IT202000002104A1 - METHOD FOR DETERMINING THE COMPOSITION OF WASTE AND SLUDGE FROM ANIMAL BREEDING OR ANAEROBIC DIGESTION - Google Patents

Description

? METHOD FOR DETERMINING THE COMPOSITION OF WASTE AND SEWAGE FROM ANIMAL BREEDING OR ANAEROBIC DIGESTION?

FIELD OF INVENTION

The present invention relates to a method for determining, with greater accuracy than that allowed by the methods currently in use, the composition of manure and sewage from animal breeding or from anaerobic digestion used as fertilizers in agriculture.

STATE OF THE TECHNIQUE

Manure and sewage from livestock farms or from the anaerobic digestion of plant waste are used as soil fertilizers that recycle nutrients from the soil, plants and livestock production sites. The use of these effluents and sewage? beneficial for the economy of farms and biogas production plants but can? cause the pollution of the aquifers if the spreading is not carried out in the correct way. For this reason, in different parts of the world there are regulations on the distribution of this waste: in Europe, the regulation is based on quantities? d? nitrogen, while in North America it is based on the quantity? of phosphorus which is brought, by spreading, to the ground.

To better respond to the implementation of these regulations, rapid analysis methods are required with in-line and continuous measurements carried out directly on the operating machines that spread fertilizers on the ground.

Research in the sector yes? practically exclusively addressed to the use of spectrophotometric analytical techniques: see, for example, the article? Monitoring biological wastewater treatment processes: recent advances in spectroscopy applications ?, D. P. Mesquita et al., Rev Environ Sci Biotechnol (DOI 10.1007 / s11157-017-9439-9; online publication of 3 August 2017), which summarizes the content of the work done in the sector from 2000 onwards, all based on spectrophotometric methods.

Among the spectrophotometric methods,? In particular, near infrared spectroscopy emerged (technique indicated in the sector with the acronym NIRS, from the English Near Infrared Spectroscopy, while NIR means the corresponding field of electromagnetic radiation); the reasons are that this analysis allows to obtain results quickly,? non-destructive and non-invasive,? suitable for online use and does not require specific sample preparation; moreover, NIR radiation has high penetration even in very dirty samples (unlike, for example, spectrophotometry in the visible); finally, NIR instruments and probes are generally robust and require very little maintenance. An example of these studies? reported in the article? Activated sludge process monitoring through in situ near-infrared spectral analysis ?, A. M. A. Dias et al., Water Science & Technology, 57.10, 2008.

Despite some good results, this technique has some limitations. For example, in determining the parameters of main agronomic interest, the results are generally good in quantifying the organic component of the wastewater, in particular dry matter and organic matter; the results obtained are discrete for the total nitrogen while they are discordant for the ammonia fraction according to the characteristics of the instrument (spectral range, signal / noise ratio, characteristics of the calibration data set); moreover, the results are generally disappointing for other mineral elements that are present in the wastewater in the form of ions, mainly with regard to phosphorus and potassium. There? can you? explain with the fact that the NIR technique is not? able to directly determine the inorganic (mineral) fraction present in a matrix. For some mineral elements good calibrations are obtained, but what about us? it is simply due to the fact that these are in some way related to the organic fraction, for example in the form of chelates.

Not all types of NIR sensors are then applicable on machines for spreading wastewater. In particular, the technology that is best suited to this type of application is based on instruments without moving parts such as diode array, able to analyze the product at the speed? of a few thousandths of a second.

However, these sensors have a working range limited to the first part of the NIR spectral band (generally between 800 and 1700 nm) and therefore are not able to benefit from the additional information that can be obtained. get with the laboratory instruments more? performing, reaching up to 2500 nm. Despite these instrumental limitations, diode array instruments are in fact the only instrumental type currently on the market, whose performance has been evaluated by the Company? German Agriculture (DLG). The performance reports (DLG6801, DLG6811_e, DLG6886_e, DLG6887_e), used as certification of the goodness? predictors of these NIR apparatuses, in fact show that especially for P and K the prediction errors are very high. Furthermore, there is evidence that the NIR measurement of the wastewater on the tank car can be distorted in the case of foaming, a not uncommon event during the recirculation of the sewage inside the tank car or during spreading in the field.

Purpose of the present invention? that of providing a method for the analysis of the composition of wastewater and recycled sewage from the livestock and agricultural sectors, which provides better results? accurate and reliable than those obtainable with a spectrophotometric technique, in particular for phosphorus and potassium.

SUMMARY OF THE INVENTION

This object is achieved with the present invention, which relates to a method for analyzing wastewater and sewage produced in the livestock and agricultural sectors which consists of:

- carry out a spectrophotometric analysis in the near infrared (NIR) of the sample; - at the same time, carry out at least a second measurement with a technique other than NIR on the same sample;

- combine the results obtained with the NIR spectrophotometric analysis and with the second measurement using a data fusion technique.

DETAILED DESCRIPTION OF THE INVENTION

The method of the invention consists in obtaining data from an NIR spectrophotometric measurement and at least a second measurement with a different technique, and combining the data thus. obtained with a data fusion technique (technique best known in the chemical analysis sector with the English wording? Data Fusion ?, which will also be used in this text).

The implementation of the NIR measurement? known to the technicians of the sector. In short, in this technique a spectrum of absorbance (or the complement to 100 and therefore of reflectance or transmittance depending on the type of instrument) of the sample is detected. The spectrum is digitized and the prediction value is calculated by multiplying the absorbance values of the spectrum for each predetermined wavelength with the previously determined calibration coefficients, obtaining a first numerical value X? from the following formula (1):

in which:

- n? the number of points of the spectrum taken into consideration;

-? a calibration parameter valid for the i-th point of the spectrum;

- me? the absorbance, reflectance or transmittance data obtained from the NIR spectrum for the i-th sampled point.

The values are obtained from previous calibration measurements carried out on known samples, in which the concentrations of the species under examination are known through traditional chemical analyzes: for example, the organic nitrogen values can be obtained from measurements with the Kjeldahl method of the previously dried sample; the total nitrogen can? always be measured with the Kjeldahl method on the sample as it is; ammonia nitrogen? determined by Kjeldahl after extraction or by photometric measurement; finally, the concentration of potassium and phosphorus pu? be determined by ICP-AES analysis, Atomic Absorption or by X-ray Fluorescence (XRF).

Probes and NIR instruments for carrying out the measurement are widely available commercially. One of the most popular tools? widespread? l? Harvest Lab 3000 <? > marketed in Italy by the company? ; other possible useful tools are the company's Corona Extreme? , Van Control 2.0 of the company? , m-u-t NIR speedspy onboard by

, or Evo NIR of

The NIR region ranges from about 700 to about 2500 nm. This spectral range can not? be covered by a single sensor, but in general the band from 700 to 1100 nm? covered with silicon sensors, InGaAS sensors cover the region 950-1700 nm, and the extended InGaAs (they are thus defined sensors with a higher indium content than the basic sensors, in which the atomic ratio between indium and gallium is about 1 ) reach up to 2500 nm together with PbS sensors.

Generally for cost reasons the instruments use only one type of sensor (even if on the market there are also those with two sensors) and for a good compromise between cost and performance the pi? employed are the non-extended InGaAs ones. The present invention has used precisely InGaAs sensors, recording the absorbance values on 256 points between 950 and 1650 nm, to each of which it will correspond? a value i used in the formula (1) above.

The same application can? be done with cheaper sensors with fewer registration points, the important thing? which fall within the NIR range preferably covering a part of the spectral range greater than 1100 nm.

The second measure of the method of the invention can? be of various types; for example, can? be a measurement of turbidimetry, pH, temperature or conductivity.

From this second measurement we obtain a value, similar to the wavelength absorbance indicated as mn + 1, to be combined with the mi NIR values. The value X will be? obtained with a formula similar to the formula (1) described above, using the values mi NIR and the value mn + 1 of the second sensor.

The value mn + 1? preferably obtained from a conductivity measurement, which is carried out with instruments, called conductivity meters, widely available commercially; instruments of this type are for example the digital conductivity sensor? inductive 3700 (sold in Italy by the WTW <TM> Cond 3110 ProfiLine <TM> (sold in Italy by and Chemitec S411TEF C (produced and sold by. This technique, in turn well known in the field of electrochemical measurements, consists in measuring the conductivity, measured in Siemens (S, equivalent to the reciprocal of Ohm, that is S = 1 /?), of a sample of a solution interposed between two electrodes. The measurement can be made specific, and therefore an intensive characteristic of the given solution, normalizing the measurement with respect to the surface of the electrodes and their distance, obtaining the conductivity of the solution, which is measured in S / m.

Unlike the NIR measurement, that of conductivity? provides already? as a result a single value expressed in Siemens. Also, this measure? more influenced by the system temperature, so this parameter must be taken into account.

The conductivity electric (at 25? C)? an indirect measure of the total concentration of ions in a solution. In samples, such as wastewater, where the pH? close to neutrality, the contribution of the ions H <+> and OH? to the conductivity? electric? and therefore the conductivity values? they are essentially attributable to the concentration of anions and cations present in the wastewater. In livestock manure the main cations are represented by Na <+>, K <+>, Ca <2+>, Mg <2+> and NH4 <+>, which are balanced by the anions SO4 <2->, PO4 < 2-> and Cl ?. The predominant cation? ammonium (NH4 <+>) followed by potassium (K <+>). The conductivity can therefore be used as an indirect measure of the concentration of ammonium and potassium in the solution. When the solutions are very concentrated, the relationship between conductivity? electrical and concentration of the ions in the solution becomes non-linear due to the electrical and ionic interactions between charged ions: for this reason it could be useful to couple the conductivity measurement? to another type of on-line measurement of dry matter, for example by turbidimetry and / or NIR. Also the conductometric measurement? much less influenced by the formation of foams for which an analysis system composed of pi? sensors could be made more? stable.

Normally commercial conductivity meters integrate a temperature probe and provide as output a compensated data and? Reported? at 25 ° C; if the conductivity meter used does not include this probe,? It is necessary to foresee in the method of the invention the addition of a temperature sensor and to compensate a posteriori the data supplied by the conductivity meter. The data obtained from the conductometric measurement, compensated for the temperature, constitutes the value mn + 1 to be used in the method of the invention.

The combination of different sensors, generically defined Data Fusion, or in the specific case Sensor Fusion can? essentially be operated at three different levels: low-level, mid-level and high-level; see for example the article? Data fusion methodologies for food and beverage authentication and quality assessment - A review ?, E. Borr? s et al., Analytica Chimica Acta, vol. 891 (2015) pp. 1-14, https://doi.org/10.1016/j.aca.2015.04.042.

In low-level fusion, the data coming from the various sources are simply concatenated sample by sample in a single matrix that has as many rows as there are samples analyzed and as many columns as there are signals (variables) measured by the different instruments. This matrix is then used as the basis for calculating a single predictive model. Although the concatenation can take place directly on the raw data, mathematical pretreatments on the various signals are normally necessary before their association in a single matrix: methods of pre-processing, selection of variables or extraction of characteristics (commonly referred to in the sector by the English term? Features ?) are the main techniques used for this purpose.

The mid-level fusion (also called feature level fusion) primarily extracts the most? relevant features from each sensor separately, then concatenates them into a single matrix which is then used for multivariate regressions. The approach more? common ? that of merging a certain number of latent variables obtained independently of the signals of each instrument, normally the scores (for which the English term? scores? is used in the sector) deriving from the analysis of the principal components or of the PLS.

In high-level data fusion, also called decision level, several regressions or models are calculated separately from each data source (sensor) and the results of each model are combined into a final model: the difficulty? in this case ? find a combination that performs better than each model separately.

Normalization? a necessary operation when coupling data from different sensors and serves to standardize the data in order to give equal weight to each. This can? take place via Mean Centering or Autoscaling. Mean Centering is applied by subtracting the average value obtained from the dataset for each column (variable) that makes up the vector (sample). For a data set R (I? J) composed of I samples each in turn defined by J discrete variables, the j-th discrete variable of the j-th sample? defined by the relation:

in which mc indicates the data Ri, j obtained from the Mean Centering procedure.

Autoscaling uses the above formula followed by dividing each column (variable) by the standard deviation of that column.

This information incorporated at various levels, such as low-level or mid-level, in a single matrix of concatenated data is processed through chemometric methods that exploit principal component algorithms (PLS; SVM; neural network, etc.) to create models predictive type.

The coupling of different instruments allows to exploit the peculiarities? of each type of sensor: for example the NIR? suitable for accurately determining the organic component of wastewater, such as dry and organic matter, while the conductivity? electric? well correlated to the presence of dissolved ions in solution such as ammonium (NH4-N) and K. Phosphorus? more correlated to the dry matter so the coupling with a turbidimeter could lead to further improvements.

The data cos? obtained are then processed using software that allow you to calculate simple, multiple or multivariate linear regressions, for example spreadsheets and chemometric software that allow you to calculate mathematical models based on multivariate data analysis.

For NIR calibrations? It is possible, for example, to use the chemometric software Ucal and multivariate algorithms such as partial regressions with the least squares method, known by the acronym PLS, from the English partial least squares regression (PLSR). The PLS? probably the method of regression to latent variables pi? used and? particularly suitable for data sets with pi? variables you sample. The PLS regression tries to maximize the covariance between the blocks of the X and the blocks of the Y in such a way that the new latent variables explain not only the variability? of the X's but are also correlated at most with the Y's.

For data fusion, the NIR spectrum (m1 ?? mn) is coupled with the EC value (mn + 1, analytical value measured by the conductivity meter) of each sample (low level data fusion). The matrix? Spectra + CE? ? was imported into UCal and the data was normalized before applying the PLS type algorithm.

The invention will be further described in the following experimental part.

Materials and methods

The experimental work? was carried out in the laboratory using the following equipment:

- AuroraNir NIR spectrophotometer equipped with a diode array sensor consisting of 256 pixels with a spectral range of 900-1650 nm, with reflectance reading; the raw wavelength values were interpolated in order to obtain a spectral scale from 950 to 1650 nm every 2 nm for a total of absorbance values n = 351.

- portable conductivity meter WTW <TM> Cond 3110 ProfiLine <TM> (Thermo Fisher Scientific) with universal probe Tetracon <? > 325 (manufactured and sold by Global Water, a division of) which incorporates a conductivity cell? with 4 electrodes and a temperature sensor for data compensation.

EXAMPLE 1

220 digestate samples from anaerobic digestion plants were analyzed with the indicated instruments. Each sample, represented by about 1 liter of wastewater,? it has been carefully mixed and then poured into an aluminum tray to a height of about 3 cm; the NIR tool? was placed immediately in direct contact with the wastewater, with the reading window immersed in the liquid for about 1 mm. While reading and recording the NIR tool? was moved manually while remaining in contact with the wastewater by operating 4 consecutive scans. As soon as the NIR reading is finished, the conductivity meter probe electrode? been immersed in the wastewater, after stirring through the probe itself and the conductivity value, expressed in mS / cm,? been recorded on the Excel sheet as soon as the value on the display yes? stabilized.

The NIR absorbance data were exported from Ucal to an Excel sheet and then integrated with the conductivity value? electrical temperature compensated: in practice for each digestate sample at 351 NIR absorbance values (expressed in Log 1 / R)? has been added the single conductivity data? compensated for temperature (expressed in mS / cm compensated at 25 ° C). From Excel the data were then reconverted to the format compatible with the Ucal chemometric software.

For reference chemical analyzes each sample? been divided into three aliquots: a 40 ml aliquot for the determination of total nitrogen; a second aliquot of 40 ml with the addition of 3 ml of 0.6 N hydrochloric acid for the analysis of ammoniacal nitrogen, after filtering the sample; a third aliquot, of about 500 ml, for the determination of the dry matter, organic matter, phosphorus and potassium,? dried at 60 ° C for at least 72 h, then ground with a mill equipped with a grid at 1 mm.

Total nitrogen? been quantified by Kjeldahl (indicated in the sector with the abbreviation TKN, Total Kjeldahl Nitrogen). The ammonia nitrogen? was determined after filtering the sample and extraction by Kjeldahl; phosphorus (P) and potassium (K) were quantified by optical emission spectrometry with excitation of the sample by inductively coupled plasma (technique indicated in analytical chemistry with the abbreviation ICP? OES). In addition, the analyzes of the dry matter were carried out, by drying in an oven at 105? C for the determination of the humidity? residual and organic matter by incineration in a muffle.

The data set? was then divided into two parts:

- a set of data used for calibration: (Table 1);

- a set of data used for validation: (Table 2).

The calibration and validation datasets were imported into a chemometric software capable of calculating the relationships between chemical composition (TKN, ammonia, P and K) and the 352 NIR absorbance (351) and conductivity values? electric (1). In particular, the 352 values were subject to standardization through Mean Centering. The predictive models were then calculated using PLS regression.

The predictive models were then applied to the validation data set and the predicted values were compared with the values of the laboratory analyzes by calculating the mean difference (BIAS), the standard error (RMSEP), the correlation squared (R <2> ), the ratio between standard deviation and error (RPD) and the coefficient of variation (CV).

The following two tables show the number of samples measured for each evaluated parameter, the minimum, average and maximum values of each set of measures (for each parameter) and the relative standard deviation. The measured parameters are the total nitrogen content (TKN), the ammonia nitrogen content (NH4-N), the phosphorus content measured as a quantity? of P2O5 and the potassium content measured as K2O.

Table 1. Composition of the calibration dataset

Table 2. Composition of the validation dataset

The results obtained were then compared with the TKN, NH4-N, P2O5 and K2O values obtained with the NIR measurement alone. As can be seen from tables 3, 4, 5 and 6, the combination of NIR and conductivity? electric with temperature compensation? improvement for all parameters with a reduction of the measurement error (Coefficient of Variation CV) of 35% for total nitrogen, 50% for ammonia nitrogen, 35% for P and 38% for K.

The statistical values reported in Tables 3-6 are:

RMSEP = Rooth Mean Square Error of Prediction;

R <2> = coefficient of determination;

RPD = ratio between standard deviation and error (RMSEP);

CV = coefficient of variation.

Table 3. Calibration Statistics for Total Nitrogen (TKN)

Table 4. Calibration statistics for ammoniacal nitrogen (NH4-N)

Table 5. Calibration statistics for phosphorus (P2O5)

Table 6. Calibration Statistics for Potassium (K2O)

As shown in tables 3-6, in which the statistical indicators of reliability are compared. of the measurements carried out with the NIR technique alone or with the combinatorial technique of the invention, the latter provides better results. accurate for all four parameters considered. In addition, the coupling of pi? different sensors is improved as it makes the system more? robust and reliable over time and leaves open the possibility? to use them individually or by merging the information according to the parameter and needs? operational.

Claims (9)

CLAIMS 1. Method for the analysis of manure and sewage produced in the zootechnical and agricultural sectors which consists of: - carry out a spectrophotometric analysis in the near infrared (NIR) of the sample; - at the same time, carry out at least a second measurement with a technique other than NIR on the same sample; - combine the results obtained with the NIR spectrophotometric analysis and with the second measurement using a data fusion technique, characterized by the fact that: the NIR spectrum is digitized and the prediction value is calculated by multiplying the absorbance values of the spectrum at each predetermined wavelength with previously determined calibration coefficients, obtaining a first numerical value X? from the following formula (1): in which: - n? the number of points of the spectrum taken into consideration; -? i? a calibration parameter valid for the i-th point of the spectrum; - me? the absorbance, reflectance or transmittance data obtained from the NIR spectrum for the i-th sampled point; And the X value? is combined with at least one value? n + 1m n + 1 obtained from at least one second measurement. Method according to claim 1, wherein the values used in formula (1) are obtained from calibration measurements carried out on known samples by chemical analysis. 3. Method according to claim 2 wherein the calibration values of the total nitrogen are obtained with the Kjeldahl method on the sample as is; the calibration values of ammonia nitrogen are obtained by means of the difference between the total and the organic one or by photometric measurement; and the calibration values for potassium and phosphorus are obtained by ICP-OES, Atomic Absorption or X-ray Fluorescence (XRF) analysis. Method according to any one of the preceding claims, wherein said second measure? choice between turbidimetry, a pH measurement and a conductivity measurement. 5. Method according to claim 4 wherein, when the second measure? a pH or conductivity measurement, and the measurement temperature? different from 25? C, said measure? compensated and brought back to the value corresponding to 25? C. 6. Method according to any one of the preceding claims, wherein the data fusion is carried out in the mode? ? low-level ?, concatenating the sample-by-sample data into a single matrix that has a number of rows equal to the number of samples analyzed and a number of columns equal to the number of variable signals measured by at least two different measuring instruments, and using then the data matrix cos? obtained as a basis for calculating a single predictive model. Method according to any one of claims 1 to 5, wherein the data fusion is carried out in mode? ? mid-level ?, preliminarily extracting the most? relevant from each sensor separately, by concatenating the values for these characteristics into a single matrix, and then using the data matrix cos? obtained as a basis for performing multivariate regressions. Method according to any one of claims 1 to 5, wherein the data fusion is carried out in the mode? ? high-level ?, by separately calculating a different regression for each dataset obtained from each measuring instrument and combining the results of each regression into a final model. Method according to any one of the preceding claims, in which the NIR calibrations are carried out using chemometric software and partial regressions with the least squares (PLS) method.