ES2356879B1 - INSTANT IDENTIFICATION AND CHARACTERIZATION OF SAMPLES THROUGH LASER ABLATION DYNAMIC COMBINATION AND MATHEMATICAL ALGORITHMS. - Google Patents

Abstract

Instantaneous identification and characterization of samples without prior treatment by means of the dynamic combination of laser ablation and mathematical algorithms. # The method is based on the analysis of a single laser pulse from which a sample spectrum is obtained that is compared with a database. Dynamic spectral using a neural network that allows to handle a large number of data in a very short time with 100% sensitivity and specificity. # The identification and characterization technique is fast, direct and allows the correlation of analytical data with organoleptic properties or of quality as well as the determination of substances, contaminating compounds, pathogens or bioactives present in samples in any state of the matter without any previous preparation.

Description

Instantaneous identification and characterization of samples through the dynamic combination of laser ablation and mathematical algorithms.

Object and Field of the Invention

This invention is part of the field of spectral analysis of chemical compounds present in liquid, solid or gaseous samples. More specifically, the invention refers to a method of identification and characterization by qualitative determination of chemical compounds in samples, without any prior treatment and that allows correlation with organoleptic or quality properties with application in various fields such as in water samples, food or biological samples. The method combines laser spectroscopic technique and spectral treatment based on mathematical algorithms.

This type of algorithm provides a suitable method with 100% sensitivity and 100% specificity in the discrimination of compounds, substances or materials that allow their identification and correlation with organoleptic or quality properties, food traceability or the detection of contaminants or pathogens. Of different nature.

This invention allows the correlation of analytical data with organoleptic or quality properties and / or the determination of substances, compounds, pollutants or pathogens, in water samples (rainwater, underground, natural fl uents, sewage, industrial waste), food (fruit juices, vegetable oils, milk and its derivatives), and biological samples (blood, serum, urine, etc.), solid samples (steels, alloys, wood, bones, rocks, minerals, etc.).

State of the art

The LIBS (Laser Induced Breakdown Spectroscopy) technique is a fast and versatile method for analyzing different types of samples that may be inaccessible or tedious by conventional analytical techniques, being particularly useful in the analysis of samples with a complex matrix, ( JO Caceres et al. Spectrochimica Acta B: Atomic Spectroscopy 56 (6) 831 (2001), WB Lee, et al, Appl Spectrosc. Rev. 39 (1) 27 (2004)).

However, this technique has been mainly applicable to elementary analysis and in many situations there is a considerable number of alternative techniques available with a sensitivity greater than LIBS. The decision on the choice of a method for elementary analysis can be made favorable to the LIBS technique due to some of its characteristics that give it a considerable advantage over conventional techniques such as a) the elimination of the need to prepare the sample for analysis , b) the analysis can be carried out in any of the states of matter (solid, liquid, gas), c) the analysis is carried out in a few seconds, d) only a very small amount of the sample, in the order of micrograms, is vaporized from the surface of the sample and e) the detection limits are found for most samples in the order of a few parts per million.

The great importance of this technique for quantitative spectrum chemical analysis is evident. However, surprisingly, much less attention has been placed on the potential of the LIBS technique as a qualitative analysis tool. In this case, a detailed chemical composition is not sought, rather an instantaneous identification of the sample using a unique feature of LIBS, which is capable of generating a "fingerprint" of the sample. This “fingerprint” is a LIBS spectrum of the sample, which comes exclusively from it and therefore depends on its matrix. In addition, due to the nature of the emission spectrum, which is dominated by ionic lines, which are often inhibited in their direct relationship between the elementary concentration in the sample and its intensities, it provides a unique spectrum and only belonging to the sample. under analysis

Using a correlation procedure, the LIBS system can be trained to recognize spectra from different samples, which means assessing the similarity of unknown spectra with a spectral bank of classified samples. It is necessary, therefore, to know which correlation technique is most suitable for the identification of samples. Your choice is usually determined by the experimental arrangement, the type of data obtained and the time requirements for analysis.

In recent years, different methodologies have been used to identify or classify compounds, in a progression of complexity ranging from the relatively simple non-parametric linear correlation procedures (IB Gornushkin et al. Chem. 71, 5157-5164, (1999), AJ Lopez et al Appl. Phys. A 83, 695698 (2006)), Fourier analysis (WH Flannery et al, Numericla recipes, the Art of Scienti fi c Computing Cambridge University Press NY (1986)), the method known as Mahalanobis distance (O. Sameck et al. LNCS, 2124,443 (2001)), up to certain types of neural networks from their first applications to detection and classification in the 1990s (I. Ruisanchez et al, J. Chem. Inf. Comput. Sci 36, 214 (1996), R. Sattmann et al Appl. Spect. 52 456 (1998)). Neural networks provide the best results, however despite their dissemination and application to various fields of research such as minerals and artistic paintings (M: M. Howari, J. Appl. Spect. 70, 782 (2003), D Anglos Appl. Spect. 51, 258A (1997)), have strong limitations and have not been able to provide a satisfactory solution to many practical problems, mainly due to uncertainties in the ability to evaluate samples with high sensitivity and specificity, or with results, with a degree of error or uncertainty in the identification that may be greater than 30% being totally unacceptable in the case, for example, of detection and identification of explosives (W. Nunes et al, Appl. Rad. Isotop. 56 , 937 (2002)).

There are methods of analysis that combine different analytical techniques with neural networks for the detection of substances in a sample. Thus, US 6,618,712 describes a method of identifying bioaerosols that combines spectra of ion masses (positive and negative) of known particles used to train a neural network and correlate the data obtained from an unknown substance, with another one identified through the training. These measures, however, require a prior treatment that introduces uncertainty in the result and lengthens the analysis time. (R.A. Gieray et al. J. Microbiol. Meth. 29, 191 (1997)).

Despite the existence of other methods of analysis that combine with neural networks, the field of neural networks can be improved in many directions.

In this invention a very high number of spectral points obtained from the plasma emission of the sample in the entire visible spectrum is used, combined with a system of mathematical algorithms suitable for handling a large number of data in a very short time and in a way Efficient and effective, ensuring 100% sensitivity and 100% specificity as will be demonstrated later.

The analysis method is based on an analysis of a single laser pulse that produces a vaporization process and subsequent formation of a plasma from the surface of the sample, obtaining the emission spectrum of this plasma in the order of a few microseconds without No type of data processing and subsequent spectral comparison with a dynamic spectral database for which trained mathematical algorithms are used and the comparison of spectra suitable for analysis with those of spectral matrices that can have at least 2048x100 points and with a total analysis time of less than 60 (sixty) seconds.

The method claimed in the patent allows to obtain great sensitivity (concentrations between parts per million (ppm) and parts per billion (ppb)), the ease of handling toxic or potentially dangerous samples, and performing in-situ analysis of samples. Given the limited calculation time necessary for the treatment of the data supplied by the laser (less than 10 seconds), they make the Laser-NNs conjunction offer online measurement possibilities unknown so far.

Description of the invention

The present invention consists of a rapid and direct method for the qualitative determination and / or complete spectral analysis of chemical compounds present in liquid or solid samples. This invention allows identification and characterization of samples, allowing the correlation of analytical data with organoleptic or quality properties and / or the determination of substances, compounds, contaminants, pathogens or bioactive agents, in the analyzed samples.

By using its own characteristics present in the LIBS technique and the mathematical algorithms used, the new methodology allows the identification of a sample without any preparation, in any state of the matter, with a single pulse, in a time less than one second and with high sensitivity and specificity.

The device used for the identification and characterization of samples, shown in Figure 1, consists of a pulsed laser (1), in this particular case an Nd: YAG laser has been used working at a frequency of 1 to 20 Hz, a fixed wavelength of 1064 nm and a pulse duration of 4 nanoseconds that can provide up to 180 mJ / pulse of output energy. This wavelength is not limiting for the use of other wavelengths, which can also emit this laser as 266 or 532 nm, or other wavelengths produced by any other type of laser both solid-state or gaseous (nitrogen laser , carbon dioxide laser, eximer laser, OPO laser, etc.) that allows to obtain sufficient energy conditions to produce a plasma. For this reason lasers can be used both in ultra violet, visible or infrared.

Appropriate mirrors (2) and lenses (3) have been used in order to focus the laser beam on the sample (4). Plasma emission is collected using a 1 meter optical fiber coupled to the detector, which in turn is activated by the laser pulse. The detector is a CCD optical sensor that provides 2048 spectral points in a range of 200 to 1100 nm. The detector signal is subsequently compared to a spectral base stored using a system of mathematical algorithms. A graphical interface allows the user to control the laser, observe the spectrum of the sample obtained, and the representation of the results for evaluation.

The spectral information corresponding to the emission of elements such as P, C, Mg, Ca, and Na provides sufficient information to achieve the identification of the samples and, subsequently, in the case of food, correlate the results of the spectral analysis with the organoleptic properties (in the case of food) or sample quality, previously determined by other analytical methods.

Although the quality of the spectrum and its reproducibility can vary considerably, they do not affect the ability of the system to identify the sample. This is so for two reasons: first, the system is trained with a set of spectra that have been stored using all possible variations of the ablation parameters, such as changes in the lens-sample distance, laser pulse energy etc. and, second, both the relationship of intensities and the width of each band is analyzed by the neural network which makes discrimination and identification possible.

In order to create a dynamic spectral database that collects the “fingerprints” of each sample, a neural network (NN) with forward connections is used, specifically it is a perceptron propagation model (back-propagating perceptron model ). The NN used consists of three layers called input, output and hidden, formed by neurons (only operating element). The input layer is used only for the input of the data matrix in the NN. In the other layers, nonlinear calculations are performed. In the hidden layer, each neuron receives signals from other input neurons, adding these through the activation function, equation 1. Then, the result is transformed by the transfer function, equation 2. Finally, the result is sent to the neurons of output (neurons in the output layer).

In equations 1 and 2, yie and k represent the output of the hidden neurons (j) and the output neurons (k) respectively, wjk represents the weight between the jth of the hidden neurons and that of the output neurons. As a transfer function, the sigmoidal hyperbolic tangent algorithm, equation 2, was used and the result used ranges from 0 to 1. The weights were adjustable parameters of the NNs associated with each of the connections between neurons and these can modify the communication signal between them. . The optimization process of the NN weight matrix was carried out using the SCG conjugate scaled gradient training algorithm. This algorithm was selected in order to prevent over-adjustment (violation of Occam’s razor) and overtraining (I.V. Tetko, et al. J. Chem. Inf. Coput. Sci. 35, 826 (1995)). The problem of overtraining refers to the fact that the network only memorizes the learning set and loses its ability to generalize. On the other hand, the SCG algorithm was developed for rapid training of neural network learning systems (H. Demuth et al, Neural Network Toolbox for use whit Matlab User's guide version 4.0.6. Ninth printing revised; The math Works: Natick , MA, 2005). An important feature of this type of high performance algorithms is that they can converge hundreds of times faster than other types of algorithms, but in turn, consumes a large amount of memory. The SCG minimizes the prediction error by using a linear combination of the mean square error (MSE), equation 3. This determines the correct combination to produce a network that generalizes new input data within the range of the learning data.

In equation 3, N corresponds to the number of observations, and k, the real value, rk, the estimates in the NN model and k NN output layer, respectively.

The basic backpropagation algorithms adjust the weights in a steep downward slope (negative gradient values), this is the direction in which the performance function decreases more rapidly. It turns out that, although the function decreases more rapidly along the negative values of the slope, it does not necessarily produce the fastest convergence. In the conjugate gradient algorithm, it performs a search along conjugate directions, which generally produces a faster convergence. (See Hagan, M.T.,

H.B. Demuth, and M.H. Beale, Neural Network Design, Boston, MA: PWS Publishing, 1996. for a broader discussion of conjugate gradient algorithms).

Each of the conjugate gradient algorithms requires a search line in each interaction. This search line is computationally expensive, because it requires the network to respond to all training entries several times for each search. The conjugate gradient algorithm developed by Moller [(M.F. Moller, Neural Networks, 6 525 (1993))], was designed to prevent time-consuming search line.

All conjugate gradient algorithms begin with the search for the descending direction of the gradient (negative gradient) in the first interaction.

Subsequently, a search line is made, to determine the optimal distance of movement along this search direction:

The general procedure to determine the new search address is the combination of the new pending address with the previous search:

The different versions of the conjugate gradient algorithms are distinguished by the way in which the constant βk is calculated, for the SCG algorithm the update procedure is:

This algorithm combines the true region approach (used in the Levenberg-Marquardt algorithm (D. Marquardt, J. Appl. Math. 11, 431-441, (1963)), with the conjugate gradient approximation, (see Moller, for a detailed explanation of this algorithm).

The topology of the NN used in this work consisted of an input layer with 2048 neurons for the input variables (LIBS spectrum). A hidden layer (optimized later) and an output layer with two neurons to determine if the recognition is positive or negative. This topology with a single hidden layer was adequate to solve the problems of all the samples studied. Each NN model used in this work was designed using the Matlab program, selected for its effectiveness in vector calculation.

Since the NN used is based on the supervised algorithm, to optimize the weight matrix it is necessary to use input and output data that properly characterize the process that will be modeled. These data have been taken directly from 50 LIBS spectra obtained from the samples. Each spectrum corresponds to a single laser pulse. Both the samples and the spectral data obtained did not have any previous treatment. The data set (LIBS spectra of the samples) was distributed randomly in the learning (80%) and in the verification (20%) of the samples, taking into account that no data set or any of its replicas should be presented in the verification of the sample.

Accuracy is the main characteristic of a recognition procedure as a resource for decision-making, which is why the metrics to evaluate detection processes are of significant importance and involve the relative frequency of the correct and incorrect acknowledgments that an observer makes from the results obtained. The basic measures are the number of positive and negative, (true and false, PV, NV, FP, FN), from which the sensitivity and specificity of the detection processes are calculated.

A true positive (PV) corresponds to the correct detection of a substance, compound or characteristic in a sample, when it really exists. A true negative (NV) corresponds to the negative detection of a substance, compound or characteristic in a sample when it does not exist. False detections (PF, NF) correspond to cases in which the detection does not correspond to the reality of the sample.

Sensitivity and specificity are two metrics of the performance of a detection process that are constructed from the number of PV, PF, NV and NF in a validation sample. The sensitivity of a detection process refers to the probability that a substance, compound or characteristic is detected when it really exists. Sensitivity is specified as a fraction between 0 and 1, or a percentage between 0 and 100.

The sum of PV and FN corresponds to the total of positives in the detection process so the sensitivity of a detection system can be calculated as:

Sensitivity = PV / (PV + FN)

A sensitivity of 1 indicates that all substances, compounds or characteristics are detected. Sensitivity is also called True Positive Fraction.

The metric that complements sensitivity is the specificity which measures the probability that a detection process correctly reports the non-existence of a substance, compound or characteristic when it does not exist. The sum of NV and PF corresponds to the total false in the detection process so the specificity in a detection system can be calculated as:

Speci fi city = NV / (NV + PF)

A specificity of 1 indicates that the existence of a substance, compound or characteristic is never reported when it does not exist. The Fraction of Positive False is defined as (1-specificity) and is the fraction of samples that are reported wrongly.

To evaluate a detection process it is necessary to have the values of its two metrics: specificity and sensitivity, since a single metric cannot correctly evaluate the process. This is because a sensitivity of 1 can be forced if our detection system reports all cases as positive (which corresponds to a specificity of 0) and a speci fi cation of 1 can also be forced if our system reports all cases as negative. (This corresponds to a sensitivity of 0).

The analysis of the Receiver Operating Characteristic Curve (ROC) developed in the context of the detection of electronic signals in the early 1950s, found an important application in the process of making medical decision The ROC curve is the standard tool for plotting all possible combinations of sensitivity and specificity of a detection process. Usually, the Fraction of True Positives (TPF) or Y-axis sensitivity is plotted, vs the Fraction of False Positives (FPF)

or (1-specificity) on the X axis. In an ROC curve the ideal operating point is the upper left corner where TPF = 1, FPF = 0.

The identification procedure allows complete separation and therefore its identification for samples with very similar matrices. The ROC curves of these samples show that the identification procedure works perfectly, and for all practical purposes they can be considered to be resolved with 100% sensitivity and 100% specificity.

The interest of the analysis ranges from the simple identification of a sample to determine whether or not it has a certain characteristic, sensory, chemical, physical or biological. Including the possibility of determining its traceability or origin of the sample (food or samples in general).

Description of the fi gures

Figure 1 shows a schematic view of the device comprising both the laser equipment (1), the mirrors (2), lenses (3), sample (4), optical fiber (5), monochromator (6), pulse delay controller (7), oscilloscope (8) and personal computer (9).

Figure 2 shows the LIBS spectra of the oil samples classified as olive1 (a) and olive2 (b).

Figure 3 shows the results of the network, the results corresponding to the olive1 sample have a value of 1 (one) while the results corresponding to the olive2 sample have a value of 0 (zero).

Figure 4 shows the ROC response curve of olive samples.

Figure 5 shows the LIBS spectra of the tomato derivative samples classified as ketchup (a) and tomato-fried (b).

Figure 6 shows the results of the network, the results corresponding to the ketchup sample have a value of 1 (one) while the results corresponding to the tomato-fried sample have a value of 0 (zero).

Figure 7 shows the ROC response curve of tomato derivative samples.

Figure 8 shows the LIBS spectra of the ketchup samples classified as ketchup1 (a) and ketchup2 (b).

Figure 9 shows the results of the network, the results corresponding to the ketchup1 sample have a value of 1 (one) while the results corresponding to the ketchup2 sample have a value of 0 (zero).

Figure 10 shows the ROC response curve of ketchup samples.

Figure 11 shows the LIBS spectra of infant milk samples of similar composition, classified as milk1 (a) and milk2 (b).

Figure 12 shows the results of the network, the results corresponding to the milk sample1 have a value of 1 (one) while the results corresponding to the milk sample2 have a value of 0 (zero).

Figure 13 shows the ROC response curve of infant milk derivative samples.

Embodiment of the invention

The present invention is illustrated by the following examples that are not in any case limiting its scope, which is defined exclusively by the claim.

Example 1

Identi fi cation and classification of olive oils

As an example, in order to identify an oil sample, LIBS spectra obtained from a single laser shot are obtained from known or identi fi ed samples. A total of 50 laser pulses of each sample are stored in a database as fingerprints of the samples. In order to prevent black body radiation generated in the first moments of the plasma, a delay of 10 μs between the laser pulse and obtaining the spectrum in the evolution of the plasma was used as the optimal time. The spectra of two olive oil samples, therefore with a very similar type of matrix and with a slight difference in flavor, were classified as an olive sample1 and an olive sample2, and are shown in Figure 2. The classification made The NN network is shown in Figure 3. With the first 50 points corresponding to the olive sample1 and the remaining points to the olive sample2, it can be seen that these samples are completely separable. The evaluation of the procedure, ROC curve (Figure 4) shows that 100% of the spectra have been correctly classified, approximating the evaluation to an ideal 100% sensitivity and 100% specificity.

Example 2

Identi fi cation and classification of tomato derivatives

As an example, in order to identify different tomato derivatives, this example is presented as an example of the wide variety of elements and substances that can be analyzed in solutions or suspensions in various types of matrices, as in the case of tomato juice and which can be extended to other types of fruit juices and even to liquid substances such as milk and biological substances such as blood, extending its application to solid samples of minerals, metals, etc. The procedure used is similar to that already described and where a sample of tomato juice was placed in the sample holder, then the laser beam was focused on the surface of the samples to obtain LIBS spectra obtained from a single laser shot. It was used as optimal time, a delay of 10 μs between the laser pulse and obtaining the spectrum in the evolution of the plasma. The spectra of the ketchup and fried tomato samples were classified as ketchup and Tfrito, these spectra are shown in Figure 5. The classification carried out by the NN network is shown in Figure 6. The first 50 points corresponding to the ketchup sample and the remaining ones to the Tfrito sample, it can be seen that these samples are completely separable. The evaluation of the procedure, ROC curve (Figure 7) shows that 100% of the spectra have been correctly classified, approximating the evaluation to an ideal 100% sensitivity and 100% specificity.

Example 3

Identi fi cation and classification of commercial ketchup

As an example, in order to identify different types of ketchup of commercial origin, it is presented with the capacity of the method used to correctly identify and classify two identical samples, whose only difference is the introduction of a known amount of carotene with a lower concentration at 100 ppm, the matrix of both samples being exactly the same. The procedure used is similar to that already described and where the samples are placed in the sample holder, then the laser beam was focused on the surface of the samples to obtain LIBS spectra obtained from a single laser shot. It was used as optimal time, a delay of 10 μs between the laser pulse and obtaining the spectrum in the evolution of the plasma. The spectra of the ketchup tomato samples were classified as ketchup1 and ketchup2, these spectra are shown in the figure

5. The classification carried out by the NN network is shown in Figure 6. With the first 50 points corresponding to the ketchup1 sample and the remaining points to the ketchup2 sample, it can be seen that these samples are completely separable. The evaluation of the procedure, ROC curve (Figure 10) shows that 100% of the spectra have been correctly classified, approximating the evaluation to an ideal 100% sensitivity and 100% specificity.

Example 4

Identi fi cation and classification of dairy products (commercial infant milk)

As an example, in order to identify a milk sample, this example is presented as an example of the wide variety of elements and substances that can be analyzed using these methods of analysis. The procedure used is similar to that already described and where the samples are placed in the sample holder, then the laser beam was focused on the surface of the samples to obtain LIBS spectra obtained from a single laser shot. It was used as optimal time, a delay of 10 μs between the laser pulse and obtaining the spectrum in the evolution of the plasma. The spectra of the samples of dairy products were classified as milk1 and milk2, these spectra are shown in Figure 11. The classification made by the NN network is shown in Figure 12. The first 50 points corresponding to the milk sample1 and those remaining in the milk sample2, it can be seen that these samples are completely separable. The evaluation of the procedure, ROC curve (Figure 13) shows that 100% of the spectra have been correctly classified, approximating the evaluation to an ideal 100% sensitivity and 100% specificity.

Claims (12)

one.

Method of analysis for instantaneous identification and characterization of samples characterized in that it comprises: (a) irradiating a sample using a laser focused on the surface of a sample, (b) detecting the radiation produced by the chemical elements of the plasma that forms on the surface on which the laser beam is focused, obtaining a single spectrum of the sample, using an optical analyzer (c) to compare the detector signal with a dynamic spectral base using a system of artificial neural networks that include supervised mathematical algorithms.

2.

Method of analysis for instantaneous identification and characterization of samples, according to claim 1, wherein the laser is a solid or gaseous state laser that emits in the ultraviolet, visible or infrared range.

3.

Method of analysis for instantaneous identification and characterization of samples, according to claims 1 and 2, wherein the laser is of the Nd: YAG type working at a frequency of 1 to 20 Hz, at a fixed wavelength of 1064 nm, with a duration of 4 nanosecond pulse

Four.

Method of analysis for instantaneous identification and characterization of samples, according to claim 1, wherein the detection of the radiation produced by the chemical elements of the plasma is carried out through a CCD optical sensor that provides 2048 spectral points in a range of 200 to 1100 nm .

5.

Analysis method for instant identification and characterization of samples, according to claim 1, wherein the neural network has forward connections based on a perceptron propagation model, three layers (input, output and hidden), a transfer function between layers that uses a Sigmoidal hyperbolic tangent algorithm and its weight matrix is optimized by the SCG conjugate scaled gradient training algorithm preventing overfitting and overtraining.

6.

Device for instantaneous identification and characterization of samples comprising: (a) laser equipment capable of producing a vaporization process and subsequent formation of a plasma on the surface of a sample,

(b) an optical analyzer that detects the radiation produced by the chemical elements of the plasma, (c) a personal computer with a graphical interface that allows the user to control the laser, observe the spectrum of the sample obtained and the representation of the results for its analysis and that includes a dynamic spectral database that uses a system of artificial neural networks that include supervised mathematical algorithms.

7.

Device for instantaneous identification and characterization of samples, according to claim 6, wherein the laser is a solid or gaseous state laser that emits in the ultraviolet, visible or infrared range.

8.

Device for instantaneous identification and characterization of samples, according to claims 6 and 7, wherein the laser is of the Nd: YAG type working at a frequency of 1 to 20 Hz, at a fixed wavelength of 1064 nm, with a pulse duration between 4 and 5 seconds.

9.

Device for instantaneous identification and characterization of samples, according to claim 6, wherein the detection of the radiation produced by the chemical elements of the plasma is carried out through a CCD optical sensor that provides 2048 spectral points in a range of 200 to 1100 nm.

10.

Device for instantaneous identification and characterization of samples, according to claim 6, wherein the neural network has forward connections based on a perceptron propagation model, three layers (input, output and hidden), a transfer function between layers using an algorithm Sigmoidal hyperbolic tangent and its weight matrix is optimized by the SCG conjugate scaled gradient training algorithm preventing overfitting and overtraining.

eleven.

Use of the claimed device for the qualitative spectral determination of chemical compounds present in a sample and their subsequent correlation with organoleptic or quality properties of said sample in water, food or biological samples, in any state of matter (solid, liquid or gas) previously determined by other analytical methods.

SPANISH OFFICE OF THE PATENTS AND BRAND Application no .: 200901443 SPAIN Date of submission of the application: 06.19.2009 Priority Date: REPORT ON THE STATE OF THE TECHNIQUE 51 Int. Cl.: See Additional Sheet RELEVANT DOCUMENTS

Category

Documents cited Claims Affected

X

RAMIL, A. et al. Application of artificial neural networks for the rapid classification of archaeological ceramics by means of laser induced breakdown spectroscopy (LIBS). Applied Physics A. Materials Science &amp; Processing 2008, Vol. 92, No. 1, pages 197-202 <DOI: 10.1007 / s00339-008-4481-7> 1-4.6-9.11

Y

5.10

Y

MOLLER, M.F. A scaled conjugated gradient algorithm for fast supervised learning. [online] 18.02.2005. [Recovered on 04.11.2010] Recovered on the Internet: <URL: http://www.sciencedirect.com/science/article/B6T08-4HC1K78-5/2/83389e738f7c78e981d 1addae6bfa827>, summary. 5.10

TO

Wikipedia (several authors). Perceptron multilayer. [online] 06/14/2009 [retrieved 04.11.2010] Recovered on the Internet: <URL: http: //en.wikipedia.org/w/index.php? title = Multilayer_perceptron &amp; oldid = 296332932 &amp; printable = yes> 1,5,10

X

MOTTO-ROS, V. et al. Quantitative multi-elementary laser-induced Breakdown spectroscopy using artificial neural networks. Journal of the European Optical Society - Rapid publications. 2008. Vol. 3, pages 08011-1 to 08011-5. ISSN 1990-2573. <DOI: 10.2971 / jeos. 2008.08011> 1-4.6-9.11

TO

Wikipedia (several authors). Nd: YAG laser. [online] 04/26/2009 [retrieved 04.11.2010] Recovered on the Internet: <URL: http://en.wikipedia.org/w/index.php?title=Nd:YAG_laser&amp;oldid=286140280> 2,3,7,8

TO

GORNUSHKIN, I.B. et al. Identification of Solid Materials by Correlation Analysis Using a Microscopic Laser-Induced Plasma Spectrometer. Analytical Chemistry November 14, 1999. Vol. 71, No. 22, pages 5157-5164. <DOI: 10.1021 / ac9905524> 3,4,8,9

X

US 2003132142 A1 (KUMAR) 07.17.2003, summary; paragraphs [2-8], [40-50], [63-65], [75-78], [83-86]; Figures 1,2,6,10. 1-4.6-9.11

TO

Wikipedia (several authors) Laser-induced breakdown spectroscopy. 06.02.2009 [retrieved 04.11.2010] Recovered on the Internet: <URL: http: //en.wikipedia.org/w/index.php? Title = Laserinduced_breakdown_spectroscopy &amp; oldid = 293 940191> 4

Category of the documents cited X: of particular relevance Y: of particular relevance combined with other / s of the same category A: reflects the state of the art O: refers to unwritten disclosure P: published between the priority date and the date of priority submission of the application E: previous document, but published after the date of submission of the application

This report has been prepared • for all claims • for claims no:

Date of realization of the report 30.03.2011

Examiner A. Figuera González Page 1/7

SPANISH OFFICE OF PATENTS AND BRANDS Application no .: 200901443 SPAIN Date of submission of the application: 06-18-2009 Priority Date: REPORT ON THE STATE OF THE TECHNIQUE 51 Int. Cl.: See Additional Sheet RELEVANT DOCUMENTS

Category

Documents cited Claims Affected

X

SNYDER, E.G. et al. Laser-induced breakdown spectroscopy for the classification of unknown powders. Applied Optics 01.11.2008. Vol. 47, No. 31, pages G80-G87 <DOI: 10.1364 / AO.47.000G80> 1-4.6-9.11

Category of the documents cited X: of particular relevance Y: of particular relevance combined with other / s of the same category A: reflects the state of the art O: refers to unwritten disclosure P: published between the priority date and the date of priority submission of the application E: previous document, but published after the date of submission of the application

This report has been prepared • for all claims • for claims no:

Date of realization of the report 30.03.2011

Examiner A. Figuera González Page 2/7

REPORT OF THE STATE OF THE TECHNIQUE Application number: 200901443 CLASSIFICATION OBJECT OF THE APPLICATION G01N21 / 71 (2006.01) G01J3 / 443 (2006.01) G05B13 / 02 (2006.01) Minimum documentation sought (classification system followed by classification symbols) G01N, G01J, G05B Electronic databases consulted during the search (name of the database and, if possible, terms of search used) INVENES, EPODOC, WPI, TXTEN, Internet State of the Art Report Page 3/7  WRITTEN OPINION Application number: 200901443 Date of Written Opinion: 30.03.2011 Statement

Novelty (Art. 6.1 LP 11/1986)

Claims 3-11 Claims 1, 2 IF NOT

Inventive activity (Art. 8.1 LP11 / 1986)

Claims Claims 1-11 IF NOT

The application is considered to comply with the industrial application requirement. This requirement was evaluated during the formal and technical examination phase of the application (Article 31.2 Law 11/1986).  Opinion Base.- This opinion has been made on the basis of the patent application as published. State of the Art Report Page 4/7  WRITTEN OPINION Application number: 200901443 1. Documents considered.- The documents belonging to the state of the art taken into consideration for the realization of this opinion are listed below.

Document

Publication or Identification Number publication date

D01

RAMIL, A. et al. Application of artificial neural networks for the rapid classification of archaeological ceramics by means of laser induced breakdown spectroscopy (LIBS). Applied Physics A. Materials Science &amp; Processing Vol. 92, No. 1, pages 197-202. 2008

D02

MOLLER, M.F. A scaled conjugated gradient algorithm for fast supervised learning. Summary. 02/18/2005

D03

Wikipedia (several authors). Perceptron multilayer. 06/14/2009

D04

MOTTO-ROS, V. et al. Quantitative multi-elementary laser-induced Breakdown spectroscopy using artificial neural networks. Journal of the European Optical Society - Rapid publications. Vol. 3, pages 08011-1 to 08011-5. 2008

D05

Wikipedia (several authors). Nd: YAG laser. 04/26/2009

D06

GORNUSHKIN, I.B. et al. Identification of Solid Materials by Correlation Analysis Using a Microscopic Laser-Induced Plasma Spectrometer. Analytical Chemistry November 14, 1999. Vol. 71, No. 22, pages 5157-5164. 11/14/1999

D07

US 2003132142 A1 (KUMAR) 07.17.2003

D08

Wikipedia (several authors) Laser-induced breakdown spectroscopy. 06.02.2009

D09

SNYDER, E.G. et al. Laser-induced breakdown spectroscopy for the classification of unknown powders. Applied Optics Vol. 47, No. 31, pages G80-G87. 01.11.2008

2. Statement motivated according to articles 29.6 and 29.7 of the Regulations for the execution of Law 11/1986, of March 20, on Patents on novelty and inventive activity; quotes and explanations in support of this statement CLAIM 1. Document D01 describes the application of artificial neural networks for rapid classification of archaeological ceramics by means of laser-induced dissociation spectroscopy (LIBS). The laser beam is focused on the surface of the sample. The plasma emission is captured and guided by an optical fiber to a spectrograph. An ICCD detects the spectrum of plasma emission. See D01, section 2.1 Experimental setup, pages 197-198. To automate and facilitate the comparison of LIBS spectra, artificial neural networks are used. For the training of artificial neural networks a backpropagation algorithm can be used which is a known (see for example reference work D03) supervised learning algorithm for neural networks. In this algorithm, a certain number of reference data whose results are known are used as input data. The calculated result is then compared with the known result. The difference between the two results is then fed back to recalculate the weights. See D01, summary and section 3.1 Designing the networks, page 198. Thus, document D01 discloses all the technical characteristics of claim 1. Therefore, claim 1 is considered to be novel in accordance with article 6 of the Patent Law. State of the Art Report Page 5/7  WRITTEN OPINION Application number: 200901443 CLAIM 2. In document D01 an Nd-YAG laser is used that operates at 355 nm. It is a solid-state laser that emits in the ultraviolet range. Thus, document D01 discloses a particular case of a device that meets the requirements of claim 2, which is dependent on claim 1, which is not new, so claim 2 is in turn devoid of novelty.  REIVIDICATION 3. In document D01, the laser used is an Nd: YAG laser that operates in the third harmonic, at 355 nm wavelength. It is known that Nd: YAG type lasers usually emit at 1064 nm (wavelength used in D04), but they can also generate laser light at 532 nm, 355 nm (wavelength used in D01) or 266 nm. See for example a reference work such as document D05. Document D04 mentions the possibility of using a variable repetition rate of up to 20 Hz, and the repetition rate can be reduced in 1Hz experiments See D04, first page, section 2. LIBS experiment. In documents D01 and D04, the pulse length is 6 ns and 7 ns respectively (see D01, page 197, section 2.1 Experimental setup and D04, first page, 2. LIBS experiment). But in document D06 the possibility of using pulses of 4 ns is mentioned (see D06, Experimental section, page 5157) and in document D07, in figure 4, the pulses are between 5 ns and 10 ns. That is to say that in claim 3 there are specific technical characteristics that are not explicitly included in document D01, but they are known technical characteristics that seem to respond to mere design options that do not produce any surprising technical effect or apparently pose any technical problem that has been You need to overcome for your employment. Therefore, claim 3, which depends on the previous claims that lack novelty, does not provide any technical characteristic that is not a design option known in the state of the art, and consequently lacks inventive activity according to the article 8 of the Patent Law. CLAIM 4. In document D01 512 intensity values are used in the spectral window between 260 nm and 340 nm detected by an iCCD. See D01, page 198 right column, first paragraph and page 199, left column, first paragraph. However, in document D06 a spectrometer is used whose second channel has a spectral range between 200 and 850 nm and a linear CCD of 2048 pixels (see D06, page 5158, right column). On the other hand, it is known in the state of the art that the spectrometer a LIBS system can have a response typically in the range of 170 nm to 1100 nm which is the approximate response range of a CCD detector. See for example reference work D08. Therefore, although the specific technical characteristics of claim 3 are not explicitly set out in document D01, these are known technical characteristics that appear to respond to mere design options that do not produce any surprising technical effect or apparently pose any technical problem that it has been necessary to overcome for its use. In short, claim 4, which depends on claim 1, which lacks novelty, does not provide any technical characteristic that is not a known design option, and therefore lacks inventive activity. CLAIM 5. The neural networks of document D01 are neural networks with forward connections with backpropagation with an input layer, an output layer and a hidden layer. A hyperbolic-sigmoidal tangent function is used as a transfer function. See D01, page 198, right column, first paragraph and page 199, left column, second paragraph. It is a multilayer perceptron as explained, for example, in a reference work such as wikipedia (see D03). Thus, the difference between what is disclosed in document D01 and the object of claim 1, is that in document D01 a backpropagation algorithm is used for the optimization of the weight matrix while in the invention object of claim 1 employs the gradient conjugate scaling algorithm. However, document D02 (see summary) explains how the conjugate scaling gradient is a supervised training algorithm that can be used instead of the backpropagation algorithm to train a neural network. The person skilled in the art could then have resorted to this alternative disclosed in document D02 of training of neural networks to train the neural networks of document D01 without overcoming any technical problem nor obtaining any result other than expected. State of the Art Report Page 6/7  WRITTEN OPINION Application number: 200901443 In conclusion, claim 5, which depends on claim 1, which lacks novelty, in turn lacks inventive activity. CLAIM 6. All the features of claim 6 common with those of claim 1 are known as explained for claim 1. The additional technical feature of claim 6 with respect to that set forth for claim 1 is that the device consists of a personal computer with a graphic interface that allows laser control, spectrum observation and representation of results. However, although in document D01 the use of a personal computer is not explicitly mentioned, an element of general use in the state of the art is treated as can be observed for example, in document D04 figure 1 and in the document D06, figure 1. Thus, independent claim 6 lacks inventive activity.  CLAIMS 7 TO 10. For a reasoning similar to that set forth for claims 2 to 5, claims 7 to 10, which depend on claim 6 lacking inventive activity, in turn lack inventive activity. CLAIM 11. In document D01 a device is used as claimed for the qualitative determination of chemical compounds in solid state. The possibility of using the same technique for samples in liquid or gaseous state is well known in the state of the art, as illustrated for example in reference work D08 or in document D04 in the introduction. As for the correlation with organoleptic or quality properties in food water samples or biological samples previously determined by other analytical means, no specific explanation of how this result is achieved is provided in the description, so it is considered that it is either The expression of a desire for results that are intended to be achieved or is an obvious correlation that has not been considered to be explained in more detail. In conclusion, claim 11 which refers to the use of a device that lacks inventive activity does not provide any technical characteristic that is not already known and therefore lacks inventive activity. State of the Art Report Page 7/7