Gravel aggregate screening method and system based on big data analysis

Abstract

The invention relates to a gravel aggregate screening method and system based on big data analysis, and the method comprises the steps: carrying out the space-time alignment and abnormal value filtering of collected data through an edge calculation node, constructing a high-reliability screening process data set, and executing a first-stage analysis process in a centralized big data platform: carrying out the first-stage analysis process based on an online machine learning model LSTM-RF hybrid model, predicting screening efficiency deviation by second-level frequency, generating a vibration parameter compensation instruction according to a prediction result, dynamically adjusting the amplitude to be + / -0.5 mm or the frequency to be + / -2Hz, and executing a third-level analysis process on a centralized big data platform; and generating a screen inclination angle-feeding rate matching matrix oriented to specific lithology, and forming an optimization strategy real-time instruction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital data processing, in particular to a sand aggregate screening method and system based on big data analysis. BACKGROUND

[0002] Sand aggregate screening is a key production link in building and infrastructure construction, and its efficiency directly affects product quality and production cost. The screening process is subject to the complex coupling of raw material properties, equipment parameters and environmental factors, showing highly nonlinear dynamic characteristics, resulting in large fluctuations in screening efficiency and easy over-standard of powder content in the screened material, making it difficult to meet high-quality standards.

[0003] In the prior art, some intelligent manufacturing system technologies do not optimize the dynamic nonlinear characteristics of the sand screening process, lack real-time processing capability of multi-source heterogeneous data, cannot accurately predict screening efficiency deviation or quickly trace the cause of over-standard powder content, and do not involve a dynamic parameter self-adaptive adjustment mechanism.

[0004] In the sand aggregate screening production line, the first technical challenge is how to collect raw material properties such as aggregate gradation, silt content, water content, rock hardness, equipment operating parameters such as vibration frequency, amplitude, screen inclination, and environmental temperature and humidity, dust concentration data in real time based on a multi-source sensor network, and solve the problem of spatio-temporal alignment caused by data heterogeneity. At the same time, the problems of screening efficiency fluctuation and over-standard powder content in the screened material need to be quickly traced to the key factors of raw material property mutation or screen damage through correlation analysis. In addition, it is still a technical difficulty to build a digital twin based on historical data to simulate parameter optimization strategies under different working conditions, and then realize parameter adaptive adjustment and full-process closed-loop control. These problems jointly restrict the intelligent level of the screening production line, and an integrated solution is needed to realize efficiency prediction, factor tracing and parameter optimization. SUMMARY

[0005] To solve the problems existing in the prior art, the present application aims to provide a sand aggregate screening method and system based on big data analysis.

[0006] The sand aggregate screening method based on big data analysis described in the present application comprises the following steps:

[0007] S101, through a multi-source sensor network deployed in the screening production line, the following field-specific data is collected in real time: raw material property data (including aggregate gradation curve, silt content, water content and rock hardness), equipment operating parameters (vibration frequency, amplitude, screen inclination, motor current and screen wear state of the vibrating screen), environmental parameters (environmental temperature and humidity and dust concentration) and screening effect indicators (real-time production rate of aggregate in each particle size interval, powder content in the screened material and screening efficiency);

[0008] S102, time and space alignment and outlier filtering of the collected field-specific data by the edge computing node to construct a high-confidence screening process dataset;

[0009] S103, performing a first-level analysis process on the centralized big data platform based on the high-confidence screening process dataset: predicting screening efficiency deviation at a second-level frequency based on an online machine learning model (LSTM-RF hybrid model), and generating vibration parameter compensation instructions (dynamically adjusting amplitude ±0.5mm or frequency ±2Hz) according to the prediction results;

[0010] S104, based on the vibration parameter compensation instructions, performing a second-level analysis process on the centralized big data platform: running an association rule mining (Apriori-PSO optimization algorithm) every 5 minutes to identify key factor combinations affecting screening efficiency, and when the oversize powder content exceeds the standard, tracing back to the raw material mud content mutation or screen damage state to generate diagnostic strategy instructions;

[0011] S105, according to the diagnostic strategy instructions, performing a third-level analysis process on the centralized big data platform: constructing a screening process digital twin based on historical data, simulating different parameter combinations under different working conditions, generating a screen inclination-feeding rate matching matrix for specific lithology, and forming real-time optimization strategy instructions;

[0012] S106, the real-time instructions directly drive the vibration motor actuator, the strategy instructions are pushed to the human-machine interface and updated after confirmation, the production line control logic is updated, at the same time, image recognition technology is used to monitor the aggregate particle size distribution in real time, the screening efficiency improvement rate before and after optimization is quantified, the verification data is fed back to the digital twin, the analysis model weight is iteratively updated, and a closed-loop optimization is formed.

[0013] Preferably, in step S101, the following field-specific data is collected in real time through the multi-source sensor network deployed in the screening production line: raw material characteristic data (including aggregate grading curve, mud content, water content, and rock hardness), equipment operating parameters (vibration frequency, amplitude, screen inclination, motor current, and screen wear state), environmental parameters (environmental temperature and humidity, and dust concentration), and screening effect indicators (real-time output rate of each particle size interval, oversize powder content, and screening efficiency), including:

[0014] Obtain multi-source sensing data of raw material mud content, water content, and rock hardness, and output standardized raw material characteristic parameter matrix after preprocessing to eliminate outliers and noise;

[0015] According to the standardized raw material characteristic parameter matrix, a support vector machine is used to establish a mapping model of raw material characteristics and equipment parameters, to predict optimal operating parameters, to monitor motor current and screen wear data in real time, to trigger equipment abnormal warning if the current exceeds the threshold and the wear reaches the critical value, to determine maintenance needs, to predict dust concentration trend by using Kalman filter, to combine with environmental temperature and humidity, to reduce screen inclination angle to reduce vibration intensity if dust concentration rises, to maintain current parameters if concentration is stable, to verify the effect according to the adjusted aggregate output efficiency and powder content, to keep the new parameters if the efficiency is improved and the powder content is reduced, otherwise to revert to the original configuration, and to optimize the screening precision result;

[0016] Based on the optimization results and equipment state data, a screening effect evaluation system is constructed, a comprehensive evaluation report of equipment running state and screening quality is generated, and a decision-making closed loop is formed.

[0017] Preferably, in step S102, the collected field-specific data is subjected to spatio-temporal alignment and outlier filtering by an edge computing node to construct a high-confidence screening process data set, including:

[0018] The original data stream of the distributed sensor network is acquired, and a synchronous data sequence is generated through timestamp calibration;

[0019] Based on the spatial coordinates of each sensor in the synchronous data sequence, a nearest neighbor algorithm is used to establish a node spatial correlation mapping table to determine the data position matching relationship;

[0020] Real-time correlation analysis is performed according to the spatial position relationship, if the difference between adjacent sensor data exceeds the threshold, it is marked as a potential outlier, through sliding window time series test, the outliers that are isolated and distributed are judged as real outliers and are removed, the cleaned data is classified according to sensor type, dimensionless and numerical range standardization is performed, and a structured data set containing source identifier, timestamp, spatial coordinates and numerical value is constructed;

[0021] The data set is evaluated by using an integrity checking algorithm to verify that the data missing rate is below a preset threshold, and the data consistency is tested, and a high-confidence screening data set is output.

[0022] Preferably, in step S103, according to the high-confidence screening process data set, a first-level analysis process is performed on the centralized big data platform: based on an online machine learning model (LSTM-RF hybrid model), the screening efficiency deviation is predicted at a frequency of seconds, and vibration parameter compensation instructions (dynamic adjustment of amplitude ±0.5mm or frequency ±2Hz) are generated according to the prediction results, including:

[0023] The time series data of vibration acceleration, rotational speed and load current are acquired, and the current screening efficiency reference value is calculated;

[0024] If the value deviates from the historical mean range, trigger the LSTM neural network to extract the time sequence characteristics of the vibration pattern;

[0025] Adopt random forest algorithm to perform regression analysis on the output features of the LSTM neural network, predict future efficiency deviation, generate amplitude reduction instructions when predicting positive deviation, and generate frequency improvement instructions when predicting negative deviation, form vibration parameter compensation instruction set, monitor efficiency change after executing compensation instructions;

[0026] If the efficiency improvement amplitude is lower than the preset threshold, recalculate the compensation parameters and perform secondary correction, continuously track long-term running data, and evaluate the stability of the parameters;

[0027] Based on the long-term effectiveness evaluation results, update the model sample library: when the prediction accuracy decreases, trigger the LSTM-RF hybrid model retraining, and continuously perform the prediction compensation task through the optimized model.

[0028] Preferably, in step S104, based on the vibration parameter compensation instruction, a second-level analysis process is performed on the centralized big data platform: an association rule mining (Apriori-PSO optimization algorithm) is run once every 5 minutes to identify the key factor combination affecting the screening efficiency, and when the oversize powder content exceeds the standard, trace back to the raw material mud content mutation or screen damage state, and generate diagnostic strategy instructions, including:

[0029] Obtain material flow, vibration frequency spectrum, power consumption, and particle size distribution multi-dimensional data, construct data set after segmented preprocessing, mine strong association rules between screening efficiency and operating parameters using Apriori algorithm, identify key factor combination patterns, and optimize rule threshold parameters adaptively through PSO algorithm;

[0030] Real-time monitor the oversize powder content, trigger abnormal identification when exceeding the limit, trace back to the raw material mud content historical data, detect mutation trend to judge raw material quality anomaly, if the mud content is stable, analyze the screen vibration frequency spectrum to diagnose damage or blockage;

[0031] According to the root cause type, automatically match the diagnostic strategy template, generate pretreatment strengthening instructions for raw material anomalies, and generate repair and replacement instructions for screen faults;

[0032] Track the device feedback data after executing the instructions, if the powder content improvement does not meet the standard, re-execute the association analysis to update the factor combination, correct the diagnostic strategy template library based on the new factor combination, construct a dynamic knowledge base, continuously optimize the association rule mining period and parameter configuration, and form an adaptive diagnosis system.

[0033] Preferably, in step S105, according to the diagnostic strategy instruction, a third-level analysis process is performed on the centralized big data platform: a screening process digital twin is constructed based on historical data, parameter combinations under different conditions are simulated, a screen inclination-angle-feeding rate matching matrix for specific lithology is generated, an optimization strategy real-time instruction is formed, including:

[0034] A multi-dimensional time series training set is constructed based on a historical database, a nonlinear mapping relationship between screening condition parameters (inclination angle, feeding rate) and performance indicators is established through a deep learning neural network, and a screening process digital twin prediction model is formed;

[0035] The digital twin prediction model is used to simulate screening performance prediction results of different lithology materials under various parameter combinations;

[0036] According to the simulation results, the correspondence between lithology characteristics and optimal parameters is determined, that is, high-hardness lithology matches a larger screen inclination angle, and low-hardness lithology matches a smaller screen inclination angle, and a “lithology classification-inclination angle-feeding rate” two-dimensional matching matrix is constructed to form a parameter optimization query table;

[0037] The current material type is detected through a lithology recognition module, the optimal parameter value is obtained by querying the matching matrix according to the recognition result, screen inclination angle adjustment instructions and feeding rate control instructions are generated, and the instructions are issued to actuators to realize automatic optimization of parameters.

[0038] Preferably, in step S106, the real-time instruction directly drives the vibration motor actuator, the strategy instruction is pushed to the human-computer interaction interface, and after confirmation, the production line control logic is updated, at the same time, image recognition technology is used to monitor the aggregate particle size distribution in real time, the screening efficiency improvement rate before and after optimization is quantified, the verification data is fed back to the digital twin, the analysis model weight is iteratively updated, and a closed-loop optimization is formed, including:

[0039] Real-time data of the jaw crusher crushing cavity pressure, motor load current, and discharge port parameters are obtained, the material jamming risk is judged based on the pressure fluctuation threshold, the discharge port gap adjustment requirement is triggered, the convolutional neural network is used to analyze the current waveform characteristics, and the equipment wear evaluation and remaining life prediction are output;

[0040] The crushing efficiency attenuation coefficient is calculated according to the wear degree, the maintenance warning and spare parts replacement list are generated when the critical value is reached, the high-hardness material is diverted to the standby equipment through the linkage production scheduling system, the load redistribution scheme is formed, and then the conveying belt speed and the stockpile discharging rhythm are adjusted synchronously to determine the production line coordination parameters;

[0041] The production line comprehensive efficiency index is calculated in real time, and when it is lower than the target value, the load distribution optimization process is restarted, and the equipment work arrangement is iteratively updated based on new operation data.

[0042] The sand and gravel aggregate screening system based on big data analysis comprises:

[0043] The multi-source sensing acquisition module is used for collecting raw material characteristic data, equipment operation parameters, environmental parameters and screening effect indexes in real time through a multi-source sensing network deployed in the screening production line, and performing spatio-temporal alignment and outlier filtering on the collected data by using an edge computing node;

[0044] The real-time compensation and prediction module is used for executing a first-level analysis process on the centralized big data platform, dynamically tracking the efficiency change after compensation, and triggering secondary correction and updating a model sample library if the improvement amplitude is lower than a preset threshold;

[0045] The correlation analysis and diagnosis module is used for executing a second-level analysis process on the centralized big data platform, dynamically updating a correlation rule threshold and a diagnosis strategy template library based on instruction execution feedback data;

[0046] The digital twin optimization module is used for executing a third-level analysis process on the centralized big data platform, obtaining a current material type through a lithology identification module, and generating a screen mesh inclination adjustment instruction and a feeding rate control instruction according to a matching matrix;

[0047] The closed-loop execution and feedback module is used for directly driving a vibration motor actuator by using real-time instructions, updating production line control logic after strategy instructions are pushed to a human-machine interaction interface for confirmation, monitoring aggregate particle size distribution in real time by using an image recognition technology, quantifying the screening efficiency improvement rate before and after optimization, feeding verification data to a digital twin, iteratively updating analysis model weights by using an incremental learning algorithm, and forming a closed-loop optimization.

[0048] The sand and gravel aggregate screening method and system based on big data analysis have the advantages that, for the core business problems of screening efficiency fluctuation caused by raw material characteristic change in a traditional screening production line, equipment parameter adjustment lag and fault diagnosis difficulty, a multi-dimensional sensing network covering raw material characteristics, equipment operation, environmental parameters and screening effect is deployed, full-factor data in a screening process is collected in real time, data preprocessing is performed by using edge computing, and a three-level analysis process is constructed: a second-level screening efficiency prediction and vibration parameter real-time compensation are realized based on a LSTM-RF hybrid model, key influence factors are mined and fault traceability diagnosis is performed by using an Apriori-PSO algorithm, a digital twin is simulated to simulate and optimize parameter combinations to generate strategy instructions, a double control mode of directly driving an actuator by using real-time instructions and strategy instructions confirmed by a human-machine is adopted, an image recognition technology is combined to quantify optimization effects, a closed-loop optimization mechanism of data feedback iteration is formed, intelligent adaptive regulation and control and continuous efficiency improvement of the screening process are realized. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1is a process of a sand and gravel aggregate screening method based on big data analysis according to the application Figure 1 ;

[0050] Figure 2 is a process of a sand and gravel aggregate screening method based on big data analysis according to the application Figure 2 . DETAILED DESCRIPTION

[0051] As shown in Figures 1-2 , the sand and gravel aggregate screening method based on big data analysis according to the application comprises the following steps:

[0052] As shown in Figures 1-2 , in step S101, the following field-specific data is collected in real time through the multi-source sensor network deployed in the screening production line: raw material characteristic data (including aggregate grading curve, silt content, water content, and rock hardness), equipment operating parameters (vibration frequency, amplitude, screen inclination angle, motor current, and screen wear state of the vibrating screen), environmental parameters (environmental temperature and humidity and dust concentration), and screening effect indicators (real-time output rate of aggregate in each particle size interval, powder content of oversize material, and screening efficiency).

[0053] Preferably, in step S101, the silt content, water content, and rock hardness data collected by the multi-source sensor are obtained, and the data preprocessing module is used to remove outliers and noise interference to obtain a standardized raw material characteristic parameter matrix;

[0054] According to the standardized raw material characteristic parameter matrix, a mapping relationship model of raw material characteristics and optimal vibration frequency, vibration amplitude, and screen inclination angle is established using a support vector machine algorithm to obtain predicted values of the equipment operating parameters. Through real-time monitoring of the motor current change amplitude and screen wear state sensor data, if the motor current exceeds the preset threshold and the screen wear degree reaches the critical value, an equipment state abnormality early warning is triggered, the equipment maintenance requirement is determined, the Kalman filtering algorithm is used for state estimation and prediction of the dust concentration time series data, the future change direction of the dust concentration is judged in combination with the environmental temperature and humidity change trend;

[0055] According to the dust concentration change trend and the current aggregate grading distribution, if the dust concentration shows an upward trend, the screen inclination angle is adjusted to reduce the vibration intensity, and if the dust concentration is stable, the current equipment parameters are maintained to obtain a dynamically adjusted equipment operating scheme. By comparing the output efficiency of aggregate in each particle size and the powder content of oversize material before and after adjustment, if the output efficiency is improved and the powder content is reduced, the current parameter setting is maintained, and if the effect is not good, the previous parameter configuration is rolled back to determine the final screening precision optimization result.

[0056] According to the screening precision optimization result and the equipment operating state data, a screening effect evaluation index system is established to generate a comprehensive evaluation report of the equipment operating state and screening quality.

[0057] Specifically, in step S101, in the multi-source sensor network deployed in the screening production line, first, the laser particle size analyzer is used to detect the feed aggregate gradation in real time. When it is detected that the proportion of 5-10 mm particle size decreases from the standard 35% to 28%, the system automatically triggers the methylene blue test sensor to detect the clay content, and finds that the value increases from 1.2% to 2.8%. At the same time, the infrared moisture meter shows that the water content is 4.5%, and the rock hardness HLD value measured by the Lelievre hardness tester is 680;

[0058] Based on these changes in raw material characteristics, the equipment operation parameter adaptive adjustment module is started, the vibration screen frequency is adjusted from the standard 16 Hz to 18 Hz, the amplitude is increased from 4.5 mm to 5.2 mm, the screen inclination angle is adjusted from 15 degrees to 17 degrees by a servo motor, and at the same time, it is monitored that the motor current rises from 45 A to 52 A, the laser displacement sensor detects that the screen wear depth reaches 0.8 mm, and the environmental monitoring system shows that the temperature is 26℃, the humidity is 65%, and the dust concentration is 12 mg / m 3 After these data are processed by Kalman filtering algorithm for noise, they are input into the screening effect prediction model;

[0059] The system uses support vector machine algorithm to analyze the correlation of each parameter, and calculates that the 0-5 mm particle size yield is 22.3%, the 5-10 mm particle size yield is 31.8%, the 10-20 mm particle size yield is 28.4%, the powder content of the oversize is controlled at 1.8%, and the comprehensive screening efficiency reaches 94.2%. Through the multiple linear regression model:

[0060] Y = 0.65X1 + 0.23X2 - 0.18X3 + 0.31X4

[0061] Real-time optimization of each parameter ratio ensures that the screening quality is stable within the target range.

[0062] As shown in Figures 1-2 In step S102, the collected field-specific data is subjected to time-space alignment and abnormal value filtering by the edge computing node to construct a high-credibility screening process data set.

[0063] Preferably, in step S102, the original screening data stream of each edge node in the distributed sensor network is obtained, and all data packets are subjected to time stamp calibration processing to obtain a synchronous data sequence with a unified time reference;

[0064] According to the spatial coordinate information of each sensor in the synchronous data sequence, a nearest neighbor algorithm is used to establish a spatial correlation mapping table between the sensor nodes to determine the spatial position matching relationship of the data;

[0065] The sensor data at different positions at the same time is correlated and analyzed through a spatial position matching relationship. If the difference between adjacent sensor data exceeds a preset threshold range, the data point is marked as a potential abnormal value;

[0066] The marked potential abnormal value is subjected to time sequence continuity test by using a sliding window mechanism. If the abnormal value presents an isolated distribution mode within the time window, the data point is judged as a true abnormal value and a rejection operation is performed;

[0067] According to the cleaned data after the rejection operation, a data classification index is established according to the sensor type and the measurement parameter, and the dimension is unified and the numerical range is standardized for the measurement data of different types;

[0068] A structured data set containing data source identification, time stamp, spatial coordinates and standardized value is constructed through the classified data after the standardization processing, a high-credibility screening process basic data set is obtained, and a data integrity verification algorithm is used to evaluate the quality of the basic data set. If the data missing rate is lower than the preset integrity threshold and the data consistency is verified, it is determined that the final credible screening data set is constructed.

[0069] Specifically, in step S102, after the distributed edge computing node receives the original data stream from each sensor of the screening production line, first, a time stamp calibration algorithm is performed to unify the sensor data of different sampling frequencies to a 100ms time window, and through the GPS time system, the clock deviation of each node is controlled within ±1ms. When it is detected that the discharge port flow sensor data of the crusher is 285t / h, the belt speed of the feeder is 1.8m / s, and the material layer thickness is 150mm, the system uses a cubic spline interpolation method to complete the missing intermediate time data;

[0070] The spatial alignment module establishes a spatial correlation matrix by using the least square method according to the accurate position of the sensor in the three-dimensional coordinate system, and obtains the overall temperature distribution diagram of the device by weightedly averaging the temperature sensor readings distributed at different positions of the screening device. The abnormal value detection adopts an improved 3σ criterion combined with a sliding window technology. When it is monitored that an abnormal point exceeding 3.2 standard deviations of the mean value appears in the particle size detection data of the undersize material, the system automatically enables the isolation forest algorithm for secondary verification. The data with a pollution index exceeding 0.85 is marked as abnormal and is rejected;

[0071] The data quality evaluation module calculates the integrity index, the consistency index and the accuracy index, and the data with a comprehensive score of 0.92 or above is included in the high-credibility data set, and finally a standardized screening process data set containing 18 key parameters, with a sampling frequency of 10Hz and a data completeness rate of 98.7% is formed.

[0072] As Figures 1-2As shown, in step S103, according to the high-trust screening process dataset, a first-level analysis process is performed on the centralized big data platform: based on an online machine learning model (LSTM-RF hybrid model), a screening efficiency deviation is predicted at a frequency of seconds, and a vibration parameter compensation instruction (dynamic adjustment of amplitude ±0.5mm or frequency ±2Hz) is generated according to the prediction result.

[0073] Preferably, in step S103, vibration acceleration, rotational speed, and load current operation data collected by the screening equipment sensor are acquired, and the multi-source data stream is arranged in time sequence according to the data acquisition time stamp;

[0074] The current screening efficiency reference value is calculated according to the operation data arranged in time sequence, and if the reference value deviates from the historical mean value range, the efficiency deviation prediction process is triggered, the LSTM neural network is used to train the time sequence operation data, and the time dependence of the vibration mode in the screening process is extracted;

[0075] Random forest algorithm is used to classify and regress the time sequence features output by the LSTM neural network, and the efficiency deviation prediction result in the future time window is obtained;

[0076] The vibration parameter compensation amount is calculated according to the efficiency deviation prediction result, if the prediction deviation is positive, a compensation instruction to reduce the amplitude is generated, if the prediction deviation is negative, a compensation instruction to increase the frequency is generated, and a parameter adjustment instruction set is formed;

[0077] The amplitude compensation value in the instruction set is used to adjust the amplitude of the exciter of the screening equipment, and the working frequency of the exciter is adjusted according to the frequency compensation value, the real-time feedback data of the adjusted equipment is obtained, and the efficiency change amount before and after parameter adjustment is calculated;

[0078] The execution effect of the compensation instruction is evaluated through the efficiency change amount, if the efficiency improvement amplitude is lower than the preset improvement threshold, the compensation parameters are recalculated;

[0079] Secondary adjustment instructions are generated according to the recalculated compensation parameters to finely correct the vibration parameters;

[0080] The screening equipment is operated with the corrected vibration parameters, the equipment operation state and the screening efficiency index are continuously monitored, the efficiency stability data in the continuous operation period is obtained, and the long-term effectiveness of the parameter compensation strategy is judged;

[0081] The training sample library of the LSTM-RF hybrid model is updated according to the long-term effectiveness evaluation result, if the model prediction accuracy decreases, the model retraining process is triggered;

[0082] The screening efficiency prediction and parameter compensation tasks are continued to be performed through the optimized model after retraining.

[0083] Specifically, in step S103, after receiving the standardized screening data set, the centralized big data platform immediately starts the LSTM-RF hybrid prediction model for real-time analysis and processing. The LSTM neural network layer first performs sequence learning on 60 consecutive seconds of historical screening efficiency data, which includes 12-dimensional feature vectors such as the pass rate of oversize material, the particle size distribution of undersize material, and the vibration acceleration of the equipment. Time series features are extracted through a bidirectional LSTM structure with 128 hidden units.

[0084] When the system detects that the current screening efficiency is 87.3% and the target efficiency is 91.5%, the time series feature vector output by the LSTM layer is passed to the random forest regressor for secondary modeling;

[0085] The random forest module uses 200 decision trees, with the maximum depth of each tree set to 15 layers. Node splitting is performed using the Gini impurity index to predict the screening efficiency deviation within the next 5 seconds, and the calculated expected deviation value is -4.2%;

[0086] Based on the predicted deviation results, the vibration parameter optimization algorithm automatically calculates the compensation strategy. When the efficiency deviation exceeds the -3% threshold, the system uses the PID control algorithm to generate vibration adjustment instructions. The controller calculates the optimal compensation parameters of 1.8Hz increase in frequency and 0.3mm decrease in amplitude based on the current reference values ​​of 48.5Hz vibration frequency and 12.8mm amplitude, combined with the material property coefficient of 0.73 and the equipment load rate of 82%, through the dynamic response matrix. The entire prediction and adjustment process is completed within 1.2 seconds, ensuring that the screening efficiency quickly returns to the target range.

[0087] like Figures 1-2 As shown, in step S104, based on the vibration parameter compensation instruction, the second-level analysis process is executed on the centralized big data platform: association rule mining (Apriori-PSO optimization algorithm) is run every 5 minutes to identify the key factor combination that affects the screening efficiency. When the powder content of the screened material exceeds the standard, it is traced back to the sudden change of the mud content of the raw material or the damage of the screen, and a diagnostic strategy instruction is generated.

[0088] Preferably, in step S104, a multi-dimensional operating parameter data stream of the screening equipment is obtained, including material flow, screen vibration spectrum, motor power consumption, and screen oversize particle size distribution data, and the data is segmented and pre-processed according to a preset time window;

[0089] Based on the collected multi-dimensional data, a correlation analysis data set is constructed, and the Apriori algorithm is used to mine the strong correlation rules between screening efficiency and various operating parameters, and the combination pattern of key factors affecting screening performance is identified;

[0090] The minimum support and confidence thresholds of the Apriori algorithm are adaptively optimized by the PSO particle swarm optimization algorithm to obtain the optimal association rule parameter configuration.

[0091] The association rule mining is re-executed according to the optimized parameter configuration to obtain an accurate factor correlation matrix, and the real-time monitoring module is used to continuously detect the powder content index of the oversize material. If the powder content value exceeds the preset normal range, an abnormal state identifier is triggered.

[0092] The traceability analysis process is started according to the abnormal state identifier, the historical data of the raw material mud content and the screen running state record before and after the abnormal time point are retrieved, and whether the mud content has a mutation trend is judged through time series data comparison and analysis. If the mud content data shows obvious fluctuations, it is determined that the raw material quality abnormality is the main root cause.

[0093] If the mud content data remains stable, further analysis of the screen vibration spectrum characteristics is performed to detect whether the screen is damaged or blocked. According to the root cause type obtained by the traceability analysis, the corresponding diagnostic strategy template is automatically matched. If the root cause is a sudden change in mud content, a raw material pretreatment strengthening instruction is generated. If the root cause is screen damage, a screen repair and replacement instruction is generated, forming a complete set of diagnostic strategy instructions.

[0094] The instruction execution monitoring module is used to track the implementation progress of the diagnostic strategy instructions and obtain the equipment state feedback data after the instructions are executed.

[0095] The feedback data is used to evaluate the improvement of the powder content index. If the improvement effect does not meet the expected target, the association rule analysis is re-executed, and the factor combination recognition result is updated.

[0096] The diagnostic strategy template library is corrected by the updated factor combination result, and a dynamically optimized strategy knowledge base is established.

[0097] The execution cycle and parameter configuration of the association rule mining are continuously optimized according to the content of the knowledge base, forming an adaptive screening efficiency diagnostic analysis system.

[0098] Specifically, in step S104, after completing real-time prediction and control, the centralized big data platform automatically starts the deep correlation analysis module. The system extracts 18-dimensional quality abnormal data samples including oversize material powder content, raw material mud content, screen tension, flushing water pressure, etc. from the historical database every 5 minutes. The Apriori algorithm first sets the minimum support threshold to 0.15 and the confidence threshold to 0.78. Through frequent item set mining, it is found that when the oversize material powder content exceeds 2.8%, the correlation degree with the raw material mud content greater than 4.2% reaches 0.83, and the confidence of the screen damage area exceeding 0.6 square meters is 0.71.

[0099] To optimize the mining precision, the system introduces a particle swarm algorithm to dynamically adjust the Apriori parameters. After 45 iterations, the optimal parameter combination is converged. When the powder content of the material on the No. 3 screen machine suddenly increases to 3.4%, the association rule engine immediately activates the traceability analysis process. Through the Bayesian inference network, it is calculated that the probability of abnormal raw material mud content is 0.76, and the probability of local damage of the screen mesh is 0.68. The system generates a double diagnosis instruction to automatically dispatch the raw material pretreatment unit to increase the desliming intensity to 1.4 times the standard value, and at the same time triggers the screen mesh integrity detection program. The entire association analysis and diagnosis decision-making process is completed independently within 280 seconds.

[0100] As shown in Figures 1-2 In step S105, according to the diagnostic strategy instruction, a third-level analysis process is performed on the centralized big data platform: a screening process digital twin is constructed based on historical data, parameter combinations under different working conditions are simulated, a screen mesh inclination-feeding rate matching matrix for specific lithology is generated, and an optimization strategy real-time instruction is formed.

[0101] Preferably, in step S105, multi-dimensional time series data in the screening equipment historical operation database is obtained, including screen mesh inclination setting values, feeding rate control parameters, and screening efficiency index records of different lithological materials, a digital twin model training data set is constructed;

[0102] A deep learning neural network is used to train the training data set to establish a nonlinear mapping relationship between the screening working condition parameters and the performance output, and a screening process digital twin prediction model is obtained;

[0103] According to the digital twin prediction model, different working condition parameter combinations are input to simulate the screening performance prediction results of various lithological materials under different screen mesh inclinations and feeding rates;

[0104] The optimal parameter configuration interval of different lithological characteristic materials is analyzed through the screening performance prediction results. If the lithological hardness is high, a larger screen mesh inclination configuration is determined, and if the lithological hardness is low, a smaller screen mesh inclination configuration is determined;

[0105] A two-dimensional matching relationship matrix of lithology classification and screen mesh inclination-feeding rate is constructed according to the optimal parameter configuration interval, and a parameter optimization query table for specific lithology is established;

[0106] A real-time lithology recognition module is used to detect the lithology type of the current feed material, and the corresponding optimal screen mesh inclination and feeding rate parameter values are queried in the matching relationship matrix according to the recognition results;

[0107] The parameter value generates a screening equipment control instruction, which is sent to a screen inclination angle adjusting actuator and a feeding rate control system, so that automatic optimization adjustment of screening process parameters is realized.

[0108] Specifically, in step S105, the centralized big data platform starts a digital twin construction module, and the system automatically extracts 1.86 million sample data covering screening working condition records of 12 lithologies including granite, limestone and basalt from a production database of the past 24 months;

[0109] The platform adopts a deep neural network (DNN) algorithm to establish a physical-digital mapping model, the input layer includes 32 process parameters such as screen inclination angle, feeding rate, vibration frequency and screen size, the hidden layer is set to a 4-layer structure with 256 neurons in each layer, the weight is updated by an Adam optimizer, the learning rate is set to 0.001, and after 1200 rounds of iteration training, the model convergence accuracy reaches 0.94;

[0110] The system then starts a Monte Carlo simulation engine, generates 15,000 groups of parameter combinations for virtual screening tests for the granite working condition, finds that when the screen inclination angle is 18.5 degrees and the feeding rate is controlled at 240 tons / hour, the screening efficiency can reach an optimal value of 92.3%, based on the simulation results, the platform automatically constructs a lithology-parameter matching matrix, and sets the optimal inclination angles of granite, limestone and basalt to 18.5 degrees, 16.2 degrees and 20.1 degrees respectively, and the corresponding feeding rate thresholds are 240, 280 and 210 tons / hour;

[0111] When it is detected that No. 2 production line is switched to limestone processing, the digital twin immediately outputs the optimization instruction, automatically adjusts the screen inclination angle to 16.2 degrees, and at the same time, increases the speed of the feeder to the corresponding feeding amount of 280 tons / hour, and the whole parameter matching and automatic adjustment process is completed within 150 seconds.

[0112] As shown in FIG. 1, Figures 1-2 In step S106, the real-time instruction directly drives the vibration motor actuator, the strategy instruction is pushed to the human-computer interaction interface, and after confirmation, the production line control logic is updated, at the same time, the image recognition technology is used to monitor the aggregate particle size distribution in real time, the screening efficiency improvement rate before and after optimization is quantified, the verification data is fed back to the digital twin, the analysis model weight is iteratively updated, and a closed-loop optimization is formed.

[0113] Preferably, in step S106, real-time running state data of the jaw crusher in the crushing production line is acquired, including values of pressure sensors in the crushing cavity, motor load current change curves and discharge port adjusting position parameters, and a device running state monitoring data set is constructed;

[0114] According to the pressure sensor value change range in the operation state monitoring data set, if the pressure fluctuation exceeds the preset threshold range, it is judged that there is a risk of material blockage in the crushing cavity, it is determined that the discharge port gap parameter needs to be adjusted, the waveform characteristics of the motor load current change curve are analyzed by using the convolutional neural network, the abnormal current peak value and frequency distribution mode are identified, and the equipment wear degree evaluation result and remaining service life prediction data are obtained;

[0115] The crushing efficiency attenuation coefficient is calculated according to the wear degree evaluation result, and if the attenuation coefficient reaches a critical value, a device maintenance warning signal and a spare part replacement suggestion list are generated;

[0116] According to the maintenance warning signal, the production scheduling system is triggered to redistribute the material flow, and high-hardness materials are diverted to standby crushing equipment to obtain an optimized production load distribution scheme;

[0117] The production load distribution scheme is used to update the working time schedule of each crushing equipment, and the running speed of the conveyor belt and the discharging rhythm of the buffer bin are adjusted synchronously to determine the coordinated operation parameters of the entire production line;

[0118] By collecting the change data of the operating parameters of each device in real time, the overall equipment comprehensive efficiency index of the production line is calculated, and if the efficiency index is lower than the target value, the load distribution optimization process is re-executed.

[0119] Specifically, in step S106, the optimization instruction is directly transmitted to the servo driver of the No. 3 vibrating screen through the industrial Ethernet protocol, and the system uses the PID control algorithm to accurately adjust the vibration frequency from the current 1450 rpm to the target value of 1680 rpm. The proportion coefficient is set to 0.8, the integral time constant is 0.15 seconds, and the differential time constant is 0.05 seconds. The entire frequency adjustment process is completed in 45 seconds and reaches a steady state;

[0120] The strategy pushing module synchronously executed pushes the optimization scheme containing 7 auxiliary parameters such as amplitude adjustment and screen surface cleaning period to the central control room monitoring system through the OPC UA communication protocol, and the operation interface automatically pops up a confirmation dialog box to display the expected efficiency improvement amplitude of 6.8%. The system waits for 8 seconds and automatically confirms and updates the PLC control logic program;

[0121] A high-speed industrial camera is deployed at the screening discharge port to collect aggregate images at a frequency of 30 frames per second, different particle sizes are identified by an improved Faster R-CNN target detection algorithm, a ResNet-50 is used as a feature extraction backbone network in the convolutional layer, the anchor frame size is set to 16x16, 32x32 and 64x64, the confidence threshold is set to 0.85, the image processing unit calculates the particle proportion in the 0-5mm, 5-10mm and 10-20mm particle size intervals in real time, it is found that the screening rate of less than 5mm fine material is increased from 89.2% to 94.7% after optimization, the screening efficiency is improved by 5.5%, the verification data is fed back to the model updating module of the digital twin in real time through the Kafka message queue, the neural network weight is fine-tuned by using the incremental learning algorithm, the learning rate is dynamically decayed to 0.0005, and after 200 times of incremental training, the model prediction accuracy is improved to 0.96, forming a complete closed-loop control system of parameter optimization-execution verification-model iteration.

[0122] The application provides a sand and gravel aggregate screening system based on big data analysis, mainly comprising:

[0123] A multi-source sensing acquisition module is used for collecting raw material characteristic data, equipment operation parameters, environmental parameters and screening effect indicators in real time through a multi-source sensing network deployed in a screening production line, and performing spatio-temporal alignment and outlier filtering on the collected data by using an edge computing node;

[0124] A real-time compensation and prediction module is used for executing a first-level analysis process on a centralized big data platform, dynamically tracking the efficiency change after compensation, and triggering secondary correction and updating a model sample library if the improvement amplitude is lower than a preset threshold;

[0125] An association analysis and diagnosis module is used for executing a second-level analysis process on the centralized big data platform, dynamically updating an association rule threshold and a diagnosis strategy template library based on instruction execution feedback data;

[0126] A digital twin optimization module is used for executing a third-level analysis process on the centralized big data platform, obtaining a current material type through a lithology identification module, and generating a screen mesh inclination adjustment instruction and a feeding rate control instruction according to a matching matrix;

[0127] A closed-loop execution and feedback module is used for directly driving a vibration motor actuator by using real-time instructions, updating production line control logic after strategy instructions are pushed to a human-computer interaction interface for confirmation, monitoring aggregate particle size distribution in real time by using image recognition technology, quantifying the screening efficiency improvement rate before and after optimization, feeding verification data to a digital twin, and iteratively updating analysis model weights by using an incremental learning algorithm, to form a closed-loop optimization.

[0128] For those skilled in the art, other various corresponding changes and modifications can be made according to the technical solutions and concepts described above, and all these changes and modifications shall belong to the protection scope of the claims of the present application.

Claims

1. A sand and gravel aggregate screening method based on big data analysis, characterized in that: The following steps are involved: S101: Through the multi-source sensor network deployed on the screening production line, raw material characteristic data, equipment operating parameters, environmental parameters and screening effect indicators are collected in real time; S102: The collected data is subjected to spatiotemporal alignment and outlier filtering via the edge computing node to construct a high-reliability screening process dataset; S103: Execute the first-level analysis process on the centralized big data platform: predict the screening efficiency deviation at a frequency of seconds based on the LSTM-RF hybrid model and generate vibration parameter compensation instructions; S104: Execute the second-level analysis process: Run the Apriori-PSO optimization algorithm every 5 minutes to identify the key factor combination that affects screening efficiency. When the powder content of the screen material exceeds the standard, trace it back to the sudden change of the mud content of the raw material or the damage of the screen, and generate diagnostic strategy instructions; S105: Execute the third-level analysis process: Build a digital twin of the screening process based on historical data, generate a screen inclination-feed rate matching matrix for specific lithologies, and form real-time instructions for optimization strategies; S106: Use real-time instructions to drive the vibration motor actuator. After the strategic instructions are confirmed by the human-computer interaction interface, the production line control logic is updated. Image recognition technology is used to monitor the aggregate particle size distribution, and the verification data is fed back to the digital twin to iteratively update the model weight.

2. The sand and gravel aggregate screening method based on big data analysis according to claim 1, characterized in that: The S101 includes: A mapping model between raw material characteristics and equipment parameters is established through support vector machines to predict optimal operating parameters. Kalman filtering is used to predict dust concentration trends, and the screen inclination is dynamically adjusted according to concentration changes. The parameter adjustment effect is verified based on aggregate output efficiency and powder content.

3. The sand and gravel aggregate screening method based on big data analysis according to claim 1, characterized in that: The S102 includes: A synchronous data sequence is generated through timestamp calibration, and a sensor spatial association mapping table is established using the nearest neighbor algorithm. Isolated outliers are eliminated based on the sliding window timing test. The cleaned data is dimensionalized and the numerical range is standardized. An integrity check algorithm is used to verify that the data missing rate is lower than the preset threshold, and a structured data set is output.

4. The sand and gravel aggregate screening method based on big data analysis according to claim 1, characterized in that: The S103 includes: When the screening efficiency benchmark value deviates from the historical average, the LSTM neural network is triggered to extract the timing characteristics of the vibration pattern, and the random forest algorithm is used to predict the efficiency deviation. The amplitude or frequency compensation instructions are generated according to the deviation direction. If the efficiency improvement is lower than the preset threshold, the compensation parameters are recalculated and a secondary correction is performed.

5. The sand and gravel aggregate screening method based on big data analysis according to claim 1, characterized in that: The S104 includes: The PSO algorithm is used to adaptively optimize the minimum support and confidence thresholds of the Apriori algorithm. When the powder content exceeds the standard, the sudden change of mud content or the abnormality of the screen vibration spectrum is judged through time series data comparison and analysis. The diagnosis strategy template is matched according to the root cause type to build a dynamic knowledge base.

6. The sand and gravel aggregate screening method based on big data analysis according to claim 1, characterized in that: The S105 includes: A nonlinear mapping relationship between screen inclination, feed rate and performance indicators is established through a deep learning neural network. The screen inclination is matched based on the rock hardness: high-hardness rock types are configured with a larger inclination angle, and low-hardness rock types are configured with a smaller inclination angle. The matching matrix is ​​queried through the rock type identification module to generate optimization instructions.

7. The sand and gravel aggregate screening method based on big data analysis according to claim 1, characterized in that: The S106 includes: The Faster R-CNN algorithm is used to identify the aggregate particle size distribution in real time and quantify the improvement rate of screening efficiency. The weights of the digital twin model are fine-tuned through the incremental learning algorithm, and the learning rate is dynamically decayed to 0.0005.

8. The method according to any one of claims 1 to 7, characterized in that The screening effect indicators include: real-time output rate of aggregate in each particle size range, powder content of oversize and screening efficiency; the equipment operating parameters include: vibration frequency, amplitude, screen inclination, motor current and screen wear status.

9. A sand and gravel aggregate screening system based on big data analysis, characterized in that: include: Multi-source sensor acquisition module, used to collect screening process data in real time and perform spatiotemporal alignment and outlier filtering through edge computing nodes; The real-time compensation and prediction module is used to perform the first-level analysis process and generate vibration parameter compensation instructions based on the LSTM-RF hybrid model; The association analysis and diagnosis module is used to perform the second-level analysis process and generate diagnostic strategy instructions through the Apriori-PSO algorithm; A digital twin optimization module to perform the third-level analysis process and generate a lithology-specific screen inclination-feed rate matching matrix; The closed-loop execution and feedback module is used to drive the actuator and use image recognition technology to feed back verification data to the digital twin.

10. The system according to claim 9, characterized in that The real-time compensation and prediction module triggers a secondary correction and updates the model sample library when the efficiency improvement after compensation is lower than a preset threshold. The closed-loop execution and feedback module includes an industrial camera and an incremental learning unit, which is used to identify the aggregate particle size distribution using the Faster R-CNN algorithm.

Images