CN114813632A - Spectroscopy measurement method and system for rock softening in water - Google Patents

Abstract

The present application relates to the technical field of using near-infrared light to test or analyze materials, and provides a method and system for measuring the intensity of water softening of rocks by spectroscopy. Axial compressive strength experiment, to obtain near-infrared spectral data and uniaxial compressive strength data of rock samples with different water contents; The relationship between the intensity and the characteristics of near-infrared spectroscopy; the obtained near-infrared spectral data is subjected to modeling sample screening and data enhancement, and a long-short-term memory fully convolutional network is used to establish a prediction model of rock water-intensity intensity. In this way, from the perspective of spectroscopy, a direct explanation method for the softening mechanism of soft rock in contact with water is established, and a non-destructive method for detecting rock strength is provided, which can achieve non-destructive, real-time, and 100% coverage, and can be used for on-site rock water strength measurement. .

Description

A spectroscopic measurement method and system for the softening of rock strength in contact with water

technical field

The present application relates to the technical field of testing or analyzing materials by using near-infrared light, and in particular, to a method and system for measuring the intensity of water softening of rocks by spectroscopy.

Background technique

Soft rock strength softening in contact with water is a common phenomenon in soft rock engineering practice. The influence and degree of action of water on rock is directly affected by the degree of rock saturation. When the minerals in the rock come into contact with water, a series of mechanical, physical and chemical effects will occur, which will change the mechanical properties of the rock and reduce the strength of the rock. In most cases, civil engineering disasters are caused by the interaction of water and rock, such as tunnel leakage, large deformation of soft rock in deep coal mines, etc. In addition, in the field of stone cultural relics and site protection, water is one of the key factors that induce various diseases in the surrounding rocks of caves in cultural relics. The content and distribution of water in rocks will cause varying degrees of damage to ancient cultural relics. Seriously affect the appearance and life of cultural relics.

The evaluation of softening strength has always been a difficult problem. At present, the research methods of softening mechanism, such as X-ray diffraction, electron microscope scanning, and pressure test, can only indirectly explain the reasons for strength softening, and cannot establish a direct relationship; The original state of the rock is destroyed, and only the intensity at a specific sampling point can be obtained, and the spatial and temporal distribution of the intensity cannot be obtained in a non-destructive, real-time, 100% covered manner.

Therefore, it is necessary to provide an improved technical solution for the deficiencies of the above-mentioned prior art.

SUMMARY OF THE INVENTION

The purpose of the present application is to provide a spectroscopic measurement method and system for softening the strength of rock in contact with water, so as to solve or alleviate the above-mentioned problems in the prior art.

In order to achieve the above purpose, the application provides the following technical solutions:

The application provides a spectroscopic measurement method for the softening of rock strength in contact with water, the method comprising:

Obtain the near-infrared spectral data and uniaxial compressive strength data of rock samples with different water contents;

The peak height of each absorption peak in the uniaxial compressive strength data and the near-infrared spectral number, and the peak area of each absorption peak in the uniaxial compressive strength data and the near-infrared spectral number, respectively. Perform nonlinear regression and fitting to determine the relationship between rock intensity and near-infrared spectral characteristics;

According to the relationship between the rock strength and the near-infrared spectral characteristics, extracting the near-infrared spectral data to obtain a training data set for a prediction model of rock strength in water;

According to the training data set, the prediction model of the rock's water-encounter strength is trained, and the trained rock's water-in-contact strength prediction model is obtained. strength is predicted.

Preferably, the near-infrared spectral data and uniaxial compressive strength data of rock samples with different water contents are obtained respectively, specifically:

Perform indoor near-infrared spectrum collection and uniaxial compressive strength test on the prepared rock samples to obtain the original near-infrared spectra of the rock samples with different water contents and the uniaxial compressive strength of the rock samples with different water contents strength data;

The original near-infrared spectra of the rock samples with different water contents are preprocessed by a combination of multi-point smoothing, first derivative and standard variable transformation, and the near-infrared spectral data of the rock samples with different water contents are obtained .

Preferably, the peak heights of the respective absorption peaks in the uniaxial compressive strength data and the near-infrared spectral numbers, and the uniaxial compressive strength data and the absorption peaks in the near-infrared spectral numbers The peak area of the peak is subjected to nonlinear regression and fitting in turn to determine the relationship between the rock intensity and the near-infrared spectral characteristics, specifically:

Taking the wavelength corresponding to each absorption peak in the near-infrared spectrum number as the center, determining a plurality of characteristic spectral segments of the near-infrared spectrum;

extracting the peak heights and peak areas of the plurality of characteristic spectral segments of the near-infrared spectrum;

The peak heights of the uniaxial compressive strength data and the plurality of characteristic spectral segments of the near-infrared spectrum, and the uniaxial compressive strength data and the plurality of characteristic spectral segments of the near-infrared spectrum, respectively. The peak area is subjected to nonlinear regression analysis and fitting in turn, and the relationship curve between the average peak heights of the characteristic spectral sections with different water contents and the average value of the rock strength, and a plurality of the features with different water contents are obtained correspondingly. The relationship curve between the average peak area of the spectrum segment and the average value of the rock intensity;

According to the relationship curve between the average peak heights of the plurality of characteristic spectral sections with different water contents and the average value of the rock strength, and the average peak area of the plurality of characteristic spectral sections with different water contents and the rock The relationship curve of the mean value of the intensity determines the relationship between the rock intensity and the near-infrared spectral feature.

Preferably, the fitting formula is:

In the formula, y represents the rock strength; x represents the peak height/peak area of the characteristic spectral section; a, b, and c are fitting coefficients.

Preferably, according to the relationship between the rock intensity and the near-infrared spectral characteristics, the near-infrared spectral data is extracted to obtain a training data set of the rock-water intensity prediction model, specifically:

The Mahalanobis distance between near-infrared spectra of rock samples with different saturations was screened based on the partial least squares method, and the screened near-infrared spectral data were obtained;

performing data enhancement on the near-infrared spectral data according to a plurality of characteristic spectral segments of the near-infrared spectrum, to obtain a plurality of training data sets for the prediction model of the rock-in-water intensity;

Correspondingly, the method further includes:

According to a plurality of training data sets of the rock water intensity prediction model, training the rock water intensity prediction model, and correspondingly obtain a plurality of trained rock water intensity prediction models;

According to the model accuracy of each of the rock water intensity prediction models in the multiple trained rock water intensity prediction models, determine an optimal rock water intensity prediction model, so as to predict the rock water intensity based on the optimal rock water intensity The model predicts the strength of rocks with different water contents.

Preferably, data enhancement is performed on the near-infrared spectral data according to a plurality of characteristic spectral segments of the near-infrared spectrum to obtain a plurality of training data sets for the prediction model of the strength of the rock in contact with water, specifically:

According to the absorbance value range corresponding to each characteristic spectral section of the near-infrared spectrum, the near-infrared spectral data corresponding to each of the characteristic spectral sections of the near-infrared spectrum in the plurality of the characteristic spectral sections of the near-infrared spectrum are respectively integrated into The increasing direction and decreasing direction of the absorbance value range are shifted by a preset threshold value, and a plurality of the training data sets are obtained correspondingly.

Preferably, the threshold parameter of the Mahalanobis distance between the near-infrared spectra of the rock samples with different saturation levels is 0.90.

Preferably, the prediction model of rock water contact strength is constructed based on a long short-term memory full convolutional neural network, and the rock water contact strength prediction model includes at least one long short-term memory module and at least one full convolution module.

Preferably, the rock water intensity prediction model is implemented through Keras library and Tensorflow.

The embodiment of the present application also provides a spectroscopic measurement system for softening the strength of rock in contact with water, including:

an acquisition unit, configured to acquire the near-infrared spectral data and uniaxial compressive strength data of rock samples with different water contents respectively;

The fitting unit is configured to compare the uniaxial compressive strength data and the peak height of each absorption peak in the near-infrared spectral number, and the uniaxial compressive strength data and each of the near-infrared spectral numbers. The peak area of the absorption peak is subjected to nonlinear regression and fitting in turn to determine the relationship between rock intensity and near-infrared spectral characteristics;

an extraction unit, configured to extract the near-infrared spectral data according to the relationship between the rock intensity and the near-infrared spectral characteristics, to obtain a training data set of a rock-water-intensity prediction model;

The prediction unit is configured to train the rock water intensity prediction model to be trained according to the training data set, and obtain the trained rock water intensity prediction model, so as to obtain the rock water intensity prediction model based on the trained rock water intensity prediction model. Predict the strength of rocks with different water contents.

Beneficial effects:

In this application, the near-infrared spectral data and uniaxial compressive strength data of rock samples with different water contents are obtained respectively; the uniaxial compressive strength data and the near-infrared spectral data are subjected to nonlinear regression and fitting to establish the relationship between rock strength and uniaxial compressive strength. The relationship between the near-infrared spectral features; then according to the relationship between the rock intensity and the near-infrared spectral features, the near-infrared spectral data is extracted to obtain a training data set for the prediction model of rock water intensity; based on the rock water intensity prediction model Predict the strength of rocks with different water contents. In this way, a strong correlation between spectral characteristics and uniaxial compressive strength is obtained through analysis. At the same time, a long short-term memory full convolution network method is used to establish a prediction model of rock water contact strength to reflect this correlation. The prediction model of rock strength in contact with water by near-infrared spectroscopy provides a new method for predicting rock strength.

Description of drawings

The accompanying drawings that form a part of the present application are used to provide further understanding of the present application, and the schematic embodiments and descriptions of the present application are used to explain the present application and do not constitute improper limitations on the present application. in:

1 is a schematic flowchart of a spectroscopic measurement method for softening of rock strength in contact with water provided according to some embodiments of the present application;

2 is a near-infrared spectrogram after preprocessing provided according to some embodiments of the present application;

3 is a schematic diagram of the failure situation of performing a uniaxial compressive strength test on sandstone samples with different saturation levels according to some embodiments of the present application;

4 is a schematic diagram of a stress-strain change curve obtained by performing a uniaxial compressive strength test on a sandstone sample with a saturation of 0% according to some embodiments of the present application;

5 is a schematic diagram of a stress-strain change curve obtained by performing a uniaxial compressive strength test on a sandstone sample with a saturation of 20% according to some embodiments of the present application;

6 is a schematic diagram of a stress-strain change curve obtained by performing a uniaxial compressive strength test on a sandstone sample with a saturation of 420% according to some embodiments of the present application;

7 is a schematic diagram of a stress-strain change curve obtained by performing a uniaxial compressive strength test on a sandstone sample with a saturation of 60% according to some embodiments of the present application;

8 is a schematic diagram of a stress-strain change curve obtained by performing a uniaxial compressive strength test on a sandstone sample with a saturation of 80% according to some embodiments of the present application;

9 is a schematic diagram of a stress-strain change curve obtained by performing a uniaxial compressive strength test on a sandstone sample with a saturation of 100% according to some embodiments of the present application;

FIG. 10 is the relationship between the peak height average value and uniaxial compressive strength of the characteristic spectrum band (peak R1) near the wavelength of 1400 nm provided according to some embodiments of the present application;

11 is the relationship between the average peak height of the characteristic spectral band (peak R2) near the wavelength of 1900 nm and the uniaxial compressive strength provided according to some embodiments of the present application;

FIG. 12 is the relationship between the average peak height of the characteristic spectral band (peak R3) near the wavelength of 2200 nm and the uniaxial compressive strength provided according to some embodiments of the present application;

FIG. 13 is the relationship between the average peak height of the characteristic spectral band (peak R4) near the wavelength of 2325 nm and the uniaxial compressive strength provided according to some embodiments of the present application;

14 shows the relationship between the peak area average value of the characteristic spectral band (peak R1 ) near the wavelength of 1400 nm and the uniaxial compressive strength provided according to some embodiments of the present application;

FIG. 15 is the relationship between the peak area average value and the uniaxial compressive strength of the characteristic spectral band (peak R2) near the wavelength of 1900 nm provided according to some embodiments of the present application;

16 is the relationship between the peak area average value and uniaxial compressive strength of the characteristic spectral band (peak R3) near the wavelength of 2200 nm provided according to some embodiments of the present application;

FIG. 17 is the relationship between the peak area average value and the uniaxial compressive strength of the characteristic spectral band (peak R4) near the wavelength of 2325 nm provided according to some embodiments of the present application;

18 is a columnar distribution diagram of Mahalanobis distance when the threshold parameter provided according to some embodiments of the present application is 0.95;

19 is a columnar distribution diagram of Mahalanobis distance when the threshold parameter provided according to some embodiments of the present application is 0.90;

FIG. 20 is a schematic structural diagram of a spectroscopic measurement system for softening the strength of rock in contact with water according to some embodiments of the present application.

Detailed ways

The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments. The various examples are provided by way of explanation of the application and do not limit the application. In fact, it will be apparent to those skilled in the art that modifications and variations can be made in the present application without departing from the scope or spirit of the application. For example, features illustrated or described as part of one embodiment can be used on another embodiment to yield yet another embodiment. Therefore, it is intended that this application cover such modifications and variations as come within the scope of the appended claims and their equivalents.

Exemplary method

FIG. 1 is a schematic flowchart of a spectroscopic method for measuring the softening of rock strength in contact with water provided according to some embodiments of the present application. As shown in FIG. 1 , the method includes:

Step S101 , respectively acquiring near-infrared spectral data and uniaxial compressive strength data of rock samples with different water contents.

In some embodiments, the near-infrared spectral data and uniaxial compressive strength data of rock samples with different water contents are respectively obtained, specifically: performing indoor near-infrared spectral collection and uniaxial compressive strength experiments on the prepared rock samples, The original near-infrared spectra of rock samples with different water contents and the uniaxial compressive strength data of rock samples with different water contents were obtained.

In the examples of this application, the near-infrared spectral data and uniaxial compressive strength data of rock samples with different water contents are obtained by performing indoor near-infrared spectral collection and uniaxial compressive strength experiments on the prepared sandstone samples, respectively. The process is:

A total of 6 sandstone samples with saturation of 100%, 80%, 60%, 40%, 20% and 0% were prepared respectively, and 5 sandstone samples were prepared for each saturation, a total of 30 samples, and the above 30 sandstone samples were prepared. The sandstone samples were subjected to near-infrared spectrum acquisition and uniaxial compressive strength experiments, and the original near-infrared spectral data and uniaxial compressive strength data of the sandstone samples during the experiment were obtained.

Since the near-infrared spectrum will include background noise and some irrelevant information caused by the surrounding environment such as natural light and fluorescent lamps during the experiment, multi-point smoothing (N Point Smooth, NPS for short), first derivative ( The method combining 1st Derivative (1st-Der for short) and Standard Normal Variate (SNV for short) is used to preprocess the original near-infrared spectra of rock samples with different water contents to obtain the Near-infrared spectroscopy data. In this way, by preprocessing the original near-infrared spectral data, the influence of background noise, sample particle size, surface scattering, baseline drift and other non-target factors in the original near-infrared spectral data on the target spectrum can be eliminated, and the original spectral data can be optimized. Spectral analysis and modeling accuracy.

FIG. 2 is a near-infrared spectrogram after preprocessing provided according to some embodiments of the present application. It can be seen from Figure 2 that there are four obvious absorption peaks for sandstones with different saturations, which are near the wavelength of 1400 nm, near the wavelength of 1900 nm, near the wavelength of 2200 nm and near the wavelength of 2325 nm. , especially at the wavelengths of 1400nm and 1900nm, the change with saturation is more obvious.

In another embodiment, uniaxial compressive strength experiments were performed on 30 sandstone samples, and uniaxial compressive strength data were collected. Among them, the uniaxial compressive strength data at least include: stress-strain change curves and physical parameters of sandstone samples with different saturation.

FIG. 3 is a schematic diagram of the damage of the samples after uniaxial compressive strength experiments are performed on sandstone samples with different saturation degrees provided according to some embodiments of the present application, wherein part (a) in FIG. 3 is a sample with a saturation of 0%. The failure of sandstone samples; (b) the failure of sandstone samples with 20% saturation; (c) the failure of sandstone samples with 40% saturation; (d) the failure of sandstone samples with 60% saturation Part (e) is the failure of the sandstone sample with 80% saturation; (f) part is the failure of the sandstone sample with 100% saturation. Figure 4, Figure 5, Figure 6, Figure 7, Figure 8, Figure 9 show the stress-strain change curves obtained by the uniaxial compressive strength test of sandstone samples with different saturation degrees.

In addition, the physical parameters of sandstone samples with different saturation include uniaxial compressive strength, average uniaxial compressive strength, softening coefficient, and percentage of strength drop. The experimental results of each physical parameter are shown in Table 1, and Table 1 is as follows :

Step S102, respectively perform nonlinear regression on the uniaxial compressive strength data and the peak height of each absorption peak in the near-infrared spectral number, and the uniaxial compressive strength data and the peak area of each absorption peak in the near-infrared spectral number, respectively. Fitting to determine the relationship between rock intensity and NIR spectral features.

In some embodiments, the uniaxial compressive strength data and the peak height of each absorption peak in the near-infrared spectral number, and the uniaxial compressive strength data and the peak area of each absorption peak in the near-infrared spectral number are sequentially analyzed. Linear regression and fitting are used to determine the relationship between rock intensity and near-infrared spectral characteristics, specifically: taking the wavelength corresponding to each absorption peak in the near-infrared spectral number as the center, to determine the characteristic spectral bands of multiple near-infrared spectra; The peak heights and peak areas of the characteristic spectral segments of the infrared spectrum; the peak heights of the uniaxial compressive strength data and the characteristic spectral segments of multiple near-infrared spectra, respectively, and the characteristics of the uniaxial compressive strength data and multiple near-infrared spectra, respectively The peak areas of the spectral sections are analyzed and fitted by nonlinear regression in turn, and the relationship curve between the average peak height of multiple characteristic spectral sections with different water contents and the average value of rock strength, as well as the average value of multiple characteristic spectral sections with different water contents, is obtained. The relationship curve between the average value of peak area and the average value of rock strength; the relationship curve between the average value of peak height and the average value of rock strength in multiple characteristic spectrum sections according to different water content, and the peak area of multiple characteristic spectrum sections with different water content The relationship curve between the mean value and the mean value of the rock intensity determines the relationship between the rock intensity and the near-infrared spectral characteristics.

In specific applications, according to the analysis of the near-infrared spectral data obtained in step S101, sandstones with different saturations mainly have four obvious absorption peaks, which are around 1400 nm, around 1900 nm, around 2200 nm and around 2325 nm, respectively. , firstly mark the four absorption peaks as peak R1, peak R2, peak R3 and peak R4, and take the wavelength corresponding to each absorption peak as the center to determine the characteristic spectral segments of multiple near-infrared spectra. Exemplarily, the wavelength interval [1350nm, 1450nm] is selected as the representative interval (that is, the characteristic spectrum) centered on the peak R1, the wavelength interval [1825nm, 1975nm] is selected as the representative interval centered on the peak R2, and the wavelength interval [ 2120nm, 2220nm] is the representative interval centered on the peak R3, and the wavelength interval [2255nm, 2380nm] is selected as the representative interval centered on the peak R4.

Then, extract the peak height and peak area of each characteristic spectral segment of the near-infrared spectrum, and compare the uniaxial compressive strength data with the peak heights of the characteristic spectral segments of multiple near-infrared spectra, as well as the uniaxial compressive strength The data and the peak areas of multiple characteristic spectral sections of the near-infrared spectrum are sequentially analyzed and fitted by nonlinear regression, and the relationship between the average peak height of multiple characteristic spectral sections with different water contents and the average value of rock intensity is obtained correspondingly. curve, and the relationship curve between the average peak area of multiple characteristic spectral sections with different water content and the average value of rock strength.

Among them, the fitting formula is as follows:

In the formula, y represents the rock strength; x represents the peak height/peak area of the characteristic spectrum segment; a, b, and c are the fitting coefficients.

The scatter diagram of the correlation between the average peak height and the average value of uniaxial compressive strength of each characteristic spectral section of sandstone samples with different saturation levels is shown in Figure 10 (a), Figure 11 (a), and Figure 12 ( As shown in part a) and part (a) of Figure 13, the correlation curve between the average peak height and the average value of uniaxial compressive strength of each characteristic spectral section of sandstone samples with different saturation levels is shown in part (b) of Figure 10 , Part (b) in Figure 11, part (b) in Figure 12, and part (b) in Figure 13. For the relationship between the rock intensity at peak R1 and peak R2 and the peak height of the characteristic spectrum in the near-infrared spectral characteristics, it can be seen from Figure 10 and Figure 11 that the average and uniaxial peak heights of the characteristic spectrum (peak R1) near the wavelength of 1400 nm The average value of compressive strength has a nonlinear negative correlation, and there is a good negative correlation between the average peak height of the characteristic spectrum band (peak R2) near the wavelength of 1900nm and the average value of uniaxial compressive strength. to approximate, approximate linear correlation, the correlation coefficient is 0.98. With the increase of saturation, the average value of uniaxial compressive strength gradually decreased, while the average value of peak height of each characteristic spectrum segment gradually increased. For the relationship between the rock intensity at peak R3 and peak R4 and the peak height of the characteristic spectrum in the near-infrared spectral characteristics, it can be seen from Figure 12 and Figure 13 that the average and uniaxial peak heights of the characteristic spectrum (peak R3) near the wavelength of 2200 nm The average value of compressive strength has a good positive correlation, and its correlation is approximated by an exponential function, showing a nonlinear positive correlation, and the correlation coefficient R ² is 0.89. With the increase of saturation, the average value of uniaxial compressive strength gradually decreases, and the average peak height of the characteristic spectrum band (peak R3) near the wavelength of 2200nm also gradually decreases; the peak height of the characteristic spectrum band (peak R4) near the wavelength of 2325nm There is also a good positive correlation between the average value and the average value of uniaxial compressive strength. With the increase of saturation, the average value of uniaxial compressive strength gradually decreases, while the average peak height of each characteristic spectrum segment gradually increases. The correlation is also approximately exponentially increasing, showing a nonlinear positive correlation with a correlation coefficient of 0.77. Compared with the absorption band near 2200 nm, its correlation with the uniaxial compressive strength is weaker.

From the above analysis, it can be seen that the correlation between the peak height of each characteristic spectrum segment and the uniaxial compressive strength is in descending order: the peak height of the characteristic spectrum segment R2 > the peak height of the characteristic spectrum segment R3 > the characteristic spectrum segment R4 The peak height of > the peak height of the characteristic spectrum band R1.

For the peak area of the characteristic spectral section of the near-infrared spectrum, the scatter diagram of the correlation between the average peak area of each characteristic spectral section of the sandstone samples with different saturation and the average value of the uniaxial compressive strength is shown in part (a), As shown in part (a) of Fig. 15, part (a) of Fig. 16, and part (a) of Fig. 17, the average value of peak area and the average value of uniaxial compressive strength of each characteristic spectral section of sandstone samples with different saturation The correlation curves of , are shown in part (b) in Figure 14, part (b) in Figure 15, part (b) in Figure 16, and part (b) in Figure 17. For the relationship between the rock intensity at peak R1 and peak R2 and the peak area of each characteristic spectrum segment, it can be seen from Figure 14 and Figure 15 that the average peak area of the characteristic spectrum segment near the wavelength of 1400 nm (peak R1) and the average uniaxial compressive strength There is a nonlinear negative correlation between the values, and there is a good negative correlation between the average value of the peak area of the characteristic spectrum band near the wavelength of 1900nm (peak R2) and the average value of the uniaxial compressive strength. Correlation, the correlation coefficient reached 0.96. With the increase of the saturation of the sandstone sample, the uniaxial compressive strength gradually decreases, and the peak area of the characteristic spectrum band near the wavelength of 1900 nm (peak R2) gradually increases. For the relationship between the rock intensity at peak R3 and peak R4 and the peak area of each characteristic spectrum segment, it can be seen from Figure 16 and Figure 17 that the average peak area of the characteristic spectrum segment near the wavelength of 2200nm (peak R3) and the average uniaxial compressive strength The value has a good positive correlation, and its correlation is approximately represented by an exponential function, showing a non-linear positive correlation, and the correlation coefficient R ² is 0.83. With the increase of the saturation of the sandstone sample, the uniaxial compressive strength gradually decreases, and the average value of the peak area of the characteristic spectrum band near the wavelength of 2200 nm (peak R3) also gradually decreases; the peak of the characteristic spectrum band near the wavelength of 2325 nm (peak R4) There is also a good positive correlation between the area average and the uniaxial compressive strength. decreases, the correlation is also approximately exponentially increasing, showing a non-linear positive correlation, and the correlation coefficient is 0.82.

From the above analysis, the correlation between peak area and uniaxial compressive strength is in descending order: peak area of characteristic spectrum R2 > peak area of characteristic spectrum R3 > peak area of characteristic spectrum R4 > characteristic Peak area of band R1.

To sum up, the peak area and peak height of each characteristic spectrum in the near-infrared spectrum are correlated with the uniaxial compressive strength, and the former is more obvious than the latter. The order of size is: characteristic spectrum segment R2>characteristic spectrum segment R3>characteristic spectrum segment R4>characteristic spectrum segment R1. In this way, from the perspective of spectroscopy, the process of rock strength softening is described, the softening mechanism is explained, and a direct method for explaining the strength softening of rocks in contact with water is established.

Step S103 , extracting the near-infrared spectral data according to the relationship between the rock intensity and the near-infrared spectral characteristics to obtain a training data set of a prediction model of the rock-in-water intensity.

In some embodiments, the near-infrared spectral data is extracted according to the relationship between the rock intensity and the near-infrared spectral characteristics to obtain a training data set for the prediction model of the rock-in-water intensity. The Mahalanobis distance between the near-infrared spectra of the rock samples is screened to obtain the filtered near-infrared spectral data; the near-infrared spectral data is enhanced according to the characteristic spectrum of multiple near-infrared spectra, and multiple rock encounters are obtained. A training dataset for a water intensity prediction model.

In the specific implementation, the PCA-MD program that comes with the software Vision is used to calculate the Mahalanobis distance of 1080 near-infrared spectral data with different saturations collected in the experiment, and the threshold parameter of the Mahalanobis distance is set. The Mahalanobis distance between the near-infrared spectra of the rock samples is screened, and the abnormal samples are removed from the samples of the training data set of the rock water intensity prediction model to obtain the filtered near-infrared spectral data. In practical applications, the threshold parameter of Mahalanobis distance can take different values, such as 0.95 or 0.90, and the corresponding screening results are shown in Figure 18 and Figure 19. Among them, when the threshold parameter is 0.95, a total of 78 abnormal samples are screened out, and when the threshold parameter is 0.90, a total of 96 abnormal samples are screened out. It can be seen that the fitting effect when the threshold parameter is 0.90 when using the partial least squares method for modeling Therefore, the 96 abnormal samples screened when the threshold parameter is 0.90 will be removed from the sample set of the training data set during modeling.

In one embodiment, data enhancement is performed on the near-infrared spectral data according to the characteristic spectral segments of multiple near-infrared spectra to obtain a plurality of training data sets for prediction models of rock strength in contact with water, specifically: according to each near-infrared spectral characteristic. The absorbance value domain range corresponding to the characteristic spectrum segment, the near-infrared spectral data corresponding to each near-infrared spectrum characteristic spectrum segment of the multiple near-infrared spectrum characteristic spectrum segments are respectively increased and decreased in the overall direction of the absorbance value range. The preset threshold size of the direction translation corresponds to obtaining multiple training data sets.

For the rock water intensity prediction model, the larger the data scale of the training data set, the better the training effect of the rock water intensity prediction model and the higher the prediction accuracy. In the embodiment of the present application, on the premise of not affecting the authenticity of the data samples of the near-infrared spectral data, each characteristic spectral segment of the near-infrared spectral data is enhanced to obtain multiple training data sets for the prediction model of rock water intensity .

In practical applications, firstly, according to the characteristic spectrum of the 4 near-infrared spectra, the near-infrared spectral data is divided into 4 groups, and the data set groups 1 to 4 are obtained; then, all the near-infrared spectral data are combined as a data set. Group 5, so as to obtain 5 training data sets of the rock water intensity prediction model. During data enhancement, according to the absorbance value range corresponding to each characteristic spectral segment of the near-infrared spectrum, the near-infrared spectral data corresponding to each characteristic spectral segment of the near-infrared spectrum are moved to the increasing direction of the absorbance range and Decrease the preset threshold size of the direction translation to obtain a training data set corresponding to each characteristic spectrum segment of the near-infrared spectrum.

The preset threshold value can be determined according to the relationship between the upper and lower limits of the absorbance range of the near-infrared spectral data in the characteristic spectral segment and the saturation category (percentage) of the sandstone sample. Taking the characteristic spectrum corresponding to peak R1 to peak R4 as an example, the wavelength range of the characteristic spectrum and the group of the training data set are shown in Table 2. Table 2 is as follows:

The steps of data enhancement are described in detail below by taking the first group of data sets (that is, the characteristic spectrum R1 [1350nm, 1450nm]) as an example. The absorbance range of the near-infrared spectral data of the sandstone samples of each saturation is shown in Table 3, and Table 3 is as follows:

Table 3 Absorbance value range table under different saturation

It can be seen from Table 3 that when the near-infrared spectral data as a whole moves to the increasing direction and the decreasing direction of the absorbance range (translation up and down) by 1e-11, the saturation category corresponding to the near-infrared spectral data is different. changes, i.e. the saturation percentage to which the NIR spectral data belongs does not change. After the near-infrared spectral data is translated twice as a whole, the sample size of the training data set will be tripled, and a total of 2952 sample data can be obtained.

Data enhancement is performed on the near-infrared spectral data of each characteristic spectral segment by the same method to obtain multiple training data sets, wherein each group corresponds to a training data set.

Step S104 , according to the training data set, train the rock water intensity prediction model to obtain the trained rock water intensity prediction model, and predict the intensity of rocks with different water contents based on the trained rock water intensity prediction model.

The strength of rocks with different water contents is predicted through the trained rock water strength prediction model, thereby providing a non-destructive method for detecting rock strength, evaluating softening strength, and achieving a non-destructive, real-time, 100% coverage method. It can be applied to the measurement of on-site rock strength in contact with water.

In one embodiment, the rock water contact intensity prediction model is constructed based on a long short-term memory full convolutional neural network LSTM-FCN model, and the rock water contact intensity prediction model includes at least one long short term memory module and at least one full convolution module.

Among them, the LSTM-FCN model is a Python open source code module in the GitHub software source code platform, including the long short-term memory LSTM module and the fully convolutional FCN module. The LSTM module is used to classify useful feature information in univariate long-term series data. Learning, the FCN module is used to assist in extracting and classifying features and improving model accuracy.

In some optional embodiments, the method further includes: training the rock water intensity prediction model according to multiple training data sets of the rock water intensity prediction model, to obtain a plurality of trained rock water intensity prediction models; According to the model accuracy of each rock water intensity prediction model in the multiple trained rock water intensity prediction models, the optimal rock water intensity prediction model is determined, and based on the best rock water intensity prediction model, the rocks with different water contents are analyzed strength is predicted.

In the specific model training process, still taking the first group of data sets as an example, first of all, a total of 2952 training data samples after data enhancement are divided by the ratio of training set: test set = 3:1. The generalization ability of the model is further divided into 1/4 of the training set as the verification set. The smaller the error of the verification set and the higher the correct rate, the stronger the generalization ability of the model. Finally, the number of samples in the training set is 1661, the number of samples in the test set is 738, and the number of samples in the validation set is 553. Then the training set, test set, and validation set are expressed by shape tensors (S, F, T), where S represents the number of samples in the training set/test set/validation set, the training set is 1661, and the test set is 738. The validation set is 553. F represents the number of variables processed at each time step, and the rock water contact strength prediction model in the embodiment of the present application is a univariate time series model, so the value of F is 1. T represents the longest time step in the sample, and the input of the rock-water-intensity prediction model is the length of the characteristic spectral section of the near-infrared absorption band of the sandstone samples with different saturation levels related to the uniaxial compressive strength data.

Before training the rock water intensity prediction model, set the training hyperparameters as follows: set the epochs to 2000, that is, use all the samples in the training set to train 2000 times; set the batchsize to 256, that is, take 256 in the training set for each training samples; normalize the data so that the data is mapped to the range of 0~1, which is conducive to the rapid convergence of the model to the optimal solution; the damage of the training set is used as the supervision value, and the model only saves the training during the training process Set the parameter weights with decreasing loss to ensure the effectiveness of model learning; the initial learning rate is set to 1e-4, the minimum learning efficiency is set to 1e-5, and the Adam algorithm is used, and the automatic adjustment learning during the training process is also used. rate method; if the supervision value does not decrease after 10 consecutive iterations, the learning rate will be reduced to help fast convergence.

In another optional embodiment, the prediction model of rock water contact strength is implemented by Keras library and Tensorflow.

In the specific implementation, the code of the rock water intensity prediction model is implemented through the Keras library and the Tensorflow backend. The five training data sets in Table 2 are respectively input into the rock water intensity prediction model for training, and the four main parameters of the model weights that are finally saved in the training process of the five rock water intensity prediction models are obtained correspondingly. The model training effect is shown in Table 4, and Table 4 is as follows:

After the training of the rock water intensity prediction model is completed, input the test set that did not participate in the training in the 5 data sets into each rock water intensity prediction model to calculate the accuracy of each model, as shown in Table 5. Table 5 is as follows:

Finally, according to the model accuracy of each rock water intensity prediction model in the multiple trained rock water intensity prediction models, the optimal rock water intensity prediction model is determined. The strength of the water volume rock is predicted.

As can be seen from Table 5, it is better to use the characteristic spectrum segments of peak R1+peak R2+peak R3+peak R4 to participate in the modeling than other characteristic spectrum segments to participate in the modeling alone, so the characteristic spectrum segment of peak R1+peak R2+peak R3+peak R4 is used. The LSTM-FCN strength evaluation model obtained by training is used as the best rock water strength prediction model, and the strength of rocks with different water contents is predicted based on the best rock water strength prediction model.

To sum up, when the long short-term memory fully convolutional network is used to establish the rock strength prediction model, the spectral data filtered by the Mahalanobis distance with a threshold parameter of 0.90 is used as the input sample set of the network, and the NPS+1st-Der+ The SNV composite processing method combines the peak R1 [1350nm, 1450nm] + peak R2 [1825nm, 1975nm] + peak R3 [2120nm, 2220nm] + peak R4 [2255nm, 2380nm] four characteristic spectral bands to jointly train the model, and obtain a near-infrared based model. The best rock water strength prediction model for sandstone sample strength prediction by spectrum, namely LSTM-FCN model, the accuracy of this model can reach 97.52%, and it can be used to predict the strength of in-situ rocks in practical applications.

In the examples of this application, the near-infrared spectral data and uniaxial compressive strength data of rock samples with different water contents are obtained respectively; and the uniaxial compressive strength data and the near-infrared spectral data are subjected to nonlinear regression and fitting to establish rock The relationship between intensity and near-infrared spectral characteristics, a direct explanation method for the softening mechanism of soft rock in contact with water is established from the perspective of spectroscopy; The training data set of the prediction model of rock strength in water is obtained; the strength of rocks with different water contents is predicted based on the prediction model of rock strength in contact with water. In this way, a strong correlation between spectral characteristics and uniaxial compressive strength is obtained through analysis, and a long-short-term memory full convolutional network method is used to establish a prediction model of rock water contact strength, forming a non-destructive, real-time, 100% coverage, full A new method for azimuthally assessing water strength of soft rocks.

Exemplary System

Embodiments of the present application further provide a spectroscopic measurement system for softening the strength of rock in contact with water. FIG. 20 is a schematic structural diagram of a spectroscopic measurement system for softening the strength of rock in contact with water provided according to some embodiments of the present application, as shown in FIG. 20 As shown, the system includes: an acquisition unit 2001 , a fitting unit 2002 , an extraction unit 2003 , and a prediction unit 2004 . in:

The acquiring unit 2001 is configured to acquire near-infrared spectral data and uniaxial compressive strength data of rock samples with different water contents, respectively.

The fitting unit 2002 is configured to compare the uniaxial compressive strength data and the peak height of each absorption peak in the near-infrared spectral number, and the uniaxial compressive strength data and the near-infrared spectral number. The peak area of each absorption peak was subjected to nonlinear regression and fitting in turn to determine the relationship between rock intensity and near-infrared spectral characteristics.

The extraction unit 2003 is configured to extract the near-infrared spectral data according to the relationship between the rock intensity and the near-infrared spectral characteristics, to obtain a training data set of a prediction model of the rock-in-water intensity.

The prediction unit 2004 is configured to train the rock water intensity prediction model according to the training data set, and obtain the trained rock water intensity prediction model, and use the rock water intensity prediction model based on the trained rock water intensity prediction model. The strength of rocks with different water contents is predicted.

The spectroscopic measurement system for the softening of rock strength in contact with water provided by the present application can realize the steps and procedures of any of the above-mentioned methods for measuring the strength of rock with water softening, and achieve the same technical effect. One more elaboration.

The above descriptions are only preferred embodiments of the present application, and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included within the protection scope of this application.

Claims (10)

1. a spectroscopic measurement method of rock strength softening in water, is characterized in that, comprises: Obtain the near-infrared spectral data and uniaxial compressive strength data of rock samples with different water contents; The peak height of each absorption peak in the uniaxial compressive strength data and the near-infrared spectral number, and the peak area of each absorption peak in the uniaxial compressive strength data and the near-infrared spectral number, respectively Perform nonlinear regression and fitting sequentially to determine the relationship between rock intensity and NIR spectral characteristics; According to the relationship between the rock strength and the near-infrared spectral characteristics, extracting the near-infrared spectral data to obtain a training data set for a prediction model of rock strength in water; According to the training data set, the prediction model of the rock's water-encounter strength is trained, and the trained rock's water-in-contact strength prediction model is obtained. strength is predicted. 2. the spectroscopic measurement method of the softening of rock strength in contact with water according to claim 1, is characterized in that, the described acquisition of near-infrared spectral data and uniaxial compressive strength data of rock samples with different water content respectively, is specifically: Perform indoor near-infrared spectrum collection and uniaxial compressive strength test on the prepared rock samples to obtain the original near-infrared spectra of the rock samples with different water contents and the uniaxial compressive strength of the rock samples with different water contents strength data; The original near-infrared spectra of the rock samples with different water contents are preprocessed by a combination of multi-point smoothing, first derivative and standard variable transformation, and the near-infrared spectral data of the rock samples with different water contents are obtained . 3. The spectroscopic measurement method of rock strength softening in contact with water according to claim 1, characterized in that, the peak height of each absorption peak in the uniaxial compressive strength data and the near-infrared spectral number is measured respectively. , and, the uniaxial compressive strength data and the peak area of each absorption peak in the near-infrared spectral data are followed by nonlinear regression and fitting to determine the relationship between the rock strength and the near-infrared spectral characteristics, specifically: : Taking the wavelength corresponding to each absorption peak in the near-infrared spectrum number as the center, determining a plurality of characteristic spectral segments of the near-infrared spectrum; extracting the peak heights and peak areas of the plurality of characteristic spectral segments of the near-infrared spectrum; The peak heights of the uniaxial compressive strength data and the plurality of characteristic spectral segments of the near-infrared spectrum, and the uniaxial compressive strength data and the plurality of characteristic spectral segments of the near-infrared spectrum, respectively. The peak area is subjected to nonlinear regression analysis and fitting in turn, and the relationship curve between the average peak heights of the characteristic spectral sections with different water contents and the average value of the rock strength, and a plurality of the features with different water contents are obtained correspondingly. The relationship curve between the average peak area of the spectrum segment and the average value of the rock intensity; According to the relationship curve between the average peak heights of the plurality of characteristic spectral sections with different water contents and the average value of the rock strength, and the average peak area of the plurality of characteristic spectral sections with different water contents and the rock The relationship curve of the average value of the intensity determines the relationship between the rock intensity and the near-infrared spectral feature. 4. the spectroscopic measurement method of rock-water strength softening according to claim 3, is characterized in that, the formula of described fitting is:

In the formula, y represents the rock strength; x represents the peak height/peak area of the characteristic spectral section; a, b, and c are fitting coefficients. 5. The spectroscopic measurement method for the softening of rock strength in contact with water according to claim 3 or 4, characterized in that, according to the relationship between the rock strength and the near-infrared spectral characteristics, the near-infrared spectral data Extraction to obtain the training data set of the rock water intensity prediction model, specifically: Based on the partial least squares method, the Mahalanobis distance between the near-infrared spectra of rock samples with different saturations was screened, and the screened near-infrared spectral data were obtained; performing data enhancement on the near-infrared spectral data according to a plurality of characteristic spectral segments of the near-infrared spectrum, to obtain a plurality of training data sets for the prediction model of the rock-in-water intensity; Correspondingly, the method further includes: According to a plurality of training data sets of the rock water intensity prediction model, training the rock water intensity prediction model, and correspondingly obtain a plurality of trained rock water intensity prediction models; According to the model accuracy of each of the rock water intensity prediction models in the multiple trained rock water intensity prediction models, an optimal rock water intensity prediction model is determined, so as to predict the rock water intensity based on the best rock water intensity prediction model. The model predicts the strength of rocks with different water contents. 6. The spectroscopic measurement method for softening the strength of rock in contact with water according to claim 5, wherein the data enhancement is performed on the near-infrared spectral data according to a plurality of characteristic spectral sections of the near-infrared spectrum, Obtain a plurality of training data sets of the rock water intensity prediction model, specifically: According to the absorbance value range corresponding to each characteristic spectral section of the near-infrared spectrum, the near-infrared spectral data corresponding to each of the characteristic spectral sections of the near-infrared spectrum in the plurality of the characteristic spectral sections of the near-infrared spectrum are respectively integrated into The increasing direction and decreasing direction of the absorbance value range are shifted by a preset threshold value, and a plurality of the training data sets are correspondingly obtained. 7 . The spectroscopic measurement method for the softening of rock strength in contact with water according to claim 5 , wherein the threshold parameter of the Mahalanobis distance between the near-infrared spectra of the rock samples with different saturation levels is 0.90. 8 . 8. The spectroscopic method for measuring the softening of rock water strength according to claim 1, wherein the rock water strength prediction model is constructed based on a long short-term memory full convolutional neural network, and the rock water strength The prediction model includes at least one long short-term memory module and at least one fully convolutional module. 9 . The spectroscopic measurement method for softening the strength of rock in contact with water according to claim 8 , wherein the prediction model for the strength of rock in contact with water is realized by Keras library and Tensorflow. 10 . 10. A spectroscopic measurement system for softening the strength of rock in contact with water, characterized in that it comprises: an acquisition unit, configured to acquire the near-infrared spectral data and uniaxial compressive strength data of rock samples with different water contents respectively; The fitting unit is configured to compare the uniaxial compressive strength data and the peak height of each absorption peak in the near-infrared spectral number, and the uniaxial compressive strength data and each of the near-infrared spectral numbers. The peak area of the absorption peak is subjected to nonlinear regression and fitting in turn to determine the relationship between rock intensity and near-infrared spectral characteristics; an extraction unit, configured to extract the near-infrared spectral data according to the relationship between the rock intensity and the near-infrared spectral characteristics, to obtain a training data set of a rock-water-intensity prediction model; The prediction unit is configured to train the rock water intensity prediction model to be trained according to the training data set, and obtain the trained rock water intensity prediction model, so as to obtain the rock water intensity prediction model based on the trained rock water intensity prediction model. Predict the strength of rocks with different water contents.

Images