Crusher discharge particle size detection method and system based on pattern recognition

Abstract

The application discloses a crusher discharge particle size detection method and system based on pattern recognition, and relates to the technical field of pattern recognition. The method comprises the following steps: in the process of coal gangue crushing of a crusher, a monitoring vibration parameter is obtained through a vibration sensor, and a broken material image is collected through an image sensor; the broken material image is input into a particle size recognition intelligent agent based on integrated learning configuration, a first recognized particle size distribution is output, a plurality of discrete parameters of the first recognized particle size are analyzed, and the discrete parameters are obtained; according to the discrete parameters, classified vibration parameters are obtained, the monitoring vibration parameters are combined, a fusion vibration parameter is obtained through processing, and the fusion vibration parameter is used as a vibration mode; according to the vibration mode, particle size vibration error compensation is performed on the first recognized particle size distribution, and a second recognized particle size is obtained. The application effectively improves the detection precision of the discharge particle size of the coal gangue crusher.

Description

Technical Field

[0001] This invention relates to the field of pattern recognition technology, and specifically to a method and system for detecting the particle size of crusher output based on pattern recognition. Background Technology

[0002] With the rapid development of intelligent mining technology in the coal industry, the automation and precision of coal and gangue crushing and separation have become crucial for improving coal mining quality and resource utilization. During coal and gangue crushing, the output particle size directly affects subsequent separation efficiency; therefore, real-time and accurate detection of the crusher's output particle size is of great significance. In existing technologies, image sensor-based visual inspection methods have been widely researched and applied, using image processing and pattern recognition algorithms to perform preliminary analysis of particle size distribution by acquiring images of the crushed material.

[0003] However, due to the continuous and intense vibrations generated by the crusher during operation, the acquired images become blurred, distorted, or lack information, severely affecting the accuracy of image-based particle size recognition. Furthermore, relying solely on image information is insufficient to overcome interference under dynamic operating conditions; the recognition results often exhibit significant dispersion and lack stability, failing to adapt to changes in the crushing state in real time, resulting in insufficient accuracy of the detection results. This, in turn, restricts the precise control and overall efficiency of the coal and gangue separation process. Summary of the Invention

[0004] This invention provides a method and system for detecting the particle size of crusher output based on pattern recognition, aiming to solve the technical problem of insufficient detection accuracy of particle size of coal gangue crusher output in the prior art.

[0005] In view of the above problems, the present invention provides a method and system for detecting the particle size of crusher output based on pattern recognition.

[0006] In a first aspect, the present invention provides a method for detecting the particle size of crusher discharge based on pattern recognition, comprising:

[0007] During the coal and gangue crushing process in the crusher, vibration parameters are monitored by vibration sensors, and images of the crushed material are acquired by image sensors. The image of the broken object is input into a particle size recognition intelligent body based on ensemble learning configuration, and the first recognition particle size distribution is output. The discrete parameters of multiple first recognition particles are analyzed to obtain discrete parameters. Based on the discrete parameters, classified vibration parameters are obtained, and combined with the monitored vibration parameters, fused vibration parameters are obtained as vibration modes. Based on the vibration mode, particle size vibration error compensation is performed on the first identification particle size distribution to obtain the second identification particle size.

[0008] Secondly, the present invention provides a crusher discharge particle size detection system based on pattern recognition, comprising: The data acquisition module is used to monitor vibration parameters through vibration sensors and acquire images of the crushed material through image sensors during the coal and gangue crushing process in the crusher. The intelligent recognition module is used to input the image of the broken object into a particle size recognition intelligent body based on integrated learning configuration, output the first recognition particle size distribution, and perform discrete parameter analysis on multiple first recognition particles to obtain discrete parameters. The vibration mode generation module is used to classify vibration parameters according to the discrete parameters, and combine them with the monitored vibration parameters to obtain fused vibration parameters as vibration modes. The vibration error compensation module is used to perform particle size vibration error compensation on the first identification particle size distribution according to the vibration mode to obtain the second identification particle size.

[0009] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention provides a method and system for detecting the particle size of crusher output material based on pattern recognition. Through the collaborative monitoring of vibration sensors and image sensors, multi-dimensional acquisition of vibration parameters and images of the crushed material is achieved during the crushing process. The crushed material images are input into a particle size recognition agent with integrated learning configuration, outputting the particle size distribution and performing discrete parameter analysis, thus improving the accuracy and reliability of particle size recognition. Vibration parameters are classified according to discrete parameters, and vibration patterns are obtained through fusion, establishing a correlation between vibration patterns and particle size recognition results. Error compensation is performed on the initially identified particle size distribution based on the vibration patterns, effectively reducing detection errors and ultimately obtaining a high-precision second particle size. This achieves accurate and efficient detection of crushed output particle size, ensuring the stability of the crushing process and the consistency of product quality. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating the crusher discharge particle size detection method based on pattern recognition provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a crusher discharge particle size detection system based on pattern recognition provided in an embodiment of the present invention; The components represented by each number in the attached diagram are explained below: Data acquisition module 11, intelligent recognition module 12, vibration mode generation module 13, vibration error compensation module 14. Detailed Implementation

[0012] This invention provides a method and system for detecting the particle size of crusher output based on pattern recognition, which is used to address the technical problem of insufficient detection accuracy of particle size of coal gangue crusher output in the prior art.

[0013] Example 1, as Figure 1 As shown, this invention provides a method for detecting the particle size of crusher discharge based on pattern recognition, the method comprising: S100: During the coal gangue crushing process, vibration parameters are monitored by vibration sensors, and images of the crushed material are acquired by image sensors. Impact crushers operate in a complex industrial environment characterized by high dust and strong vibration. On one hand, equipment vibration directly affects the particle size distribution of the crushed material, and vibration parameters reflect the equipment's operating status, serving as the basis for subsequent particle size identification error compensation. On the other hand, dust easily adheres to the surface of the image sensor lens, causing the acquired images of the crushed material to become blurry and lose details, making it impossible to accurately extract particle features. Therefore, relying solely on image sensors is insufficient to obtain reliable data. To address this issue, this step aims to simultaneously acquire two types of key data: quantifying the vibration state of the equipment itself through vibration sensors to provide a basis for subsequent visual error compensation; and acquiring clear images of the crushed material in harsh environments using specially protected image sensors to provide raw materials for particle size visual identification.

[0014] Step S100 in the method provided in this embodiment of the invention includes: During the crushing of coal and gangue in the crusher, monitoring is carried out at a preset frequency, and the vibration parameters are obtained by monitoring through vibration sensors. Images of the broken material are acquired using an image sensor, which is equipped with a dust removal module.

[0015] First, during the coal and gangue crushing process, monitoring is performed at a preset frequency using vibration sensors to obtain vibration parameters. A vibration sensor is a device that converts mechanical vibration signals into electrical signals. The core monitoring parameter is vibration acceleration, a key indicator reflecting the intensity of equipment vibration. The preset frequency refers to the pre-defined data acquisition interval of the sensor, which must match the crusher's operating cycle to ensure data integrity. Monitored vibration parameters include vibration acceleration and vibration frequency. A piezoelectric vibration sensor suitable for heavy-duty industrial environments is selected; the piezoelectric vibration sensor is fixed to the crusher rotor bearing housing; based on the rotor speed of the impact crusher, the preset acquisition frequency is 50Hz, meaning 50 sets of data are collected per second to ensure coverage of the equipment's vibration cycle; the vibration sensor converts the mechanical vibration signal into a voltage signal in real time, which is then quantized by an analog-to-digital converter (ADC) and transmitted to a data acquisition card for storage.

[0016] For example, taking an impact crusher that processes coal gangue with a particle size of 50-100mm as an example: a piezoelectric vibration sensor is installed in the vertical direction of the rotor bearing seat on the left side of the crusher; the preset acquisition frequency is 50Hz. When the crusher runs at a speed of 1200r / min, the sensor acquires one set of vibration acceleration data every 20ms. The vibration acceleration acquired at a certain moment is 8.5g, where g is the acceleration due to gravity. The data is stored in real time to the local server in CSV format, and the fields include acquisition timestamp, vibration acceleration, and vibration frequency.

[0017] Secondly, images of the crushed material are acquired using an image sensor, which is equipped with a dust removal module. The image sensor specifically refers to an industrial camera, capable of converting optical images into digital images, including frame rate, resolution, and protection level. The dust removal module is a dust removal device installed on the outside of the image sensor lens, used to remove dust from the lens surface; this embodiment employs high-pressure airflow dust removal. The crushed material image refers to a real-time digital image of the coal gangue particles at the crusher's discharge port, used for subsequent particle segmentation and particle size identification. An industrial camera with IP67 protection rating, frame rate ≥100fps, and resolution 1920×1080 was selected. The camera was fixed 30cm above the crusher's discharge port, with the lens pointing vertically downwards at the material's trajectory to ensure complete coverage of the discharge area. The dust removal module was triggered synchronously with the industrial camera's data acquisition, using high-pressure nitrogen as the gas source. It was set to spray air for 1 second every 30 frames of images acquired, with the air pressure controlled at 0.3MPa, to remove coal and gangue dust adhering to the lens surface. The industrial camera converted the optical signal into an 8-bit grayscale image, which was transmitted to the image processing unit in real time via Ethernet and stored in JPG format.

[0018] For example, an industrial camera with IP67 protection, a frame rate of 100fps, and a resolution of 1920×1080 was selected and installed 30cm above the discharge port. The dust removal module was supplied with gas from a small high-pressure nitrogen tank, and the system was set to spray gas for 1 second every 30 frames of images collected. When the coal gangue fell from the discharge port at a flow rate of 5kg / s, the industrial camera collected 100 frames of images per second, one of which clearly showed coal gangue particles with a diameter of 5-20mm. All 100 frames clearly recorded the size, shape, and distribution of particles in the material flow, providing raw data for subsequent visual analysis.

[0019] In this embodiment of the invention, a piezoelectric vibration sensor collects parameters such as vibration acceleration at a preset frequency, fully capturing the dynamic changes in the crusher's operating state. This provides a precise vibration data source for subsequent vibration mode construction and particle size error compensation. The industrial camera with a dust removal module effectively overcomes the interference of high-dust environments, and the collected images of crushed materials have clear particle boundaries and complete details, solving the problem of low recognition accuracy caused by dust in traditional image acquisition. The simultaneous acquisition by dual sensors enables multi-source data matching between the equipment's operating state and the morphology of the crushed materials, providing a reliable data foundation for subsequent integrated learning of particle size recognition and vibration error compensation.

[0020] S200: Input the image of the broken object into the particle size recognition intelligent body based on the integrated learning configuration, output the first recognition particle size distribution, and perform discrete parameter analysis on multiple first recognition particles to obtain discrete parameters.

[0021] In this embodiment of the invention, the image of the crushed material is input into a particle size recognition intelligent body configured based on ensemble learning, and a first particle size distribution is output. Discrete parameters of multiple first particle sizes are then analyzed to obtain discrete parameters. In high-dust, high-vibration coal and gangue crushing conditions, a single image recognition model is easily affected by image blurring and detail loss caused by vibration, resulting in low particle size recognition accuracy and unstable results. Simultaneously, there is a strong correlation between vibration intensity and the discreteness of the recognition results: the greater the vibration intensity, the worse the image quality, and the more random and less similar the recognition results of multiple independent recognition models, resulting in stronger discreteness. Therefore, it is necessary to integrate the advantages of multiple recognition networks through ensemble learning to improve the robustness and basic accuracy of particle size recognition; furthermore, by analyzing the discrete parameters of the recognition results, the degree of interference of vibration on recognition accuracy can be indirectly quantified, providing a basis for subsequent vibration error compensation.

[0022] Step S200 in the method provided in this embodiment of the invention includes: Specifically, the image of the broken object is input into a granularity recognition intelligent body based on ensemble learning configuration, and the first recognition granularity distribution is output, including: Invoke a granularity recognition agent pre-configured based on ensemble learning, wherein the granularity recognition agent includes multiple granularity recognition networks; The image of the broken object is input into the particle size recognition agent to obtain multiple first recognition particles output by multiple particle size recognition networks, which serve as the first recognition particle size distribution.

[0023] First, a pre-configured granularity recognition agent based on ensemble learning is invoked. This granularity recognition agent comprises multiple granularity recognition networks. Ensemble learning is a machine learning method that improves overall recognition accuracy and robustness by fusing the prediction results of multiple base models. The granularity recognition agent is a recognition module built based on ensemble learning, consisting of multiple granularity recognition networks with identical structures but different training data. A granularity recognition network refers to a base model based on a convolutional neural network (CNN) used to extract particle features from images of broken objects and output predicted particle sizes.

[0024] The configuration steps of the granularity recognition agent include: Based on historical particle size detection data of the crusher, a set of sample crushed material images is collected, and the particle size of the crushed material in each sample crushed material image is labeled to obtain a sample particle size set. Based on convolutional neural networks, construct multiple granularity recognition networks with the same structure; The set of fragmented sample images and the set of sample particle sizes are divided multiple times with replacement to obtain multiple sets of particle size recognition training data. The multiple sets of granularity recognition training data are used to train the multiple granularity recognition networks. After convergence, a granularity recognition agent is obtained and configured.

[0025] First, based on historical particle size detection data from the crusher, a set of sample crushed material images is collected, and the particle size of the crushed material in each sample image is labeled to obtain a sample particle size set. The sample crushed material image set refers to a dataset of historically collected coal gangue crushed material images covering different vibration intensities and dust concentrations. The sample particle size set refers to the real particle size dataset formed after labeling the actual particle size of each particle in the sample crushed material images. The D50 equivalent particle size (median particle size) is obtained through a standard sieving method, representing the overall particle size level of the material flow corresponding to a single image, and serving as a label for supervised learning.

[0026] Specifically, 10,000 images of the discharge port were extracted from 1-2 years of historical data from the crusher. The selection criteria were: discharge area coverage ≥80%, no severe obstruction; complete corresponding equipment operating parameters; and coverage of the 5-20mm target particle size range. For each image acquisition period, 5kg of material was collected simultaneously and dry-screened using standard sieves with aperture sizes of 2.5mm, 5mm, 8mm, 12mm, 15mm, and 20mm. The mass of material remaining on each sieve layer was weighed, and the cumulative mass fraction of material smaller than the corresponding sieve aperture size was calculated. The particle size corresponding to the cumulative mass fraction reaching exactly 50% was identified as the D50 value of that batch of material. The images, D50 values, and operating parameters of the crushed material were organized and stored as a CSV file, ensuring that each image of the crushed material had a unique corresponding D50 label.

[0027] For example, based on historical particle size detection data from an impact crusher, 10,000 valid images were extracted to obtain a sample crushed material image set. Among these, 2,000 images were taken under a vibration acceleration of 8.5g, all of which met the requirement of over 85% coverage of the discharge area. A 5kg sample of material from a specific image acquisition period was sieved, yielding a D50 of 6.5mm. The remaining sample crushed material images were processed similarly, ultimately resulting in 10,000 sample particle sizes corresponding to the 10,000 sample crushed material images.

[0028] Secondly, multiple granularity recognition networks with the same structure were constructed based on convolutional neural networks (CNNs). CNNs are deep learning models adept at extracting spatial features from images. Each granularity recognition network is a single base model, taking an image of fragmented material as input and outputting a predicted granularity. The number of network layers, filters, and activation functions are identical across all granularity recognition networks; only the training data differs. Multiple base models were determined, employing a unified structural design. The input layer uses a 1920×1080×1 grayscale image to match the resolution of industrial cameras, and single-channel processing reduces computational cost. The feature extraction layer consists of five convolutional layers and three max-pooling layers, arranged in the order Conv1, MaxPool, Conv2, MaxPool, Conv3, Conv4, MaxPool, and Conv5. All convolutional layers use the ReLU activation function to enhance feature representation. The output layer consists of a single neuron with linear activation, directly outputting the predicted granularity value. All networks initialize weights using a He normal distribution, with biases initialized to 0, ensuring a consistent training starting point.

[0029] For example, three granularity recognition networks are constructed, numbered Net1, Net2, and Net3, with identical structures. The input is a 1920×1080 grayscale image, corresponding to the image format of the fragmented sample. In the feature extraction stage, Conv1 uses 64 3×3 filters, Conv2 uses 128 3×3 filters, and Conv3 to Conv5 use 256, 256, and 512 3×3 filters respectively. Every two convolutional layers are followed by a max-pooling layer to compress the feature dimension. The output layer consists of a single neuron with linear activation, corresponding to the granularity prediction value.

[0030] Next, the set of fragmented sample images and the set of sample granularities are split multiple times with replacement to obtain multiple sets of granularity recognition training data. Split with replacement means randomly selecting samples from the sample set and then replacing them with the original set, ensuring that each sample distribution is independent. This is used to generate multiple sets of training data to improve the generalization ability of the ensemble model. The number of splits is consistent with the number of granularity recognition networks, generating multiple sets of independent training data. Each time, 8000 images are randomly selected with replacement from 10000 sample images and their corresponding granularity labels, with a sampling ratio set to 80% to ensure sufficient data. From each set of 8000 images, 20% (1600 images) are randomly selected as the validation set to determine if the model has converged, and the remaining 80% (6400 images) are used as the training set for model parameter optimization. Each set of data is stored in the form of an image folder plus a labeled CSV file. The images are in JPG format, and the label file contains the image path and the actual granularity value.

[0031] For example, the data is split three times, with a total sample set containing 10,000 images and corresponding D50 labels, ranging from 5 to 20 mm. In the first sampling, 8,000 images are drawn with replacement, resulting in a training set of 6,400 images and a validation set of 1,600 images, forming training data set Data1. In the second sampling, another 8,000 images are drawn with replacement. Because of the sampling with replacement feature, samples can be drawn repeatedly, resulting in training data set Data2. The third sampling follows the same process to form training data set Data3.

[0032] Furthermore, the multiple sets of granularity recognition training data are used to train the multiple granularity recognition networks. After convergence, a granularity recognition agent is obtained and configured. The network weights are continuously optimized through backpropagation, so that the granularity prediction value gradually approaches the true granularity value. The granularity recognition agent refers to a combination of multiple converged granularity recognition networks, which can subsequently process the same image in parallel and output multiple granularity prediction values. The optimizer used is AdamW, which has a weight decay function to avoid model overfitting; the learning rate is initially set to 0.001, and decays by half every 50 rounds to adapt to later fine optimization; the batch size is set to 32 to balance training speed and stability. The loss function uses mean squared error to measure the deviation between the predicted value and the true D50; the convergence criterion is that the mean squared error of the validation set fluctuates less than 0.01 for 10 consecutive rounds, at which point the granularity recognition network is considered to have converged; after multiple granularity recognition networks have converged, their network structures and trained weight files are packaged and deployed to the Jetson Xavier NX embedded platform to support parallel inference calls.

[0033] For example, during training, Data1 is input into Net1, with a learning rate of 0.001 for the first 50 epochs, and the model's prediction error gradually decreases. After 120 epochs, the mean squared error of Net1 on the validation set remains less than 0.01 for 10 consecutive epochs, meeting the preset convergence criterion, and Net1 is considered to have completed training. Similarly, Net2 is trained using Data2, and at 135 epochs, it meets the condition that the mean squared error remains less than 0.01 for 10 consecutive epochs, achieving convergence. Net3 is trained using Data3, and reaches convergence at 128 epochs. After all three granularity recognition networks have converged, their respective network structure configuration files and trained weight files are packaged together and deployed to the Jetson Xavier NX embedded platform, forming a granularity recognition agent that can be called with one click. When a single fragmented image is input, the three granularity recognition networks can be started simultaneously for parallel processing, ultimately outputting three independent granularity prediction values.

[0034] Next, the image of the broken object is input into the particle size recognition agent to obtain multiple first recognition granularities output by multiple particle size recognition networks, which serve as the first recognition granularity distribution. The first recognition granularity refers to the predicted particle size in the broken object image by a single particle size recognition network. The first recognition granularity distribution is the set of multiple first recognition granularities output by multiple particle size recognition networks, reflecting the multi-dimensional prediction results of ensemble learning. The input image is standardized and resized, with pixel values ​​normalized to [0,1] and the size maintained at 1920×1080 resolution, consistent with the training data format. The preprocessed image is simultaneously input into multiple particle size recognition networks for parallel inference; each network outputs one first recognition granularity, which is then aggregated to form the first recognition granularity distribution.

[0035] For example, the image of the broken material collected by S100 is input into the particle size recognition agent. First, the image pixel value is normalized from 0-255 to 0-1 to maintain a resolution of 1920×1080. Three particle size recognition networks process the image simultaneously. Net1 outputs a first recognition particle size of 11.8mm, Net2 outputs a first recognition particle size of 9.2mm, and Net3 outputs a first recognition particle size of 10.5mm, forming a first recognition particle size distribution: [11.8mm, 9.2mm, 10.5mm].

[0036] This involves performing discrete parameter analysis on multiple first-level identification granularities to obtain discrete parameters, including: Calculate the mean of multiple first recognition granularities to obtain the first average recognition granularity; The discrete parameters are calculated based on multiple first recognition granularities and the first average recognition granularity.

[0037] First, the mean of multiple first recognition granularities is calculated to obtain the first average recognition granularity. The first average recognition granularity refers to the arithmetic mean of all values ​​in the first recognition granularity distribution, reflecting the basic prediction level of ensemble learning. For example, based on the first recognition granularity distribution [11.8mm, 9.2mm, 10.5mm], the first average recognition granularity = (11.8 + 9.2 + 10.5) / 3 = 10.5mm.

[0038] Secondly, based on multiple first recognition granularities and a first average recognition granularity, the dispersion parameter is calculated. The dispersion parameter is an index characterizing the degree of numerical dispersion in the distribution of the first recognition granularities. If variance is used, the larger the variance, the higher the degree of dispersion, and the stronger the corresponding vibration. Variance refers to the average of the sum of squares of the differences between each first recognition granularity and the first average recognition granularity. The difference between each first recognition granularity and the first average recognition granularity is calculated one by one; each difference is squared to eliminate positive and negative effects; the sum of the squared differences is calculated and then divided by the number of first recognition granularities, i.e., the number of recognition networks, to obtain the variance, i.e., the dispersion parameter.

[0039] For example, the first recognition particle size distribution is [11.8mm, 9.2mm, 10.5mm], the first average recognition particle size is 10.5mm, the difference is calculated as follows: 11.8-10.5=1.3mm, 9.2-10.5=-1.3mm, 10.5-10.5=0mm; the square operation is 1.3. 2 =1.69, (-1.3) 2 =1.69, 0 2 =0; Sum of values ​​and average: (1.69 + 1.69 + 0) / 3 ≈ 1.13 mm 2 That is, the variance is 1.13mm. 2If the subsequently acquired images correspond to a vibration acceleration of 12g, indicating a greater vibration intensity, the initial particle size distribution might be [13.2mm, 8.1mm, 10.8mm], with a calculated variance of ≈4.21mm. 2 The greater the vibration, the greater the discrete parameter.

[0040] In this embodiment of the invention, the particle size recognition agent integrates the prediction results of multiple independent convolutional neural networks through ensemble learning, effectively reducing the interference caused by dust and vibration, improving the basic recognition accuracy, and adapting to the real-time particle size recognition requirements under complex working conditions; the discrete parameter can quantitatively reflect the degree of interference of vibration on image recognition accuracy, establish an indirect correlation between the discreteness of recognition results and vibration intensity, and provide data support for subsequent acquisition of classification vibration parameters and construction of vibration patterns; by using parallel inference and lightweight computing methods, the timing requirements of online real-time detection are met.

[0041] S300: Based on the discrete parameters, classify and obtain classified vibration parameters, and combine them with the monitored vibration parameters to obtain fused vibration parameters, which are used as vibration modes.

[0042] In this embodiment of the invention, based on the discrete parameters, classified vibration parameters are obtained, and combined with the monitored vibration parameters, fused vibration parameters are obtained as vibration modes. During crusher vibration monitoring, the monitored vibration parameters directly collected by vibration sensors have inherent errors and cannot fully and accurately reflect the actual vibration state of the equipment. Discrete parameters, however, have a clear correlation with vibration intensity, and classified vibration parameters that characterize the actual vibration situation can be derived from these discrete parameters. Monitored vibration parameters have strong real-time performance but are limited by sensor accuracy. Classified vibration parameters are derived based on granularity identification results and have a strong correlation with vibration interference; both characterize vibration features from different dimensions. By fusing these two types of vibration parameters, their respective shortcomings can be complemented, improving the accuracy and reliability of the vibration parameters and providing precise data support for subsequent vibration mode construction and granular vibration error compensation.

[0043] Step S300 in the method provided in this embodiment of the invention includes: Input the discrete parameters into the discrete vibration classification table to obtain the classified vibration parameters; Based on the classified vibration parameters and the monitored vibration parameters, the fused vibration parameters are calculated.

[0044] First, the discrete parameters are input into the discrete vibration classification table to obtain the classified vibration parameters.

[0045] The discrete parameters are input into the discrete vibration classification table to obtain the classified vibration parameters, including: Test data for particle size recognition based on particle size recognition intelligent agent is used to obtain a set of sample discrete parameters, and the actual vibration parameters of the crusher operation test under different sample discrete parameters are collected and labeled as a set of sample vibration parameters. A mapping relationship is constructed between the set of discrete parameters of the samples and the set of vibration parameters of the samples to obtain a discrete vibration classification table; The discrete parameters are input into the discrete vibration classification table for indexing, and the sample vibration parameters corresponding to the same or most similar sample discrete parameters are output as the classification vibration parameters.

[0046] First, based on the particle size recognition intelligent agent, test data for particle size recognition is used to obtain a set of sample discrete parameters. Actual vibration parameters from crusher operation tests under different sample discrete parameters are also collected and labeled as the sample vibration parameter set. Test data refers to the particle size recognition data and vibration monitoring data generated for establishing mapping relationships during specialized tests, covering different vibration conditions. The sample discrete parameter set refers to the dataset composed of discrete parameters calculated by the particle size recognition intelligent agent after recognizing each set of test images during specialized tests, reflecting the degree of dispersion of the recognition results under different vibration intensities. Actual vibration parameters refer to the crusher vibration data measured using high-precision equipment during specialized tests, with vibration acceleration as the core, serving as a true reference for vibration intensity. The sample vibration parameter set is a dataset composed of actual vibration parameters that correspond one-to-one with the sample discrete parameters, used to establish a benchmark for the correlation between dispersion and vibration intensity.

[0047] Specifically, the design covers multiple test conditions with low, medium, and high vibration intensities, corresponding to different rotor speeds and feeding rates, ensuring the diversity of operating conditions. Under each condition, 100 images of the crushed material are collected using an industrial camera equipped with a dust removal module, ensuring complete coverage of the discharge area. Each set of test images is input into a trained particle size recognition agent to obtain three particle size prediction values. The discrete parameters of each set are calculated and summarized to form a sample discrete parameter set. During the same acquisition period, a standard vibration sensor with a higher accuracy level than conventional sensors is installed at the crusher rotor bearing seat to synchronously collect vibration acceleration, forming a sample vibration parameter set. It is ensured that each set of discrete parameters is paired with the measured vibration acceleration of the corresponding time period, and invalid samples with asynchronous acquisition or missing data are eliminated.

[0048] For example, five vibration acceleration conditions (5g, 8g, 10g, 12g, and 15g) are designed, corresponding to rotor speeds of 800 r / min, 1000 r / min, 1200 r / min, 1400 r / min, and 1600 r / min, respectively. 100 images of the broken material are collected for each condition, totaling 500 images. These images are then input into a particle size recognition agent to calculate discrete parameters. For instance, the discrete parameter for a certain image under the 8g condition is 1.1 mm.2 The discrete parameter of a certain set of images under 12g conditions is 4.3mm. 2 A set of discrete parameters for 500 data points was compiled; these parameters were simultaneously acquired using a standard vibration sensor, and the corresponding discrete parameter under 8g operating conditions was 1.1mm. 2 The measured vibration acceleration is 8.2g, corresponding to a discrete parameter of 4.3mm under the 12g condition. 2 The measured vibration acceleration was 11.8g, and a sample vibration parameter set of 500 data points was compiled. The correspondence between the test image group, sample discrete parameters, and sample vibration parameters was established to ensure that there were no misalignments or omissions.

[0049] Secondly, a mapping relationship is constructed between the set of discrete parameters and the set of vibration parameters of the samples, resulting in a discrete vibration classification table. The mapping relationship refers to the fixed correspondence between the discrete parameters and the vibration parameters of the samples, reflecting the law of variation of discrete parameters with vibration intensity. The discrete vibration classification table is a structured table storing the mapping relationship, containing key information about the discrete parameters and corresponding vibration parameters of the samples, used for subsequent rapid querying and matching. The paired sample data is filtered to remove abnormal data where the discrete parameters are negatively correlated with vibration acceleration or whose values ​​exceed a reasonable range. The preprocessed valid paired data is sorted by the discrete parameters from smallest to largest, making the data present a regular distribution. The sorted discrete parameters are individually labeled with vibration acceleration, clarifying the unique true vibration acceleration corresponding to each discrete parameter, establishing a one-to-one mapping relationship between discrete parameters and vibration acceleration. The mapping relationship is organized in a structured format of sequence number, sample discrete parameter, and sample vibration acceleration, and stored as a quickly indexable file, ensuring clear fields and convenient querying.

[0050] For example, based on 500 pairs of data: 3 outliers with increased discrete parameters but decreased vibration acceleration were removed, leaving 497 valid data sets; the samples were then sorted by discrete parameter from smallest to largest, resulting in a discrete parameter range of 0.3-4.5 mm. 2 The corresponding vibration acceleration range is 3.1-14.8g; the discrete parameter for serial number 50 is 0.5mm. 2 The corresponding vibration acceleration is 5.1g; the discrete parameter for serial number 300 is 1.1mm. 2 The corresponding vibration acceleration is 8.2g; the discrete parameter for serial number 400 is 4.3mm. 2 The corresponding vibration acceleration is 11.8g; a discrete vibration classification table is formed, which clearly presents the discrete parameters and vibration acceleration corresponding to each group number, which is convenient for subsequent indexing.

[0051] Then, the discrete parameters are input into the discrete vibration classification table for indexing, and the sample vibration parameters corresponding to the same or most similar sample discrete parameters are output as classification vibration parameters. Classification vibration parameters are sample vibration parameters obtained through index matching, and are vibration parameters that reflect the correlation between vibration and identification interference, derived from the discrete parameters. The current discrete parameters calculated in S200 are input into the data processing unit; the discrete vibration classification table is first searched for sample data that is completely consistent with the current discrete parameters. If such data exists, the corresponding sample vibration parameters are directly extracted; if no completely consistent data exists, the absolute value of the difference between the current discrete parameters and the discrete parameters of all samples in the table is calculated, and the vibration parameter corresponding to the sample with the smallest difference is selected; the matched vibration parameters are used as classification vibration parameters, stored, and used for subsequent fusion calculations.

[0052] For example, based on the calculation results of S200, the current discrete parameter is 1.13mm. 2 The discrete parameter is input into the data processing unit, which then calls the existing discrete vibration classification table. No two samples in the table have completely identical discrete parameters. After calculating the difference, it is found that the discrete parameter of sample number 300 is 1.1 mm. 2 The sample with the smallest absolute difference from the current value is selected; the sample vibration acceleration corresponding to this sample, 8.2g, is extracted as the classification vibration parameter and stored in the data processing unit for fusion.

[0053] Finally, based on the classified vibration parameters and the monitored vibration parameters, a fused vibration parameter is calculated. The fused vibration parameter refers to the final vibration parameter obtained by integrating the classified vibration parameters and the monitored vibration parameters, combining the advantages of both. Mean fusion refers to a fusion method that takes the arithmetic mean of the two types of vibration parameters. This method is suitable for scenarios where the accuracy of both types of vibration parameters is similar and their errors are independent, enabling data complementarity. The classified vibration parameters obtained above and the monitored vibration parameters obtained in S100 are extracted from the data storage unit; the arithmetic mean method is used to fuse the two types of parameters, integrating the effective information of both types of parameters; one decimal place of the fusion result is retained as the final fused vibration parameter, which is the vibration mode characterizing the actual vibration state of the equipment.

[0054] For example, 8.2g of classified vibration parameters and 8.5g of monitored vibration parameters are extracted from the data storage unit; the arithmetic mean method is used to calculate and integrate the advantages of the two types of parameters to offset the error of a single parameter; the calculation result is 8.4g after being rounded to one decimal place. This value is the fused vibration parameter, which is used as the vibration mode for subsequent particle size vibration error compensation.

[0055] In this embodiment of the invention, by constructing a stable mapping relationship between sample discrete parameters and actual vibration parameters, the classified vibration parameters can accurately reflect the interference characteristics of vibration on particle size identification, making up for the deficiency of insufficient correlation of conventional monitoring vibration parameters. At the same time, by using mean fusion, the advantages of classified vibration parameters and monitoring vibration parameters are complemented, effectively reducing the influence of inherent sensor errors and improving the accuracy and reliability of vibration parameters. The resulting vibration pattern provides a precise and reliable basis for subsequent particle size vibration error compensation, ensuring the closed-loop accuracy of the entire detection method and meeting the technical goal of intelligent detection and precise control of crushers under complex working conditions.

[0056] S400: Based on the vibration mode, perform particle size vibration error compensation on the first identified particle size distribution to obtain the second identified particle size.

[0057] In this embodiment of the invention, particle size vibration error compensation is performed on the first identified particle size distribution according to the vibration mode to obtain a second identified particle size. Vibration during crusher operation causes afterimages in the crushed material images acquired by the image sensor. These afterimages stretch the visual contours of coal and gangue particles, leading to an overestimation of particle size and affecting detection accuracy. The first average identified particle size is obtained based on image recognition affected by vibration, which has inherent systematic errors. Vibration modes accurately characterize vibration intensity, and there is a fixed correlation between vibration intensity and the increase in particle size caused by afterimages. Therefore, it is necessary to establish a mapping relationship between vibration parameters and error coefficients using historical data. Based on the vibration mode matching the corresponding error coefficients, reverse compensation is performed on the first average identified particle size to offset the overestimation of particle size caused by afterimages, obtaining accurate particle size detection results that conform to reality, and providing a reliable basis for the adjustment of sorting process parameters.

[0058] Step S400 in the method provided in this embodiment of the invention includes: Based on the vibration mode, the vibration particle size error is classified to obtain the vibration particle size error coefficient; The first recognition granularity distribution is subjected to mean regression processing to obtain the first average recognition granularity. Using the vibration particle size error coefficient, error compensation calculation is performed on the first average identification particle size to obtain the second identification particle size.

[0059] First, based on the vibration mode, the vibration particle size error is classified to obtain the vibration particle size error coefficient.

[0060] Based on the vibration mode, vibration particle size error is classified to obtain vibration particle size error coefficients, including: Based on historical test data of crusher particle size identification through image recognition, a set of sample vibration parameters is extracted, and the increase in the size of the crushed material image under different sample vibration parameters is collected and labeled as the sample vibration particle size error coefficient, thus obtaining a set of sample vibration particle size error coefficients. Construct a mapping relationship between the sample vibration parameter set and the sample vibration particle size error coefficient set to obtain a vibration particle size error classification table; The vibration mode is input into the vibration particle size error classification table to obtain the vibration particle size error coefficient.

[0061] First, based on historical test data of crusher particle size identification via image recognition, a sample vibration parameter set is extracted. The magnitude of the increase in image size of the crushed material under different sample vibration parameters is collected and labeled as the sample vibration particle size error coefficient, thus obtaining the sample vibration particle size error coefficient set. Historical test data refers to past particle size identification tests conducted by the crusher, covering different vibration conditions, encompassing vibration parameters, corresponding images, and actual particle size records. The sample vibration parameter set is a dataset composed of vibration parameters extracted from historical test data, covering different vibration intensities. The sample vibration particle size error coefficient refers to the magnitude of the increase in identified particle size relative to the actual particle size due to image retention in the crushed material image under different vibration parameters, reflecting the degree of interference of vibration on particle size identification. The sample vibration particle size error coefficient set is a dataset composed of error coefficients corresponding one-to-one with the sample vibration parameters, used to establish a correlation benchmark between vibration intensity and error magnitude.

[0062] Specifically, standard-sized coal gangue samples were selected as the test reference material; multiple vibration conditions were designed, covering a vibration acceleration range of 3-15g. Under each condition, the crusher was started, and the standard-sized sample was fed into the crushing chamber; images of the crushed material of the sample were acquired through an image sensor, and the vibration parameters of the current condition were recorded as the sample vibration parameters; the acquired images were input into the particle size recognition agent to obtain the recognized particle size, and the difference between the recognized particle size and the actual standard particle size was calculated. This difference was then divided by the actual standard particle size to obtain the particle size increase, which is the sample vibration particle size error coefficient; the sample vibration parameters and corresponding error coefficients under all conditions were summarized to form two sets that were paired one-to-one, and invalid samples with abnormal recognition or missing data were removed.

[0063] For example, a coal gangue sample with an actual standard particle size of 6.5mm is selected; five sets of operating conditions are designed with vibration accelerations of 5g, 8g, 8.4g, 10g, and 12g, corresponding to rotor speeds of 800r / min, 1000r / min, 1050r / min, 1200r / min, and 1400r / min; under the 8.4g vibration condition, sample images are collected and vibration parameters of 8.4g are recorded, and the particle size recognition agent is input to obtain a recognized particle size of 6.825mm; the error coefficient = (6.825-6.5) / 6.5 = 0.05; the data from the five sets of operating conditions are summarized to form a sample vibration parameter set [5g, 8g, 8.4g, 10g, 12g] and a sample vibration particle size error coefficient set [0.02, 0.04, 0.05, 0.07, 0.09], ensuring a one-to-one correspondence.

[0064] Secondly, a mapping relationship is constructed between the sample vibration parameter set and the sample vibration granularity error coefficient set to obtain a vibration granularity error classification table. The mapping relationship refers to the fixed correspondence between sample vibration parameters and sample vibration granularity error coefficients, reflecting the rule that the greater the vibration intensity, the larger the error coefficient. The vibration granularity error classification table is a structured table storing the above mapping relationship, containing key information about vibration parameters and their corresponding error coefficients for subsequent rapid matching and querying. The paired sample vibration parameter set and sample vibration granularity error coefficient set are filtered to remove abnormal data where the error coefficient is negatively correlated with the vibration intensity; the valid data are sorted in ascending order of sample vibration parameters to ensure a regular data distribution; each sorted vibration parameter and error coefficient is associated and labeled to clarify the unique error coefficient corresponding to each vibration parameter, establishing a one-to-one mapping relationship between vibration parameters and error coefficients; the mapping relationship is organized in a structured format of sequence number, sample vibration parameter, and sample vibration granularity error coefficient, and stored as a quickly indexable file to ensure clear fields and convenient querying.

[0065] For example, there are no abnormal data in the vibration parameters and error coefficients. The samples are sorted by vibration parameters from smallest to largest, in the order of 5g, 8g, 8.4g, 10g, and 12g, with corresponding error coefficients of 0.02, 0.04, 0.05, 0.07, and 0.09. The record labeled with serial number 3 is "3-8.4g-0.05", and the record labeled with serial number 5 is "5-12g-0.09". A vibration particle size error classification table is formed, which clearly shows the vibration parameters and error coefficients corresponding to each group of serial numbers, facilitating rapid indexing later.

[0066] Next, the vibration mode is input into the vibration particle size error classification table to obtain the vibration particle size error coefficient. The vibration particle size error coefficient refers to the particle size increase coefficient corresponding to the current vibration mode, obtained through index matching, and is used for subsequent error compensation calculation. The fused vibration parameters obtained in S300 are input into the data processing unit; the vibration particle size error classification table is first searched for whether there are sample vibration parameters that are completely consistent with the vibration mode. If they exist, the corresponding sample vibration particle size error coefficient is directly extracted; if there is no completely consistent data, the absolute value of the difference between the current vibration mode and all sample vibration parameters in the table is calculated, and the error coefficient corresponding to the sample with the smallest difference is selected; the error coefficient obtained by matching is used as the final vibration particle size error coefficient, stored, and used for subsequent compensation calculation.

[0067] For example, if the current fused vibration parameter is 8.4g, input 8.4g into the data processing unit and call the constructed vibration granularity error classification table; if there is a completely consistent sample vibration parameter of 8.4g in the table, directly extract the corresponding sample vibration granularity error coefficient of 0.05; if the current vibration mode is 8.3g, there is no completely matching data, calculate the difference and find that the absolute value of the difference with 8.4g is the smallest, extract the corresponding error coefficient of 0.05, and store it in the data processing unit for compensation.

[0068] Then, mean regression processing is performed on the first identification granularity distribution to obtain the first average identification granularity. Mean regression processing refers to calculating the arithmetic mean of multiple granularity prediction values ​​in the first identification granularity distribution to eliminate the random error of individual prediction values ​​and obtain a more stable basic identification result. The first average identification granularity refers to the arithmetic mean of the first identification granularity distribution, which is the basic data for error compensation. The first identification granularity distribution obtained in S200 is extracted from the data storage unit; the average value of the distribution is calculated using the arithmetic mean formula, i.e., first average identification granularity = sum of all first identification granularity values ​​ / number of first identification granularities; the calculation result is retained to two decimal places as input data for error compensation.

[0069] For example, following the result of S200, the first recognition particle size distribution is [11.8mm, 9.2mm, 10.5mm], and the arithmetic mean is (11.8+9.2+10.5) / 3=10.50mm; the first average recognition particle size is 10.50mm, which serves as the basis for subsequent compensation calculations.

[0070] Finally, the vibration particle size error coefficient is used to perform error compensation calculation on the first average identified particle size to obtain the second identified particle size. Error compensation calculation refers to the process of reverse correction of the first average identified particle size based on the vibration particle size error coefficient to offset the particle size overestimation error caused by residual images. The second identified particle size refers to the final particle size detection result obtained after error compensation, which more closely matches the actual particle size of the crushed material. The vibration particle size error coefficient and the first average identified particle size are extracted from the data storage unit; the error compensation formula is used for calculation: Second identified particle size = First average identified particle size / (1 + Vibration particle size error coefficient), where (1 + Vibration particle size error coefficient) corresponds to the overestimation ratio of the identified particle size, which is reversed through division; one decimal place of the calculation result is retained as the final actual particle size detection value of the crushed material.

[0071] For example, the vibration particle size error coefficient is 0.05, the first average identified particle size is 10.50mm, and the compensation formula is calculated as: 10.50 / (1+0.05)=10.0mm; therefore, the second identified particle size is 10.0mm. This value accurately offsets the particle size error caused by vibration afterimage and matches the actual crushed particle size of coal gangue.

[0072] In this embodiment of the invention, a stable mapping relationship between vibration parameters and vibration particle size error coefficients is established based on historical test data. The constructed vibration particle size error classification table can quickly and accurately match the error coefficients corresponding to the current vibration mode. Combined with mean regression processing, the random error of a single identification result is reduced. Furthermore, through targeted reverse compensation calculation, the problem of oversized particle size recognition caused by image retention due to vibration is effectively offset, improving the accuracy and reliability of the final particle size detection result. This not only provides realistic particle size data for precise parameter control in subsequent heavy media separation processes but also provides a key guarantee for the stable control of crushed particle size in crushers.

[0073] Through the specific implementation methods described above, the embodiments of the present invention achieve the following technical effects: This invention provides a method and system for detecting the particle size of crusher output based on pattern recognition. By using an image sensor with a dust removal module and a vibration sensor working in tandem, clear and stable image data and real-time vibration data are provided for detection, laying a reliable foundation for perception. An integrated learning-driven particle size recognition scheme effectively improves the robustness of particle size recognition under complex working conditions. Simultaneously, discrete parameter analysis establishes an intrinsic correlation between the recognition results and vibration intensity, achieving a deep understanding of the equipment's state. Vibration parameter fusion modeling, by integrating backward-inferred classified vibration parameters and directly monitored vibration parameters, effectively reduces the inherent error of the sensors, forming a vibration pattern that accurately represents the actual vibration state of the equipment. A targeted error compensation strategy successfully offsets the problem of over-recognition of particle size caused by vibration afterimages, ultimately outputting accurate particle size data that closely matches reality. The entire technical process constructs a complete closed loop of perception, cognition, and control, realizing the transformation from reliance on manual experience to data-driven precision detection. It not only improves the accuracy and real-time performance of output particle size detection, but also builds an intelligent linkage mechanism between equipment status and particle size detection, providing reliable support for parameter optimization of subsequent heavy media separation processes. It fully meets the technical goals of intelligent upgrading of crushers, green and efficient development of the coal industry, and seamless connection of subsequent processes.

[0074] Example 2, as Figure 2 As shown, this invention provides a crusher discharge particle size detection system based on pattern recognition, the system comprising: The data acquisition module 11 is used to monitor vibration parameters through vibration sensors and acquire images of the crushed material through image sensors during the coal and gangue crushing process of the crusher. The intelligent recognition module 12 is used to input the image of the broken object into a particle size recognition intelligent body based on integrated learning configuration, output the first recognition particle size distribution, and perform discrete parameter analysis on multiple first recognition particles to obtain discrete parameters. The vibration mode generation module 13 is used to classify and obtain classified vibration parameters according to the discrete parameters, and process them in combination with the monitored vibration parameters to obtain fused vibration parameters as vibration modes. The vibration error compensation module 14 is used to perform particle size vibration error compensation on the first identification particle size distribution according to the vibration mode to obtain the second identification particle size.

[0075] In one embodiment, the data acquisition module 11 is further configured to: During the crushing of coal and gangue in the crusher, monitoring is carried out at a preset frequency, and the vibration parameters are obtained by monitoring through vibration sensors. Images of the broken material are acquired using an image sensor, which is equipped with a dust removal module.

[0076] In one embodiment, the intelligent recognition module 12 is further configured to: Specifically, the image of the broken object is input into a granularity recognition intelligent body based on ensemble learning configuration, and the first recognition granularity distribution is output, including: Invoke a granularity recognition agent pre-configured based on ensemble learning, wherein the granularity recognition agent includes multiple granularity recognition networks; The image of the broken object is input into the particle size recognition agent to obtain multiple first recognition particles output by multiple particle size recognition networks, which serve as the first recognition particle size distribution.

[0077] The configuration steps of the granularity recognition agent include: Based on historical particle size detection data of the crusher, a set of sample crushed material images is collected, and the particle size of the crushed material in each sample crushed material image is labeled to obtain a sample particle size set. Based on convolutional neural networks, construct multiple granularity recognition networks with the same structure; The set of fragmented sample images and the set of sample particle sizes are divided multiple times with replacement to obtain multiple sets of particle size recognition training data. The multiple sets of granularity recognition training data are used to train the multiple granularity recognition networks. After convergence, a granularity recognition agent is obtained and configured.

[0078] This involves performing discrete parameter analysis on multiple first-level identification granularities to obtain discrete parameters, including: Calculate the mean of multiple first recognition granularities to obtain the first average recognition granularity; The discrete parameters are calculated based on multiple first recognition granularities and the first average recognition granularity.

[0079] In one embodiment, the vibration mode generation module 13 is further configured to: Input the discrete parameters into the discrete vibration classification table to obtain the classified vibration parameters; Based on the classified vibration parameters and the monitored vibration parameters, the fused vibration parameters are calculated.

[0080] The discrete parameters are input into the discrete vibration classification table to obtain the classified vibration parameters, including: Test data for particle size recognition based on particle size recognition intelligent agent is used to obtain a set of sample discrete parameters, and the actual vibration parameters of the crusher operation test under different sample discrete parameters are collected and labeled as a set of sample vibration parameters. A mapping relationship is constructed between the set of discrete parameters of the samples and the set of vibration parameters of the samples to obtain a discrete vibration classification table; The discrete parameters are input into the discrete vibration classification table for indexing, and the sample vibration parameters corresponding to the same or most similar sample discrete parameters are output as the classification vibration parameters.

[0081] In one embodiment, the vibration error compensation module 14 is further configured to: Based on the vibration mode, the vibration particle size error is classified to obtain the vibration particle size error coefficient; The first recognition granularity distribution is subjected to mean regression processing to obtain the first average recognition granularity. Using the vibration particle size error coefficient, error compensation calculation is performed on the first average identification particle size to obtain the second identification particle size.

[0082] Based on the vibration mode, vibration particle size error is classified to obtain vibration particle size error coefficients, including: Based on historical test data of crusher particle size identification through image recognition, a set of sample vibration parameters is extracted, and the increase in the size of the crushed material image under different sample vibration parameters is collected and labeled as the sample vibration particle size error coefficient, thus obtaining a set of sample vibration particle size error coefficients. Construct a mapping relationship between the sample vibration parameter set and the sample vibration particle size error coefficient set to obtain a vibration particle size error classification table; The vibration mode is input into the vibration particle size error classification table to obtain the vibration particle size error coefficient.

[0083] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting the discharge particle size of a crusher based on pattern recognition, characterized in that, The method includes: During the coal and gangue crushing process in the crusher, vibration parameters are monitored by vibration sensors, and images of the crushed material are acquired by image sensors. The image of the broken object is input into a particle size recognition intelligent body based on ensemble learning configuration, and the first recognition particle size distribution is output. The discrete parameters of multiple first recognition particles are analyzed to obtain discrete parameters. Based on the discrete parameters, classified vibration parameters are obtained, and combined with the monitored vibration parameters, fused vibration parameters are obtained as vibration modes. Based on the vibration mode, particle size vibration error compensation is performed on the first identification particle size distribution to obtain the second identification particle size. Based on the discreteness parameters, classified vibration parameters are obtained. Combined with the monitored vibration parameters, fused vibration parameters are obtained, including: Input the discrete parameters into the discrete vibration classification table to obtain the classified vibration parameters; Based on the classified vibration parameters and the monitored vibration parameters, the fused vibration parameters are calculated. The discrete parameters are input into the discrete vibration classification table to obtain the classified vibration parameters, including: Test data for particle size recognition based on particle size recognition intelligent agent is used to obtain a set of sample discrete parameters, and the actual vibration parameters of the crusher operation test under different sample discrete parameters are collected and labeled as a set of sample vibration parameters. A mapping relationship is constructed between the set of discrete parameters of the samples and the set of vibration parameters of the samples to obtain a discrete vibration classification table; The discrete parameters are input into the discrete vibration classification table for indexing, and the sample vibration parameters corresponding to the same or most similar sample discrete parameters are output as the classification vibration parameters. Based on the vibration mode, particle size vibration error compensation is performed on the first identified particle size distribution to obtain a second identified particle size, including: Based on the vibration mode, the vibration particle size error is classified to obtain the vibration particle size error coefficient; The first recognition granularity distribution is subjected to mean regression processing to obtain the first average recognition granularity. Using the vibration particle size error coefficient, error compensation calculation is performed on the first average identification particle size to obtain the second identification particle size; Based on the vibration mode, vibration particle size error is classified to obtain vibration particle size error coefficients, including: Based on historical test data of crusher particle size identification through image recognition, a set of sample vibration parameters is extracted, and the increase in the size of the crushed material image under different sample vibration parameters is collected and labeled as the sample vibration particle size error coefficient, thus obtaining a set of sample vibration particle size error coefficients. Construct a mapping relationship between the sample vibration parameter set and the sample vibration particle size error coefficient set to obtain a vibration particle size error classification table; The vibration mode is input into the vibration particle size error classification table to obtain the vibration particle size error coefficient.

2. The method for detecting the discharge particle size of a crusher based on pattern recognition according to claim 1, characterized in that, During the coal and gangue crushing process in the crusher, vibration parameters are monitored by vibration sensors, and images of the crushed material are acquired by image sensors, including: During the crushing of coal and gangue in the crusher, monitoring is carried out at a preset frequency, and the vibration parameters are obtained by monitoring through vibration sensors. Images of the broken material are acquired using an image sensor, which is equipped with a dust removal module.

3. The method for detecting the discharge particle size of a crusher based on pattern recognition according to claim 1, characterized in that, The image of the fragmented object is input into a granularity recognition intelligent body based on ensemble learning configuration, and the first recognition granularity distribution is output, including: Invoke a granularity recognition agent pre-configured based on ensemble learning, wherein the granularity recognition agent includes multiple granularity recognition networks; The image of the broken object is input into the particle size recognition agent to obtain multiple first recognition particles output by multiple particle size recognition networks, which serve as the first recognition particle size distribution.

4. The method for detecting the discharge particle size of a crusher based on pattern recognition according to claim 1, characterized in that, The configuration steps for the granularity recognition agent include: Based on historical particle size detection data of the crusher, a set of sample crushed material images is collected, and the particle size of the crushed material in each sample crushed material image is labeled to obtain a sample particle size set. Based on convolutional neural networks, construct multiple granularity recognition networks with the same structure; The set of fragmented sample images and the set of sample particle sizes are divided multiple times with replacement to obtain multiple sets of particle size recognition training data. The multiple sets of granularity recognition training data are used to train the multiple granularity recognition networks. After convergence, a granularity recognition agent is obtained and configured.

5. The method for detecting the discharge particle size of a crusher based on pattern recognition according to claim 1, characterized in that, Discrete parameter analysis was performed at multiple first-level identification granularities to obtain discrete parameters, including: Calculate the mean of multiple first recognition granularities to obtain the first average recognition granularity; The discrete parameters are calculated based on multiple first recognition granularities and the first average recognition granularity.

6. A crusher discharge particle size detection system based on pattern recognition, characterized in that, For implementing the pattern recognition-based crusher discharge particle size detection method according to any one of claims 1-5, the system comprises: The data acquisition module is used to monitor vibration parameters through vibration sensors and acquire images of the crushed material through image sensors during the coal and gangue crushing process in the crusher. The intelligent recognition module is used to input the image of the broken object into a particle size recognition intelligent body based on integrated learning configuration, output the first recognition particle size distribution, and perform discrete parameter analysis on multiple first recognition particles to obtain discrete parameters. The vibration mode generation module is used to classify vibration parameters according to the discrete parameters, and combine them with the monitored vibration parameters to obtain fused vibration parameters as vibration modes. The vibration error compensation module is used to perform particle size vibration error compensation on the first identification particle size distribution according to the vibration mode to obtain the second identification particle size.

Images