Automatic viscous gold concentrate sampling system based on intelligent sensing

Abstract

The invention relates to the technical field of viscous ore sampling, and discloses an automatic viscous gold concentrate sampling system based on intelligent sensing. The system comprises an ore flow state sensing module, an ore sample characteristic fusion module, a mineral topological graph module, a mineral characteristic screening module and a sampling anomaly detection module. The ore flow state sensing module collects rheological characteristic dynamic parameters through a multi-source sensing array, and an ore pulp rheological parameter matrix is output through heterogeneous data fusion. The ore sample characteristic fusion module is used for performing multi-dimensional characteristic decoupling on the ore sample characteristic, generating an ore pulp characteristic tensor and constructing a local association topology; the mineral topological graph module extracts key mineral feature nodes, calculates cross-dimension correlation strength, and forms a global rheological feature topological graph through integration. The mineral feature screening module analyzes the feature vector entropy value and screens out a mineral feature core entropy sequence in combination with the cross-dimensional correlation strength. And the sampling anomaly detection module triggers sampling anomaly judgment based on the sequence and generates a sampling action regulation and control instruction.

Description

Technical Field

[0001] The invention relates to the technical field of sticky ore sampling, in particular to an automatic sticky gold concentrate sampling system based on intelligent sensing. Background Art

[0002] In the processing of sticky gold concentrate, sampling is a crucial means of obtaining mineral information, the results of which are directly related to subsequent process adjustments and product quality control. Traditional sampling methods rely heavily on manual operation or simple mechanical sampling devices, which are difficult to adapt to the complex physical properties of sticky gold concentrate.

[0003] Viscous gold concentrate slurry has high viscosity and yield stress, making it prone to uneven velocity distribution and localized accumulation during flow. Manual sampling makes it difficult for operators to accurately determine the slurry's uniformity, and the selection of sampling points is often subjective, resulting in samples that fail to truly reflect the overall slurry composition and characteristics. Furthermore, the frequency and timeliness of manual sampling are significantly affected by human factors, making it difficult to meet the demand for real-time mineral quality information during continuous production. Existing mechanical sampling devices typically employ fixed sampling cycles and volumes, lacking the ability to dynamically perceive the rheological state of the slurry. When parameters such as slurry viscosity and flow rate change, fixed sampling patterns can easily lead to over- or under-sampling, or even slurry blockage of the sampling device. Furthermore, these devices are unable to deeply process the collected sample information, making it difficult to identify abnormal conditions in the slurry, significantly compromising the reliability of the sampling results. As mining production evolves toward intelligent and automated processes, higher requirements are placed on the accuracy and adaptability of sampling systems. Traditional sampling methods are increasingly limited by the complex and ever-changing characteristics of viscous gold concentrates, failing to provide stable and accurate basic data for mineral analysis. An automated sampling system is needed that can dynamically sense slurry conditions and intelligently adjust sampling strategies. Summary of the Invention

[0004] The purpose of the present invention is to provide an automatic sampling system for viscous gold concentrate based on intelligent sensing to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides an automatic sampling system for viscous gold concentrate based on intelligent sensing, the system comprising: an ore flow state sensing module configured to collect dynamic parameters of rheological characteristics of viscous gold concentrate slurry in real time through a multi-source sensor array, perform heterogeneous data fusion processing on the dynamic parameters of rheological characteristics, and output a slurry rheological parameter matrix; an ore sample characteristic fusion module, connected to the ore flow state perception module, configured to perform multi-dimensional feature decoupling on the slurry rheological parameter matrix to generate a slurry characteristic tensor, and construct a local correlation topology of each rheological dimension based on the slurry characteristic tensor; A mineral topology map module is configured to extract key mineral characteristic nodes of each rheological dimension, calculate the cross-dimensional correlation strength between the key mineral characteristic nodes, and integrate all local correlation topologies to form a global rheological characteristic topology map of the viscous gold concentrate; a mineral feature screening module configured to parse the characteristic vector entropy value of each rheological dimension from the slurry feature tensor, and screen out a core entropy sequence of mineral features by combining the cross-dimensional correlation strength in the global rheological feature topology map; The sampling anomaly detection module is connected to the mineral feature screening module and is configured to trigger sampling anomaly judgment of the sticky gold concentrate based on the mineral feature core entropy sequence and generate a sampling action control instruction.

[0006] Preferably, the dynamic parameters of rheological characteristics included in the ore flow state perception module specifically include slurry viscosity gradient data, mineral particle size distribution data, solid-liquid two-phase concentration data, pipeline flow velocity pulsation data and temperature field distribution data.

[0007] Preferably, the operation of performing multi-dimensional feature decoupling by the ore sample characteristic fusion module includes: performing noise suppression and data normalization processing on the slurry rheological parameter matrix to generate a standardized slurry rheological parameter matrix; Separating independent characteristic components of each dimension in the standardized slurry rheological parameter matrix through a characteristic decoupling engine; The independent feature components of all dimensions are aggregated to generate the slurry feature tensor.

[0008] Preferably, the process of calculating the cross-dimensional correlation strength by the mineral topology map module includes: Select any two key mineral characteristic nodes, and extract characteristic response sequences of the key mineral characteristic nodes in the slurry characteristic tensor; Calculating a morphological similarity measure of the feature response sequence using a dynamic time warping algorithm; The morphological similarity measure value is mapped to a cross-dimensional association strength weight coefficient.

[0009] Preferably, the operation of analyzing the characteristic vector entropy value by the mineral characteristic screening module includes: Obtain the mineral distribution dispersion coefficient for each rheological dimension; Calculate the statistical weight distribution of the internal feature components of the feature vector of each dimension; A quantified result of the characteristic vector entropy value is generated based on the mineral distribution dispersion coefficient and the statistical weight distribution.

[0010] Preferably, the step of screening the mineral characteristic core entropy sequence by the mineral characteristic screening module includes: Extracting cross-dimensional correlation strength weight coefficients of all key mineral characteristic nodes in the global rheological characteristic topology map; Perform weighted fusion of the quantized result of the feature vector entropy value and the corresponding cross-dimensional correlation strength weight coefficient; The highly significant mineral feature nodes are screened out by setting the entropy threshold; All highly significant mineral feature nodes are arranged in descending order of entropy value to generate the mineral feature core entropy sequence.

[0011] Preferably, the sampling anomaly detection module further comprises: a sampling execution feedback unit configured to collect motion state parameters of the sampling manipulator in real time and generate a motion state feedback signal; an instruction optimization unit, connected to the sampling execution feedback unit, configured to fuse the mineral characteristic core entropy sequence with the action state feedback signal to generate a dynamic compensation instruction; The instruction fusion unit is configured to superimpose and integrate the sampling action control instruction and the dynamic compensation instruction, and output a final sampling control signal.

[0012] Preferably, the action state feedback signal specifically includes displacement trajectory deviation data of the sampling robot arm, hydraulic execution pressure fluctuation data, sampling container filling rate data and mechanical vibration spectrum data.

[0013] Preferably, the system further comprises: an adaptive sampling execution mechanism, connected to the sampling anomaly detection module, configured to receive the sampling action control instruction and perform a physical sampling operation; The self-adaptive sampling actuator comprises a rotary cutting sampling head, a vacuum negative pressure stabilization unit, a multi-stage filtering and separation chamber, and a sample packaging manipulator.

[0014] Preferably, the operating parameters of the rotary cutting sampling head include cutting speed gradient, cutting angle offset, reciprocating motion frequency and cutting depth increment; The parameters adjusted in real time by the vacuum negative pressure stabilization unit include the negative pressure intensity threshold, the airflow pulsation suppression coefficient, and the air pressure balance value of the sample transmission pipeline; The separation parameters configured in the multi-stage filtration separation bin include a screen aperture combination sequence, a vibration separation amplitude curve, and a mineral residue cleaning cycle.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The ore flow state perception module utilizes a multi-source sensor array to collect dynamic parameters of the rheological properties of viscous gold concentrate slurry in real time, then performs heterogeneous data fusion processing to output a slurry rheological parameter matrix. This process comprehensively captures the various changes in the slurry's physical properties during flow, providing a more comprehensive and detailed understanding of the slurry's state and eliminating the one-sided perception of slurry state found in traditional sampling methods. The Ore Sample Characteristic Fusion module decouples the multi-dimensional features of the slurry rheological parameter matrix, generating a slurry characteristic tensor and constructing a local correlation topology for each rheological dimension. This decomposes complex slurry parameters into characteristic information of different dimensions, clearly presenting the internal correlations within each rheological dimension. This enables the system to understand slurry characteristics from multiple perspectives, providing a richer basis for subsequent mineral quality analysis. The mineral topology map module extracts key mineral characteristic nodes, calculates cross-dimensional correlation strengths, and integrates them to form a global rheological characteristic topology map. By constructing this global map, the system can grasp the inherent connections between different rheological dimensions, breaking the isolation of information in each dimension and providing a more systematic and comprehensive understanding of the overall characteristics of the slurry. The mineral feature screening module analyzes the eigenvector entropy values ​​from the slurry feature tensor and, combined with cross-dimensional correlation strength, selects a core entropy sequence of mineral features. This screening process eliminates redundant information and focuses on core features that are critical to slurry properties, making the system more targeted when processing information and reducing the impact of invalid data on subsequent analysis. The sampling anomaly detection module triggers sampling anomaly determinations and generates control instructions based on the core entropy sequence of mineral characteristics. When the slurry exhibits an abnormal state, the system promptly identifies and adjusts the sampling action to align it with the actual state of the slurry, preventing the adverse effects of abnormal slurry on sampling results and ensuring that the sampling process is more closely aligned with the dynamic changes of the slurry. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a working principle diagram of the automatic sampling system for viscous gold concentrate based on intelligent sensing according to the present invention; Figure 2 Flowchart for feature decoupling of the ore sample characteristic fusion module; Figure 3 Flowchart for the calculation of cross-dimensional correlation strength for the mineral topology map module; Figure 4 Flowchart for screening of core entropy sequences for mineral characteristics; Figure 5 Flowchart for optimizing the sampling anomaly detection module instructions. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0018] See also Figure 1 The present invention provides an automatic sampling system for viscous gold concentrate based on intelligent sensing, the system comprising: The ore flow state perception module collects the dynamic parameters of the rheological characteristics of the viscous gold concentrate slurry in real time through a multi-source sensor array, performs heterogeneous data fusion processing on the dynamic parameters of the rheological characteristics, and outputs the slurry rheological parameter matrix.

[0019] The ore sample characteristic fusion module is connected to the ore flow state perception module to perform multi-dimensional feature decoupling on the slurry rheological parameter matrix, generate a slurry characteristic tensor, and construct the local correlation topology of each rheological dimension based on the slurry characteristic tensor.

[0020] The mineral topology map module extracts key mineral characteristic nodes of each rheological dimension, calculates the cross-dimensional correlation strength between key mineral characteristic nodes, and integrates all local correlation topologies to form a global rheological characteristic topology map of viscous gold concentrate.

[0021] The mineral feature screening module parses the characteristic vector entropy value of each rheological dimension from the slurry feature tensor, and combines the cross-dimensional correlation strength in the global rheological feature topology map to screen out the core entropy sequence of mineral features.

[0022] The sampling anomaly detection module is connected to the mineral feature screening module, triggers the sampling anomaly judgment of the sticky gold concentrate based on the mineral feature core entropy sequence, and generates sampling action control instructions.

[0023] Example 1: See Figure 2 The dynamic parameters of rheological characteristics processed by the ore flow state perception module specifically include slurry viscosity gradient data, mineral particle size distribution data, solid-liquid two-phase concentration data, pipeline flow velocity pulsation data, and temperature field distribution data. The slurry viscosity gradient data is collected by multiple groups of viscosity sensors arranged in layers along the radial direction of the pipeline. Each group of sensors contains three probes of different depths, which can synchronously obtain real-time viscosity values ​​at different radial positions within the pipeline cross section, and then calculate the viscosity gradient change by the viscosity difference between adjacent positions. These sensors adopt a high-frequency response design, and the sampling interval can be automatically adjusted according to the slurry flow state. The sampling interval is extended when the slurry flow is stable, and the sampling interval is shortened when the flow fluctuates violently, so as to ensure the timeliness and integrity of the data.

[0024] Mineral particle size distribution data is collected by a laser particle size analyzer installed alongside the pipeline. This device transmits a laser beam of a specific wavelength through the slurry flow. Particles of varying sizes in the slurry scatter the laser light at varying angles. An array of photodetectors captures the intensity and angular distribution of the scattered light. After signal conversion and data processing, the size distribution range of the mineral particles and the percentage of particles in each size range are determined. To minimize the impact of bubbles in the slurry on the measurement, both the laser transmitter and receiver are equipped with bubble recognition and rejection algorithms. When a bubble is detected passing through the measurement area, the data for that period is automatically marked and corrected in subsequent processing.

[0025] Solid-liquid concentration data is acquired using microwave transmission measurement. A microwave transmitter and receiver are installed on either side of the pipeline. The microwave signal emitted by the transmitter passes through the slurry and is captured by the receiver. Due to the different microwave absorption and reflection characteristics of solid particles and liquid in the slurry, the solid-liquid concentration in the slurry is calculated by measuring the attenuation and phase change of the microwave signal and combining it with a pre-established concentration calibration model. The microwave frequency of the measurement device can be adjusted according to the viscosity of the slurry. For high-viscosity slurries, a lower frequency is used to enhance signal penetration, while for low-viscosity slurries, a higher frequency is used to improve measurement accuracy.

[0026] Pipeline flow velocity pulsation data is collected using an ultrasonic Doppler flowmeter. This instrument features two pairs of ultrasonic transducers mounted on the outer wall of the pipeline: one pair transmits ultrasonic waves, and the other receives echo signals reflected by particles in the slurry. The Doppler effect is then used to calculate the slurry's flow velocity. Simultaneously, the instrument's built-in high-frequency sampling module records changes in flow velocity on a millisecond timescale, thereby capturing the pulsating characteristics of the flow velocity. To cover flow velocity conditions in different areas of the pipeline, the transducers are mounted at multiple angles, monitoring flow velocities in the pipeline's center and near the wall. Data fusion is then used to determine the flow velocity pulsation distribution throughout the pipeline.

[0027] Temperature distribution data is acquired using an infrared thermal imaging sensor, installed in the non-contact measurement area of ​​the pipeline. This sensor captures real-time images of the temperature distribution on the pipeline surface, and then uses a temperature field reconstruction algorithm to invert the temperature distribution of the slurry inside the pipeline. The sensor's field of view can be adjusted using a motorized adjustment mechanism to ensure full coverage of the pipeline's cross-section. Dust protection and cooling devices ensure continuous and stable operation in the complex mining environment, preventing dust accumulation and high temperatures from affecting measurement accuracy.

[0028] When the ore sample characteristic fusion module performs multi-dimensional feature decoupling, it first performs noise suppression on the slurry rheological parameter matrix. This noise suppression utilizes a wavelet transform-based denoising method, decomposing the matrix data into wavelet coefficients of varying frequencies. The high-frequency coefficients are thresholded to remove noise, and the de-noised slurry rheological parameter matrix is ​​reconstructed through an inverse wavelet transform. Different thresholding strategies are employed for different types of parameter data. For velocity pulsation data and temperature field distribution data, which are significantly affected by random noise, a more stringent hard thresholding approach is employed. For viscosity gradient data and particle size distribution data, which are primarily affected by system noise, a soft thresholding approach is employed to preserve more detailed information.

[0029] Data normalization converts the data in the noise-suppressed slurry rheological parameter matrix to a unified data interval. Specifically, the min-max normalization method is used to calculate the minimum and maximum values ​​of each parameter, mapping the raw data to the range [0, 1] to generate a standardized slurry rheological parameter matrix. During the normalization process, extreme outliers that may appear in some parameters are truncation-based, forcing values ​​outside the reasonable range to the maximum or minimum value to prevent outliers from affecting the overall normalization results.

[0030] The feature decoupling engine separates the independent eigenvalues ​​of each dimension in the standardized slurry rheological parameter matrix. This engine, based on the independent component analysis (ICA) algorithm, whitens the standardized matrix to remove second-order correlations between the data dimensions. It then uses iterative optimization to find a separation matrix that maximizes the statistical independence of each component, thereby separating the mixed eigenvalues ​​into independent single eigenvalues. Each separated independent eigenvalue is validated by calculating its correlation coefficient with the original parameter matrix. Components with higher correlations are retained, while interfering components with lower correlations are removed.

[0031] When aggregating independent feature components across all dimensions to generate a slurry feature tensor, the dimensional structure of the tensor is first determined. Using sampling time, feature type, and spatial location as the three fundamental dimensions, each independent feature component is assigned to the corresponding position in the tensor according to its corresponding time, type, and location information. For parameter data with different sampling frequencies, interpolation or downsampling is used to unify the resolution of the temporal dimension, ensuring consistency across the tensor. The resulting slurry feature tensor is stored as a three-dimensional array, intuitively reflecting the temporal and spatial relationships between the features of each dimension.

[0032] Example 2: See Figure 3When calculating cross-dimensional correlation strength, the mineral topology map module first extracts key mineral feature nodes for each rheological dimension from the slurry feature tensor. This extraction is based on feature importance assessment. By analyzing the sensitivity of each feature during the slurry rheological process, the module selects characteristic points that significantly affect the changes in slurry properties as nodes. These nodes are distributed across different rheological dimensions, covering aspects such as slurry viscosity, particle size distribution, and solid-liquid concentration. Each node corresponds to a characteristic behavior of the slurry under specific conditions.

[0033] After selecting any two key mineral characteristic nodes, the characteristic response sequence of each key mineral characteristic node in the slurry characteristic tensor is extracted. The characteristic response sequence is extracted based on the time axis, and the characteristic data related to each node in the slurry characteristic tensor is arranged in chronological order to form a continuous sequence of data. The length of the sequence is determined based on the actual sampling period and data volume to ensure that it can fully reflect the characteristic change trend of the node over time. For example, for a node related to slurry viscosity, its characteristic response sequence will include viscosity gradient change data at different times, while the sequence of a node related to particle size distribution will consist of particle size distribution parameters at each time.

[0034] When calculating the morphological similarity metric for characteristic response sequences using the dynamic time warping algorithm, the two sequences to be compared are first preprocessed. This preprocessing involves removing outliers from the sequences, which may be caused by transient sensor failures or sudden changes in slurry flow. A sliding window method is used to identify and replace data points that fall outside the normal fluctuation range. Subsequently, a time warping path for the sequences is constructed. This path uses dynamic programming to find the optimal match between the two sequences, allowing for nonlinear scaling of the sequences along the time axis to accommodate time delays or advances in characteristic responses at different nodes. During the calculation process, path constraints are set to limit the slope range of the warping path to avoid overly distorted matching results.

[0035] When calculating the morphological similarity metric, the sum of the squared differences between corresponding points on a regular path between two sequences is used as the cumulative distance. A smaller cumulative distance indicates greater morphological similarity between the two sequences. After the cumulative distance is calculated, it is normalized to a morphological similarity metric. The normalization process is adjusted based on sequence length to enable effective similarity comparisons between sequences of different lengths.

[0036] When mapping morphological similarity metrics to cross-dimensional association strength weight coefficients, the first step is to determine the mapping range, typically set within the interval [0, 1]. The mapping rules are tailored to the required association strength in practical applications. When the morphological similarity metric is small, the corresponding weight coefficient is higher, indicating a closer association between the dimensions of the two nodes. When the metric is large, the weight coefficient is correspondingly lower, reflecting a weaker association. A nonlinear transformation function is introduced during the mapping process, assigning more sensitive weight changes to regions of high similarity to highlight the differences between strongly associated nodes. At the same time, regions of low similarity are smoothed to avoid drastic fluctuations in the weight coefficient.

[0037] When the mineral feature screening module analyzes the entropy value of the characteristic vector, it first obtains the mineral distribution dispersion coefficient of each rheological dimension. When calculating the dispersion coefficient, the mean and standard deviation of all characteristic data under this dimension are first counted. The mean reflects the central trend of the characteristic data, and the standard deviation reflects the degree of dispersion of the data. The ratio of the two is the mineral distribution dispersion coefficient. For dimensions with large data volumes, a segmented calculation method is used to divide the data into several segments by time or space, and the dispersion coefficient of each segment is calculated separately and then the average is taken to reduce the impact of extreme values ​​on the results. For example, when processing pipeline flow velocity pulsation data, the data is divided into multiple time periods according to the sampling time, and the mean and standard deviation of the flow velocity are calculated in each time period, and then the overall dispersion coefficient of the dimension is obtained.

[0038] When calculating the statistical weight distribution of characteristic components within each dimension's eigenvector, the composition of the eigenvector is first clarified. Each eigenvector is composed of multiple interrelated characteristic components, which reflect the mineral characteristics of that dimension from different perspectives. The statistical weight distribution is calculated based on the information contribution of each component in the eigenvector. The weight of each component is determined by analyzing the component's frequency of occurrence, degree of variation, and correlation with other components. Components that are active in the changes in slurry characteristics are assigned higher weights, while components that are more slowly changing or have less impact are assigned lower weights. An iterative adjustment mechanism is employed during the calculation process to optimize the weight distribution through multiple iterations, ensuring that the weight distribution more accurately reflects the actual importance of each component.

[0039] When generating the eigenvector entropy quantification results based on the mineral distribution discrete coefficient and the statistical weight distribution, the two are used as input parameters for comprehensive calculation. First, the mineral distribution discrete coefficient is standardized to keep it consistent with the value range of the statistical weight distribution, and then the two are fused through a weighted combination to form a preliminary entropy quantification result. During the fusion process, the weight ratio of the two is adjusted according to the characteristics of different rheological dimensions. For dimensions where the degree of discreteness has a greater impact on the mineral characteristics, the weight of the mineral distribution discrete coefficient is increased; for dimensions where the difference between the characteristic components is significant, the influence ratio of the statistical weight distribution is increased. Finally, the preliminary results are normalized to obtain the final eigenvector entropy quantification result, which can comprehensively reflect the uncertainty and information richness of the eigenvector, and provide a quantitative basis for subsequent mineral characteristic screening.

[0040] Example 3: See Figure 4 The process of screening the core entropy sequence of mineral characteristics by the mineral characteristic screening module begins with the extraction of the cross-dimensional correlation strength weight coefficients of all key mineral characteristic nodes in the global rheological characteristic topology map. In the global rheological characteristic topology map, each key mineral characteristic node is associated with multiple other nodes. The extraction of the cross-dimensional correlation strength weight coefficient requires traversing the association records of each node with all other nodes. For each node, its cross-dimensional correlation strength values ​​with other nodes are collected. These values ​​are values ​​in the [0,1] interval previously calculated and mapped by the dynamic time warping algorithm. These values ​​are arithmetically averaged to obtain the average correlation strength of the node. This average correlation strength is the cross-dimensional correlation strength weight coefficient of the key mineral characteristic node, and its numerical value reflects the overall degree of correlation between the node and other dimensional nodes in the global topology structure.

[0041] The quantified results of the eigenvector entropy values ​​are weightedly fused with the corresponding cross-dimensional correlation strength weight coefficients. The quantified results of the eigenvector entropy values ​​are calculated through the mineral distribution discrete coefficient and the statistical weight distribution, reflecting the characteristic information of a single node in its own rheological dimension; the cross-dimensional correlation strength weight coefficient reflects the correlation importance of the node in the global topology. In the weighted fusion process, the corresponding weight ratio needs to be assigned to the two, and the ratio is dynamically adjusted according to the actual characteristics of the slurry. For gold concentrate slurry with higher viscosity, the proportion of the cross-dimensional correlation strength weight coefficient is appropriately increased to highlight the mutual influence between nodes; for slurry with relatively good fluidity, the proportion of the quantified results of the eigenvector entropy values ​​is appropriately increased. The following formula is used for fusion:

[0042] in, represents the comprehensive evaluation value of the feature, Represents the weight coefficient of the quantized result of the feature vector entropy value, Represents the quantization result of the feature vector entropy value, The weight coefficient representing the cross-dimensional correlation strength weight coefficient, represents the cross-dimensional correlation strength weight coefficient, and .

[0043] Highly significant mineral feature nodes are screened out by presetting the entropy threshold. The setting of the entropy threshold needs to refer to the distribution of the comprehensive evaluation values ​​of the characteristics in the historical data, and be determined in combination with the requirements for the representativeness of the mineral samples in actual production. First, the comprehensive evaluation values ​​of the characteristics of all key mineral feature nodes in the past period of time are collected, and their distribution histograms are drawn to analyze the central trend and dispersion of the data. According to the distribution characteristics of the histogram, a suitable value is selected as the threshold. The threshold should be able to distinguish those nodes with high comprehensive evaluation values ​​and greater influence on the overall characteristics of the slurry. When the comprehensive evaluation value of a key mineral feature node is higher than the threshold, it is judged as a highly significant mineral feature node; otherwise, it is regarded as a low-significance node and is not temporarily included in the core entropy sequence.

[0044] The core entropy sequence of mineral characteristics is generated by sorting all highly significant mineral characteristic nodes in descending order of entropy value. The sorting process uses the comprehensive evaluation value of the characteristics as the primary basis, and all highly significant nodes are ranked from high to low. If nodes with the same comprehensive evaluation value exist, their eigenvector entropy values ​​are further compared, with the node with the higher entropy value ranked first. If the eigenvector entropy values ​​are also the same, their cross-dimensional correlation strength weight coefficients are compared, with the node with the higher coefficient ranked first. After the sorting is completed, an ordered node sequence is formed, which is the core entropy sequence of mineral characteristics.

[0045] After generating the core entropy sequence, its stability must be verified. During this verification process, multiple sets of slurry sample data are continuously collected, and the core entropy sequence of mineral characteristics corresponding to each set of data is calculated. The changes in the order of nodes in different sequences are compared. If the relative positions of the nodes remain stable across multiple sequences, the core entropy sequence has good reliability. If there are large fluctuations, the entropy threshold or the weight ratio of the weighted fusion must be readjusted until the sequence stabilizes. In addition, the length of the core entropy sequence must be determined based on actual needs. A sequence that is too long may contain excessive redundant information, increasing the complexity of subsequent processing; a sequence that is too short may omit important nodes, affecting the accuracy of sampling anomaly detection. Typically, based on the accuracy requirements of ore sample analysis in production, the top 20%-30% of the nodes in the sequence are retained as the core entropy sequence. The specific ratio can be dynamically adjusted based on changes in slurry viscosity.

[0046] The selected core entropy sequence of mineral characteristics must be updated in real time to adapt to changes in slurry properties. The update frequency is consistent with the sampling frequency of the ore flow state perception module, ensuring that the core entropy sequence can promptly reflect the latest state of the slurry. During each update, only the newly added slurry data is processed, and a new comprehensive feature evaluation value is calculated. This is compared with the existing core entropy sequence, and nodes that no longer meet the high significance criteria are replaced. Newly emerged high-significance nodes are added to the sequence and reordered. Through this dynamic update mechanism, the core entropy sequence of mineral characteristics can continuously and accurately reflect the core characteristics of the slurry, providing a reliable basis for the sampling anomaly detection module.

[0047] Example 4: See Figure 5 The sampling anomaly detection module includes a sampling execution feedback unit, an instruction optimization unit, and an instruction fusion unit. The sampling execution feedback unit collects the motion state parameters of the sampling robot arm in real time and generates a motion state feedback signal. The displacement trajectory deviation data in the motion state feedback signal is obtained by the encoder and the visual positioning device installed at the joint of the robot arm. The encoder records the actual rotation angle of each joint of the robot arm and calculates the theoretical position of the end effector in combination with the kinematic model of the robot arm. The visual positioning device obtains the actual position by capturing the image of the end effector in the workspace and analyzing the coordinates of the feature points in the image. The difference between the two is the displacement trajectory deviation data, which is in millimeters and contains deviation components in the three directions of X, Y, and Z.

[0048] Hydraulic actuator pressure fluctuation data is collected by pressure sensors integrated into the hydraulic pipeline. The sensor's sampling frequency is set according to the hydraulic system's response speed, typically collecting data every millisecond. This data records the pressure changes of the hydraulic oil during different movements. The pressure fluctuation data is measured in megapascals, forming a time-varying pressure curve. The sampling container fill rate data is acquired by a flow sensor installed at the inlet of the sampling pipeline. Based on the principle of electromagnetic induction, this sensor measures the volume of slurry flowing into the sampling container per unit time. The fill rate data is measured in liters per minute and varies with the viscosity of the slurry and the rhythm of the sampling arm's movement. Mechanical vibration spectrum data is collected by accelerometers fixed to the arm's base and end effector. The sensors capture the vibration acceleration of the arm during movement. Signal conversion converts the time-domain signals into the frequency domain, resulting in vibration amplitudes at different frequencies, ranging from 0 to 1000 Hz, and the amplitudes are measured in meters per second squared.

[0049] The instruction optimization unit is connected to the sampling and execution feedback unit, fusing the mineral characteristic core entropy sequence with the action state feedback signal to generate dynamic compensation instructions. During this fusion process, each node value in the mineral characteristic core entropy sequence is first correlated with various parameters in the action state feedback signal to determine which node changes are related to fluctuations in the action state parameters. For example, when the node value reflecting the viscosity of the mineral characteristic core entropy sequence increases, its correlation with the hydraulic execution pressure fluctuation data is analyzed. If a significant synchronous change is found between the two, the hydraulic pressure is adjusted accordingly when generating the dynamic compensation instruction to reflect the change in the viscosity node.

[0050] Dynamic compensation instructions are generated using a hierarchical processing approach. For displacement trajectory deviation data, the angle increments required to adjust each joint of the robotic arm are calculated based on the direction and magnitude of the deviation. For hydraulic execution pressure fluctuation data, the hydraulic valve opening adjustment amount is determined based on the amplitude and frequency of the pressure fluctuation. For sampling container filling rate data, the slurry inflow rate is changed by adjusting the valve opening on the sampling pipeline. For mechanical vibration spectrum data, the robotic arm component corresponding to the main vibration frequency is identified, and the vibration is suppressed by adjusting the motion parameters of the component.

[0051] The instruction fusion unit superimposes and integrates the sampling action control instructions and the dynamic compensation instructions to output the final sampling control signal. The sampling action control instructions are basic control instructions generated based on the core entropy sequence of mineral characteristics, and include basic parameters such as the robot's motion path, sampling time, and sampling volume; the dynamic compensation instructions are adjustment instructions for deviations in the action state feedback signal. When superimposed and integrated, the two types of instructions need to be converted into the same parameter space. For example, the angle adjustment amount and displacement compensation amount of the robot arm can be uniformly converted into the number of pulses of the joint motor, and the hydraulic pressure adjustment amount can be converted into the control current of the hydraulic valve.

[0052] During the integration process, different adjustment instructions for the same control parameter are prioritized. When a parameter in a sampling action control instruction conflicts with a parameter in a dynamic compensation instruction, the priority is determined based on the parameter's impact on sampling accuracy. The parameter with the greater impact is adopted, while the other parameter is appropriately corrected to reduce the conflict. The resulting sampling control signal is a series of digital pulses and analog voltage signals, which are transmitted via a cable to the sampling robot's control system, directly driving each actuator to complete the sampling operation.

[0053] The following is an example of collecting some parameters in the action status feedback signal:

[0054] This table records some action state parameters at five consecutive sampling moments. Through these parameters, the action changes of the robot arm in a short period of time can be observed, providing real-time feedback information for the instruction optimization unit.

[0055] Example 5: The system includes an adaptive sampling actuator connected to the sampling anomaly detection module, which receives sampling action control instructions and performs physical sampling operations. The rotary cutting sampling head, vacuum negative pressure stabilization unit, multi-stage filtration and separation chamber, and sample packaging robot in the adaptive sampling actuator work together to complete the entire process from sample acquisition to sample packaging in the slurry.

[0056] The rotary cutting sampling head is installed at the sampling port of the slurry pipeline. Its cutting action is driven by a servo motor, and its operating parameters can be adjusted according to the sampling action control instructions. The cutting speed gradient refers to the rate of change of the sampling head speed over time from startup to reaching the preset rated speed. This rate of change is set according to the viscosity characteristics of the slurry. When the slurry viscosity is high, a lower speed gradient is used to slowly accelerate the sampling head to avoid sample splashing or damage to the sampling head due to excessive instantaneous impact force. When the slurry viscosity is low, a higher speed gradient can be used to shorten the time it takes for the sampling head to reach the rated speed. The cutting angle offset is the angular deviation between the cutting plane of the sampling head and the pipeline axis. By adjusting this angle, the sampling head can adapt to the flow direction of the slurry in the pipeline. When the slurry flows in a spiral pattern in the pipeline, the cutting angle offset is adjusted accordingly to ensure that the sampling head cuts into the core area of ​​the slurry flow. The reciprocating frequency is the number of times the sampling head moves back and forth between the sampling position and the initial position after completing a cut. This frequency is set based on the sampling interval requirements. When multiple consecutive samples are required, the reciprocating frequency is increased to shorten the sampling interval. The cutting depth increment is the change in the depth of the sampling head inserted into the pipe with each cut. The initial cutting depth is determined by the pipe diameter and slurry flow rate. For each subsequent cut, the depth increment is adjusted based on the previous sampling volume. If the previous sampling volume is insufficient, the cutting depth increment is increased; otherwise, it is reduced.

[0057] The vacuum negative pressure stabilization unit is connected to the rotary cutting sampling head via a pipeline, which sucks in the cut mineral samples and transports them to the multi-stage filtration and separation chamber. The negative pressure intensity threshold is the minimum negative pressure value required to maintain stable mineral sample transportation. This value is determined based on the particle size and viscosity of the mineral sample. For mineral samples with larger particles and higher viscosity, a higher negative pressure intensity threshold is set to ensure that the mineral sample can pass smoothly through the transportation pipeline. For mineral samples with smaller particles and lower viscosity, a lower negative pressure intensity threshold is set to avoid excessive evaporation of liquid components in the mineral sample due to excessive negative pressure. The airflow pulsation suppression coefficient is achieved by adjusting the opening of the pressure regulating valve in the vacuum system. When airflow pulsation is detected in the transportation pipeline, the suppression coefficient is adjusted according to the amplitude of the pulsation. The larger the coefficient, the larger the adjustment amplitude of the pressure regulating valve, so as to quickly suppress the airflow pulsation, ensure that the mineral sample is evenly stressed during transportation, and avoid the mineral sample from being deposited in the pipeline due to pulsation. The air pressure balance value of the sample transmission pipeline refers to the air pressure difference at the inlet and outlet of the pipeline. This value is set according to the length and diameter of the transmission pipeline. When the pipeline is long and the diameter is small, a larger air pressure balance value is set to overcome the pipeline resistance; when the pipeline is short and the diameter is large, a smaller air pressure balance value is set to reduce energy consumption.

[0058] The multi-stage filtration separation chamber consists of multiple filter units connected in sequence. Each filter unit is equipped with screens of different apertures. After the mineral sample enters, it is filtered through each level of screen in turn to achieve the separation of particles of different particle sizes. The screen aperture combination sequence is a sequence in which the aperture sizes of each level of screen are arranged in the order of filtration. This sequence is determined based on the expected particle size distribution of the mineral particles in the mineral sample. It is usually arranged from large to small, so that larger particles are separated first and smaller particles enter the subsequent filtration unit. The vibration separation amplitude curve is a curve that shows the change of vibration amplitude over time during the filtration process of the separation chamber. The shape of the curve is determined by the viscosity of the mineral sample and the agglomeration of particles. For mineral samples with high viscosity and easy particle agglomeration, a curve with gradually increasing amplitude is used to break up the particle agglomeration through continuously increasing vibration. For mineral samples with low viscosity and good particle dispersion, a steady vibration amplitude curve is used to reduce unnecessary energy consumption. The mineral residue cleaning cycle is the time interval for regular cleaning of the screen, which is determined according to the blockage condition of the screen. When it is detected that the filtration efficiency of the screen has dropped to a certain level, the cleaning cycle is triggered. The cleaning methods include reverse airflow blowing and mechanical scraping to remove the residual mineral particles on the screen and restore the filtration performance of the screen.

[0059] The sample packaging robot is installed at the exit of the multi-stage filtration and separation chamber. Once the filtered and separated mineral sample enters the designated location, the robot moves to place the sample into a sample container and then seals it. The robot's motion parameters, including gripping force, movement speed, and sealing pressure, are adjusted based on the material of the sample container and the characteristics of the mineral sample. For fragile containers, lower gripping force and sealing pressure are used; for heavier mineral samples, higher movement speeds are used to shorten sealing time. Once sealed, the sample packaging robot places the sealed sample in a designated storage location for subsequent processing.

[0060] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0061] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An automatic sampling system for viscous gold concentrate based on intelligent sensing, characterized in that: include: an ore flow state sensing module configured to collect dynamic parameters of rheological characteristics of viscous gold concentrate slurry in real time through a multi-source sensor array, perform heterogeneous data fusion processing on the dynamic parameters of rheological characteristics, and output a slurry rheological parameter matrix; an ore sample characteristic fusion module, connected to the ore flow state perception module, configured to perform multi-dimensional feature decoupling on the slurry rheological parameter matrix to generate a slurry characteristic tensor, and construct a local correlation topology of each rheological dimension based on the slurry characteristic tensor; A mineral topology map module is configured to extract key mineral characteristic nodes in each rheological dimension, calculate the cross-dimensional correlation strength between the key mineral characteristic nodes, and integrate all local correlation topologies to form a global rheological characteristic topology map of the viscous gold concentrate; a mineral feature screening module configured to parse the characteristic vector entropy value of each rheological dimension from the slurry feature tensor, and screen out a core entropy sequence of mineral features by combining the cross-dimensional correlation strength in the global rheological feature topology map; The sampling anomaly detection module is connected to the mineral feature screening module and is configured to trigger sampling anomaly judgment of the sticky gold concentrate based on the mineral feature core entropy sequence and generate a sampling action control instruction.

2. The automatic sampling system for viscous gold concentrate based on intelligent sensing according to claim 1 is characterized in that: The dynamic parameters of rheological characteristics included in the ore flow state perception module specifically include slurry viscosity gradient data, mineral particle size distribution data, solid-liquid two-phase concentration data, pipeline flow velocity pulsation data and temperature field distribution data.

3. The automatic sampling system for viscous gold concentrate based on intelligent sensing according to claim 1 is characterized in that: The operations of the ore sample characteristic fusion module to perform multi-dimensional feature decoupling include: performing noise suppression and data normalization processing on the slurry rheological parameter matrix to generate a standardized slurry rheological parameter matrix; Separating independent characteristic components of each dimension in the standardized slurry rheological parameter matrix through a characteristic decoupling engine; The independent feature components of all dimensions are aggregated to generate the slurry feature tensor.

4. The automatic sampling system for viscous gold concentrate based on intelligent sensing according to claim 1 is characterized in that: The process of calculating the cross-dimensional correlation strength by the mineral topology map module includes: Select any two key mineral characteristic nodes, and extract characteristic response sequences of the key mineral characteristic nodes in the slurry characteristic tensor; Calculating a morphological similarity measure of the feature response sequence using a dynamic time warping algorithm; The morphological similarity measure value is mapped to a cross-dimensional association strength weight coefficient.

5. The automatic sampling system for viscous gold concentrate based on intelligent sensing according to claim 1 is characterized in that: The operation of the mineral feature screening module to analyze the feature vector entropy value includes: Obtain the mineral distribution dispersion coefficient for each rheological dimension; Calculate the statistical weight distribution of the internal feature components of the feature vector of each dimension; A quantified result of the characteristic vector entropy value is generated based on the mineral distribution dispersion coefficient and the statistical weight distribution.

6. The automatic sampling system for viscous gold concentrate based on intelligent sensing according to claim 1 is characterized in that: The step of screening the mineral characteristic core entropy sequence by the mineral characteristic screening module includes: Extracting cross-dimensional correlation strength weight coefficients of all key mineral characteristic nodes in the global rheological characteristic topology map; Perform weighted fusion of the quantized result of the feature vector entropy value and the corresponding cross-dimensional correlation strength weight coefficient; The highly significant mineral feature nodes are screened out by setting the entropy threshold; All highly significant mineral feature nodes are arranged in descending order of entropy value to generate the mineral feature core entropy sequence.

7. The automatic sampling system for viscous gold concentrate based on intelligent sensing according to claim 1 is characterized in that: The sampling anomaly detection module further comprises: a sampling execution feedback unit configured to collect motion state parameters of the sampling manipulator in real time and generate a motion state feedback signal; an instruction optimization unit, connected to the sampling execution feedback unit, configured to fuse the mineral characteristic core entropy sequence with the action state feedback signal to generate a dynamic compensation instruction; The instruction fusion unit is configured to superimpose and integrate the sampling action control instruction and the dynamic compensation instruction, and output a final sampling control signal.

8. The automatic sampling system for viscous gold concentrate based on intelligent sensing according to claim 7 is characterized in that: The action state feedback signal specifically includes displacement trajectory deviation data of the sampling robot arm, hydraulic execution pressure fluctuation data, sampling container filling rate data and mechanical vibration spectrum data.

9. The automatic sampling system for viscous gold concentrate based on intelligent sensing according to claim 1 is characterized in that: The system further comprises: an adaptive sampling execution mechanism, connected to the sampling anomaly detection module, configured to receive the sampling action control instruction and perform a physical sampling operation; The self-adaptive sampling actuator comprises a rotary cutting sampling head, a vacuum negative pressure stabilization unit, a multi-stage filtering and separation chamber, and a sample packaging manipulator.

10. The automatic sampling system for viscous gold concentrate based on intelligent sensing according to claim 9, characterized in that: The operating parameters of the rotary cutting sampling head include cutting speed gradient, cutting angle offset, reciprocating motion frequency and cutting depth increment; The parameters adjusted in real time by the vacuum negative pressure stabilization unit include the negative pressure intensity threshold, the airflow pulsation suppression coefficient, and the air pressure balance value of the sample transmission pipeline; The separation parameters configured in the multi-stage filtration separation bin include a screen aperture combination sequence, a vibration separation amplitude curve, and a mineral residue cleaning cycle.

Images