Comparison of Different Calibration Techniques by Laser Induced Breakdown Spectroscopy in Bakery Products

28 Laser induced breakdown spectroscopy (LIBS) is a rapid optical spectroscopy technique for 29 elemental determination, which has been used for quantitative analysis in many fields. 30 However, the calibration involving atomic emission intensity and sample concentration, is still 31 a challenge due to physical-chemical matrix effect of samples and fluctuations of experimental 32 parameters. To overcome these problems, various chemometric data analysis techniques have 33 been combined with LIBS technique. In this study, LIBS was used to show its potential as a 34 routine analysis for Na measurements in bakery products. A series of standard bread samples 35 containing various concentrations of NaCl (0.025% – 3.5%) was prepared to compare different 36 calibration techniques. Standard calibration curve (SCC), artificial neural network (ANN) and 37 partial least square (PLS) techniques were used as calibration strategies. Among them, PLS was 38 found to be more efficient for predicting the Na concentrations in bakery products with an 39 increase in coefficient of determination value from 0.961 to 0.999 for standard bread samples 40 and from 0.788 to 0.943 for commercial products. 41


46
Laser induced breakdown spectroscopy (LIBS) is an atomic emission spectroscopy 47 technique in which laser beam excites and intensively heats the surface of sample. Excited 48 sample is taken to a gaseous plasma state and dissociated to all molecules and fine particles, 49 which produces a characteristic plasma light. Intensity of this plasma light is associated with 50 concentration of the elements in the sample. LIBS has many advantages as it allows for rapid, 51 real-time and in situ field analysis without the need for sample preparation [1-10]. Moreover, 52 its application has expanded to the fields such as metallurgy, mining, environmental analysis 53 and pharmacology [11][12][13][14]. 54 Intensity of LIBS signal is influenced by various factors including laser energy, detection 55 time window, distance between lens and chemical and physical matrix [15]. Chemical matrix 56 effect is the most important one since the molecular and chemical composition of the sample is 57 directly related with chemical matrix, and it perturbs the LIBS plasma [16]. Minor elements in 58 the sample structure can cause matrix effects and interactions on the major element spectral 59 lines. Furthermore, LIBS signal intensity is influenced by atmospheric composition, and 60 occurred plasma products are interacted with sample surface. To overcome matrix effect, many 61 approaches have been developed. Traditionally, spectral peak intensity or peak area is analyzed 62 through LIBS data versus concentration of samples for quantitative analyses, which is the 63 standard calibration curve method (SCC) [17]. Chemometric techniques are being used more 64 widely in order to enhance analytical performance of LIBS [18]. Recent works have shown that 65 multivariate analysis such as partial least square (PLS) and artificial neural network (ANN) 66 give promising results for quantitative analysis [19][20][21]. These advanced techniques reduce the 67 complexity of spectra and enable valuable information. In LIBS analysis, many fluctuating 68 experimental parameters decrease the relation between elemental composition and LIBS [27]. Thus, sodium levels in food should be controlled. In a human diet, 70-75% of the total 81 sodium chloride (NaCl) intake is obtained from processed foods, out of which cereal and cereal 82 products constitute approximately 30% [28]. Therefore, NaCl content in bread, the most 83 consumed food all over the world, should be lowered and adhered to Codex Alimentarus. Na 84 content can be controlled by using standard methods such as flame atomic absorption 85 spectrometry (AAS), titration and potentiometry [29][30]. These methods are time consuming and complex due to their sample preparation process and their inconvenience for in situ and 87 point detection analyses. Therefore, new, rapid and practical techniques are required.

88
Quantitative analysis of bakery products has been analyzed by standard calibration curve   Laser repetition rate is adjustable in the range of 1-100 Hz, but the experiment is performed at 104 nm Na line was detected by gating the spectrometer 0.5 µs after the laser pulse and with a 20 112 µs gate width. All measurements were performed under ambient conditions and exposed to

Na detection in bakery products by atomic absorption spectroscopy
134 Na content of standard bread samples and commercial samples were analyzed by atomic 136 absorption spectroscopy (reference method for Na measurements). Samples were prepared 137 based on the EPA Method 3051A through microwave-assisted digestion for atomic absorption 138 spectroscopy measurements [33]. At the beginning, 0.3 g of the dried sample and 10 ml In addition, we used RSD as a prediction precision indicator. includes 20 spectra) for calibration, 13 data for prediction of SCC method were used. Following 175 that, LIBS spectra of commercial products were analyzed via SCC method, and the results were 176 compared with measured Na concentrations by AAS.

177
In this study we used two different multivariate analysis methods. One of them is PLS in 178 which we used the same data set as in previous work. Data of LIBS spectra ranged between 179 538.424 nm and 800.881 nm were used instead of whole spectrum because the most quantitative 180 data could be obtained from this region. LIBS data matrix was obtained by analyzing the spectra 181 of 39 standard bread samples (26 samples for calibration, 13 samples for the validation) for Software (Version Solo 6.5 for Windows 7, Eigenvector Research Inc., Wenatchee, WA, USA).

184
Data matrix of selected LIBS data and concentration was embedded into the software as 185 calibration data, and PLS algorithm was performed using different components between 1-15. 186 Mean center was applied as pre-processing to calibration input data. Prediction ability of  For the calibration study, a total of 780 spectra (20 spectra for each pellet, 3 pellet for each 222 sample, 13 standard bread sample) and for commercial products a total of 360 spectra (20 223 spectra for each pellet, 3 pellet for each sample, 13 standard bread sample) were recorded by 224 LIBS. Fig.2. illustrates the LIBS spectra of different amount of NaCl containing standard bread 225 samples. The peak at 588.599 nm belongs to Na and peak at 769 nm belongs to K according to 226 NIST atomic data base [35]. Fig.2 shows that as the intensity of Na band at 588.599 nm 227 increases, the NaCl levels in breads rises, as well.

228
Calibration models were developed according to three different methods, which are SCC, samples). Then, prediction ability of the regression of obtained model was evaluated via 231 validation data set (standard bread samples, excluding the calibration data set, were treated as 232 unknown) to test the accuracy and precision of calibration model. After that, Na concentrations 233 of commercial products were predicted and compared with results of standard method AAS.

234
This treatment is found to be useful for evaluating the matrix effect and the potential of LIBS 235 in commercial samples.

236
The traditional way to obtain calibration curve is using reference samples which contain  (Fig.3(c)). RSD and REP values of PLS is presented in Table1.

252
In addition to the PLS method, ANN was also used for Na quantification in standard bread 253 samples. Same calibration and validation data set were used for ANN model. The networks that 254 had maximum R 2 values between predicted and actual data were selected as the best trained network. Then, the best-trained network was used for prediction of Na content in standard 256 breads with ANN. The predicted calibration and validation data sets were compared with 257 experimental data sets and high correlations were obtained for Na concentrations (Fig.3 (b)). in LIBS applications [40]. Availability of data and materials: There is no data that needs to be shared.    Figure 1 Schematic presentation of LIBS experimental setup.   Correlation between AAS and LIBS method for commercial products with SCC (a), ANN (b), PLS (c) data analysis techniques.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. Tables.docx