination that can occur during the manufacturing process were considered. This involved applying machine grease or universal machine oil on the copper or Hilumin surface. Defect case F2 manifested itself as a delayed start of welding, where there was no welding process activity until 0.0015 s. This delay was evident in the backreflection signal and acoustic signal, both of which consistently remained below their respective reference signals. The contamination hindered interaction between the laser radiation and the copper surface, resulting in reduced backreflected radiation and the absence of a vapor capillary formation. The low signal amplitudes in the acoustic signal are due to the lack of mass flow rate from the keyhole. At approximately 0.0015 s, the welding process initiated, showing signal behaviors similar to a defect-free weld start, although with higher values in the plasma and temperature signals, which were attributed to the evaporation of the machine grease.

In defect case F3, contamination of the copper surface with universal machine oil resulted in spatter formation and irregularities in the weld seam. While a complete weld failure was not observed, an unstable process behavior was evident, particularly between 0.002 and 0.0025 s in Fig. 3. The plasma and temperature signals deviated significantly from the reference in this time span, indicating a high signal level caused by the evaporation process of the machine oil. In contrast, no abnormalities were observed in the acoustic signal.

Another defect case, F4, involved a contamination of the Hilumin surface with machine grease. This defect led to an irregular weld seam surface and unstable process behavior, as depicted in Fig. 3. An analysis of the signal characteristics showed a slightly elevated P-signal, while the T-signal exhibited clear deviations from the reference. The backreflection signal did not exhibit significant differences, and occasional outliers were visible in the acoustic signal, but these could not be attributed to a specific defect based on visual weld assessment.

The final defect case involved welding over the copper sample, representing a misalignment of the weld seam. When the laser radiation interacted with the Hilumin sample, a rise in plasma, temperature, and sound signals was observed. However, no anomalies were observed in the backreflection signal. The amplitude of the acoustic signal occurred slightly delayed with respect to the time of the error due to the difference in the speed of light and sound.

The detection of weld seam defects can be approached through the comparison of a signal with the reference. An alternative involves process monitoring using statistical signal characteristics. The selection of these statistical quantities was based on the current state of technology and scientific knowledge. In addition to conventional measures for describing central tendency (mean, median, and mode) and variability (interquartile range, range, variance, and standard deviation), the application of L-moments was considered. These moments have been used by Yusof et al. for the detection of irregularities.22 Additionally, the AE count (AEc) as a measure developed by Luo et al. was applied.21 Originally, this measure indicated the total number of signal amplitudes exceeding a predefined threshold value of 0.5 V. However, in the present study, the AE count was used as a relative value, representing the proportion of data points surpassing a defined threshold.

In order to determine the threshold value, the absolute values of all data points were initially calculated, followed by the computation of the arithmetic mean. This mean value was then multiplied by 1.5 to establish the threshold for the AE count 1 (AEc-1) measure and by 2.0 for the AE count 2 (AEc-2) measure. These thresholds were applied to the spectral and the acoustic signals. Another metric used for signal analysis was the band power, which indicates the average power within a specific frequency band. In this study, the entire acquired frequency range of the respective signal was utilized for calculating the band power.

The statistical ratios were computed based on the raw signals for the reference signals and the weld seams with weld defects, resulting in four values for each statistical parameter due to the recording of four emission signals for each welding process. Signal features sensitive to weld defects were identified by comparing the calculated features with those of the reference welds using a boxplot diagram, considering their positions within the area covered by the reference weld features. If a feature was within this range, it indicated that the weld defect was within the variation of the defect-free reference laser beam welding process, thus demonstrating the defect undetectable based on that feature.

This evaluation methodology using boxplots was conducted for all four sensors. An overview of the signal features suitable for detecting weld defects, as determined by the described evaluation method, is presented in Table II. The suitable statistical quantities are highlighted in green.

TABLE II.

Suitability (shown in green) of selected statistical features for the detection of defect cases 1–5.

 
 

An analysis of the signal characteristics presented in Table II reveals that weld defect F5—characterized by misalignment with the weld end adjacent to the copper material—cannot be detected using the signal characteristics of the P- and T-sensor. However, the detection is possible through the L-moments, L S k e w, and L S c a l e, as well as the 75% quartile of the R-sensor. Similarly, the variance and standard deviation of the acoustic signal can also facilitate the detection of this weld defect. In contrast, weld defect F1 (corresponding to misalignment with the weld start adjacent to the ribbon material) can be detected using several features of the spectral emissions. Measures, such as skewness, kurtosis, and L-Scale of the T- and P-sensor prove effective, whereas the acoustic measurements do not enable the detection of this error. For defects F2 (contamination of the copper surface by GLEITMO® 588M) and F3 (contamination of the copper surface by machine oil), a multitude of potential features are available across all signals.

Furthermore, defect F4, involving a contamination of the Hilumin surface by GLEITMO 588M, can be detected through numerous statistical quantities derived from the T- and P-signals. However, the R-signal and the A-signal do not enable the detection of this error. The failure of detecting weld defect F4 might be attributed to the similarity between the defective welding process and the reference, resulting in a comparable rate of vapor capillary-induced sound generation. This hypothesis is supported by a failure force of 82.50 N for the defective welded joint, which closely aligns with the 25% quartile of the reference weld at 89.68 N. Consequently, the contamination has a minimum impact on the mechanical properties of the welded joint. However, the identification of the defect based on the T- and P-sensors remains possible due to the brighter plasma radiation resulting from the evaporation of the machine oil during the welding process.

The previous analysis has demonstrated a correlation between process defects and specific signal features, where the assessment of weld quality has been predominantly qualitative. Another approach involves the quantitative evaluation of weld seams based on statistical signal characteristics. In order to establish a relationship between the weld depth and statistical signal features, the welding speeds were varied from 360 to 510 mm/s in increments of 10 mm/s, while maintaining a constant laser power of 1000 W. The weld penetration depth was subsequently determined based on transverse sections. Figure 3 illustrates the average weld depths obtained from five welding repetitions for each test point. The curve indicates a nearly linear decrease in penetration depth with an increasing feed rate. After considering the weld penetration depths, the statistical variables were investigated, focusing on those that exhibited observed correlations.

In the case of the P-signal, an examination of the statistical variables in Fig. 4 revealed a positive correlation between the weld penetration depth and the mean and median characteristics. The variance also demonstrates a positive correlation with the weld depth. Similarly, the mean, median, and variance statistical quantities showed a positive correlation for the T-sensor. Notably, irregularities are evident in all curves at a speed of 440 mm/s for all characteristics. These anomalies could potentially be attributed to sources of interference, such as slight contaminations of the copper sample during the experimental series.

FIG. 4.

Influence of the welding speed on the weld depth and on the statistical features of the plasma signal with its minimum and maximum values for each measurement.

FIG. 4.

Influence of the welding speed on the weld depth and on the statistical features of the plasma signal with its minimum and maximum values for each measurement.

Close modal

An analysis of the statistical quantities for the R-signal reveals a multitude of measures that are associated with the weld penetration depth. Positive correlations were observed for the S k e w, L S k e w, K u r t, and L K u r t characteristics. In contrast, the statistical measures of m e a n, m e d i a n, S D, M A D, I Q R, L S c a l e, and B P displayed negative correlations with the weld penetration depth.

An examination of the acoustic signal characteristics depicted in Fig. 4 revealed a positive correlation between the statistical quantities B P, L S c a l e, and I Q R, and the weld penetration depth. Conversely, the behavior of the L-moments, L K u r t, and L C V exhibited a negative correlation. Therefore, a decrease in weld penetration depth corresponds to an increase in the signal characteristics. These correlations align with the findings demonstrated by Yusof et al.22 and are further supported by the results of their study.

This study aimed to investigate the potential for detecting and identifying weld defects in the battery cell contacting process using spectral and acoustic process emissions. A reference welding process was established, which demonstrated stability and reproducibility through repeated tests and shear pull tests, confirming its reliability. The emission signals of the reference weld seams exhibited distinct characteristics associated with different phases of the welding process. Subsequently, welding processes with intentionally introduced defects, including misplacement and surface contamination, were analyzed. The signals of these defective welding processes showed variations in signal characteristics compared to the reference weld seams. Furthermore, the study explored the use of statistical signal characteristics for weld defect detection. Various statistical measures, including Hosking's L-moments, were applied for this purpose. By comparing these measures between reference and defective weld seams using boxplots, appropriate statistical quantities that correlated with the presence of weld defects were identified. Moreover, the study examined the quantitative evaluation of weld quality by determining the average weld penetration depth and exploring its relationship with statistical signal characteristics. Positive correlations were observed between the specific statistical variables and the weld penetration depth. In conclusion, this study demonstrated the effectiveness of analyzing statistical measures for the detection and evaluation of weld seam defects. The findings contribute to the understanding of signal behaviors during the welding process and provide valuable insights for process monitoring and quality control in welding applications.

Further investigations can explore whether the approach using statistical features can be transferred to other applications. It should also be investigated how a division of the weld seam in different sections can enhance the detection of weld failures.

The authors gratefully acknowledge the financial support by the German Federal Ministry for Economic Affairs and Climate Action (BMWK) as part of the research project FlaMe (Grant No. 03EN4008A-G) and the research project ultraBatt (Grant No. 01MV21015D).

The authors have no conflicts to disclose.

Johannes Heilmeier: Conceptualization (lead); Formal analysis (lead); Investigation (lead); Methodology (lead); Software (lead); Validation (lead); Visualization (lead); Writing – original draft (lead); Writing – review & editing (equal). Michael K. Kick: Conceptualization (supporting); Formal analysis (supporting); Methodology (supporting); Writing – review & editing (equal). Sophie Grabmann: Writing – review & editing (equal). Tatek Muschol: Investigation (supporting); Methodology (supporting). Franz Schlicht: Methodology (supporting); Writing – review & editing (equal). Felix von Hundelshausen: Methodology (equal). Hans-Georg von Ribbeck: Methodology (supporting); Writing – review & editing (equal). Tony Weiss: Writing – review & editing (equal). Michael F. Zaeh: Funding acquisition (lead); Supervision (lead); Writing – review & editing (equal).

1.
N.
Wassiliadis
,
M.
Ank
,
L.
Wildfeuer
,
M. K.
Kick
et al, “
Experimental investigation of the influence of electrical contact resistance on lithium-ion battery testing for fast-charge applications
,”
Appl. Energy
295
,
117064
(
2021
).
2.
Tony Weiß, Michael Kick, Sophie Grabmann, Christian Geiger, Lukas Mayr, Katrin Wudy, and Michael Zaeh
, “
A holistic approach for an intelligent laser beam welding architecture using machine learning for the welding of metallic bipolar plate for polymer electrolyte membrane fuel cells
, ”
Procedia CIRP
111, 810–815.
3.
J.
Stavridis
,
A.
Papacharalampopoulos
, and
P.
Stavropoulos
, “
Quality assessment in laser welding: A critical review
,”
Int. J. Adv. Manuf. Technol.
94
, 1825–1847 (
2018
).
4.
H.
Krauss
,
Qualitätssicherung Beim Laserstrahlschmelzen Durch Schichtweise Thermografische In-Process-Überwachung
(
Herbert Utz Verlag GmbH
,
München
,
2016
).
5.
B.
Fischer
,
N.
Panzer
, and
W.
Rohringer
,
Optisches Mikrofon: Neue Ohren für Kontaktfreie Industrielle Prüfverfahren
(Magazin Verlag Hightech Publications KG, Bad Nenndorf,
2017
).
6.
H.
Bernstein
,
Elektroakustik: Mikrofone, Klangstufen, Verstärker, Filterschaltungen und Lautsprecher
, 2nd edn. (
Springer Fachmedien
,
Wiesbaden
,
2019
).
7.
J.
Lange
and
T.
Lange
,
Fourier-Transformation Zur Signal- und Systembeschreibung: Kompakt, Visuell, Intuitiv Verständlich
(
Vieweg
,
Wiesbaden
,
2019
).
8.
D.
Selvamuthu
and
D.
Das
,
Introduction to Statistical Methods, Design of Experiments and Statistical Quality Control
(
Springer
,
Singapore
,
2018
).
9.
J. R. M.
Hosking
, “
L-moments: Analysis and estimation of distributions using linear combinations of order statistics
,”
J. R. Stat. Soc.: Ser. B (Methodol.)
52
,
105–124
(
1990
).
10.
J. R. M.
Hosking
, “
Moments or L moments? An example comparing two measures of distributional shape
,”
Am. Stat.
46
,
186
–189 (
1992
).
11.
Q. J.
Wang
, “
Direct sample estimators of L moments
,”
Water Resour. Res.
32
,
3617
–3619 (
1996
).
12.
R.
Vallat
, see https://raphaelvallat.com/bandpower.html for “Bandpower of an EEG Signal.”
13.
M. G.
Müller
,
Prozessüberwachung Beim Laserstrahlschweißen Durch Auswertung der Reflektierten Leistung
(
Herbert Utz Verlag GmbH
,
München
,
2002
).
14.
F.
Bardin
,
A.
Cobo
,
J. M.
Lopez-Higuera
,
O:
Collin
et al, “
Optical techniques for real-time penetration monitoring for laser welding
,”
Appl. Opt.
44
,
3869
(
2005
).
15.
H.
Kang
,
J.
Suh
, and
T.-D.
Cho
, “
Study on quality monitoring of laser welding
,” in
2009 IEEE International Symposium on Industrial Electronics, Seoul, South Korea, 5–8 July 2009
(IEEE, New York, 2009), pp. 1679–1682.
16.
P.
Bono
,
C.
Allen
,
G.
D'Angelo
, and
A.
Cisi
, “
Investigation of optical sensor approaches for real-time monitoring during fibre laser welding
,”
J. Laser Appl.
29
,
022417
(
2017
).
17.
G.
Chianese
,
P.
Franciosa
,
J.
Nolte
,
D.
Ceglarek
, and
S
Patalano
, “
Characterization of photodiodes for detection of variations in part-to-part gap and weld penetration depth during remote laser welding of copper-to-steel battery tab connectors
,”
J. Manuf. Sci. Eng
. 144, 071004 (2022).
18.
J.
Gedicke
,
B.
Regaard
,
K.
Klages
,
A.
Olowinsky
, and
S.
Kaierle
Comparison of different process monitoring methods for laser beam micro welding
,” in
International Congress on Applications of Lasers & Electro-Optics (ICALEO®)
, Scottsdale, AZ, 30 October-2 November 2006 (
Laser Institute of America
, Orlando, FL,
2006
),
262
269
.
19.
W.
Huang
and
R.
Kovacevic
, “Noise reduction during acoustic monitoring of laser welding of high-strength steels,” in
Proceedings of the ASME 2009 International Manufacturing Science and Engineering Conference, West Lafayette, IN, 4–7 October 2009, Volume 1
(ASME, New York, 2009), pp. 763–770.
20.
S.
Shevchik
,
Q.
T. Le
,
B.
Meylan
, and
K.
Wasmer
, “Acoustic emission for in situ monitoring of laser processing,” in
Conference Proceedings: EWGAE 2018. 33rd European Conference on Acoustic Emission Testing, Senlis, France, 12–14 September 2018
(
CETIM, Senlis
,
France
,
2018
).
21.
Y.
Luo
,
L.
Zhu
,
J.
Han
,
X.
Xie
,
R.
Wan
,
Y.
Zhu
et al, “
Study on the acoustic emission effect of plasma plume in pulsed laser welding
,”
Mech. Syst. Signal Proc.
124
,
715
–723 (
2019
).
22.
M. F. M.
Yusof
,
M.
Ishak
, and
M. F.
Ghazali
, “
L-statistical analysis of sound signal acquired from pulse mode laser welding for characterising weld geometry
,”
Int. J. Automot. Mech. Eng.
16
,
6987–7006
(
2019
).
23.
M. F. M.
Yusof
,
M.
Ishak
, and
M. F.
Ghazali
, “
Detection of irregularities on weld bead from the L-statistic analysis of the acquired sound during pulse mode laser welding process
,”
IOP Conf. Ser.: Mater. Sci. Eng.
788
,
012015
(
2020
).
24.
B.
Fischer
,
W.
Rohringer
,
N.
Panzer
, and
S.
Hecker
, “
Acoustic process control for laser material processing
,”
Laser Tech. J.
14
,
21
–25 (
2017
).
25.
C.
Prieto
,
R.
Fernandez
,
C.
Gonzalez
,
J. Arias, R. Sommerhuber, and F. Lücking
(
2020
). “
In situ process monitoring by optical microphone for crack detection in laser metal deposition applications
,”
Zenodo
. https://doi.org/10.5281/zenodo.4043548.
26.
M.
Bastuck
,
B.
Valeske
,
B.
Wolter
, and
H.-G.
Herrmann
, “
Überwachung und Qualitätssicherung des Laserschweißprozesses ultrahoch-fester Karosseriestähle durch integrierte ZfP-Verfahren
,” in DGZfP Jahrestagung 2013, Dresden, Germany, 6–8 May 2013, https://www.ndt.net/?id=15314
27.
Y.-M.
Zhang
, “
An analysis of welding process monitoring and control
,” in
Real-Time Weld Process Monitoring
, edited by Y.-M. Zhang (
Woodhead Publishing
,
Cambridge, UK
,
2008
), pp.
1
11
.
28.
M.
Yusof
,
M.
Ishak
, and
M. F.
Ghazali
, “
Classification of weld penetration condition through synchrosqueezed-wavelet analysis of sound signal acquired from pulse mode laser welding process
,”
J. Mater. Process. Technol
279
,
116559
(
2020
).
29.
L.
Tomcic
,
A.
Ederer
,
S.
Grabmann
,
M. Kick, J. Kriegler, and M. F. Zaeh
, “
Interpreting acoustic emissions to determine the weld depth during laser beam welding
,”
J. Laser Appl.
34, 042052 (2022).
30.
N.
Authier
,
E.
Touzet
,
F.
Lücking
,
R.
Sommerhuber
,
V.
Bruyere
, and
P.
Namy
, “
Coupled membrane free optical microphone and optical coherence tomography keyhole measurements to setup welding laser parameters
,”
SPIE Proc.
11273
, 1127308 (
2020
).
31.
M. F. M.
Yusof
,
M.
Ishak
and
M. F.
Ghazali
, “
Acoustic methods in real-time welding process monitoring: Application and future potential advancement
,”
JMES
15
,
8490–8507
(
2021
).
32.
P.
Norman
,
I.
Eriksson
, and
A.
Kaplan
, “Basic study of photodiode signals from laser welding emissions” (unpublished).
33.
I.
Eriksson
,
J.
Powell
, and
A. F. H.
Kaplan
, “
Signal overlap in the monitoring of laser welding
, ”
Meas. Sci. Technol
. 21, 105705 (
2010
).
34.
H.
Hügel
and
T.
Graf
,
Laser in der Fertigung Strahlquellen, Systeme, Fertigungsverfahren
, 2nd ed. (
Vieweg + Teubner
,
Wiesbaden
,
2009
).
35.
G.
Herziger
,
H.
Weber
, and
E.
Beyer
,
Schweißen mit Laser: Grundlagen
(
Springer
,
Berlin
,
1995
).
36.
P.
Berger
,
H.
Hügel
,
A.
Hess
,
R.
Weber
, and
T.
Graf
, “
Understanding of humping based on conservation of volume flow
,”
Phys. Procedia
12
,
232
–240 (
2011
).
37.
K.
Siebertz
,
D.
van Bebber
, and
T.
Hochkirchen
,
Statistische Versuchsplanung
(
Springer
,
Berlin
,
2017
).