AI intelligent metal material classification proximity sensing device and method

Abstract

The invention discloses an AI intelligent metal material classification proximity sensing device and method.The device comprises a power module, a sensing module, a central controller, a communication module, a display module and a storage module.Metal proximity sensing detection is conducted through an LC high-frequency oscillation circuit, and metal material category judgment is achieved through AI intelligent fusion classification; an application environment simulation platform is built, an LC signal self-adaptive calibration mechanism is established to carry out anti-interference compensation, metal material classification requirements in a dynamic industrial environment are met, the application performance of the device is ensured, data transmission is carried out through multiple types of external communication circuits, and the data transmission efficiency is improved. And a liquid crystal screen is used for visual display management, so that man-machine interaction of metal material detection data is realized, the category of the metal material can be quickly identified in a short-distance range, the classification requirement of the novel metal material is dynamically met, and the classification precision and efficiency are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent sensing technology, and in particular to a proximity sensing device and method with AI intelligent metal material classification. Background Art

[0002] In recent years, significant technological progress has been made in metal material sorting devices. The application areas of metal material sorting devices have continued to expand, and they are widely used in many fields such as construction, transportation, automobiles, railways, water conservancy, energy, and electricity. With the advancement of technology and the growth of demand, the metal products industry has achieved remarkable development, the market scale has continued to expand, and the number of enterprises has continued to increase. This has also led to technological progress in the steel industry, mainly reflected in shortening processes, reducing steps, reducing energy consumption and costs, and improving quality. This has promoted the development of the steel industry in an intensified direction. Especially in terms of environmental protection and energy conservation, the research and application of metal material sorting devices have also received increasing attention to meet the global requirements for sustainable development and environmental protection.

[0003] However, traditional metal material classification methods rely primarily on manual experience or simple proximity sensor switch quantity identification of metal categories, resulting in low efficiency, high misjudgment rates, and an inability to distinguish complex alloys. Existing technologies, such as X-ray fluorescence spectroscopy (XRF), can identify some metal components, but the equipment is expensive and relies on expert analysis, making real-time classification difficult and impacting metal material classification efficiency. Furthermore, the complex use scenarios of metal materials mean that the sensor's application environment may interfere with the accuracy and stability of metal detection, resulting in a decrease in the sensor's metal detection performance. In view of the above technical defects, a solution is now proposed. Summary of the Invention

[0004] The purpose of the present invention is to solve the above-mentioned technical defects of low recognition efficiency, high misjudgment rate, expensive equipment and reliance on expert analysis, difficulty in achieving real-time classification, affecting the efficiency of metal material classification, and weak anti-interference ability of the sensing device through proximity sensor technology and AI learning.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A proximity sensing method for metal material classification with AI intelligence includes the following steps: Step 1: Metal proximity sensing detection and collection of multi-dimensional monitoring information: The device is powered on and starts the metal sensing detection mode. The metal object to be detected is brought close to the induction magnetic tank and sensed through the LC high-frequency oscillation circuit. The multi-dimensional monitoring information includes LC oscillation signal data and application environment data. Step 2: LC oscillation signal data preprocessing: Preprocess the LC oscillation signal data to obtain LC effective signal data and extract key features, thereby establishing a fusion feature vector for classification training; Step 3: AI intelligent fusion classification and output of metal material detection data: By building a hierarchical multimodal classifier, the probability distribution of the metal material to be detected and the category label is obtained, and it is integrated into the metal material detection data, thereby outputting the prediction results of the metal material and performing metal material classification judgment; Step 4: Establish an LC signal adaptive calibration mechanism: Compare and evaluate the metal classification accuracy by comparing the predicted and actual results. Combined with the metal classification response speed, the metal material detection performance is comprehensively evaluated. An application environment simulation platform is then built and the application environment data is analyzed to establish an LC signal adaptive calibration mechanism for anti-interference compensation and dynamically adjust and update weight parameters. Step 5: The metal material detection data is stored regularly, transmitted and displayed externally, thereby performing human-computer interaction of the metal material detection data.

[0006] Furthermore, the collection and preprocessing process of multi-dimensional monitoring information is as follows: The application environment data includes temperature parameters, electromagnetic parameters and vibration parameters, which are collected by monitoring sensor equipment in the application environment where the device is located; Acquire LC oscillation signal data through the output signal of the LC high-frequency oscillation circuit; The LC oscillation signal data is preprocessed to obtain the LC effective signal data. The specific process is as follows: Eliminate interference signals through IIR band-stop filters and normalize the signal amplitude; Extract time domain features and frequency domain features through normalized LC oscillation signal data; Time domain characteristics include amplitude parameters, time parameters and statistical parameters; Among them, the amplitude parameters include peak amplitude and amplitude variance; the time parameters include rise time and settling time; the statistical parameters include mean, standard deviation, skewness and kurtosis of the signal amplitude; Frequency domain features include spectrum parameters and time-frequency parameters; Among them, the spectrum parameters include center frequency offset, bandwidth and harmonic components; the time-frequency parameters include signal time-frequency distribution histogram and frequency change rate; The time domain features and frequency domain features of the LC oscillation signal data are integrated and marked as LC effective signal data.

[0007] Furthermore, the phase-amplitude joint key features are extracted through the LC effective signal data. The specific process is as follows: A set V is established through the LC effective signal data, and the elements of the set V refer to the characteristic parameter indicators of the LC effective signal data; The number of characteristic parameter indicators of LC effective signal data is marked as a, any characteristic parameter indicator is marked as X, and the value of the characteristic parameter indicator X is marked as x; Establish a set Y of category labels of metal materials, mark the number of category labels of metal materials as n, and mark any category label of metal materials as y; By setting N0 metal material samples and performing probability statistics, the mutual information value MI of the characteristic parameter index of the LC effective signal data and the category label is calculated, and the threshold of the mutual information value MI is set. , extract mutual information value MI not less than the threshold Characteristic parameter indicators; The extracted mutual information value MI is not less than the threshold The characteristic parameter index is integrated and marked as the initial characteristic vector ; Then the initial eigenvector The original signal data Performing short-time Fourier transform to obtain LC signal amplitude information, where the LC signal amplitude information includes signal amplitude; Original signal data Signal undergoing short-time Fourier transform , through the signal Get the signal amplitude; Then, the original signal data is transformed by Hilbert transform Performing analysis to extract LC signal phase information, including instantaneous phase; By initial eigenvector The amplitude information and phase information of the LC signal are fused to generate a fusion feature vector .

[0008] Furthermore, a hierarchical multimodal classifier is established. The specific process is as follows: Set the number of categories of the first category of metal materials to Ma, use the fully connected neural network as the first layer classifier, and input the fusion feature vector , let the number of neurons in the input layer of the first classifier be Na, the number of neurons in the hidden layer be Nb, and the number of neurons in the output layer be Ma; The weight matrix from the input layer to the hidden layer is obtained as W1, and the bias vector is marked as b1; The weight matrix from the hidden layer to the output layer is W2, and the bias vector is marked as b2; Then the output of the hidden layer is h1, and the output of the output layer is u1: , It refers to the probability that the metal material belongs to the Math category; Determine the category of the first category as g, and set the number of the second subcategories of category g of the first category to Mg; Use the fully connected neural network as the second layer classifier and follow the above operation to obtain the output of the hidden layer as h2 and the output layer as u2; Then, through the hierarchical classification of the hierarchical multimodal classifier, the complete probability distribution P of metal materials in all categories is obtained: , n refers to the total number of categories of metal materials, Pn refers to the probability of being in the nth category; Any metal material is marked as i, so as to obtain the probability distribution of the i and n categories of metal materials to be detected, and integrate them into the metal material detection data.

[0009] Furthermore, the specific process of evaluating the metal classification accuracy is as follows: The actual category label of the metal material is marked as E: ; The predicted class label of metal material is marked as P: ; By combining the actual category label E of the metal material with the predicted category label P, the cross entropy loss function F is obtained: ; Among them, Ei is the actual probability of any label i, and Pi is the predicted probability of any label i; The metal classification accuracy is evaluated by the cross entropy loss function F.

[0010] Furthermore, the specific process of evaluating the detection performance of metal materials is as follows: By detecting the time from the metal material approaching to the output of the classification result, the response time Tx of the metal material is monitored to evaluate the metal classification response speed; By analyzing the volatility of the cross entropy loss function F and the metal classification response time Tx, the application performance score is comprehensively obtained. , thereby evaluating the metal material detection performance.

[0011] Furthermore, the specific process of establishing the LC signal adaptive calibration mechanism is as follows: The temperature parameters, electromagnetic parameters and vibration parameters of the application environment data are used as independent variables, and the number of categories of the independent variables is marked as , the application performance score As the dependent variable, the multiple regression function G is constructed by fitting; Then, by dynamically adjusting the weight matrix W and the bias vector b, an adaptive calibration mechanism for the LC signal is established, and the application performance is scored in real time through application environment data evaluation. The influence of the time domain and frequency domain features of the hierarchical multimodal classifier are dynamically adjusted to perform adaptive adjustment optimization and anti-interference compensation of the device.

[0012] A proximity sensing device with AI intelligent metal material classification includes a power module, a sensor module, a central controller, a communication module, a display module, and a storage module, wherein the modules are electrically connected. The device applies the aforementioned proximity sensing method with AI intelligent metal material classification. The power module is used to supply power to the device, and the power module includes a power circuit; The sensor module is used for metal proximity sensing detection: the sensor module includes an LC high-frequency oscillation circuit, which is used to sense the metal object when it is close to the induction magnetic tank. The central controller is used for data core control processing: the central controller uses an MCU microcontroller and sets a core CPU circuit to perform data processing and control, and sets a clock circuit to control the execution order and duration of each operation; The communication module is used to communicate the metal material detection data to the outside world: the communication module includes Ethernet communication circuit, RS485 communication circuit and RS232 communication circuit; The display module is used for data output and human-computer interaction: the display module includes an LCD display circuit, and the LCD screen uses a TFT LCD screen; The storage module is used to store metal material detection data: the storage module includes a data power-off storage circuit, which is used to store the detection data of different metal materials after being processed by the central controller.

[0013] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: The present invention uses an LC high-frequency oscillation circuit for metal proximity sensing detection and AI intelligent fusion classification to determine the metal material category, achieving a layer-by-layer fine classification effect from the major metal material categories to specific subcategories. Furthermore, by building an application environment simulation platform and establishing an LC signal adaptive calibration mechanism for anti-interference compensation, the device adapts to the metal material classification requirements in dynamic industrial environments and ensures the application performance of the device. Furthermore, through multiple types of external communication circuits for data transmission and a liquid crystal display for visual display management, human-computer interaction of metal material detection data is achieved. The present invention can quickly identify the type of metal materials at close range and is suitable for scenarios such as industrial sorting, waste recycling, and quality control. By responding to the metal material situation and using a self-learning algorithm, it can dynamically adapt to the classification needs of new metal materials, significantly improving classification accuracy and efficiency. The module structures of the present invention are interconnected, and the hardware cost is low, and the communication types are diverse, which facilitates direct data interaction input and output with external modules. It is also equipped with a liquid crystal display, which makes human-computer interaction more convenient, improves self-learning capabilities, supports online learning and incremental updates, and adapts to the metal material classification needs in dynamic industrial environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 shows a schematic diagram of the connection of the device modules of the present invention; Figure 2 A schematic diagram showing the steps of the workflow of the present invention is shown; Figure 3 A schematic diagram showing the circuit principle of the present invention is shown; Figure 4 Shows the overall circuit design diagram of the device of the present invention; Figure 5 A partial enlarged view of the power supply circuit, data power-off storage circuit, LC oscillation proximity sensor sensing circuit, and liquid crystal display circuit of the circuit design diagram of the present invention is shown; Figure 6 A partial enlarged view of the communication module circuit of the circuit design diagram of the present invention is shown; Figure 7 A partial enlarged view of the core CPU circuit and clock circuit of the circuit design diagram of the present invention is shown. DETAILED DESCRIPTION

[0015] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0016] Example 1: like Figure 1-Figure 7 As shown, an AI intelligent metal material classification proximity sensing device includes a power module, a sensor module, a central controller, a communication module, a display module and a storage module, and each module is electrically connected; The power module is used to supply power to the device, and the power module includes a power circuit; The sensor module is used for metal proximity sensing detection: the sensor module includes an LC high-frequency oscillation circuit, which is used to sense the metal object when it is close to the induction magnetic tank. The central controller is used for data core control processing: the central controller uses an MCU microcontroller and sets a core CPU circuit to perform data processing and control, and sets a clock circuit to control the execution order and duration of each operation; The communication module is used to communicate the metal material detection data to the outside world: the communication module includes Ethernet communication circuit, RS485 communication circuit and RS232 communication circuit; The display module is used for data output and human-computer interaction: the display module includes an LCD display circuit, and the LCD screen uses a TFT LCD screen; The storage module is used to store metal material detection data: the storage module includes a data power-off storage circuit, which is used to store the detection data of different metal materials after being processed by the central controller.

[0017] The hardware installation environment requirements for the device are as follows: Electromagnetic environment: The device should be kept away from high-power electromagnetic equipment, such as inverters and transformers, at least 1 meter away to prevent electromagnetic interference from affecting the stability of the LC oscillation signal; Temperature and humidity: The working environment temperature range is -10℃~50℃, and the relative humidity does not exceed 85% (non-condensing). If it exceeds this range, constant temperature and humidity equipment must be installed; Vibration conditions: The vibration amplitude at the installation location should be less than 0.5g0 (g0 is the acceleration of gravity). If the vibration is severe, a shock-absorbing base must be used to fix the sensor to prevent mechanical displacement from affecting detection accuracy.

[0018] The installation steps of each module of the device are as follows: Power module: Connect the AC220V power cord to the input of the power module. Connect the output to the central controller and other modules through a shielded cable, ensuring the correct polarity. The power module must be fixed separately to the metal base plate and properly grounded. Sensing module: Fix the induction magnetic tank of the LC high-frequency oscillation circuit on the detection bracket. Adjust the bracket height so that the induction magnetic tank is perpendicular to the surface of the metal material to be measured. Set the initial detection distance to 10mm. Use a shielded cable to connect the sensing module to the signal input terminal of the central controller. Central controller: Install the MCU microcontroller motherboard in a metal chassis and connect the module interfaces through cables. The core CPU circuit needs to be equipped with a heat sink. Avoid dense wiring near the clock circuit crystal oscillator to prevent electromagnetic coupling. Communication module: The Ethernet communication circuit is connected to the industrial switch through the RJ45 interface. The RS485 / RS232 communication circuit uses a dedicated cable to connect to the host computer or PLC. The communication cable must be twisted and shielded, and the terminal resistor must be connected according to the specifications. Among them, the RS485 terminal resistor is 120Ω; Display module: The TFT LCD screen is connected to the display interface of the central controller through an FPC cable and fixed to the front panel of the chassis. Make sure the cable is not bent. The touch screen needs to be calibrated. Storage module: A data power-off storage circuit, such as an EEPROM or Flash chip, is installed on the central controller motherboard, close to the CPU interface to ensure that the data read and write speeds match. It is then connected to an external storage expansion interface, such as an SD card slot. It must also be dustproof and waterproof.

[0019] The system wiring inspection process is as follows: Use a multimeter to check whether the power supply output voltage is stable and the fluctuation range does not exceed ±5%; Check whether the connection cables of each module are secure and whether the shielding layer is reliably grounded to avoid signal interference caused by loose connections; Perform an open-circuit test on the LC oscillation circuit of the sensor module and observe the oscillation waveform using an oscilloscope. Under normal circumstances, it should present a stable sine wave with a frequency deviation of no more than ±1%.

[0020] The device applies the following AI intelligent metal material classification proximity sensing method, including the following steps: S1, metal proximity sensing detection and collection of multi-dimensional monitoring information: The device is powered on and starts the metal sensing detection mode. The metal object to be detected is brought close to the induction magnetic tank and sensed through the LC high-frequency oscillation circuit. The multi-dimensional monitoring information includes LC oscillation signal data and application environment data; The collection and preprocessing process of multi-dimensional monitoring information is as follows: The application environment data includes temperature parameters, electromagnetic parameters and vibration parameters, which are collected by monitoring sensor equipment in the application environment where the device is located; The LC oscillation signal data is obtained through the output signal of the LC high-frequency oscillation circuit.

[0021] S2, LC oscillation signal data preprocessing: Preprocess the LC oscillation signal data to obtain LC effective signal data and extract the phase-amplitude joint key features to establish a fusion feature vector for classification training; S2-1, pre-processing is performed on the LC oscillation signal data to obtain the LC effective signal data. The specific process is as follows: Eliminate interference signals through IIR band-stop filters and normalize the signal amplitude; Extract time domain features and frequency domain features through normalized LC oscillation signal data; Time domain characteristics include amplitude parameters, time parameters and statistical parameters; Among them, the amplitude parameters include peak amplitude and amplitude variance; the time parameters include rise time and settling time; the statistical parameters include mean, standard deviation, skewness and kurtosis of the signal amplitude; It should be noted that: Peak amplitude refers to the maximum amplitude of the oscillation signal, which is used to monitor the degree of attenuation of the metal to the magnetic field; Amplitude variance is used to monitor the fluctuation of the signal amplitude, thereby distinguishing metals with different surface roughness; Rise time refers to the time it takes for the signal to go from valley to peak, which is used to monitor the approach speed of the metal; Settling time refers to the time it takes for the signal to enter a steady state, which is used to evaluate the conductivity of the metal; Statistical parameters are used to monitor the distribution characteristics of the signal; Frequency domain features include spectrum parameters and time-frequency parameters; Among them, the spectrum parameters include center frequency offset, bandwidth and harmonic components; the time-frequency parameters include signal time-frequency distribution histogram and frequency change rate; It should be noted that the center frequency offset refers to the difference between the oscillation frequency and the empty field frequency, and is used to evaluate the magnetic permeability and electrical conductivity of metals; the bandwidth is used to monitor the quality factor Q of the oscillation circuit; the harmonic component is used to distinguish nonlinear materials such as alloys; and the time-frequency parameter is used to analyze the time-varying frequency characteristics of metals. The time domain features and frequency domain features of the LC oscillation signal data are integrated and marked as LC effective signal data.

[0022] S2-2, extract phase-amplitude joint key features through LC effective signal data. The specific process is as follows: A set V is established through the LC effective signal data. The elements of the set V refer to characteristic parameter indicators of the LC effective signal data, including peak amplitude, amplitude variance, rise time, and settling time. The number of characteristic parameter indicators of LC effective signal data is marked as a, any characteristic parameter indicator is marked as X, and the value of the characteristic parameter indicator X is marked as x; Establish a set Y of category labels of metal materials, mark the number of category labels of metal materials as n, and mark any category label of metal materials as y; By setting N0 metal material samples and performing probability statistics, the mutual information value MI of the characteristic parameter index of the LC effective signal data and the category label is calculated, and the threshold of the mutual information value MI is set. , extract mutual information value MI not less than the threshold Characteristic parameter indicators; ; in, It refers to the joint probability that the feature parameter index X takes the value x and the class label is y; It refers to the marginal probability of the characteristic parameter index X taking the value x; It refers to the marginal probability of the class label y; The extracted mutual information value MI is not less than the threshold The characteristic parameter index is integrated and marked as the initial characteristic vector ; Then the initial eigenvector The original signal data Performing short-time Fourier transform to obtain LC signal amplitude information, where the LC signal amplitude information includes signal amplitude; Original signal data Signal undergoing short-time Fourier transform : ; f is the frequency, is the window function; Through the signal Get the signal amplitude; Then, the original signal data is transformed by Hilbert transform Performing analysis to extract LC signal phase information, including instantaneous phase; Among them, the signal after Hilbert transform for: ; Analyzing the signal for: ; Instantaneous phase for: ; By initial eigenvector The amplitude information and phase information of the LC signal are fused to generate a fusion feature vector .

[0023] S3, AI intelligent fusion classification and output of metal material detection data: By building a hierarchical multimodal classifier, the probability distribution of categories i and n of the metal materials to be detected is obtained and integrated into the metal material detection data, thereby outputting the prediction results of the metal materials and realizing the classification and judgment of the metal materials; Through the probability distribution of the hierarchical multimodal classifier, a layer-by-layer fine classification effect from the major categories of metal materials to specific subcategories is achieved. The specific process is as follows: S3-1, set the number of categories of the first category of metal materials to Ma, and the categories of the first category include ferromagnetic, non-ferromagnetic, and alloy; Through the fully connected neural network as the first layer classifier, the fusion feature vector is input , let the number of neurons in the input layer of the first classifier be Na, the number of neurons in the hidden layer be Nb, and the number of neurons in the output layer be Ma; The input layer is used to analyze the dimension of the associated feature vector; the hidden layer is used to analyze nonlinear complex features; Then the weight matrix from the input layer to the hidden layer is W1: , R represents the set of real numbers; label the bias vector as b1: ; The bias vector b1 is used to avoid the extreme case where the input feature is 0 and enhance the model fitting ability; The weight matrix from the hidden layer to the output layer is W2: ; Label the bias vector b2: ; The bias vector b2 is used to adjust the baseline of the output probability to avoid probability prediction deviation caused by the integration offset of the hidden layer output; The output of the hidden layer is h1: ; Among them, σ is the activation function, and the ReLU function is selected, that is, , thus retaining the positive correlation characteristics; It means transposing the rows and columns of the W1 matrix; The output of the output layer is u1: ; Among them, the softmax function is used to convert the output into a probability distribution, that is ; zj is the jth element of the input vector, K is the dimension of the output vector, k is any dimension, K=Ma in the first layer classifier, and the probability distribution u1 of the input metal material belonging to Ma major categories is obtained through the softmax function: , It refers to the probability that the metal material belongs to the Math category; The first category of metal materials can be determined by sorting in descending order of probability.

[0024] S3-2, determining the category of the first category to be g, and setting the number of the second subcategories of category g of the first category to Mg; Through the fully connected neural network as the second layer classifier, the fusion feature vector is input , set the number of input neurons of the second layer classifier to Nc, the number of hidden layer neurons to Nd, and the number of output layer neurons to Mg; The weight matrix from the input layer to the hidden layer is W3: ; Label the bias vector b3: ; The weight matrix from the hidden layer to the output layer is W4: ; Label the bias vector b4: ; The output of the hidden layer is h2: ; The output of the output layer is u2: ; Therefore, the category of the second largest category of metal materials can be determined by outputting the descending order of u2 probability.

[0025] S3-3, through hierarchical classification of hierarchical multimodal classifier, the complete probability distribution P of metal materials in all categories is obtained: , n refers to the total number of categories of metal materials, Pn refers to the probability of being in the nth category; Mark any metal material as i to obtain the probability distribution of the i and n categories of metal materials to be detected, and integrate them into metal material detection data to output the prediction results of the metal material and realize the classification and judgment of the metal material.

[0026] S4, establish an LC signal adaptive calibration mechanism: By evaluating the metal material detection performance, building an application environment simulation platform and analyzing the application environment data, an LC signal adaptive calibration mechanism is established to perform anti-interference compensation and dynamically adjust and update weight parameters; S4-1, mark the actual category label of the metal material as E: ; The predicted class label of metal material is marked as P: ; By combining the actual category label E of the metal material with the predicted category label P, the cross entropy loss function F is obtained: ; Among them, Ei is the actual probability of any label i, and Pi is the predicted probability of any label i; The metal classification accuracy is evaluated by the cross-entropy loss function F. When the true label is i, Ei=1. At this time: if the probability Pi of the predicted label i is higher and closer to 1, log(Pi) is closer to 0, the closer the cross-entropy loss function F is to 0, and the metal classification accuracy is higher; if the probability Pi of the predicted label i is lower and closer to 0, log(Pi) is closer to negative infinity, the higher the cross-entropy loss function F is, and the lower the metal classification accuracy.

[0027] S4-2, the specific process of evaluating the detection performance of metal materials is as follows: By detecting the time from the metal material approaching to the output of the classification result, the response time Tx of the metal material at different detection distances is monitored to evaluate the metal classification response speed; When the response time Tx is shorter, the metal classification response speed of the metal material at the detection distance is higher; The application performance score is obtained by combining the cross entropy loss function F with the metal classification response time Tx. : ; in, and They are the fluctuation coefficients of metal classification accuracy and metal classification response speed respectively. By setting the data evaluation period Tz, the metal classification accuracy and metal classification response speed corresponding to Nz time nodes are calculated regularly, thereby calculating the standard deviation of metal classification accuracy and metal classification response speed, and marking them as the fluctuation coefficients of metal classification accuracy respectively. Fluctuation coefficient of metal classification response speed ; β1 and β2 are weight factors of metal classification accuracy and metal classification response speed respectively. The weight factors are preset after a large amount of experimental data testing, and the preset values ​​of β1 and β2 are both greater than 0. ; When the cross entropy loss function F and the metal classification response time Tx are lower, and the fluctuation coefficient and The lower it is, the higher the metal classification accuracy and metal classification response speed are, and the better the stability is, thus comprehensively evaluating the higher the application performance of the sensing device. Then, by dynamically adjusting the weight matrix W and the bias vector b, the application performance score Gradually improve, thereby improving the application performance of the sensing device.

[0028] S4-3, the specific process of establishing the LC signal adaptive calibration mechanism is as follows: The temperature parameters, electromagnetic parameters and vibration parameters of the application environment data are used as independent variables, and the number of categories of the independent variables is marked as , the application performance score As the dependent variable, fit and construct the multiple regression function G: ;in, For the independent variables; Through the LC signal adaptive calibration mechanism, the application performance score of different characteristics is evaluated in real time according to the changes in application environment data such as temperature, electromagnetic and vibration parameters. The influence of the time domain and frequency domain features of the hierarchical multimodal classifier is dynamically adjusted to achieve adaptive adjustment and optimization of the device; By continuously interacting and learning with the environment, the classifier autonomously learns the optimal parameter adjustment strategy under different circumstances, so that the sensor device can maintain good performance in various complex environments. For example, when the ambient temperature changes significantly, the frequency of the LC oscillation signal may drift. At this time, the importance of the frequency-related frequency domain features may increase, and the model can automatically increase the weight of the frequency domain features. When there is strong electromagnetic interference in the environment, it may affect the phase and amplitude of the signal. At this time, the weights related to the time domain features may need to be adjusted. The specific adjustment operation needs to be dynamically controlled according to actual needs, and the adjustment operation results need to make the application performance score Improve, if the application performance score If it decreases, the feature-related weights will not be adjusted; By using the adaptive weight adjustment mechanism to perform anti-interference compensation, the classifier can better adapt to environmental changes, improve anti-interference ability and classification accuracy, and thus improve application efficiency.

[0029] S5, regularly stores the metal material detection data, transmits and displays it to the outside world, and realizes human-computer interaction of the metal material detection data; Among them, through the intuitive and friendly human-computer interaction interface, users can conveniently set system parameters, view classification results, perform data management and other operations through the graphical interface, which reduces the difficulty of operation and improves work efficiency; at the same time, through the multiple communication interfaces supported by the communication module, the connection between the sensor device and the remote monitoring center is realized, and users can remotely monitor the operating status of the sensor device in real time through the network, view classification data, perform remote parameter setting and fault diagnosis, and realize the intelligent management and remote operation and maintenance of the metal material classification process.

[0030] In summary, the present invention uses an LC high-frequency oscillation circuit to perform metal proximity sensing detection, and uses AI intelligent fusion classification to determine the metal material category, achieving a layer-by-layer fine classification effect from the major metal material categories to specific subcategories. Furthermore, by building an application environment simulation platform and establishing an LC signal adaptive calibration mechanism for anti-interference compensation, the device adapts to the metal material classification requirements in dynamic industrial environments and ensures the application performance of the device. Furthermore, through multiple types of external communication circuits for data transmission and a liquid crystal screen for visual display management, human-computer interaction of metal material detection data is achieved. The present invention can quickly identify the type of metal materials at close range and is suitable for scenarios such as industrial sorting, waste recycling, and quality control. By responding to the metal material situation and using a self-learning algorithm, it can dynamically adapt to the classification needs of new metal materials, significantly improving classification accuracy and efficiency. The module structures of the present invention are interconnected, and the hardware cost is low, and the communication types are diverse, which facilitates direct data interaction input and output with external modules. It is also equipped with a liquid crystal display, which makes human-computer interaction more convenient, improves self-learning capabilities, supports online learning and incremental updates, and adapts to the metal material classification needs in dynamic industrial environments.

[0031] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0032] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0033] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A proximity sensing method for metal material classification with AI, characterized by: The following steps are involved: Step 1: Metal proximity sensing detection and collection of multi-dimensional monitoring information: The device is powered on and starts the metal sensing detection mode. The metal object to be detected is brought close to the induction magnetic tank and sensed through the LC high-frequency oscillation circuit. The multi-dimensional monitoring information includes LC oscillation signal data and application environment data. Step 2: LC oscillation signal data preprocessing: Preprocess the LC oscillation signal data to obtain LC effective signal data and extract key features, thereby establishing a fusion feature vector for classification training; Step 3: AI intelligent fusion classification and output of metal material detection data: By building a hierarchical multimodal classifier, the probability distribution of the metal material to be detected and the category label is obtained, and it is integrated into the metal material detection data, thereby outputting the prediction results of the metal material and performing metal material classification judgment; Step 4: Establish an LC signal adaptive calibration mechanism: Compare and evaluate the metal classification accuracy by comparing the predicted and actual results. Combined with the metal detection response speed, the metal material detection performance is comprehensively evaluated. An application environment simulation platform is then built and the application environment data is analyzed to establish an LC signal adaptive calibration mechanism for anti-interference compensation and dynamically adjust and update the weight parameters. Step 5: The metal material detection data is stored regularly, transmitted and displayed externally, thereby performing human-computer interaction of the metal material detection data.

2. The AI ​​intelligent metal material classification proximity sensing method according to claim 1 is characterized by: The collection and preprocessing process of multi-dimensional monitoring information is as follows: The application environment data includes temperature parameters, electromagnetic parameters and vibration parameters, which are collected by monitoring sensor equipment in the application environment where the device is located; Acquire LC oscillation signal data through the output signal of the LC high-frequency oscillation circuit; The LC oscillation signal data is preprocessed to obtain the LC effective signal data. The specific process is as follows: Eliminate interference signals through IIR band-stop filters and normalize the signal amplitude; Extract time domain features and frequency domain features through normalized LC oscillation signal data; Time domain characteristics include amplitude parameters, time parameters and statistical parameters; Among them, the amplitude parameters include peak amplitude and amplitude variance; the time parameters include rise time and settling time; the statistical parameters include mean, standard deviation, skewness and kurtosis of the signal amplitude; Frequency domain features include spectrum parameters and time-frequency parameters; Among them, the spectrum parameters include center frequency offset, bandwidth and harmonic components; the time-frequency parameters include signal time-frequency distribution histogram and frequency change rate; The time domain features and frequency domain features of the LC oscillation signal data are integrated and marked as LC effective signal data.

3. The AI ​​intelligent metal material classification proximity sensing method according to claim 2 is characterized by: The phase-amplitude joint key features are extracted from the LC effective signal data. The specific process is as follows: A set V is established through the LC effective signal data, and the elements of the set V refer to the characteristic parameter indicators of the LC effective signal data; The number of characteristic parameter indicators of LC effective signal data is marked as a, any characteristic parameter indicator is marked as X, and the value of the characteristic parameter indicator X is marked as x; Establish a set Y of category labels of metal materials, mark the number of category labels of metal materials as n, and mark any category label of metal materials as y; By setting N0 metal material samples and performing probability statistics, the mutual information value MI of the characteristic parameter index of the LC effective signal data and the category label is calculated, and the threshold of the mutual information value MI is set. , extract mutual information value MI not less than the threshold Characteristic parameter indicators; The extracted mutual information value MI is not less than the threshold The characteristic parameter index is integrated and marked as the initial characteristic vector ; Then the initial eigenvector The original signal data Performing short-time Fourier transform to obtain LC signal amplitude information, where the LC signal amplitude information includes signal amplitude; The original signal data The signal undergoing short-time Fourier transform is marked as , through the signal Get the signal amplitude; Then, the original signal data is transformed by Hilbert transform Performing analysis to extract LC signal phase information, including instantaneous phase; By initial eigenvector The amplitude information and phase information of the LC signal are fused to generate a fusion feature vector .

4. The AI ​​intelligent metal material classification proximity sensing method according to claim 3 is characterized by: Establish a hierarchical multimodal classifier. The specific process is as follows: Set the number of categories of the first category of metal materials to Ma, use the fully connected neural network as the first layer classifier, and input the fusion feature vector , let the number of neurons in the input layer of the first classifier be Na, the number of neurons in the hidden layer be Nb, and the number of neurons in the output layer be Ma; The weight matrix from the input layer to the hidden layer is obtained as W1, and the bias vector is marked as b1; The weight matrix from the hidden layer to the output layer is W2, and the bias vector is marked as b2; Then the output of the hidden layer is h1, and the output of the output layer is u1: , It refers to the probability that the metal material belongs to the Math category; Determine the category of the first category as g, and set the number of the second subcategories of category g of the first category to Mg; Use the fully connected neural network as the second layer classifier, and follow the same operation as the first layer classifier to obtain the output of the hidden layer as h2 and the output layer as u2; Then, through the hierarchical classification of the hierarchical multimodal classifier, the complete probability distribution P of metal materials in all categories is obtained: , n refers to the total number of categories of metal materials, Pn refers to the probability of being in the nth category; Any metal material is marked as i, so as to obtain the probability distribution of the i and n categories of metal materials to be detected, and integrate them into the metal material detection data.

5. The AI ​​intelligent metal material classification proximity sensing method according to claim 4 is characterized by: The specific process for evaluating metal classification accuracy is: The actual category label of the metal material is marked as E: ; The predicted class label of metal material is marked as P: ; By combining the actual category label E of the metal material with the predicted category label P, the cross entropy loss function F is obtained: ; Among them, Ei is the actual probability of any label i, and Pi is the predicted probability of any label i; The metal classification accuracy is evaluated by the cross entropy loss function F.

6. The AI ​​intelligent metal material classification proximity sensing method according to claim 5, characterized in that: The specific process of evaluating the detection performance of metal materials is as follows: By detecting the time from the metal material approaching to the output of the classification result, the response time Tx of the metal material is monitored, thereby evaluating the metal classification response speed; By analyzing the volatility of the cross entropy loss function F and the metal classification response time Tx, the application performance score is comprehensively obtained. , thereby evaluating the metal material detection performance.

7. The AI ​​intelligent metal material classification proximity sensing method according to claim 6, characterized in that: The specific process of establishing the LC signal adaptive calibration mechanism is as follows: The temperature parameters, electromagnetic parameters and vibration parameters of the application environment data are used as independent variables, and the number of categories of the independent variables is marked as , the application performance score As the dependent variable, the multiple regression function G is constructed by fitting; Then, by dynamically adjusting the weight matrix W and the bias vector b, an adaptive calibration mechanism for the LC signal is established, and the application performance is scored in real time through application environment data evaluation. The influence of the time domain and frequency domain features of the hierarchical multimodal classifier are dynamically adjusted to perform adaptive adjustment optimization and anti-interference compensation of the device.

8. An AI-enabled intelligent metal material classification proximity sensing device, characterized by: The device comprises a power module, a sensor module, a central controller, a communication module, a display module and a storage module, and each module is electrically connected to each other; the device applies the AI ​​intelligent metal material classification proximity sensing method described in any one of claims 1 to 7 above; The power module is used to supply power to the device, and the power module includes a power circuit; The sensor module is used for metal proximity sensing detection: the sensor module includes an LC high-frequency oscillation circuit, which is used to sense the metal object when it is close to the induction magnetic tank. The central controller is used for data core control processing: the central controller uses an MCU microcontroller and sets a core CPU circuit to perform data processing and control, and sets a clock circuit to control the execution order and duration of each operation; The communication module is used to communicate the metal material detection data to the outside world: the communication module includes Ethernet communication circuit, RS485 communication circuit and RS232 communication circuit; The display module is used for data output and human-computer interaction: the display module includes an LCD display circuit, and the LCD screen uses a TFT LCD screen; The storage module is used to store metal material detection data: the storage module includes a data power-off storage circuit, which is used to store the detection data of different metal materials after being processed by the central controller.

Images