Intelligent testing system and method for abrasion of filling slurry conveying pipeline based on big data

Abstract

The invention belongs to the technical field of intelligent testing systems, and discloses an intelligent testing system and method for abrasion of a filling slurry conveying pipeline based on big data, and the system comprises a pipeline testing module, a database module and a data analysis module. The database module comprises a multi-dimensional pipeline database and a preprocessing module, the database module receives and stores the pipeline signals transmitted by the pipeline testing module and constructs the multi-dimensional pipeline database, and the preprocessing module receives the pipeline signals and performs data preprocessing to form feature signals. According to the intelligent testing system and method for the abrasion of the filling slurry conveying pipeline based on the big data, the problems that abrasion monitoring of the filling slurry conveying pipeline is not timely, and the abrasion evaluation precision is generally lower than the actual requirement can be solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent testing systems, and particularly relates to an intelligent testing system and method for wear of filling slurry conveying pipelines based on big data. Background Art

[0002] The wear monitoring and control of filling slurry conveying pipelines are crucial in mine filling operations, but the existing technologies have significant defects. Traditional methods rely on regular manual inspections and single sensors to collect basic parameters such as local pressure and flow rate, making it difficult to comprehensively reflect the true state of pipeline wear. Due to the lack of comprehensive analysis of multi-dimensional data such as slurry concentration, particle size distribution, and pipe material characteristics, the accuracy of wear assessment is generally lower than the actual requirements, resulting in maintenance decision-making deviations. Most existing wear prediction models are based on static empirical formulas and cannot dynamically respond to complex working condition changes. Measured data shows that the prediction error often exceeds 30%, especially when the rheological characteristics of the slurry change suddenly, the failure risk increases.

[0003] In addition, the problem of lag in manual regulation response is prominent, and the handling of abnormal working conditions is delayed by more than 2 hours. A case study of an iron mine shows that the unplanned shutdown rate caused by this is as high as 17%. The data island phenomenon further restricts the system efficiency. The effective integration of multi-source heterogeneous data such as historical operation data, equipment parameter libraries, and geological data is lacking. In the practical application of a gold mine, the data utilization rate is less than 40%, seriously limiting the space for intelligent upgrading.

[0004] The patent application with the publication number CN118294303A in the prior art discloses a pipeline wear degree detection method, detection device, and detection system, in which a wear detection model based on historical data is constructed. However, this model has poor analytical ability when processing signals with different spatio-temporal characteristics, and the wear detection model based only on historical wear data cannot adapt to complex and changeable pipeline wear scenarios. The patent application with the publication number CN119063667A in the prior art discloses a detection system for boiler water wall wear, in which the wear condition of the pipe wall is judged by processing the acoustic wave signal and the acoustic wave signal reflected back through the pipe wall, but it cannot predict the wear state of the pipe wall and cannot detect other wear states of the pipeline. Summary of the Invention

[0005] In view of the above disadvantages of the prior art, the present invention provides an intelligent testing system and method for wear of filling slurry conveying pipelines based on big data, which is used to solve the problems of untimely wear monitoring of filling slurry conveying pipelines and generally lower wear assessment accuracy than actual requirements. The present invention systematically overcomes the limitations of traditional technologies by integrating multi-dimensional data acquisition, dynamic simulation analysis and intelligent closed-loop regulation. Specifically, this solution constructs an intelligent monitoring system based on a multi-physical field sensing network and an adaptive learning framework, and its technical implementation path includes four core modules: a high-density sensing array realizes the holographic perception of pipeline status, a distributed data processing architecture completes the spatio-temporal alignment and feature extraction of multi-modal data, a hybrid modeling method combines the dual advantages of physical mechanism and data-driven, and a closed-loop optimization mechanism realizes the co-evolution of parameter regulation and model iteration.

[0006] The intelligent testing system for wear of filling slurry conveying pipelines based on big data provided by the present invention includes: A pipeline testing module, which forms pipeline signals. A database module, which includes a multi-dimensional pipeline database and a preprocessing module. The database module receives and stores the pipeline signals transmitted by the pipeline testing module, constructs a multi-dimensional pipeline database by integrating the historical pipeline signal data set and the external pipeline operation data obtained through networking, and the preprocessing module receives the pipeline signals and performs data preprocessing to form feature signals. A data analysis module, which constructs a pipeline wear simulation model based on the multi-dimensional pipeline database. The feature signals are input into the pipeline wear simulation model for dynamic wear analysis and form simulation prediction signals and regulation parameter signals. The regulation parameter signals are transmitted to the pipeline testing module to adjust the slurry conveying parameters, and the pipeline testing module forms slurry regulation signals. A feedback optimization module, which receives and compares the slurry regulation signals and the simulation prediction signals, extracts the difference feature values between the two for analysis to generate feedback optimization signals, and transmits them to the data analysis module and the database module.

[0007] In an embodiment of the present invention, the pipeline testing module includes a test pipeline unit, a slurry circulation unit and a data acquisition unit. The slurry circulation unit controls the slurry to be conveyed to the test pipeline unit, and the data acquisition unit is arranged in the test pipeline unit. The data acquisition unit collects various data when the slurry flows through the test pipeline unit and forms pipeline signals.

[0008] In one embodiment of the present invention, when the database module integrates the historical pipeline signal dataset and the external pipeline operation data obtained through networking, it includes a data fusion stage, a data cleaning stage, and a data repair stage. In the data fusion stage, a distributed stream processing engine is used to perform real-time access and format alignment on multi-source heterogeneous pipeline signals. The pipeline signals include, but are not limited to, pressure waveforms, vibration modal spectra, and pipe wall thickness measurement data under different geological conditions, pipe diameter specifications, and transportation media. In the data cleaning stage, an adaptive threshold filtering algorithm is used to eliminate impulse noise interference, and based on time series decomposition technology, the trend term and random fluctuation component in the pipeline signal are separated. In the data repair stage, a spatio-temporal correlation modeling method is adopted. By constructing a tensor filling model with the pipe section position as the spatial dimension and the operation cycle as the time dimension, the objective function is optimized using the alternating least squares method to complete data repair. The processed standardized data establishes a spatial index according to the pipe section topology structure, and a time series database is used to compress and store high-frequency sampling data to form the database module.

[0009] In one embodiment of the present invention, when the preprocessing module receives the pipeline signal and performs data preprocessing to form a feature signal, first, multi-scale feature extraction is performed on the pipeline signal. The wavelet packet transform is used to decompose the time-domain signal of the pipeline signal into different frequency subspaces, and the energy entropy of each sub-band is calculated as the frequency-domain feature quantity. At the same time, the empirical mode decomposition is used to obtain the intrinsic mode function components of the signal, and its instantaneous amplitude-frequency characteristics are extracted to form a time-frequency joint feature vector. After normalizing the feature vector, the principal component analysis algorithm is used to reduce the feature dimension, retaining the feature components with a cumulative contribution rate exceeding a preset threshold, and forming a feature signal. The generated feature signal is pushed to the data analysis module in real time through a message queue.

[0010] In one embodiment of the present invention, the integrated learning method is adopted to construct the pipeline wear simulation model. The prediction accuracy is improved by stacking multiple base models. The pipeline wear simulation model includes the first-layer base model, the second-layer base model, and the third-layer base model. The first-layer base model includes a fluid dynamics equation solver based on physical mechanisms to calculate the slurry flow field distribution and particle collision energy. The second-layer base model uses a gradient boosting decision tree to capture the non-linear relationship in historical data. The third-layer base model is a three-dimensional convolutional neural network to process the microscopic wear morphology features in the pipeline cross-section detection image. Finally, the weight coefficients of each base model are dynamically adjusted through a meta-learning framework to form the pipeline wear simulation model.

[0011] In one embodiment of the present invention, when performing dynamic wear analysis, according to the time-varying characteristics of the slurry transportation parameters, the sliding time window mechanism is used to divide the continuous pipeline signal into analysis segments of a fixed duration. After the feature signal within each time window is input into the pipeline wear simulation model, the predicted value of the pipe wall thickness loss rate for the next three operation cycles is output, and a simulation prediction signal is formed.

[0012] In an embodiment of the present invention, the generation of the slurry regulation signal adopts a multi-objective optimization algorithm. The defined objective function variables include but are not limited to three optimization dimensions: minimizing the wear rate, maximizing the conveying efficiency, and minimizing the energy consumption cost. The Pareto front search algorithm is used to find the optimal solution set, and the parameter combination with the highest comprehensive score is selected based on the fuzzy decision theory. Among them, the fuzzy membership function dynamically adjusts the shape parameters according to the real-time working conditions to form the slurry regulation signal.

[0013] In an embodiment of the present invention, when the feedback optimization module receives and compares the slurry regulation signal with the simulation prediction signal, the dynamic mode decomposition technology is used to extract the dominant oscillation modes of the two, and the regulation effect is evaluated by calculating the mode amplitude correlation coefficient. When the correlation coefficient is lower than the preset threshold, the differential feature depth analysis process is started. When performing the differential feature depth analysis process, after the phase space reconstruction of the differential slurry regulation signal, the largest Lyapunov exponent is calculated to judge the chaotic characteristics of the system, and the key modal components are extracted based on the singular value decomposition to form the differential feature value.

[0014] In an embodiment of the present invention, a causal reasoning framework is introduced for the differential eigenvalue analysis. A Bayesian network with the slurry parameters as the dependent variable and the wear characteristics as the result variable is constructed. The conditional probability distribution is estimated by the Markov chain Monte Carlo method, the regulation parameters with a contribution degree to the wear difference exceeding the preset value are identified, a feedback optimization signal including a parameter adjustment priority list is generated, and the parameter weights of the pipeline wear simulation model are updated through an incremental learning mechanism.

[0015] The present invention also includes an intelligent testing method for the wear of the filling slurry conveying pipeline based on big data, including: S100: Real-time collect the pipeline signals of the slurry flow in the conveying pipeline; S200: Receive and store the pipeline signals, construct a multi-dimensional pipeline database by integrating the historical pipeline signal dataset and the external pipeline operation data obtained through networking, and perform data preprocessing on the pipeline signals to generate feature signals; S300: Construct a pipeline wear simulation model based on the multi-dimensional pipeline database, input the feature signals into the simulation model for dynamic wear analysis, and generate a simulation prediction signal including predicted wear parameters and a slurry regulation signal representing the optimal operating parameters; S400: Transmit the slurry regulation signal to the execution unit of the slurry conveying system to adjust the slurry parameters; S500: Real-time obtain the adjusted pipeline operation feedback signal, perform feature comparison between the feedback signal and the simulation prediction signal, extract the differential feature value between the two, and generate an optimization instruction including a parameter correction strategy; S600: Synchronously feedback the optimization instructions to the pipeline wear simulation model and the multi-dimensional pipeline database, and drive the pipeline wear simulation model to perform parameter iterative updates.

[0016] Beneficial effects: The intelligent test system and method for wear of filling slurry conveying pipelines based on big data provided by the present invention constructs an intelligent monitoring system based on a multi-physical field sensing network and an adaptive learning framework. Its technical implementation path includes four core modules: a high-density sensing array realizes holographic perception of pipeline states, a distributed data processing architecture completes spatio-temporal alignment and feature extraction of multi-modal data, a hybrid modeling method integrates the dual advantages of physical mechanisms and data-driven approaches, and a closed-loop optimization mechanism realizes the co-evolution of parameter regulation and model iteration. The system captures multi-dimensional physical signals such as wall vibration acceleration, radial strain, and surface temperature through electromagnetic-acoustic composite sensors deployed at key pipeline nodes, and cooperates with a high-speed industrial bus to achieve microsecond-level timing synchronization, constructing a dynamic monitoring network covering the entire length of the pipeline. The data processing layer adopts a stream-batch integrated computing framework to perform preprocessing operations on the original signals such as wavelet denoising, outlier removal, and construction of feature vectors, and maps discrete sampling points to a continuous three-dimensional field distribution model through spatio-temporal coding technology. It solves the problems of untimely wear monitoring of filling slurry conveying pipelines and generally low wear assessment accuracy lower than actual requirements. Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is the system architecture diagram of the intelligent test system for wear of filling slurry conveying pipelines based on big data; Figure 2 It is the step flow chart of the intelligent test method for wear of filling slurry conveying pipelines based on big data. Detailed Embodiments

[0019] The following illustrates the embodiments of the present invention through specific specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through different specific embodiments. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0020] The intelligent test system and method for wear of filling slurry conveying pipeline based on big data of the present invention are mainly applied to the working scenario of pipeline transportation. As one of the core ways of modern industrial material transportation, pipeline transportation plays a key role in fields such as mine filling, metallurgical smelting, petrochemical industry, etc. When the filling slurry flows at high speed inside the pipeline, the continuous friction between solid particles and the pipe wall causes progressive wear. This wear process has complex characteristics of non-linearity and multi-factor coupling, and it is difficult for traditional detection means to achieve accurate monitoring and trend prediction. The current commonly used periodic manual detection method in the industry has significant lag. Usually, problems can only be found when the wall thickness reduction exceeds the safety threshold, resulting in frequent unplanned shutdowns and rising maintenance costs. Although online monitoring devices based on technologies such as vibration sensing and ultrasonic thickness measurement have been gradually popularized in recent years, problems such as single data analysis dimension and insufficient model generalization ability have not been fundamentally solved. Especially when dealing with wear prediction under multi-condition and multi-material composite action, existing systems generally show defects such as poor adaptability and high false alarm rate. The main limitations of the existing technical system are reflected in three levels: the spatio-temporal resolution in the data acquisition link is insufficient. Most systems adopt a sparse point layout strategy and it is difficult to capture the wear gradient distribution characteristics of the entire pipeline area; the data processing level lacks the ability of deep fusion of multi-source heterogeneous data. Key parameters such as vibration signals, pressure waveforms, and material properties are often stored in an isolated form, and an association model reflecting the wear mechanism has not been constructed; the decision support level relies on static empirical models and cannot adapt to the complexity of actual working conditions such as dynamic changes in slurry composition and degradation of pipe material properties. Typical cases show that a conventional monitoring system adopted by a copper mine continuously showed that the wear rate was within the safe range for three consecutive months, but actual detection found that the local wear amount at the elbow had exceeded the critical value. The fundamental reason was that the system algorithm failed to identify the turbulent impact characteristics hidden in the high-frequency vibration signal, resulting in the key warning information being mis-eliminated by the noise filtering mechanism.

[0021] Please refer to Figure 1 - Figure 2, the intelligent wear testing system for the filling slurry conveying pipeline based on big data of the present invention includes a pipeline testing module, a database module, and a data analysis module. The pipeline testing module forms pipeline signals. The database module includes a multi-dimensional pipeline database and a preprocessing module. The database module receives and stores the pipeline signals transmitted by the pipeline testing module, constructs the multi-dimensional pipeline database by integrating the historical pipeline signal data set and the external pipeline operation data obtained through networking. The preprocessing module receives the pipeline signals and performs data preprocessing to form characteristic signals. The data analysis module constructs a pipeline wear simulation model based on the multi-dimensional pipeline database. The characteristic signals are input into the pipeline wear simulation model for dynamic wear analysis and form simulation prediction signals and regulation parameter signals. The regulation parameter signals are transmitted to the pipeline testing module to adjust the slurry conveying parameters, and the pipeline testing module forms slurry regulation signals. The feedback optimization module receives and compares the slurry regulation signals and the simulation prediction signals, extracts the difference characteristic values between the two for analysis to generate feedback optimization signals, and transmits them to the data analysis module and the database module.

[0022] In the intelligent wear testing system for the filling slurry conveying pipeline based on big data, the pipeline testing module serves as the front-end perception layer of the entire system and undertakes the key task of obtaining the original data of the pipeline operation state in real time. This module is composed of three parts: a test pipeline unit, a slurry circulation unit, and a data acquisition unit. Through a highly coordinated hardware layout and intelligent control logic, it realizes the accurate capture and dynamic regulation of the pipeline wear characteristics under complex working conditions. The test pipeline unit is the core carrier of physical experiments. Its design needs to consider both the simulation of real working conditions and parameter adjustability. Usually, a modular structure design is adopted, and the pipe section material, pipe diameter specification, and connection form can be quickly replaced according to different experimental requirements. For example, for the scenario of highly abrasive slurry, the test pipeline may adopt wear-resistant pipe sections with an inner lining of ceramic composite materials, while for high-pressure conveying conditions, high-strength alloy steel materials will be selected. Profiled structures (such as corrugations, protrusions, etc.) can be preset inside the pipeline to simulate the common local wear hot spots in actual engineering. At the same time, the stress monitoring device at the flange connection can sense the impact of mechanical vibration on the integrity of the pipe wall in real time. Parameters such as the installation angle and bending radius of the test pipeline can be dynamically adjusted through an electric adjustment mechanism to reproduce the influence of the common spatial orientation changes in the mine filling system on the slurry flow pattern.

[0023] Specifically, as a power and medium supply system, the core function of the slurry circulation unit is to accurately control the physical properties and flow parameters of the slurry. The unit usually includes key components such as a high-pressure plunger pump, a slurry mixing bin, a concentration adjustment device and a temperature control system. The high-pressure plunger pump adopts variable frequency drive technology, which can steplessly adjust the slurry flow rate within the range of 0.5-8m / s, and cooperate with a digital flowmeter to achieve closed-loop control of the delivery volume. The slurry mixing bin is equipped with a double-axis agitator and an ultrasonic dispersion device to ensure the uniform suspension of solid particles (such as tailings, aggregates, etc.) in the liquid medium. The concentration adjustment device monitors the solid content of the slurry in real time through an online density meter, and links the dry powder feeder and the dilution water valve to achieve dynamic balance of concentration, with an adjustment accuracy of up to ±1.5%. The temperature control system maintains the slurry in a set temperature range (usually 5-60°C) through a coil heat exchanger to simulate the effects of different seasons or underground environments on the rheological properties of the slurry. In particular, the unit integrates a particle size analyzer and a rheometer, which can obtain the slurry particle grading curve and apparent viscosity data in real time, providing multi-dimensional input parameters for subsequent wear mechanism analysis. As an information perception network, the data acquisition unit realizes comprehensive monitoring of the pipeline operation status through a distributed sensor array. Sensor nodes are arranged at key positions of the test pipeline according to the spatial topological structure, including straight pipe sections, elbows, reducer sections and other vulnerable areas. The pressure sensor adopts the piezoelectric principle to capture the dynamic pressure fluctuations of the pipe wall at a sampling frequency of 1000Hz. The measurement range covers 0-25MPa and the resolution reaches 0.01MPa. Vibration monitoring uses a combination of a triaxial accelerometer and a fiber Bragg grating (FBG) sensor. The former captures the mechanical vibration spectrum with a high-frequency sampling of 10kHz, and the latter analyzes the pipeline strain distribution through wavelength offset, with a spatial resolution of up to 5mm. The pipe wall thickness measurement adopts pulsed eddy current and electromagnetic ultrasonic composite detection technology to achieve online thickness measurement with an accuracy of 0.1mm under non-contact conditions, and is not affected by surface coatings or deposits. The environmental parameter sensor group continuously records the changes in ambient temperature, humidity and temperature gradient on the pipe surface, providing auxiliary data for multi-physical field coupling analysis. All sensor signals are transmitted to the data acquisition card via anti-interference shielded cables, and are processed through 24-bit AD conversion and digital filtering to form a standardized data packet. The time synchronization error is less than 1ms, ensuring the time and space alignment of multi-modal data.

[0024] In an embodiment of the present invention, the three major sub-units achieve deep collaboration through industrial Ethernet and fieldbus. When starting the test process, the slurry circulation unit first prepares the slurry according to preset parameters (such as a concentration of 65% and a flow rate of 4 m / s) and establishes a stable flow. The test pipeline unit synchronously adjusts to the target inclination angle (such as a 30° elevation angle) and bending radius (such as a 3D elbow) to simulate the spatial characteristics of the actual conveying line. The data acquisition unit then starts full-channel sampling and uploads the original signals such as pressure pulsation, vibration spectrum, and wall thickness change to the central controller in real time. During this process, the intelligent control algorithm continuously analyzes the data characteristics and dynamically adjusts the slurry parameters to cover a wider range of working conditions. For example, when it is detected that the vibration energy abnormally accumulates at a certain elbow section, the system automatically triggers a gradient pressure boosting program, gradually increasing the pumping pressure until the resonance frequency shift of the pipe wall is observed, so as to determine the structural strength critical point of this pipe section. This active test mode breaks through the limitations of traditional passive monitoring, can stimulate potential failure modes under controllable conditions, and provides rich data samples for wear mechanism research.

[0025] As Figure 1 shown, the construction of the database module and the data processing flow are the core basis for its intelligent analysis. This module constructs a pipeline database covering multiple dimensions and multiple spatio-temporal scales by integrating real-time signals from the pipeline test module, the historical accumulated pipeline operation data set, and the external pipeline operation data obtained through networking. During the data integration process, first, the data fusion stage is carried out, and a distributed stream processing engine is used to connect multi-source heterogeneous data streams. These data sources include pressure waveforms, vibration modal spectra, pipe diameter specification parameters, conveying medium properties (such as slurry concentration, particle size distribution), and periodic measurement data of the wall thickness under different geological conditions. Due to the differences in the acquisition frequencies, data formats, and communication protocols of different sensors, the system uniformly encodes metadata such as timestamps, spatial coordinates, and data types through a custom format alignment algorithm and establishes a spatio-temporal correlation index between the data. For example, the millisecond-level pulse signal of the pressure sensor and the hourly-level data of the wall thickness detector are time-aligned through an interpolation algorithm, and the vibration spectrum data of different pipe sections establish a spatial mapping relationship through the pipeline topology structure.

[0026] Specifically, in the data cleaning stage, an adaptive threshold filtering algorithm is adopted to eliminate impulse noise interference. This algorithm dynamically calculates the statistical features (such as mean and variance) within the signal window and adjusts the filtering threshold in real time in combination with the sliding window mechanism. For sudden spike noises, the system introduces morphological filtering technology to identify and remove abnormal pulses while retaining the true pressure fluctuation characteristics. For the common baseline drift problem in vibration signals, time series decomposition technology is used to separate the original signal into a trend term, a periodic term, and a random fluctuation component. The trend term reflects the stiffness degradation caused by pipe fatigue, the periodic term corresponds to the natural frequencies of pump operation or slurry pulses, and the random term is used to capture abnormal events. The data repair stage focuses on solving the problem of data loss caused by sensor failures or communication interruptions. By constructing a spatio-temporal correlation tensor filling model, the pipeline network is regarded as a three-dimensional structure (pipe segment location, operation time, measurement parameters), and the alternating least squares method is used to optimize the objective function to iteratively estimate the missing values. For example, when 12 hours of vibration data are lost due to a fiber break in a certain pipe segment, the system reconstructs the data distribution during this period based on the similarity of vibration modes in adjacent pipe segments and the time evolution law of historical data under the same working conditions. The processed standardized data is indexed by an R-tree according to the spatial topology structure of the pipe segments. The high-frequency sampled data is compressed and stored in a time series database, and the storage space is reduced to less than 30% of the original data through differential coding and run-length encoding. The preprocessing module extracts multi-scale features from the original pipeline signal and decomposes the time-domain signal into different frequency subspaces using wavelet packet transform. Taking the pressure pulsation signal as an example, the db4 wavelet basis is selected for 5-layer decomposition to obtain 32 frequency band components, and the energy entropy of each sub-band is calculated to characterize the complexity of the frequency-domain energy distribution. For non-stationary vibration signals, empirical mode decomposition is used to adaptively decompose them into multiple intrinsic mode functions, and the instantaneous amplitude envelope and Hilbert spectrum of each IMF component are extracted to form a time-frequency joint feature vector. For example, after the vibration signal at a certain elbow is decomposed by EMD, the amplitude mutation of the 3rd IMF component in a specific frequency band (such as 800 - 1200 Hz) is identified as a typical feature of particle group collisions. After the feature vector is standardized by Z-score, principal component analysis is used for dimensionality reduction, and the principal components with a cumulative contribution rate exceeding 95% are retained, compressing the original 128-dimensional features to 18 dimensions, significantly improving the computational efficiency of subsequent models. The dimension-reduced feature signals are pushed to the data analysis module in real time through the Kafka message queue to ensure a throughput capacity of processing more than 100,000 feature data streams per second.

[0027] Furthermore, an integrated learning method is adopted to construct the pipeline wear simulation model, and multi-level feature fusion is achieved by stacking physical mechanism models, machine learning models, and deep learning models. The first-layer base model is a physical simulator based on computational fluid dynamics, which solves the Navier-Stokes equations to simulate the slurry flow field distribution and combines the discrete element method to calculate the collision energy distribution of solid particles. The key physical quantities output by this layer include wall shear stress, particle impact frequency, and kinetic energy transfer efficiency. The second-layer base model uses the XGBoost gradient boosting decision tree, which inputs the operating parameters (such as flow rate and concentration) and wear amount labels in historical data to capture the non-linear interaction effects between parameters. For example, when the slurry concentration increases from 60% to 70%, the decision tree model can identify that the flow rate needs to be reduced by 0.8 m / s synchronously to maintain the equivalent wear rate. The third-layer base model is a three-dimensional convolutional neural network, which processes the microscopic wear images taken by the pipeline endoscope and extracts surface morphology features such as pits and scratches through multi-scale convolutional kernels. Finally, the weight coefficients of each base model are dynamically adjusted through the meta-learning framework. For example, a higher weight is assigned to the CFD model under stable flow conditions, while the decision-making contribution of the CNN model is enhanced under complex turbulent conditions to form an adaptive hybrid simulation model. The dynamic wear analysis uses a sliding time window mechanism to process continuous data streams, and the window length is dynamically adjusted according to the change frequency of slurry transportation parameters, usually adaptively selected within the range of 10 minutes to 2 hours. The feature signals within each time window are normalized and then input into the simulation model to output the predicted values of the wall thickness loss rate for the next three operating cycles (usually 72 hours). For example, when it is detected that the vibration energy entropy in the current time window increases by 15% compared with the previous period, the model will combine the flow field simulation results to predict that the wear rate on the outer side of the elbow will increase to 0.12 mm / h within the next three days, exceeding the safety threshold of 0.1 mm / h, thus triggering a warning signal. The Monte Carlo method is used for uncertainty quantification during the prediction process, and a wear rate probability distribution curve is generated through 500 random samplings to provide a confidence interval reference for risk decision-making. In terms of the iterative optimization of the simulation model, the system introduces an online learning mechanism. Whenever new pipeline detection data is stored in the database, the model incremental training process is automatically started. For example, during a certain on-site maintenance, it is found that the actual wear amount is 8% lower than the predicted value. The system adds this difference data to the training set, fine-tunes the convolution kernel weights of the CNN network through the backpropagation algorithm, and reduces the wear rate estimation value under this operating condition in the next prediction. This continuous learning mechanism enables the model prediction error to gradually converge over time. The measured data in a certain copper mine shows that the average prediction error decreases from the initial 9.2% to 4.7% after the system runs for 6 months. The generation of slurry control signals relies on a multi-objective optimization algorithm, and the defined objective function covers three dimensions: wear rate, transportation efficiency, and energy consumption cost.Specifically, the wear rate target term adopts an exponential penalty function. When the predicted wear rate approaches the safety threshold, the objective function value rises sharply to force parameter adjustment. The conveying efficiency target term is positively correlated with the slurry flow rate and concentration, but is limited by the pump power. The energy consumption cost term is related to the pumping pressure and the operation duration, forming a non-linear constraint relationship. The optimal solution set is found through the Pareto front search algorithm, and 200 - 300 non-dominated solutions are screened out from millions of parameter combinations, and then the final parameters are selected based on the fuzzy decision-making theory. The fuzzy membership function is dynamically adjusted according to the real-time working conditions. For example, during high-load periods of the equipment, the energy consumption weight is preferentially reduced, and during the safety warning state, the wear suppression priority is greatly increased. In a certain actual case, the system completed the multi-objective optimization calculation within 30 seconds, generating a control plan that reduced the flow rate from 4.5 m / s to 4.1 m / s and added 0.3% drag reducer at the same time, reducing the wear rate by 22% while only increasing the energy consumption by 5%, achieving the optimal balance.

[0028] As Figure 1 shown, the feedback optimization module compares the difference features between the control signal and the predicted signal through the dynamic mode decomposition technology. First, the time series data of both are mapped into a high-dimensional phase space, and the dominant spatio-temporal modes are extracted through singular value decomposition (SVD). For example, there is a phase difference between the decay mode of the actual wear rate after control and the predicted exponential decay mode, indicating that the model does not fully consider the material fatigue accumulation effect. The difference feature values are analyzed through the causal inference framework, and a Bayesian network is constructed to infer the key influencing factors. In a certain analysis, the system identified that the measurement error of the slurry viscosity was the main cause of the prediction deviation, and immediately initiated the data review process, and found that there was a systematic error of 0.8 Pa·s in the viscosity sensor. The optimization instruction then updated the sensor calibration coefficient and triggered the parameter retraining of the simulation model, improving the subsequent prediction accuracy by 12%. The implementation of the entire technical chain has significantly improved the intelligent level of pipeline wear management. At the data level, the efficient integration and repair technology of multi-source heterogeneous data has broken through the information island limitation of traditional systems; at the model level, the integrated learning framework has achieved the complementary advantages of physical mechanism and data-driven methods; at the control level, the dynamic optimization and feedback mechanism ensure the adaptive ability of the system under complex working conditions. An industrial test in an iron ore mine shows that the system has advanced the early warning time of pipeline abnormal wear from 48 hours of traditional methods to 72 hours, the wear prediction accuracy of key parts reaches 96.3%, and at the same time, the pumping energy consumption is reduced by 18% through the optimized control strategy, saving more than 1.5 million yuan in electricity costs annually.

[0029] Specifically, the generation and optimization process of the slurry control signal is the core link of the entire intelligent control chain. This process achieves a dynamic balance among wear control, conveying efficiency, and energy consumption cost through a multi-objective optimization algorithm. The key lies in constructing a mathematical model that can reflect complex working condition constraints and designing an efficient solution strategy. The objective function defined by the system includes three main dimensions: minimizing the wear rate requires suppressing the impact damage of slurry particles on the pipe wall; maximizing the conveying efficiency requires maintaining a reasonable slurry flow rate and concentration; minimizing the energy consumption cost involves the optimal configuration of pumping power. There is a non-linear coupling relationship among these three objectives. For example, increasing the flow rate can increase the conveying volume but will exacerbate wear and increase energy consumption, while decreasing the concentration can reduce particle impact but may cause slurry segregation and affect the filling quality. To solve such multi-objective conflicts, the system uses an improved NSGA-II algorithm for Pareto front search, and screens out the optimal solution set from a large number of parameter combinations through fast non-dominated sorting and crowding degree calculation. Specifically, in the initialization stage of the algorithm, a population of 5000 individuals is randomly generated, and each individual's encoding includes 10 control parameters such as flow rate, concentration, and pump pressure. During the iteration process, the population diversity is maintained through simulated binary crossover and polynomial mutation operations, and at the same time, a constraint handling mechanism is introduced to eliminate invalid solutions that violate the pipe strength or equipment power limits. After 100 generations of evolution, the algorithm outputs a Pareto front containing 200 - 300 non-dominated solutions, and these solutions form an optimal trade-off surface in the three-dimensional objective space. The fuzzy decision-making theory plays a key role at this stage, and it adapts to real-time working condition changes by dynamically adjusting the shape parameters of the membership function. The system presets triangular and trapezoidal membership functions to correspond to the satisfaction intervals of each objective. For example, the membership function of the wear rate objective reaches complete satisfaction (membership degree 1) at 0.08 mm / h, and the satisfaction degree drops to 0 when the rate exceeds 0.12 mm / h. During the decision-making process, the fuzzy inference engine calculates the comprehensive score by integrating the membership degree values of each objective, and selects the parameter combination with the highest total score as the final control scheme. To cope with the dynamic nature of the working conditions, the system monitors external factors (such as grid electricity price fluctuations and production task urgency) in real time and adjusts the objective weight coefficients. For example, during peak electricity consumption periods, the priority of the energy consumption cost objective is automatically increased, and during safety warning states, the wear suppression weight is set to the highest. In an actual case, when the system detected that the wear rate of a certain pipe section was approaching the threshold, it chose a solution through fuzzy decision-making to reduce the flow rate from 4.3 m / s to 3.9 m / s and add 0.2% drag reducer at the same time, resulting in an 18% decrease in the wear rate and only a 7% increase in energy consumption, successfully avoiding unplanned shutdowns.

[0030] In an embodiment of the present invention, the feedback optimization module deeply analyzes the difference between the control effect and the prediction result through the dynamic mode decomposition technique. After the control signal acts on the actual system, the data acquisition unit continuously monitors key parameters such as the wall thickness and vibration spectrum to form a real-time feedback signal. After the system spatially and temporally aligns the feedback signal with the simulation prediction signal, the DMD technique is applied to extract the dominant spatio-temporal modes of the two. Specifically, the time series data is constructed into a high-dimensional state matrix, and the low-rank approximation space is obtained through singular value decomposition, and then the eigenvalues and modes of the linear dynamics operator are solved. For example, after a certain regulation, the decay mode of the actual wear rate shows an exponential characteristic, while the prediction signal shows a linear decay. The correlation coefficient of the mode amplitudes of the two is calculated to be 0.65, which is lower than the preset threshold of 0.8, triggering the in-depth analysis process of the difference characteristics. In this process, the system first performs phase space reconstruction on the difference signal, maps the one-dimensional sequence to the high-dimensional phase space by the time delay embedding method, and judges the dynamic characteristics of the system by calculating the largest Lyapunov exponent. When the exponent is greater than zero, it indicates that the system is in a chaotic state and a non-linear analysis method needs to be adopted; if the exponent approaches zero, a linear approximation model is applicable. Subsequently, singular value decomposition is performed on the reconstructed phase space trajectory, and the first 5 dominant mode components are extracted as the difference eigenvalues, which usually correspond to the physical mechanisms not fully captured by the model (such as the material fatigue accumulation effect or unmodeled environmental interference factors). In the difference eigenvalue analysis link, the system introduces a causal inference framework to construct a Bayesian network to reveal the causal relationship between the control parameters and the wear characteristics. The network nodes include 35 variables in total, such as slurry parameters (flow rate, concentration, etc.), equipment status (pump pressure, temperature, etc.) and wear indicators (thickness loss rate, surface roughness, etc.). The conditional probability distribution is estimated by the Markov chain Monte Carlo method, and the parameter space is traversed by Gibbs sampling. After 5000 iterations, a stable probability network is obtained. For example, when analyzing a certain prediction deviation, the network shows that the contribution of the measurement error of the slurry viscosity to the wear difference reaches 42%, far exceeding other factors. The system immediately starts data tracing, discovers that there is a system drift of 0.6 Pa·s in the viscosity sensor, and immediately triggers the calibration program and updates the historical data tags in the database. At the same time, the difference eigenvalues are encoded as feature vectors and input into the incremental learning module, and the parameters of the pipeline wear simulation model are fine-tuned through the online backpropagation algorithm. In specific implementation, the weights of the last fully connected layer of the three-dimensional convolutional neural network are adaptively adjusted at a learning rate of 0.01, and the subtree structure of the gradient boosting decision tree incorporates new feature interaction relationships by adding split nodes. This progressive optimization strategy enables the model to quickly adapt to new working conditions without forgetting the existing knowledge. The application data of an iron ore mine shows that after three incremental learning processes, the prediction error of the model under similar working conditions is reduced from 12.3% to 6.8%.

[0031] Furthermore, the generation of the feedback optimization signal not only includes parameter adjustment suggestions but also involves system-level state assessment and maintenance strategy optimization. After identifying the key influencing factors, the system combines the equipment health model to predict the remaining service life and generates a multi-level response strategy: for short-term reversible deviations (such as sensor drift), it automatically performs online calibration and model retraining; for medium-term trend changes (such as material property degradation), it suggests adjusting the maintenance cycle or spare part procurement plan; for long-term systematic risks (such as design defects), it triggers engineering modification suggestions. For example, in a certain case, the system detected that the wear rate at the elbow part continuously exceeded the predicted value. Through causal analysis, it was found that the insufficient curvature radius of the elbow led to the intensification of secondary flow. Finally, it was suggested to increase the curvature radius from 2D to 3D. After implementation, the wear rate in this area decreased by 37%. The entire feedback optimization process forms a closed-loop control system, and its innovation is reflected in three aspects: First, the combination of dynamic mode decomposition and causal reasoning realizes the traceability analysis from data differences to physical mechanisms, breaking through the limitation of traditional statistical methods that can only find correlations but cannot determine causal relationships; second, the incremental learning mechanism ensures the continuous evolution of the model and can track the time-varying characteristics of the pipeline system; finally, the multi-level response strategy organically connects the microscopic parameter adjustment and macroscopic maintenance decision-making, significantly improving the overall reliability of the system. The operation data of a demonstration project shows that this module reduces the number of unexpected shutdowns by 55% and the unplanned maintenance cost by 41%, fully verifying its technical advantages. At the technical implementation level, the system adopts a distributed computing architecture to handle the massive data processing requirements. The multi-objective optimization algorithm is deployed on the GPU cluster, and the CUDA parallel computing is used to accelerate the evolution process, controlling the search time in the million-level parameter space within 30 seconds. The dynamic mode decomposition and Bayesian network reasoning run on the CPU cluster, and the parallelization of large-scale matrix operations is achieved through the MPI protocol. The incremental learning process uses the Elastic Weight Consolidation (EWC) algorithm to prevent catastrophic forgetting, and the weight update direction is constrained by calculating the parameter importance matrix to ensure the compatibility of old and new knowledge. All optimization instructions and feedback data are seamlessly integrated with the industrial control system through the OPC UA protocol to achieve end-to-end automation from decision-making to execution.

[0032] Such as Figure 2As shown in the figure, the intelligent test method for wear of the filling slurry conveying pipeline based on big data according to the present invention includes: S100: Real-time collect the pipeline signals of the slurry flow in the conveying pipeline. S200: Receive and store the pipeline signals, construct a multi-dimensional pipeline database by integrating the historical pipeline signal data set and the external pipeline operation data obtained through networking, and perform data preprocessing on the pipeline signals to generate characteristic signals. S300: Construct a pipeline wear simulation model based on the multi-dimensional pipeline database, input the characteristic signals into the simulation model for dynamic wear analysis, and generate a simulation prediction signal containing predicted wear parameters and a slurry regulation signal characterizing the optimal operation parameters. S400: Transmit the slurry regulation signal to the execution unit of the slurry conveying system to adjust the slurry parameters. S500: Real-time obtain the adjusted pipeline operation feedback signal, compare the characteristics of the feedback signal with the simulation prediction signal, extract the difference characteristic value between the two and generate an optimization instruction containing a parameter correction strategy. S600: Synchronously feedback the optimization instruction to the pipeline wear simulation model and the multi-dimensional pipeline database, and drive the pipeline wear simulation model to perform parameter iterative update.

[0033] Furthermore, the present invention constructs a complete closed-loop control system, realizing intelligence and adaptability throughout the whole process from data acquisition to model iteration. The core of this method lies in the synergistic effect of multi-source data fusion, dynamic simulation modeling, and real-time feedback optimization, breaking through the static analysis and experience-dependent mode in traditional pipeline wear management and providing a systematic solution for pipeline health management under complex working conditions. During specific implementation, the system first captures multi-physical field signals of the slurry flow inside the pipeline in real time through a distributed sensor network, including pressure pulsation, vibration spectrum, wall thickness change, and environmental parameters, etc. These raw data are continuously input into the central processor at a sampling rate of tens of thousands of points per second. For example, in a test case of a DN300 pipeline in an iron mine, 48 sensor nodes arranged along a 200-meter pipe section generate a data packet containing 128-dimensional features such as pressure, vibration, and temperature every 10 seconds, which is transmitted to the edge computing node through industrial Ethernet for preliminary filtering and timestamp alignment to ensure the temporal consistency of subsequent analysis. In the data storage and preprocessing stage, the system adopts a hierarchical architecture to process the massive data stream. The original signal first enters the buffer pool for format standardization, unifying data with different sampling frequencies to a 10ms time base through linear interpolation, and at the same time attaching spatial coordinate tags (such as pipeline mileage stake numbers) to each data point. Historical data and real-time data are mixedly stored through a distributed database (such as Cassandra), where high-frequency vibration data uses columnar storage to optimize query efficiency, while low-frequency thickness detection data establishes a time series index to support fast backtracking analysis. The preprocessing module eliminates on-site electromagnetic interference through wavelet threshold denoising and applies empirical mode decomposition technology to separate the trend term and random fluctuation components in the signal. For example, the interference from a crusher operation (characteristic frequency 25Hz) mixed in the vibration signal of a copper mine pipeline is successfully filtered out, retaining the components in the 0.5 - 15Hz frequency band that truly reflect the slurry flow state. The feature extraction link adopts a multi-scale analysis framework, performs wavelet packet decomposition on the pressure signal to obtain the energy entropy of 32 sub-bands, calculates the time-domain kurtosis index and frequency-domain centroid frequency for the vibration signal, and finally generates a vector set containing 56-dimensional features, which is compressed to 18 dimensions through principal component analysis to reduce the model complexity. The core of dynamic wear analysis lies in constructing a hybrid simulation model, which innovatively combines the advantages of physical mechanism and data-driven methods. During the model training stage, the system integrates historical operation data, laboratory wear test data, and external engineering case libraries, and maps the wear laws of different pipe diameters and materials to a unified feature space through transfer learning technology. For example, after dimensionless processing, the wear data of a DN200 steel pipe is used to assist in training the prediction model of a DN250 composite pipeline, improving the initial model accuracy by 23%. During online prediction, the model adopts a sliding time window mechanism to process the real-time data stream, and the length of each window is dynamically adjusted according to the working condition volatility (usually 10 - 30 minutes). The feature vectors within the window are normalized and then input into the integrated model.The application case of a gold mine shows that when the vibration energy entropy of a certain elbow section suddenly increases by 20%, the model combines the results of fluid flow simulation to predict that the wear rate of this part will rise to 0.15 mm / h within the next 72 hours, exceeding the safety threshold of 0.1 mm / h, and then triggers the generation process of the control instruction. During the prediction process, uncertainty quantification is carried out synchronously. The probability distribution curve of the wear rate is generated through Monte Carlo simulation, providing confidence interval support for risk assessment. For example, a certain prediction shows that the 95% confidence interval is [0.12, 0.18] mm / h, providing a key basis for operation and maintenance decisions. In the optimization link of the control parameters, a multi-objective evolutionary algorithm is adopted to seek the optimal balance between transportation efficiency and energy consumption cost on the premise of ensuring wear safety. The algorithm encodes 10 adjustable parameters such as flow velocity, concentration, and pump pressure as decision variables, defines a fitness function including wear inhibition, production capacity requirements, and energy consumption constraints, and searches for the Pareto optimal set in the million-level solution space through the NSGA-III algorithm. During the actual operation of a lead-zinc mine, the system completes 500 generations of evolutionary calculations within 45 seconds, screens out 153 non-dominated solutions, and then selects the solution of reducing the flow velocity from 5.1 m / s to 4.7 m / s and adding 0.15% drag reducer through fuzzy decision-making, reducing the wear rate by 19% while only increasing the energy consumption by 5%. The control instruction is sent to the PLC control system through the OPC UA protocol, driving the actuators such as variable frequency pumps and dosing devices to complete parameter adjustment, and the whole process is fully automated with a manual intervention rate of less than 2%.

[0034] In an embodiment of the present invention, the feedback optimization mechanism constitutes the self-evolution ability of the system. After each regulation and control is implemented, the data acquisition module continuously monitors the actual wear progress and performs spatio-temporal alignment and comparison with the predicted values. For the difference analysis, the dynamic mode decomposition technique is used to extract the dominant mode features. For example, after a certain regulation and control, the actual wear curve shows an exponential decay characteristic, while the output of the prediction model is a linear decay. The system identifies the material fatigue effect not considered by the model through the modal correlation coefficient (0.68 is lower than the threshold of 0.8). Subsequently, the causal inference engine constructs a Bayesian network and analyzes that the measurement error of the pipe hardness is the main influencing factor (contribution degree of 37%). After triggering the material detection process, it is found that the actual Brinell hardness is 8.2 HB lower than the design value. The system then starts incremental learning, adds the new material property data to the training set, and updates the neural network parameters through the elastic weight consolidation algorithm, reducing the subsequent prediction error by 6.5%. At the same time, a new dimension of "hardness compensation coefficient" is added to the optimization instruction library to ensure that the regulation and control strategies under similar working conditions automatically adapt to the changes in material properties. The implementation case of an iron ore mine shows that the system shortens the early warning response time of pipeline abnormal wear from 36 hours of the traditional method to 8 hours, and the wear prediction accuracy of key parts reaches 94.7%. By continuously optimizing the regulation and control strategies, the average service life of the pipeline is extended from 11 months to 16 months, and the annual maintenance cost is reduced by 2.8 million yuan. More notably, the self-learning ability of the system enables the model prediction error to converge from the initial 8.9% to 4.3% after 6 months of operation, demonstrating a strong environmental adaptation ability. This "perception - decision - execution - optimization" closed-loop system not only solves the problems of data islands and model rigidity in traditional methods, but also realizes the continuous evolution of the system through the real-time feedback mechanism, providing reliable technical support for the construction of intelligent mines.

[0035] The intelligent wear testing system and method for filling slurry conveying pipelines based on big data of the present invention systematically overcome the limitations of traditional technologies by integrating multi-dimensional data acquisition, dynamic simulation analysis, and intelligent closed-loop regulation and control. Specifically, this solution constructs an intelligent monitoring system based on a multi-physical field sensing network and an adaptive learning framework, and its technical implementation path includes four core modules: a high-density sensing array realizes the holographic perception of the pipeline state, a distributed data processing architecture completes the spatio-temporal alignment and feature extraction of multi-modal data, a hybrid modeling method combines the dual advantages of physical mechanism and data-driven, and a closed-loop optimization mechanism realizes the co-evolution of parameter regulation and model iteration.

[0036] Therefore, through the intelligent wear testing system and method for filling slurry conveying pipelines based on big data of the present invention, the problems of untimely wear monitoring of filling slurry conveying pipelines and generally low wear assessment accuracy lower than the actual requirements can be solved.

[0037] The above embodiments are only illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical idea disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. Intelligent testing system for wear of filling slurry conveying pipeline based on big data, characterized by: include: A pipeline testing module, wherein the pipeline testing module forms a pipeline signal; A database module, the database module includes a multi-dimensional pipeline database and a preprocessing module, the database module receives and stores the pipeline signal transmitted by the pipeline testing module, and constructs the multi-dimensional pipeline database by integrating the historical pipeline signal data set and the external pipeline operation data obtained through the network, and the preprocessing module receives the pipeline signal and performs data preprocessing to form a characteristic signal; A data analysis module, wherein the data analysis module constructs a pipeline wear simulation model based on a multi-dimensional pipeline database, wherein the characteristic signal is input into the pipeline wear simulation model for dynamic wear analysis and forms a simulation prediction signal and a control parameter signal, wherein the control parameter signal is transmitted to the pipeline testing module to adjust the slurry delivery parameters, and the pipeline testing module forms a slurry control signal; A feedback optimization module receives and compares the slurry control signal and the simulation prediction signal, extracts the difference characteristic value between the two, analyzes and generates a feedback optimization signal, and inputs it to the data analysis module and the database module.

2. The intelligent testing system for filling slurry conveying pipeline wear based on big data according to claim 1 is characterized in that: The pipeline testing module includes a testing pipeline unit, a slurry circulation unit and a data acquisition unit. The slurry circulation unit controls the slurry to be delivered to the testing pipeline unit. The data acquisition unit is arranged in the testing pipeline unit. The data acquisition unit collects various data when the slurry flows through the testing pipeline unit and forms a pipeline signal.

3. The intelligent testing system for filling slurry conveying pipeline wear based on big data according to claim 1 is characterized in that: When the database module integrates the historical pipeline signal data set and the external pipeline operation data obtained through the network, it includes a data fusion stage, a data cleaning stage and a data repair stage. In the data fusion stage, the multi-source heterogeneous pipeline signals are accessed and format aligned in real time through a distributed stream processing engine. The pipeline signals include but are not limited to pressure waveforms, vibration modal spectra and pipe wall thickness measurement data under different geological conditions, pipe diameter specifications and conveying media. In the data cleaning stage, an adaptive threshold filtering algorithm is used to eliminate pulse noise interference, and the trend item and random fluctuation component in the pipeline signal are separated based on the time series decomposition technology. In the data repair stage, a spatiotemporal correlation modeling method is used to construct a tensor filling model with the pipe section position as the spatial dimension and the operation cycle as the time dimension, and the alternating least squares method is used to optimize the objective function to complete data repair. The processed standardized data establishes a spatial index according to the pipe section topological structure, and a time series database is used to compress and store high-frequency sampling data to form the database module.

4. The intelligent testing system for wear of filling slurry conveying pipeline based on big data according to claim 1 is characterized in that: When the preprocessing module receives the pipeline signal and performs data preprocessing to form a characteristic signal, it first performs multi-scale feature extraction on the pipeline signal, uses wavelet packet transform to decompose the time domain signal of the pipeline signal into different frequency band subspaces, and calculates the energy entropy of each subband as the frequency domain feature quantity, and at the same time obtains the intrinsic mode function component of the signal through empirical mode decomposition, and extracts its instantaneous amplitude-frequency characteristics to form a time-frequency joint feature vector; after normalizing the feature vector, the principal component analysis algorithm is used to reduce the feature dimension, retain the feature components whose cumulative contribution rate exceeds the preset threshold, and form the feature signal, and the generated feature signal is pushed to the data analysis module in real time through the message queue.

5. The intelligent testing system for filling slurry conveying pipeline wear based on big data according to claim 1 is characterized in that: The construction of the pipeline wear simulation model adopts an integrated learning method to improve the prediction accuracy by stacking multiple base models. The pipeline wear simulation model includes a first-layer base model, a second-layer base model and a third-layer base model. The first-layer base model contains a fluid dynamics equation solver based on physical mechanisms to calculate the slurry flow field distribution and particle collision energy; the second-layer base model uses a gradient boosting decision tree to capture the nonlinear relationship in historical data; the third-layer base model is a three-dimensional convolutional neural network, which processes the microscopic wear morphology features in the pipeline cross-section detection image, and finally dynamically adjusts the weight coefficients of each base model through a meta-learning framework to form the pipeline wear simulation model.

6. The intelligent testing system for filling slurry conveying pipeline wear based on big data according to claim 1 is characterized in that: When performing dynamic wear analysis, a sliding time window mechanism is used to divide the continuous pipeline signal into analysis segments of fixed length according to the time-varying characteristics of the slurry conveying parameters. After the characteristic signal in each time window is input into the pipeline wear simulation model, the output contains the predicted value of the pipe wall thickness loss rate for the next three operating cycles, and a simulation prediction signal is formed.

7. The intelligent testing system for filling slurry conveying pipeline wear based on big data according to claim 1 is characterized in that: The slurry control signal is generated by adopting a multi-objective optimization algorithm. The defined objective function variables include but are not limited to three optimization dimensions: minimizing wear rate, maximizing conveying efficiency, and minimizing energy consumption cost. The optimal solution set is found through the Pareto frontier search algorithm, and the parameter combination with the highest comprehensive score is selected based on fuzzy decision theory. The fuzzy membership function dynamically adjusts the shape parameters according to the real-time working conditions and forms a slurry control signal.

8. The intelligent testing system for filling slurry conveying pipeline wear based on big data according to claim 1 is characterized in that: When the feedback optimization module receives and compares the slurry control signal with the simulation prediction signal, the dynamic mode decomposition technology is used to extract the dominant oscillation modes of the two, and the control effect is evaluated by calculating the mode amplitude correlation coefficient. When the correlation coefficient is lower than the preset threshold, the difference feature deep analysis process is started. When performing the difference feature deep analysis process, the phase space of the slurry control signal with differences is reconstructed, and the maximum Lyapunov exponent is calculated to determine the chaotic characteristics of the system, and the key modal components are extracted based on singular value decomposition to form difference feature values.

9. The intelligent testing system for filling slurry conveying pipeline wear based on big data according to claim 1 is characterized in that: A causal reasoning framework is introduced when performing the difference characteristic value analysis, and a Bayesian network with slurry parameters as dependent variables and wear characteristics as result variables is constructed. The conditional probability distribution is estimated by the Markov chain Monte Carlo method, and the control parameters whose contribution to the wear difference exceeds the preset value are identified. A feedback optimization signal containing a parameter adjustment priority list is generated, and the parameter weights of the pipeline wear simulation model are updated through an incremental learning mechanism.

10. A method for using the intelligent testing system for wear of filling slurry conveying pipeline based on big data according to any one of claims 1 to 9, characterized in that: include: S100: real-time collection of pipeline signals of slurry flow in the conveying pipeline; S2 00: receiving and storing the pipeline signal, building a multi-dimensional pipeline database by integrating historical pipeline signal data sets and external pipeline operation data obtained through networking, and performing data preprocessing on the pipeline signal to generate a characteristic signal; S300: constructing a pipeline wear simulation model based on the multi-dimensional pipeline database, inputting the characteristic signal into the simulation model for dynamic wear analysis, and generating a simulation prediction signal including predicted wear parameters and a slurry control signal representing optimal operating parameters; S400: transmitting the slurry control signal to an execution unit of a slurry delivery system to adjust slurry parameters; S5 00: obtaining the adjusted pipeline operation feedback signal in real time, performing feature comparison between the feedback signal and the simulation prediction signal, extracting the difference feature value between the two and generating an optimization instruction including a parameter correction strategy; S600: Synchronously feeding back the optimization instruction to the pipeline wear simulation model and the multi-dimensional pipeline database, and driving the pipeline wear simulation model to iteratively update parameters.

Images