Laser‐Induced Breakdown Spectroscopy Online Quantitative Analysis for Laser Processing

Plasma fluctuations, the uncertainty of laser ablation, and bremsstrahlung limit the identification of online element analysis during laser processing and cause difficulty in achieving concentration results with sufficient accuracy and repeatability. A laser‐induced breakdown spectroscopy (LIBS) online monitoring system with plasma spatial filtering and spectral screening is proposed to solve this problem. In this system, the high‐frequency ablation noise component of the plasma is eliminated using a specially designed optical Fourier filtering structure, and a spectral screening system based on plasma time waveform monitoring is used to suppress the influence of plasma fluctuations. Without noise filters or algorithm optimizations and based only on the basic internal standard method, the calibration curves of all nine elements in the alloy sample exhibits a Pearson's R2 value ranged from 0.91 to 0.99, with a mean of 0.94. The relative standard deviations are all in the range of 3.5%–8.4% with a mean of 5.4%. The accuracy and repeatability are comparable to those of typical LIBS systems.


Introduction
Plasmas produced by laser processing contain valuable information.[6][7] Plasmas in laser processing undergo severe fluctuations, making quantitative analysis difficult. [8]15] The microscopic mechanism of the interaction between the laser and material determines the difference between shortand long-pulse laser ablation.For metals, short-pulse laser ablation occurs primarily through the avalanche and multiphoton ionization mechanisms, instantly producing several high-temperature free electrons.Under the action of bremsstrahlung radiation, a large amount of laser energy is absorbed, causing a rapid increase in temperature, and the lattice electrons are removed to produce plasma. [16][19] More importantly, in a typical LIBS system, spectra are obtained with a gate delay to avoid plasma fluctuations and bremsstrahlung.For laser processing monitoring, the spectrum acquisition process should be performed simultaneously with the laser ablation and bremsstrahlung processes.Laser ablation and bremsstrahlung cause the plasma state parameters to fluctuate violently and further overlap with the uncertainty factors of the random ablation surface, sample composition, and environment, making the error of the quantitative analysis unacceptable.

Experimental Section
A schematic of the LIBS system used for online laser processing monitoring is shown in Figure 1.A laser-diode-pumped freerunning laser was used as the laser processing source, and the output energies were 1.74, 3.20, 5.03, and 6.59 J at working currents of 100, 150, 200, and 250 A, respectively.The pulse width was 2.2 ms, and the working frequency was 1 Hz.The total pumping power of the laser diode arrays was 36.0 kW at a working current of 250 A, and the laser working medium was a 15 mm-diameter Nd:YAG rod.The length of the resonant cavity was 280 mm, and the reflectance of the output mirror was 80% at 1064 nm.The power supply of the free-running laser comprised DOI: 10.1002/adpr.202300293Plasma fluctuations, the uncertainty of laser ablation, and bremsstrahlung limit the identification of online element analysis during laser processing and cause difficulty in achieving concentration results with sufficient accuracy and repeatability.A laser-induced breakdown spectroscopy (LIBS) online monitoring system with plasma spatial filtering and spectral screening is proposed to solve this problem.In this system, the high-frequency ablation noise component of the plasma is eliminated using a specially designed optical Fourier filtering structure, and a spectral screening system based on plasma time waveform monitoring is used to suppress the influence of plasma fluctuations.Without noise filters or algorithm optimizations and based only on the basic internal standard method, the calibration curves of all nine elements in the alloy sample exhibits a Pearson's R 2 value ranged from 0.91 to 0.99, with a mean of 0.94.The relative standard deviations are all in the range of 3.5%-8.4% with a mean of 5.4%.The accuracy and repeatability are comparable to those of typical LIBS systems.a 15-kW DC power source (GEN3U, TDKlambda) and pulse modulator (eDriver, Northrop Grumman).The pulse modulator changes the continuous DC output into a pulse output and sets the pulse width and working current.The laser beam ablated the sample using a lens with a focal length of 500 mm.The dichroic mirror reflected the signal light of the plasma, which was converged by a 90°off-axis parabolic mirror A (focal length 101.6 mm, diameter: 50 mm, MPD249-F01, Thorlabs) to a parallel beam.The optical Fourier filtering structure comprised two 90°off-axis parabolic mirrors B and C (the same as A) and a 1.0 mm-diameter aperture.A Fourier spatial filter structure was used to filter out high-frequency noise in the spectral signals.Subsequently, the signal was split using a splitter.One part of the photodetector with a filter (high reflection to 1064 nm ablation laser) and the other part was converged by a parabolic mirror D (the same as A) into the fiber collimator for the spectrometer.The temporal pulse shape was obtained using an Si-biased photodetector (DET025A, Thorlabs) and oscilloscope (DPO4104, Tektronix).The parabolic mirror D directly converged the plasma light into the fiber collimator.The spectrum was collected by a collimator coupled to a four-channel spectrometer (Avs-Desktop-USB2, Avantes); the spectrometer had a spectral range of 190-557 nm and resolution of 0.06 nm at 190-310 nm and 0.04 nm at 308-557 nm.
In this experimental setup the coaxial excitation and mirror collection schemes are advantageous mainly for spatial filtering.If a separate optical path scheme is used for laser excitation and spectrum acquisition, aberrations increase the complexity of the optical structure.The experiments showed that the collection method shown in Figure 1 could better transfer the image of the ablated area to the surface of the aperture, and the all-mirror optical structure could effectively avoid chromatic aberration of the broadband spectrum signals.This was mainly because the Fourier spatial filter structure was used to eliminate the jitter of the high-frequency components of the plasma's spatial distribution; however, chromatic aberration could lead to broadband spectrum transmission efficiency fluctuations, which was considered a fluctuating factor that should be avoided.
If the spectral signal acquisition structure is regarded as an imaging system for the plasma, this aperture can be regarded as a field stop.The plasma light emitted from points that deviate from the field of vision is not ideally focused at the center of the aperture.The aperture determines the spatial range of the plasma light that can be collected, and the aperture diameter determines the size of the field of vision, that is, the collection area and intensity of the plasma light.If the aperture is sufficiently small, the light emitted by the plasma at the edge of the ablation zone will not be collected, which means that the detection of plasma signals at specific spatial positions can be achieved.The aperture diameter in our experimental system was 1.0 mm.The design of a more precise structure or selection of a higher-peak-power light source can significantly reduce the aperture diameter.This scheme can detect regions with more consistent states in the plasma, which is of significant importance for suppressing plasma fluctuations.

Data Screening
The uncertainty of laser ablation can cause fluctuations in the plasma state parameters; however, data screening is an effective method to solve this problem.In the experiments, the time-and frequency-domain signals from the plasma were collected individually.The instantaneous intensity of the plasma time-domain signals is closely related to the status parameters such as the plasma emission rate and temperature.Selecting plasma signals with similar instantaneous intensities for analysis can ensure consistent plasma states, which reduces inherent errors.Therefore, the validity of the spectral data was determined based on the strength and stability of the plasma time-domain signal waveform.Specifically, the area of the time-domain signal over the spectrometer integration period, which is the sum of the intensities of the oscilloscope time waveform at each sample point during the spectrometer integration time, was calculated, and data that were too strong or too weak were filtered out.Simultaneously, the area intensity of each spectrum was calculated, and a certain intensity fluctuation range was set for screening.Additionally, if the temperature of the plasma can be accurately calculated each time it is excited, it can be introduced for screening.
The spectral intensity screening factor k 1 is defined as the proportion of data retention based on the spectral signal intensity.When k 1 = 70%, 15% of the strongest and 15% of the weakest signal strengths of all the spectral data were excluded.The time-stability filter factor k 2 is defined as the relative standard deviation (RSD) of the intensity of each sampling point of the time-series waveform collected by the oscilloscope over the spectrometer integration time range.When k 2 = 15%, the corresponding spectral data were filtered when the RSD was higher than 15%.For comparison, the spectral data in our experiments were roughly screened using the above k 1 and k 2 values, and the values of k 1 and k 2 were reduced to 50% and 10%, respectively, for fine screening.
The influence of screening on the quantitative analysis results is also provided for comparison.The original, rough-screened, and fine-screened spectral data were limited to a single-line internal standard and used the same background correction, normalization, integration, and curve fitting as in the overlapping spectra method; their average R 2 values for calibration were 0.83, 0.87, and 0.91, respectively.Simultaneously, the RSDs were 15.4%, 11.9%, and 6.9% for the original, rough-screened, and fine-screened samples, respectively.Figure 2 shows the internal standard curves of some elements under the three screening conditions.We selected the univariate calibration curve method under the same data processing and parameters to compare the screening effects, as this method could more reasonably reflect the screening value for quantitative analysis.
It should be noted that the spectrum treatment followed the following steps: 1) background correction; 2) spectrum peak searching; 3) spectrum fitting; 4) certified compositions of all the standard samples were obtained; 5) calibration curve construction; 6) selection of the inner relative standard and calibration wavelengths; and 7) linear fitting.The background was removed with reference to the baseline, a spectral line was selected, and a calibration curve based on the peak area was established.

Results and Discussion
We constructed a plasma spatial filtering structure that could form and maintain plasma at a steady state during the ablation period.Combined with our proposed data screening method based on the plasma intensity and time-domain waveform, the problem of sharp plasma fluctuations caused by long-pulse ablation could be significantly improved.Figure 3 shows the calibration curves and RSD of the Ni-based alloy sample elements.Without a noise filter or algorithm optimization and based on the basic internal standard method, the Pearson's R 2 value for the calibration curve of all nine elements in an alloy sample ranged from 0.91 to 0.99, with an average of 0.94.The RSDs of all nine elements in the alloy sample were in the range of 3.5%-8.4%,with an average of 5.4%.The accuracy of this quantitative analysis was comparable to that of nanosecond-or picosecond-laser LIBS.To the best of our knowledge, this is the first study to solve the problem of severe fluctuations in plasma signals during online analysis of laser processing, and the accuracy and repeatability can reach the level of LIBS induced by short-pulse lasers.This method of decreasing the fluctuation of plasma signals is also valuable for improving the quantitative accuracy of LIBS analyses.
The use of a chemometric algorithm is a valid method for improving the accuracy and precision of quantitative analysis of LIBS.In this study, we established the most common calibration curve for comparison with traditional LIBS technology.Based on the quantitative analysis capabilities of traditional LIBS technology using similar data processing, our results have relatively good accuracy and repeatability. [7,17,20]However, if algorithms such as the noise suppression methods of a Fourier transform filter [20] and wavelet threshold, [21,22] calibration methods of machine learning, [23] strategy optimization, [24] wavelet transform, [25] multienergy calibration, [26] and iterative multi-energy calibration [27] are used, the precision can be improved further.
In terms of the operating current modulation scheme, if the power supply system can achieve the modulation output of an arbitrary pulse waveform, the plasma can be controlled more stably.The use of the above algorithms can significantly improve the accuracy and repeatability of quantitative analysis not only in the processing of spectral data, but also in controlling, filtering, and screening time-waveform signals, especially when the number of samples and spectra is sufficiently large.

Conclusion
A LIBS methodology was proposed to solve the problem of large errors in the quantitative analysis of online plasma monitoring.The experimental results showed that the plasma spatial filtering and spectral screening method had good feasibility and could effectively improve the precision of the quantitative analysis.

Figure 1 .
Figure 1.Experimental scheme and data screening method of long-pulse LIBS for laser processing.

Figure 2 .
Figure 2. Comparison of the single-spectrum internal standard results under three screening conditions.a) Internal standard results and RSDs without spectral data screening.b) Rough-screened spectral data internal standard results and RSDs: intensity filter ratio factor k 1 = 70% and RSD filter ratio factor k 2 = 15%.c) Finely screened spectral data, internal standard results, and RSDs: intensity filter ratio factor k 1 = 50% and RSD filter ratio factor k 2 = 10%.