Material particle size monitoring methods, devices, equipment and storage media

Abstract

This application discloses a method, apparatus, device, and storage medium for material particle size monitoring, relating to the field of particle size monitoring technology. The method includes: acquiring material flow data and raw material vibration signals, filtering the raw material vibration signals to obtain pure material vibration signals; extracting multi-dimensional vibration feature signals from the pure material vibration signals in the time domain, frequency domain, and nonlinear dimension; inputting the multi-dimensional vibration feature signals and the pure material vibration signals into a particle size inversion system for deduction to obtain an initial material particle size percentage; calculating a compensation coefficient based on the initial material particle size percentage, the pure material vibration signals, and the material flow data; and calibrating the initial material particle size percentage based on the compensation coefficient to determine the target material particle size percentage. This application is adaptable to harsh industrial environments such as high-dust environments, reduces hardware costs and maintenance difficulty, automates the entire process from signal processing to particle size calibration, and features fast response speed and high accuracy.

Description

Technical Field

[0001] This application relates to the field of particle size monitoring technology, and in particular to a method, apparatus, equipment and storage medium for monitoring particle size of materials. Background Technology

[0002] Currently, the industry mainly relies on three technical solutions for monitoring the particle size of materials conveyed by belt conveyors: manual sampling and screening, laser particle size analyzer monitoring, and machine vision system monitoring.

[0003] However, manual sampling and sieving requires stopping the machine to collect samples and send them to the laboratory for analysis. This process is not only time-consuming, resulting in significant delays in monitoring results and failing to meet the needs of real-time production guidance, but also makes it difficult to reflect the particle size distribution of materials along the entire conveyor belt, leading to large data fluctuations and errors. Laser particle size analyzers are significantly affected by environmental factors. In humid or dusty conditions, the lens is easily contaminated and malfunctions, requiring regular cleaning or machine shutdown to replace parts. This maintenance is complex and costly. Furthermore, when the moisture content of the material exceeds 5%, changes in surface reflectivity can cause a sharp increase in measurement errors. In high humidity conditions, powder particle agglomeration can also lead to misjudging the proportion of large particles. Machine vision systems are highly dependent on lighting conditions. In low-brightness environments, additional lighting is required, while scattered light in high-dust environments can cause image blurring, significantly reducing the effective recognition rate. When encountering sticky and wet materials adhering to the conveyor belt in patches, they can be misjudged as oversized particles, further affecting the accuracy of monitoring.

[0004] The shortcomings of these existing technologies mean that they cannot meet the actual needs of industrial production in terms of real-time performance, environmental adaptability, cost control, anti-interference ability, and measurement accuracy, thus restricting the closed-loop control and optimization of the production process. Summary of the Invention

[0005] The main purpose of this application is to provide a material particle size monitoring method, device, equipment and storage medium, which is designed to be suitable for harsh industrial environments such as high dust and humidity, reduce hardware costs and maintenance difficulty, automate the entire process from signal processing to particle size calibration, have fast response speed and reduce human error.

[0006] To achieve the above objectives, this application proposes a method for monitoring material particle size, the method comprising:

[0007] Acquire material flow data and raw material vibration signal, and filter the raw material vibration signal to obtain a pure material vibration signal;

[0008] Extract the multi-dimensional vibration feature signals of the pure material vibration signal in the time domain, frequency domain, and nonlinear dimension;

[0009] The multi-dimensional vibration characteristic signal and the pure material vibration signal are input into the particle size inversion system for deduction to obtain the initial material particle size ratio.

[0010] A compensation coefficient is calculated based on the initial material particle size distribution, the vibration signal of the pure material, and the material flow rate data. Based on the compensation coefficient, the initial material particle size distribution is calibrated to determine the target material particle size distribution.

[0011] In one possible implementation, filtering the original material vibration signal to obtain a purified material vibration signal includes:

[0012] The initial material vibration signal is obtained by filtering out belt mechanical vibration noise with a frequency lower than a preset first frequency threshold and high-frequency noise generated by motor rotation with a frequency higher than a preset second frequency threshold from the original material vibration signal using a bandpass filter.

[0013] By using a power isolation circuit to suppress grounding interference in the industrial field and reduce the influence of electromagnetic noise on the initial material vibration signal, a pure material vibration signal is obtained.

[0014] In one possible implementation, extracting the multi-dimensional vibration feature signals of the pure material vibration signal in the time domain, frequency domain, and nonlinear dimension includes:

[0015] For the time domain dimension, N consecutive vibration signals within a preset time period are selected, and the root mean square value of the amplitude and the kurtosis corresponding to each vibration signal are calculated, where N is a positive integer;

[0016] For the frequency domain dimension, the vibration waveform of each vibration signal is transformed from the time domain to the frequency domain by fast Fourier transform, the frequency interval centered on the preset third frequency threshold and the preset fourth frequency threshold is located, and the total vibration energy in each frequency interval is calculated to obtain the vibration energy ratio.

[0017] For the aforementioned nonlinear dimension, calculate the sample entropy of the vibration signal of the pure material;

[0018] The analysis results from the time domain, frequency domain, and nonlinear dimensions are integrated to form a multidimensional vibration characteristic signal.

[0019] In one possible implementation, the step of converting the vibration waveforms of each vibration signal from the time domain to the frequency domain using a fast Fourier transform, locating frequency intervals centered on a preset third frequency threshold and a preset fourth frequency threshold, and calculating the sum of vibration energy within each frequency interval to obtain the vibration energy ratio includes:

[0020] After zero-padding of N consecutive vibration signals within a preset time period, a fast Fourier transform is performed to decompose each vibration signal in the time domain into a superposition of sinusoidal waves with different frequency components, and output the frequency amplitude correspondence.

[0021] Based on the frequency amplitude correspondence, the low-frequency range centered on the preset third frequency threshold and the high-frequency range centered on the preset fourth frequency threshold are located, and the sum of squares of the amplitudes of all frequency points in the low-frequency range and the high-frequency range are calculated to obtain the total energy of the low-frequency range and the total energy of the high-frequency range.

[0022] The ratio of the total energy in the low-frequency range to the total energy in the high-frequency range is taken as the vibration energy ratio.

[0023] In one possible implementation, the granular inversion system includes a convolutional neural network, a long short-term memory network, and a fully connected layer;

[0024] The step of inputting the multi-dimensional vibration characteristic signal and the pure material vibration signal into the particle size inversion system for deduction to obtain the initial material particle size ratio includes:

[0025] The multi-dimensional vibration feature signal and the pure material vibration signal are input into the particle size inversion system, and the pure material vibration signal is scanned by the convolutional neural network to extract local detail features.

[0026] The temporal context of the multi-dimensional vibration characteristic signal is analyzed by the long short-term memory network to determine whether the change in material particle size is a smooth switch or an instantaneous disturbance, and the temporal judgment result is output.

[0027] The multi-dimensional vibration characteristic signal, the local detail feature, and the timing judgment result are input into the fully connected layer so that the fully connected layer can output the mass ratio of coarse particles, medium particles, and fine particles to generate the initial material particle size ratio.

[0028] In one possible implementation, the step of analyzing the temporal context of the multi-dimensional vibration characteristic signal through the long short-term memory network to determine whether the material particle size change is a smooth transition or a transient disturbance, and outputting the temporal judgment result, includes:

[0029] A time window is set for the long short-term memory network, and the historical multidimensional vibration feature signal sequence within the time window is acquired and used as the historical feature sequence.

[0030] The multidimensional vibration feature signal and the historical feature sequence are input into the hidden layer of the long short-term memory network, and the weight ratio of the historical feature sequence and the multidimensional vibration feature signal is adaptively adjusted through a gating mechanism.

[0031] Calculate the gradient value of the change between the multi-dimensional vibration feature signal and the historical feature sequence. If the gradient value is within the range of a preset smoothing threshold, it is determined that the material particle size is smoothly switched. The multi-dimensional vibration feature signal is retained for particle size deduction, and the time series judgment result is output.

[0032] If the change gradient value exceeds the preset smoothing threshold range, and the multi-dimensional vibration feature signal is a single pulse change, it is determined to be an instantaneous interference from a large interfering object that was accidentally mixed in. The influence of the multi-dimensional vibration feature signal on the deduction result is suppressed by the forget gate, and the timing judgment result is output.

[0033] In one possible implementation, the calculation of the compensation coefficient based on the initial material particle size distribution, the pure material vibration signal, and the material flow rate data includes:

[0034] The belt speed of the belt conveyor is obtained, and several material layer thickness-related parameters are determined by combining historical experimental data.

[0035] The compensation coefficient is obtained by substituting the material layer thickness correlation parameter, the material flow rate data, the belt speed, and the initial material particle size ratio into a preset empirical formula.

[0036] Furthermore, to achieve the above objectives, this application also proposes a material particle size monitoring device, which includes:

[0037] The signal acquisition and conditioning module is used to acquire material flow data and raw material vibration signal, and to filter the raw material vibration signal to obtain a pure material vibration signal.

[0038] The feature extraction module is used to extract multi-dimensional vibration feature signals of the pure material vibration signal in the time domain, frequency domain, and nonlinear dimension.

[0039] The particle size inversion module is used to input the multi-dimensional vibration characteristic signal and the pure material vibration signal into the particle size inversion system for deduction to obtain the initial material particle size ratio.

[0040] The three-dimensional compensation and calibration module is used to calculate the compensation coefficient based on the initial material particle size ratio, the pure material vibration signal, and the material flow rate data, and to calibrate the initial material particle size ratio based on the compensation coefficient to determine the target material particle size ratio.

[0041] In addition, to achieve the above objectives, this application also proposes a material particle size monitoring device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the material particle size monitoring method as described above.

[0042] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the material particle size monitoring method described above.

[0043] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the material particle size monitoring method described above.

[0044] This application provides a material particle size monitoring method, apparatus, device, and storage medium. The material particle size monitoring method acquires material flow data and raw material vibration signals, filters the raw material vibration signals to obtain purified material vibration signals, and then extracts multi-dimensional vibration feature signals from the purified material vibration signals in the time domain, frequency domain, and nonlinear dimension. These multi-dimensional vibration feature signals and the purified material vibration signals are then input into a particle size inversion system for deduction to obtain an initial material particle size ratio. Finally, based on the initial material particle size ratio, the purified material vibration signals, and the material flow data, further... The calculation yields a compensation coefficient, and based on this coefficient, the initial material particle size distribution is calibrated to determine the target material particle size distribution. This filters out interference from belt vibration, motor noise, and other sources, ensuring that the characteristic signal accurately reflects the material particle size characteristics. The particle size inversion system further improves the accuracy of the initial particle size distribution deduction, reducing instantaneous interference and misjudgment. Combined with the compensation coefficient, the final particle size monitoring accuracy is further improved. The entire process requires no optical components, making it suitable for harsh industrial environments such as high dust and humidity, reducing hardware costs and maintenance difficulty. The entire process from signal processing to particle size calibration is automated, with fast response speed, no need for manual intervention, and reduced human error. Attached Figure Description

[0045] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart illustrating an embodiment of the material particle size monitoring method of this application.

[0048] Figure 2 This is a flowchart illustrating Embodiment 2 of the material particle size monitoring method of this application.

[0049] Figure 3 This is a schematic diagram of the module structure of the material particle size monitoring device according to an embodiment of this application;

[0050] Figure 4 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the material particle size monitoring method in the embodiments of this application.

[0051] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0052] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0053] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0054] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device, big data service platform, or material particle size monitoring system capable of performing the above functions. The following description uses a material particle size monitoring system as an example to illustrate this embodiment and the subsequent embodiments.

[0055] Based on this, the embodiments of this application provide a method for monitoring material particle size, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the material particle size monitoring method of this application.

[0056] In this embodiment, the material particle size monitoring method includes steps S11 to S14:

[0057] Step S11: Obtain material flow data and raw material vibration signal, and filter the raw material vibration signal to obtain pure material vibration signal;

[0058] It should be noted that material flow rate data refers to the mass of material transported by the belt conveyor per unit time, and its calculation is based on the relationship between material weight and belt speed, for example, a value in tons per hour (t / h); raw material vibration signal refers to the original signal collected by vibration sensor, which includes material collision and friction vibration and various interference noises, and multiple vibration intensity data can be obtained per second; pure material vibration signal refers to the vibration signal after filtering, which retains only the effective vibration components generated by material collision and friction and removes irrelevant interference.

[0059] The core purpose of this step is to ensure the spatiotemporal consistency of the data by synchronously acquiring material flow data and raw material vibration signals, and then to eliminate irrelevant interference through filtering to avoid noise signals affecting the accuracy of subsequent feature extraction and granularity inference, ultimately improving the overall monitoring accuracy.

[0060] Specifically, in this embodiment, the acquisition of material flow data and raw material vibration signals must follow specific spatial layout requirements to ensure data validity. In one possible implementation, the electronic belt scale for acquiring material weight information is installed approximately 5 meters downstream of the material drop point, avoiding the impact zone of the drop point. The triaxial vibration sensor for acquiring raw material vibration signals is installed on the idler roller bearing seat approximately 5 meters downstream of the electronic belt scale. This layout avoids cross-interference between the two types of sensors due to belt tension fluctuations, ensuring the independence and accuracy of signal acquisition. Filtering is achieved using a signal conditioning module integrating a bandpass filter and power isolation circuit, improving signal purification through a dual filtering mechanism.

[0061] For example, in a certain ore conveyor belt system, an electronic belt scale is installed 5 meters downstream of the material drop point, and a triaxial vibration sensor is installed on the idler roller bearing seat 5 meters downstream of the electronic belt scale. When the ore material is conveyed by the belt conveyor, the electronic belt scale collects the real-time weight information of the material, and combined with the belt speed set at 2.5 m / s, calculates the material flow rate as 500 t / h. At the same time, the triaxial vibration sensor collects the vibration signal of the raw material at a frequency of 5000 Hz, acquiring 5000 vibration intensity data per second. Subsequently, the signal conditioning module filters the raw material vibration signal: a bandpass filter removes belt mechanical vibration noise with a frequency less than 10 Hz and high-frequency noise generated by motor rotation with a frequency greater than 2000 Hz, and a power isolation circuit suppresses grounding interference in the industrial field and eliminates the influence of electromagnetic noise, finally obtaining a pure material vibration signal containing only the vibration components of ore collision and friction.

[0062] Step S12: Extract the multi-dimensional vibration feature signals of the pure material vibration signal in the time domain, frequency domain, and nonlinear dimension;

[0063] It should be noted that the time domain dimension refers to the signal analysis dimension with time as the horizontal axis and vibration intensity as the vertical axis, used to reflect the characteristics of the signal changing over time; the frequency domain dimension refers to the signal analysis dimension after the time domain signal is transformed by Fourier transform, with frequency as the horizontal axis and vibration energy as the vertical axis, used to reflect the frequency distribution characteristics of the signal; the nonlinear dimension refers to the analysis dimension used to analyze the complexity and non-periodic characteristics of the signal, which does not rely on the linear assumption; the multi-dimensional vibration characteristic signal refers to the feature set that integrates the analysis results of the time domain, frequency domain, and nonlinear dimensions, and can comprehensively reflect the particle size characteristics of the material, including specific characteristic parameters such as the root mean square value of amplitude, kurtosis, vibration energy ratio, and sample entropy. This step transforms the abstract pure material vibration signal into quantifiable feature indicators that can be used for particle size inference. By extracting features from three different dimensions, the specificity of vibration of materials with different particle sizes is comprehensively captured, avoiding feature omissions caused by single-dimensional analysis, thereby improving the accuracy of particle size ratio inference.

[0064] For details, please refer to steps S31-S43, which will not be repeated here.

[0065] Step S13: Input the multi-dimensional vibration characteristic signal and the pure material vibration signal into the particle size inversion system for deduction to obtain the initial material particle size ratio;

[0066] It should be noted that the particle size inversion system refers to an analysis system with a built-in AI accelerator that runs a fusion algorithm model of convolutional neural networks (CNN) and long short-term memory networks (LSTM). It is used to infer the particle size ratio of materials based on input signals. Inference refers to the process of obtaining the particle size ratio of materials through calculation and analysis based on a preset algorithm model and massive training data. The initial particle size ratio of materials refers to the mass ratio of coarse, medium and fine particles obtained by the particle size inversion system based on the input signals without compensation and calibration, such as "coarse:medium:fine = 45%:25%:30%".

[0067] The core purpose of this step is to leverage the strong fitting and time-series analysis capabilities of AI algorithm models, combined with the quantification features of multi-dimensional vibration characteristic signals and the original details of pure material vibration signals, to achieve accurate estimation of material particle size distribution. By extracting local detail features through convolutional neural networks to supplement the deficiencies of multi-dimensional vibration characteristic signals, and by filtering transient interference through long short-term memory networks to avoid estimation errors caused by accidental factors, the final output is an initial proportion result that reflects the particle size distribution of the material.

[0068] For details, please refer to steps S51-S64, which will not be repeated here.

[0069] Step S14: Calculate the compensation coefficient based on the initial material particle size ratio, the pure material vibration signal, and the material flow rate data; and calibrate the initial material particle size ratio based on the compensation coefficient to determine the target material particle size ratio.

[0070] It should be noted that the compensation coefficient refers to a coefficient calculated using a preset empirical formula to correct the weakening effect of material layer thickness on vibration signal propagation; the target material particle size ratio refers to the mass ratio of coarse, medium, and fine particles that reflects the true particle size distribution of the material after compensation and calibration. The purpose of this step is to eliminate the interference of material layer thickness changes on the vibration signal. Since material layer thickness weakens the vibration propagation effect, it leads to a deviation in the initial material particle size ratio. By integrating the initial material particle size ratio, the pure material vibration signal, and the material flow rate data to calculate the compensation coefficient, and then using this coefficient for calibration, the above deviation can be corrected, ultimately outputting a true and accurate target material particle size ratio, ensuring the reliability of the overall monitoring results.

[0071] Specifically, in one embodiment, following the initial material particle size distribution [coarse:medium:fine = 45%:25%:30%] obtained in step S13, the material flow rate data of 500t / h obtained in step S11, and the corresponding pure material vibration signal, the system first acquires the belt speed of the belt conveyor, 2.5m / s. Combining this with the undetermined coefficients a, b, c, d, e, and f (determined through extensive experimental fitting) determined from historical experimental data, the system substitutes the material flow rate data Q = 500t / h, the belt speed v = 2.5m / s, and the initial material particle size distribution of 45% coarse particles, 25% medium particles, and 30% fine particles into the empirical formula:

[0072]

[0073] Where k represents the compensation coefficient for the initial material particle size, and a, b, c, d, e, and f are undetermined coefficients in the correction empirical formula. The compensation coefficient k = 1.15 is then calculated. Subsequently, the automatic calibration module calibrates the initial material particle size ratio based on this compensation coefficient, multiplying each particle size ratio by the compensation coefficient (or adjusting according to preset calibration logic) to finally determine the target material particle size ratio [coarse:medium:fine = 50%:25%:25%].

[0074] This embodiment filters out interference from belt mechanical vibration and motor noise to ensure that the characteristic signal can accurately reflect the particle size characteristics of the material. It also improves the accuracy of the initial particle size ratio estimation through a particle size inversion system, reduces instantaneous interference and misjudgment, and further improves the accuracy of final particle size monitoring by combining compensation coefficients. The entire process does not require optical components, is suitable for harsh industrial environments such as high dust and humidity, reduces hardware costs and maintenance difficulty, and is fully automated from signal processing to particle size calibration. It has a fast response speed, requires no manual intervention, and reduces human error.

[0075] In one feasible implementation, filtering the original material vibration signal to obtain a purified material vibration signal includes:

[0076] Step S21: Filter out belt mechanical vibration noise with a frequency lower than a preset first frequency threshold and high-frequency noise generated by motor rotation with a frequency higher than a preset second frequency threshold from the original material vibration signal using a bandpass filter to obtain the initial material vibration signal;

[0077] It should be noted that a bandpass filter is an electronic filter that only allows signals within a specific frequency range to pass through while blocking signals outside that frequency range. In this method, it is used to filter the effective frequency components of material vibration. The original material vibration signal refers to the initial signal directly acquired by a triaxial vibration sensor, which includes material collision and friction vibration and various interference noises (such as belt mechanical vibration and motor high-frequency noise). The preset first frequency threshold refers to the frequency critical value determined in advance through experiments to distinguish between belt mechanical vibration noise and effective material vibration. This threshold is based on the lowest frequency setting of vibration generated by material collision and friction. Belt mechanical vibration noise refers to the low-frequency vibration signal generated during the operation of the belt conveyor due to belt tension fluctuations, slight roller offsets, and belt-roller friction. Its frequency is usually lower than the effective vibration frequency of the material.

[0078] Furthermore, the preset second frequency threshold refers to the critical frequency value determined in advance through experiments to distinguish between the high-frequency noise of motor rotation and the effective vibration of the material, based on the highest frequency setting of vibration generated by material collision and friction; the high-frequency noise generated by motor rotation refers to the high-frequency vibration and electromagnetic noise generated by the motor winding electromagnetic induction, high-speed rotation of bearings, and motor rotor imbalance when the belt conveyor is running, and its frequency is usually higher than the effective vibration frequency of the material; the initial material vibration signal refers to the intermediate processed signal that has been filtered by a bandpass filter, eliminating low-frequency belt mechanical vibration noise and high-frequency motor rotation noise, and retaining only the signal within the effective vibration frequency range of the material.

[0079] The core purpose of this step is to perform precise noise reduction on the original material vibration signal for the first time. By using the frequency filtering function of a bandpass filter, the two main types of excessive frequency interference (low-frequency mechanical noise from the belt conveyor and high-frequency noise from the motor) are cut off, initially retaining the effective vibration signal that reflects the particle size characteristics of the material. This prevents invalid noise from entering subsequent processing stages and ensures real-time signal processing, meeting the requirement of an overall monitoring delay of less than 1 second. In this embodiment, the preset first frequency threshold and preset second frequency threshold need to be adjusted according to the material characteristics of the specific application scenario and are not fixed values.

[0080] Specifically, in one embodiment, for materials with high hardness and a wide range of collision vibration frequencies, such as ores and coal, a first preset frequency threshold is set to 10Hz, and a second preset frequency threshold is set to 2000Hz. This frequency range has been verified by a large number of experiments and can completely cover the effective vibration frequencies (10Hz-2000Hz) generated by collision friction of such materials. For materials with low hardness and relatively concentrated vibration frequencies, such as grains and feed, the frequency range of analysis should be reduced to extract the effective vibration signal more accurately. The bandpass filter adopts an active bandpass filter circuit design, which has the characteristics of low insertion loss and high out-of-band rejection ratio. The insertion loss is ≤1dB and the out-of-band rejection ratio is ≥40dB, which can effectively screen frequencies while reducing the attenuation of the effective vibration signal of the material.

[0081] For example, in a coal conveyor belt system, a triaxial vibration sensor collects the vibration signal of the raw material at a frequency of 5000Hz. This signal is mixed with 8Hz of belt mechanical vibration noise (generated by belt tension fluctuations) and 2500Hz of motor rotation high-frequency noise (generated by the high-speed rotation of the motor bearings). The system uses an active bandpass filter to process the raw material vibration signal, with a preset first frequency threshold of 10Hz and a preset second frequency threshold of 2000Hz. The bandpass filter first blocks signals with frequencies lower than 10Hz, filtering out the 8Hz belt mechanical vibration noise; then it blocks signals with frequencies higher than 2000Hz, filtering out the 2500Hz motor rotation high-frequency noise; only signals in the frequency range of 10Hz-2000Hz are allowed to pass through. The signals in this range are mainly the effective vibration signals generated by coal particles colliding with the idlers and belt, and the final output is the initial material vibration signal with a frequency between 10Hz and 2000Hz.

[0082] Step S22: Suppress grounding interference in the industrial field through power isolation circuit, reduce the influence of electromagnetic noise on the initial material vibration signal, and obtain a pure material vibration signal.

[0083] It should be noted that power isolation circuits refer to circuit devices that cut off the electrical connection between the signal transmission link and the power supply network and grounding circuit through opto-isolation, electromagnetic isolation, or capacitive isolation, etc., to suppress grounding interference and power coupling noise in industrial sites. Industrial site grounding interference refers to interference signals caused by potential differences or unstable grounding resistance in the grounding circuits of different equipment sharing the same grounding system in industrial production environments. This interference can mix into the vibration signal through the sensor power line or signal transmission line. Electromagnetic noise refers to the electromagnetic radiation generated by high-power electrical equipment such as motors, frequency converters, and high-voltage transmission lines in industrial sites, as well as the noise signals generated by power coupling between equipment. The frequency of this type of noise may fall within the effective frequency range of bandpass filters and cannot be removed by bandpass filters.

[0084] The core purpose of this step is to perform secondary deep noise reduction on the initial material vibration signal, solving the problem of "co-frequency interference" that the bandpass filter cannot handle—that is, grounding interference and electromagnetic noise with frequencies falling in the range of 10Hz to 2000Hz. By cutting off the interference propagation path through the power isolation circuit, this type of noise is prevented from affecting the accuracy of subsequent vibration feature extraction, and finally outputting a high-quality, pure material vibration signal. At the same time, it is suitable for complex industrial environments such as high dust and humidity, ensuring the stability of signal processing.

[0085] Specifically, in the embodiments of this application, the isolation method of the power isolation circuit can be selected according to the interference intensity of the industrial site. In one possible implementation, for conventional industrial sites with low interference intensity, opto-isolation is adopted, using optocouplers to achieve photoelectric conversion of signals, cutting off electrical connections, with an isolation voltage ≥2500V and an isolation resistance ≥10¹²Ω, which can effectively suppress general grounding interference; for heavy industrial sites with high interference intensity (such as steel and mining), electromagnetic isolation is adopted, using isolation transformers and shielding layer designs to achieve dual isolation between signals and power supply and grounding loops, with a common-mode rejection ratio ≥60dB, which can resist strong electromagnetic radiation and high potential difference grounding interference; the power isolation circuit and bandpass filter are integrated in the same signal conditioning module to form a dual filtering mechanism of "frequency screening + interference isolation", ensuring the integration and efficiency of signal processing.

[0086] For example, the initial material vibration signal from coal conveying obtained in step S21, although having eliminated the 8Hz belt mechanical noise and the 2500Hz high-frequency motor noise, still contains a 150Hz grounding interference signal caused by the grounding potential difference between the inverter and the vibration sensor in the workshop, and a 500Hz power coupling electromagnetic noise generated during motor start-up and shutdown (both frequencies are in the range of 10Hz-2000Hz). The system processes the initial material vibration signal through an electromagnetically isolated power isolation circuit integrated in the signal conditioning module: the isolation transformer disconnects the electrical connection between the signal transmission link and the power supply network, the shielding layer blocks external electromagnetic radiation interference, and the common-mode suppression circuit cancels out the interference signal generated by the grounding potential difference. After processing, the 150Hz grounding interference signal and the 500Hz electromagnetic noise are significantly eliminated, and the final output is a pure material vibration signal containing components of coal particle collision and friction vibration (such as 120Hz coarse coal particle impact vibration and 800Hz fine coal particle friction vibration).

[0087] This embodiment uses a bandpass filter to specifically remove belt mechanical noise below a preset first frequency threshold and motor high-frequency noise above a preset second frequency threshold, initially preserving the effective frequency components of material vibration and accurately filtering out specific frequency interference. Furthermore, it uses a power isolation circuit to suppress grounding interference in industrial settings, significantly eliminating electromagnetic noise that the bandpass filter cannot handle, further purifying the signal and preventing interference from affecting the extraction of material vibration features. This dual filtering forms a complementary mechanism of "frequency screening + interference isolation," greatly improving signal purity and ensuring that subsequent multi-dimensional feature extraction accurately reflects the true vibration characteristics of the material. Therefore, this embodiment designs a filtering scheme to address common problems in industrial settings such as belt mechanical vibration, motor noise, and grounding interference, enhancing the signal stability of the device under harsh operating conditions.

[0088] Based on this, the embodiments of this application provide a method for monitoring material particle size, referring to... Figure 2 , Figure 2 This is a flowchart illustrating Embodiment 2 of the material particle size monitoring method of this application.

[0089] In one feasible implementation, the extraction of multi-dimensional vibration feature signals of the pure material vibration signal in the time domain, frequency domain, and nonlinear dimension includes:

[0090] Step S31: For the time domain dimension, select N consecutive vibration signals within a preset time period, and calculate the root mean square value of the amplitude and the kurtosis corresponding to each vibration signal, where N is a positive integer;

[0091] It should be noted that the preset time refers to a fixed time window determined experimentally beforehand for extracting vibration signal samples. This window must balance signal stability and real-time requirements, avoiding large characteristic fluctuations due to an excessively short window or monitoring delays due to an excessively long window. The N consecutive vibration signals refer to uninterrupted vibration intensity data collected at a fixed sampling frequency within the preset time window. The value of N is determined by both the preset time and the sampling frequency (e.g., N=1000 when the preset time is 0.2s and the sampling frequency is 5000Hz). The root mean square (RMS) value of the amplitude refers to the statistical value obtained by first squaring, then averaging, and finally taking the square root of the amplitude data of the N consecutive vibration signals within the preset time. The formula is:

[0092]

[0093] Where RMS is the root mean square value of the amplitude, x1, x2, x3...x N These represent the amplitudes of N consecutive signals, where N is the number of amplitude samples. These amplitudes are used to quantify the overall intensity of material collision and friction (the RMS value increases with increasing flow rate or particle count). Kurtosis refers to a statistical measure calculated from the amplitude data of N consecutive vibration signals within a preset time period, with the formula: Kurtosis = [Σ(x...] i -μ) 4 / N] / σ 4 (where μ is the average amplitude and σ is the standard deviation of amplitude), used to capture pulse anomalies in the signal, sensitive to the occasional huge impact vibration anomalies generated by large particles, and is a key indicator for identifying large particles.

[0094] The core purpose of this step is to extract quantitative indicators that reflect the vibration intensity and large particle characteristics of the material from the time domain perspective. The overall vibration level of the material is grasped by the root mean square value of the amplitude, and the impact signal of large particles is accurately located by the kurtosis. At the same time, the design of the preset time and the value of N can take into account both real-time performance and characteristic stability, and meet the requirement that the overall monitoring delay is less than 1 second.

[0095] Specifically, in this embodiment, the values ​​of the preset time and N need to be dynamically adjusted according to the material type and sampling frequency. For materials with large vibration signal fluctuations, such as ores and coal, the preset time is set to 0.2s and the sampling frequency is set to 5000Hz. At this time, N=1000, which can ensure that the sample size is sufficient to reflect the vibration characteristics and control the calculation time. For materials with relatively stable vibration signals, such as grains and feed, the preset time can be shortened to 0.1s, the sampling frequency is kept at 5000Hz, and N=500, which can further improve the processing speed while ensuring the accuracy of the features. The calculation process is completed by the FPGA (Field-Programmable Gate Array) parallel processing circuit to avoid the delay caused by serial calculation and ensure that the calculation time of the root mean square value of amplitude and kurtosis is less than 0.05s.

[0096] For example, in a certain ore conveying system, the feature extraction engine (FPGA parallel processing circuit) performs analysis in the time domain: the preset time is set to 0.2s, the sampling frequency is 5000Hz, and 1000 consecutive vibration signals (N=1000) are selected within this time window, with amplitude data of x1, x2, ..., x 1000 First, calculate the root mean square value of the amplitude, then sum the squares of each amplitude data point (x1). 2 +x2 2 +...+x 1000 2 The sum is 20250, which is divided by the sample size of 1000 to get the mean of 20.25. Taking the square root of the mean gives the root mean square value of amplitude, RMS = 4.5. Then, the kurtosis is calculated. First, the mean μ = 3.8 and the standard deviation σ = 1.2 of the 1000 amplitude data are calculated. Then, the fourth power of the difference between each amplitude data point and the mean is calculated and summed (Σ(x...)). i -3.8) 4 The sum is 12468.48, which, divided by the sample size of 1000, gives 24.93696. Then, divided by σ... 4 (1.2) 4 =2.0736), resulting in a kurtosis of 8.5. Finally, the temporal dimension feature extraction was completed, yielding an amplitude root mean square value of 4.5 and a kurtosis of 8.5, providing temporal feature data for subsequent steps.

[0097] Step S32: For the frequency domain dimension, the vibration waveform of each vibration signal is converted from the time domain to the frequency domain by fast Fourier transform, the frequency interval centered on the preset third frequency threshold and the preset fourth frequency threshold is located, and the total vibration energy in each frequency interval is calculated to obtain the vibration energy ratio.

[0098] It should be noted that Fast Fourier Transform (FFT) is a fast calculation algorithm that can convert continuous vibration waveform data in the time domain into frequency-energy distribution data in the frequency domain. It can significantly shorten the calculation time of Fourier Transform and meet the needs of real-time processing. Vibration waveform refers to the curve formed by the change of vibration intensity over time in the time domain. Time domain to frequency domain conversion refers to converting the "time-amplitude" signal representation in the time domain into the "frequency-energy" signal representation in the frequency domain through Fast Fourier Transform, thereby identifying vibration components of different frequencies. The preset third frequency threshold refers to the core frequency value that can represent the vibration characteristics of coarse particles and is determined in advance through experiments. The vibration generated by the rolling and impact of coarse particles is distributed around this frequency.

[0099] Furthermore, the preset fourth frequency threshold refers to the core frequency value that represents the vibration characteristics of fine particles, determined in advance through experiments. The vibrations generated by the friction and collision of fine particles are distributed around this frequency. The frequency range centered on the frequency threshold refers to the range around the preset third and fourth frequency thresholds that includes the main frequency components of the corresponding particle vibrations. The total vibration energy refers to the value obtained by summing the vibration energy of all frequency points within a certain frequency range. The vibration energy ratio refers to the ratio of the total vibration energy of the frequency range centered on the preset third frequency threshold to the total vibration energy of the frequency range centered on the preset fourth frequency threshold. The magnitude of this ratio directly reflects the proportion of coarse and fine particles (the larger the ratio, the higher the proportion of coarse particles).

[0100] The core objective of this step is to distinguish the vibration characteristics of coarse and fine particles through frequency analysis, and to achieve efficient conversion from the time domain to the frequency domain using Fast Fourier Transform. By locating specific frequency ranges and calculating energy ratios, the frequency distribution characteristics are transformed into quantifiable particle size correlation indicators. At the same time, the application of Fast Fourier Transform ensures the real-time nature of the frequency domain analysis, meeting the overall monitoring delay requirements.

[0101] Specifically, in this embodiment, the preset third and fourth frequency thresholds and their corresponding frequency ranges need to be adjusted according to the material characteristics. In one possible implementation, for materials such as ores and rocks with a high proportion of coarse particles, the preset third frequency threshold is set to 63Hz (core frequency of coarse particle vibration), corresponding to a frequency range of 58Hz-68Hz, and the preset fourth frequency threshold is set to 250Hz (core frequency of fine particle vibration), corresponding to a frequency range of 240Hz-260Hz; for fine particle materials such as cement and flour, the preset third frequency threshold can be lowered to 50Hz, corresponding to a range of 45Hz-55Hz, and the preset fourth frequency threshold can be raised to 300Hz, corresponding to a range of 290Hz-310Hz; before the fast Fourier transform, the vibration signal is padded with zeros to ensure that the data length is an integer power of 2, improving the transformation accuracy. At the same time, the transformation calculation is performed by the FPGA parallel processing circuit to ensure that the time taken is less than 0.08s.

[0102] Step S33: Calculate the sample entropy of the vibration signal of the pure material for the nonlinear dimension;

[0103] It should be noted that sample entropy refers to a nonlinear dynamic index used to quantify the complexity of a signal. It reflects the degree of irregularity of a signal by calculating the repetition probability of similar patterns in the signal. The formula involves parameters such as embedding dimension and similarity tolerance. The calculation logic is as follows: first, the signal is reconstructed into a high-dimensional vector, then the similarity between the vectors is statistically analyzed, and finally the entropy value is obtained through probability calculation. The larger the entropy value, the higher the signal complexity and the stronger the randomness, while the smaller the entropy value, the stronger the signal regularity.

[0104] The core purpose of this step is to supplement the characteristic information of material vibration from a nonlinear perspective. Because the motion mode of fine particles is more random (such as frequent friction and collision between multiple particles), their vibration signal is highly complex and the sample entropy is large; the motion mode of coarse particles is more regular (such as rolling and single impact), their vibration signal is highly regular and the sample entropy is small. By calculating the sample entropy, the differences in motion characteristics between coarse and fine particles can be effectively distinguished. At the same time, the calculation of sample entropy does not depend on the linear relationship of frequency or amplitude, and can capture the features missed by traditional time domain and frequency domain analysis, further improving the comprehensiveness of particle size correlation characteristics.

[0105] Specifically, the parameters for calculating sample entropy need to be determined through experimental optimization. In one implementation, the embedding dimension is set to 2 (balancing computational complexity and feature sensitivity), the similarity tolerance is set to 0.2 × signal standard deviation (ensuring effective identification of similar patterns), and the calculation window is selected based on a preset time consistent with the time-domain analysis (e.g., 0.2s) to avoid feature mismatch due to time window differences. The calculation of sample entropy is completed by the FPGA parallel processing circuit, which quickly executes vector reconstruction, similarity statistics, and other steps through pre-stored calculation logic to ensure that the calculation time is less than 0.06s, meeting the real-time monitoring requirements. For possible local mutations in the signal (such as accidentally mixed large impurities), the vibration signal of the pure material is smoothed and preprocessed (e.g., moving average filtering) before calculation to avoid the mutation signal affecting the accuracy of sample entropy.

[0106] For example, to acquire vibration signals of pure ore material (1000 data points within 0.2s), the feature extraction engine calculates sample entropy for non-linear dimensions: First, the embedding dimension is set to m=2, and the similarity tolerance is r=0.2×signal standard deviation (the signal standard deviation is 1.2, so r=0.24); then, the 1000 vibration data points are reconstructed into an m-dimensional vector, resulting in 999 2-dimensional vectors (x1x2, x2x3, ..., x...). 999 x 1000 Next, the number of vectors whose Euclidean distance to other vectors is less than r is counted, and the probability of similar patterns is calculated. Finally, the sample entropy value is calculated based on the probability, and the sample entropy of the vibration signal of the pure material is 1.2. This entropy value indicates that there is a certain proportion of fine particles in the ore material (the entropy value is greater than the sample entropy of pure coarse particles, which is 0.8), while coarse particles still account for a certain proportion (the entropy value is less than the sample entropy of pure fine particles, which is 1.8), thus completing the feature extraction of the nonlinear dimension.

[0107] Step S34: Integrate the analysis results of the time domain dimension, the frequency domain dimension, and the nonlinear dimension to form a multi-dimensional vibration characteristic signal.

[0108] It should be noted that the core purpose of this step is to transform the dispersed features of the three dimensions into a unified feature set that can comprehensively reflect the particle size characteristics of the material, thereby avoiding the limitations of single-dimensional features. Since time-domain features cannot directly distinguish the proportion of coarse and fine particles, frequency-domain features are difficult to capture signal randomness, and nonlinear features cannot quantify vibration intensity, feature integration is used to achieve feature complementarity and ensure that the multi-dimensional vibration feature signal contains all the key information required for material particle size deduction.

[0109] Specifically, in one embodiment, the integration process must ensure the spatiotemporal consistency and format uniformity of the feature parameters. During integration, a timestamp (consistent with the signal acquisition time) is added to each feature parameter to ensure that the analysis results of the three dimensions correspond to the material vibration signals of the same time period, avoiding feature association errors caused by spatiotemporal misalignment; the feature format is uniformly set to numerical data, retaining one decimal place, to facilitate numerical calculations by the granular inversion system; the integrated multi-dimensional vibration feature signals are stored in JSON format, and a data verification mechanism checks whether the feature parameters are within a reasonable range. If they exceed the range, they are marked as abnormal and feature extraction is retried to ensure the validity of the feature signals.

[0110] This embodiment accurately quantifies the vibration intensity and large particle characteristics of materials in the time domain; distinguishes the vibration characteristics of coarse and fine particles in the frequency domain; and supplements the signal complexity characteristics in the nonlinear dimension. This allows the results from the time domain, frequency domain, and nonlinear dimension to complement each other, avoiding the limitations of single-dimensional analysis and comprehensively covering the vibration specificity of materials of different particle sizes. Furthermore, by integrating multiple dimensions, complementary features are achieved, and the resulting multi-dimensional vibration feature signal contains rich particle size correlation information, ensuring that the subsequent particle size inversion system can accurately deduce the initial material particle size ratio and reduce the risk of misjudgment.

[0111] In one feasible implementation, the step of converting the vibration waveforms of each vibration signal from the time domain to the frequency domain using a fast Fourier transform, locating frequency intervals centered on a preset third frequency threshold and a preset fourth frequency threshold, and calculating the sum of vibration energy within each frequency interval to obtain the vibration energy ratio includes:

[0112] Step S41: After zero-padding the N consecutive vibration signals within a preset time, perform a fast Fourier transform to decompose each vibration signal in the time domain into a superposition of sinusoidal waves with different frequency components, and output the frequency amplitude correspondence.

[0113] It should be noted that zero-padding refers to the preprocessing operation of adding several data points with values ​​of 0 to the end of N consecutive vibration signal data, so that the total data length is adjusted to an integer power of 2 (e.g., padding 1000 data points with zeros to 1024). Vibration signals in the time domain refer to the original vibration data in the form of "time-amplitude", reflecting the change of vibration intensity over time. The decomposition into a superposition of sine waves of different frequency components refers to the decomposition of the complex vibration waveform in the time domain into multiple sine waves of different frequencies and amplitudes through fast Fourier transform. These sine waves can be superimposed to restore the original vibration waveform. The frequency-amplitude correspondence refers to the output dataset with "frequency-amplitude" as the core, where each frequency point corresponds to an amplitude, reflecting the intensity of the vibration signal at that frequency, such as "frequency 58Hz - amplitude 3.2, frequency 63Hz - amplitude 5.8", etc.

[0114] Specifically, in this embodiment, the number of zeros padded and the calculation accuracy of the Fast Fourier Transform need to be balanced according to the actual scenario. In one possible implementation, if the length of N consecutive vibration signal data is 1000, the nearest power of 2 is 1024 (2^2 + 10 ... 10 The number of zeros padded is 24, which avoids the frequency resolution decrease caused by excessive zero padding and ensures computational efficiency. The fast Fourier transform adopts the radix-2 FFT algorithm, which is executed by the FPGA parallel processing circuit. The time for a single transformation is less than 0.03s. At the same time, the validity of the frequency amplitude correspondence after transformation is verified. If the amplitude of a certain frequency range exceeds the preset reasonable range (such as far exceeding the upper limit of the amplitude of normal material vibration), it is judged as abnormal and the signal is re-acquired to ensure data reliability.

[0115] For example, in a coal conveying system, the feature extraction engine acquires 1000 consecutive vibration signals (N=1000, sampling frequency 5000Hz) within a preset time of 0.2s. First, zero-padding is performed on these 1000 data points, adding 24 zeros to the end of the data to make the total data length 1024 (2... 10 The calculation meets the requirements of the radix-2 FFT algorithm. Then, a Fast Fourier Transform is performed to decompose the "time-amplitude" vibration signal in the time domain (e.g., amplitude 4.2 at t=0.0001s, amplitude 3.8 at t=0.0002s) into a superposition of sine waves of different frequencies. The final output shows the frequency amplitude correspondence: covering a frequency range of 0-2500Hz (half of the sampling frequency), where frequency 58Hz corresponds to amplitude 3.2, frequency 63Hz corresponds to amplitude 5.8, frequency 68Hz corresponds to amplitude 3.5, frequency 240Hz corresponds to amplitude 2.1, frequency 250Hz corresponds to amplitude 4.3, and frequency 260Hz corresponds to amplitude 2.0, etc., fully reflecting the frequency distribution characteristics of the coal vibration signal.

[0116] Step S42: Based on the frequency amplitude correspondence, locate the low-frequency range centered on the preset third frequency threshold and the high-frequency range centered on the preset fourth frequency threshold, and perform a sum of squares operation on the amplitudes of all frequency points in the low-frequency range and the high-frequency range respectively to obtain the total energy of the low-frequency range and the total energy of the high-frequency range.

[0117] It should be noted that the amplitude at a frequency point refers to the vibration intensity value corresponding to each frequency point in the frequency amplitude correspondence; the total energy in the low-frequency range refers to the value obtained by summing the squares of the amplitudes of all frequency points in the low-frequency range. Since vibration energy is proportional to the square of the amplitude, this value directly reflects the total energy contribution of coarse particle vibration; the total energy in the high-frequency range refers to the value obtained by summing the squares of the amplitudes of all frequency points in the high-frequency range, which directly reflects the total energy contribution of fine particle vibration.

[0118] Specifically, in this embodiment, the widths of the low-frequency and high-frequency ranges need to be dynamically adjusted according to the material type. In a specific implementation, based on the correspondence between the vibration frequency amplitudes of coal output in step S41, a third frequency threshold is preset to 63Hz, corresponding to a low-frequency range of 58Hz-68Hz; a fourth frequency threshold is preset to 250Hz, corresponding to a high-frequency range of 240Hz-260Hz.

[0119] First, locate the frequency points and corresponding amplitudes of the two intervals based on the frequency-amplitude correspondence:

[0120] The low-frequency range includes 58Hz (amplitude 3.2), 59Hz (amplitude 3.8), 60Hz (amplitude 4.5), 61Hz (amplitude 5.1), 62Hz (amplitude 5.5), 63Hz (amplitude 5.8), 64Hz (amplitude 5.4), 65Hz (amplitude 4.9), 66Hz (amplitude 4.2), 67Hz (amplitude 3.7), and 68Hz (amplitude 3.5).

[0121] The high-frequency range includes 240Hz (amplitude 2.1), 241Hz (amplitude 2.3), 242Hz (amplitude 2.5), 243Hz (amplitude 2.8), 244Hz (amplitude 3.1), 245Hz (amplitude 3.5), 246Hz (amplitude 3.8), 247Hz (amplitude 4.0), 248Hz (amplitude 4.2), 249Hz (amplitude 4.3), 250Hz (amplitude 4.3), 251Hz (amplitude 4.2), 252Hz (amplitude 4.0), 253Hz (amplitude 3.7), 254Hz (amplitude 3.4), 255Hz (amplitude 3.0), 256Hz (amplitude 2.7), 257Hz (amplitude 2.4), 258Hz (amplitude 2.2), 259Hz (amplitude 2.1), and 260Hz (amplitude 2.0).

[0122] Then, the sum of squares was performed: the sum of the squares of each amplitude in the low-frequency range (3.2² + 3.8² + ... + 3.5²) was obtained, resulting in a total energy of 1800 in the low-frequency range; the sum of the squares of each amplitude in the high-frequency range (2.1² + 2.3² + ... + 2.0²) was obtained, resulting in a total energy of 600 in the high-frequency range.

[0123] Step S43: The ratio of the total energy in the low-frequency range to the total energy in the high-frequency range is taken as the vibration energy ratio.

[0124] Specifically, in this embodiment, the vibration energy ratio needs to be mapped to the particle size distribution based on experimental data. In one possible implementation, through controlled variable experiments, the vibration energy ratios corresponding to materials with different known proportions of coarse and fine particles (e.g., 60% coarse particles, 40% fine particles, 40% coarse particles, 60% fine particles, etc.) are tested respectively. A "vibration energy ratio - coarse particle proportion" mapping table is established (e.g., energy ratio 3.0 corresponds to 45% coarse particles, energy ratio 2.0 corresponds to 35% coarse particles), and stored in the database of the particle size inversion system for subsequent deduction. At the same time, a reasonable range for the vibration energy ratio is set (e.g., 0.5-10.0). If the calculated ratio exceeds this range, it is determined to be an anomaly in the frequency domain analysis, and the processing steps S41-S42 are retried to ensure the validity of the indicator.

[0125] For example, following step S42, the total energy of the coal vibration in the low-frequency range is 1800, and the total energy in the high-frequency range is 600. The feature extraction engine calculates the vibration energy ratio: dividing the total energy in the low-frequency range by the total energy in the high-frequency range, i.e., 1800 ÷ 600 = 3.0, yields a vibration energy ratio of 3.0. Combined with the "vibration energy ratio - coarse particle proportion" mapping table established beforehand through experiments (it is known that an energy ratio of 3.0 corresponds to a coarse particle proportion of approximately 45%), this ratio clearly reflects the high proportion of coarse particles in the current coal material, providing a key frequency domain basis for the subsequent particle size inversion system to output the initial material particle size proportion.

[0126] This embodiment uses zero padding to adapt the vibration signal data length to the calculation requirements of Fast Fourier Transform (FFT), reducing frequency aliasing and picket fence effects, ensuring a more accurate correspondence between the output frequency amplitudes. By decomposing the time-domain vibration signal into a superposition of sine waves of different frequencies, it can clearly distinguish the frequency components of low-frequency vibration of coarse particles (preset third frequency threshold interval) and high-frequency vibration of fine particles (preset fourth frequency threshold interval), achieving a precise correlation between particle type and vibration frequency. Furthermore, by performing a sum-of-squares operation on the amplitudes of all frequency points in the low-frequency and high-frequency intervals (the square of the amplitude is positively correlated with the vibration energy), it can accurately quantify the total vibration energy of the two intervals, avoiding errors caused by fluctuations in data at a single frequency point, and more realistically reflecting the vibration contribution of coarse and fine particles.

[0127] In one feasible implementation, the particle size inversion system includes a convolutional neural network, a long short-term memory network, and a fully connected layer; the step of inputting the multi-dimensional vibration feature signal and the pure material vibration signal into the particle size inversion system for deduction to obtain the initial material particle size ratio includes:

[0128] Step S51: Input the multi-dimensional vibration feature signal and the pure material vibration signal into the particle size inversion system, and scan the pure material vibration signal through the convolutional neural network to extract local detail features;

[0129] It should be noted that Convolutional Neural Network (CNN) refers to a deep learning network with local perception and parameter sharing characteristics. In this step, it is used to capture local correlation features from a continuous vibration signal sequence. Local detail features refer to the local correlation information extracted by the CNN from the vibration signal of pure material that is not covered by the multi-dimensional vibration feature signal. Examples include "the high-frequency micro-vibration of fine particles following the impact vibration of large particles" and "the pattern of alternating vibration of coarse and fine particles in a short period of time". These features can supplement and reflect the subtle changes in the particle size of the material.

[0130] The core purpose of this step is to overcome the "global limitation" of multi-dimensional vibration feature signals. Multi-dimensional feature signals are a global statistical summary of vibration signals, which may miss local subtle related features. By using the local scanning capability of convolutional neural networks, these key detailed features can be extracted. At the same time, the parameter sharing characteristic of convolutional neural networks can reduce computational complexity. With the help of TPU accelerators, the processing time can be ensured to be less than 0.1s, which meets the requirements of real-time monitoring.

[0131] Specifically, in this embodiment, the structural parameters of the convolutional neural network need to be optimized experimentally to adapt to the vibration signal features. A 1D-CNN (one-dimensional convolutional neural network) can be used to adapt to the temporal characteristics of the vibration signal, with two convolutional layers: the first layer has a kernel size of 3 and a number of 16, used to capture short-distance local vibration correlations; the second layer has a kernel size of 5 and a number of 32, used to capture mid-distance local vibration correlations; the activation function is the ReLU function to avoid the gradient vanishing problem; the pooling layer uses max pooling to retain key features while reducing the amount of data; the network is trained with massive labeled vibration signal data (such as vibration signals corresponding to materials with known particle size distributions) to ensure the accuracy of local detail feature extraction, for example, to stably identify the combined vibration mode of "coarse particle impact + fine particle friction".

[0132] For example, in a certain ore conveying system, multi-dimensional vibration characteristic signals [RMS=4.5, kurtosis=8.5, energy ratio=3.0, sample entropy=1.2] are simultaneously input into the particle size inversion system along with the vibration signals of pure materials (1000 vibration intensity data points within 0.2s, some data points being "3.8, 4.2, 5.9, 2.1, 1.8, 2.3, ..., 4.5"). The 1D-CNN in the system begins scanning the vibration signals of the pure materials: the first layer of convolutional kernels (size 3) slides through the signal, capturing the local sequence "5.9 (large particle impact) → 2.1 → 1.8 (fine particle high-frequency friction)," extracting the preliminary feature of "large particle impact followed by fine particle high-frequency vibration"; the second layer of convolutional kernels (size 5) further scans, capturing the mid-range sequence "3.8 → 4.2 → 5.9 → 2.1 → 1.8," extracting the local detail feature of "alternating vibration of coarse and fine particles with the impact amplitude of large particles being significantly higher than that of fine particles." Ultimately, the convolutional neural network outputs a local detail feature vector containing 32 feature dimensions. This vector fully reflects the local correlation information in the ore vibration signal and supplements the lack of details in the multi-dimensional vibration feature signal.

[0133] Step S52: Analyze the temporal context of the multi-dimensional vibration characteristic signal through the long short-term memory network to determine whether the material particle size change is a smooth switch or an instantaneous disturbance, and output the temporal judgment result.

[0134] It should be noted that Long Short-Term Memory (LSTM) refers to a recurrent neural network with gating mechanisms (input gate, forget gate, output gate), which can effectively capture long-term dependencies in sequence data. In this step, it is used to analyze the temporal variation of multidimensional vibration feature signals. Temporal context refers to the correlation between the current multidimensional vibration feature signal and the multidimensional vibration feature signals at multiple consecutive historical moments, such as "the energy ratio gradually increases from 2.0 to 3.0 within 10 consecutive seconds" or "the kurtosis suddenly spikes from 5.0 to 12.0 at a certain moment but only lasts for 0.2 seconds". Smooth switching of material particle size refers to the gradual change of the proportion of coarse and fine particles in the material over time, and the corresponding multidimensional vibration feature signals also show a continuous and slow trend of change, such as "the material mainly composed of fine particles gradually switches to a mixture of coarse and fine particles". At this time, the gradient of the feature signal change is within a preset reasonable range.

[0135] Furthermore, transient interference refers to the accidental introduction of large impurities (such as stones or metal blocks) into the material or the brief false triggering of the sensor, which causes sudden and brief abnormal fluctuations in the multi-dimensional vibration characteristic signals, but is not a change in the actual particle size distribution of the material. For example, "a large stone is accidentally mixed into fine particles, causing the kurtosis to rise momentarily but return to normal after 1 second." The temporal judgment result refers to the judgment conclusion output by the Long Short-Term Memory Network based on the temporal context analysis, including "smooth switching (preserving the current characteristic signal)" or "transient interference (suppressing the current abnormal characteristic signal)," and the corresponding confidence level.

[0136] The core purpose of this step is to filter out abnormal signals caused by non-real particle size changes of the material, and to avoid instantaneous interference affecting the accuracy of the initial material particle size ratio estimation. Multi-dimensional vibration characteristic signals may contain random interference, which may lead to misjudgment of particle size ratio if used directly for estimation. By using the temporal analysis capability of the Long Short-Term Memory network, we can distinguish between real smooth particle size switching and instantaneous interference, ensuring that the subsequent fully connected layer is based only on effective characteristic signals for estimation. At the same time, the gating mechanism of LSTM can efficiently process temporal data, and with the TPU accelerator, the analysis time can be ensured to be less than 0.1s, which meets the requirements of real-time monitoring.

[0137] Specifically, the time window length and judgment threshold of the Long Short-Term Memory (LSTM) network need to be adjusted according to the material conveying characteristics. In one possible implementation, the time window length is set to 10-30 seconds to ensure coverage of typical cycles of material particle size changes (e.g., particle size switching in ore conveying typically takes 15-20 seconds). The gradient threshold is determined experimentally; for example, the gradient threshold for the root mean square amplitude is set to 0.5 / second (i.e., a change of no more than 0.5 per second), and the gradient threshold for the kurtosis is set to 2.0 / second. If the gradient of the characteristic signal exceeds the threshold and the duration is less than 1 second, it is judged as transient interference. The network is trained using historical abnormal interference data (e.g., recorded scenarios of impurity contamination or sensor malfunction) to improve the accuracy of timing judgment. For example, for a scenario of "sudden increase in kurtosis and duration < 0.5 seconds", the confidence level for judging it as transient interference can reach 98%.

[0138] For example, given the current multidimensional vibration feature signal [RMS=4.5, kurtosis=8.5, energy ratio=3.0, sample entropy=1.2], the Long Short-Term Memory (LSTM) network first acquires a historical multidimensional vibration feature signal sequence from the past 20 seconds (time window length 20 seconds). This sequence shows that: within 1-15 seconds, the energy ratio gradually increases from 2.0 to 2.8, and the kurtosis gradually increases from 5.0 to 7.8 (slow change); within 16-19 seconds, the energy ratio increases from 2.8 to 3.0, and the kurtosis increases from 7.8 to 8.5 (continuous slow change). Subsequently, the LSTM calculates the gradient of the current feature signal and the historical sequence through a gating mechanism: the gradient of the energy ratio change is (3.0-2.8) / 1 second = 0.2 / second (less than the threshold of 0.5 / second), and the gradient of the kurtosis change is (8.5-7.8) / 1 second = 0.7 / second (less than the threshold of 2.0 / second), and the change lasts for 4 seconds (not a brief fluctuation). Finally, the LSTM outputs the timing judgment result: "It is determined that the material particle size is smoothly switched, the current multi-dimensional vibration characteristic signal is valid, and the confidence level is 96%". This result will be used for the deduction of the subsequent fully connected layer.

[0139] Step S53: Input the multi-dimensional vibration feature signal, the local detail feature and the timing judgment result into the fully connected layer so that the fully connected layer can output the mass ratio of coarse particles, medium particles and fine particles to generate the initial material particle size ratio.

[0140] It should be noted that the fully connected layer refers to the deep learning network layer in the granularity inversion system that receives multi-source feature inputs and outputs granularity proportions. Its neurons are fully connected to all preceding feature nodes, enabling the fusion of multi-source information to establish a mapping relationship between features and granularity proportions. The mass proportions of coarse, medium, and fine particles refer to the proportional relationship of the three types of particles in the material calculated by mass. The core purpose of this step is to achieve accurate inference of granularity proportions through multi-source information fusion. The fully connected layer can integrate global statistical features, local detailed features, and validity judgments, avoiding the limitations of single feature inputs—using only global features will miss details, using only local features will lack a global perspective, and adding time-series judgment results can eliminate invalid features. After the three are fused, the non-linear mapping capability of the fully connected layer can accurately output the initial granularity proportions. At the same time, the parameters of the fully connected layer are trained and optimized through massive labeled experimental data (multi-source features corresponding to materials with known granularity proportions) to ensure the accuracy of the mapping relationship. With the help of a TPU accelerator, the inference time can be ensured to be less than 0.1 seconds, meeting the requirements of real-time monitoring.

[0141] Specifically, in this embodiment, the structure and granularity classification criteria of the fully connected layer need to be adjusted according to the application scenario. In one possible implementation, the fully connected layer has two hidden layers: the first layer contains 64 neurons, and the second layer contains 32 neurons. The activation function is the Softmax function, ensuring that the sum of the proportions of the three types of particles in the output is 100%. The particle size classification criteria are determined according to the material type. For example, in coal materials, coarse particles are defined as particles > 8 mm, medium particles as 3-8 mm, and fine particles as < 3 mm; in grain materials, coarse particles are defined as particles > 5 mm, medium particles as 2-5 mm, and fine particles as < 2 mm. The fully connected layer is trained using the cross-entropy loss function to minimize the error between the predicted particle size proportion and the actual particle size proportion, ensuring that the particle size proportion prediction error on the test set is less than 5%. For example, when the actual coarse particle proportion is 45%, the predicted value is in the range of 42%-48%.

[0142] For example, the local detail feature vector (32-dimensional) obtained in step S51 and the time-series judgment result ("smooth switching + 96% confidence") obtained in step S52, along with the multi-dimensional vibration feature signal [RMS=4.5, kurtosis=8.5, energy ratio=3.0, sample entropy=1.2], are input into the fully connected layer. The fully connected layer first concatenates the multi-dimensional vibration feature signal (4-dimensional), the local detail feature vector (32-dimensional), and the time-series judgment result (converted to 1-dimensional confidence data 0.96) into a 37-dimensional input vector. This vector undergoes a nonlinear transformation through the first hidden layer (64 neurons), and is then fed into the second hidden layer (32 neurons) for further optimization of the feature mapping. Finally, the output layer (3 neurons, corresponding to coarse, medium, and fine particles) outputs the probability distribution. After calculation, the output result is "coarse particles 45%, medium particles 25%, fine particles 30%", which is the initial particle size distribution of the material.

[0143] This embodiment extracts local details through a convolutional neural network; filters transient interference through a long short-term memory network; and uses a fully connected layer to simultaneously receive multi-dimensional vibration features, local detail features, and time-series judgment results. This multi-source information complementarity verification reduces the impact of single feature bias and improves the output accuracy of coarse / medium / fine particle quality ratio.

[0144] In one feasible implementation, the step of analyzing the temporal context of the multi-dimensional vibration characteristic signal through the long short-term memory network to determine whether the material particle size change is a smooth switch or a transient disturbance, and outputting the temporal judgment result, includes:

[0145] Step S61: Set the time window of the long short-term memory network, and acquire and use the historical multidimensional vibration feature signal sequence within the time window as the historical feature sequence.

[0146] It should be noted that the time window refers to a pre-set fixed time range for selecting historical data. Its length should be adapted to the typical cycle of material particle size change to avoid insufficient historical information due to being too short or redundant data interference due to being too long. The historical multi-dimensional vibration characteristic signal sequence refers to the set of multi-dimensional vibration characteristic signals continuously collected and processed in chronological order within the set time window. Each signal contains characteristic parameters in the time domain, frequency domain, and nonlinear dimension. The historical feature sequence refers to the structured sequence data that can be input into the long short-term memory network after the historical multi-dimensional vibration characteristic signal sequence within the time window is sorted in chronological order.

[0147] The core purpose of this step is to provide effective time-series analysis basis for Long Short-Term Memory (LSTM) networks. By setting a reasonable time window, historical feature data that is highly relevant to the current analysis task is filtered out. This avoids irrelevant historical data (such as signals from a few hours ago) or excessively short historical data (such as signals from only 1 second ago) from affecting the judgment of the time-series context relationship. This ensures that subsequent smooth switching of material granularity and instantaneous interference can be accurately distinguished. At the same time, the setting of the time window can also control the amount of data and avoid delays caused by excessive computation, thus meeting the needs of real-time monitoring.

[0148] Specifically, in this embodiment, the length of the time window needs to be dynamically adjusted according to the material type and conveying scenario. In one possible implementation, for materials such as ore and coal with slow particle size changes, the time window is set to 20 seconds to ensure that the entire process of particle size switching from one state to another is covered; for materials such as flour and cement with rapid particle size changes (e.g., easy to mix evenly during conveying), the time window can be shortened to 10 seconds to reduce redundant data; the starting time of the time window is T seconds before the current multi-dimensional vibration characteristic signal acquisition time (T is the length of the time window), and the acquisition frequency of the historical multi-dimensional vibration characteristic signal sequence is consistent with the current signal (e.g., 1 acquisition per second) to ensure the continuity and consistency of the time sequence; when acquiring the historical characteristic sequence, the data integrity is checked, and if the signal at a certain moment is missing, linear interpolation is used to supplement it to avoid data gaps affecting the analysis.

[0149] For example, in a certain ore conveying system, the time window of the Long Short-Term Memory network is set to 20 seconds, and the current acquisition time of the multi-dimensional vibration characteristic signal is t. 20 (14:30:20). The system retrieves t0 (14:30:00) to t from the database. 19 (14:30:19) This 20-second period contains 20 historical multi-dimensional vibration characteristic signals. The system organizes these data into a historical feature sequence in chronological order, ensuring that the signal at each moment corresponds one-to-one with the timestamp, thus completing this step of processing.

[0150] Step S62: Input the multi-dimensional vibration feature signal and the historical feature sequence into the hidden layer of the long short-term memory network, and adaptively adjust the weight ratio of the historical feature sequence and the multi-dimensional vibration feature signal through a gating mechanism;

[0151] It should be noted that the hidden layer refers to the core layer in a Long Short-Term Memory (LSTM) network used to store and process temporal information. It includes cell states and hidden states, and can control the inflow, forgetting, and output of information through gating mechanisms. The gating mechanism refers to the structure in the LTM network used to regulate information transmission, including input gates, forget gates, and output gates. The input gate controls the input weight of the current signal, the forget gate controls the retention weight of historical signals, and the output gate controls the weight of the hidden layer output. Adaptive adjustment of weight ratio refers to the gating mechanism dynamically allocating the importance of the current signal and historical sequence in the hidden layer processing based on their correlation and stability. If the historical sequence is stable and consistent with the trend of the current signal, the weight of the historical feature sequence will be increased; if the current signal contains new granularity change features (such as a sudden increase in energy ratio), the weight of the current multi-dimensional vibration feature signal will be increased.

[0152] The core objective of this step is to achieve efficient integration of historical and current information, avoiding the limitation of "treating all data with equal weight" in traditional sequence analysis. Through the adaptive adjustment of the gating mechanism, the long short-term memory network focuses more on signals that are valuable for granular judgment (such as stable historical trends or key current changes), reducing the interference of redundant information. At the same time, the parallel computing characteristics of the gating mechanism can also ensure processing efficiency and meet the requirements of real-time monitoring.

[0153] Specifically, in this embodiment, the weight adjustment logic of the gating mechanism is optimized through training with massive experimental data. In one possible implementation, the input gate calculates the weight coefficient (between 0 and 1) of the current signal using the Sigmoid activation function. If the current signal follows the same trend as the historical sequence (e.g., the energy ratio increases slowly), the weight coefficient is set to 0.3-0.5. If the current signal exhibits new characteristics (e.g., the kurtosis begins to increase rapidly), the weight coefficient is increased to 0.6-0.8. The forget gate also calculates the weight coefficient of the historical sequence using the Sigmoid activation function. If the historical sequence is continuous and stable (e.g., RMS fluctuation is less than 0.2 / second), the weight coefficient is set to 0.6-0.8. If the historical sequence fluctuates (e.g., the energy ratio changes abruptly at a certain moment), the weight coefficient is reduced to 0.3-0.5. The output gate combines the cell state and the hidden state, and outputs the fused feature representation through the Tanh activation function to ensure that the fused features accurately reflect the temporal correlation. During the weight adjustment process, the coefficient changes of each gate are recorded for subsequent traceability and verification.

[0154] For example, the historical feature sequence (t0-t) obtained from step S61 19 ) and the current multidimensional vibration characteristic signal (t 20 (RMS=4.5, kurtosis=8.5, energy ratio=3.0, sample entropy=1.2), both are input into the hidden layer of the Long Short-Term Memory network. First, the forgetting gate analyzes the historical feature sequence: t0-t 19 In the signal, the RMS slowly increases from 3.2 to 4.3 (fluctuation less than 0.2 / second), and the energy ratio gradually increases from 2.0 to 2.8 (stable trend). The historical sequence is determined to be stable, and the forgetting gate weight coefficient is calculated to be 0.7, meaning 70% of the historical sequence information is retained. Subsequently, the input gate analyzes the current signal: the current RMS = 4.5 (compared to t...). 19 (4.3 increased by 0.2), energy ratio = 3.0 (compared to t) 19 The value increased from 2.8 to 0.2, showing a trend consistent with historical data but incorporating the latest granular features. The calculated input gate weight coefficient was 0.4, assigning the current signal 40% of the input weight. Finally, the output gate combines the cell state (stored core features of the historical sequence) and the hidden state (features of the current signal), outputting a fused feature vector through the Tanh activation function. This vector contains 70% historical sequence features and 40% current signal features, achieving adaptive weight adjustment and providing a foundation for fused features in subsequent steps.

[0155] Step S63: Calculate the change gradient value between the multi-dimensional vibration feature signal and the historical feature sequence. If the change gradient value is within the preset smoothing threshold range, it is determined that the material particle size is smoothly switched. The multi-dimensional vibration feature signal is retained for particle size deduction, and the time series judgment result is output.

[0156] It should be noted that the gradient value refers to the rate of change between the current multidimensional vibration feature signal and the latest signal in the historical feature sequence (usually the signal at the last moment of the time window). It is calculated separately for each feature parameter, and the formula is "gradient value = (current parameter value - latest historical parameter value) / time interval", where the time interval is the difference between the acquisition time of the current signal and the latest historical signal (e.g., 1 second). The preset smoothing threshold refers to the critical value determined in advance through experiments to judge whether the change of the feature parameter is stable. Each feature parameter corresponds to a threshold.

[0157] Specifically, the preset smoothing threshold needs to be set differently according to the material characteristics and the type of feature parameters, and there is no restriction here; when calculating the change gradient value, if the latest moment signal of the historical feature sequence is missing, the signal of the second to last moment in the time window is used to replace it to ensure the continuity of the calculation; when making a judgment, the change gradient value of all feature parameters must be within the corresponding smoothing threshold range before it is judged as a smooth switch. If a parameter exceeds the threshold, it is necessary to combine the subsequent steps to further judge whether it is an instantaneous interference.

[0158] For example, the historical feature sequence following step S61 (latest time t) 19 (Signal: RMS=4.3, kurtosis=7.8, energy ratio=2.8, sample entropy=1.3) and the current signal t 20 (RMS=4.5, kurtosis=8.5, energy ratio=3.0, sample entropy=1.2), with a time interval of 1 second. First, the gradient values ​​of each feature parameter were calculated: RMS gradient = (4.5-4.3) / 1 = 0.2 / second; kurtosis gradient = (8.5-7.8) / 1 = 0.7 / second; energy ratio gradient = (3.0-2.8) / 1 = 0.2 / second; sample entropy gradient = (1.2-1.3) / 1 = -0.1 / second (absolute value 0.1 / second). The preset smoothing thresholds were: RMS ≤ 0.5 / second, kurtosis ≤ 2.0 / second, energy ratio ≤ 0.3 / second, and sample entropy ≤ 0.2 / second. Comparison showed that the gradient values ​​of all parameters were within the corresponding threshold ranges. Therefore, it is determined to be a smooth material particle size switch. The current multi-dimensional vibration characteristic signal is retained for subsequent particle size deduction, and the time series judgment result is output as follows: "Judgment type: smooth material particle size switch; gradient of each parameter: RMS 0.2 / sec, kurtosis 0.7 / sec, energy ratio 0.2 / sec, sample entropy 0.1 / sec (absolute value); threshold comparison: all satisfied; confidence level: 96%", completing this step of processing.

[0159] Step S64: If the change gradient value exceeds the preset smoothing threshold range and the multi-dimensional vibration feature signal is a single pulse change, it is determined to be an instantaneous interference from a large interfering object that was accidentally mixed in. The influence of the multi-dimensional vibration feature signal on the deduction result is suppressed by the forget gate, and the timing judgment result is output.

[0160] It should be noted that suppressing the impact of the multi-dimensional vibration feature signals on the inference results refers to the weighting coefficient of the forget gate output being close to 0, which significantly reduces the contribution of the current abnormal signal in subsequent particle size inference and avoids misjudgment. The core purpose of this step is to accurately identify and suppress abnormal signals caused by non-real particle size changes of the material, and to avoid instantaneous interference affecting the accuracy of particle size inference. If a single pulse-type abnormal signal is misjudged as a real particle size change, it will lead to a serious deviation in the output of the initial material particle size ratio. Through the dual judgment criteria of "gradient over-threshold + single pulse", instantaneous interference and real changes can be accurately distinguished. Then, the abnormal signal is suppressed by the forget gate to ensure that subsequent inference is based on valid data. At the same time, this process is fully automated and requires no manual intervention, which meets the high efficiency requirements of industrial real-time monitoring.

[0161] Specifically, in this embodiment, determining a "single-pulse change" requires considering the signal data at the next moment. In one possible implementation, the system pre-buffers the multi-dimensional vibration characteristic signal of the next moment (1 second after the current signal acquisition). If the gradient value of the abnormal parameter of the signal at the next moment recovers to within the preset smoothing threshold range and is consistent with the trend of the historical sequence, it is determined to be a single-pulse change. If the signal at the next moment is still abnormal, it is necessary to further investigate whether it is a device malfunction (such as sensor damage) rather than instantaneous interference. The suppression coefficient of the forget gate is set according to the degree of abnormality. If the gradient value of the abnormal parameter is 1.5-2 times the threshold, the coefficient is set to 0.1-0.2; if it is more than 2 times the threshold, the coefficient is set to 0.05-0.1 to ensure that strong abnormal signals are more strongly suppressed. The output timing judgment result is synchronously stored in the log for easy tracing of the cause of the abnormality (such as checking whether large interfering objects are mixed in).

[0162] For example, in a certain ore conveying system, the current multi-dimensional vibration characteristic signal t 20 The parameters are: RMS = 4.5, kurtosis = 10.0, energy ratio = 2.8, sample entropy = 1.3; the latest time step t of the historical feature sequence. 19 The signal has the following parameters: RMS = 4.3, kurtosis = 7.8, energy ratio = 2.8, and sample entropy = 1.3; the time interval is 1 second. The gradient value is calculated as follows: kurtosis gradient = (10.0 - 7.8) / 1 = 2.2 / second, exceeding the preset smoothing threshold of 2.0 / second; other gradient parameters are normal. The system retrieves the next signal t. 21Given RMS = 4.4, kurtosis = 8.0, energy ratio = 2.9, sample entropy = 1.3, kurtosis gradient = (8.0 - 10.0) / 1 = -2.0 / second, the signal recovers to the threshold range and is consistent with the historical sequence trend. Therefore, the current signal t is determined to be... 20 The change was a single pulse. Further analysis revealed only an anomaly in kurtosis, with no coordinated changes in other parameters, consistent with the signal characteristics of "accidental intrusion of a large interfering element." Therefore, it was determined to be a transient interference caused by an accidental intrusion of a large interfering element. The forgetting gate output suppression coefficient of the Long Short-Term Memory network was 0.1, significantly reducing t. 20 Signal weights. Final output timing judgment result: "Judgment type: transient interference from a large, randomly introduced object; Abnormal parameters: kurtosis gradient 2.2 / second (>threshold 2.0 / second); Impulse verification: t" 21 The kurtosis returned to normal, indicating a single pulse; forget gate inhibition coefficient: 0.1; confidence level: 98%; thus completing this step.

[0163] This embodiment uses a time window to lock valid historical data, a gating mechanism to achieve dynamic weight adjustment, gradient judgment to accurately distinguish particle size change types, and a forget gate to suppress instantaneous interference, thereby outputting standardized time-series judgment results. The overall process is highly efficient, the time window and threshold can be flexibly adjusted, and it can quickly respond to changes in material particle size, meeting the industrial real-time requirement of monitoring delay of less than 1 second.

[0164] In one feasible implementation, the calculation of the compensation coefficient based on the initial material particle size distribution, the pure material vibration signal, and the material flow rate data includes:

[0165] Step S71: Obtain the belt speed of the belt conveyor and, in conjunction with historical experimental data, determine several material layer thickness-related parameters.

[0166] It should be noted that belt speed refers to the distance the belt travels per unit time during the operation of the belt conveyor, usually measured in meters per second (m / s). This parameter is set by the belt conveyor's drive system or collected in real time by a speed sensor and is a key factor affecting the material layer thickness (the faster the belt speed, the thinner the material layer thickness at the same flow rate). Historical experimental data refers to the set of experimental data conducted in advance in the laboratory or industrial field, which examines the relationship between material layer thickness and vibration signal under different material types, belt speeds, and material flow rates. It includes the correspondence between "belt speed - material flow rate - material layer thickness - vibration signal attenuation". Material layer thickness-related parameters refer to parameters that are related to the material layer thickness and can affect the propagation of vibration signals. These parameters need to be determined by fitting historical experimental data, such as the density of the material, the hardness coefficient of the material particles, and the propagation attenuation coefficient of the vibration signal in the material layer. The values ​​of the material layer thickness-related parameters are different for different material types.

[0167] Specifically, the determination of the material layer thickness correlation parameters needs to be differentiated for different material types. There are no restrictions here, and they can be set according to the actual situation.

[0168] Step S72: Substitute the material layer thickness correlation parameter, the material flow rate data, the belt speed, and the initial material particle size ratio into a preset empirical formula to calculate the compensation coefficient.

[0169] It should be noted that the preset empirical formula refers to the mathematical formula used to calculate the compensation coefficient, which is obtained by fitting a large amount of historical experimental data. Its core logic is to quantify the weakening effect of the material layer thickness on the vibration signal, as can be seen in step S14.

[0170] Specifically, in this embodiment, the undetermined coefficients of the preset empirical formula need to be optimized through a large number of experiments. In one possible implementation, for iron ore materials, 500 sets of experiments are conducted using the controlled variable method: other parameters are fixed, only one parameter (such as material flow rate) is changed, and the initial particle size ratio and the actual particle size ratio under different parameter combinations are recorded (obtained through manual sampling and sieving). The required compensation coefficients are calculated, and then the undetermined coefficients of the empirical formula are obtained by fitting using the least squares method, for example, a=0.85, b=0.12 (for density parameters), c=0.35, d=0.28, e=0.15, f=0.08. Before substituting the parameters into the calculation, all parameters are standardized (such as unifying the material flow rate unit to t / h and the belt speed to m / s) to avoid calculation errors caused by inconsistent units. After the compensation coefficients are calculated, it is verified whether they are within the preset reasonable range (such as 1.0-1.5, specifically determined according to the material type). If they exceed the range, the parameter input and formula coefficients are rechecked to ensure that the compensation coefficients are effective.

[0171] This embodiment uses the dynamic changes in parameters such as belt speed and material flow rate during the conveying process to calculate in real time, ensuring that the compensation coefficient can dynamically adapt to changes in working conditions and guaranteeing the effectiveness of compensation under different working conditions.

[0172] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0173] This application also provides a material particle size monitoring device; please refer to [reference needed]. Figure 3 The material particle size monitoring device includes:

[0174] The signal acquisition and conditioning module is used to acquire material flow data and raw material vibration signal, and to filter the raw material vibration signal to obtain a pure material vibration signal.

[0175] The feature extraction module is used to extract multi-dimensional vibration feature signals of the pure material vibration signal in the time domain, frequency domain, and nonlinear dimension.

[0176] The particle size inversion module is used to input the multi-dimensional vibration characteristic signal and the pure material vibration signal into the particle size inversion system for deduction to obtain the initial material particle size ratio.

[0177] The three-dimensional compensation and calibration module is used to calculate the compensation coefficient based on the initial material particle size ratio, the pure material vibration signal, and the material flow rate data, and to calibrate the initial material particle size ratio based on the compensation coefficient to determine the target material particle size ratio.

[0178] The material particle size monitoring device provided in this application, employing the material particle size monitoring method in the above embodiments, can solve the technical problems in the background art. Compared with the prior art, the beneficial effects of the material particle size monitoring device provided in this application are the same as the beneficial effects of the material particle size monitoring method provided in the above embodiments, and other technical features in the material particle size monitoring device are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0179] This application provides a material particle size monitoring device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the material particle size monitoring method in the first embodiment described above.

[0180] The following is for reference. Figure 4 The diagram illustrates a structural schematic suitable for implementing the material particle size monitoring device in the embodiments of this application. The material particle size monitoring device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 4 The material particle size monitoring device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0181] like Figure 4As shown, the material particle size monitoring device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the material particle size monitoring device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, a touch screen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows the material particle size monitoring device to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows material particle size monitoring devices with various systems, it should be understood that it is not required to implement or possess all of the systems shown. More or fewer systems may be implemented alternatively.

[0182] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0183] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0184] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the material particle size monitoring methods provided by the methods described above.

[0185] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. 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 the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0186] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0187] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for monitoring the particle size of materials, characterized in that, include: Material flow data and raw material vibration signals are acquired, and the raw material vibration signals are filtered to obtain a clean material vibration signal. Specifically, a bandpass filter is used to filter out belt mechanical vibration noise with a frequency lower than a preset first frequency threshold and high-frequency noise generated by motor rotation with a frequency higher than a preset second frequency threshold from the raw material vibration signal to obtain an initial material vibration signal. A power isolation circuit is used to suppress grounding interference in the industrial field and reduce the influence of electromagnetic noise on the initial material vibration signal to obtain a clean material vibration signal. Multi-dimensional vibration feature signals of the pure material vibration signal are extracted in the time domain, frequency domain, and nonlinear dimension. For the time domain, N consecutive vibration signals within a preset time period are selected, and the root mean square value of the amplitude and kurtosis of each vibration signal are calculated, where N is a positive integer. For the frequency domain, the vibration waveform of each vibration signal is transformed from the time domain to the frequency domain using a fast Fourier transform. Frequency intervals centered on preset third and fourth frequency thresholds are located, and the total vibration energy within each frequency interval is calculated to obtain the vibration energy ratio. For the nonlinear dimension, the sample entropy of the pure material vibration signal is calculated. The analysis results of the time domain, frequency domain, and nonlinear dimensions are integrated to form a multi-dimensional vibration feature signal. The multi-dimensional vibration characteristic signal and the pure material vibration signal are input into the particle size inversion system for deduction to obtain the initial material particle size ratio. A compensation coefficient is calculated based on the initial material particle size distribution, the vibration signal of the pure material, and the material flow rate data. Based on the compensation coefficient, the initial material particle size distribution is calibrated to determine the target material particle size distribution.

2. The material particle size monitoring method as described in claim 1, characterized in that, The process involves converting the vibration waveforms of each vibration signal from the time domain to the frequency domain using a Fast Fourier Transform (FFT), locating frequency intervals centered on a preset third frequency threshold and a preset fourth frequency threshold, and calculating the sum of vibration energy within each frequency interval to obtain the vibration energy ratio. This includes: After zero-padding of N consecutive vibration signals within a preset time period, a fast Fourier transform is performed to decompose each vibration signal in the time domain into a superposition of sinusoidal waves with different frequency components, and output the frequency amplitude correspondence. Based on the frequency amplitude correspondence, the low-frequency range centered on the preset third frequency threshold and the high-frequency range centered on the preset fourth frequency threshold are located, and the sum of squares of the amplitudes of all frequency points in the low-frequency range and the high-frequency range are calculated to obtain the total energy of the low-frequency range and the total energy of the high-frequency range. The ratio of the total energy in the low-frequency range to the total energy in the high-frequency range is taken as the vibration energy ratio.

3. The material particle size monitoring method as described in claim 1, characterized in that, The granularity inversion system includes a convolutional neural network, a long short-term memory network, and a fully connected layer; The step of inputting the multi-dimensional vibration characteristic signal and the pure material vibration signal into the particle size inversion system for deduction to obtain the initial material particle size ratio includes: The multi-dimensional vibration feature signal and the pure material vibration signal are input into the particle size inversion system, and the pure material vibration signal is scanned by the convolutional neural network to extract local detail features. The temporal context of the multi-dimensional vibration characteristic signal is analyzed by the long short-term memory network to determine whether the change in material particle size is a smooth switch or an instantaneous disturbance, and the temporal judgment result is output. The multi-dimensional vibration characteristic signal, the local detail feature, and the timing judgment result are input into the fully connected layer so that the fully connected layer can output the mass ratio of coarse particles, medium particles, and fine particles to generate the initial material particle size ratio.

4. The material particle size monitoring method as described in claim 3, characterized in that, The step of analyzing the temporal context of the multi-dimensional vibration characteristic signals through the long short-term memory network to determine whether the material particle size change is a smooth switch or a transient disturbance, and outputting the temporal judgment result, includes: A time window is set for the long short-term memory network, and the historical multidimensional vibration feature signal sequence within the time window is acquired and used as the historical feature sequence. The multidimensional vibration feature signal and the historical feature sequence are input into the hidden layer of the long short-term memory network, and the weight ratio of the historical feature sequence and the multidimensional vibration feature signal is adaptively adjusted through a gating mechanism. Calculate the gradient value of the change between the multi-dimensional vibration feature signal and the historical feature sequence. If the gradient value is within the range of a preset smoothing threshold, it is determined that the material particle size is smoothly switched. The multi-dimensional vibration feature signal is retained for particle size deduction, and the time series judgment result is output. If the change gradient value exceeds the preset smoothing threshold range, and the multi-dimensional vibration feature signal is a single pulse change, it is determined to be an instantaneous interference from a large interfering object that was accidentally mixed in. The influence of the multi-dimensional vibration feature signal on the deduction result is suppressed by the forget gate, and the timing judgment result is output.

5. The material particle size monitoring method as described in claim 1, characterized in that, The compensation coefficient is calculated based on the initial material particle size distribution, the vibration signal of the pure material, and the material flow rate data, including: Obtain the belt speed of the belt conveyor and, in conjunction with historical experimental data, determine several material layer thickness-related parameters; The compensation coefficient is obtained by substituting the material layer thickness correlation parameter, the material flow rate data, the belt speed, and the initial material particle size ratio into a preset empirical formula.

6. A material particle size monitoring device, characterized in that, include: The signal acquisition and conditioning module is used to acquire material flow data and raw material vibration signals, and to filter the raw material vibration signals to obtain a clean material vibration signal. Specifically, a bandpass filter is used to filter out belt mechanical vibration noise with a frequency lower than a preset first frequency threshold and high-frequency noise generated by motor rotation with a frequency higher than a preset second frequency threshold from the raw material vibration signal to obtain an initial material vibration signal. A power isolation circuit is used to suppress grounding interference in the industrial field and reduce the influence of electromagnetic noise on the initial material vibration signal to obtain a clean material vibration signal. The feature extraction module is used to extract multi-dimensional vibration feature signals of the pure material vibration signal in the time domain, frequency domain, and nonlinear dimension. For the time domain, N consecutive vibration signals within a preset time period are selected, and the root mean square value of the amplitude and kurtosis of each vibration signal are calculated, where N is a positive integer. For the frequency domain, the vibration waveform of each vibration signal is converted from the time domain to the frequency domain using a fast Fourier transform, and frequency intervals centered on preset third and fourth frequency thresholds are located. The total vibration energy within each frequency interval is calculated to obtain the vibration energy ratio. For the nonlinear dimension, the sample entropy of the pure material vibration signal is calculated. The analysis results of the time domain, frequency domain, and nonlinear dimensions are integrated to form a multi-dimensional vibration feature signal. The particle size inversion module is used to input the multi-dimensional vibration characteristic signal and the pure material vibration signal into the particle size inversion system for deduction to obtain the initial material particle size ratio. The three-dimensional compensation and calibration module is used to calculate the compensation coefficient based on the initial material particle size ratio, the pure material vibration signal, and the material flow rate data, and to calibrate the initial material particle size ratio based on the compensation coefficient to determine the target material particle size ratio.

7. A material particle size monitoring device, characterized in that, The material particle size monitoring device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the material particle size monitoring method as described in any one of claims 1 to 5.

8. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the material particle size monitoring method as described in any one of claims 1 to 5.

Images