Air cushion belt dynamic balance control system and control method

Abstract

The invention discloses an air cushion belt dynamic balance control system and a management and control method, and belongs to the field of air cushion belt conveyors. Original signals of a sensor are obtained, a Fluent noise filtering algorithm is used for eliminating airflow interference, synchronous tension and air pressure data are aligned through timestamps, a reliability model is used for filling temperature field missing values, and standardized data streams are formed; based on the standardized data flow, analyzing an air pressure gradient by means of a fluid mechanics equation, extracting abnormal vibration modal characteristics of the carrier roller, calculating a fluctuation coefficient in combination with tension simulation, and obtaining a multi-dimensional parameter set containing dynamic distribution of a pressure field and a belt deviation angle; inputting the multi-dimensional parameter set into the digital twin, updating the fatigue damage parameters of the belt through Fluent simulation driving, rendering the visualization result of the flow field of the air chamber in real time, and outputting a simulation result containing the dynamic cloud picture of the pressure field of the air chamber and the fatigue damage distribution of the belt; and calculating the pressure balance degree of the air chamber according to a simulation result.

Description

Technical Field

[0001] The present invention relates to the field of air-cushion belt conveyors, and more specifically, to an air-cushion belt dynamic balance control system and a management and control method thereof. Background Art

[0002] The working principle of an air-cushion conveyor is to form a stable air-cushion between the conveyor belt and the bearing surface, so that the conveyor belt runs in suspension. Compared with the traditional conveying equipment that directly supports the conveying by rollers or idlers, there is almost no direct contact between the conveyor belt and the bearing surface in the air-cushion conveyor, which greatly reduces the friction force, thereby reducing the energy consumption. At the same time, this suspension operation mode can also effectively reduce the breakage rate of materials during the conveying process, because the materials are no longer squeezed and rubbed by rollers or idlers.

[0003] With the development of industrial production towards stability, safety and intelligence, the requirements for equipment operation monitoring are constantly increasing. The monitoring methods of traditional air-cushion conveyors mainly rely on manual inspections and simple sensor monitoring. This kind of detection has certain limitations. For example, in manual inspections, although the appearance and partial operation status of the equipment can be observed intuitively, there are problems of low efficiency and strong subjectivity. It is difficult for the inspection personnel to conduct a comprehensive and detailed inspection of all parts of the equipment in a short time, and it is easily affected by personal experience and judgment ability, and some potential fault hazards may be missed. At the same time, manual inspections cannot obtain the operation data of the equipment in real time, and cannot detect the abnormal changes during the operation of the equipment in time.

[0004] Although simple sensor monitoring can collect some equipment operation parameters, the monitored parameters are often limited to a few indicators such as the running speed of the belt and the motor current. Although these parameters can reflect the operation status of the equipment to a certain extent, they are far from being able to comprehensively and real-time obtain the multi-dimensional status information of the air-cushion conveyor during operation. For example, key parameters such as the belt tension, vibration spectrum, air pressure gradient, temperature field distribution and displacement trajectory of the air-cushion conveyor are difficult to accurately and real-time collect and analyze by traditional monitoring means. Due to the lack of accurate grasp of the overall operation status of the equipment, potential fault hazards cannot be detected in time. Once a fault occurs, it often causes relatively serious consequences, easily leading to the expansion of the fault, affecting the normal operation and production efficiency of the equipment, and bringing huge economic losses to the enterprise. Summary of the Invention

[0005] Aiming at the problems existing in the prior art, the purpose of the present invention is to provide an air-cushion belt dynamic balance control system and a management and control method thereof. This management and control method realizes the comprehensive, real-time monitoring and precise regulation of the operation status of the air-cushion belt conveyor by integrating advanced sensor technology, fluid mechanics models, digital twin simulations and machine learning algorithms.

[0006] To solve the above problems, the present invention adopts the following technical solutions.

[0007] In a first aspect, a method for controlling dynamic balance of an air cushion belt comprises:

[0008] Step S1: Obtain the original sensor signal, use the Fluent noise filtering algorithm to eliminate airflow interference, synchronize tension and pressure data through timestamp alignment, and use the reliability model to fill in the missing values of the temperature field to form a standardized data stream;

[0009] Step S2: Based on the standardized data stream, the pressure gradient is analyzed by relying on the fluid mechanics equation to extract the abnormal vibration modal characteristics of the roller. The fluctuation coefficient is calculated by combining the tension simulation to obtain a multi-dimensional parameter set including the dynamic distribution of the pressure field and the belt deviation angle.

[0010] Step S3: Input the multi-dimensional parameter set into the digital twin, update the belt fatigue damage parameters through Fluent simulation drive, render the air chamber flow field visualization results in real time, and output simulation results including the dynamic cloud map of the air chamber pressure field and the belt fatigue damage distribution;

[0011] Step S4: Calculate the pressure balance of the air chamber based on the simulation results, derive the load mutation safety threshold through fault tree probability analysis, and construct an air pressure shortage warning criterion based on the air hole flow coefficient. Output a safety situation label containing the tension fluctuation limit value and the pressure imbalance level.

[0012] Step S5: respond in a graded manner based on the safety situation label: Level 1 warning initiates dual-arc air chamber pressure compensation; Level 2 warning generates a maintenance work order with air chamber deformation detection priority; Level 3 warning executes an emergency shutdown and triggers belt tear protection, outputting control action records and post-control data;

[0013] Step S6: Fusing historical pressure data with safety thresholds, predicting pressure anomaly trends based on fluid simulation training machine learning models, optimizing tension compensator gain parameters through bypass system effectiveness calculations, and outputting updated safety threshold library and optimized parameters;

[0014] In step S7, based on the regulated data and optimized parameters, the pressure balance is verified, the effectiveness of the safety threshold optimization is evaluated, a closed-loop regulation is formed through reinforcement learning iteration, and an initialization parameter set is output.

[0015] Furthermore, in step S11, the air chamber pressure fluctuation signal and the belt vibration time domain signal collected by the array sensor are acquired to form a raw data set;

[0016] Step S12: Based on the original data set, the turbulence noise of the air chamber pressure signal is removed using the Fluent simulation noise filtering algorithm, and the belt vibration signal is processed using an adaptive median filter to output preliminary noise reduction data;

[0017] Step S13, based on the Markov process correction method, use the change in belt tension as a node to synchronize the time reference of the preliminarily noise-reduced data;

[0018] Step S14, use the parallel system interpolation model to fill in the missing points of the synchronized noise-reduced data in combination with historical and redundant information;

[0019] Step S15, extract features from the filled data, correct outliers, and output a standardized data stream containing the time series corresponding to the air chamber pressure and the belt vibration frequency spectrum matrix.

[0020] Further, in step S21, construct multi-source data including the time series corresponding to the air chamber pressure and the belt vibration frequency spectrum matrix according to the standardized data stream. Based on the hydrodynamics equation, spatially discretize the air chamber pressure sequence, divide the air chamber into x equally spaced regions, calculate the pressure difference between adjacent regions, obtain the air pressure gradient distribution, and then use the finite element method combined with real-time data to correct the model and output the dynamic distribution data of the air chamber pressure field;

[0021] Step S22, take the output belt vibration frequency spectrum matrix as the input. Through vibration frequency spectrum envelope analysis, extract the fault characteristic frequencies of the idler bearings in the belt drive system through Hilbert transform and frequency spectrum analysis, compare and identify the abnormal vibration modes of the idlers, output the abnormal vibration information, and associate it with the air chamber pressure field data;

[0022] Step S23, based on the multi-source data, combine the data related to the belt operation and the tension simulation results, establish a model to calculate the tension fluctuation coefficient, use machine vision to monitor the edge of the belt, calculate the deviation angle increment, and output the mechanical and position deviation data;

[0023] Step S24, integrate the dynamic distribution data of the air chamber pressure field, the abnormal vibration modes of the idlers, the tension fluctuation coefficient, and the belt deviation angle increment, and construct and output a multi-dimensional parameter set.

[0024] Further, in step S31, according to the multi-dimensional parameter set, use the Fluent simulation data to drive the digital twin. Map the time series data corresponding to the air chamber pressure and the air pressure gradient distribution data in the multi-dimensional parameters to the three-dimensional model of the air chamber through the data mapping algorithm. Inside the three-dimensional model, calculate the eddy current region inside the air chamber based on the hydrodynamics equation, and through mesh division and iterative calculation, realize the dynamic presentation of the eddy current region in the digital twin, and output the preliminary simulation data of the internal flow field of the air chamber;

[0025] Step S32: Based on the preliminary simulation data of the internal flow field of the air chamber and the belt vibration frequency spectrum matrix and tension fluctuation coefficient in the multi-dimensional parameters, construct a Markov state transition matrix. According to the parameter changes of vibration and tension during the belt operation, set different state nodes and transition probabilities, dynamically update the belt fatigue damage parameters, and output the belt fatigue damage state data;

[0026] Step S33: Combine the preliminary simulation data of the internal flow field of the air chamber and the output belt fatigue damage state data, and use visualization rendering technology to perform isosurface extraction and streamline drawing on the internal flow field data of the air chamber to generate a dynamic cloud map of the air chamber pressure field. Perform color coding and texture mapping on the belt fatigue damage data to display the belt fatigue damage distribution, and finally output the digital twin simulation results including the dynamic cloud map of the air chamber pressure field and the belt fatigue damage distribution.

[0027] Furthermore, in step S41, based on the dynamic cloud map of the air chamber pressure field output by the digital twin, use the space grid division method to divide the air chamber into m×n regions, extract the pressure values of each region, calculate the ratio Kp of the pressure standard deviation to the average pressure to quantify the air chamber pressure balance degree, and divide it into three levels of high, medium, and low balance according to the Kp value, and output the air chamber pressure balance degree level result;

[0028] Step S42: Relying on the output air chamber pressure balance degree level, combined with the belt fatigue damage distribution data, use the fault tree analysis method to construct a fault tree model, use the Bayesian network to calculate the probability of the bottom event of the motor bearing failure, and then deduce the safety threshold LST of the load mutation rate, and output the safety threshold data;

[0029] Step S43: Synthesize the safety threshold of the load mutation rate and the air chamber pressure balance degree level, analyze the orifice flow data, calculate the orifice flow coefficient Cd according to Bernoulli's equation, establish a warning criterion for insufficient air pressure by optimizing the orifice parameters, and at the same time, compare the tension fluctuation coefficient with the preset threshold and mark the overlimit value, and finally output the safety situation label covering the overlimit value of the tension fluctuation coefficient and the air chamber pressure imbalance degree level.

[0030] Furthermore, in step S51, analyze the warning level of the safety situation label. When it is a first-level warning, start the double-arc air chamber pressure compensation algorithm, calculate the compensation pressure value according to the dynamic cloud map of the pressure field, adjust the intake valve to perform compensation, record the gas volume, and output the first-level warning regulation record;

[0031] Step S52: When it is a second-level warning, combined with the drive system fault diagnosis, if the motor is overloaded, adjust the motor speed, determine the priority of the air chamber deformation detection according to the pressure imbalance degree and the belt damage using a risk matrix, generate a maintenance work order, and output the second-level warning regulation record and the maintenance work order;

[0032] Step S53: When a level-3 warning is issued, execute the emergency shutdown protocol, analyze the fault propagation based on the Markov model, lock the time and location, trigger the belt tearing protection, cut off the motor power supply, start braking, record the shutdown and fault information, and output the level-3 warning control record and fault information;

[0033] Step S55: Integrate the control records of level-1 to level-3 warnings, summarize the information on the air chamber pressure adjustment amount and belt tension compensation value, and form the control action record and the data after control.

[0034] Further, in step S61, for the data related to the air chamber pressure, use the intuitionistic fuzzy integration operator to construct an intuitionistic fuzzy set from the historical air chamber pressure data and the safety threshold. The membership degree and non-membership degree are used to reflect whether the pressure meets the safety standard. Through the IFHA operator, the fuzzy sets at different time points are weighted and fused according to the time proximity and credibility, the pressure safety analysis module of the historical knowledge base is updated, and the optimized pressure safety data is output;

[0035] Step S62: Based on the optimized pressure safety data, combine the pressure curve data corresponding to the flow rates of different orifice shapes in the fluid simulation as the training set. Use the random forest algorithm with the flow rate, pressure, and orifice shape as the inputs and the abnormal air pressure state as the output. After cross-validation and parameter tuning, train the model to learn the mapping relationship and predict the abnormal trend. Store the model and rules in the knowledge base and output the abnormal air pressure prediction model;

[0036] Step S63: Combine the abnormal air pressure prediction model with the effectiveness result of the standby system, establish the correlation equation between the gain parameter and the air pressure and the standby system state, simulate the system operation under different parameter combinations, use the particle swarm algorithm to search for the optimal parameters with the stability and compensation efficiency as the goals, update the control parameters of the tension compensator, improve the equipment control strategy in the knowledge base, and output the optimized gain parameter and control strategy.

[0037] Further, in step S71, according to the data after control and the optimized parameters, align the dynamic cloud map of the air chamber pressure field after control with the cloud map before control in space and time, calculate the pressure balance degree improvement rate ΔK, set the effective threshold of ΔK. If it meets the standard, generate a pressure optimization report and output the pressure balance degree improvement evaluation result, which is used to measure the pressure control effect;

[0038] Step S72: Based on the fault tree analysis method, recalculate the system failure probability after control, compare the top event probabilities before and after control, calculate the failure probability reduction rate ΔPf of the optimized safety threshold, set the effective threshold of ΔPf. If it meets the standard, update the safety threshold database and output the safety threshold optimization evaluation result to evaluate the safety control effectiveness.

[0039] Furthermore, in step S721, the pressure balance improvement assessment results and the safety threshold optimization assessment results are used as state feedback, and the control strategy parameter adjustment is used as the action space. A Q-learning algorithm is used to set a reward function R with a weight coefficient. After multiple rounds of iterative learning, the control strategy parameters are optimized, a strategy optimization report is generated, and the control strategy parameters optimized by reinforcement learning are output.

[0040] In step S722, based on the optimized control strategy parameters, combined with the system physical characteristics and safety standards, the air chamber pressure PID control parameters and the belt tension compensator initial gain values are calculated. After verifying the parameter stability through simulation, an initialization parameter set document is formed, and the initialization parameter set including the PID control parameters and tension compensation gain is output.

[0041] In a second aspect, a control system for dynamic balance control of an air cushion belt is applied to the above-mentioned method for dynamic balance control of an air cushion belt, comprising:

[0042] The data acquisition module acquires the original sensor signal, uses the Fluent noise filtering algorithm to eliminate airflow interference, synchronizes tension and pressure data through timestamp alignment, and uses a reliability model to fill in missing values in the temperature field to form a standardized data stream;

[0043] The data processing module, based on standardized data streams and relying on fluid mechanics models to analyze air pressure gradients, extracts abnormal vibration modal characteristics of rollers, and calculates fluctuation coefficients in conjunction with tension simulation to obtain a multi-dimensional parameter set including the dynamic distribution of the pressure field and the belt deviation angle.

[0044] The digital modeling module inputs a multi-dimensional parameter set into the digital twin, updates the belt fatigue damage parameters through Fluent simulation, renders the air chamber flow field visualization results in real time, and outputs simulation results including a dynamic cloud map of the air chamber pressure field and the distribution of belt fatigue damage;

[0045] The safety assessment module calculates the pressure balance of the air chamber based on simulation results, derives the load mutation safety threshold through fault tree probability analysis, and constructs the air pressure shortage warning criterion based on the air hole flow coefficient. It outputs a safety situation label containing the tension fluctuation limit value and the pressure imbalance level;

[0046] The control execution module responds in a graded manner based on the safety situation label: Level 1 warning activates dual-arc air chamber pressure compensation; Level 2 warning generates a maintenance work order with air chamber deformation detection priority; Level 3 warning executes an emergency shutdown and triggers belt tear protection, outputting control action records and post-control data;

[0047] Optimization and update module, which integrates historical pressure data and safety thresholds, trains a machine learning model based on fluid simulation to predict the trend of abnormal air pressure, calculates and optimizes the gain parameters of the tension compensator through the effectiveness calculation of the parallel system, outputs the updated safety threshold library and optimized parameters, verifies the pressure balance degree according to the regulated data and optimized parameters, evaluates the effectiveness of safety threshold optimization, forms a regulation closed-loop through reinforcement learning iteration, and outputs the initial parameter set.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] (1) The dynamic balance control system and management method of this air-cushion belt realize comprehensive, real-time monitoring and precise regulation of the operating state of the air-cushion belt conveyor by integrating advanced sensor technology, fluid mechanics models, digital twin simulations, and machine learning algorithms. This method can automatically eliminate air flow interference, synchronize tension and air pressure data, and fill in the missing values of the temperature field to form a standardized data stream, providing a high-quality data basis for subsequent analysis. Based on the fluid mechanics model to analyze the air pressure gradient, extract the abnormal vibration mode characteristics of the idler, and combine the tension simulation to calculate the fluctuation coefficient to obtain a multi-dimensional parameter set, enabling the system to comprehensively and accurately grasp the operating state of the air-cushion belt conveyor. Through digital twin simulation, the visualization results of the air chamber flow field are rendered in real time, and the simulation results including the dynamic cloud map of the air chamber pressure field and the distribution of belt fatigue damage are output, providing intuitive and comprehensive equipment state information for operation and maintenance personnel, helping to detect and handle potential faults in a timely manner, and improving the operating reliability and safety of the equipment.

[0050] (2) The dynamic balance control system and management method of this air-cushion belt also have intelligent early warning and regulation functions, which can respond at different levels according to the safety situation label and execute different regulation measures. The first-level early warning starts the double-arc air chamber pressure compensation, the second-level early warning generates a maintenance work order with the priority of air chamber deformation detection, and the third-level early warning executes an emergency stop and triggers the belt tearing protection, effectively preventing the expansion of faults and ensuring production safety. In addition, this method integrates historical pressure data and safety thresholds, trains a machine learning model based on fluid simulation to predict the trend of abnormal air pressure, and optimizes the gain parameters of the tension compensator through the effectiveness calculation of the parallel system, realizing the continuous optimization of the regulation strategy. Through reinforcement learning iteration, a regulation closed-loop is formed, continuously verifying the pressure balance degree, evaluating the effectiveness of safety threshold optimization, and outputting the initial parameter set, providing a strong guarantee for the long-term stable operation of the system. These functions together improve the intelligent level of the air-cushion belt conveyor, reduce the operation and maintenance costs, and improve the production efficiency. Description of the Drawings

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0052] Figure 1 is a flowchart of a method for dynamic balance control of an air cushion belt of the present invention;

[0053] Figure 2 is a module diagram of a control system for dynamic balance control of an air cushion belt of the present invention. Detailed implementation manners

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention; obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0055] Please refer to Figures 1 to 2 , a method for dynamic balance control of an air cushion belt, including:

[0056] Step S1, obtain the original sensor signals, use the Fluent noise filtering algorithm to eliminate air flow interference, synchronize the tension and air pressure data through timestamp alignment, and use a reliability model to fill in the missing values of the temperature field to form a standardized data stream;

[0057] Step S2, based on the standardized data stream, rely on the hydrodynamic equation to analyze the air pressure gradient, extract the abnormal vibration mode characteristics of the idler rollers, combine the tension simulation to calculate the fluctuation coefficient, and obtain a multi-dimensional parameter set including the dynamic distribution of the pressure field and the belt deviation angle;

[0058] Step S3, input the multi-dimensional parameter set into the digital twin, drive the update of the belt fatigue damage parameters through Fluent simulation, render the visualization result of the air chamber flow field in real time, and output the simulation results including the dynamic cloud map of the air chamber pressure field and the distribution of belt fatigue damage;

[0059] Step S4, calculate the air chamber pressure balance degree according to the simulation results, deduce the safety threshold of load mutation through fault tree probability analysis, combine the orifice flow coefficient to construct an early warning criterion for insufficient air pressure, and output a safety situation label including the tension fluctuation overrun value and the pressure imbalance degree level;

[0060] Step S5, perform hierarchical response according to the security situation tags: for a first-level warning, initiate the double-arc chamber pressure compensation; for a second-level warning, generate a maintenance work order including the priority of the chamber deformation detection; for a third-level warning, execute an emergency shutdown and trigger the belt tearing protection, and output the regulation action record and the data after regulation;

[0061] Step S6, fuse the historical pressure data and the safety threshold, predict the abnormal trend of the air pressure based on the fluid simulation training machine learning model, calculate the effectiveness of the bypass system and optimize the gain parameters of the tension compensator, and output the updated safety threshold library and the optimized parameters;

[0062] Step S7, verify the pressure balance degree according to the data after regulation and the optimized parameters, evaluate the effectiveness of the safety threshold optimization, form a regulation closed-loop through reinforcement learning iteration, and output the initialization parameter set;

[0063] Among them, step S1 includes the following:

[0064] Step S11, obtain the air chamber pressure fluctuation signal and the belt vibration time-domain signal collected by the array sensor, and form the original data set;

[0065] Step S12, according to the original data set, use the Fluent simulation noise filtering algorithm to remove the turbulent noise of the air chamber pressure signal, and process the belt vibration signal with the adaptive median filtering, and output the preliminary noise-reduced data;

[0066] Step S13, based on the Markov process correction method, use the change of the belt tension as the node to synchronize the time reference of the preliminary noise-reduced data;

[0067] Step S14, use the bypass system interpolation model to fill the missing points of the synchronized noise-reduced data in combination with the historical and redundant information;

[0068] Step S15, extract features from the filled data and correct the outliers, and output the standardized data stream including the time series corresponding to the air chamber pressure and the belt vibration frequency spectrum matrix.

[0069] In this embodiment, the air chamber pressure fluctuation signal and the belt vibration time-domain signal are collected by the array sensor. These sensors are accurately deployed at the key positions of the air-cushion belt conveyor to ensure that the changes of relevant physical quantities can be accurately captured. The air chamber pressure fluctuation signal reflects the dynamic change of the gas pressure in the air-cushion system, while the belt vibration time-domain signal directly reflects the vibration state of the belt during operation. Integrate these collected signals to form the original data set, providing the basic data for subsequent analysis and processing.

[0070] The Fluent simulation noise filtering algorithm is used to process the air chamber pressure signal. The Fluent software is widely used in the field of fluid dynamics simulation and provides various methods for calculating aerodynamic noise. In this method, an acoustic analogy model is selected to identify and remove the turbulent noise in the air chamber pressure signal, and the noise components generated by gas turbulence in the pressure signal are analyzed and filtered. The principle is to distinguish the noise signal from the real air chamber pressure change signal by simulating the flow characteristics of the fluid, so as to effectively remove the turbulent noise and output a purer air chamber pressure signal. For the belt vibration time-domain signal, an adaptive median filtering method is adopted. When processing the belt vibration signal, the principle is to sort the data in a moving window, judge whether the data is a noise point by comparing the relationship between the center value and the median of the window. If it is a noise point, the median in the window is used to replace the data; if it is not a noise point, the original data is retained. As the window slides on the signal, the above process is continuously repeated, so as to effectively remove the noise while maximizing the retention of the detailed features of the belt vibration signal and output the preliminarily denoised belt vibration signal. Combining the preliminary denoising results of the air chamber pressure signal and the belt vibration signal, the preliminarily denoised data is obtained.

[0071] The time base synchronization of the preliminarily denoised data is carried out with the change of belt tension as the node. The Markov process has no aftereffect, that is, the state of the system at a future moment only depends on the current state and has nothing to do with the past historical state. In this application, the change of belt tension is regarded as the trigger point of state transition. When the belt tension changes, it corresponds to a change of state. By establishing a Markov chain model, the transition probability between different states is determined. Based on these state transition nodes, the air chamber pressure signal and the belt vibration signal after preliminary denoising are time-aligned to make them consistent in the time dimension, ensuring that the subsequent analysis can be based on synchronized data and avoiding analysis errors caused by time asynchronization.

[0072] Use the parallel connection system interpolation model to fill in the missing points of the synchronous noise reduction data by combining historical and redundant information. The parallel connection system interpolation model is constructed based on a comprehensive analysis of the system operation data. This model utilizes the system operation laws and redundant information contained in the historical data, such as the correlated data collected by sensors at different positions. When a missing point is detected in the synchronous noise reduction data, the model first analyzes the change trend of the data around the missing point according to the data characteristics under similar working conditions in the historical data, and then estimates and fills in the value of the missing point through appropriate interpolation algorithms, such as Lagrange interpolation method, spline interpolation method, etc., in combination with the supplementary constraint conditions provided by the redundant information. For example, if the air chamber pressure data at a certain moment is missing, the model will refer to the air chamber pressure values at adjacent moments and other relevant sensors at the same moment, such as the data of the flow sensor associated with the air chamber and other redundant information, and calculate the estimated value of the air chamber pressure at the missing point through interpolation to obtain the complete filled data. Feature extraction is performed on the filled data. For the air chamber pressure data, features such as pressure change trend, fluctuation amplitude, frequency, etc. are extracted; for the belt vibration data, features such as vibration frequency, amplitude, phase, etc. are extracted. During the feature extraction process, outliers in the data are simultaneously identified. Outliers may be caused by reasons such as sensor failures and sudden interferences. By setting a reasonable threshold range and using statistical analysis methods, such as the 3σ criterion, that is, data falling outside the range of the mean plus or minus 3 times the standard deviation is regarded as an outlier, to determine whether the data is abnormal. For outliers, methods such as data smoothing and regression analysis are used for correction to ensure the accuracy and reliability of the data.

[0073] After feature extraction and outlier correction, the time series corresponding to the air chamber pressure and the belt vibration frequency spectrum matrix are integrated to form a standardized data stream. The standardization process includes normalizing the data to make data with different dimensions at the same order of magnitude for convenient subsequent analysis and comparison. For example, data such as air chamber pressure values and belt vibration amplitudes are mapped into standard intervals such as [0,1] or [-1,1] through linear transformation and other methods. The finally output standardized data stream contains the comprehensively processed and analyzed information related to the air chamber pressure and belt vibration, providing high-quality data support for the further assessment and control of the dynamic balance state of the air cushion belt conveyor.

[0074] In a preferred embodiment of the present invention, step S2 includes the following:

[0075] Step S21, construct multi-source data including the time series corresponding to the air chamber pressure and the belt vibration frequency spectrum matrix according to the standardized data stream. Discretize the air chamber pressure sequence spatially based on the fluid mechanics equation, divide the air chamber into x equally spaced regions, calculate the pressure difference between adjacent regions, obtain the air pressure gradient distribution, and then use the finite element method to correct the model in combination with real-time data to output the dynamic distribution data of the air chamber pressure field;

[0076] Step S22: Taking the output belt vibration spectrum matrix as the input, through vibration spectrum envelope analysis, extracting the fault characteristic frequencies of the idler bearings in the belt drive system via Hilbert transform and spectrum analysis, comparing and identifying the abnormal vibration modes of the idlers, outputting the abnormal vibration information, and correlating with the air chamber pressure field data;

[0077] Step S23: Based on the multi-source data, combining the data related to the belt operation and the tension simulation results, establishing a model to calculate the tension fluctuation coefficient, using machine vision to monitor the edge of the belt, calculating the deviation angle increment, and outputting the mechanical and position deviation data;

[0078] Step S24: Integrating the dynamic distribution data of the air chamber pressure field, the abnormal vibration modes of the idlers, the tension fluctuation coefficient, and the belt deviation angle increment, constructing and outputting a multi-dimensional parameter set.

[0079] In this embodiment, based on the time series corresponding to the air chamber pressure in the standardized data stream, according to the fluid mechanics equation, the air chamber is spatially divided into x equally spaced regions. This is a process of discretizing the continuous air chamber pressure distribution. By calculating the pressure values in each region, the pressure difference between adjacent regions can be obtained. The pressure difference is the key factor driving the fluid flow, and its calculation follows basic principles such as Bernoulli's equation and continuity equation. For example, in the case of steady flow, the continuity equation is used to ensure the conservation of fluid mass, and Bernoulli's equation is combined to analyze the relationships between pressure, velocity, and height in different regions, so as to accurately obtain the air pressure gradient distribution and reflect the trend of internal pressure changes in the air chamber. The finite element method is used to construct the air chamber pressure model. When constructing the air chamber pressure model, the air chamber is divided into multiple units, and the solution is carried out according to the boundary conditions and initial conditions of fluid mechanics. Then, combined with the real-time collected air chamber pressure data, the model is corrected. When there is a deviation between the actually collected pressure data and the model prediction data, by adjusting the parameters of the model, such as boundary conditions, material properties, etc., the model can more accurately reflect the actual distribution of the air chamber pressure, and finally output the dynamic distribution data of the air chamber pressure field.

[0080] Perform Hilbert transform on the vibration signal. Hilbert transform can convert a real-valued signal into an analytic signal, thereby obtaining information such as the instantaneous amplitude and instantaneous phase of the signal. By performing spectrum analysis on the analytic signal, the periodic components in the signal can be highlighted, and then the fault characteristic frequencies of the idler bearings in the belt drive system can be extracted. Different types of idler bearing faults, such as rolling element damage, inner ring faults, outer ring faults, etc., will generate vibration signals with specific frequencies. These characteristic frequencies are closely related to the structural parameters of the bearing (such as the number of rolling elements, diameter, bearing pitch circle diameter, etc.). The corresponding fault characteristic frequency formulas can be obtained through calculation and used as the basis for judging the fault type. For example, for the rolling element fault characteristic frequency formula, let the rotational speed of the bearing be (unit: revolutions per minute, rpm), the number of rolling elements is Z, the diameter of the rolling element is d, the pitch diameter of the bearing is D, and the contact angle is , according to the bearing kinematics principle, the revolution speed of the rolling element and the bearing speed The relationship is: , the fault characteristic frequency of the rolling element is the number of times the rolling element passes through the fault point in one revolution multiplied by the revolution speed, that is: , where Z determines the number of times the rolling element passes through the fault point per unit time, and reflects the difference between the revolution speed of the rolling element and the bearing speed.

[0081] Compare the extracted characteristic frequency with the frequency range of the idler vibration in the normal state to identify the abnormal vibration mode of the idler. When an abnormal vibration mode is detected, associate it with the air chamber pressure field data, because changes in the air chamber pressure may affect the running state of the belt, which in turn causes abnormal idler vibration; conversely, abnormal idler vibration may also reflect the instability of the air chamber pressure. By correlating these two sets of data, the running condition of the air cushion belt conveyor can be analyzed more comprehensively, and abnormal vibration information can be output.

[0082] Based on multi-source data, combined with belt operation-related data (such as belt speed, conveyed material weight, etc.) and the tension simulation results, establish a model to calculate the tension fluctuation coefficient. During the belt operation, the tension is affected by various factors. By comprehensively analyzing these factors and using mechanical principles to establish a mathematical model. For example, consider factors such as the friction between the belt and the air chamber, the pressure of the material on the belt, and the elasticity of the belt itself. By analyzing the balance relationship of these forces, obtain the mathematical expression between the tension and each influencing factor, and then calculate the tension fluctuation coefficient to reflect the change degree of the belt tension.

[0083] The tension fluctuation coefficient is specifically calculated as follows: Select the tension of the belt under stable working conditions (such as uniform no-load operation) as the reference value, which can be obtained by measuring with a sensor. The tension fluctuation coefficient is defined as the relative change degree between the real-time tension and the reference tension , and the calculation formula is: In actual calculation, it is necessary to substitute the data such as the real-time collected belt speed, material weight, and air chamber pressure into the tension mathematical expression to obtain , and then compare it with to calculate , for example, when > 10%, it indicates that the belt tension fluctuation exceeds the normal range and there may be abnormal working conditions.

[0084] The edge of the belt is monitored using machine vision technology. The machine vision system collects image information of the belt edge through a camera, and then uses image processing algorithms, such as edge detection algorithms and contour extraction algorithms, to process the image, accurately identify the position of the belt edge. By comparing the positions of the belt edge at different times, the belt deviation angle increment is calculated. For example, a line detection algorithm based on the Hough transform is used to identify the linear features of the belt edge, and the belt deviation angle increment is determined by calculating the angle change of the line, and the position deviation data of the belt during operation is output.

[0085] Integrate the dynamic distribution data of the air chamber pressure field, the abnormal vibration mode of the idler, the tension fluctuation coefficient, and the belt deviation angle increment to construct a multi-dimensional parameter set. These parameters reflect the operating state of the air-cushion belt conveyor from different angles. The air chamber pressure field data reflects the working performance of the air-cushion system, the abnormal vibration information of the idler reflects the health status of the drive system, and the tension fluctuation coefficient and the belt deviation angle increment show the mechanical and position states of the belt. By integrating these parameters, it is possible to provide comprehensive and accurate data support for the operation monitoring, fault diagnosis, and optimization control of the air-cushion belt conveyor, and finally output a multi-dimensional parameter set.

[0086] In a preferred embodiment of the present invention, step S3 includes the following:

[0087] Step S31, according to the multi-dimensional parameter set, use Fluent simulation data to drive the digital twin. Map the time series data corresponding to the air chamber pressure and the air pressure gradient distribution data in the multi-dimensional parameters to the three-dimensional model of the air chamber in real time through a data mapping algorithm. In the three-dimensional model, calculate the eddy current region inside the air chamber based on the fluid mechanics equation. Through mesh generation and iterative calculation, realize the dynamic presentation of the eddy current region in the digital twin, and output the preliminary simulation data of the internal flow field of the air chamber.

[0088] Step S32, based on the output preliminary simulation data of the internal flow field of the air chamber and the belt vibration frequency spectrum matrix and tension fluctuation coefficient in the multi-dimensional parameters, construct a Markov state transition matrix. Set different state nodes and transition probabilities according to the parameter changes of vibration and tension during the belt operation, and dynamically update the belt fatigue damage parameters, and output the belt fatigue damage state data.

[0089] Step S33, combine the output preliminary simulation data of the internal flow field of the air chamber and the output belt fatigue damage state data, and use visualization rendering technology to perform isosurface extraction and streamline drawing on the internal flow field data of the air chamber to generate a dynamic cloud map of the air chamber pressure field, and perform color coding and texture mapping on the belt fatigue damage data to display the belt fatigue damage distribution. Finally, output the digital twin simulation results including the dynamic cloud map of the air chamber pressure field and the belt fatigue damage distribution.

[0090] In this embodiment, multi-dimensional parameters such as the pressure time series data of the air chamber, such as the pressure values at different times, the air pressure gradient distribution data, and the spatial pressure change rate, are collected as the input source of the Fluent simulation data. Through data mapping algorithms such as interpolation and coordinate transformation, the time series and spatial gradient data are mapped in real time to the spatial coordinates of the three-dimensional model of the air chamber to ensure a one-to-one correspondence between the data and the geometric position of the model.

[0091] The calculation principle of the internal flow field of the air chamber is based on the Navier-Stokes equation (N-S equation) and the continuity equation, which describe the flow state of the gas inside the air chamber:

[0092] The S equation: , which is used to calculate the fluid velocity field, pressure field, and eddy current region, where The gas density, whose unit is generally , which describes the mass per unit volume of the gas in the air chamber. For example, the density of air under standard conditions is about 1.29 , which affects the mechanical properties such as the inertia of the gas. The greater the density, the greater the inertial force and other forces will change when the same state of motion changes; The velocity vector The partial derivative of with respect to time, that is, the local acceleration, reflects the change rate of the fluid velocity over time at the same spatial point. For example, when the gas flow velocity at a certain point in the air chamber increases over time, this term is not zero, which reflects the characteristics of unsteady flow; is the velocity vector The dot product of and the velocity gradient , which is called the convective acceleration, reflects the change rate of velocity caused by the change in the position of the fluid particles, that is, the flow non-uniformity. When the flow velocities are different in different regions of the air chamber, that is, there is a velocity gradient, and the fluid flows from the region with a fast flow velocity to the region with a slow flow velocity, this term will come into play, reflecting the migration and convection characteristics of the flow; is the pressure gradient, which is a vector pointing in the direction of the fastest pressure increase, and its magnitude is the pressure change rate per unit distance. When the pressure distribution in the air chamber is uneven, the pressure gradient will drive the gas flow. For example, from the high-pressure region to the low-pressure region, the pressure gradient pushes the gas like a "force" and acts together with other force terms in the equation to determine the fluid motion; is the dynamic viscosity, which reflects the magnitude of the gas viscosity. Viscosity will generate internal frictional forces inside the gas and hinder the relative motion between fluid layers. For example, for gases with high viscosity, such as some viscous gas mixtures, the internal frictional forces between different flow velocity layers are more significant. The larger the μ value, the stronger the viscous effect;

[0093] is expressed as the velocity vector The Laplace operator, when expanded in a rectangular coordinate system, is to take the second-order partial derivatives of each component of the velocity and then sum them up. It reflects the influence of viscous diffusion on the velocity distribution and, together with the dynamic viscosity constitutes the viscous force term , which describes the effect of viscosity in making the velocity tend to be uniform in space. For example, in a gas chamber, regions with large differences in flow velocity originally will gradually become more uniform in flow velocity due to viscosity; is the external force vector per unit mass of the fluid, such as the manifestation of external forces like gravity and electromagnetic force on the fluid per unit mass. In the conventional flow analysis of a gas chamber, if special external forces, such as electromagnetic force, are not considered and mainly gravity is considered, it corresponds to the vector related to the gravitational acceleration. However, in some gas chamber flow field analyses, if the gas density is small and the influence of gravity is negligible compared to the flow driving force, it may also be approximately considered .

[0094] Continuity equation: , which ensures the conservation of fluid mass, where represents the partial derivative of density with respect to time, reflecting the rate of change of fluid density with time at the same spatial point. If the gas flow in the gas chamber is steady, that is, the flow state does not change with time, this term is zero; if the flow is unsteady, such as during the intake and exhaust processes of the gas chamber, the local density changes with time, and this term is not zero; represents density multiplied by the divergence of the velocity vector . Divergence is an operation of a vector field, reflecting the "divergence" or "convergence" characteristics of the vector field. For it can be understood as the mass flux vector, that is, the mass passing through a unit area per unit time. Its divergence represents the net outflow or inflow rate of the mass flux per unit volume. If the divergence is positive, it means the mass in this volume is flowing out and the density has a decreasing trend; if the divergence is negative, it means the mass is flowing in and the density has an increasing trend. The entire continuity equation is based on the law of conservation of mass, indicating that during the flow process, the sum of the rate of change of density per unit volume with time and the divergence of the mass flux is zero, that is, mass is neither created nor destroyed out of thin air.

[0095] Divide the three-dimensional model of the gas chamber into fine computational grids, such as structured or unstructured grids, apply the fluid mechanics equations to each grid cell for numerical solution, use an iterative algorithm, such as the SIMPLE algorithm, to gradually approach the convergent solution, update parameters such as pressure and velocity through multiple iterations until the calculation accuracy requirements are met, finally determine the position and intensity of the eddy current region, and output the preliminary simulation data of the flow field, such as velocity contour maps and pressure distributions.

[0096] Based on the belt vibration spectrum matrix, such as the amplitudes of different frequency components and the tension fluctuation coefficient, which reflect the degree of tension change, state nodes are set, such as "normal", "slight damage", and "severe damage". Each node corresponds to a specific parameter threshold range. Analyze the historical data of vibration parameters during belt operation, such as amplitude, frequency, and tension fluctuation. Statistically calculate the transition probabilities between different states. For example, when the tension fluctuation coefficient exceeds the threshold, the probability of transitioning from the "normal" state to the "slight damage" state can be obtained by fitting historical fault data. Using the Markov model, according to the changes in current vibration and tension parameters, the state transition probabilities are updated in real time, and then fatigue damage parameters of the belt are derived, such as the damage degree and remaining life. For example, when the high-frequency components in the vibration spectrum increase and the tension fluctuation intensifies, the model will increase the probability of damage state transition and output data of a more severe fatigue damage state.

[0097] Extract the isosurfaces from the preliminary simulation data of the internal flow field of the air chamber, such as pressure and velocity, such as extracting isobaric surfaces and drawing streamlines, tracking the movement trajectories of fluid particles, visually displaying the pressure distribution and flow direction. Through color mapping, such as warm colors representing high pressure and cold colors representing low pressure and transparency adjustment, generate dynamic cloud maps to reflect the spatio-temporal changes of the air chamber pressure field in real time. Encode the fatigue damage state data, such as the damage degree, through color coding, such as red representing high damage areas and texture mapping, such as crack textures, and overlay it on the surface of the belt three-dimensional model to visually present the damage distribution. Integrate the air chamber pressure field cloud map with the belt fatigue damage distribution to form the simulation results of the digital twin containing multi-physical field information, providing a visual basis for equipment state assessment and fault warning. Through the combination of fluid mechanics equations and the Markov model, cross-physical field correlation analysis of the air chamber pressure and belt damage is realized. Based on multi-dimensional parameters collected in real time, such as pressure, vibration, and tension, etc., drive the dynamic update of the digital twin to ensure that the simulation results are consistent with the actual equipment state. Through technologies such as cloud maps and color coding, convert complex physical processes into intuitive information to assist maintenance personnel in quickly identifying anomalies, such as uneven pressure and concentrated areas of fatigue damage.

[0098] In a preferred embodiment of the present invention, step S4 includes the following:

[0099] Step S41, based on the dynamic cloud map of the air chamber pressure field output by the digital twin, divide the air chamber into m×n regions using the spatial grid division method, extract the pressure values of each region, calculate the ratio Kp of the pressure standard deviation to the average pressure to quantify the air chamber pressure balance degree, and divide it into three levels of high, medium, and low balance according to the Kp value, and output the air chamber pressure balance degree level result;

[0100] Step S42: Based on the output air chamber pressure balance level, combined with the belt fatigue damage distribution data, use the fault tree analysis method to construct a fault tree model, calculate the probability of the bottom event of motor bearing failure using the Bayesian network, and then derive the safety threshold LST of the load mutation rate, and output the safety threshold data;

[0101] Step S43: Synthesize the safety threshold of the load mutation rate and the air chamber pressure balance level, analyze the orifice flow data, calculate the orifice flow coefficient Cd according to Bernoulli's equation, establish an air pressure deficiency warning criterion by optimizing the orifice parameters, and at the same time, compare the tension fluctuation coefficient with the preset threshold, mark the overlimit value, and finally output a safety situation label covering the overlimit value of the tension fluctuation coefficient and the air chamber pressure imbalance level.

[0102] In this embodiment, based on the dynamic cloud map of the air chamber pressure field generated by the digital twin, the physical space of the air chamber is divided into m×n regular rectangular regions by the spatial grid division method. Each grid region corresponds to a calculation unit in the digital twin model, and the pressure values at the center points of each region are extracted through the interpolation algorithm , where (i = 1~m, j = 1~n), to form a two-dimensional pressure matrix. This process is essentially to discretize the continuous pressure field into a finite number of numerical points for subsequent statistical analysis. First, calculate the average pressure value of all grid regions , , represents the row index of the grid, corresponding to the region number of the air chamber in a certain dimension, such as the length or width direction, with a value range of 1 to m; j represents the column index of the grid, corresponding to the region number of the air chamber in another dimension, such as the width or height direction, with a value range of 1 to n. Calculate the pressure standard deviation , which characterizes the degree of dispersion of the pressure values: , the balance index is the ratio of the standard deviation to the average pressure, and the calculation formula is: ; The smaller the value, the more uniform the pressure distribution; the larger the value, the more significant the pressure fluctuation. For example, divide the balance according to the following standard levels:

[0103] Highly balanced : The pressure field is evenly distributed, the air film support force is consistent, and the belt runs stably;

[0104] Moderately balanced 3: There are local pressure fluctuations, and the potential risk of air flow disorder needs to be concerned;

[0105] Lowly balanced 3: The pressure distribution is significantly uneven, which may cause belt deviation or abnormal vibration.

[0106] Taking the failure of the motor bearing as the top event, a fault tree is constructed that includes the following basic events:

[0107] The air chamber pressure is unbalanced, which is characterized by level;

[0108] The belt is fatigued and damaged, which is quantified by the strain sensor data;

[0109] The load changes suddenly, which is characterized by the tension fluctuation coefficient K;

[0110] The lubrication system is abnormal, and the oil sensor data.

[0111] Each event is connected by a logic gate to form a causal relationship network. For example, "the air chamber pressure is unbalanced" and "the belt is highly fatigued" jointly cause the intermediate event of "bearing additional load" through an "AND gate".

[0112] Based on the historical operation and maintenance data, the prior probability P(basic event) of each basic event is set. For example, P(the air chamber pressure is low and balanced) = 0.05.

[0113] Define the conditional probability relationship between the intermediate event and the basic event:

[0114] For example, P(bearing additional load | low air chamber balance, high belt fatigue) = 0.8. When the air chamber pressure balance level is known, such as low balance and the belt fatigue damage distribution data, the probability of the basic event is updated through the Bayes' formula: P(basic event | observed data) = , and finally the posterior probability of "motor bearing failure" is deduced.

[0115] The load mutation rate L is obtained by fitting historical data. The function relationship between the tension change rate per unit time and the bearing failure probability .

[0116] Set the acceptable bearing failure risk probability , such as 0.01, and through inverse solution, the corresponding load mutation rate LST is obtained, that is: . When the actual load mutation rate exceeds LST, it is determined as a high-risk working condition.

[0117] Calculation of the orifice flow coefficient Cd. For a single orifice, when the influence of gravity is ignored, the Bernoulli equation is:

[0118] where is the internal pressure of the air chamber, is the external atmospheric pressure, is the air flow velocity in the air chamber, approximately 0, ​is the pore outlet flow rate, is the local resistance coefficient.

[0119] Theoretical derivation of flow coefficient Cd, theoretical flow assumes that the gas is inviscid , no energy loss, the flow velocity is derived only through the Bernoulli equation;

[0120] Simplifying the Bernoulli equation: , the Bernoulli equation simplifies to: ,in is the theoretical outlet flow rate, the flow rate when there is no resistance;

[0121] Solve for the theoretical flow rate: ;

[0122] Theoretical flow calculation: flow is defined as "flow velocity × cross-sectional area", the pore cross-sectional area is , then the theoretical flow rate: ;

[0123] Actual flow Q1 and flow coefficient definition:

[0124] The actual flow rate needs to take into account the viscous resistance , actual outlet flow rate Less than theoretical flow rate ;

[0125] The actual flow rate Q1 is determined by the actual outlet flow rate calculate: ;

[0126] Flow coefficient definition: Flow coefficient It is the ratio of actual flow rate to theoretical flow rate, which is used to correct the difference between theoretical assumptions and actual flow: , Q1 and Substituting the definition into the equation and modifying the resistance term in combination with the Bernoulli equation, we can obtain ;

[0127] Solving the full Bernoulli equation (including the resistance term) for the actual flow rate : , sorted out to get: , substitute into middle, By comparing the measured flow rate with the theoretical calculation, the Cd value can be calibrated, usually 0.6~0.85. When the Cd value is lower than the threshold, such as 0.6, and the average pressure of the gas chamber is When the rated pressure is less than 80%, the low air pressure warning is triggered.

[0128] Compare the real-time tension fluctuation coefficient K with the preset threshold, such as 15%. When K>15%, it is marked as "tension over limit". The calculation results are mapped to "high / medium / low balance" labels, and a rule engine is used to integrate the multi-parameter results:

[0129] If "tension exceeds the limit" and "air chamber is in low balance", output the "red warning" label;

[0130] If "tension is close to the threshold" or "air chamber is in medium balance", output the "yellow warning" label;

[0131] In other cases, output the "normal" label.

[0132] In a preferred embodiment of the present invention, step S5 includes the following:

[0133] Step S51, analyze the security situation label to judge the warning level. In the case of a first-level warning, start the double-arc air chamber pressure compensation algorithm. Based on the dynamic cloud map of the pressure field, calculate the compensation pressure value, adjust the intake valve to perform compensation, record the gas volume, and output the first-level warning control record;

[0134] Step S52, in the case of a second-level warning, combine the drive system fault diagnosis. If the motor is overloaded, adjust the motor speed. Based on the pressure imbalance degree and belt damage, use the risk matrix to determine the priority of air chamber deformation detection, generate a maintenance work order, and output the second-level warning control record and the maintenance work order;

[0135] Step S53, in the case of a third-level warning, execute the emergency shutdown protocol. Analyze the fault propagation based on the Markov model, lock the time and location, trigger the belt tear protection, cut off the motor power supply, start braking, record the shutdown and fault information, and output the third-level warning control record and the fault information;

[0136] Step S55, integrate the first-level to third-level warning control records, summarize information such as the air chamber air pressure adjustment amount and the belt tension compensation value, and form a control action record and the data after control.

[0137] In this embodiment, the first-level warning: corresponding to mild anomalies, such as the air chamber pressure imbalance degree being "medium balance", and the tension fluctuation coefficient being close to the threshold but not exceeding it; for the double-arc air chamber pressure compensation algorithm based on the dynamic cloud map data of the pressure field, adopt the reverse engineering optimization idea. By comparing the measured pressure distribution with the ideal uniform distribution in the dynamic cloud map of the pressure field, use the pressure difference matrix to locate the pressure weak area, and based on Bernoulli's equation and the fluid continuity equation, establish a pressure compensation mathematical model: , where is the target pressure value, determined according to the air chamber design parameters; is the measured pressure value of the current area. The calculated compensation pressure value is converted into an intake valve opening command, and the valve is precisely adjusted through a PID closed-loop control system to achieve dynamic compensation of gas flow. The gas volume change, pressure adjustment time, and pressure field data before and after adjustment during the compensation process are recorded in real time to generate a first-level early warning control record for subsequent effect evaluation and fault tracing.

[0138] Secondary warning: Represents a medium risk. For example, the air chamber pressure imbalance reaches "low balance", or the tension fluctuation coefficient exceeds the limit and is accompanied by an increase in belt fatigue damage. When the motor overload is detected, the system starts a dynamic speed regulation strategy. Based on the motor load-speed characteristic curve and combined with the current overload degree, the motor speed is gradually reduced through the frequency converter. While maintaining the basic conveying capacity, it avoids continuous overload damage to the motor. The adjustment process follows the formula , where is the target speed, is the current speed, is the overload force, is the rated load force of the motor. Taking the pressure imbalance degree level as the abscissa (low / medium / high balance) and the belt damage degree as the ordinate (slight / medium / severe damage), the risk level area is divided. Each area corresponds to a different priority for air chamber deformation detection. For example, when the pressure is low balance and the belt is severely damaged, it is determined as a high-risk area, and the air chamber deformation detection is immediately arranged. According to the risk matrix evaluation results, a maintenance work order containing information such as inspection items, priorities, and estimated time-consuming is automatically generated to guide the maintenance personnel to work efficiently. At the same time, the secondary warning control measures, such as the speed adjustment amount and the risk assessment process, are recorded, and the control record and maintenance work order are output.

[0139] Tertiary warning: For serious faults, it is triggered when there are risks such as belt tearing, extremely high probability of motor bearing failure, or sudden drop in air chamber pressure that directly threaten the system safety. When the tertiary warning is triggered, the system uses a pre-trained Markov model. Based on the current fault state, such as belt tearing and severe air chamber leakage, it analyzes the fault propagation path. According to the state transition probability matrix constructed from historical fault data, it predicts the probability of the fault spreading to other components within the future time step. For example, if the belt tears, the model can calculate the probability of the tear spreading and causing the driving roller to jam. Combining sensor data, such as the position of sudden tension drop and the area of abnormal vibration, it locks the time point and physical location of the fault occurrence, providing accurate guidance for subsequent maintenance.

[0140] When a level-three warning is triggered, immediately execute the emergency shutdown protocol, cut off the power supply of the motor, activate the mechanical braking device, ensure that the belt stops running within the shortest distance, prevent the expansion of the fault, trigger the belt tearing protection device, such as the linkage of the tearing sensor, and fix the belt by mechanical clamping or pneumatic braking to avoid further spread of the tear. Record in detail information such as the shutdown time, the cause of the fault trigger, and the results of the Markov model analysis, generate a level-three warning regulation record and a fault information report, and provide a complete data chain for subsequent accident analysis.

[0141] Integrate the regulation records of level-one to level-three warnings, summarize the key information, including the air chamber pressure adjustment amount, the belt tension compensation value, the motor speed adjustment record, the valve opening and closing status, etc., record the air chamber pressure distribution, the stable value of the belt tension, the equipment operation parameters, etc. after regulation, and form a complete data closed-loop from warning trigger to regulation execution to support subsequent system optimization and fault prediction model iteration.

[0142] In a preferred embodiment of the present invention, step S6 includes the following:

[0143] Step S61, for the data related to the air chamber pressure, use the intuitionistic fuzzy integration operator to construct an intuitionistic fuzzy set from the historical air chamber pressure data and the safety threshold, and use the membership degree and non-membership degree to reflect whether the pressure meets the safety standard. Through the IFHA operator, fuse the fuzzy sets at different time points with weights according to the proximity in time and credibility, update the pressure safety analysis module in the historical knowledge base, and output the optimized pressure safety data;

[0144] Step S62, based on the optimized pressure safety data, combine the pressure curve data corresponding to the flow rates of different orifice shapes in the fluid simulation as the training set, use the random forest algorithm, take the flow rate, pressure, orifice shape, etc. as the input, and the abnormal air pressure state as the output. After cross-validation and parameter tuning, train the model to learn the mapping relationship to predict the abnormal trend, store the model and rules in the knowledge base, and output the abnormal air pressure prediction model;

[0145] Step S63, combine the abnormal air pressure prediction model with the effectiveness result of the parallel system, establish the correlation equation between the gain parameter and the air pressure and the state of the parallel system, simulate the operation of the system under different parameter combinations, use the particle swarm algorithm to search for the optimal parameters with the stability and compensation efficiency as the goals, update the control parameters of the tension compensator, improve the equipment regulation strategy in the knowledge base, and output the optimized gain parameter and regulation strategy.

[0146] In this embodiment, the intuitionistic fuzzy set describes the relationship between elements and the set through three dimensions. Compared with the traditional fuzzy set, it introduces the hesitation degree to enhance the ability to express uncertainty. In the processing of air chamber pressure data, the membership degree is used to measure the degree to which the pressure value , and an extended boundary is introduced and , which is used for smooth transition. The specific calculation formula is as follows: .

[0147] When the pressure value is within the safe range , the membership degree is 1; when the pressure value is between the extended boundary and the safety threshold, the membership degree changes linearly with the distance; when it exceeds the extended boundary, the membership degree is 0.

[0148] For the calculation of the non-membership function, the non-membership degree represents the degree to which the pressure value does not meet the safety standard, and it needs to satisfy . Its calculation formula is: .

[0149] The non-membership degree corresponds to the membership degree. It is 0 within the safe range, and as the pressure value deviates from the safe range, the non-membership degree gradually increases to 1.

[0150] For the calculation of the hesitation degree, the hesitation degree reflects the uncertainty in judging the pressure safety state. For example, when the pressure value is close to the threshold boundary, both the membership degree and the non-membership degree are in an intermediate state, resulting in a higher hesitation degree, which means there is a greater uncertainty in judging whether the pressure is safe.

[0151] According to the intuitionistic fuzzy hybrid average (IFHA) operator, by weighted aggregation of intuitionistic fuzzy sets at different time points, highlighting the role of recent data and high-confidence data, the weights are assigned.

[0152] For the time weight, an exponential decay function is used to assign higher weights to recent data. The formula is: ;

[0153] where t is the time interval, α is the decay coefficient. The larger α is, the faster the weight of recent data grows, and the more the timeliness of the data is emphasized. represents the total number of time points participating in the fusion. is expressed as the time point.

[0154] For the confidence weight, the confidence score s is calculated according to factors such as the historical error rate and calibration status of the sensor, and then the weight is obtained through normalization , is expressed as the confidence score at the th time point; for example, the data of sensors with a low historical error rate and a good calibration status has a higher confidence and a corresponding larger weight.

[0155] Fuse the formula. Let is an intuitionistic fuzzy set of time points, where is expressed as the membership degree at time points; is expressed as the non - membership degree at time points; is expressed as the hesitation degree at time points; The comprehensive weight , is a regulation parameter used to balance the influence of time weight and credibility weight. The fused intuitionistic fuzzy set has the following calculation formula: , , . Through this formula, the intuitionistic fuzzy sets at different time points and with different credibilities are weighted and fused to obtain optimized data that more accurately reflects the pressure safety state, which is used to update the pressure safety analysis module of the historical knowledge base.

[0156] When constructing the air pressure anomaly prediction model, first pre - process the data. Normalize continuous variables such as the actual flow rate Q1 of the air hole and the internal pressure P1 of the air chamber, and map the processed data to the interval [0, 1] to eliminate the influence of dimension and improve the model training efficiency. For the shape of the air hole, such as circular, square, and oval, perform one - hot encoding and convert it into a binary vector. For example, a circle is encoded as [1, 0, 0], a square is encoded as [0, 1, 0], and an oval is encoded as [0, 0, 1], enabling the model to process non - numerical data. Divide the air pressure state into categories such as "normal", "mild anomaly", and "severe anomaly", and label the training set labels through historical fault data. For example, according to factors such as the degree of pressure deviation from the safety threshold and the duration of the anomaly, assign corresponding labels to each piece of data.

[0157] For a random forest composed of multiple decision trees, multiple training subsets are generated through bootstrap sampling. Each subset trains a decision tree. During the decision tree node splitting process, the Gini Index or information gain ratio is used to select the optimal feature to divide the data set. Taking the Gini Index as an example, its calculation formula is , where is the number of categories; is the proportion of samples of class in the node;

[0158] The training set is divided into a subset, and each time the model is trained with subsets, and the remaining 1 subset is used for testing. Repeat times and take the average performance. This can more comprehensively evaluate the performance of the model on different data subsets and avoid overfitting. Preset the parameter ranges such as the number of decision trees and the maximum depth. Evaluate the accuracy of different combinations through cross-validation and select the optimal parameter combination. For example, set the range of the number of decision trees to [50, 200] and the range of the maximum depth to [3, 10]. Traverse all parameter combinations and select the parameters that optimize the model performance for the final model training. The trained model learns the mapping relationship between input features such as flow rate, pressure, and pore shape and the output of the abnormal air pressure state, which can be used to predict the trend of abnormal air pressure, and store the model and rules in the knowledge base.

[0159] The effectiveness of the parallel system represents the degree to which the parallel system can operate normally and perform its intended functions at a certain moment or time period. It is usually represented by a value between 0 and 1. The closer the value is to 1, the more reliable the parallel system is and the higher the probability of normal operation; the closer the value is to 0, it indicates that the parallel system has a high failure risk and poor normal operation ability. Regarding the calculation of effectiveness , where represents the mean time between failures of the device, reflecting the reliability of the system; r represents the average repair time of the device, reflecting the maintainability of the system; in practical applications, these data can be obtained through statistical analysis of the historical operation data of the system, such as counting the number of system failures, the time intervals between failures, and the repair durations in the past period of time, and then calculating and r, and finally obtaining the effectiveness result.

[0160] Establish a state-space equation that includes the gain parameter , the air pressure change , and the state S of the parallel system: , where is the system state vector, including elements such as the gain parameter, air pressure change, and parallel system state, is the input vector, such as the adjustment value of the gain parameter, is the output vector, such as indicators like pressure stability and compensation time, , , , is the coefficient matrix. Through this equation, the change of the system state over time and the influence of the input on the output are described, and the correlation between the gain parameter, the air pressure, and the state of the bypass system is established. Then, using the Fluent simulation software, different combinations of gain parameters and bypass system states are input, the system operation is simulated, and the stability and compensation efficiency indicators are recorded. For example, by setting different gain parameter values and the working states of the bypass system, data such as the standard deviation of pressure fluctuations and the pressure recovery time are obtained after running the simulation, providing a basis for subsequent optimization.

[0161] Let the particle swarm size be , and the position vector of each particle z represents a set of gain parameters, and the velocity vector controls the position update. Initially, the positions and velocities of the particles are randomly generated so that they are distributed within the feasible solution space. Using the stability index , that is, the standard deviation of pressure fluctuations, and the compensation efficiency index , that is, the pressure recovery time, to construct the fitness function: ;

[0162] The larger the value of this function, the better the system performance corresponding to the combination of gain parameters. Among them, and are weight coefficients used to adjust the relative importance of stability and compensation efficiency in the optimization goal. and can be determined by clarifying the priorities of "pressure stability" and "compensation efficiency". For example, when transporting precision electronic components, even a small pressure fluctuation can cause product scrapping, so can be set to to give priority to ensuring stability; when transporting bulk materials such as coal, occasional pressure fluctuations have little impact, but the shutdown recovery cost is high, so can be set to to give priority to quickly restoring the pressure; it can also be adjusted according to the scenario adaptation. For example, in winter, the airtightness of the air chamber decreases and the pressure is prone to fluctuations, so the value is temporarily increased to strengthen the stability optimization; if the equipment is aging recently and the pressure compensation response is slow, the value is increased to give priority to optimizing the recovery time.

[0163] Particle update rules:

[0164] Velocity update: ;

[0165] Position update: ;

[0166] Among them, is the inertia weight, which controls the tendency of the particle to maintain its original velocity; It is used to distinguish different particle individuals; It represents the multi-dimensional parameters of the particles; It is represented as the th particle, the current velocity of the It is represented as the th particle, the current position of the It is represented as the th particle, the self-historical optimal position of the It is represented as the global optimal position of the , are learning factors, which guide the particles to learn from their own historical optimal positions and the global optimal position . The larger is, the more "persistent" the particle is in the good position it found in the past, and the weaker the motivation to explore new areas. The larger is, the more the particle tends to follow the global optimal position, and it is easy to converge quickly, but it may also miss better solutions around. Therefore, the values of and are essentially to find a balance between "exploring new solutions" and "converging to known optimal solutions", so that the algorithm will neither search blindly nor fall into local optimality prematurely; , are random numbers of the th dimension parameter, which increases the randomness of the search. The fitness is calculated repeatedly, and the particle positions and velocities are updated until the particle swarm converges. The finally output optimal gain parameters and control strategies are used to update the control parameters of the tension compensator, improve the knowledge base equipment control strategies, and enhance the performance of the air chamber pressure control system.

[0167] In a preferred embodiment of the present invention, step S7 includes the following:

[0168] Step S71, according to the regulated data and optimized parameters, align the dynamic cloud map of the regulated air chamber pressure field with the pre-regulated cloud map in space and time, calculate the pressure balance degree improvement rate ΔK, set the effective threshold of ΔK, and if it meets the standard, generate a pressure optimization report and output the evaluation result of the pressure balance degree improvement, which is used to measure the pressure regulation effect;

[0169] Step S72, based on the fault tree analysis method, recalculates the failure probability of the system after regulation, compares the top event probabilities before and after regulation, calculates the failure probability reduction rate ΔPf for safety threshold optimization, sets an effective threshold for ΔPf, and updates the safety threshold database if it meets the standard, outputs the safety threshold optimization evaluation results, and evaluates the effectiveness of safety regulation. Step S72 includes the following:

[0170] Step S721: Using the pressure balance improvement assessment results and the safety threshold optimization assessment results as state feedback, and the control strategy parameter adjustment as the action space, the Q-learning algorithm is used to set a reward function R with a weight coefficient. After multiple rounds of iterative learning, the control strategy parameters are optimized, a strategy optimization report is generated, and the control strategy parameters optimized by reinforcement learning are output;

[0171] In step S722, based on the optimized control strategy parameters, combined with the system physical characteristics and safety standards, the air chamber pressure PID control parameters and the belt tension compensator initial gain values are calculated. After verifying the parameter stability through simulation, an initialization parameter set document is formed, and the initialization parameter set including the PID control parameters and tension compensation gain is output.

[0172] In this embodiment, the dynamic cloud map of the air chamber pressure field is a visualization result of the pressure distribution based on the time and space dimensions. There are differences in the time and space between the cloud maps before and after regulation. To achieve accurate comparison, the system adopts image registration technology and uses SIFT or SURF algorithms to extract key feature points in the cloud maps before and after regulation. These algorithms detect extreme points in the image and calculate their scale and direction information to generate feature descriptors with scale invariance and rotation invariance. For example, stable feature points are found in areas with drastic pressure changes as the basis for subsequent alignment. Based on the extracted feature points, the geometric transformation parameters such as translation, rotation and scaling between the cloud maps are calculated. Through affine transformation or perspective transformation, the cloud map after regulation is adjusted to the same spatial position and time scale as the cloud map before regulation, ensuring that the pressure data of the two cloud maps correspond one-to-one in spatial coordinates and time points, laying the foundation for subsequent quantitative analysis.

[0173] Regarding the calculation of pressure balance improvement rate, pressure balance Used to measure the uniformity of air chamber pressure distribution, The smaller the value, the more uniform the pressure distribution; the system calculates the pressure before and after adjustment. and after regulation Value, pressure balance improvement rate , which reflects the improvement of pressure distribution uniformity after regulation compared with that before regulation. For example =30%, which means that the pressure balance is improved by 30% after regulation. The effective threshold of If it is greater than the threshold, it is determined that the pressure regulation effect is significant. At this time, the system will integrate the pressure data before and after regulation, the calculation process, the comparison of the dynamic cloud map of the pressure field, etc., to generate a pressure optimization report. The report intuitively shows the improvement effect brought by the pressure regulation in the form of visual charts and data tables, providing a basis for subsequent decision-making.

[0174] Regarding the update of the fault tree model, according to the actual operating status after regulation, the fault tree model is updated. For example, if the pressure balance degree in the air chamber is improved after regulation, the occurrence probability of the basic event of "uneven air chamber pressure" is correspondingly reduced; if the belt tension is more stable, the probability parameters of the basic events related to "belt fatigue damage" are adjusted. In this way, the fault tree model better fits the actual situation of the system after regulation and accurately reflects the failure risks of various parts of the system. Using the updated fault tree model, through the fault tree analysis method, the probability of system failure (top event) is gradually calculated from the probabilities of basic events. The system calculates the probability of system failure before regulation and the probability of system failure after regulation and the probability of system failure after regulation Failure probability reduction rate This value reflects the degree of reduction in the system failure risk after regulation. The larger it is, the more obvious the effect of regulation on improving the system safety. Set in advance the effective threshold. When the calculated reaches or exceeds this threshold, it is determined that the safety regulation effect is significant. At this time, the system updates the safety threshold database, stores the relevant data such as the probabilities of basic events and top events after regulation, and outputs the safety threshold optimization evaluation results. The evaluation results include the analysis of the change in the system failure probability, suggestions for adjusting the safety threshold, etc., to help evaluate the actual effect of the safety regulation measures and provide a reference for formulating subsequent safety strategies.

[0175] Integrate the evaluation results of the improvement of the pressure balance degree , the evaluation results of the optimization of the safety threshold , as well as the current pressure distribution in the air chamber, the belt tension state, etc. into the state space of the system. These state parameters comprehensively describe the operating condition of the system at a certain moment, providing a decision-making basis for the intelligent agent; the action space is defined as the adjustment operation of the regulation strategy parameters, including changing the PID control parameters, the gain value of the belt tension compensator, etc. Each action corresponds to a specific set of parameter adjustment schemes. The intelligent agent changes the system operating state by selecting different actions. The reward function R comprehensively considers factors such as the improvement of the pressure balance degree, the optimization of the safety threshold, and the regulation cost. By setting weight coefficients , , , and performs weighted summation on different factors. The formula is as follows: ;

[0176] Among them, is the regulation cost, such as energy consumption, equipment loss, etc. By adjusting the weight coefficient, the agent is guided to balance cost - effectiveness while pursuing the regulation effect. For example, if is larger, the algorithm is more inclined to select regulation actions with lower costs. Regarding the values of , , , the following examples are given. For example, in the initial stage, that is, the debugging period, the goal is to quickly balance the pressure and verify the safety optimization effect, and the cost is temporarily secondary. The coefficients can be , , ; during the stable operation period of the equipment, the goal is to balance the optimization effect and the operation cost, and the coefficients can be: , , ; during the cost reduction and breakthrough period, the goal is to strictly control the cost and the optimization effect is not lower than the baseline, and the coefficients can be: , , .

[0177] Create a Q - table to store the expected cumulative rewards for performing different actions in each state. Initially, all Q - values are set to 0. The number of rows of the Q - table corresponds to the number of states in the state space, and the number of columns corresponds to the number of actions in the action space. In each round of iteration, the agent selects an action according to the current state. The ε - greedy strategy is adopted. With probability ε, an action is randomly selected for exploration to discover new strategies; with probability 1 - ε, the action with the largest current Q - value is selected for exploitation to execute the known optimal strategy. As the iteration progresses, the value of ε is gradually decreased, making the agent more dependent on the learned optimal strategy. After executing an action, the system enters a new state, and the reward value R is calculated according to the reward function. The Q - values in the Q - table are adjusted using the Q - learning update formula,

[0178] New Q - value = Old Q - value+Learning rate×[Immediate reward + Discount factor×Best future expectation of the new state - Old Q - value];

[0179] The discount factor determines the importance of future rewards. Through continuous iteration, the Q - values gradually converge to the expected cumulative rewards corresponding to the optimal strategy. When the Q - table is stable, that is, the Q - values no longer change significantly or reach the set number of iterations, it is considered that the algorithm has learned the optimal regulation strategy. The system generates a strategy optimization report based on the final Q - table and outputs the regulation strategy parameters optimized by reinforcement learning to guide the actual regulation operation.

[0180] According to the air chamber pressure regulation requirements and the physical characteristics of the system, combined with the optimized regulation strategy, the Ziegler-Nichols method is used to calculate the PID control parameters.

[0181] Based on the belt tension compensation requirements and the system safety standards, comprehensively considering factors such as the change of belt load and the influence of air chamber pressure on the belt, the initial gain value of the belt tension compensator is calculated through a mechanical model. For example, according to parameters such as the elastic modulus, cross-sectional area, and running speed of the belt, combined with the relationship between tension and pressure, a mechanical equation is established to solve the appropriate gain value to ensure that the compensator can effectively and stably adjust the belt tension.

[0182] Substitute the calculated parameters into the air chamber pressure simulation model and belt mechanics simulation model based on Fluent, simulate various actual operating conditions, such as different load conditions and air chamber pressure fluctuations. During the simulation process, real-time monitor indicators such as the air chamber pressure response curve and the change trend of belt tension, and evaluate the stability and control accuracy of the parameters. If problems such as slow system response, excessive pressure overshoot, or severe tension fluctuations are found, adjust the parameters and then re-run the simulation. Through repeated iteration and optimization, until the parameters can make the system operate stably under various conditions and meet the control accuracy requirements, finally form an initialization parameter set document containing the PID control parameters of the air chamber pressure and the belt tension compensation gain, providing an accurate parameter configuration basis for the actual system operation.

[0183] A control system for the dynamic balance control of an air cushion belt, which is applied to the above-mentioned method for the dynamic balance control of an air cushion belt, includes:

[0184] A data acquisition module, which acquires the original sensor signals, uses the Fluent noise filtering algorithm to eliminate air flow interference, synchronizes the tension and air pressure data through timestamp alignment, and fills in the missing values of the temperature field using a reliability model to form a standardized data stream;

[0185] A data processing module, based on the standardized data stream, relying on the fluid mechanics model to analyze the air pressure gradient, extract the abnormal vibration mode characteristics of the idler, and combine the tension simulation to calculate the fluctuation coefficient to obtain a multi-dimensional parameter set including the dynamic distribution of the pressure field and the belt deviation angle;

[0186] A digital modeling module, which inputs the multi-dimensional parameter set into the digital twin, updates the belt fatigue damage parameters through Fluent simulation drive, and renders the real-time visualization result of the air chamber flow field, and outputs the simulation results including the dynamic cloud map of the air chamber pressure field and the belt fatigue damage distribution;

[0187] A safety assessment module, which calculates the air chamber pressure balance degree according to the simulation results, deduces the safety threshold of load mutation through fault tree probability analysis, constructs an air pressure deficiency warning criterion in combination with the orifice flow coefficient, and outputs a safety situation label including the tension fluctuation overlimit value and the pressure imbalance degree level

[0188] The control execution module responds hierarchically according to the security situation label: for a first-level warning, it starts the double-arc gas chamber pressure compensation; for a second-level warning, it generates a maintenance work order including the priority of gas chamber deformation detection; for a third-level warning, it executes an emergency shutdown and triggers the belt tearing protection, and outputs the control action record and the data after control.

[0189] The optimization and update module integrates historical pressure data and safety thresholds, predicts the abnormal trend of air pressure based on a machine learning model trained by fluid simulation, optimizes the gain parameters of the tension compensator through the effectiveness calculation of the bypass system, outputs the updated safety threshold library and optimized parameters, verifies the pressure balance degree according to the data after control and the optimized parameters, evaluates the effectiveness of safety threshold optimization, forms a control closed-loop through reinforcement learning iteration, and outputs the initialization parameter set.

[0190] As described above, it is only a preferred specific embodiment of the present invention; however, the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution of the present invention and its improved concept, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A method for dynamically balancing and controlling an air-cushion belt, characterized in that, Including: Step S1: Obtain the original sensor signals, eliminate airflow interference using the Fluent noise filtering algorithm, synchronize the tension and air pressure data through timestamp alignment, fill in the missing values of the temperature field, and form a standardized data stream; Step S2: Based on the standardized data stream, rely on the hydrodynamic equation to analyze the air pressure gradient, extract the abnormal vibration mode characteristics of the idler, calculate the fluctuation coefficient by combining tension simulation, and obtain a multi-dimensional parameter set including the dynamic distribution of the pressure field and the belt deviation angle; Step S3: Input the multi-dimensional parameter set into the digital twin, update the belt fatigue damage parameters through Fluent simulation drive, render the visualization result of the air chamber flow field in real time, and output the simulation results including the dynamic cloud map of the air chamber pressure field and the belt fatigue damage distribution; Step S4: Calculate the air chamber pressure balance degree according to the simulation results, deduce the safety threshold of load mutation through fault tree probability analysis, combine the orifice flow coefficient to construct an air pressure deficiency warning criterion, and output a safety situation label including the tension fluctuation overrun value and the pressure imbalance degree level; Step S5: Respond according to the classification of the safety situation label: initiate the double-arc air chamber pressure compensation for the first-level warning; generate a maintenance work order including the priority of air chamber deformation detection for the second-level warning; For the third-level warning, execute an emergency stop and trigger the belt tearing protection, and output the regulation action record and the data after regulation; Step S6: Integrate the historical pressure data and the safety threshold, predict the air pressure anomaly trend based on the fluid simulation training machine learning model, calculate and optimize the gain parameters of the tension compensator through the availability calculation of the parallel system, and output the updated safety threshold library and the optimized parameters; Step S7: Verify the pressure balance degree according to the data after regulation and the optimized parameters, evaluate the effectiveness of the safety threshold optimization, form a regulation closed-loop through reinforcement learning iteration, and output the initialization parameter set.

2. The dynamic balance control method of an air cushion belt according to claim 1, wherein Obtain the original sensor signals, eliminate airflow interference using the Fluent noise filtering algorithm, synchronize the tension and air pressure data through timestamp alignment, fill in the missing values of the temperature field using a reliability model, and form a standardized data stream, including: Step S11: Obtain the air chamber pressure fluctuation signal and the belt vibration time-domain signal collected by the array sensor to form an original data set; Step S12: According to the original data set, eliminate the turbulent noise of the air chamber pressure signal using the Fluent simulation noise filtering algorithm, process the belt vibration signal with adaptive median filtering, and output the preliminary noise-reduced data; Step S13: Based on the Markov process correction method, use the change of belt tension as a node to synchronize the time base of the preliminary noise-reduced data; Step S14: Use the interpolation model of the parallel system to fill in the missing points of the synchronized and noise-reduced data by combining historical and redundant information; Step S15: Extract features and correct outliers from the filled data, and output a standardized data stream including the time series corresponding to the air chamber pressure and the belt vibration frequency spectrum matrix.

3. The dynamic balance control method of an air cushion belt according to claim 2, wherein Based on the standardized data stream, rely on the hydrodynamic equation to analyze the air pressure gradient, extract the abnormal vibration mode characteristics of the idler, calculate the fluctuation coefficient by combining tension simulation, and obtain a multi-dimensional parameter set including the dynamic distribution of the pressure field and the belt deviation angle, including: Step S21: Construct multi-source data including the time series corresponding to the air chamber pressure and the belt vibration frequency spectrum matrix based on the standardized data stream. Discretize the air chamber pressure sequence spatially based on the fluid mechanics equation, divide the air chamber into x equally spaced regions, calculate the pressure difference between adjacent regions to obtain the air pressure gradient distribution, and then use the finite element method combined with real-time data to correct the model and output the dynamic distribution data of the air chamber pressure field. Step S22: Take the output belt vibration frequency spectrum matrix as the input. Through vibration frequency spectrum envelope analysis, extract the fault characteristic frequencies of the idler bearings in the belt drive system via Hilbert transform and frequency spectrum analysis, compare and identify the abnormal vibration modes of the idlers, output the abnormal vibration information, and associate it with the air chamber pressure field data. Step S23: Based on the multi-source data, combine the data related to the belt operation and the tension simulation results, establish a model to calculate the tension fluctuation coefficient, use machine vision to monitor the edge of the belt, calculate the deviation angle increment, and output the mechanical and position deviation data. Step S24: Integrate the dynamic distribution data of the air chamber pressure field, the abnormal vibration modes of the idlers, the tension fluctuation coefficient, and the belt deviation angle increment, and construct and output a multi-dimensional parameter set.

4. A dynamic balance control method for an air cushion belt according to claim 3, characterized in that Input the multi-dimensional parameter set into the digital twin. Drive and update the belt fatigue damage parameters through Fluent simulation, and render the visualization result of the air chamber flow field in real time. Output the simulation results including the dynamic cloud map of the air chamber pressure field and the belt fatigue damage distribution, including: Step S31: According to the multi-dimensional parameter set, use the Fluent simulation data to drive the digital twin. Map the time series data corresponding to the air chamber pressure and the air pressure gradient distribution data in the multi-dimensional parameters to the three-dimensional model of the air chamber through the data mapping algorithm in real time. Inside the three-dimensional model, calculate the eddy current region inside the air chamber based on the fluid mechanics equation, and through mesh generation and iterative calculation, realize the dynamic presentation of the eddy current region in the digital twin, and output the preliminary simulation data of the internal flow field of the air chamber. Step S32: Based on the output preliminary simulation data of the internal flow field of the air chamber and the belt vibration frequency spectrum matrix and the tension fluctuation coefficient in the multi-dimensional parameters, construct a Markov state transition matrix. Set different state nodes and transition probabilities according to the parameter changes of vibration and tension during the belt operation, and dynamically update the belt fatigue damage parameters, and output the belt fatigue damage state data. Step S33: Combine the output preliminary simulation data of the internal flow field of the air chamber and the output belt fatigue damage state data, and use the visualization rendering technology to perform operations such as isosurface extraction and streamline drawing on the internal flow field data of the air chamber to generate a dynamic cloud map of the air chamber pressure field, perform color coding and texture mapping on the belt fatigue damage data, display the belt fatigue damage distribution, and finally output the digital twin simulation results including the dynamic cloud map of the air chamber pressure field and the belt fatigue damage distribution.

5. A dynamic balance control method for an air cushion belt according to claim 4, characterized in that, Calculate the air chamber pressure balance degree based on the simulation results, deduce the safety threshold for load mutation through fault tree probability analysis, and combine the orifice flow coefficient to construct an air pressure deficiency warning criterion. Output the safety situation label including the tension fluctuation overlimit value and the pressure imbalance degree level, including: Step S41: Based on the dynamic cloud map of the air chamber pressure field output by the digital twin, the air chamber is divided into m×n regions using a spatial grid division method. The pressure value of each region is extracted. The ratio of the pressure standard deviation to the average pressure, Kp, is calculated to quantify the air chamber pressure balance. The pressure balance is divided into three levels according to the Kp value: high, medium, and low balance. The air chamber pressure balance level result is output; Step S42: Based on the outputted air chamber pressure balance level and combined with the belt fatigue damage distribution data, a fault tree analysis method is used to construct a fault tree model, and the probability of motor bearing failure bottom event is calculated using a Bayesian network. Then, the load mutation rate safety threshold LST is derived and the safety threshold data is outputted. Step S43, comprehensively considers the load mutation rate safety threshold and the air chamber pressure balance level, analyzes the pore flow data, calculates the pore flow coefficient Cd based on the Bernoulli equation, establishes the air pressure insufficient warning criterion by optimizing the pore parameters, and at the same time, compares the tension fluctuation coefficient with the preset threshold, marks the exceeded value, and finally outputs a safety situation label covering the tension fluctuation coefficient exceeded value and the air chamber pressure imbalance level.

6. A dynamic balance control method for an air cushion belt according to claim 5, characterized in that, Graded response based on security situation labels: Level 1 warning activates dual-arc air chamber pressure compensation; Level 2 warning generates a maintenance work order with air chamber deformation detection priority; The third-level warning executes an emergency stop and triggers belt tear protection, outputting control action records and post-control data, including: Step S51: Analyze the security situation label to determine the warning level. When the warning level is level 1, activate the double arc air chamber pressure compensation algorithm, calculate the compensation pressure value based on the pressure field dynamic cloud map, adjust the intake valve to perform the compensation, record the gas volume, and output the level 1 warning control record. Step S52: In the case of a secondary warning, combined with the drive system fault diagnosis, the motor speed is adjusted if the motor is overloaded. Based on the pressure imbalance and belt damage, the risk matrix is used to determine the priority of the air chamber deformation detection, generate a maintenance work order, and output the secondary warning control record and maintenance work order; Step S53: When the third-level warning is reached, the emergency shutdown protocol is executed, the fault propagation is analyzed based on the Markov model, the time and location are locked, the belt tear protection is triggered, the motor power is cut off, the brake is activated, the shutdown and fault information is recorded, and the third-level warning control record and fault information are output; Step S55: Integrate the first to third level warning control records, summarize the air chamber pressure adjustment amount and belt tension compensation value information, and form the control action record and post-control data.

7. A dynamic balance control method for an air cushion belt according to claim 6, characterized in that, The system integrates historical pressure data with safety thresholds, trains a machine learning model based on fluid simulation, and predicts pressure anomaly trends. The system then optimizes the tension compensator gain parameters through effective calculations of the bypass system, outputting an updated safety threshold library and optimized parameters, including: Step S61: For the air chamber pressure-related data, the historical air chamber pressure data and safety thresholds are constructed into an intuitionistic fuzzy set using an intuitionistic fuzzy integration operator. The membership and non-membership degrees are used to reflect whether the pressure meets the safety standards. The IFHA operator is used to weight the fuzzy sets at different time points according to time proximity and credibility. The historical knowledge base pressure safety analysis module is updated to output the optimized pressure safety data. Step S62: Based on the optimized pressure safety data, the pressure curve data corresponding to the flow rate of different pore shapes in the fluid simulation is used as a training set. The random forest algorithm is used to input the flow rate, pressure, and pore shape, and output the abnormal pressure state. After cross-validation and parameter adjustment, the training model learns the mapping relationship to predict the abnormal trend. The model and rules are stored in the knowledge base, and the pressure anomaly prediction model is output. In step S63, the pressure anomaly prediction model and the bypass system effectiveness results are combined to establish the correlation equation between the gain parameter and the pressure and bypass system status. The system operation under different parameter combinations is simulated. With stability and compensation efficiency as the goals, the particle swarm algorithm is used to search for the optimal parameters, update the tension compensator control parameters, improve the knowledge base equipment control strategy, and output the optimized gain parameters and control strategy.

8. A dynamic balance control method for an air cushion belt according to claim 7, characterized in that Based on the regulated data and optimized parameters, the pressure balance is verified, the effectiveness of the safety threshold optimization is evaluated, and a closed-loop regulation is formed through reinforcement learning iterations. The initialization parameter set is output, including: Step S71: Based on the post-control data and optimization parameters, the dynamic cloud map of the air chamber pressure field after control is spatiotemporally aligned with the cloud map before control, and the pressure balance improvement rate ΔK is calculated. An effective threshold for ΔK is set, and if the threshold is met, a pressure optimization report is generated, outputting the pressure balance improvement evaluation result to measure the pressure control effect. In step S72, based on the fault tree analysis method, the failure probability of the system after regulation is recalculated, the top event probabilities before and after regulation are compared, and the failure probability reduction rate ΔPf of the safety threshold optimization is calculated. The effective threshold of ΔPf is set. If it meets the standard, the safety threshold database is updated, the safety threshold optimization evaluation results are output, and the effectiveness of safety regulation is evaluated.

9. A dynamic balance control method for an air cushion belt according to claim 8, characterized in that, include: Step S721: Using the pressure balance improvement assessment results and the safety threshold optimization assessment results as state feedback, and the control strategy parameter adjustment as the action space, the Q-learning algorithm is used to set a reward function R with a weight coefficient. After multiple rounds of iterative learning, the control strategy parameters are optimized, a strategy optimization report is generated, and the control strategy parameters optimized by reinforcement learning are output; In step S722, based on the optimized control strategy parameters, combined with the system physical characteristics and safety standards, the air chamber pressure PID control parameters and the belt tension compensator initial gain values are calculated. After verifying the parameter stability through simulation, an initialization parameter set document is formed, and the initialization parameter set including the PID control parameters and tension compensation gain is output.

10. A control system for dynamic balance control of an air cushion belt, which is applied to the method for dynamic balance control of an air cushion belt according to any one of claims 1-9, characterized in that, include: The data acquisition module acquires the original sensor signal, uses the Fluent noise filtering algorithm to eliminate airflow interference, synchronizes tension and pressure data through timestamp alignment, and uses a reliability model to fill in missing values in the temperature field to form a standardized data stream; The data processing module, based on standardized data streams and relying on fluid mechanics models to analyze air pressure gradients, extracts abnormal vibration modal characteristics of rollers, and calculates fluctuation coefficients in conjunction with tension simulation to obtain a multi-dimensional parameter set including the dynamic distribution of the pressure field and the belt deviation angle. The digital modeling module inputs a multi-dimensional parameter set into the digital twin, updates the belt fatigue damage parameters through Fluent simulation drive, renders the visualization result of the air chamber flow field in real time, and outputs the simulation results including the dynamic cloud map of the air chamber pressure field and the distribution of belt fatigue damage; The safety assessment module calculates the air chamber pressure balance degree based on the simulation results, derives the safety threshold for load mutation through fault tree probability analysis, constructs an early warning criterion for insufficient air pressure in combination with the orifice flow coefficient, and outputs a safety situation label including the overlimit value of tension fluctuation and the pressure imbalance degree level The control execution module responds according to the classification of the safety situation label: for a first-level warning, start the double-arc air chamber pressure compensation; for a second-level warning, generate a maintenance work order including the priority of air chamber deformation detection; For a third-level warning, execute an emergency shutdown and trigger belt tear protection, and output the regulation action record and the data after regulation; The optimization and update module integrates historical pressure data and safety thresholds, predicts the abnormal air pressure trend based on the fluid simulation training machine learning model, optimizes the gain parameters of the tension compensator through the effectiveness calculation of the bypass system, outputs the updated safety threshold library and optimized parameters, verifies the pressure balance degree according to the data after regulation and the optimized parameters, evaluates the effectiveness of safety threshold optimization, forms a regulation closed-loop through reinforcement learning iteration, and outputs the initialization parameter set.

Images