Construction waste sample collection and analysis method

Abstract

This invention discloses a method for collecting and analyzing construction waste soil samples, belonging to the field of construction waste soil detection technology. The method includes: acquiring three-dimensional spatial distribution data of the construction waste soil dump and automatically planning a non-uniform sampling grid based on this data; using drilling parameters to invert the in-situ physical properties of the waste soil at each node in real time, and dynamically adjusting the drilling sampling depth according to the inversion results; rapidly determining the moisture content and performing laser particle size scanning on the collected waste soil samples to generate a digital fingerprint; performing fuzzy matching of the digital fingerprint with a pre-stored database of target performance parameters for improved soil, and automatically outputting a collaborative solidification operation instruction containing the baseline dosage of solidifying agent and recommended compaction process parameters based on the matching results. This method achieves fully automated processing from differentiated sampling and in-situ drilling sensing to intelligent solidification decision-making.

Description

Technical Field

[0001] This invention relates to the field of construction waste soil testing technology, specifically a method for collecting and analyzing construction waste soil samples. Background Technology

[0002] Scientific sampling and characteristic analysis of construction waste soil are prerequisites for the resource utilization of engineering spoil. Existing technical solutions for sampling waste soil dumps often rely on manual, experience-based pre-defined, evenly distributed sampling points, which fails to accurately reflect the non-uniform spatial distribution of waste soil characteristics in different areas within the dump. Evenly distributed sampling methods, when dealing with dumps with complex sources and varying filling histories, yield samples with insufficient representativeness and are prone to missing soil property variations in key local areas. Regarding drilling depth control, conventional techniques often employ fixed-depth sampling or rely on a single empirical threshold to determine whether to terminate drilling. When randomly distributed hard or weak interlayers exist in the dump fill, fixed-depth sampling schemes cannot respond to changes in strata in real time, leading to excessive sampling in ineffective soft layers or over-drilling even when reaching hard layers, reducing sample collection efficiency and the accuracy of obtaining samples from the target strata. Meanwhile, after obtaining waste soil samples, subsequent formulation of improved solutions often relies on manual review of charts and trial mixing. The entire process, from sampling to the formation of usable solidification instructions, is fragmented, lacking a technological means to automatically integrate spatial differentiation perception, in-situ real-time decision-making, and the final generation of improved solutions. This invention aims to solve the problems of how to autonomously adjust the density of sampling points based on the macroscopic morphological differences of construction waste soil dumps with uneven composition, and how to actively control the sampling layer based on real-time feedback of in-situ mechanical properties during drilling. Summary of the Invention

[0003] This invention provides a method for collecting and analyzing construction waste soil samples. Its purpose is to achieve adaptive densification and sparseness of the sampling grid according to the variation characteristics of the landfill terrain, real-time dynamic adjustment of the drilling sampling depth based on the in-situ waste soil strength, and direct generation of collaborative solidification instructions driven by the digital characterization results of the waste soil, so as to form an integrated processing chain from spatial difference perception to automatic output of improvement scheme.

[0004] The objective of this invention can be achieved through the following technical solutions:

[0005] This invention provides a method for collecting and analyzing construction waste soil samples, comprising the following steps:

[0006] Acquire three-dimensional spatial distribution data of construction waste dumps and automatically plan non-uniform sampling grids based on this three-dimensional spatial distribution data.

[0007] As a technical solution of the present invention, the method of acquiring three-dimensional spatial distribution data of construction waste dumps and automatically planning non-uniform sampling grids based on the three-dimensional spatial distribution data specifically involves: using a UAV equipped with a lidar to scan the construction waste dump to obtain point cloud data of the dump; triangulating the point cloud data to construct a digital elevation model of the dump; based on the elevation variation coefficient in the digital elevation model, areas with an elevation variation coefficient greater than a preset threshold are designated as dense sampling areas, and areas with an elevation variation coefficient less than or equal to the preset threshold are designated as sparse sampling areas; sampling nodes are arranged in the dense sampling areas at a first grid spacing, and sampling nodes are arranged in the sparse sampling areas at a second grid spacing, wherein the first grid spacing is less than the second grid spacing, thereby forming the non-uniform sampling grid. This method can reasonably allocate the sampling point density according to the adaptive differences in the dump morphology, improving sampling efficiency while ensuring comprehensive detection.

[0008] At each node of the non-uniform sampling grid, the in-situ physical properties of the excavated soil are inverted in real time using drilling parameters, and the drilling sampling depth is dynamically adjusted based on the inversion results of these in-situ physical properties.

[0009] As a technical solution of the present invention, at each node of the non-uniform sampling grid, the in-situ physical characteristics of the excavated soil are inverted in real time using drilling parameters, and the drilling sampling depth is dynamically adjusted according to the inversion results of the in-situ physical characteristics. Specifically, during the lowering of the drill rod, the torque value, drilling rate value, and vibration frequency value of the drill bit are continuously collected; the torque value, drilling rate value, and vibration frequency value are input into a pre-trained radial basis function neural network model, which outputs the predicted value of the unconfined compressive strength of the excavated soil at the corresponding depth; the predicted value of the unconfined compressive strength is compared with preset hard layer thresholds and soft layer thresholds; when the predicted value of the unconfined compressive strength is greater than the hard layer threshold at three consecutive sampling depth points, it is determined that the bearing hard layer has been reached and the drilling sampling is terminated; when the predicted value of the unconfined compressive strength is less than the soft layer threshold at five consecutive sampling depth points, it is determined that a weak interlayer has been entered and the drill rod is automatically raised to skip the weak interlayer and continue drilling. This method utilizes the mechanical response during drilling to invert soil strength in real time, allowing for preliminary perception of soil layer variations without the need to remove rock cores. It also enables immediate intervention in drilling depth based on strength changes, thereby avoiding invalid sampling layers and accurately stopping at the target bearing layer.

[0010] Preferably, the radial basis function neural network model employs an online incremental learning method during drilling, dynamically correcting the model parameters using measured values ​​of actual sampled physical properties. This allows the model's predictive ability to continuously optimize as the geological strata change at the engineering site, improving the matching degree between the in-situ inversion results and the actual soil properties.

[0011] The collected waste soil samples were subjected to rapid moisture content determination and particle size laser scanning to generate a digital fingerprint of the sample.

[0012] As a technical solution of the present invention, the method of rapidly determining the moisture content and performing laser particle size scanning on the collected waste soil sample to generate a digital fingerprint of the sample specifically involves: laying the collected waste soil sample flat in a detection chamber with a rotating disk; during the rotation of the rotating disk, an array-type near-infrared spectral probe is used to illuminate the waste soil sample from multiple angles and collect the reflectance spectrum; the moisture content value is calculated based on the intensity of the characteristic absorption peak of water molecules in the reflectance spectrum; simultaneously, a linear array laser scanner is used to perform a full-surface scan of the waste soil sample to obtain two-dimensional particle size distribution data; the moisture content value and the two-dimensional size distribution data are used to extract features, and the extracted moisture content feature vector and particle size feature vector are concatenated into a unique hash code sequence, which is the digital fingerprint. This process transforms the physical state of the soil into an irreversible coded identifier, providing a unified data carrier for subsequent rapid matching while retaining key characterization information.

[0013] The digital fingerprint is matched with a pre-stored database of various improved soil target performance parameters, and a collaborative curing operation instruction containing the standard dosage of curing agent and recommended compaction process parameters is automatically output based on the matching results.

[0014] As a technical solution of the present invention, the digital fingerprint is fuzzily matched with a pre-stored database of target performance parameters for various improved soils, and a collaborative curing operation instruction containing the reference dosage of curing agent and recommended compaction process parameters is automatically output based on the matching results. Specifically, this involves: calculating the weighted Euclidean distance between the digital fingerprint and the target performance digital fingerprint of each standard improved soil in the database; selecting the three standard improved soils with the smallest weighted Euclidean distance as candidate matching objects; extracting the reference dosage of curing agent and recommended compaction process parameters from each candidate matching object; calculating the confidence-weighted average of the three reference dosages of curing agent to obtain the initial curing agent dosage value; and voting on the three recommended compaction process parameters using a majority voting method to obtain the recommended number of compaction passes and the recommended type of compaction equipment. This solution searches for the historical mix design with the most similar engineering characteristics in a multi-dimensional neighborhood space, and suppresses the volatility of a single match by fusing multiple nearest neighbor results, thereby outputting curing and compaction operation parameters that are both representative and reliable.

[0015] Furthermore, the calculation of the weighted Euclidean distance between the digital fingerprint and the target performance digital fingerprint of each standard improved soil in the improved soil target performance parameter library specifically involves: parsing the generated digital fingerprint into a one-dimensional feature sequence composed of a water content feature vector and a particle size feature vector concatenated end-to-end; simultaneously reading the target performance digital fingerprint corresponding to each standard improved soil from the improved soil target performance parameter library and parsing it into a one-dimensional target feature sequence of the same length; for each feature component in this one-dimensional feature sequence, reading the weight coefficient corresponding to that component from a preset weight configuration table, where the weight... The coefficients are pre-calibrated based on the influence of the feature component on the curing agent dosage and compaction process. A difference operation is performed between the feature components at the same position in the one-dimensional feature sequence and the one-dimensional target feature sequence to obtain a difference sequence. Each difference in this difference sequence is first squared and then multiplied by the weight coefficient corresponding to that position to obtain a weighted squared difference sequence. All elements in this weighted squared difference sequence are summed to obtain a total value. The square root of this total value is then taken to obtain the weighted Euclidean distance between the digital fingerprint and the target performance digital fingerprint derived from this standard improvement. The weighted Euclidean distance can be expressed by the following formula:

[0016]

[0017] For weighted Euclidean distance, The length of the one-dimensional feature sequence. For the first The weight coefficients corresponding to each feature component The first in the one-dimensional feature sequence Each feature component The first one in the one-dimensional target feature sequence Each feature component. This distance metric strengthens the role of important features that are highly sensitive to mix design, while mitigating noise interference that may be introduced by weakly correlated features.

[0018] Furthermore, the process of using a majority vote to determine the recommended compaction passes and recommended compaction equipment types for the three compaction process recommended parameters is as follows: The compaction process recommended parameters include graded optimization parameters for different compaction degree targets of the subgrade; when the subgrade design compaction degree requirement is the first level, the first recommended compaction passes and the first recommended compaction equipment type for the first level are extracted from the three compaction process recommended parameters; the median of the three first recommended compaction passes is taken to obtain the target compaction passes for the first level; the mode of the three first recommended compaction equipment types is calculated, and the equipment type that appears most frequently is taken as the target compaction equipment for the first level; when the subgrade design compaction degree requirement is the second level, which is higher than the first level, the second recommended compaction passes and the second recommended compaction equipment type for the second level are extracted from the three compaction process recommended parameters, and the same median and mode calculation operations are performed to obtain the target compaction passes and the target compaction equipment for the second level. This method refines construction parameters independently according to compaction level, ensuring that different compaction levels can obtain a matching rolling process combination.

[0019] As a preferred embodiment of the above technical solution, during the dynamic adjustment of drilling and sampling depth, the in-situ physical properties obtained through inversion are associated and stored with the corresponding depth coordinates to generate a three-dimensional mechanical profile of the stockpile. This profile visually displays the three-dimensional distribution of strength parameters within the stockpile, providing a spatial reference for the zoned utilization of spoil and subsequent excavation.

[0020] As a further optimization of the above technical solution, the collaborative solidification operation command is output to the intelligent mixing plant and the unmanned road roller, realizing automatic batching of the solidifying agent and autonomous planning of the compaction path. This achieves closed-loop control of the operating equipment by directly driving the spoil analysis results, reducing manual proportioning and path setting steps, and improving the automation level of on-site construction.

[0021] The beneficial effects of this invention are:

[0022] Point cloud data of the stockpile was acquired using UAV LiDAR and a digital elevation model (DEM) was constructed. The elevation variation coefficient was calculated, and the area was divided into zones. Zones with an elevation variation coefficient greater than a preset threshold were designated as denser sampling zones with a denser first grid spacing. Zones with an elevation variation coefficient less than or equal to the preset threshold were designated as sparser sampling zones with a sparser second grid spacing. This non-uniform sampling grid correlated the density of sampling nodes with the intensity of the stockpile's topographic relief. In areas with significant surface morphology variations, the probability of differences in the composition and properties of the excavated soil was higher; denser sampling could capture detailed variations in the properties of the excavated soil in these areas. In areas with gentle surface morphology, the probability of uniform macroscopic properties of the excavated soil increased; sparse sampling could maintain representative sampling coverage while reducing ineffective duplicate sampling. During drill rod lowering, torque, drilling rate, and vibration frequency values ​​were continuously collected and input into a radial basis function neural network model, which output real-time predicted values ​​of the unconfined compressive strength of the excavated soil at the corresponding depth. The predicted value is compared with hard layer thresholds and soft layer thresholds. When the predicted value at three consecutive sampling depths is greater than the hard layer threshold, it is determined that a bearing hard layer has been reached, and drilling and sampling are terminated. When the predicted value at five consecutive sampling depths is less than the soft layer threshold, it is determined that a weak interlayer has been entered, and the drill pipe is automatically raised to skip this layer and continue drilling. Through multi-parameter fusion and drilling-while-drilling inversion, the zero-delay characteristic of the drilling dynamic response signal is utilized to actively identify and avoid sampling termination layers and weak spoil heap sections. This ensures that the sampling action and the vertical changes in formation mechanical properties are tracked in real time, replacing the blindness of fixed-depth sampling, and directionally obtaining representative spoil heap samples while reducing the interference of weak interlayers on sample batches. Attached Figure Description

[0023] The invention will now be further described with reference to the accompanying drawings.

[0024] Figure 1 This is a flowchart of the method for collecting and analyzing construction waste soil samples;

[0025] Figure 2 This is a flowchart of online prediction and drilling control for the unconfined compressive strength of excavated soil;

[0026] Figure 3 This is a flowchart of matching the dosage of soil stabilizer and compaction process parameters based on weighted Euclidean distance. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] See Figure 1 This invention provides a method for collecting and analyzing construction waste soil samples, including acquiring three-dimensional spatial distribution data of the construction waste soil dump, and automatically planning a non-uniform sampling grid based on the three-dimensional spatial distribution data; at each node of the non-uniform sampling grid, using drilling parameters to invert the in-situ physical properties of the waste soil in real time, and dynamically adjusting the drilling sampling depth according to the inversion results of the in-situ physical properties; rapidly determining the moisture content and performing laser scanning of the particle size of the collected waste soil samples to generate a digital fingerprint of the sample; performing fuzzy matching of the digital fingerprint with a pre-stored library of multiple modified soil target performance parameters, and automatically outputting a collaborative solidification operation instruction containing the reference dosage of solidifying agent and recommended compaction process parameters based on the matching results.

[0029] The drone, equipped with a lidar system, scans the construction waste disposal site. Flying along a pre-planned flight path, the lidar system emits laser pulses onto the surface of the site and receives the reflected echo signals. Combined with the position and attitude data recorded in real time by the drone's onboard positioning and attitude determination system, the three-dimensional spatial coordinates of each laser footprint are calculated, forming point cloud data of the disposal site.

[0030] A digital elevation model (DEM) of the stockyard was constructed by triangulating the point cloud data. The Delaunay triangulation algorithm was used to generate non-overlapping triangular patches using discrete points in the point cloud data as vertices, establishing an irregular triangular mesh on the stockyard surface. Based on this irregular triangular mesh, a regular grid structure DEM was generated through spatial interpolation. The planar resolution of the grid cells was set to 0.5 meters, and each cell stored the elevation value of its corresponding location.

[0031] The sampling area was divided based on the elevation variation coefficient in the digital elevation model (DEM). The DEM coverage area was divided into square grid cells with sides of 3 meters. For each grid cell, the elevation values ​​of all pixels within the cell were extracted, and the average elevation value was calculated. with standard deviation Elevation variation coefficient The calculation formula is:

[0032]

[0033] in: The elevation variation coefficient of the grid cell. The standard deviation of the elevation values ​​of all pixels within the grid cell. This is the average elevation value of all pixels within the grid cell.

[0034] The preset threshold is set to 0.2. For each grid cell, when the grid cell's... When the value is greater than 0.2, the area where the grid cell is located is designated as the encrypted sampling area; when the value of the grid cell is greater than 0.2, the area where the grid cell is located is designated as the encrypted sampling area. When the value is less than or equal to 0.2, the area where the grid cell is located is designated as a sparse sampling area.

[0035] Sampling nodes are deployed in the encrypted sampling area at a first grid spacing of 5 meters; in the sparse sampling area, sampling nodes are deployed at a second grid spacing of 10 meters. The first grid spacing is smaller than the second grid spacing, resulting in a higher sampling node density in the encrypted sampling area compared to the sparse sampling area. When deploying sampling nodes, square grids are generated in the encrypted sampling area at 5-meter intervals, with the grid intersections serving as sampling nodes; similarly, square grids are generated in the sparse sampling area at 10-meter intervals, with the grid intersections serving as sampling nodes. These grid nodes with different spacings together constitute a non-uniform sampling grid.

[0036] During the drill pipe lowering process, torque, drilling rate, and vibration frequency values ​​of the drill bit are continuously collected at a sampling frequency of 10 Hz using torque, displacement, and acceleration sensors installed on the top of the drill pipe. The torque value is obtained by measuring the torque transmitted by the drill pipe using the torque sensor, and the unit is kilonewton-meter; the drilling rate value is obtained by recording the amount of downward movement of the drill pipe per unit time using the displacement sensor, and the unit is meters per minute; the vibration frequency value is obtained by extracting the dominant frequency after performing a fast Fourier transform on the time-domain signal collected by the acceleration sensor, and the unit is Hz.

[0037] See Figure 2 The real-time collected torque, drilling rate, and vibration frequency values ​​are used as input vectors and fed into a pre-trained radial basis function (RBF) neural network model. The RBF neural network model consists of an input layer, one hidden layer, and an output layer. The input layer contains three neurons, corresponding to the torque, drilling rate, and vibration frequency values, respectively. The hidden layer contains 20 RBF neurons, each using a Gaussian kernel function as its activation function. The output of each hidden layer neuron is represented as ,in: For input vectors , This is the torque value. This is the drilling rate value. This represents the vibration frequency value. For the first The center vector of each hidden layer neuron has the same dimension as the input vector, and each component corresponds to the feature center value of torque, drilling rate and vibration frequency. For the first The width parameter of each hidden layer neuron controls the radial range of the Gaussian kernel function. The output layer contains one neuron, which uses a linear weighted summation method to output the predicted value of the unconfined compressive strength of the excavated soil. The calculation formula is:

[0038]

[0039] in: The value represents the predicted unconfined compressive strength of the excavated soil, in megapascals (MPa). For the connection in the output layer The weight coefficients of each hidden layer neuron and the output neuron; For the input vector, For the first The center vector of each hidden layer neuron For the first The width parameter of each hidden layer neuron.

[0040] The pre-training process of the radial basis function neural network model is as follows: Historical drilling data from construction waste dumps are collected to construct a training sample set. Each training sample contains torque, drilling rate, vibration frequency, and measured unconfined compressive strength values ​​obtained through laboratory unconfined compressive strength tests at a given depth. The input vectors of all training samples are clustered using the K-means clustering algorithm, with a cluster size of 20. The centers of the 20 clusters are then used as the center vectors of the 20 hidden layer neurons. The initial value is then used. For each cluster, the average distance from the samples within the cluster to the cluster center is calculated, and one-third of this average distance is used as the width parameter of the corresponding hidden layer neuron. The initial values ​​are then determined. After the parameters of the hidden layer neurons are determined, the least squares method is used to solve the system of linear equations to obtain the weight coefficients of the output layer. The initial values ​​are then used. Backpropagation is performed using all training samples for fine-tuning, with mean squared error as the loss function, a learning rate of 0.001, and 500 iterations to obtain the parameters of the pre-trained radial basis function neural network model.

[0041] During the drilling process, the real-time collected torque value, drilling rate value, and vibration frequency value are used to form an input vector. Input a pre-trained radial basis function neural network model, and the model outputs the predicted value of the unconfined compressive strength of the excavated soil at the corresponding depth. The threshold for hard layers is set at 15 MPa, and the threshold for soft layers is set at 2 MPa. The 15 MPa threshold for hard layers is based on the fact that when the unconfined compressive strength reaches 15 MPa, the soil layer, as a bearing layer, can meet the bearing capacity design requirements of general highway subgrades, and further sampling is unnecessary. The 2 MPa threshold for soft layers is based on the fact that when the unconfined compressive strength is below 2 MPa, the soil layer is considered a weak interlayer, and its mechanical properties do not meet the requirements for improvement and utilization; therefore, the collected samples have no reference value for subsequent design of the curing agent dosage.

[0042] The predicted unconfined compressive strength of the excavated soil generated at each sampling depth point Comparison with hard layer thresholds and soft layer thresholds. Predicted unconfined compressive strength of excavated soil at three consecutive sampling depths along the drilling depth direction. When all values ​​are greater than 15 MPa, it is determined that the drill bit has reached the bearing hard layer, and the control system issues a command to the drilling rig to terminate the drilling and sampling operation. The predicted unconfined compressive strength of the excavated soil at five consecutive sampling depths along the drilling depth direction is then considered. When all values ​​are less than 2 MPa, it is determined that the drill bit has entered a weak interlayer. The control system sends an automatic drill rod lifting command to the drilling rig. After lifting the drill rod to skip the weak interlayer section, the control system continues to drill downwards until it encounters a bearing hard layer again or reaches the maximum designed drilling depth.

[0043] During drilling, the radial basis function neural network model employs an online incremental learning approach, dynamically correcting model parameters using measured values ​​of actual sampled physical properties. After acquiring a soil sample at each depth, an unconfined compressive strength test is immediately performed on the sample to obtain the measured unconfined compressive strength value. The torque value, drilling rate value, vibration frequency value, and measured unconfined compressive strength value recorded at the corresponding depth are used as new training samples. When the cumulative number of new training samples reaches 10, a model parameter correction process is triggered. During correction, the 10 new training samples are merged with 100 historical training samples generated from the last 100 drilling operations to form the current incremental training set. Using the current model parameters as initial values, 50 backpropagation iterations are performed on the current incremental training set, with a learning rate set to 0.0005, updating the center vectors of the hidden layer neurons. Width parameters and output layer weight coefficients After the corrections are completed, the updated model parameters will be used to predict the unconfined compressive strength of the excavated soil at subsequent drilling depths.

[0044] In practice, the excavated soil samples collected during drilling are removed from the sampling tube and placed in a testing chamber equipped with a rotating disk. The testing chamber is a sealed metal box with a rotating disk installed at the center of its bottom surface. The rotating disk is driven by a stepper motor, and its top surface is a horizontal circular bearing surface. The excavated soil samples are evenly spread on the top surface of the rotating disk, with a thickness of 15 mm. The rotating disk rotates at a constant speed of 10 revolutions per minute.

[0045] During the rotation of the rotating disk, an array of near-infrared spectral probes illuminates the waste soil sample from multiple angles and collects its reflectance spectra. The array of near-infrared spectral probes consists of eight probes evenly distributed circumferentially along the inner wall of the detection chamber. The optical axis of each probe points towards the center of the rotating disk, and the vertical distance between each probe and the surface of the waste soil sample is 80 mm. Each near-infrared spectral probe incorporates a halogen tungsten lamp light source and a spectral acquisition module. The halogen tungsten lamp light emits near-infrared light with a wavelength range of 900 nm to 1700 nm, illuminating the surface of the waste soil sample at a 45-degree incident angle. The spectral acquisition module receives the diffuse reflected light from the waste soil sample in the vertical direction. Each rotation of the rotating disk results in one reflectance spectrum acquisition by each of the eight near-infrared spectral probes. The exposure time for each acquisition is 100 milliseconds, with a spectral resolution of 8 nm. A total of 80 reflectance spectra are obtained after 10 consecutive rotations.

[0046] Eighty reflectance spectra were averaged to obtain an average reflectance spectrum. The absorbance value at 1450 nm was extracted from this average reflectance spectrum. Water molecules exhibit a first-order overtone absorption peak at 1450 nm due to the stretching vibration of the OH bond. (Water content value) The calculation formula is:

[0047]

[0048] in: The moisture content of the waste soil sample is expressed as a percentage by mass. This represents the reflected light intensity at a wavelength of 1450 nm in the average reflectance spectrum. The value represents the reference reflected light intensity at a wavelength of 1100 nm, where 1100 nm is the reference wavelength for which water absorption is insensitive. The slope coefficient, with a value of -18.7, was obtained through a pre-established standard soil sample moisture content calibration experiment. The calibration experiment used five standard soil samples with known moisture contents, each covering a moisture content range of 5% to 35%. The slope of each standard soil sample was measured. and ,by A linear regression was performed with the x-axis representing the slope and the y-axis representing the known moisture content, to obtain the slope coefficient. The value is -18.7; The intercept coefficient is 2.4, which is also obtained from the linear regression results of the calibration experiment mentioned above.

[0049] Simultaneously, a linear laser scanner was used to perform a full-surface scan of the waste soil sample. The linear laser scanner was mounted on the top of the detection chamber, and its scanning beam width covered the diameter of the rotating disk. The scanner emitted a 650-nanometer wavelength linear laser beam, which projected onto the surface of the waste soil sample, forming a bright line. The CMOS image sensor built into the scanner captured an image of the bright line at a 45-degree angle, with an image resolution of 2048 x 1536 pixels. The scanner acquired one image frame for every 12 degrees of rotation of the rotating disk, collecting 30 images in a complete rotation. The contour information of the laser line was extracted from each image, and the height values ​​of various points on the surface of the waste soil sample were calculated using the principle of triangulation, forming a three-dimensional point cloud of the waste soil sample surface. The three-dimensional topographic point cloud is segmented into particles. The watershed algorithm is used to identify the boundary of each independent particle. The equivalent projected circle diameter of each independent particle is extracted as the particle size. The equivalent projected circle diameter values ​​of all particles are statistically analyzed, and a particle size distribution histogram is generated with an interval width of 0.1 mm. This particle size distribution histogram is the two-dimensional size distribution data of the particles.

[0050] Moisture content value Feature extraction was performed on the two-dimensional size distribution data. Moisture content values. The moisture content feature vector is directly used as the feature vector, which is a one-dimensional vector containing one element. For the two-dimensional size distribution data, the particle size values ​​corresponding to cumulative passing percentages of 10%, 30%, 50%, 60%, and 90% are calculated and denoted as follows: , , , and And calculate the curvature coefficient. Non-uniform coefficient ,Will , , , , The particle size feature vector consists of seven values: curvature coefficient and non-uniformity coefficient.

[0051] One element from the moisture content feature vector is concatenated with seven values ​​from the particle size feature vector, with the moisture content feature vector preceding the particle size feature vector, resulting in a feature sequence containing eight elements. Each element in the feature sequence is multiplied by a preset normalization coefficient, mapping each element value to an integer range of 0 to 255. The eight mapped integer values ​​are then concatenated sequentially, and the SHA-256 hash algorithm is used to perform a hash operation on the concatenated string, generating a 64-bit hexadecimal hash code sequence. This hash code sequence is the digital fingerprint of the waste soil sample.

[0052] See Figure 3 In practice, the generated digital fingerprint is parsed into a one-dimensional feature sequence composed of a moisture content feature vector and a particle size feature vector concatenated end-to-end. The moisture content feature vector contains one element, which is the moisture content value. The normalized integer value. The particle size feature vector contains 7 values, representing the particle size corresponding to a cumulative passing percentage of 10%. Particle size value corresponding to a cumulative pass rate of 30% Particle size value corresponding to a cumulative pass rate of 50% Particle size value corresponding to a cumulative pass rate of 60% Particle size value corresponding to a cumulative pass rate of 90% curvature coefficient Non-uniform coefficient The curvature coefficient The calculation expression is as follows Non-uniformity coefficient The calculation expression is as follows One element of the moisture content feature vector is concatenated sequentially with the seven values ​​of the particle size feature vector, resulting in a one-dimensional feature sequence of length 8, denoted as . ,in The normalized integer value of moisture content. for value, for value, for value, for value, for value, Curvature coefficient value, Non-uniform coefficient value.

[0053] The target performance digital fingerprint for each standard improved soil is retrieved from the target performance parameter database. This database is a pre-built relational database, where each record corresponds to complete information about a standard improved soil. Each standard improved soil has been verified through indoor mix proportion tests and field test sections to determine the required baseline dosage of curing agent and recommended compaction process parameters to achieve specific road performance indicators. The database stores the target performance digital fingerprint for each standard improved soil. The structure of the target performance digital fingerprint is consistent with the generated digital fingerprint structure, and it is also parsed into a one-dimensional target feature sequence of length 8, denoted as... , where: superscript This refers to the number of the standard improved soil in the target performance parameter library for improved soil. The value range is from 1 to the total number of standard improved soils in the target performance parameter library of improved soils. For the first The target moisture content of the standard improved soil is a normalized integer value. For the first The goal of standard improved soil value, For the first The goal of standard improved soil value, For the first The goal of standard improved soil value, For the first The goal of standard improved soil value, For the first The goal of standard improved soil value, For the first The target curvature coefficient value of the standard improved soil. For the first The target nonuniformity coefficient value of a standard improved soil.

[0054] For one-dimensional feature sequences For each feature component, the corresponding weight coefficient is read from a preset weight configuration table. The preset weight configuration table is an array containing 8 weight coefficients, denoted as... .in: The weighting coefficient corresponding to the moisture content characteristic component is set to 0.35. This value is based on the fact that moisture content has the greatest impact on the curing agent dosage and compaction process, and its weight is set to the highest among multiple influencing factors. for The weight coefficients corresponding to the feature components are set to 0.08. for The weight coefficients corresponding to the feature components are set to 0.08. for The weight coefficients corresponding to the feature components are set to 0.13. for The weight coefficients corresponding to the feature components are set to 0.13. for The weight coefficients corresponding to the feature components are set to 0.08. This is the weighting coefficient corresponding to the curvature coefficient feature component, with a value of 0.08; The weighting coefficients corresponding to the non-uniform coefficient characteristic components are set to 0.07. These weighting coefficients were obtained by averaging the eigenvectors of the judgment matrices filled out by 12 geotechnical engineering experts using the analytic hierarchy process (AHP). The sum of all weighting coefficients is 1. The weighting configuration table is stored as a static file in the control system's memory and is directly read and used each time fuzzy matching is performed.

[0055] One-dimensional feature sequence With the One-dimensional target feature sequence of standard improved soil The feature components at the same position are interpolated to obtain a difference sequence. The first feature component in the difference sequence is the first feature component in the difference sequence. The difference expression is as follows , The value of is from 1 to 8. Each difference in the difference sequence is first squared, then multiplied by the weight coefficient corresponding to that position to obtain a weighted squared difference sequence. The th value in the weighted squared difference sequence... The weighted squared difference expression is as follows: The weighted squared difference sequence is summed to obtain a total value. The square root of this sum is then taken; the resulting value is the generated digital fingerprint. The weighted Euclidean distance between the target performance digital fingerprints of the standard improved soil and the weighted Euclidean distance. The calculation formula is:

[0056]

[0057] in: For the generated digital fingerprint and the first The weighted Euclidean distance between the target performance digital fingerprints of various standard improved soils; For the weight configuration table, the first The weight coefficients corresponding to each feature component; One-dimensional feature sequence The Middle The values ​​of each feature component; For the first One-dimensional target feature sequence of standard improved soil The Middle The values ​​of each feature component; The position index of the feature component, with values ​​from 1 to 8; superscript This refers to the number of the standard improved soil in the target performance parameter library for improved soil.

[0058] Based on the weighted Euclidean distance calculation formula described above, the corresponding weighted Euclidean distance is calculated by iterating through each standard improved soil in the target performance parameter library of improved soil. Sort all weighted Euclidean distance values ​​from smallest to largest, and select the three standard improved soils with the smallest weighted Euclidean distance as candidate matching objects.

[0059] The baseline dosage of the curing agent and the recommended compaction process parameters for each of the three candidate matching objects were extracted from the target performance parameter library of improved soil. The baseline dosage of the curing agent is expressed as the mass of curing agent required to be added per cubic meter of excavated soil, in kilograms per cubic meter. The recommended compaction process parameters include tiered optimization parameters, which are stored separately for different subgrade compaction target levels. Each level corresponds to a set of recommended compaction passes and recommended compaction equipment types.

[0060] A confidence-weighted average was calculated for the three baseline curing agent dosages corresponding to the three candidate matching objects. The method for calculating the confidence-weighted average was as follows: the weighted Euclidean distances between the three candidate matching objects and the generated digital fingerprints were obtained, denoted as follows: , , ,in Take the reciprocal of the three weighted Euclidean distances, normalize the reciprocal, and use it as the confidence weight. Confidence weights of candidate matching objects The expression is The baseline dosage of the curing agent for three candidate matching objects was extracted from the target performance parameter library of improved soil, and denoted as follows: , , The unit is kilograms per cubic meter. Initial curing agent dosage value. The calculation expression is as follows .

[0061] The recommended compaction process parameters for the three candidate matching objects were determined by majority voting. In practice, the recommended compaction process parameters include graded optimization parameters for different compaction targets of the subgrade. The graded optimization parameters are divided into a first level and a second level according to the compaction requirements from low to high. The first level corresponds to the design requirement of 94% compaction of the subgrade, and the second level corresponds to the design requirement of 96% compaction of the subgrade.

[0062] When the subgrade design compaction requirement is Level 1, the recommended number of compaction passes and the recommended type of compaction equipment for Level 1 are extracted from the compaction process recommendations of the three candidate matching objects. The three candidate matching objects provide three recommended number of compaction passes; the median of these three numbers is taken as the target number of compaction passes for Level 1. The three candidate matching objects also provide three recommended types of compaction equipment; the mode of these three types is used, and the type that appears most frequently is taken as the target compaction equipment for Level 1. If the three equipment types are all different, the recommended compaction equipment type of the candidate matching object with the smallest weighted Euclidean distance is selected as the target compaction equipment.

[0063] When the subgrade design compaction requirement is level two, the recommended number of compaction passes and the recommended type of compaction equipment for level two are extracted from the compaction process recommendations of the three candidate matching objects. The same median operation is performed on the three recommended compaction passes to obtain the target number of compaction passes for level two. The same mode operation is performed on the three recommended compaction equipment types to obtain the target compaction equipment for level two.

[0064] In practice, during the process of lowering the drill rod and dynamically adjusting the drilling sampling depth, the predicted value of the unconfined compressive strength of the excavated soil obtained by inverting the radial basis function neural network model is used. And obtain the predicted value of the unconfined compressive strength of the excavated soil. The depth coordinates corresponding to each sampling depth are associated and stored. The depth coordinates are obtained from real-time readings of displacement sensors installed on the drill rod. The depth coordinates are zeroed out at the surface of the stockpile, with the vertically downward direction as the positive direction, and the unit is meters. Each sampling depth point generates a data record, and each data record contains three fields: depth coordinate value, predicted unconfined compressive strength of the excavated soil, and so on. and the moisture content of the excavated soil sample collected at that depth coordinate. Data records are sequentially written to the three-dimensional mechanical database of the stockyard, indexed by timestamps. The three-dimensional mechanical database of the stockyard is a real-time database with a spatial index structure, and the database is deployed in the field control industrial computer.

[0065] After drilling and sampling are completed at each sampling node of the non-uniform sampling grid, the three-dimensional mechanical database of the stockpile stores multiple data records along the full borehole depth at that sampling node. For a single sampling node, the depth coordinate value is plotted on the vertical axis, and the predicted value of the unconfined compressive strength of the excavated soil is plotted on the horizontal axis. Using the horizontal axis, plot the single-hole mechanical profile curve. Spatial interpolation is performed on the single-hole mechanical profile curves of all sampling nodes according to their planar coordinates. The interpolation algorithm uses the Kriging method, with a planar resolution of 0.5 meters and a depth-direction interpolation interval of 0.1 meters. After interpolation, a three-dimensional mechanical volume data of the stockpile is generated. This data is a three-dimensional matrix, and each voxel of the matrix stores the interpolated unconfined compressive strength of the excavated soil at that spatial location.

[0066] The 3D mechanical data of the stockpile is visualized and rendered to generate a 3D mechanical profile. The 3D mechanical profile displays the spatial distribution of the unconfined compressive strength of the excavated soil in pseudo-color. The unconfined compressive strength values ​​from 0 MPa to 20 MPa are mapped to a gradient color band from blue to red, with 20 color levels, each corresponding to a strength range of 1 MPa. The 3D mechanical profile supports cross-section display along any plane, and operators can drag the cross-section position on the control terminal to view the distribution of mechanical properties at a specified location within the stockpile in real time.

[0067] After the collaborative curing operation instruction is generated, it is output to the intelligent mixing plant and the unmanned road roller via a wireless local area network in the form of a structured data message. The structured data message adopts JSON format, and the message content includes the initial curing agent dosage value. The target number of compaction passes, the target type of compaction equipment, and the planar coordinates of each sampling node in the stockpile and the corresponding predicted values ​​of the unconfined compressive strength of the spoil. .

[0068] After receiving the structured data message, the intelligent mixing plant parses out the initial curing agent dosage value. The automatic batching system of the intelligent mixing plant is based on the initial curing agent dosage value. The system calculates the required mass of curing agent based on the total mass of the current batch of excavated soil and the metering accuracy of the curing agent silo, and controls the screw conveyor and electronic weighing scale to automatically weigh and feed the curing agent. The total mass of the excavated soil is obtained in real time by the weighing sensor at the feed inlet of the mixing plant, and the metering accuracy of the curing agent silo is 0.5 kg.

[0069] After receiving the structured data message, the unmanned road roller parses out the target number of compaction passes, the target compaction equipment type, the plane coordinates of each sampling node in the stockpile, and the corresponding predicted values ​​of the unconfined compressive strength of the spoil. The unmanned road roller predicts the unconfined compressive strength of the excavated soil based on the sampling nodes. The stockpile was divided into multiple compaction sub-zones, and the predicted value of the unconfined compressive strength of the excavated soil was obtained. Areas with a compressive strength below 5 MPa are considered weak foundation zones. The predicted unconfined compressive strength of the excavated soil is... The region between 5 MPa and 10 MPa is designated as the medium-base zone, and the predicted value of the unconfined compressive strength of the excavated soil is given. Areas with a subbase density higher than 10 MPa are designated as strong subbase zones. The unmanned road roller's path planning module, based on the sub-regional division of the stockpile compaction area, uses a covered path planning algorithm to generate compaction paths with a smooth transition between adjacent sub-regions as a constraint. The compaction paths start in weak subbase zones and end in strong subbase zones, with the overlap width of adjacent compaction strips being 30% of the wheel width. The unmanned road roller autonomously performs compaction operations according to the generated compaction paths and the calculated target number of compaction passes. During compaction, the roller uses its onboard global navigation satellite system to locate and record the compacted sections in real time, preventing under-compaction or over-compaction.

[0070] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for collecting and analyzing construction waste soil samples, characterized in that, Includes the following steps: Acquire three-dimensional spatial distribution data of construction waste dumps and automatically plan non-uniform sampling grids based on this three-dimensional spatial distribution data; At each node of the non-uniform sampling grid, the in-situ physical properties of the excavated soil are inverted in real time using the drilling parameters, and the drilling sampling depth is dynamically adjusted based on the inversion results of the in-situ physical properties. The collected waste soil samples were subjected to rapid moisture content determination and particle size laser scanning to generate a digital fingerprint of the sample. The digital fingerprint is matched with a pre-stored database of various improved soil target performance parameters, and a collaborative curing operation instruction containing the standard dosage of curing agent and recommended compaction process parameters is automatically output based on the matching results.

2. The method for collecting and analyzing construction waste soil samples according to claim 1, characterized in that, The specific steps for obtaining the three-dimensional spatial distribution data of the construction waste dump and automatically planning a non-uniform sampling grid based on the three-dimensional spatial distribution data are as follows: The construction waste dump was scanned by a drone equipped with a lidar to obtain point cloud data of the dump. The point cloud data was triangulated to construct a digital elevation model of the storage yard. Based on the elevation variation coefficient in the digital elevation model, areas with an elevation variation coefficient greater than a preset threshold are set as dense sampling areas, and areas with an elevation variation coefficient less than or equal to a preset threshold are set as sparse sampling areas. Sampling nodes are arranged at a first grid spacing in the encrypted sampling area and at a second grid spacing in the sparse sampling area, wherein the first grid spacing is smaller than the second grid spacing, thereby forming the non-uniform sampling grid.

3. The method for collecting and analyzing construction waste soil samples according to claim 1, characterized in that, The specific steps for inverting the in-situ physical properties of the excavated soil in real time at each node of the non-uniform sampling grid using drilling parameters, and dynamically adjusting the drilling sampling depth based on the inversion results of these in-situ physical properties, are as follows: During the drill pipe lowering process, the torque value, drilling rate value and vibration frequency value of the drill bit are continuously collected; The torque value, drilling rate value, and vibration frequency value are input into a pre-trained radial basis function neural network model, which outputs the predicted value of the unconfined compressive strength of the excavated soil at the corresponding depth. The predicted value of the unconfined compressive strength is compared with the preset hard layer threshold and soft layer threshold; When the predicted value of the unconfined compressive strength is greater than the hard layer threshold at three consecutive sampling depths, it is determined that the bearing hard layer has been reached and drilling sampling is terminated. When the predicted value of the unconfined compressive strength is less than the soft layer threshold at five consecutive sampling depths, it is determined that a weak interlayer has been entered, and the drill pipe is automatically raised to skip the weak interlayer and continue drilling.

4. The method for collecting and analyzing construction waste soil samples according to claim 3, characterized in that, The radial basis function neural network model adopts an online incremental learning method during the drilling process, and dynamically corrects the model parameters using the actual sampled physical characteristics measured values.

5. The method for collecting and analyzing construction waste soil samples according to claim 1, characterized in that, The specific steps for rapidly determining the moisture content and performing laser particle size scanning on the collected waste soil samples to generate the digital fingerprint of the samples are as follows: The collected waste soil samples were laid flat inside a testing chamber equipped with a rotating disk; During the rotation of the rotating disk, an array-type near-infrared spectroscopy probe is used to irradiate the waste soil sample from multiple angles and collect the reflectance spectrum. The water content value is calculated based on the intensity of the characteristic absorption peak of water molecules in the reflectance spectrum. Meanwhile, a linear laser scanner was used to scan the entire surface of the waste soil sample to obtain two-dimensional size distribution data of the particles; The moisture content value and the two-dimensional size distribution data are used to extract features. The extracted moisture content feature vector and particle size feature vector are then concatenated into a unique hash code sequence, which is the digital fingerprint.

6. The method for collecting and analyzing construction waste soil samples according to claim 1, characterized in that, The specific steps for performing fuzzy matching between the digital fingerprint and a pre-stored database of various improved soil target performance parameters, and automatically outputting a collaborative curing operation instruction containing the baseline dosage of curing agent and recommended compaction process parameters based on the matching results, are as follows: Calculate the weighted Euclidean distance between this digital fingerprint and the target performance digital fingerprint of each standard improved soil in the target performance parameter library; The three standard improved soils with the smallest weighted Euclidean distance were selected as candidate matching objects; Extract the baseline curing agent dosage and recommended compaction process parameters for each candidate match; The initial curing agent dosage value was obtained by calculating the confidence-weighted average of the baseline dosages of the three curing agents. The recommended parameters for the three compaction processes were determined by majority voting to obtain the recommended number of compaction passes and the recommended type of compaction equipment.

7. The method for collecting and analyzing construction waste soil samples according to claim 6, characterized in that, The specific steps for calculating the weighted Euclidean distance between the digital fingerprint and the target performance digital fingerprint of each standard improved soil in the improved soil target performance parameter library are as follows: The generated digital fingerprint is parsed into a one-dimensional feature sequence composed of a water content feature vector and a particle size feature vector concatenated end to end. At the same time, the target performance digital fingerprint corresponding to each standard improved soil is read from the improved soil target performance parameter library and parsed into a one-dimensional target feature sequence of the same length. For each feature component in the one-dimensional feature sequence, the weight coefficient corresponding to the component is read from the preset weight configuration table. The weight coefficient is pre-calibrated according to the influence of the feature component on the curing agent dosage and compaction process. The difference sequence is obtained by performing a difference operation between the feature components at the same position in the one-dimensional feature sequence and the one-dimensional target feature sequence. Squaring each difference in the difference sequence and then multiplying it by the weight coefficient corresponding to that position yields a weighted squared difference sequence. The sum of all elements in the weighted squared difference sequence is obtained by summing the total value. The square root of this sum is the weighted Euclidean distance between the digital fingerprint and the target performance digital fingerprint of the standard improved soil.

8. The method for collecting and analyzing construction waste soil samples according to claim 6, characterized in that, The specific steps for determining the recommended number of compaction passes and the recommended type of compaction equipment by majority voting on the recommended parameters for the three compaction processes are as follows: The recommended compaction process parameters include graded optimization parameters for different compaction targets of the roadbed; When the subgrade design compaction requirement is the first level, the first recommended number of compaction passes and the first recommended type of compaction equipment for the first level are extracted from the recommended parameters of the three compaction processes respectively. The median of the three recommended compaction passes is taken to obtain the target compaction pass number for the first level. The mode of the three first recommended compaction equipment types was statistically analyzed, and the equipment type that appeared most frequently was selected as the target compaction equipment of the first level. When the subgrade design compaction requirement is higher than the first level, the second recommended number of compaction passes and the second recommended type of compaction equipment for the second level are extracted from the three recommended compaction process parameters, and the same median and mode statistical operations are performed to obtain the target number of compaction passes and the target compaction equipment for the second level.

9. The method for collecting and analyzing construction waste soil samples according to claim 1, characterized in that, During the process of dynamically adjusting the drilling and sampling depth, the in-situ physical properties obtained by inversion are associated and stored with the corresponding depth coordinates to generate a three-dimensional mechanical profile of the stockpile.

10. A method for collecting and analyzing construction waste soil samples according to claim 1, characterized in that, The collaborative curing operation command is output to the intelligent mixing plant and the unmanned road roller, realizing automatic batching of curing agent and autonomous planning of compaction path.

Images