A method and device for identifying and controlling magnetic ore based on a hall sensor

Abstract

This invention discloses a magnetic ore identification and control method and device based on Hall sensors, belonging to the field of magnetic ore identification technology. It includes: synchronously acquiring the magnetic field signal of the ore to be detected using a modular matrix Hall sensor array system. The array system includes multiple Hall effect sensor units, each integrating a temperature compensation module. The spacing between the sensor units is between 1mm and 2mm. A high-speed multi-channel synchronous acquisition module is used to control the synchronous acquisition of the sensor units with a unified clock signal. This invention improves signal accuracy through a modular matrix Hall sensor array, followed by adaptive two-stage filtering and dual calibration, and generates a color magnetic field cloud map. Subsequently, a three-dimensional feature system integrating 256-dimensional feature vectors in the spatial-temporal-frequency domain is constructed, solving the problems of single-row sensor signal characteristics and signal distortion under heavy load.

Description

Technical Field

[0001] This invention relates to the field of magnetic ore identification technology, specifically to a magnetic ore identification control method and device based on a Hall sensor. Background Technology

[0002] In the field of magnetic ore identification and screening, Hall effect sensors are a commonly used detection device. Their application is based on the Hall effect principle: Hall effect sensors can detect changes in the surrounding magnetic field. When a magnetic ore passes near the sensor, the sensor converts the detected magnetic field signal into a quantifiable electrical signal. By analyzing this electrical signal, it is possible to initially determine whether the ore has any commercial value. This technical approach has found some application in the initial screening of magnetic ores due to its clear fundamental principles and low operational threshold.

[0003] The current mainstream magnetic ore identification solution in the industry uses a single-row Hall sensor array of 48 Hall sensors for detection. The process involves allowing the ore to pass perpendicular to the sensor array at a fixed speed in a straight line above it. The corresponding sensors detect the ore's magnetic field and convert it into a voltage signal. The ore's properties are then determined by comparing the output voltage signal to a preset initial threshold. If the voltage signal is greater than the threshold, it is considered usable ore; otherwise, it is considered tailings. However, this solution has significant drawbacks: first, the signal features captured by a single-row sensor array are relatively limited, and the judgment is mainly based on voltage signals, failing to comprehensively acquire information about the ore's magnetic distribution; second, relying on only one row of sensors significantly increases the workload of the sensors when the density of the ore is high, potentially leading to signal distortion or abnormal operation, thus affecting the accuracy of the identification results. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a magnetic ore identification and control method and device based on Hall sensors, which solves the problems in existing technologies where a single row of sensor groups can only capture relatively simple signal features and where signal distortion or abnormal operating status of the sensors occurs when the workload increases.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a magnetic ore identification and control method based on a Hall sensor, comprising: S1. A modular matrix Hall sensor array system is used to synchronously acquire the magnetic field signal of the ore to be tested. The array system includes multiple Hall effect sensor units, each sensor unit integrates a temperature compensation module, and the spacing between the sensor units is between 1mm and 2mm. A high-speed multi-channel synchronous acquisition module is used to control the sensor units to acquire data synchronously with a unified clock signal. S2. Adaptive two-stage filtering and dual calibration processing are performed on the magnetic field signal. The dual calibration processing includes zero-point drift calibration and environmental magnetic field calibration. A color magnetic field cloud map of the ore is generated based on the calibrated data. S3. Extracting multi-dimensional features: Based on the magnetic field cloud map and magnetic field signal, a three-dimensional feature system of "space-time domain-frequency domain" is constructed to extract multi-dimensional magnetic field features of the ore and integrate them into a 256-dimensional feature vector; S4. A hierarchical classification model with dual-model fusion is used to classify and identify feature vectors. The hierarchical classification model includes a first-level model for fast screening and a second-level model for fine classification. S5. Based on the classification and identification results, trigger the industrial control module to link with the sorting mechanism to sort the ore; S6. Perform incremental training on the hierarchical classification model that integrates the two models to update the model parameters.

[0006] Furthermore, in S1, the modular matrix Hall sensor array system supports multi-unit splicing expansion to adapt to the ore particle size requirements of different production lines; and the high-speed multi-channel synchronous acquisition module is connected to the sensor unit through low-impedance wires to ensure that the signal synchronization error of each sensor is less than 1ms.

[0007] Furthermore, in S2, the adaptive two-stage filtering includes: a first-stage adaptive low-pass filter that automatically adjusts the filter cutoff frequency according to the ore movement speed; and a second-stage wavelet denoising that specifically filters out pulse interference in the industrial environment.

[0008] Furthermore, in S2, zero-point drift calibration includes periodically and automatically collecting sensor output values ​​when no ore passes through and updating the zero-point reference; environmental magnetic field calibration includes collecting the environmental background magnetic field in real time by setting an environmental magnetic field reference sensor at the edge of the detection area, and removing background magnetic field interference from the detection data.

[0009] Furthermore, in S3, the multi-dimensional magnetic field features extracted from the ore include: spatial dimension features, including spatial gradient, spatial texture and edge contour; temporal dimension features, including magnetic field signal duration and magnetic field peak change rate; and frequency domain dimension features, which are extracted by performing Fourier transform on the magnetic field signal to extract feature frequencies.

[0010] Furthermore, in S4, the first-level model is a support vector machine model, which performs fast screening based on spatial gradient and magnetic field peak features; the second-level model is a lightweight convolutional neural network model, which performs fine classification based on the input magnetic field cloud map and feature vector.

[0011] Furthermore, in step S5, controlling the sorting mechanism to perform ore sorting based on the classification and identification results includes: If the magnetic ore is determined to be usable, the industrial control module triggers the sorting mechanism to guide the magnetic ore into the concentrate channel. If the magnetic ore is determined to be tailings, the industrial control module triggers the sorting mechanism to guide the magnetic ore into the tailings channel.

[0012] Furthermore, in step S6, incremental training of the hierarchical classification model fused with the dual models includes: manually labeling new samples through sample labeling units, integrating the new samples into the training set, and performing incremental training to update the dual model parameters.

[0013] The present invention also provides a magnetic ore identification and control device based on a Hall sensor, used to execute the magnetic ore identification and control method based on a Hall sensor described in any one of the above claims, comprising: The magnetic field acquisition module is used to synchronously acquire the magnetic field signal of the ore to be tested using a modular matrix Hall sensor array system that includes multiple Hall effect sensor units, each of which integrates a temperature compensation module. The spacing between the sensor units is between 1mm and 2mm. The array system uses a high-speed multi-channel synchronous acquisition module to control the sensor units to acquire the data synchronously with a unified clock signal. The data preprocessing module is used to perform adaptive two-stage filtering and two-calibration processing on the magnetic field signal. The two-calibration processing includes zero-point drift calibration and environmental magnetic field calibration, and generates a color magnetic field cloud map of the ore based on the calibrated data. The feature extraction module is used to construct a three-dimensional feature system of "space-time domain-frequency domain" based on magnetic field cloud map and magnetic field signal, extract multi-dimensional magnetic field features of ore and integrate them into a 256-dimensional feature vector; The intelligent recognition module is used to classify and recognize feature vectors using a hierarchical classification model that integrates two models. The hierarchical classification model includes a first-level model for fast screening and a second-level model for fine classification. The control linkage module is used to trigger the industrial control module to link the sorting mechanism to perform ore sorting based on the classification and identification results. The model self-update module is used to incrementally train the hierarchical classification model fused with the two models in order to update the model parameters.

[0014] The present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform any of the above-described methods for identifying and controlling magnetic ores based on Hall sensors.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention utilizes a modular matrix Hall sensor array, followed by adaptive dual-stage filtering and dual calibration to improve signal accuracy and generate a color magnetic field cloud map. Subsequently, it constructs a three-dimensional feature system integrating 256-dimensional feature vectors in the spatial-temporal-frequency domain, solving the problems of single-column sensor signal features and signal distortion under heavy load. It combines dual models to improve recognition accuracy and efficiency. Furthermore, it uses incremental training to update the model to adapt to ore changes, controlling the linkage to accurately separate concentrate and tailings, thus comprehensively improving the accuracy, stability, and adaptability of magnetic ore identification and sorting. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0017] 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.

[0018] Please see Figure 1 This invention provides a magnetic ore identification and control method based on a Hall sensor, comprising: S1. A modular matrix Hall sensor array system is used to synchronously acquire the magnetic field signal of the ore to be tested. The array system includes multiple Hall effect sensor units, each sensor unit integrates a temperature compensation module, and the spacing between the sensor units is between 1mm and 2mm. A high-speed multi-channel synchronous acquisition module is used to control the sensor units to acquire data synchronously with a unified clock signal. S2. Adaptive two-stage filtering and dual calibration processing are performed on the magnetic field signal. The dual calibration processing includes zero-point drift calibration and environmental magnetic field calibration. A color magnetic field cloud map of the ore is generated based on the calibrated data. S3. Extracting multi-dimensional features: Based on the magnetic field cloud map and magnetic field signal, a three-dimensional feature system of "space-time domain-frequency domain" is constructed to extract multi-dimensional magnetic field features of the ore and integrate them into a 256-dimensional feature vector; S4. A hierarchical classification model with dual-model fusion is used to classify and identify feature vectors. The hierarchical classification model includes a first-level model for fast screening and a second-level model for fine classification. S5. Based on the classification and identification results, trigger the industrial control module to link with the sorting mechanism to sort the ore; S6. Perform incremental training on the hierarchical classification model that integrates the two models to update the model parameters.

[0019] Specifically, the construction of the modular matrix Hall sensor array system needs to be designed in conjunction with the ore detection requirements. For example, 64 A1324 Hall effect sensor units are selected, each unit integrating a temperature compensation module, which can offset the influence of temperature on detection within the range of -40℃ to 150℃, and arranged in an 8×8 matrix structure with a spacing of 1.5mm. If it is necessary to adapt to the detection of larger particle size ores, it can be expanded into a 16×8 matrix by splicing. The high-speed multi-channel synchronous acquisition module uses the AD7606 data acquisition chip from ADI, and is paired with an STM32H743 microcontroller to generate a unified clock signal of 30MHz, controlling all sensor units to synchronously acquire magnetic field signals at a sampling rate of 1kHz. To ensure synchronization, the sensor units and the acquisition module are connected through copper core low-impedance wires with a cross-sectional area of ​​0.3mm². As measured by an oscilloscope, the synchronization error of each sensor signal is less than 0.8ms, effectively avoiding the detection deviation caused by signal asynchrony in traditional single-row sensors. At the same time, the matrix structure can comprehensively capture the spatial magnetic distribution of the ore, solving the problem of single-row sensor signal characteristics.

[0020] The adaptive two-stage filtering process is implemented as follows: The first-stage adaptive low-pass filter uses an infinite impulse response (IIR) filter, whose cutoff frequency is adjusted in conjunction with the ore movement speed, and the relationship formula is:

[0021] in, This is the low-pass filter cutoff frequency, in Hz. This is a proportionality coefficient, with a value of 500 Hz·s / m, determined based on on-site industrial testing. The velocity of the ore, measured in m / s, is obtained by a photoelectric velocity measurement module positioned in front of the sensor array. For example, if the photoelectric module detects a time difference of 0.5 seconds between the ore passing two phototubes spaced 0.1 m apart, then... .when hour, This can filter out high-frequency noise higher than the ore movement frequency. The second-stage wavelet denoising uses the db4 wavelet basis, decomposing the filtered signal into four layers and employing soft thresholding. The threshold calculation formula is:

[0022] in, The noise standard deviation is estimated from the signal detail coefficients. N = 1000, representing the number of signal sampling points. The processed signal is reconstructed to filter out pulse interference from the industrial environment, such as interference caused by motor start-stop. In the dual calibration process, zero-point drift calibration is performed once per hour. After the system pauses ore delivery, the sensor output values ​​are continuously collected 20 times when there is no ore, with a sampling interval of 100ms, using the formula:

[0023] Update the zero-point baseline. As the new zero-point benchmark, The value is the i-th acquisition value, n=20; for environmental magnetic field calibration, one SS49E reference sensor is set at each of the four corners of the detection area to acquire the background magnetic field in real time. Through the formula:

[0024] Eliminate interference. To detect the magnetic field, The magnetic field values ​​are calibrated, and all units are in mT. Based on the calibrated data, the magnetic field strength from 0 to 5 mT is mapped to RGB colors to generate a color magnetic field cloud map, which visually presents the magnetic distribution of the ore.

[0025] A three-dimensional feature system is constructed, encompassing the spatial, temporal, and frequency domains. The spatial dimension's gradient is calculated using the Sobel operator, with the following formula:

[0026] in, Texture is calculated for the magnetic field values ​​and gray-level co-occurrence matrix corresponding to the cloud map pixels, and contours are extracted using the Canny operator, totaling 100 dimensions. The time domain dimension calculates the signal duration, i.e., the time interval during which the signal exceeds the 0.2 mT threshold and the peak change rate, using the following formula:

[0027] in, For adjacent peak values, There are 24 dimensions in total, corresponding to the time period; The FFT transform of the time-domain signal in the frequency domain is given by the following formula:

[0028] in, For time-domain magnetic field, The angular frequency is used as the starting point, and the amplitudes of the first 132 characteristic frequencies are extracted, resulting in a total of 132 dimensions. These three are then integrated into a 256-dimensional feature vector, comprehensively covering the magnetic characteristics of the ore.

[0029] Dual-model fusion hierarchical classification: The first-level SVM model uses the RBF kernel function, with the following formula:

[0030] in, The first stage uses the mean of the spatial gradient and the peak value of the magnetic field as inputs. The training samples consist of 10,000 sets of historical data, and the filtering is completed within 0.1ms to exclude obvious tailings. The second stage is a lightweight CNN containing 2 convolutional layers, 2 pooling layers and 2 fully connected layers. It takes a cloud map and a 256-dimensional vector as inputs and initializes the parameters through transfer learning. The single sample processing time is 0.5ms, achieving fine classification while balancing speed and accuracy.

[0031] The control linkage uses a Siemens S7-1200 PLC. If the ore is determined to be concentrate, the PLC triggers the electromagnetic baffle to be energized, and the ore is introduced into the concentrate channel. If the ore is tailings, the baffle is de-energized and reset, and the ore is introduced into the tailings channel. The action is delayed by 200ms to match the conveying speed and reduce mixing.

[0032] Incremental training involves accumulating 500 manually labeled new samples, labeling them correctly using an industrial screen, merging the new samples with the original training set, and training with the SGD optimizer. The initial learning rate is 0.01, decaying by 0.5 every 10 rounds. Once the accuracy on the validation set meets the target, the model parameters are updated to ensure the model adapts to changes in ore composition.

[0033] In this embodiment, in S1, the modular matrix Hall sensor array system supports multi-unit splicing expansion to adapt to the ore particle size requirements of different production lines; and the high-speed multi-channel synchronous acquisition module is connected to the sensor unit through low-impedance wires to ensure that the signal synchronization error of each sensor is less than 1ms.

[0034] Specifically, the modular matrix Hall sensor array system allows for multi-unit splicing and expansion, which can be flexibly adjusted according to the ore particle size. For example, a 16×16 basic array module is used for 5mm ore particles; when the ore particle size increases to 10mm, another 16×16 module is added to create a 16×32 array, covering a larger detection area. Between the high-speed multi-channel synchronous acquisition module and the sensor units, low-impedance copper core wires with an impedance of less than 0.1Ω / m and a cross-sectional area of ​​0.5mm² are used. After connection, the rise time difference of each sensor signal is measured using an oscilloscope to ensure that the synchronization error is less than 1ms. This design can adapt to the ore particle size requirements of different production lines, while avoiding signal delay caused by excessive wire impedance, ensuring the consistency of acquired data, and further improving detection reliability.

[0035] In this embodiment, in S2, the adaptive two-stage filtering includes: a first-stage adaptive low-pass filter that automatically adjusts the filter cutoff frequency according to the ore movement speed; and a second-stage wavelet denoising that specifically filters out pulse interference in the industrial environment.

[0036] Specifically, the first stage of the adaptive dual-stage filtering, the adaptive low-pass filter, dynamically adjusts its cutoff frequency according to the ore's moving speed. For example, when the photoelectric velocimetry module detects an ore moving speed of 0.15 m / s, the cutoff frequency is adjusted according to the formula... A cutoff frequency is set to precisely filter out high-frequency noise unrelated to ore movement. The second-stage wavelet denoising uses a db4 wavelet basis, with a decomposition level of 5, and a threshold set to... Calculation, such as If N=1000, then By applying soft thresholding to the wavelet coefficients, pulse interference generated by motors and frequency converters in the industrial environment can be effectively filtered out, making the magnetic field signal purer and laying the foundation for subsequent calibration and feature extraction.

[0037] In this embodiment, in S2, zero-point drift calibration includes periodically and automatically collecting the sensor output value when no ore passes through and updating the zero-point reference; environmental magnetic field calibration includes setting an environmental magnetic field reference sensor at the edge of the detection area, collecting the environmental background magnetic field in real time, and removing background magnetic field interference from the detection data.

[0038] Specifically, zero-point drift calibration is performed periodically, for example, once per hour. After the system pauses ore conveying, each sensor unit continuously collects 20 output values ​​within 1 second, with a sampling interval of 50ms, using the formula... A new zero-point reference is calculated to replace the original reference, avoiding zero-point shift caused by long-term temperature changes in the sensor. During environmental magnetic field calibration, four environmental magnetic field reference sensors are set up at the edges of the detection area, approximately 10 cm in all directions, to collect background magnetic field data in real time. For example, at a certain moment When the sensor array detects the magnetic field At that time, through the formula By eliminating background interference, the test data is ensured to reflect only the ore's own magnetism, thus improving data accuracy.

[0039] In this embodiment, in S3, the multi-dimensional magnetic field features extracted from the ore include: spatial dimension features, including spatial gradient, spatial texture and edge contour; temporal dimension features, including magnetic field signal duration and magnetic field peak change rate; and frequency domain features, which are extracted by performing Fourier transform on the magnetic field signal to extract feature frequencies.

[0040] Specifically, in spatial dimension feature extraction, the spatial gradient uses the Sobel operator to calculate the gradients in the x and y directions, and integrates the absolute values ​​of the gradients into a 32-dimensional feature; spatial texture is calculated using the gray-level co-occurrence matrix, replacing gray values ​​with the brightness values ​​of the magnetic field cloud map, and calculating four types of features, including contrast and correlation, with each type of feature corresponding to 16 directions, for a total of 64 dimensions; edge contours are detected using the Canny operator, extracting eight features such as contour length and curvature, for a total of eight dimensions, resulting in a total of 104 spatial dimensions. In the temporal dimension features, the duration of the magnetic field signal is the time interval from the signal exceeding 0.2 mT to below 0.2 mT, for a total of eight dimensions; the rate of change of the magnetic field peak is calculated according to... The rate of change of the first 16 adjacent peaks is calculated, resulting in 16 dimensions, for a total of 24 dimensions in the time domain. The frequency domain features are extracted using FFT transformation, yielding the amplitude of the first 128 characteristic frequencies within the 0-500Hz range, resulting in 128 dimensions. These three dimensions are integrated into a 256-dimensional feature vector, comprehensively capturing the spatial, temporal, and frequency characteristics of the ore's magnetism, providing rich evidence for the classification model and improving the comprehensiveness of identification.

[0041] In this embodiment, in S4, the first-level model is a support vector machine model, which performs fast screening based on spatial gradient and magnetic field peak features; the second-level model is a lightweight convolutional neural network model, which performs fine classification based on the input magnetic field cloud map and feature vector.

[0042] Specifically, the first-level support vector machine (SVM) model uses the radial basis function (RBF) as the kernel function, and the kernel function parameters are... The input features are the spatial gradient mean and magnetic field peak (2D). The training samples are historical data from 8000 sets of concentrate and 2000 sets of tailings. Parameters were optimized using 5-fold cross-validation. After training, the single-sample screening time is less than 0.1ms, which can quickly exclude tailings with a magnetic field peak below 0.5mT, reducing the load on the second-level model. The second-level lightweight convolutional neural network (CNN) takes a 32×32×3 color magnetic field cloud map and a 256-dimensional feature vector as input. The network structure contains two convolutional layers: the first layer has 16 3×3 convolutional kernels, and the second layer has 32 3×3 convolutional kernels, two 2×2 max-pooling layers, and two fully connected layers, outputting 128-dimensional and 2-dimensional features respectively. The softmax function is used to output the classification probability. The model is initialized with pre-trained weights on the ImageNet dataset through transfer learning, and then fine-tuned with 5000 sets of samples after the first-level screening. The single-sample processing time is 0.5ms, and the classification accuracy is significantly higher than that of a single model, achieving a combination of fast screening and fine classification.

[0043] In this embodiment, step S5, controlling the sorting mechanism to perform ore sorting based on the classification and identification results, includes: If the magnetic ore is determined to be usable, the industrial control module triggers the sorting mechanism to guide the magnetic ore into the concentrate channel. If the magnetic ore is determined to be tailings, the industrial control module triggers the sorting mechanism to guide the magnetic ore into the tailings channel.

[0044] Specifically, when controlling the sorting mechanism based on the classification and identification results, the industrial control module uses a Mitsubishi FX5UPLC. The sorting mechanism consists of an electromagnetic baffle and a conveyor belt. If the intelligent identification module determines that the magnetic ore is usable, the PLC receives a "concentrate" signal and outputs a 24V DC voltage to energize the solenoid of the electromagnetic baffle. The baffle rotates to a position parallel to the conveyor belt, guiding the ore into the concentrate channel. If it is determined to be tailings, the PLC outputs a signal to de-energize the solenoid, and the baffle rotates to a position perpendicular to the conveyor belt under the action of a return spring, guiding the ore into the tailings channel. To match the ore conveying speed, such as 0.3m / s, the PLC is set with a 200ms action delay to ensure that the baffle completes switching when the ore reaches the sorting position, reducing the mixing of concentrate and tailings and improving the sorting effect.

[0045] In this embodiment, S6 involves incremental training of the hierarchical classification model fused with the dual models, including: manually labeling new samples using sample labeling units, integrating the new samples into the training set, and performing incremental training to update the dual model parameters.

[0046] Specifically, during incremental training of the hierarchical classification model fused with dual models, the sample labeling unit consists of an industrial display screen and a manual labeling button. Operators can view the ore's magnetic field cloud map, classification results, and original signals in real time. If a misclassification is detected, such as concentrate being misclassified as tailings, pressing the labeling button stores the sample's 256-dimensional feature vector, magnetic field cloud map, and correct label. The operator then selects either "concentrate" or "tailings." Every 500 newly labeled samples, the system initiates incremental training. The new samples are divided into an incremental training set and a validation set in a 7:3 ratio, and merged with the original 10,000 training sets. Training uses the SGD optimizer with an initial learning rate of 0.01, decaying by 0.5 every 10 training epochs to minimize the cross-entropy loss function. After training, the model is validated using the validation set. If the accuracy improves or remains unchanged, the dual model parameters are updated; if the accuracy decreases, the update is abandoned. This method allows the model to continuously learn new features, avoiding accuracy drops due to changes in ore composition and ensuring long-term stable operation.

[0047] The present invention also provides a magnetic ore identification and control device based on a Hall sensor, for executing any one of the above-mentioned magnetic ore identification and control methods based on a Hall sensor, comprising: The magnetic field acquisition module is used to synchronously acquire the magnetic field signal of the ore to be tested using a modular matrix Hall sensor array system that includes multiple Hall effect sensor units, each of which integrates a temperature compensation module. The spacing between the sensor units is between 1mm and 2mm. The array system uses a high-speed multi-channel synchronous acquisition module to control the sensor units to acquire the data synchronously with a unified clock signal. The data preprocessing module is used to perform adaptive two-stage filtering and two-calibration processing on the magnetic field signal. The two-calibration processing includes zero-point drift calibration and environmental magnetic field calibration, and generates a color magnetic field cloud map of the ore based on the calibrated data. The feature extraction module is used to construct a three-dimensional feature system of "space-time domain-frequency domain" based on magnetic field cloud map and magnetic field signal, extract multi-dimensional magnetic field features of ore and integrate them into a 256-dimensional feature vector; The intelligent recognition module is used to classify and recognize feature vectors using a hierarchical classification model that integrates two models. The hierarchical classification model includes a first-level model for fast screening and a second-level model for fine classification. The control linkage module is used to trigger the industrial control module to link the sorting mechanism to perform ore sorting based on the classification and identification results. The model self-update module is used to incrementally train the hierarchical classification model fused with the two models in order to update the model parameters.

[0048] Specifically, the magnetic field acquisition module uses an A1324 Hall effect sensor unit with an integrated temperature compensation module. The compensation range is -40℃ to 150℃. The modules are arranged in a 16×16 basic matrix with a 1.5mm spacing. It supports multi-module splicing, such as splicing two basic modules into a 16×32 array. The high-speed multi-channel synchronous acquisition module uses an AD7606 data acquisition chip and an STM32H743 microcontroller to generate a 40MHz unified clock signal to control the sensor to acquire data synchronously at a 1kHz sampling rate. The sampled data is transmitted to subsequent modules via SPI.

[0049] Data preprocessing module: Using Xilinx Zynq-7020 FPGA, it implements adaptive two-stage filtering and dual calibration in hardware. At the same time, it converts the calibrated data into RGB format color magnetic field cloud map, with the magnetic field strength from 0 to 5 mT corresponding to the blue to red gradient. The cloud map data is transmitted via Ethernet.

[0050] Feature extraction module: Using TITMS320C6748DSP, a program is written to calculate the spatial gradient Sobel operator, the peak change rate in the time domain, and the frequency domain FFT transform to extract multi-dimensional features and integrate them into a 256-dimensional feature vector, which is then sent to the intelligent recognition module via SPI.

[0051] Intelligent recognition module: It adopts the NVIDIA Jetson Nano embedded AI chip, runs the SVM and lightweight CNN dual model deployed by the TensorFlow Lite framework, and completes the classification and recognition within 0.6ms after receiving the feature vector and magnetic field cloud map. The recognition result is sent to the control linkage module via Ethernet.

[0052] Control and linkage module: adopts Siemens S7-1200 PLC. After receiving the identification results, it controls the electromagnetic baffle of the sorting mechanism through the digital output interface. At the same time, it links with the conveyor belt controller through the Modbus protocol to coordinate the conveying speed and sorting action.

[0053] Model self-updating module: It adopts an industrial computer, configured with an Intel Core i5 processor and 8GB of memory, runs a Python incremental training program, receives new labeled samples via FTP, starts training after accumulating 500 samples, and transmits the updated model parameters to the intelligent recognition module after training is completed.

[0054] Each module is connected via industrial Ethernet and SPI, adapting to dusty and vibrating industrial environments to achieve automated and high-precision operation of magnetic ore identification and sorting.

[0055] The present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is running, it controls the device containing the computer-readable storage medium to execute any of the above-mentioned magnetic ore identification and control methods based on Hall sensors.

[0056] Specifically, the computer-readable storage medium uses an industrial-grade SSD with a capacity of 128GB, a read / write speed of >500MB / s, and a temperature tolerance of -40℃ to 85℃. The stored computer program is written in C++ and Python and includes code for six major functional modules: magnetic field acquisition driver, data preprocessing, feature extraction, intelligent recognition, control linkage, and model self-updating.

[0057] When the storage medium is connected to the industrial computer or embedded AI chip of the above-mentioned device, the program runs automatically after the device is powered on: First, the hardware such as sensors, FPGA, and PLC is initialized, and then the acquisition drive is called to control the sensor array to collect magnetic field signals; the signal is transmitted to the preprocessing module to complete filtering and calibration and generate a cloud map; then the feature extraction module calculates a 256-dimensional vector; the intelligent recognition module loads dual-model classification; the control linkage module sends the results to the PLC to control sorting; the model self-update module monitors the number of new samples, and after reaching the target, it starts incremental training and updates the parameters.

[0058] This storage medium connects to devices via USB or industrial bus, facilitating program updates and maintenance, while its industrial-grade design ensures stable operation in the field.

[0059] In summary, this invention utilizes a modular matrix Hall sensor array, followed by adaptive dual-stage filtering and dual calibration to improve signal accuracy and generate a color magnetic field cloud map. Subsequently, it constructs a three-dimensional feature system integrating 256-dimensional feature vectors in the spatial-temporal-frequency domain, solving the problems of single-column sensor signal characteristics and signal distortion under heavy load. The invention also combines dual models to improve recognition accuracy and efficiency. Furthermore, it uses incremental training to update the model to adapt to ore changes, controlling the linkage to accurately separate concentrate and tailings, thus comprehensively improving the accuracy, stability, and adaptability of magnetic ore identification and sorting.

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

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

Claims

1. A magnetic ore identification and control method based on a Hall sensor, characterized in that, include: S1. A modular matrix Hall sensor array system is used to synchronously acquire the magnetic field signal of the ore to be tested. The array system includes multiple Hall effect sensor units, each sensor unit integrates a temperature compensation module, and the spacing between the sensor units is between 1mm and 2mm. A high-speed multi-channel synchronous acquisition module is used to control the sensor units to acquire data synchronously with a unified clock signal. S2. Adaptive two-stage filtering and dual calibration processing are performed on the magnetic field signal. The dual calibration processing includes zero-point drift calibration and environmental magnetic field calibration. A color magnetic field cloud map of the ore is generated based on the calibrated data. S3. Extracting multi-dimensional features: Based on the magnetic field cloud map and magnetic field signal, a three-dimensional feature system of "space-time domain-frequency domain" is constructed to extract multi-dimensional magnetic field features of the ore and integrate them into a 256-dimensional feature vector; S4. A hierarchical classification model with dual-model fusion is used to classify and identify feature vectors. The hierarchical classification model includes a first-level model for fast screening and a second-level model for fine classification. S5. Based on the classification and identification results, trigger the industrial control module to link with the sorting mechanism to sort the ore; S6. Perform incremental training on the hierarchical classification model that integrates the two models to update the model parameters.

2. The magnetic ore identification and control method based on a Hall sensor according to claim 1, characterized in that, In S1, the modular matrix Hall sensor array system supports multi-unit splicing expansion to adapt to the ore particle size requirements of different production lines; and the high-speed multi-channel synchronous acquisition module is connected to the sensor unit through low-impedance wires to ensure that the signal synchronization error of each sensor is less than 1ms.

3. The magnetic ore identification and control method based on a Hall sensor according to claim 1, characterized in that, In S2, the adaptive two-stage filtering includes: a first-stage adaptive low-pass filter that automatically adjusts the filter cutoff frequency according to the ore movement speed; and a second-stage wavelet denoising that specifically filters out pulse interference in the industrial environment.

4. The magnetic ore identification and control method based on a Hall sensor according to claim 1, characterized in that, In S2, zero-point drift calibration includes periodically and automatically collecting sensor output values ​​when no ore passes through and updating the zero-point reference; environmental magnetic field calibration includes setting an environmental magnetic field reference sensor at the edge of the detection area, collecting the environmental background magnetic field in real time, and removing background magnetic field interference from the detection data.

5. The magnetic ore identification and control method based on a Hall sensor according to claim 1, characterized in that, In S3, the multi-dimensional magnetic field features extracted from the ore include: spatial dimension features, including spatial gradient, spatial texture and edge contour; temporal dimension features, including magnetic field signal duration and magnetic field peak change rate; and frequency domain features, which are extracted by performing Fourier transform on the magnetic field signal to extract feature frequencies.

6. The magnetic ore identification and control method based on a Hall sensor according to claim 1, characterized in that, In S4, the first-level model is a support vector machine model, which performs fast screening based on spatial gradient and magnetic field peak features; the second-level model is a lightweight convolutional neural network model, which performs fine classification by inputting magnetic field cloud map and feature vector.

7. The magnetic ore identification and control method based on a Hall sensor according to claim 1, characterized in that, In step S5, controlling the sorting mechanism to perform ore sorting based on the classification and identification results includes: If the magnetic ore is determined to be usable, the industrial control module triggers the sorting mechanism to guide the magnetic ore into the concentrate channel. If the magnetic ore is determined to be tailings, the industrial control module triggers the sorting mechanism to guide the magnetic ore into the tailings channel.

8. The magnetic ore identification and control method based on a Hall sensor according to claim 1, characterized in that, In step S6, incremental training of the hierarchical classification model fused with dual models includes: manually labeling new samples through sample labeling units, integrating the new samples into the training set, and performing incremental training to update the dual model parameters.

9. A magnetic ore identification and control device based on a Hall sensor, used to execute the magnetic ore identification and control method based on a Hall sensor as described in any one of claims 1 to 8, characterized in that, include: The magnetic field acquisition module is used to synchronously acquire the magnetic field signal of the ore to be tested using a modular matrix Hall sensor array system that includes multiple Hall effect sensor units, each of which integrates a temperature compensation module. The spacing between the sensor units is between 1mm and 2mm. The array system uses a high-speed multi-channel synchronous acquisition module to control the sensor units to acquire the data synchronously with a unified clock signal. The data preprocessing module is used to perform adaptive two-stage filtering and two-calibration processing on the magnetic field signal. The two-calibration processing includes zero-point drift calibration and environmental magnetic field calibration, and generates a color magnetic field cloud map of the ore based on the calibrated data. The feature extraction module is used to construct a three-dimensional feature system of "space-time domain-frequency domain" based on magnetic field cloud map and magnetic field signal, extract multi-dimensional magnetic field features of ore and integrate them into a 256-dimensional feature vector; The intelligent recognition module is used to classify and recognize feature vectors using a hierarchical classification model that integrates two models. The hierarchical classification model includes a first-level model for fast screening and a second-level model for fine classification. The control linkage module is used to trigger the industrial control module to link the sorting mechanism to perform ore sorting based on the classification and identification results. The model self-update module is used to incrementally train the hierarchical classification model fused with the two models in order to update the model parameters.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform a magnetic ore identification and control method based on a Hall sensor as described in any one of claims 1 to 8.

Images