A method and system for detecting the three-dimensional morphology of aggregates based on two-dimensional particle parameters

Abstract

A method and system for detecting the three-dimensional morphology of aggregates based on two-dimensional particle parameters is disclosed, relating to the fields of aggregate detection and image processing technology. The method mainly involves acquiring two-dimensional morphological data of aggregates using an aggregate image measurement system (AIMS), acquiring three-dimensional morphological data of aggregates using a three-dimensional blue light scanner and visualization software (AVIZO), and then establishing a correlation mapping between two-dimensional and three-dimensional indicators using a gradient boosting regression tree method. This enables rapid and accurate prediction of the three-dimensional morphology of aggregates, facilitating refined management of mineral aggregate classification, identification, storage, and accurate retrieval, providing scientific support for engineering design, and promoting the intelligent and automated development of engineering management.

Description

Technical Field

[0001] This invention relates to the field of aggregate detection and image processing technology, and in particular to a method and system for detecting the three-dimensional morphology of aggregates based on two-dimensional particle parameters. Background Technology

[0002] In road construction, aggregates are the core component of asphalt mixtures, and their morphological characteristics directly affect road performance. Aggregate morphology includes particle shape, size, and surface properties, all of which collectively determine the workability, high-temperature stability, skid resistance, and fatigue resistance of the mixture. Therefore, accurately measuring and evaluating aggregate morphology is crucial for ensuring road quality and extending its service life.

[0003] The morphology of natural aggregates is inherently variable due to factors such as their source, lithology, and crushing method. This is especially true for recycled asphalt mixtures, where the morphology variability is further amplified by traffic loads, abrasion, and the cutting action of milling tools. The current technical specification for asphalt pavement construction, JTG F40-2004, clearly limits the content of needle-like and flaky particles, crushing value, and Los Angeles abrasion value. However, current techniques for assessing aggregate morphology, such as the vernier caliper method, flow time method, and void ratio method, suffer from low efficiency and high dispersion, failing to meet the needs of large-scale production and quality control, and making it difficult to guarantee data consistency and reliability. Summary of the Invention

[0004] One of the objectives of this invention is to address the shortcomings of the prior art by providing a method for detecting the three-dimensional morphology of aggregates based on two-dimensional particle parameters, thereby improving the efficiency and accuracy of aggregate morphology measurement and providing more reliable data support for the design, optimization, and on-site management of asphalt mixtures and aggregate particles.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: a method for detecting the three-dimensional morphology of aggregates based on two-dimensional particle parameters, comprising the following steps:

[0006] Step 1: Select aggregate particles within a predetermined particle size range and classify them into multiple particle size subcategories;

[0007] Step 2: Obtain two-dimensional morphological data of the aggregate particles using an image measurement system;

[0008] Step 3: Obtain the three-dimensional morphological data of the aggregate particles using three-dimensional scanning technology;

[0009] Step 4: Establish a correlation model between the two-dimensional morphological data and the three-dimensional morphological data using computer algorithms;

[0010] Step 5: Based on the difference between the predicted value and the actual value of the current association model, calculate the residual and use the residual as training data to train the association model;

[0011] Step 6: Input the two-dimensional morphology data of the aggregate particles into the trained correlation model to predict the corresponding three-dimensional morphology data.

[0012] Preferably, in step one, representative aggregate particles with different particle sizes of 2.36-16.0mm are selected, and the aggregates are divided into four categories according to the sieve aperture size: 2.36-4.75mm, 4.75-9.5mm, 9.5-13.2mm and 13.2-16.0mm, with each category containing at least 1,000 aggregate particle samples.

[0013] More preferably, in step two, the two-dimensional morphological data of each grade of aggregate is acquired using the aggregate image measurement system AIMS, and the two-dimensional morphological indicators such as aggregate particle angularity, micro-texture, particle shape, and flatness are quantitatively evaluated.

[0014] More preferably, in step three, a projection image of the three-dimensional shell of the aggregate is captured by a three-dimensional blue light scanner. The obtained image is imported into AVIZO software for morphological processing and binarization is achieved by threshold segmentation to identify the planar region containing the aggregate and the shell point cloud data. A stereolithography STL file containing the coarse aggregate shell point cloud data is generated according to the geometric node distance to create a virtual coarse aggregate particle shell of the discrete element model.

[0015] In steps two and three, using AIMS to acquire two-dimensional indicators has the advantages of simple operation, high efficiency, and fast calculation. Three-dimensional indicators acquired using a 3D blue light scanner and AVIZO software have the advantage of high precision.

[0016] More preferably, in step four, based on the known two-dimensional and three-dimensional morphological data of the aggregate sample data, the gradient boosting regression tree neural network algorithm is used to establish the correlation between the two-dimensional and three-dimensional indicators. First, a CART tree model is constructed as an initialization model, and the current model is used to predict the training data.

[0017] More preferably, in step five, based on the difference between the current model's predicted value and the actual value, the residual is calculated, and the residual is used as training data to construct a new CART tree. The prediction result of the new CART tree is then integrated into the model, and the predicted value is adjusted. By adding a new CART tree, the previous residual is minimized, so that the loss function of each iteration will develop in the direction of the negative gradient. Then, the above steps are repeated continuously, and a new CART tree is trained each time to fit the residual and combined with the existing model to gradually improve the performance of the model until the preset conditions are met and the aforementioned associated model is obtained.

[0018] The association model construction method in step five includes the following steps:

[0019] (1) Establish the initial model using the initial values ​​obtained through the following methods:

[0020]

[0021] In the formula: L(y i (γ) is the loss function;

[0022] (2) In the gradient boosting algorithm, the gradient descent method is used to reduce the residual. The residual in the gradient direction is represented as:

[0023]

[0024] In the formula: n is the number of iterations and n = 1, 2, ..., N;

[0025] (3) Obtain the initial constant model T by fitting the sample data. i (x i ;α n And calculate the parameter α using the least squares method. n for:

[0026]

[0027] (4) By minimizing the loss function, the current model weights are expressed as:

[0028]

[0029] (5) The model is iteratively updated as follows:

[0030] F n (x)=F n-1 (x)+γ n T n (x i ;α n ).

[0031] Preferably, in step six, all iteration results are summarized to obtain the final predicted value. A portion of the data that was not used in training is used as a validation set to evaluate the correlation model, including mean squared error and mean absolute error. Based on the evaluation results, the parameters of the correlation model are adjusted. When the performance of the correlation model reaches a satisfactory level, the corresponding three-dimensional morphological data is predicted by inputting the two-dimensional morphological data of the aggregate and using the trained correlation model.

[0032] In addition, the present invention also provides a three-dimensional morphology detection system for aggregates based on two-dimensional particle parameters, which includes:

[0033] The image measurement module is used to acquire two-dimensional morphological data of aggregate particles;

[0034] The 3D scanning module is used to acquire the 3D morphological data of aggregate particles;

[0035] The data processing module is used to process the two-dimensional morphological data and the three-dimensional morphological data;

[0036] The model building module is used to establish a correlation model between two-dimensional morphological data and three-dimensional morphological data.

[0037] The detection module is used to detect the three-dimensional morphological data of aggregate particles based on the association model.

[0038] The system operates using the detection method described above.

[0039] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0040] 1. The test results of this invention are accurate and reasonable. By combining digital image processing technology, the particle morphology characteristics of new and old aggregates can be accurately acquired, overcoming the limitations of traditional reliance on manual observation and enabling quantitative analysis of aggregate morphology. For example, parameters such as particle shape coefficient and angularity can be accurately measured, providing a reliable data foundation for in-depth research on aggregate morphology differences and the establishment of high-precision numerical simulation models for asphalt mixtures.

[0041] 2. The present invention provides real-time and rapid test results. By constructing a database of two-dimensional and three-dimensional aggregate indicators, the three-dimensional indicator data can be estimated simply by scanning the two-dimensional indicator data. This method enables rapid identification of aggregate particle morphology, which is helpful for the performance design and optimization of asphalt mixtures.

[0042] 3. This invention facilitates the management of aggregate particles on-site. Through efficient technical means, it achieves refined management of mineral aggregate classification, identification, on-site storage, and accurate retrieval, providing scientific and accurate technical support for engineering design and enabling the rapid development of engineering management towards intelligence and automation. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the gradient boosting regression tree method of the present invention;

[0044] Figure 2 This is a comparison chart of the measured and predicted three-dimensional aggregate index data in the test set of this invention. Detailed Implementation

[0045] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to embodiments and accompanying drawings. The content mentioned in the embodiments is not intended to limit the present invention.

[0046] A method for detecting the three-dimensional morphology of aggregates based on two-dimensional particle parameters includes the following steps:

[0047] (I) Collection and processing of aggregate samples.

[0048] Four thousand typical aggregate samples from different sources and batches were collected and pre-treated by washing and drying. First, the two-dimensional morphological characteristics of the aggregates were acquired using the AIMSII aggregate image measurement system to measure the size of coarse and fine aggregates, and to quantitatively evaluate two-dimensional morphological indicators such as aggregate edges, microtexture, and particle shape. Then, the morphological characteristics of the coarse aggregates were acquired using an ATOS Core 135 3D blue light scanner. Next, 3D images of the aggregates were acquired and digitized to obtain relevant 3D morphological characteristic indicators, ensuring that the images clearly and accurately reflect the particle morphology.

[0049] (II) Digital Image Processing and Parameter Calculation.

[0050] Two-dimensional images of various aggregate grades were acquired using the Artificial Image Measurement System (AIMS), and two-dimensional morphological indicators such as particle angularity, microtexture, particle shape, and flatness were quantitatively evaluated. Projected images of the three-dimensional shells of the aggregates, captured by a 3D blue light scanner, were imported into AVIZO software for morphological processing, and a stereolithography STL file containing point cloud data of the coarse aggregate shells was generated based on the geometric node distances.

[0051] The basic principles for measuring the two-dimensional morphological characteristics of aggregates, including the angularity index, shape index, texture index, and sphericity, are as follows:

[0052] (1) Angularity index

[0053] The gradient angle is calculated as the inclination of the gradient vector at the aggregate boundary points starting from the horizontal axis in the image. The gradient vector inclination is calculated as shown below. The gradient angle quantifies the change in the shape boundary of aggregate particles; a higher gradient value indicates a larger shape angle. The relative range of the gradient angle is 0 to 10000.

[0054]

[0055] In the formula, AI is the edge index; θ is the gradient vector angle of the edge point of the aggregate particle outline; n is the number of edge points of the aggregate particle outline; and i is the i-th point of the edge of the aggregate particle outline.

[0056] (2) Shape index

[0057] The more angular and protruding the aggregate, the higher its shape index value. Conversely, the more rounded and smooth the aggregate, the lower its shape index value.

[0058]

[0059] In the formula: R θ Let θ be the radius of the aggregate at angle θ; Δθ is the increment of the angle.

[0060] (3) Texture Index

[0061]

[0062] In the formula: D i The function is jdecomposition, with i-th level decomposition; N is the total number of points on the aggregate boundary profile; i = 1, 2, 3 are the texture directions; j is the wavelet coefficient exponent; x, y are the position coordinates of the points in the measured area.

[0063] (4) Sphericity

[0064] The relative proportion of sphericity ranges from 0 to 1, with a sphericity of 1 indicating that the aggregate is closer to a cube. Sphericity calculation:

[0065]

[0066] In the formula: d s ,d I ,d L These refer to the lengths of the short axis, middle axis, and long axis of the aggregate, respectively.

[0067] (III) Model building and training.

[0068] like Figure 1 As shown, a relational model is established using the gradient boosting regression tree neural network algorithm, and the model is trained and optimized to improve its prediction accuracy. The residuals are calculated based on the difference between the current model's predicted values ​​and the actual values.

[0069] Using the residuals as training data, a new CART tree is constructed. The predictions from the new CART tree are then incorporated into the model to adjust the predicted values. By adding new CART trees, the residuals from previous iterations are minimized, ensuring that the loss function in each iteration moves towards the negative gradient. This process is repeated continuously, training a new CART tree to fit the residuals each time and combining it with the existing model to gradually improve performance until the preset conditions are met. All iteration results are then summarized to obtain the final predicted value.

[0070] A CART tree model is constructed as the initialization model, and this model is used to predict the relevant three-dimensional data of aggregate particles. Its core content consists of data on four two-dimensional index variables: angular characteristics, microstructure, particle morphology, and flatness ratio of aggregate particles, used to predict the three-dimensional index data of the aggregate. To accurately evaluate the performance of the machine learning model, the dataset is typically divided into two subsets: a training set and a test set. First, approximately 80% of the original measurement data is selected as the training dataset, and 20% as the test dataset. The training set is mainly used to estimate the model parameters, while the test set is used to test the model's generalization performance. Second, a five-fold cross-validation method is used to fine-tune the model's hyperparameters, and then the GBRT model is fitted using the optimal hyperparameter configuration based on the training set. Next, the GBRT model can predict the three-dimensional index data of the aggregate based on long-term collected data of multiple variables related to aggregate particles. Finally, the prediction results of the GBRT model are compared and verified using the test set data. Therefore, by inputting the particle angular characteristics, microstructure, particle morphology, and flatness ratio from the test set into the GBRT model, the predicted results of the aggregate's three-dimensional index data can be obtained. Finally, by comparing the actual measured data with the predicted data, the model's performance on unknown data is evaluated. If the model's predictive performance meets the requirements, it can be used for practical deployment.

[0071] This invention utilizes the GBRT algorithm to construct a predictive model for three-dimensional aggregate index data. The model's input includes four two-dimensional index variables of aggregate particles: particle angularity, microtexture, particle shape, and flattening ratio. The entire computation process is performed in a Python 3.7 environment based on scikit-learn. Hyperparameter tuning plays a crucial role in the prediction accuracy of GBRT. The hyperparameters used in this model mainly include the number of CART trees, the shrinkage coefficient of each tree (ranging from 0 to 1), and the depth of the CART trees. GBRT has low sensitivity to parameters, requiring minimal hyperparameter tuning; as long as the tree depth is set appropriately and the number of iterations is sufficient, good prediction results can be achieved.

[0072] In the GBRT model, the actual aggregate particle index data are derived from stereolithography STL files generated from point cloud data obtained by 3D scanning of 4000 typical aggregate samples from different sources and batches. The 2D index variables are obtained by using the Artificial Image Measurement System (AIMS) to acquire 2D images of various aggregate grades and to quantitatively evaluate the angular characteristics, microstructure, particle morphology, and flatness ratio of the aggregate particles. Before training the model, the data needs to be processed, with key steps including removing outliers and missing values ​​from the measured data and related auxiliary data.

[0073] To evaluate the performance of the GBRT model, it is necessary to design an index to quantify its strengths and weaknesses. The coefficient of determination (R²) is needed. 2 R² and mean absolute error (MAE) were chosen as performance metrics for the model. Generally, a smaller MAE value indicates better predictive performance; while R²... 2 A higher value indicates a better predictive performance from the model. The following are the formulas for calculating the two performance metrics:

[0074]

[0075] In the formula: where y i,predicted y is the predicted value of the i-th sample; i,real It is the actual (measured) value corresponding to n samples;

[0076] In the GBRT model, the test set data is primarily used to validate the model's prediction results. The test set includes four two-dimensional index variables of aggregate particles: particle angularity, microtexture, particle morphology, and flatness ratio. Inputting this data into the trained GBRT model yields predictions of the aggregate's three-dimensional index data. This model is widely used for various typical aggregate particles with diameters ranging from 2.36 to 16.0 mm. Figure 2 As shown, the prediction results indicate that there is no significant difference in prediction accuracy between the 80% training set and the 20% test set, suggesting that the model does not suffer from overfitting. All performance metrics are very similar across different datasets.

[0077] This invention focuses on the differences in particle morphology between new and old aggregates and their impact on material properties. It innovatively integrates digital image processing, mathematical statistics, micromechanical simulation, and machine learning techniques to develop a highly efficient and accurate analytical method. The method first takes aggregate particles of different sizes (2.36-16.0 mm), sieves and processes the obtained asphalt aggregates, and uses AIMS to acquire the aggregate morphology and calculate its two-dimensional indices. A blue light 3D scanner is used to acquire the three-dimensional morphology of the aggregates, and three-dimensional shape indices such as fractal dimension, shape factor, and sphericity are calculated. A three-dimensional voxel model of the coarse aggregate particles is created using the visualization software AVIZO, point cloud data is acquired, and a three-dimensional virtual morphology database of recycled aggregates is accessed. The three-dimensional voxel model of the coarse aggregate particles is created using the image data software GOMInspect and the visualization software AVIZO. A gradient boosting regression tree neural network algorithm is used to establish the correlation between the two-dimensional and three-dimensional indices, thus enabling the estimation of three-dimensional indices using only two-dimensional index data.

[0078] To facilitate understanding by those skilled in the art of the improvements of this invention over the prior art, some of the accompanying drawings and descriptions have been simplified. The above embodiments are preferred implementations of this invention. In addition, this invention can be implemented in other ways. Any obvious substitutions without departing from the concept of this technical solution are within the protection scope of this invention.

Claims

1. A method for detecting the three-dimensional morphology of aggregates based on two-dimensional particle parameters, characterized in that, Includes the following steps: Step 1: Select aggregate particles within a predetermined particle size range and classify them into multiple particle size subcategories; Step 2: Obtain two-dimensional morphological data of the aggregate particles using an image measurement system; Step 3: Obtain the three-dimensional morphological data of the aggregate particles using three-dimensional scanning technology; Step 4: Establish a correlation model between the two-dimensional morphological data and the three-dimensional morphological data using computer algorithms. Based on the known aggregate sample data of the two-dimensional and three-dimensional morphological data, use the gradient boosting regression tree neural network algorithm to establish the correlation between the two-dimensional and three-dimensional indicators. First, construct a CART tree model as an initialization model, and use the current model to predict the training data. Step 5: Based on the difference between the predicted and actual values ​​of the current association model, calculate the residuals and use the residuals as training data to train the association model; the association model construction method includes the following steps: (1) Establish the initial model using the initial values ​​obtained through the following methods: ； In the formula: L(y i (γ) is the loss function; (2) In the gradient boosting algorithm, the gradient descent method is used to reduce the residual. The residual in the gradient direction is represented as: ； In the formula: n is the number of iterations and n = 1, 2, …, N; (3) Obtain the initial constant model T by fitting the sample data. n (x i ;α n And calculate the parameter α using the least squares method. n for: ； (4) By minimizing the loss function, the current model weights are expressed as: ； (5) The model is iteratively updated as follows: ； Step 6: Input the two-dimensional morphology data of the aggregate particles into the trained correlation model to predict the corresponding three-dimensional morphology data.

2. The method for detecting the three-dimensional morphology of aggregates based on two-dimensional particle parameters according to claim 1, characterized in that: In step one, representative aggregate particles with different particle sizes of 2.36-16.0 mm are selected, and the aggregates are divided into four categories according to the sieve aperture size: 2.36-4.75 mm, 4.75-9.5 mm, 9.5-13.2 mm and 13.2-16.0 mm, with each category containing at least 1,000 aggregate particle samples.

3. The method for detecting the three-dimensional morphology of aggregates based on two-dimensional particle parameters according to claim 1, characterized in that: In step two, the two-dimensional morphological data of each grade of aggregate is acquired using the aggregate image measurement system AIMS, and the two-dimensional morphological indicators are quantitatively evaluated. The two-dimensional morphological indicators include aggregate particle angularity, micro-texture, particle shape, and flatness.

4. The method for detecting the three-dimensional morphology of aggregates based on two-dimensional particle parameters according to claim 1, characterized in that: In step three, a 3D blue light scanner is used to capture the projected image of the 3D shell of the aggregate. The obtained image is then imported into AVIZO software for morphological processing. Binarization is achieved through threshold segmentation to identify the planar region containing the aggregate and the point cloud data of the shell. Based on the geometric node distance, a stereolithography STL file containing the point cloud data of the coarse aggregate shell is generated to create a virtual coarse aggregate particle shell for the discrete element model.

5. The method for detecting the three-dimensional morphology of aggregates based on two-dimensional particle parameters according to claim 1, characterized in that: In step five, based on the difference between the current model's predicted value and the actual value, the residual is calculated. The residual is used as training data to construct a new CART tree, and the prediction result of the new CART tree is integrated into the model. The predicted value is adjusted, and the previous residual is minimized by adding a new CART tree, so that the loss function of each iteration will develop in the direction of the negative gradient. Then, the above steps are repeated continuously, and a new CART tree is trained each time to fit the residual and combined with the existing model to gradually improve the performance of the model until the preset conditions are met and the aforementioned correlation model is obtained.

6. The method for detecting the three-dimensional morphology of aggregates based on two-dimensional particle parameters according to claim 1, characterized in that: In step six, all iteration results are summarized to obtain the final predicted value. A portion of the data that was not used in training is used as a validation set to evaluate the correlation model, including mean squared error and mean absolute error. Based on the evaluation results, the parameters of the correlation model are adjusted. When the performance of the correlation model reaches a satisfactory level, the corresponding three-dimensional morphological data is predicted by inputting the two-dimensional morphological data of the aggregate and using the trained correlation model.

7. A three-dimensional morphology detection system for aggregates based on two-dimensional particle parameters, characterized in that, include: The image measurement module is used to acquire two-dimensional morphological data of aggregate particles; The 3D scanning module is used to acquire the 3D morphological data of aggregate particles; The data processing module is used to process the two-dimensional morphological data and the three-dimensional morphological data; The model building module is used to establish a correlation model between two-dimensional morphological data and three-dimensional morphological data. The detection module is used to detect the three-dimensional morphological data of aggregate particles based on the association model; The detection method described in any one of claims 2-6 shall be used.

Images