A tailings thickening layering interface intelligent detection device and method based on model driving and adaptive sampling

Abstract

The application discloses a tailings thickening layering interface intelligent detection method and device based on model driving and adaptive sampling. The device comprises a detection mechanism, an execution mechanism, a control and data acquisition unit and a data processing and decision unit, and each component cooperates with each other to realize the detection of the slurry concentration distribution and the layering interface in the tailings thickening machine. Through the cooperation of the model-driven data analysis method and the adaptive sampling control strategy, the stability and accuracy of the slurry concentration prediction result are improved, and the detection efficiency is significantly improved while the interface recognition precision is ensured.

Description

Technical Field

[0001] This invention relates to the field of mineral processing and tailings treatment, specifically to an intelligent detection device and method for tailings density stratification interface based on model-driven and adaptive sampling. Background Technology

[0002] Tailings thickeners are commonly used solid-liquid separation equipment in mineral processing and tailings treatment. They are mainly used to thicken tailings slurry to achieve sedimentation of solid particles and recycling of the supernatant. During the operation of a tailings thickener, the slurry typically forms different functional zones along the vertical direction, such as a clear water zone, a sedimentation zone, and a compression zone. Clear stratification interfaces exist between these zones, and their positional changes directly reflect the sedimentation state and operating conditions inside the thickener. Therefore, accurate detection and identification of the stratification interfaces within the tailings thickener are crucial for optimizing operating parameters, improving thickening efficiency, and ensuring stable system operation.

[0003] Existing methods for detecting concentration in tailings thickeners largely rely on single-point or limited fixed-location detection, making it difficult to balance accuracy and efficiency. When the sampling interval is set too large, it fails to accurately reflect the concentration changes along the depth direction of the slurry, resulting in insufficient accuracy in identifying stratification interfaces. Conversely, a small sampling interval leads to an excessive number of sampling points, prolonging the detection cycle and reducing overall efficiency. Furthermore, the detection results are easily affected by slurry heterogeneity, impurity disturbances, and fluctuations in operating conditions. Judgment methods based on raw monitoring data or simple statistical characteristics lack stability and are prone to misjudgment. Regarding stratification interface identification, existing technologies largely rely on empirical thresholds or manual judgment, which are poorly adaptable to interface location identification under complex operating conditions, making reliable automatic identification of stratification interfaces difficult.

[0004] Therefore, it is necessary to propose a tailings density detection method and device that can improve detection efficiency while ensuring detection accuracy and achieve intelligent identification of layered interfaces. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes an intelligent detection device and method for tailings thickening and stratification interfaces based on model-driven and adaptive sampling. By introducing a concentration prediction method based on feature engineering and model-driven approaches during the detection process, and performing unified modeling and analysis on the detection data acquired along the depth direction, the device effectively characterizes the variation patterns of the detection signal data, reducing the impact of slurry heterogeneity, impurity disturbance, and random noise on the detection results, thereby improving the stability and accuracy of the slurry concentration prediction results. Simultaneously, the sampling position and sampling resolution of the detection probe are dynamically adjusted based on the prediction results, significantly improving detection efficiency while ensuring interface recognition accuracy.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an intelligent detection device for tailings density stratification interface based on model-driven and adaptive sampling, characterized in that it comprises the following parts: The detection mechanism is used to detect the slurry at different liquid levels in the tailings thickener and output detection signals. It consists of detection probes. An actuator, used to drive the detection probe to be horizontally positioned and vertically lowered and retrieved within the tailings thickener, to achieve detection at different horizontal positions and liquid levels, includes: a winding wheel, a fixed pulley, a servo motor, a telescopic mechanism, a rotating mechanism, a track sliding platform, and a track. The control and data acquisition unit is used to control the motion state of the actuator and acquire the detection signals output by the detection mechanism, including: a PLC integrated controller, a servo controller, a motor encoder, and an ultrasonic controller; The data processing and decision-making unit, which is communicatively connected to the control and data acquisition unit, is used to process and analyze the acquired detection data, generate detection results, and send control commands to the control and data acquisition unit to achieve adaptive regulation of the detection process. It is implemented by a computer and includes: a feature engineering processing module, a Conv-LSTM-based concentration mapping module, an adaptive sampling control module, an interface recognition module, and a result visualization module.

[0007] in: The feature engineering processing module is used to preprocess and construct features from the raw detection data, forming a sequence of feature vectors arranged along the depth direction; The concentration mapping module based on Conv-LSTM is used to map the feature vector sequence to the slurry concentration prediction value corresponding to each sampling depth position, and generate a slurry concentration-depth prediction curve. The mapping relationship is obtained by supervised training using detection data obtained under known slurry concentration conditions. An adaptive sampling control module is used to analyze concentration change characteristics based on concentration-depth prediction results, identify suspected depth intervals where there may be stratification transitions, and dynamically adjust the sampling step size of the detection probe along the depth direction within the interval to achieve adaptive switching from coarse sampling to fine sampling. The interface recognition module is used to generate a set of candidate interface points based on the high-resolution concentration-depth curve obtained in the fine sampling stage, and construct an interface discrimination feature vector. Through a multi-class interface discrimination model, it outputs the probability value of the candidate interface points belonging to the clear water zone-sedimentation zone interface, the sedimentation zone-compression zone interface, or a non-interface category, and determines the final stratified interface position based on the probability results. The results visualization module is used to integrate and display the concentration distribution results and the layered interface recognition results.

[0008] Furthermore, the detection probe is connected to a winding wheel via a fixed pulley and a wire. The winding wheel is driven by a servo motor to retract and extend the wire, thereby controlling the vertical lowering and retraction of the detection probe. The wire is a composite cable that combines mechanical load-bearing and signal transmission functions. The ultrasonic controller is located above the winding wheel and electrically connected to the wire end at the winding wheel to transmit signals to the detection probe via the wire. The fixed pulley is fixed to the end of the telescopic mechanism, and the wire passes through the telescopic mechanism and connects to the winding wheel. The telescopic mechanism can control the extension and retraction of the fixed pulley. The telescopic mechanism is fixed to a rotating mechanism, which can control the telescopic mechanism to rotate 180° horizontally. The rotating mechanism, the protective shell, and other internal components are respectively fixed on two sides to a track sliding platform. The track sliding platform can slide along the track direction. Through the above structure, the detection probe can be positioned at different locations in the thickener.

[0009] Furthermore, the detection probe includes a housing, a transmitting transducer, a receiving transducer, a temperature sensor, and a counterweight; the housing has a hollow structure with a through hole in the middle; the transmitting and receiving transducers are respectively disposed at both ends of the through hole for transmitting and receiving ultrasonic signals; the temperature sensor is disposed inside the through hole for detecting the slurry temperature; the housing is a sealed structure for protecting the internal electrical connection lines; the counterweight is used to maintain the stability of the detection probe during the lowering process.

[0010] Furthermore, the feature engineering processing module is used to preprocess and construct features from the original detection data, forming a sequence of feature vectors arranged along the depth direction. The specific implementation steps are as follows: The system receives raw detection data uploaded by the PLC integrated controller, extracts parameters from the raw ultrasonic waveform signal to obtain a detection signal parameter vector characterizing the ultrasonic propagation characteristics, and performs temperature compensation and normalization processing on the detection signal parameter vector. Based on continuous sampling data along the depth direction, it constructs a rate of change feature reflecting the trend of change in the depth direction, a change amplitude feature reflecting the intensity of local fluctuations, and a statistical feature reflecting local stability. The detection signal parameter vector is combined with the rate of change feature, change amplitude feature, and statistical feature to form a feature vector sequence corresponding to each sampling depth position, which serves as the input to the concentration mapping module based on Conv-LSTM.

[0011] Furthermore, the concentration mapping module based on Conv-LSTM includes convolutional neural network and long short-term memory network algorithms. The convolutional neural network is used to extract the spatial variation features of the feature sequence within the local depth range, and the long short-term memory network is used to model the continuous dependency of the feature sequence in the depth direction to establish a mapping relationship between the feature vector sequence and the slurry concentration. The mapping relationship is obtained through supervised training using detection data obtained under known slurry concentration conditions.

[0012] Furthermore, the specific steps for the adaptive sampling control module to dynamically adjust the sampling step size are as follows: During the probe lowering process, coarse sampling is performed according to a preset large sampling step size. Based on the concentration-depth prediction results obtained in the coarse sampling stage, the concentration change characteristics are analyzed. The concentration change rate between adjacent sampling points is calculated by analyzing the concentration-depth prediction curve, and depth segments with significantly increased and continuously distributed concentration change rates are identified based on the distribution of these rates along the depth direction. Simultaneously, the concentration distribution within adjacent depth ranges on both sides of the depth segment is analyzed. By calculating the mean concentration and standard deviation of the adjacent segments, it is determined whether there are differences in concentration levels and stability on both sides of the depth segment. When the depth segment meets the concentration... When the rate of change of concentration increases significantly and the rate of change of concentration is continuously distributed in the depth direction, and there are significant differences in the concentration distribution characteristics on both sides of the depth section, the depth section is identified as a suspected depth interval where there may be a transition of slurry stratification, and is thus determined as a candidate region for fine sampling. During the detection probe recovery process, the sampling step size is dynamically adjusted according to the candidate region for fine sampling. A smaller sampling step size is used to perform fine sampling operation in the suspected depth interval, and the detection data obtained during the fine sampling process is fed back to the feature engineering processing module and the concentration mapping module based on Conv-LSTM to obtain a higher resolution slurry concentration-depth prediction curve in the corresponding depth interval.

[0013] Furthermore, the specific implementation steps of the interface recognition module in identifying and determining the slurry layer interface are as follows: Based on the high-resolution slurry concentration-depth prediction curve obtained in the fine sampling stage, depth locations where the concentration changes significantly along the depth direction or where the trend of change shows a clear inflection are extracted to generate a candidate interface point set. For the candidate interface point set, an interface discrimination feature vector is constructed for each candidate interface point. The interface discrimination feature vector is used to characterize the concentration statistical characteristics and trend characteristics within a preset depth neighborhood above and below the candidate interface point. Based on the interface discrimination feature vector, a multi-class interface discrimination model is used to output whether the candidate interface point belongs to the clear water zone-sedimentation zone interface, the sedimentation zone-compression zone interface, or a non-interface type. Probability value; Based on the probability output, candidate interface points belonging to non-interface categories and with probability values ​​less than a preset threshold are removed; In the clear water zone-settling zone interface category and the settling zone-compression zone interface category, the candidate interface point with the highest probability value is selected as the corresponding interface position, and the depth order constraint relationship is satisfied, and the final layered interface depth information is output; When the maximum category probability of the candidate interface point is lower than the preset threshold, or when there are multiple candidate interface points with similar probability values ​​and simultaneously higher than the preset threshold in the same interface category, the corresponding depth position is resampled locally or the neighborhood is expanded and the interface discrimination is performed again.

[0014] Secondly, the present invention provides a model-driven and adaptive sampling-based intelligent detection method for tailings dense stratification interfaces, applied to the aforementioned model-driven and adaptive sampling-based intelligent detection device for tailings dense stratification interfaces, comprising the following steps: S1. Initial parameter setting and equipment initialization: In the initial stage of detection, the detection position, detection depth range and sampling parameters are set according to the requirements of the detection task, and the detection system initialization is completed. S2. Horizontal Positioning and Coarse Sampling: Position the detection probe at the predetermined horizontal position, and then lower the detection probe vertically. Use a large sampling step size to perform coarse sampling and obtain multi-point detection data covering the target detection depth range. S3. Fine sampling interval determination: Feature engineering is performed on the detection data obtained in the coarse sampling stage, and a concentration-depth curve is generated based on the model-driven concentration prediction method. Based on this, the suspected depth intervals where there may be layered interfaces are adaptively determined according to the characteristics of concentration change along the depth direction, and the fine sampling interval is determined accordingly. S4. Fine sampling execution: During the detection probe retrieval process, fine sampling operation is performed with a small sampling step size for the determined fine sampling interval to obtain high-resolution detection data within the corresponding interval; S5. Layered Interface Recognition: Based on high-resolution concentration-depth data obtained within the fine sampling interval, the layered interfaces between the clear water zone, settling zone, and compression zone inside the slurry are identified and determined, and the corresponding depth position of each layered interface is determined. S6. Multi-location result fusion and visualization output: Repeat steps S2 to S5 at different horizontal detection locations to fuse the multi-location detection results and output the visualization results of the slurry concentration distribution and layer interface inside the tailings thickener.

[0015] Beneficial effects The beneficial effects of this invention are as follows: (1) This invention combines model-driven data analysis with adaptive sampling control strategy through the systematic collaborative design of the detection device structure and detection method to form a complete intelligent detection process, which can realize the automated and intelligent detection of the slurry concentration distribution and layer interface inside the tailings thickener; (2) In the detection process, the present invention introduces a concentration prediction method based on feature engineering and model-driven method, and performs unified modeling and analysis on the detection data obtained along the depth direction. This can effectively characterize the change law of detection signal data, reduce the influence of slurry heterogeneity, impurity disturbance and random noise on the detection results, thereby improving the stability and accuracy of slurry concentration prediction results. (3) The present invention adopts an adaptive sampling strategy that combines coarse sampling and fine sampling. By analyzing the coarse sampling results, the fine sampling interval is adaptively determined. High-resolution sampling is performed in the key depth range where there may be layered interfaces, avoiding the problem of low detection efficiency caused by high-density sampling throughout the process. While ensuring the accuracy of interface recognition, the detection efficiency is significantly improved.

[0016] (4) Based on the high-resolution concentration-depth data obtained in the fine sampling stage, this invention performs programmed identification and judgment of the layered interface between the clear water zone, the settling zone and the compression zone, reducing the reliance on human experience for interface judgment results; at the same time, by repeatedly performing detection at multiple horizontal detection positions and merging the detection results, it can obtain the spatial distribution characteristics of the slurry concentration inside the tailings thickener and the overall morphology of the layered interface, providing more intuitive and reliable data support for the operation status analysis, process control and operation optimization of the tailings thickening process. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the device structure principle provided by the present invention; Figure 2 This is a top view of the device provided by the present invention; Figure 3 The present invention provides a front view and a side sectional view of the detection probe of the device. Figure 4 This is a schematic diagram of the detection and control information transmission structure of the device provided by the present invention; Figure 5 This invention provides a schematic diagram of the concentration mapping method based on feature engineering and Conv-LSTM in the present invention. Figure 6 This invention provides an overall flowchart of the method. Figure 7 The present invention provides a concentration-depth curve obtained from a single measurement in the method, and a concentration distribution cloud map determined by fitting multiple concentration curves; The meanings of the symbols marked in the figure are as follows: 1-Computer; 2-Protective Housing; 3-PLC Integrated Controller; 4-Ultrasonic Controller; 5-Winding Wheel; 6-Servo Motor; 7-Telescopic Mechanism; 8-Fixed Pulley; 9-Detection Probe; 10-Motor Encoder; 11-Servo Controller; 12-Rail Sliding Platform; 13-Rail; 14-Rotation Mechanism; 901-Wire; 902-Transmitting Transducer; 903-Temperature Sensor; 904-Receiving Transducer; 905-Housing; 906-Weighted Weight; 101-Feature Engineering Processing Module; 102-Concentration Mapping Module Based on Conv-LSTM; 103-Adaptive Sampling Control Module; 104-Interface Recognition Module; 105-Result Visualization Module. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1 like Figure 1-2 As shown, this invention provides an intelligent detection device for tailings density stratification interface based on model-driven and adaptive sampling, which consists of the following parts: The detection mechanism, which is used to detect the slurry at different liquid levels in the tailings thickener and output detection signals, consists of detection probe 9; The actuator, used to drive the detection probe 9 to perform horizontal positioning and vertical lowering and retrieval within the tailings thickener, to achieve detection at different horizontal positions and different liquid levels, includes: a winding wheel 5, a fixed pulley 8, a servo motor 6, a telescopic mechanism 7, a rotating mechanism 14, a track sliding platform 12, and a track 13. The control and data acquisition unit is used to control the motion state of the actuator and acquire the detection signal output by the detection mechanism, including: PLC integrated controller 3, servo controller 11, motor encoder 10 and ultrasonic controller 4; The data processing and decision-making unit, which is communicatively connected to the control and data acquisition unit, is used to process and analyze the acquired detection data, generate detection results, and send control commands to the control and data acquisition unit to achieve adaptive regulation of the detection process. It is implemented by computer 1 and includes: feature engineering processing module 101, concentration mapping module 102 based on Conv-LSTM, adaptive sampling control module 103, interface recognition module 104, and result visualization module 105.

[0020] Among them, the feature engineering processing module 101 is used to preprocess and construct features from the original detection data to form a sequence of feature vectors arranged along the depth direction; The concentration mapping module 102 based on Conv-LSTM is used to map the feature vector sequence to the slurry concentration prediction value corresponding to each sampling depth position, and generate a slurry concentration-depth prediction curve. The mapping relationship is obtained by supervised training using detection data obtained under known slurry concentration conditions. The adaptive sampling control module 103 is used to analyze the concentration change characteristics based on the concentration-depth prediction results, identify the suspected depth intervals where there may be stratification transitions, and dynamically adjust the sampling step size of the detection probe along the depth direction within the interval to achieve adaptive switching from coarse sampling to fine sampling. The interface recognition module 104 is used to generate a set of candidate interface points based on the high-resolution concentration-depth curve obtained in the fine sampling stage, and construct an interface discrimination feature vector. Through a multi-class interface discrimination model, it outputs the probability value of the candidate interface points belonging to the clear water zone-sedimentation zone interface, the sedimentation zone-compression zone interface, or a non-interface category, and determines the final layered interface position based on the probability results. The results visualization module 105 is used to integrate and display the concentration distribution results and the hierarchical interface recognition results.

[0021] Furthermore, the detection probe 9 is connected to the winding wheel 5 via a wire 901 and a fixed pulley 8. The winding wheel 5 is driven by a servo motor 6 to retract and extend the wire 901, thereby controlling the vertical lowering and retraction of the detection probe 9 to meet the detection requirements at different liquid levels. The wire 901 is a composite cable that combines mechanical load-bearing and signal transmission functions. The ultrasonic controller 4 is located above the winding wheel 5 and is electrically connected to the wire end at the winding wheel 5 to transmit signals to the detection probe 9 via the wire 901.

[0022] Furthermore, the fixed pulley 8 is fixed to the end of the telescopic mechanism 7, and the wire 901 passes through the inside of the telescopic mechanism 7 and connects to the winding wheel 5, so that the telescopic mechanism can control the extension and retraction of the fixed pulley 8; further still, the telescopic mechanism 7 is fixed to the rotating mechanism 14, so that the rotating mechanism can control the telescopic mechanism 7 to rotate 180° in the horizontal direction; the rotating mechanism 14 and the protective shell 2 and other internal components are respectively fixed on both sides to the track sliding platform 12, and the track sliding platform 12 can slide along the track 13. Through the cooperation of the above-mentioned telescopic mechanism, rotating mechanism, track sliding platform, and track, they are used to achieve the horizontal positioning of the detection probe 9, so as to realize the arrangement and switching of different horizontal detection positions.

[0023] Furthermore, such as Figure 3 As shown, the detection probe 9 includes a housing 905, a transmitting transducer 902, a receiving transducer 903, a temperature sensor 904, and a counterweight 906. The housing 905 has a hollow structure with a through hole in the middle. The transmitting transducer 902 and the receiving transducer 903 are respectively disposed at both ends of the through hole. The transmitting transducer 902 is used to receive the resonant signal sent by the ultrasonic controller 4 and emit ultrasonic waves into the slurry. The receiving transducer 904 is used to receive the ultrasonic signal after it propagates through the slurry medium. The temperature sensor is disposed in the through hole and is used to detect the temperature of the slurry. The transmitting transducer 902, the receiving transducer 903, and the temperature sensor 904 are all mounted on the housing 905. The mounting parts are provided with a sealed structure to isolate the electrical connection parts of the sensors from the external medium, while their detection ends can interact with the slurry medium to realize the detection of corresponding physical parameters.

[0024] Preferably, the transmitting transducer 902, receiving transducer 903, and temperature sensor 904 are waterproof devices to meet the detection requirements in slurry environments. The housing 905 is a sealed structure to protect the internal electrical connection lines; the internal electrical connection lines are used to connect the conductor 901 to the transmitting transducer 902, receiving transducer 903, and temperature sensor 904 to provide operating power to each sensor and complete the transmission of detection signals. The counterweight 906 is used to maintain the stability of the detection probe during the lowering process. Depending on the type of slurry being tested, different masses of counterweights can be selected to ensure that the conductor and detection probe are vertically pulled.

[0025] Furthermore, such as Figure 4As shown, the PLC integrated controller 3 is used for centralized control of the device's operating status and is connected to the servo controller 11, the motor encoder 10, and the ultrasonic controller 4. The servo controller 11 controls the operation of the servo motor 6; the motor encoder 10 detects the operating status of the servo motor 6 and outputs position information to the PLC integrated controller 3; the ultrasonic controller 4 controls the transmission and reception of ultrasonic signals and outputs the acquired ultrasonic detection signals to the PLC integrated controller 3. The computer 1 is communicatively connected to the PLC integrated controller 3, receives the detection data transmitted by the PLC integrated controller 3, and stores and processes the detection data; the computer 1 sends control commands to the PLC integrated controller 3 based on the processing results to adjust the motion parameters and sampling strategy of the detection probe.

[0026] Furthermore, the feature engineering processing module 101 is used to preprocess and construct features from the original detection data to form a sequence of feature vectors arranged along the depth direction. The specific implementation steps are as follows: (1) Receive the raw detection data uploaded by the PLC integrated controller 3, wherein the data of the i-th sampling point along the depth direction in the raw detection data includes the sampling depth. Ultrasonic detection of the original waveform signal and the slurry temperature at the corresponding sampling point The original waveform signal of the ultrasonic detection is the ultrasonic echo signal received by the detection probe at the corresponding depth position; (2) Parameter extraction is performed on the original waveform signal of the ultrasonic detection to obtain a detection signal parameter vector characterizing the ultrasonic propagation characteristics, and the detection signal parameter vector is subjected to temperature compensation and normalization processing; specifically: The original ultrasonic detection waveform signal is subjected to signal processing to extract ultrasonic detection signal parameters characterizing the ultrasonic propagation characteristics. The signal processing preferably includes time-domain or frequency-domain analysis of the original waveform to extract parameters such as propagation time, echo amplitude, signal energy, attenuation characteristics, or correlation coefficient, and these parameters are then combined to form an ultrasonic detection signal parameter vector. ,in This indicates the dimension of the extracted ultrasonic detection signal parameters.

[0027] To address the impact of temperature changes on ultrasonic propagation characteristics, temperature compensation processing is performed on the ultrasonic detection signal parameter vector. The compensated detection signal parameter vector is expressed as follows: ,in The temperature compensation function can be implemented based on experimental data calibration models. Temperature compensation processing is already existing technology and will not be elaborated upon here. The components of the temperature-compensated ultrasonic detection signal parameter vector are normalized to eliminate differences in the dimensions and numerical ranges of different ultrasonic detection signal parameters, resulting in a normalized detection signal parameter vector. Based on the sampling order, the above data are matched and sorted according to the sampling order to form a normalized detection signal sequence arranged along the depth direction.

[0028] (3) Based on continuous sampling data along the depth direction, a rate of change feature reflecting the trend of change in the depth direction, an amplitude feature reflecting the intensity of local fluctuations, and a statistical feature reflecting local stability are constructed. The detection signal parameter vector is then combined with the rate of change feature, amplitude feature, and statistical feature to form a feature vector sequence corresponding to each sampling depth position, which serves as the input to the concentration mapping module based on Conv-LSTM. Specifically: Based on the normalized detection signal sequence formed by continuous sampling along the depth direction, the difference calculation is performed on the detection signal parameter vectors corresponding to adjacent sampling points to construct a feature vector reflecting the rate of change of the detection signal with depth. , used to characterize the rate of change of the detection signal in the depth direction; Set a local sampling window along the depth direction .in, Indicates the first A set of continuous sampling points centered on a sampling point; This indicates the number of sampling points selected upwards and downwards; when the sampling point is located at the sequence boundary, the local sampling window is adjusted according to the actual range of available sampling points. The normalized detection signal parameter vector within the window is analyzed to construct a feature vector reflecting the variation amplitude of local fluctuations in the detection signal. This is used to characterize the fluctuation of the detection signal within a local depth range. Represents a local sampling window Index of any sampling point within.

[0029] Simultaneously, statistical analysis is performed on the normalized detection signal parameter vector within the local sampling window to construct a local statistical feature vector reflecting the signal stability and fluctuation level, including the mean vector of the detection signal within the window. and standard deviation vector ; Normalized detection signal parameter vector eigenvector of rate of change , Variation amplitude feature vector and local statistical eigenvectors , Combined, forming a system with sampling depth corresponding feature vector The system outputs a sequence of feature vectors in order of sampling depth, which serves as the input data for the concentration mapping module based on Conv–LSTM.

[0030] It should be noted that, given the continuous variation of slurry concentration along the depth direction within the tailings thickener and the spatial correlation between adjacent sampling points, features reflecting the depth-direction variation trend and local stability are further constructed based on the normalized detection signal parameter vector. These features, including parameter features, trend features, and statistical features, are fused and organized into a feature vector sequence according to sampling depth. This achieves a multi-scale expression of the slurry concentration distribution, providing a spatially correlated structured input for the subsequent concentration mapping model. This approach enhances the characterization of slurry concentration variation along the depth direction while retaining the physical parameter feature information extracted from the original detection signal. Furthermore, compared to independent prediction based on a single sampling point, introducing adjacent depth information and sequence feature modeling effectively utilizes the continuity constraint along the depth direction, improving the stability and noise resistance of the concentration prediction results and enhancing the concentration mapping model's ability to identify and predict concentration distribution changes.

[0031] Furthermore, the concentration mapping module based on Conv-LSTM includes convolutional neural network and long short-term memory network algorithms. The convolutional neural network is used to extract the spatial variation features of the feature sequence within the local depth range, and the long short-term memory network is used to model the continuous dependency of the feature sequence in the depth direction to establish the mapping relationship between the feature vector sequence and the slurry concentration.

[0032] The mapping relationship is obtained through supervised training using detection data acquired under known slurry concentration conditions. Before the device operates, several ultrasonic detection data under known slurry concentration conditions are obtained experimentally. After feature engineering, feature sequence samples are formed, and the actual concentration values ​​at corresponding depth positions are used as supervision labels to train the parameters of convolutional neural networks and long short-term memory networks to establish the mapping relationship between feature sequences and slurry concentration. After training, the model parameters are stored in the system, and the trained model is called to calculate concentration during the online prediction phase.

[0033] Specifically, the concentration mapping module based on Conv-LSTM performs the following steps for concentration mapping on the feature-engineered data: Receive the feature vector sequence arranged along the depth direction output by the feature engineering processing module 101. A feature sequence window that slides along the depth direction is constructed on the feature vector sequence according to a preset sequence length. The feature sequence window is input into a convolutional neural network, and a one-dimensional convolution operation is performed in the sampling depth dimension to extract convolutional features that reflect the variation law of slurry concentration in the local depth range. Then, the convolutional features are input into a long short-term memory network to model the continuous dependency relationship of the feature sequence in the depth direction to obtain sequence features that characterize the trend of slurry concentration change. Based on the output of the long short-term memory network, the predicted value of slurry concentration corresponding to each sampling depth position is obtained by regressing the output structure, and arranged in the order of sampling depth to form a slurry concentration-depth prediction curve.

[0034] Furthermore, the adaptive sampling control module 103 can dynamically adjust the sampling step size, where the sampling step size refers to the distance between the upper and lower sampling point positions along the depth direction. The specific steps are as follows: During the lowering of the detection probe, the detection probe is controlled to perform coarse sampling according to a preset large sampling step size in order to obtain overall information on the distribution of slurry concentration along the depth direction inside the tailings thickener. Based on the slurry concentration-depth prediction results obtained in the coarse sampling stage, the concentration change characteristics are analyzed to identify the suspected depth range where there may be a slurry stratification interface. Specifically, the concentration change rate between adjacent sampling points is calculated for the concentration-depth prediction curve to characterize the gradient change characteristics of concentration along the depth direction. Based on the distribution of the concentration change rate along the depth direction, depth segments with significantly increased and continuously distributed concentration change rates are identified. Furthermore, the concentration distribution within adjacent depth ranges on both sides of the segment is analyzed. The adjacent depth ranges are segments formed by selecting preset depth ranges or several consecutive sampling points above and below the boundary of the depth segment. The mean concentration and standard deviation of the adjacent segments are calculated to determine whether there are differences in concentration levels and stability on both sides of the depth segment. When a depth segment satisfies the condition of a significantly increased concentration change rate and a continuously distributed concentration change rate along the depth direction (where continuous distribution means that the concentration change rate within the segment remains relatively large), and there are significant differences in concentration distribution characteristics between the segments on both sides, the depth segment is identified as a potential depth interval where slurry stratification transition may occur, and is thus determined as a candidate region for fine sampling.

[0035] During the detection probe retrieval process, the sampling step size is dynamically adjusted according to the fine sampling candidate region, and fine sampling operation is performed with a smaller sampling step size within the suspected depth range; The detection data obtained during the fine sampling process is fed back to the feature engineering module and the Conv-LSTM-based concentration mapping module to obtain higher resolution slurry concentration-depth prediction curves within the corresponding depth range. Higher resolution refers to using a smaller sampling step size to acquire more depth sampling points in the suspected stratification interface section, thereby obtaining denser concentration prediction values ​​and making the concentration-depth curve more refined in that section.

[0036] Furthermore, the interface recognition module 104 is used to identify and determine the slurry layering interface, and the specific implementation steps are as follows: Based on the high-resolution slurry concentration-depth prediction curve obtained in the fine sampling stage, the depth locations where the concentration changes significantly along the depth direction or where the trend of change changes sharply are extracted, and a candidate interface point set is generated; wherein, the candidate interface point set is defined as: a set of multiple "depth points that may belong to the interface" detected along the depth axis in a single detection process.

[0037] For the set of candidate interface points, an interface discrimination feature vector is constructed for each candidate interface point. The interface discrimination feature vector is used to characterize the concentration statistics and change trend characteristics within the preset depth neighborhood above and below the candidate interface point.

[0038] Specifically, the depth corresponding to each candidate interface point Centered on the target, the interface discrimination feature vector is constructed by calculating characteristic parameters such as the concentration mean, concentration standard deviation, concentration change rate, and concentration mean difference within the preset depth neighborhood above and below it. The upper and lower preset depth neighborhoods are adjacent segments formed by selecting preset depth ranges or several consecutive sampling points above and below the candidate interface point, respectively, with the candidate interface point as the center. The average concentration is used to characterize the overall concentration level of the segment, the average concentration difference is used to characterize the degree of concentration difference between the segments on both sides of the candidate interface point, the concentration change rate is used to reflect the gradient characteristics of concentration change along the depth direction, and the standard deviation of concentration is used to characterize the stability of the concentration distribution within the segment. Through the joint characterization of the above feature parameters, a comprehensive description of the concentration distribution characteristics of the segments on both sides of the candidate interface point is achieved.

[0039] Based on the interface discrimination feature vector, a multi-category interface discrimination model outputs the probability values ​​of candidate interface points belonging to the clear water zone-settling zone interface, the settling zone-compression zone interface, or a non-interface type. The multi-category interface discrimination model is obtained through supervised training using detection data collected under known slurry stratification conditions. Specifically, detection data with clearly defined stratification interface locations are acquired under experimental or field testing conditions. An interface discrimination feature vector is constructed from the detection data, and the samples are labeled with interface types according to the actual stratification state at the corresponding depth. The interface discrimination feature vector is used as the input sample, and the interface type labels are used as supervisory information to train the multi-category interface discrimination model, thereby establishing a mapping relationship between the interface discrimination feature vector and the interface category.

[0040] Based on the probability output, candidate interface points belonging to non-interface categories and with probability values ​​less than a preset threshold are removed. In the clear water-settling area interface category and the settling area-compression area interface category, the candidate interface point with the highest probability value is selected as the corresponding interface position, and the depth order constraint relationship is satisfied, and the final layered interface depth information is output. When the maximum category probability of a candidate interface point is lower than the preset threshold, or when there are multiple candidate interface points with similar probability values ​​and simultaneously higher than the preset threshold in the same interface category, the corresponding depth position is resampled locally or the neighborhood is expanded before the interface discrimination is performed again.

[0041] Furthermore, the result visualization module 105 is used to comprehensively display the slurry concentration prediction results output by the Conv-LSTM-based concentration mapping module and the layered interface recognition results output by the interface recognition module.

[0042] Specifically, this module can visualize the slurry concentration-depth curves obtained from a single or multiple detections, and supports the comparative display of multiple concentration-depth curves obtained at different detection locations or at different times; Simultaneously, based on the detection results from multiple detection locations and combined with the spatial location information of the detection probes, the slurry concentration data is spatially reconstructed to generate a concentration distribution cloud map reflecting the slurry concentration distribution state inside the tailings thickener. The layered interface positions of the clear water zone, settling zone, and compression zone determined by the interface recognition module 104 are superimposed and marked on the curve or cloud map, thereby realizing an intuitive and visual display of the slurry concentration distribution and layered interface inside the tailings thickener, providing auxiliary decision-making basis for operation status analysis and process control.

[0043] Example 2 like Figure 6 , Figure 7 As shown, based on the device described in Embodiment 1, the present invention provides an intelligent detection method for the tailings thickening stratification interface in a tailings thickener, the method comprising the following steps: S1. Initial parameter setting and equipment initialization: In the initial stage of detection, the detection position, detection depth range and sampling parameters are set according to the requirements of the detection task, and the detection system initialization is completed. Specifically, in the initial stage of detection, the required parameters are set via computer 1, including sampling horizontal position, sampling depth range, sampling step size, and sampling frequency, and these control parameters are input to the PLC integrated controller 3. The detection mechanism, actuator, and control and data acquisition unit are initialized and self-tested and calibrated according to the set parameters to ensure that the system operates under normal working conditions.

[0044] S2. Horizontal Positioning and Coarse Sampling: Position the detection probe at the predetermined horizontal position, and then lower the detection probe vertically. Use a large sampling step size to perform coarse sampling and obtain multi-point detection data covering the target detection depth range. Specifically, according to the control instructions of computer 1, the telescopic mechanism 7, the track sliding platform 12, and the rotating mechanism 14 work together to drive the detection probe 9 to be positioned horizontally along the tailings thickener, so that the detection probe 9 is at the preset horizontal position for target detection; the winding wheel 5, the fixed pulley 8, and the servo motor 6 drive the detection probe 9 to be lowered step by step vertically according to the preset large sampling step size. At each sampling depth position, the detection probe 9 stays for a preset time to complete stable sampling. During the stay of the detection probe 9, the ultrasonic controller 4 controls the transmitting transducer 902 to emit ultrasonic signals into the slurry, and the receiving transducer 904 receives the ultrasonic echo signals after propagation through the slurry medium; at the same time, the temperature sensor 903 collects the temperature information of the slurry at the corresponding depth. The collected ultrasonic detection signals, temperature data, and detection probe depth position information obtained by the motor encoder 10 are uniformly transmitted to the PLC integrated controller 3, and further sent to computer 1 for storage and processing.

[0045] S3. Fine sampling interval determination: Feature engineering is performed on the detection data obtained in the coarse sampling stage, and a concentration-depth curve is generated based on the model-driven concentration prediction method. Based on this, the suspected depth intervals where there may be layered interfaces are adaptively determined according to the characteristics of concentration change along the depth direction, and the fine sampling interval is determined accordingly.

[0046] Specifically, based on the coarse sampling detection data along the depth direction obtained in step S2, computer 1 calls feature engineering processing module 101 to preprocess and construct features from the detection data; the constructed feature sequence is input into concentration mapping module 102 based on Conv-LSTM to obtain the slurry concentration prediction value corresponding to each sampling depth position, and a concentration-depth prediction curve is formed; adaptive sampling control module 103 analyzes the concentration-depth prediction curve, identifies the suspected depth intervals where there may be slurry stratification transition, and determines the suspected depth intervals as fine sampling intervals.

[0047] S4. Fine sampling execution: During the detection probe retrieval process, fine sampling operation is performed with a small sampling step size for the determined fine sampling interval to obtain high-resolution detection data within the corresponding interval; Specifically, based on the fine sampling interval determined in step S3, the adaptive sampling control module 103 outputs fine sampling control commands. The winding wheel 5, fixed pulley 8, and servo motor 6 work together to drive the detection probe 9 to move vertically upwards and downwards, collecting data within the determined fine sampling interval with small sampling steps. The probe 9 retracts step by step according to the fine sampling step, stopping at each depth position within the fine sampling interval for a preset time to ensure stable and reliable data collection. The detection data within each fine sampling interval is uniformly transmitted to the PLC integrated controller 3 and further sent to the computer 1 for storage and processing.

[0048] S5. Layered Interface Recognition: Based on high-resolution concentration-depth data obtained within the fine sampling interval, the layered interfaces between the clear water zone, settling zone, and compression zone inside the slurry are identified and determined, and the corresponding depth position of each layered interface is determined. Specifically, based on the high-resolution detection data within each fine sampling interval obtained in step S4, computer 1 calls feature engineering processing module 101 to process the detection data, and inputs the processed feature data into concentration mapping module 102 based on Conv-LSTM to generate slurry concentration-depth prediction curves corresponding to each fine sampling interval. On this basis, computer 1 calls interface recognition module 104 to analyze and process the concentration-depth prediction curves within each fine sampling interval, completing the identification and determination of the clear water zone-sedimentation zone interface and the sedimentation zone-compression zone stratification interface, and outputting the corresponding interface depth or liquid level height information.

[0049] S6. Multi-location result fusion and visualization output: Repeat steps S2 to S5 at different horizontal detection locations to fuse the multi-location detection results and output the visualization results of the slurry concentration distribution and layer interface inside the tailings thickener.

[0050] Specifically, after completing the layered interface identification at a single horizontal position, the control and data acquisition unit, according to a preset detection scheme, repeats steps S2 to S5 sequentially at different horizontal detection positions to obtain the concentration-depth curves and layered interface identification results corresponding to multiple horizontal positions. The purpose of setting multiple horizontal detection positions for repeated detection is to avoid insufficient representativeness caused by single-point detection and to provide data support for constructing a three-dimensional cloud map of slurry concentration and the spatial morphology of the layered interface inside the thickener.

[0051] Computer 1 calls the result visualization module 105 to uniformly manage and integrate the detection results obtained at different horizontal positions. Based on the concentration distribution characteristics and interface depth information of each detection position, it generates a visualization result that reflects the spatial distribution characteristics of the slurry concentration inside the tailings thickener, and outputs the liquid level height and layered interface morphology corresponding to the clear water zone, settling zone and compression zone for operation status analysis and process control.

[0052] In summary, this invention utilizes a detection mechanism, an execution mechanism, a control and data acquisition unit, and a data processing and decision-making unit. These components work together to achieve intelligent detection of the slurry interface inside the tailings thickener. By combining model-driven data analysis methods with adaptive sampling control strategies, the stability and accuracy of slurry concentration prediction results are improved, significantly increasing detection efficiency while ensuring interface recognition accuracy.

[0053] Matters not covered in this invention are common knowledge. The above description is only for illustrating the technical concept and features of this invention, and its purpose is to enable those skilled in the art to understand the content of this invention and implement it accordingly, and should not be construed as limiting the scope of protection of this invention. All equivalent changes or modifications made in accordance with the spirit and essence of this invention should be covered within the scope of protection of this invention.

Claims

1. A smart detection device for tailings density stratification interface based on model-driven and adaptive sampling, characterized in that, It consists of the following parts, including: The detection mechanism is used to detect the slurry at different liquid levels in the tailings thickener and output detection signals. It consists of detection probes. An actuator, used to drive the detection probe to be horizontally positioned and vertically lowered and retrieved within the tailings thickener, to achieve detection at different horizontal positions and liquid levels, includes: a winding wheel, a fixed pulley, a servo motor, a telescopic mechanism, a rotating mechanism, a track sliding platform, and a track. The control and data acquisition unit is used to control the motion state of the actuator and acquire the detection signals output by the detection mechanism, including: a PLC integrated controller, a servo controller, a motor encoder, and an ultrasonic controller; The data processing and decision-making unit, which is communicatively connected to the control and data acquisition unit, is used to process and analyze the acquired detection data, generate detection results, and send control commands to the control and data acquisition unit to achieve adaptive regulation of the detection process. It is implemented by a computer and includes: a feature engineering processing module, a Conv-LSTM-based concentration mapping module, an adaptive sampling control module, an interface recognition module, and a result visualization module.

2. The intelligent detection device for tailings density stratification interface based on model-driven and adaptive sampling according to claim 1, characterized in that, The feature engineering processing module is used to preprocess and construct features from the raw detection data, forming a sequence of feature vectors arranged along the depth direction; The concentration mapping module based on Conv-LSTM is used to map the feature vector sequence to the slurry concentration prediction value corresponding to each sampling depth position, and generate a slurry concentration-depth prediction curve. The mapping relationship is obtained by supervised training using detection data obtained under known slurry concentration conditions. An adaptive sampling control module is used to analyze concentration change characteristics based on concentration-depth prediction results, identify suspected depth intervals where there may be stratification transitions, and dynamically adjust the sampling step size of the detection probe along the depth direction within the interval to achieve adaptive switching from coarse sampling to fine sampling. The interface recognition module is used to generate a set of candidate interface points based on the high-resolution concentration-depth curve obtained in the fine sampling stage, and construct an interface discrimination feature vector. Through a multi-class interface discrimination model, it outputs the probability value of the candidate interface points belonging to the clear water zone-sedimentation zone interface, the sedimentation zone-compression zone interface, or a non-interface category, and determines the final stratified interface position based on the probability results. The results visualization module is used to integrate and display the concentration distribution results and the layered interface recognition results.

3. The intelligent detection device for tailings density stratification interface based on model-driven and adaptive sampling according to claim 1, characterized in that, The detection probe is connected to a winding wheel via a wire and a fixed pulley. The winding wheel is driven by a servo motor to raise and lower the wire, controlling the vertical lowering and retraction of the detection probe. The wire is a composite cable that combines mechanical load-bearing and signal transmission functions. The ultrasonic controller is located above the winding wheel and electrically connected to the wire end at the winding wheel to transmit signals to the detection probe via the wire. The fixed pulley is fixed to the end of the telescopic mechanism, and the wire passes through the telescopic mechanism and connects to the winding wheel. The telescopic mechanism can control the extension and retraction of the fixed pulley. The telescopic mechanism is fixed to a rotating mechanism, which can control the telescopic mechanism to rotate 180° horizontally. The rotating mechanism, the protective shell, and other internal components are respectively fixed on two sides to a track sliding platform. The track sliding platform can slide along the track direction. Through the above structure, the detection probe can be positioned at different locations in the thickener.

4. The intelligent detection device for tailings density stratification interface based on model-driven and adaptive sampling according to claim 1, characterized in that, The detection probe includes a housing, a transmitting transducer, a receiving transducer, a temperature sensor, and a counterweight. The housing has a hollow structure with a through hole in the middle. The transmitting and receiving transducers are respectively located at both ends of the through hole for transmitting and receiving ultrasonic signals. The temperature sensor is located inside the through hole for detecting the slurry temperature. The housing is a sealed structure for protecting the internal electrical wiring. The counterweight is used to maintain the stability of the detection probe during the lowering process.

5. The intelligent detection device for tailings density stratification interface based on model-driven and adaptive sampling according to claim 2, characterized in that, The feature engineering processing module is used to preprocess and construct features from the raw detection data, forming a sequence of feature vectors arranged along the depth direction. The specific implementation steps are as follows: The system receives raw detection data uploaded by the PLC integrated controller, extracts parameters from the raw ultrasonic waveform signal to obtain a detection signal parameter vector characterizing the ultrasonic propagation characteristics, and performs temperature compensation and normalization processing on the detection signal parameter vector. Based on continuous sampling data along the depth direction, it constructs a rate of change feature reflecting the trend of change in the depth direction, a change amplitude feature reflecting the intensity of local fluctuations, and a statistical feature reflecting local stability. The detection signal parameter vector is combined with the rate of change feature, change amplitude feature, and statistical feature to form a feature vector sequence corresponding to each sampling depth position, which serves as the input to the concentration mapping module based on Conv-LSTM.

6. The intelligent detection device for tailings density stratification interface based on model-driven and adaptive sampling according to claim 2, characterized in that, The concentration mapping module based on Conv-LSTM includes convolutional neural network and long short-term memory network algorithms. The convolutional neural network is used to extract the spatial variation features of the feature sequence within the local depth range, and the long short-term memory network is used to model the continuous dependency of the feature sequence in the depth direction to establish a mapping relationship between the feature vector sequence and the slurry concentration. The mapping relationship is obtained through supervised training using detection data obtained under known slurry concentration conditions.

7. The intelligent detection device for tailings density stratification interface based on model-driven and adaptive sampling according to claim 2, characterized in that, The specific steps for the adaptive sampling control module to dynamically adjust the sampling step size are as follows: During the probe lowering process, coarse sampling is performed according to a preset large sampling step size. Based on the concentration-depth prediction results obtained in the coarse sampling stage, the concentration change characteristics are analyzed. The concentration change rate between adjacent sampling points is calculated by analyzing the concentration-depth prediction curve, and depth segments with significantly increased and continuously distributed concentration change rates are identified based on the distribution of these rates along the depth direction. Simultaneously, the concentration distribution within adjacent depth ranges on both sides of the depth segment is analyzed. By calculating the mean concentration and standard deviation of the adjacent segments, it is determined whether there are differences in concentration levels and stability on both sides of the depth segment. When the depth segment meets the concentration... When the rate of change of concentration increases significantly and the rate of change of concentration is continuously distributed in the depth direction, and there are significant differences in the concentration distribution characteristics on both sides of the depth section, the depth section is identified as a suspected depth interval where there may be a transition of slurry stratification, and is thus determined as a candidate region for fine sampling. During the detection probe recovery process, the sampling step size is dynamically adjusted according to the candidate region for fine sampling. A smaller sampling step size is used to perform fine sampling operation in the suspected depth interval, and the detection data obtained during the fine sampling process is fed back to the feature engineering processing module and the concentration mapping module based on Conv-LSTM to obtain a higher resolution slurry concentration-depth prediction curve in the corresponding depth interval.

8. The intelligent detection device for tailings density stratification interface based on model-driven and adaptive sampling according to claim 1, characterized in that, The specific steps for the interface recognition module to identify and determine the slurry layer interface are as follows: Based on the high-resolution slurry concentration-depth prediction curve obtained in the fine sampling stage, the depth positions where the concentration changes significantly along the depth direction or where the trend of change changes clearly turns are extracted to generate a candidate interface point set; for the candidate interface point set, an interface discrimination feature vector is constructed for each candidate interface point, and the interface discrimination feature vector is used to characterize the concentration statistical characteristics and trend characteristics in the preset depth neighborhood above and below the candidate interface point. Based on the interface discrimination feature vector, a multi-category interface discrimination model is used to output the probability values ​​of candidate interface points belonging to the clear water zone-settling zone interface, the settling zone-compression zone interface, or the non-interface type, respectively; according to the probability output results, candidate interface points belonging to the non-interface category and those with probability values ​​less than a preset threshold are eliminated. In the clear water zone-settlement zone interface category and the settlement zone-compression zone interface category, the candidate interface point with the highest probability value is selected as the corresponding interface position, and the depth order constraint relationship is satisfied, and the final layer interface depth information is output. When the maximum class probability of a candidate interface point is lower than the preset threshold, or when there are multiple candidate interface points of the same interface class with similar and simultaneously higher than the preset threshold, the interface discrimination is performed again after local fine sampling or expansion of the neighborhood at the corresponding depth position.

9. A model-driven and adaptive sampling-based intelligent detection method for tailings thickening and stratification interfaces, applied to the model-driven and adaptive sampling-based intelligent detection device for tailings thickening and stratification interfaces as described in any one of claims 1-8, characterized in that, Includes the following steps: S1. Initial parameter setting and equipment initialization: In the initial stage of detection, the detection position, detection depth range and sampling parameters are set according to the requirements of the detection task, and the detection system initialization is completed. S2. Horizontal Positioning and Coarse Sampling: Position the detection probe at the predetermined horizontal position, and then lower the detection probe vertically. Use a large sampling step size to perform coarse sampling and obtain multi-point detection data covering the target detection depth range. S3. Fine sampling interval determination: Feature engineering is performed on the detection data obtained in the coarse sampling stage, and a concentration-depth curve is generated based on the model-driven concentration prediction method. Based on this, the suspected depth intervals where there may be layered interfaces are adaptively determined according to the characteristics of concentration change along the depth direction, and the fine sampling interval is determined accordingly. S4. Fine sampling execution: During the detection probe retrieval process, fine sampling operation is performed with a small sampling step size for the determined fine sampling interval to obtain high-resolution detection data within the corresponding interval; S5. Layered Interface Recognition: Based on high-resolution concentration-depth data obtained within the fine sampling interval, the layered interfaces between the clear water zone, settling zone, and compression zone inside the slurry are identified and determined, and the corresponding depth position of each layered interface is determined. S6. Multi-location result fusion and visualization output: Repeat steps S2 to S5 at different horizontal detection locations to fuse the multi-location detection results and output the visualization results of the slurry concentration distribution and layer interface inside the tailings thickener.

Images