A method for measuring the volume of particles based on gap-corrected integration

Abstract

The application discloses a kind of granule volume measurement methods based on gap correction integration, it is related to granular medium measurement technical field, comprising the following steps: S1, different time point cloud acquisition and uniform modeling;S2, nonlinear gap correction integration model establishment;S3, gap feature extraction and input construction;S4, neural network gap correction factor prediction;S5, volume change calculation and dynamic application.The application adopts the above-mentioned kind of granule volume measurement method based on gap correction integration, by introducing time variable and nonlinear correction mapping in traditional voxel integration framework, constructs the space-time nonlinear model capable of representing the dynamic change of internal porosity of granular medium, in combination with neural network learns the measurement method of complex mapping relationship between internal porosity and volume change of granular medium, realizes high-precision, nonlinear, adaptive detection to the real volume of granular medium.

Description

Technical Field

[0001] This invention relates to the field of particulate media measurement technology, and in particular to a method for measuring particle volume based on void correction integral. Background Technology

[0002] In existing technologies, the volume measurement of particulate media is mainly based on apparent calculation methods of external geometry, including indirect estimation based on the geometric parameters of the storage chamber, 3D reconstruction based on depth sensor point clouds, and spatial integration methods based on voxel division. The first two types of methods usually assume that the packing is a dense solid and ignore the changes in porosity within the particles; while the voxel integration method can refine the spatial resolution, it still assumes that each voxel is uniformly filled and has a constant density. When the particulate medium undergoes compaction and loosening under the action of gravity, vibration disturbance, or changes in moisture content, its local porosity exhibits nonlinear, time-dependent, and location-dependent variation characteristics, leading to significant deviations between the measured volume and the true volume.

[0003] Furthermore, particulate media in storage or transportation environments are subject to the combined effects of wall friction, airflow disturbance, temperature and humidity coupling, and loading history, resulting in a complex nonlinear distribution of voids. External geometry or linear regression alone cannot adequately characterize the changes in the internal packing structure. At the same apparent height, different regions (such as the wall-adjacent region and the central region) exhibit significant differences in density. Traditional integral models cannot distinguish between these "appearingly consistent but density-different" states, thus limiting the accuracy and adaptability of volumetric measurements.

[0004] In summary, existing technologies for measuring the volume of particulate media mainly suffer from the following problems: First, the calculation models are still based on linear or constant density assumptions, which cannot reflect the nonlinear dynamic changes in porosity within the packing. Second, under complex working conditions such as compaction, loosening, and settling of particulate media, there are significant differences between apparent volume changes and actual volume changes. Third, traditional methods lack the ability to adaptively model environmental disturbances and historical loading states, and cannot achieve the characterization and learning of nonlinear characteristics. Summary of the Invention

[0005] The purpose of this invention is to provide a particle volume measurement method based on void correction integral, which overcomes the problem in the prior art that the particle medium volume measurement is based only on the apparent geometric volume calculation and ignores the dynamic changes of the porosity inside the packing with time and space, resulting in measurement errors.

[0006] To achieve the above objectives, the present invention provides a particle volume measurement method based on void correction integral, comprising the following steps:

[0007] S1. Acquire point cloud data at different times and model them in a unified manner. Collect point cloud data of particulate media accumulation at different time points and unify them to the same spatial reference coordinate system through global registration and local fine matching.

[0008] S2. Establish a nonlinear void correction integral model. Introduce a voxel-level nonlinear void correction mapping under the voxel partitioning integral framework to perform nonlinear correction of porosity for traditional voxel integral volume calculation, and describe the spatiotemporal variation characteristics of particulate media during the accumulation, compaction or settling process.

[0009] S3. Gap feature extraction and input construction: Extract local point density, depth variance, normal direction variance, and reflection intensity gap-related features from the point cloud at two time points, calculate temporal difference and perform neighborhood aggregation to form voxel-level input features;

[0010] S4. Neural network gap correction factor prediction: Input features into a neural network model that has been externally calibrated and trained, learn nonlinear mapping relationships and obtain voxel-level gap correction factor distribution, and adaptively model complex stacking states.

[0011] S5. Volume Change Calculation and Dynamic Application: Combining a nonlinear void correction integral model and correction factor, the actual volume change of particulate media at different times is calculated and used for dynamic monitoring and replenishment control in particulate material storage and transportation or agricultural machinery silo systems.

[0012] Preferably, in S1, a global registration algorithm based on geometric features is used to achieve coarse alignment, and a point-to-surface iterative nearest point algorithm is used for local fine alignment to minimize the residual between corresponding points. After registration, the point cloud data at each time step are unified in the same spatial reference frame.

[0013] Preferably, in S2, the space is divided into a set of voxels, and a nonlinear void correction mapping function is established for each voxel to characterize the nonlinear relationship between the internal void ratio of the voxel and its local characteristics, neighborhood state and time change; through this mapping, the solid filling assumption of the traditional voxel integral is corrected, so that the volume calculation result is converted from the apparent volume to the real volume.

[0014] Preferably, the voxel-level input features in S3 are obtained by the difference and aggregation of voxel features at two time points. The voxel features can also be fused with weight information, acoustic response, inertial measurement unit data, ambient temperature and humidity and / or flow signals, and then normalized as neural network input.

[0015] Preferably, the voxel features include one or more of the following: point density, depth or distance variance, normal direction variance, curvature index, and reflection intensity statistics.

[0016] Preferably, the neural network model in S4 is a spatiotemporally sparse convolutional network, a voxel graph neural network, and a multilayer perceptron structure combined with an attention mechanism to achieve nonlinear modeling and prediction of the gap correction factor; during the training phase, externally calibrated data is used as supervision labels, and the loss function combines robust regression error and physical consistency constraints, including nonnegativity, conservation, and spatial smoothness constraints.

[0017] Preferably, the supervision labels for S4 are obtained through weighing conversion, volume scale, liquid level measurement and / or water injection / discharge measurement; the training samples consist of voxel-level input features paired with corresponding calibrated volume changes or mass conversion values, and the network parameters are iteratively optimized through stochastic gradient descent or its adaptive variant until the training loss converges.

[0018] Preferably, in S5, the volume change calculation is based on a weighted integral of the nonlinear void correction integral model and the void correction factor predicted by the neural network to obtain the corrected volume change result. The result outputs a confidence level or uncertainty index for quality control, online calibration or subsequent dynamic compensation.

[0019] Therefore, this invention employs a particle volume measurement method based on void correction integrals. By introducing a time dimension and a nonlinear void correction mapping into the traditional voxel integral model, a nonlinear void correction integral model with spatiotemporal coupling characteristics is established. This reveals the dynamic evolution law of porosity during the volume change of granular media, overcoming the shortcomings of existing technologies that rely solely on apparent volume calculations and cannot reflect the dynamic changes in density and porosity within particle packs. By learning and approximating the nonlinear mapping relationship through a neural network, adaptive correction and high-precision dynamic measurement of the true volume change of granular media are achieved. This method is applicable to fertilizer silos, grain silos, seeder hoppers, and various granular material storage and transportation equipment, significantly improving the accuracy and intelligence level of volume measurement.

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0021] Figure 1 This is a flowchart of a particle volume measurement method based on void correction integral according to the present invention;

[0022] Figure 2 This is a schematic diagram illustrating the change of the nonlinear porosity coefficient of the present invention over time. (a) is... (b) is ;

[0023] Figure 3 This is a comparison diagram of the linearity assumption of this invention and the actual nonlinear void coefficient;

[0024] Figure 4The diagram shows the volume change measurement of the integral model at different times in this invention. (a) is the integral model at time t1, (b) is the integral model at time t2, and (c) is the volume change integral model at time t1-t2. Detailed Implementation

[0025] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0026] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0027] Example

[0028] In this invention, for ease of explanation, the relevant terms and symbols are defined as follows:

[0029] The "time point" mentioned in this invention is denoted as (It can also be extended to a maximum of times) ( ), representing the state of the particulate media accumulation at different times. At each time point, a set of raw point cloud data is acquired synchronously through multiple sensors, and after time synchronization and spatial calibration, they are fused to obtain the fused point cloud dataset for the corresponding time.

[0030] The "time period" mentioned in this invention refers to a time interval. This represents the overall volume change process of the particulate medium within a given time range. By comparing the fused point cloud or integral volume model at different times, the actual volume change of the particulate medium can be calculated.

[0031] To unify the spatial reference of multi-sensor data, this invention establishes a unified world coordinate system, denoted as . ;

[0032] The coordinate systems corresponding to each sensor (such as structured light camera, LiDAR, or depth camera) are denoted as follows: ,in Number the sensor.

[0033] The coordinate systems of each sensor can be obtained through external parameter calibration. To the world coordinate system The transformation matrix enables a unified representation of data from different perspectives under the same spatial reference.

[0034] The above symbol definitions provide a unified temporal and spatial description basis for modeling, integration, and volume calculation in subsequent steps S1 to S5.

[0035] Please see Figures 1-4 This invention provides a particle volume measurement method based on void correction integral, comprising the following steps:

[0036] S1. Point cloud acquisition and unified modeling at different times: Point cloud data of particulate media accumulation is collected at different time points and unified to the same spatial reference coordinate system through global registration and local fine matching.

[0037] Multiple spatially distributed three-dimensional sensor systems are arranged above or around the particulate media accumulation. These sensors can be structured light depth cameras, laser scanners, or phase-shifting projection devices, used to acquire spatial point cloud information of the particulate accumulation surface. At each time point... Within the system, all sensors synchronously acquire depth and reflection intensity data of the particle accumulation, generating individual sensor point clouds. After calibration of internal and external parameters and time synchronization processing, the point clouds from each sensor are uniformly transformed to the world coordinate system. This results in the formation of a complete and spatially consistent fused point cloud data at each point in time.

[0038] At different times After repeated acquisition, two or more frames of fused point clouds are obtained. To ensure the spatial registration accuracy between temporal point clouds, a global registration algorithm based on geometric features (e.g., based on...) is first employed. FPFH or SHOT Features RANSA C-matching is used to achieve coarse alignment, and then the nearest point is obtained by iterating through the point-to-surface area. ICP The algorithm performs local fine-tuning to minimize the residuals between corresponding points. After registration, the point cloud data at each time step are unified under the same spatial reference frame, forming a temporally continuous 3D point cloud sequence. Subsequently, the fused point cloud is filtered, denoised, clipped at boundaries, and voxelized to divide the continuous surface model into regular voxel units, establishing a voxel mesh model containing local attributes such as point density, normal vectors, and reflection intensity. This voxelized model is spatially consistent and temporally comparable, providing a unified and stable geometric input for the subsequent calculation of the nonlinear gap correction integral model.

[0039] S2. Establishment of a nonlinear void correction integral model: In the voxel-based integral calculation framework, a voxel-level nonlinear void correction mapping is introduced to perform nonlinear correction of the void ratio in the traditional voxel integral volume calculation, so as to reflect the nonlinear characteristics of the spatiotemporal variation of the internal void ratio of the particulate medium during the accumulation, compaction or settling process.

[0040] Under a unified coordinate system, the measurement space is divided into sets of voxels. A nonlinear porosity correction mapping function is established for each voxel to characterize the nonlinear relationship between the internal porosity of the voxel and its local characteristics, neighborhood state, and time variation. The geometric volume of each voxel unit is denoted as... Traditional volumetric calculation methods assume that the voxel is completely filled with material, ignoring the existence of internal voids. The calculation results can be expressed as follows: However, in actual particle packings, each voxel contains a porosity of varying degrees, the volume fraction of which can be defined as porosity. Therefore, this invention introduces a voxel-level void correction factor into the voxel integral calculation framework to correct the effective volume of each voxel, thereby obtaining a more realistic corrected volume:

[0041] ;

[0042] As the particulate medium evolves over time, the porosity within the voxel... Influenced by factors such as compaction, settlement, and vibration, it exhibits nonlinear time-varying characteristics, which can be expressed as: .in, For local accumulation rate field, The distribution of interparticle contact forces. Let be the energy function of the external disturbance. This is the set of parameters learned. This nonlinear mapping function can reflect the spatiotemporal variation characteristics of the internal porosity of particulate media during accumulation, compaction, or flow, realizing dynamic correction of traditional voxel integral volume calculations and providing a physically consistent porosity correction basis for subsequent volume change solutions.

[0043] S3. Gap feature extraction and input construction: Extract gap-related features such as local point density, depth variance, normal direction variance, and reflection intensity from point cloud data at two time points. Calculate temporal difference features and perform neighborhood aggregation to construct a voxel-level input feature vector with spatiotemporal correlation.

[0044] To achieve adaptive estimation of porosity in a nonlinear porosity correction integral model, this invention constructs a porosity feature extraction and input construction method based on temporal voxel difference, building upon a unified voxel partitioning framework. This step aims to extract statistical features and temporal difference information reflecting changes in the packing state of the granular medium from fused point clouds at different times, forming a voxel-level input feature vector to provide data support for subsequent neural network prediction of the porosity correction factor. Through this process, the porosity evolution trend can be implicitly derived from the dynamic changes of geometric voxels without directly measuring the porosity, achieving a cross-domain feature mapping from "geometric changes" to "physical properties."

[0045] S301, Voxel difference feature extraction and variable volume difference annotation.

[0046] Based on a unified coordinate system and voxel division, time points are... The fused point cloud is voxelized into a voxel set. To describe particulate media over a time period To extract temporal features from the internal stacking variation characteristics, this invention employs the voxel difference method. Voxel features include at least one or more of the following: point density, depth or distance variance, normal direction variance, curvature index, and reflection intensity statistics.

[0047] For each voxel Calculate basic statistics such as local point density, depth variance, normal direction variance, and reflection intensity: ；

[0048] ;

[0049] in, The number of points within a voxel. The direction of the normal. Let be the reflection intensity. The time-series changes of each statistic are defined as follows:

[0050] ;

[0051] ;

[0052] The difference in geometric volume change can be calculated based on the characteristics of time-series difference voxels:

[0053] ;

[0054] and with This serves as a label for subsequent neural network models. Instead of directly measuring voxel-level porosity at this stage, prior constraints based on geometric difference volume changes provide sample support for the network to learn nonlinear porosity correction factors.

[0055] S302, Integrate optional external features with data normalization.

[0056] To improve the model's adaptability to different particulate media and sensing conditions, this invention further integrates external feature information based on voxel difference features. External features include sensor reflection energy distribution, IMU Acceleration signal, weight information, acoustic response, inertial measurement unit data, vibration amplitude, temperature, humidity, and other auxiliary parameters are denoted as vectors. After fusion, voxel-level integrated input features are constructed:

[0057]

[0058] To ensure scale consistency across different features, the input features are standardized.

[0059]

[0060] in These are the mean and standard deviation of the features, respectively. The standardized feature set is obtained after normalization. It incorporates both spatiotemporal difference characteristics and integrates multi-source sensing information, providing a unified and stable input data foundation for neural network models.

[0061] S4. Neural network gap correction factor prediction: The voxel-level input features are input into a neural network model trained with externally calibrated data to obtain the voxel-level gap correction factor distribution; the neural network is used to learn nonlinear mapping relationships and adaptively model complex particle accumulation environments.

[0062] To achieve adaptive estimation of voxel-level porosity, this invention constructs a neural network model based on the voxel-level input features extracted in step S3. The network model is a spatiotemporally sparse convolutional network, a voxel graph neural network, and / or a multilayer perceptron structure combined with an attention mechanism. This network is used to establish a nonlinear mapping relationship between voxel difference features and porosity correction factors, enabling automatic inference of porosity changes in particle packings at different spatial locations and times.

[0063] S401, Construction and Training of the Gap Correction Factor Network.

[0064] This invention employs a lightweight neural network structure with voxel-level input features. As the input vector, the corresponding output is the gap correction factor. The network structure can consist of a multilayer perceptron (MLP). MLP ), one-dimensional convolutional networks or graph convolutional networks ( GCN It consists of ) and its basic mapping relationship is:

[0065] ;

[0066] in, The parameter set is The nonlinear mapping model.

[0067] During the training phase, the volume change difference obtained in S3 is used. As a monitoring signal, combined with the constraint relationship of the corrected volume:

[0068] ;

[0069] The network parameters are updated by minimizing the error between the predicted volume change and the geometric volume difference. Its loss function is defined as:

[0070]

[0071] in The smoothing regularization coefficients are used to constrain the continuity of correction factors among neighboring voxels. The network can be trained offline using experimental sample sets or simulation data, and can be fine-tuned online according to the packing characteristics of different particulate media.

[0072] During the training phase, externally calibrated data serves as supervisory labels. The loss function incorporates robust regression error and physical consistency constraints, including nonnegativity, conservation, and spatial smoothness constraints. Supervisory labels are obtained through weighing conversion, volume calibration, liquid level measurement, and / or water inflow / outflow measurement. Training samples consist of voxel-level input features paired with corresponding calibrated volume changes or mass conversion values. Network parameters are iteratively optimized using stochastic gradient descent or its adaptive variant until the training loss converges.

[0073] S402, Output of gap correction factor.

[0074] After the model training is completed, the point cloud data acquired in real time at different times are processed by S1–S3 to extract voxel-level input features, which are then input into the trained network model to obtain the distribution results of the gap correction factor for each voxel. .

[0075] To ensure stability, the output results can be smoothed over a neighborhood and weighted by confidence. The calculation method is as follows:

[0076]

[0077] in Other voxels within the voxel neighborhood, This is the gap correction factor for network prediction. Spatial Gaussian weights, The smoothing factor ultimately used for integral correction. ， , The smoothing coefficient. The gap correction factor after weighted smoothing. This will be input into the subsequent integration calculation step (S5) to correct the voxel-level true volume calculation, realizing the dynamic application of the nonlinear void correction integral model for particulate media.

[0078] S5. Volume Change Calculation and Dynamic Application: Combining the void correction integral model and void correction factor, the volume change of particulate media at different times is calculated and used for dynamic monitoring and replenishment control in particulate media storage and transportation or agricultural machinery silo systems.

[0079] After obtaining the porosity correction factor for each voxel, it is combined with the aforementioned voxel partitioning integral calculation framework to obtain the corrected volume distribution of the granular medium packing at different time points. By comparing different time points (e.g. By correcting the volume difference, the particulate medium can be controlled over a time period. Accurate measurement of internal true volume change. The volume change results of this invention not only reflect the geometric changes on the surface of the accumulation body, but also comprehensively consider the nonlinear evolution characteristics of the internal porosity, and can effectively distinguish the types of volume change caused by factors such as material flow, compaction or settling.

[0080] In practical applications, the volume change measurement method of this invention can be embedded in agricultural machinery silos, grain storage and transportation devices, or bulk material monitoring systems to achieve dynamic monitoring of the material loading and unloading process. When the system detects a volume change exceeding a set threshold, it can trigger automatic replenishment or discharge control, thereby achieving intelligent adjustment of the stacking state and material balance management. Furthermore, this method maintains high measurement accuracy and stability under complex lighting, vibration, and humidity conditions, making it suitable for dynamic volume measurement and automated control scenarios for various particulate media such as fertilizers, grains, and feed.

[0081] Therefore, this invention employs a particle volume measurement method based on void correction integrals. By introducing a time dimension and a nonlinear void correction mapping into the traditional voxel integral model, a nonlinear void correction integral model with spatiotemporal coupling characteristics is established. This model reveals the dynamic evolution law of porosity during the volume change of granular media, overcoming the shortcomings of existing technologies that rely solely on apparent volume calculations and cannot reflect the dynamic changes in density and porosity within particle packs. By learning and approximating the nonlinear mapping relationship through a neural network, adaptive correction and high-precision dynamic measurement of the true volume change of granular media are achieved. This method is applicable to fertilizer silos, grain silos, seeder hoppers, and various granular material storage and transportation equipment, significantly improving the accuracy and intelligence level of volume measurement.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for measuring particle volume based on void correction integral, characterized in that, Includes the following steps: S1. Acquire point cloud data at different times and model them in a unified manner. Collect point cloud data of particulate media accumulation at different time points and unify them to the same spatial reference coordinate system through global registration and local fine-tuning. S2. Establish a nonlinear void correction integral model. Introduce a voxel-level nonlinear void correction mapping under the voxel partitioning integral framework to perform nonlinear correction of porosity for traditional voxel integral volume calculation, and describe the spatiotemporal variation characteristics of particulate media during the accumulation, compaction or settling process. S3. Gap feature extraction and input construction: Extract local point density, depth variance, normal direction variance, and reflection intensity gap-related features from the point cloud at two time points, calculate temporal difference and perform neighborhood aggregation to form voxel-level input features; S4. Neural network gap correction factor prediction: Input features into a neural network model that has been externally calibrated and trained, learn nonlinear mapping relationships and obtain voxel-level gap correction factor distribution, and adaptively model complex stacking states. S5. Volume Change Calculation and Dynamic Application: Combining a nonlinear void correction integral model and correction factor, the actual volume change of particulate media at different times is calculated and used for dynamic monitoring and replenishment control in particulate material storage and transportation or agricultural machinery silo systems.

2. The particle volume measurement method based on void correction integral according to claim 1, characterized in that: In S1, a global registration algorithm based on geometric features is used to achieve coarse alignment, and a point-to-surface iterative nearest point algorithm is used for local fine alignment to minimize the residual between corresponding points. After registration, the point cloud data at each time step are unified in the same spatial reference frame.

3. The particle volume measurement method based on void correction integral according to claim 2, characterized in that: In S2, the space is divided into a set of voxels. A nonlinear void correction mapping function is established for each voxel to characterize the nonlinear relationship between the internal void ratio of the voxel and its local features, neighborhood state and time variation. This mapping corrects the solid-filling assumption of traditional voxel integrals, transforming the volume calculation result from apparent volume to actual volume.

4. The particle volume measurement method based on void correction integral according to claim 3, characterized in that: In S3, the voxel-level input features are obtained by the difference and aggregation of voxel features at two time points. The voxel features can also be fused with weight information, acoustic response, inertial measurement unit data, ambient temperature and humidity and / or flow signals, and then normalized as neural network input.

5. The particle volume measurement method based on void correction integral according to claim 4, characterized in that: Voxel characteristics include one or more of the following: point density, depth or distance variance, normal direction variance, curvature index, and reflection intensity statistics.

6. The particle volume measurement method based on void correction integral according to claim 5, characterized in that: The neural network model in S4 consists of a spatiotemporal sparse convolutional network, a voxel graph neural network, and a multilayer perceptron structure combined with an attention mechanism, which realizes nonlinear modeling and prediction of the gap correction factor. During the training phase, externally calibrated data is used as supervisory labels. The loss function combines robust regression error with physical consistency constraints, including nonnegativity, conservation, and spatial smoothness constraints.

7. The particle volume measurement method based on void correction integral according to claim 6, characterized in that: The supervision labels for S4 are obtained through weighing conversion, volume scale, liquid level measurement, and / or water injection / discharge measurement; the training samples consist of voxel-level input features paired with corresponding calibrated volume changes or mass conversion values, and the network parameters are iteratively optimized through stochastic gradient descent or its adaptive variant until the training loss converges.

8. The particle volume measurement method based on void correction integral according to claim 7, characterized in that: In S5, the volume change calculation is based on a weighted integral of a nonlinear void correction integral model and a void correction factor predicted by a neural network to obtain the corrected volume change result. The result outputs a confidence level or uncertainty index, which is used for quality control, online calibration or subsequent dynamic compensation.

Images