Scraper conveyor chain tension self-adaptive control method based on intelligent welding

Abstract

The application discloses a scraper conveyor chain tension self-adaptive control method based on intelligent welding and relates to the technical field of automatic control of industrial equipment, which comprises the following steps: carrying out thermal-mechanical coupling finite element simulation on the conveyor chain to obtain residual stress digital fingerprints; acquiring sensor array signals and obtaining equivalent stiffness, sag and catenary observed tension values through processing; carrying out recursive least square identification on the residual stress digital fingerprints to obtain residual stress release rates; calculating feedforward compensation tension and comparing the feedforward compensation tension with the observed tension values to obtain tension tracking errors; obtaining PID adjustment coefficients through index interpolation and calculating adaptive feedback control amounts; and generating the timing of the advance action of the tensioning mechanism based on a time series prediction proxy model and combining the feedforward compensation tension and the adaptive feedback control amounts to generate control instructions. The application incorporates the welding residual stress into the tension control model, realizes the adaptive adjustment of the control parameters and the advance action of the tensioning mechanism.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology for industrial equipment, and in particular to an adaptive control method for chain tension of scraper conveyors based on intelligent welding. Background Technology

[0002] Scraper conveyors are crucial conveying equipment in underground coal mines, and their chain tension directly affects operational safety and conveying efficiency. When the chain is too loose, the connection between the sprocket and chain is damaged, causing vibration or even chain derailment; when the chain is too tight, operating friction increases, conveying energy consumption increases, and the risk of equipment damage rises. Currently, the main chain tension monitoring technologies used domestically and internationally include those based on the relationship between tension and power or cylinder pressure, those based on chain sag, those based on micro-strain, those based on sliding mode control, and those based on current methods. However, due to the complex and harsh working environment, existing chain tension monitoring technologies suffer from problems such as large errors between monitored and actual values ​​and relatively low control accuracy. Chain tension control systems have complex characteristics such as time-varying behavior, making it difficult to accurately establish a model of the controlled object. Conventional PID control relies on an accurate model of the system, which cannot meet the increasingly stringent accuracy requirements under actual working conditions.

[0003] During the manufacturing process, chains undergo welding, which generates residual stress within the chain due to welding heat. This residual stress, combined with the working stress, promotes the occurrence and propagation of brittle fracture. Existing tension control methods do not consider welding residual stress as a factor affecting chain stiffness and tension characteristics. Current methods lack real-time estimation and tracking of the chain's equivalent stiffness, failing to detect the slow degradation of chain mechanical properties caused by residual stress release. Control strategies are mostly based on simple feedback regulation, lacking a feedforward compensation mechanism based on stiffness changes, resulting in a response lag behind real-time changes in the chain's state. The tensioning mechanism's action commands are typically generated based on the current tension deviation, lacking the ability to predict tension change trends and thus failing to achieve preemptive action. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an adaptive control method for scraper conveyor chain tension based on intelligent welding. It aims to solve the problems of insufficient control accuracy and response lag caused by neglecting the influence of welding residual stress, fixed control parameters, and lack of state prediction mechanism in scraper conveyor chain tension control.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an adaptive control method for chain tension of a scraper conveyor based on intelligent welding, comprising the following steps: S1. Perform thermo-mechanical coupled finite element simulation on the conveyor chain, extract the residual stress distribution field, and perform model order reduction and compression on the residual stress distribution field to obtain the residual stress digital fingerprint; S2. Obtain the original analog signal from the sensor array, perform analog-to-digital conversion and digital filtering on the original analog signal to obtain the digital signal, perform channel analysis and calculation on the digital signal to obtain the equivalent stiffness, sag and catenary observation tension value; S3. Perform low-pass filtering on the equivalent stiffness through the edge controller to obtain the filtered stiffness value; perform recursive least squares identification on the filtered stiffness value and the digital fingerprint of residual stress to obtain the residual stress release rate. S4. When the hardware timer of the edge controller reaches the preset servo cycle, the feedforward compensation tension is calculated based on the filter stiffness value. The deviation between the feedforward compensation tension and the observed tension value of the catenary is calculated to obtain the tension tracking error. S5. Using the residual stress release rate and tension tracking error as indexes, interpolate the pre-stored fuzzy inference digital lookup table to obtain the PID adjustment coefficient, and calculate the adaptive feedback control quantity based on the PID adjustment coefficient through the feedback control law. S6. The edge controller receives the pre-built timing prediction proxy model, and generates the tensioning mechanism advance action timing based on the timing prediction proxy model and the filter stiffness value, sag, residual stress release rate and tension tracking error. Based on the tensioning mechanism advance action timing and the feedforward compensation tension and adaptive feedback control quantity, the tensioning mechanism control command is generated.

[0006] Furthermore, the process of performing thermo-mechanical coupled finite element simulation on the conveyor chain to extract the residual stress distribution field, and then performing model order reduction and compression on the residual stress distribution field to obtain a digital fingerprint of the residual stress, includes: We collect welding current, welding voltage, welding speed, and geometric model of the conveyor chain. The double ellipsoidal heat source model is used as the welding input load. Zero displacement constraint boundary conditions are applied to the non-welding area in the geometric model. Thermo-mechanical coupling simulation is performed to obtain the residual stress distribution field. The von Mises equivalent stress at each node in the residual stress distribution field is extracted as the residual stress value; The residual stress values ​​of all spatial nodes corresponding to each simulation time step constitute a column vector. The column vectors are arranged sequentially according to the time step order of the thermo-mechanical coupling simulation to construct a residual stress snapshot matrix. Singular value decomposition is performed on the residual stress snapshot matrix to obtain the singular value decomposition results; Extract the left singular vector matrix and singular value vector from the singular value decomposition results; The singular values ​​in the singular value vector are arranged in descending order of their numerical values, and the sequence number is the order corresponding to each singular value. The cumulative energy percentage is obtained by calculating the sum of squares of each singular value in the singular value vector and the cumulative percentage of the sum of squares. The smallest order at which the cumulative energy percentage first reaches or exceeds the preset energy threshold is determined as the cutoff order. The reduced-order basis matrix is ​​obtained by truncating the left singular vectors of the first truncation order from the left singular vector matrix. The low-dimensional projection coefficient matrix is ​​obtained by multiplying the transpose of the reduced basis matrix with the residual stress snapshot matrix. Extract the low-dimensional projection coefficient column vector corresponding to the cooling end time step of the thermo-mechanical coupling simulation and use it as the equivalent residual stress mode coordinate vector; The equivalent residual stress modal coordinate vector and the reduced-order basis matrix are encapsulated to obtain the residual stress digital fingerprint.

[0007] Furthermore, the process of acquiring the original analog signal from the sensor array, performing analog-to-digital conversion and digital filtering on the original analog signal to obtain a digital signal, and performing channel analysis and calculation on the digital signal to obtain the equivalent stiffness, sag, and catenary observation tension values ​​includes: Raw analog voltage signals are acquired using a sensor array; The original analog voltage signal is pre-filtered by an analog anti-aliasing filter and then converted from analog to digital to obtain the original digital sequence. The original digital sequence is low-pass filtered by a fixed-point infinite impulse response filter to obtain a digital signal; The digital signal is analyzed by channel analysis to obtain the channel analysis results, which include the tension value at the first end of the chain, the tension value at the second end of the chain, and the mid-span displacement value of the chain. The equivalent stiffness is obtained by least squares regression fitting of the tension values ​​at the first and second ends of the chain and the mid-span displacement of the chain. The chain sag is obtained by estimating the catenary sag based on the mid-span displacement value of the chain. Based on the tension values ​​at the first and second ends of the chain, the horizontal tension of the catenary is calculated using the static equilibrium equation of the catenary, thus obtaining the observed tension value of the catenary.

[0008] Furthermore, the equivalent stiffness is low-pass filtered using an edge controller to obtain a filtered stiffness value; the filtered stiffness value and the residual stress digital fingerprint are then recursively least squared to identify the residual stress release rate, including: The equivalent stiffness is low-pass filtered to obtain the filtered stiffness value. The filtered stiffness values ​​are arranged in chronological order to form a sequence of filtered stiffness values. The reduced-order basis matrix in the residual stress digital fingerprint is transposed to obtain the transpose of the reduced-order basis matrix. The reduced basis matrix is ​​transposed and multiplied with the preset strain-stiffness projection coefficient vector to obtain the projection observation column vector; Transpose the projected observation column vector to obtain the equivalent observation row vector; Using the filtered stiffness value sequence as the observation for least squares identification, the equivalent residual stress modal coordinate vector in the residual stress digital fingerprint as the initial state vector for recursive least squares identification, and the equivalent observation row vector as the observation row vector for recursive least squares identification, recursive least squares identification is performed to obtain the residual stress estimation sequence; the strain-stiffness projection coefficient vector is a coefficient vector obtained in advance through calibration experiments, used to map the spatial node stress values ​​to the overall stiffness change of the chain; A moving average filter is applied to the residual stress estimation sequence to obtain a smoothed residual stress estimation sequence. A first-order backward difference operation is performed on the smoothed residual stress estimation sequence, and the difference result is divided by the preset sampling period to obtain the residual stress release rate.

[0009] Furthermore, when the hardware timer of the edge controller reaches the preset servo cycle, the feedforward compensation tension is calculated based on the filter stiffness value, and the deviation between the feedforward compensation tension and the observed tension value of the catenary is calculated to obtain the tension tracking error, including: When the hardware timer of the edge controller reaches the preset servo cycle, it obtains the current filtered stiffness value obtained by performing low-pass filtering on the equivalent stiffness; Based on the current filter stiffness value, a feedforward compensation calculation is performed on the preset reference tension to obtain the feedforward compensation tension. The tension tracking error is obtained by calculating the difference between the feedforward compensation tension and the observed tension value of the catenary.

[0010] Furthermore, the step of performing feedforward compensation calculation on the preset reference tension based on the current filter stiffness value to obtain the feedforward compensation tension includes: The difference between the current filter stiffness value and the preset reference filter stiffness value is calculated to obtain the filter stiffness value deviation; the preset reference filter stiffness value is determined by taking the arithmetic mean of the initial filter stiffness values ​​obtained by performing low-pass filtering on the equivalent stiffness under the calibration conditions before the chain is put into operation. The stiffness compensation amount is obtained by multiplying the filter stiffness value deviation with the preset feedforward compensation gain. The stiffness compensation amount is added to the preset reference tension to obtain the feedforward compensation tension. The preset reference tension is the tension reference value of the scraper conveyor chain determined offline based on the design tension requirements before it is put into operation.

[0011] Furthermore, the step of interpolating the PID adjustment coefficients from a pre-stored fuzzy inference digital lookup table using the residual stress release rate and tension tracking error as indexes, and calculating the adaptive feedback control quantity based on the PID adjustment coefficients through a feedback control law, includes: Using the residual stress release rate and tension tracking error as indexes, two-dimensional linear interpolation is performed in a pre-stored fuzzy inference digital lookup table to obtain the PID adjustment coefficients, which include the proportional gain adjustment coefficient, integral time adjustment coefficient, and derivative gain adjustment coefficient. The proportional gain, integral time, and derivative gain are read from the edge controller. The proportional gain, integral time, and derivative gain are obtained offline by digital proportional-integral-derivative feedback control law before the chain is put into operation. The adaptive proportional gain is obtained by multiplying the proportional gain adjustment coefficient with the proportional gain. The adaptive integral time is obtained by multiplying the integral time adjustment coefficient by the integral time. The adaptive differential gain is obtained by multiplying the differential gain adjustment coefficient with the differential gain. Based on adaptive proportional gain, adaptive integral time, and adaptive derivative gain, a digital proportional-integral-derivative feedback control law is used to calculate the tension tracking error, resulting in an adaptive feedback control quantity.

[0012] Furthermore, the edge controller receives a pre-built timing prediction proxy model, and based on the timing prediction proxy model and the filtered stiffness value, sag, residual stress release rate, and tension tracking error, generates a timing sequence for the early action of the tensioning mechanism. Based on the timing sequence for the early action of the tensioning mechanism and the feedforward compensation tension and adaptive feedback control quantity, it generates control commands for the tensioning mechanism, including: Receive the time-series prediction agent model sent from the cloud through the industrial Ethernet gateway; The filter stiffness value, chain sag, residual stress release rate and tension tracking error within the continuous sampling period are written into the ring thermal data buffer of the edge controller in the order of acquisition time to obtain the edge controller cache data. Using the filtered stiffness value in the edge controller cache data as the measurement input, and the sag, residual stress release rate and tension tracking error in the edge controller cache data as auxiliary inputs, a linear Kalman filter state estimation is performed through a time-series prediction surrogate model to obtain a state estimation vector; the state estimation vector includes a trend intercept state component, a trend slope state component and an autoregressive state component. The trend intercept state component and the trend slope state component are respectively used as the current trend intercept and the current trend slope. Construct the current trend line based on the current trend intercept and the current trend slope; Perform linear extrapolation on the current trend line according to the preset extrapolation step size to obtain the advance action sequence of the tensioning mechanism, which includes the predicted filtered stiffness values ​​at each extrapolation time point in the future. The sum of the feedforward compensation tension and the adaptive feedback control quantity is used as the basic control amplitude at the current moment. Based on the predicted filter stiffness value at each extrapolated time point in the advance action sequence of the tensioning mechanism, the tensioning mechanism is triggered when the predicted filter stiffness value reaches the preset action threshold, and a tensioning mechanism control command with a timestamp is generated.

[0013] Furthermore, the pre-built time-series prediction proxy model is generated according to the following steps: Acquire historical operating data of the scraper conveyor chain. The historical operating data includes historical filter stiffness value time series, historical sag time series, historical residual stress release rate time series, and historical tension tracking error time series. A significance test of the trend term was performed on the time series of historical filter stiffness values, and the results of the significance test of the trend term were obtained. When the trend term significance test result is significant, the trend intercept state component and the trend slope state component are included in the state vector; when the trend term significance test result is not significant, the initial estimates of the trend intercept state component and the trend slope state component are set to zero. Autocorrelation and partial autocorrelation functions were performed on the time series of historical filter stiffness values ​​to obtain the autoregression order. The autoregression order is used as the dimension of the autoregressive state components in the state vector; A state-space model is constructed using historical sag time series, historical residual stress release rate time series and historical tension tracking error time series as auxiliary input variables, historical filtered stiffness value time series as measurement output variable, and state vector containing trend intercept state component, trend slope state component and autoregressive state component as state variable. Joint identification of the state-space model yields the state transition matrix, measurement matrix, and trend transition coefficients; The state transition matrix, measurement matrix, and trend transition coefficients are encapsulated into a time series prediction proxy model.

[0014] The adaptive control method for chain tension of scraper conveyors based on intelligent welding provided by this invention has the following main advantages: 1. This invention extracts the residual stress distribution field from welding through thermo-mechanical coupled finite element simulation and obtains a digital fingerprint of residual stress using model order reduction compression, thus incorporating the influence of the welding process on the mechanical properties of the chain into the entire tension control process. Compared with existing tension control methods that ignore residual stress factors, this invention can sense the continuous impact of residual stress release on chain stiffness, providing more complete state information of the controlled object for tension adjustment.

[0015] 2. This invention performs low-pass filtering on the equivalent stiffness using an edge controller and then performs recursive least-squares identification with the residual stress digital fingerprint to estimate the residual stress release rate in real time. Using the residual stress release rate and tension tracking error as indexes, the PID adjustment coefficients are obtained by interpolating the fuzzy inference digital lookup table, achieving adaptive adjustment of control parameters. Compared to conventional PID control that relies on fixed parameters, the control parameters of this invention can be dynamically adjusted according to changes in the chain state. Simultaneously, this invention uses a time-series prediction surrogate model to estimate the state and extrapolate the trend of the filtered stiffness value, generating an advance action sequence for the tensioning mechanism, enabling the tensioning mechanism to perform adjustment actions before the tension deviation exceeds a threshold. Compared to existing methods that generate control commands based solely on the current tension deviation, the control commands of this invention include predictions of future chain state changes, improving the timeliness of tension adjustment. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the adaptive control method for chain tension of a scraper conveyor based on intelligent welding, as described in this invention. Figure 2 This is a flowchart illustrating the residual stress release rate identification process of the present invention. Figure 3 This is a flowchart illustrating the generation of control commands for the tensioning mechanism of the present invention. Figure 4 This is a flowchart illustrating the construction process of the time-series prediction proxy model of this invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figures 1-4 This invention provides an adaptive control method for chain tension of a scraper conveyor based on intelligent welding, comprising the following steps: S1. Perform thermo-mechanical coupled finite element simulation on the conveyor chain, extract the residual stress distribution field, and perform model order reduction and compression on the residual stress distribution field to obtain the residual stress digital fingerprint; In this embodiment, the process of performing thermo-mechanical coupled finite element simulation on the conveyor chain, extracting the residual stress distribution field, and performing model order reduction and compression on the residual stress distribution field to obtain a digital fingerprint of the residual stress includes: Welding current, welding voltage, welding speed, and geometric model of the conveyor chain are collected. The double ellipsoidal heat source model is used as the welding input load. Zero displacement constraint boundary conditions are applied to the non-welding area in the geometric model. Thermo-mechanical coupling simulation is performed to obtain the residual stress distribution field.

[0019] Specifically, welding current, welding voltage, and welding speed jointly determine the welding heat input power and heating range. The geometric model is a three-dimensional solid discretized mesh containing the geometric dimensions of the chain links. The shape parameters of the double ellipsoidal heat source model are determined based on the molten pool size, and its heat flux density spatial distribution function is used to simulate the thermal loading of the welding arc on the workpiece surface. Zero-displacement constraint boundary conditions are applied to the contact surface between the tooling fixture and the chain in the geometric model to constrain its translational degrees of freedom in three directions, simulating the rigid fixation during actual welding.

[0020] Performing thermo-mechanical coupling simulation involves sequentially solving the transient temperature field and quasi-static stress field using a finite element solver. At each time step, the temperature field calculation results are applied as a thermal load to the mechanical analysis, and the material in the weld pool region is set to an elastoplastic constitutive relationship where the yield stress changes with temperature, thereby obtaining the residual stress distribution field after the weld cools to room temperature.

[0021] The von Mises equivalent stress at each node in the residual stress distribution field is extracted as the residual stress value.

[0022] Specifically, a node is the intersection of topological connections of three-dimensional solid elements, and each node has a unique number and spatial coordinates. The von Mises equivalent stress is a scalar equivalent value synthesized from the six stress tensor components of the node according to the formula of the fourth strength theory. This value is used to quantitatively assess the equivalent strength level of the spatial location under multiaxial stress.

[0023] The residual stress values ​​of all spatial nodes corresponding to each simulation time step constitute a column vector. The column vectors are arranged sequentially according to the time step order of the thermo-mechanical coupling simulation to construct a residual stress snapshot matrix.

[0024] Specifically, the simulation time step refers to the computational progression step in solving the temperature and stress fields during the thermo-mechanical coupling simulation. Each step outputs a set of full-field stress data that strictly corresponds to the node numbering order. This set of data is arranged in ascending order of node number to form a high-dimensional column vector, the dimension of which is equal to the total number of mesh nodes. The column vectors of all time steps are arranged from left to right according to the computational order to form the residual stress snapshot matrix.

[0025] Singular value decomposition is performed on the residual stress snapshot matrix to obtain the singular value decomposition results.

[0026] Specifically, singular value decomposition decomposes the residual stress snapshot matrix into the product of the left singular vector matrix, the singular value vector, and the right singular vector matrix. This decomposition maps the original high-dimensional data to a set of orthonormal bases in the least squares sense, and the basis vectors are arranged in descending order of energy contribution.

[0027] Extract the left singular vector matrix and singular value vector from the singular value decomposition results.

[0028] Specifically, each column of the left singular vector matrix is ​​an orthogonal spatial basis vector, and the singular value vector is composed of the non-negative real number corresponding to that column. Both are directly used in the decomposition result.

[0029] The singular values ​​in the singular value vector are arranged in descending order of their numerical values, and the sequence number is the order corresponding to each singular value.

[0030] Specifically, the elements in the singular value vector are arranged in descending order of value, and the position number of each element is the order of that singular value. The order is used to identify the contribution ranking of each orthogonal mode in the total energy of the original data.

[0031] The cumulative energy percentage is obtained by calculating the sum of squares of each singular value in the singular value vector.

[0032] Specifically, the cumulative energy ratio refers to the sum of the squares of the singular values ​​of each order starting from the first order, and the sum divided by the sum of the squares of the singular values ​​of all orders. This ratio is used to characterize the degree to which the first few modes retain the information of the original data.

[0033] The smallest order at which the cumulative energy percentage first reaches or exceeds the preset energy threshold is determined as the cutoff order.

[0034] Specifically, the preset energy threshold is a percentage set before simulation based on the data compression accuracy requirements. It is usually between 99% and 99.9%. This threshold represents the lower limit of information integrity required to retain the mode. The energy percentage is accumulated step by step starting from the first order. When the percentage is greater than or equal to the preset energy threshold for the first time, the current accumulated order is determined as the truncation order.

[0035] By extracting the left singular vectors of the previous truncation order from the left singular vector matrix, a reduced-order basis matrix is ​​obtained.

[0036] Specifically, truncation means retaining all column vectors from the first column to the truncation order column in the left singular vector matrix, discarding the remaining higher-order column vectors. These retained column vectors are arranged in their original order to form a reduced-order basis matrix. The number of rows in this matrix is ​​equal to the total number of nodes, and the number of columns is equal to the truncation order.

[0037] The low-dimensional projection coefficient matrix is ​​obtained by multiplying the transpose of the reduced basis matrix with the residual stress snapshot matrix.

[0038] Specifically, the transpose of the reduced-order basis matrix has a size equal to the truncation order multiplied by the total number of nodes, and the size of the residual stress snapshot matrix has a size equal to the total number of nodes multiplied by the total number of time steps. After performing matrix multiplication on the two, a low-dimensional projection coefficient matrix with a size equal to the truncation order multiplied by the total number of time steps is obtained. Each column vector of this matrix is ​​the projection coordinate of the original high-dimensional residual stress field on the low-dimensional subspace spanned by the reduced-order basis matrix.

[0039] Extract the low-dimensional projection coefficient column vector corresponding to the cooling end time step of the thermo-mechanical coupling simulation, and use it as the equivalent residual stress modal coordinate vector.

[0040] Specifically, the cooling end time step refers to the final simulation output moment when the workpiece temperature recovers to room temperature and reaches thermal equilibrium during the welding thermal cycle. The column vector corresponding to this moment is extracted from the low-dimensional projection coefficient matrix. The values ​​of each element in this vector represent the participation weight of each reduced-order basis mode in the final residual stress field.

[0041] The equivalent residual stress modal coordinate vector and the reduced-order basis matrix are encapsulated to obtain the residual stress digital fingerprint.

[0042] Specifically, the equivalent residual stress modal coordinate vector and the reduced-order basis matrix are stored as binary data blocks. The reduced-order basis matrix is ​​used to map the modal coordinates back to the spatial node stress distribution, and the equivalent residual stress modal coordinate vector is used to characterize the projection position of the current residual stress state in the low-dimensional space. Together, they constitute the residual stress digital fingerprint and are solidified in the edge controller for subsequent recursive identification and retrieval.

[0043] S2. Obtain the original analog signal from the sensor array, perform analog-to-digital conversion and digital filtering on the original analog signal to obtain the digital signal, perform channel analysis and calculation on the digital signal to obtain the equivalent stiffness, sag and catenary observation tension value; In this embodiment, the process of acquiring the original analog signal from the sensor array, performing analog-to-digital conversion and digital filtering on the original analog signal to obtain a digital signal, and performing channel analysis and calculation on the digital signal to obtain the equivalent stiffness, sag, and catenary observation tension values ​​includes: Raw analog voltage signals are acquired through a sensor array.

[0044] Specifically, the sensor array consists of a tension sensor installed at the connection between the head and tail of the scraper conveyor chain and a laser displacement sensor installed at the mid-span of the chain. The tension sensor converts the tension into a millivolt-level differential voltage, and the laser displacement sensor converts the mid-span vertical offset into a linear voltage. The original analog voltage signal refers to the unconditioned continuous-time voltage waveform directly output by the aforementioned sensors.

[0045] The original analog voltage signal is pre-filtered by an analog anti-aliasing filter and then converted from analog to digital to obtain the original digital sequence.

[0046] Specifically, an analog anti-aliasing filter is a low-pass analog filter with a cutoff frequency set to 0.4 to 0.5 times the analog-to-digital conversion sampling frequency. It is used to attenuate frequency domain components in the original analog voltage signal that are higher than the cutoff frequency to prevent spectral aliasing after sampling. Analog-to-digital conversion refers to quantizing the amplitude of the pre-filtered analog voltage into a binary digital quantity with a finite bit width according to the sampling frequency set by the edge controller hardware timer. A digital quantity is output at each sampling moment, and the digital quantities at each moment are concatenated into the original digital quantity sequence according to the sampling order.

[0047] The original digital sequence is low-pass filtered by a fixed-point infinite impulse response filter to obtain a digital signal.

[0048] Specifically, the fixed-point infinite impulse response filter is a digital filter that uses a fixed-point number format to store filter coefficients and intermediate variables. Its system function contains poles and can achieve the same transition band attenuation steepness with a lower order than the finite impulse response filter. This filter suppresses the fluctuation components in the original digital sequence that are higher than the cutoff frequency. The cutoff frequency is tuned according to the effective bandwidth of the chain tension signal, and the filtered output sequence is a digital signal.

[0049] The digital signal is analyzed by channel analysis to obtain the channel analysis results, which include the tension value at the first end of the chain, the tension value at the second end of the chain, and the displacement value at the mid-span of the chain.

[0050] Specifically, channel parsing refers to the edge controller identifying and extracting the sampled values ​​of each channel from the time slice data of the digital signal based on the correspondence between the sensor channel number and the physical quantity in the configuration file. The channel data corresponding to the head tension sensor is marked as the tension value at the first end of the chain, the channel data corresponding to the tail tension sensor is marked as the tension value at the second end of the chain, and the channel data corresponding to the laser displacement sensor is marked as the mid-span displacement value of the chain.

[0051] The equivalent stiffness is obtained by least squares regression fitting of the tension values ​​at the first and second ends of the chain and the mid-span displacement value of the chain.

[0052] Specifically, least squares regression fitting refers to using the difference between the tension values ​​at the first and second ends of the chain collected at the same time as the independent variable and the chain mid-span displacement value at that time as the dependent variable. Data from multiple consecutive sampling times are combined into a sample set, and the linear regression coefficient is estimated according to the criterion of minimizing the sum of squared residuals. This regression coefficient is the axial tensile flexibility of the chain in the current state, and its reciprocal is the equivalent stiffness. The larger the equivalent stiffness, the stronger the axial tensile resistance of the chain.

[0053] The chain sag is obtained by estimating the catenary sag based on the mid-span displacement value of the chain.

[0054] Specifically, the estimation of catenary sag refers to the subtraction of the fixed offset component introduced by the height difference between the two support points of the chain from the vertical offset value measured by the laser displacement sensor at the mid-span. This height difference is obtained by on-site installation measurement and pre-stored in the edge controller. The result after subtraction is the actual sag amplitude formed by the chain under its own weight, that is, the sag of the chain.

[0055] Based on the tension values ​​at the first and second ends of the chain, the horizontal tension of the catenary is calculated using the static equilibrium equation of the catenary, thus obtaining the observed tension value of the catenary.

[0056] Specifically, the static equilibrium equation of the catenary describes the quantitative relationship between the sag shape and horizontal tension of a uniform flexible chain in static equilibrium in a gravitational field. The tension values ​​at the first and second ends of the chain are substituted into the equation as boundary conditions for the tension at both ends of the catenary. The horizontal tension parameter is solved numerically. This parameter is the component value of the tension along the conveying direction at any cross-section of the chain, denoted as the observed tension value of the catenary, and used as the measured tension input for feedback control.

[0057] S3. Perform low-pass filtering on the equivalent stiffness through the edge controller to obtain the filtered stiffness value; perform recursive least squares identification on the filtered stiffness value and the digital fingerprint of residual stress to obtain the residual stress release rate. In this embodiment, the step of performing low-pass filtering on the equivalent stiffness through the edge controller to obtain the filtered stiffness value; and performing recursive least-squares identification on the filtered stiffness value and the residual stress digital fingerprint to obtain the residual stress release rate includes: The equivalent stiffness is low-pass filtered to obtain the filtered stiffness value. The filtered stiffness values ​​are then arranged in chronological order to form a sequence of filtered stiffness values.

[0058] Specifically, the low-pass filter adopts a first-order infinite impulse response digital filter structure. Its cutoff frequency is set according to the effective variable frequency band contained in the equivalent stiffness measurement value. The filter output is the filtered stiffness value after filtering out the high-frequency noise of the tension sensor. At the end of each data acquisition cycle, the edge controller appends the current filtered stiffness value to the tail of the queue. The queue stores the filtered stiffness values ​​in ascending order according to the sampling time to form a filtered stiffness value sequence.

[0059] The reduced-order basis matrix in the residual stress digital fingerprint is transposed to obtain the reduced-order basis matrix transpose.

[0060] Specifically, the reduced-order basis matrix is ​​a component of the residual stress digital fingerprint. Its number of rows is equal to the total number of nodes in the finite element model and its number of columns is equal to the truncation order. The reduced-order basis matrix transpose obtained by the transpose operation after interchanging the rows and columns of the original matrix has the dimensional characteristics that the number of rows is equal to the truncation order and the number of columns is equal to the total number of nodes.

[0061] The reduced basis matrix is ​​transposed and multiplied with the preset strain-stiffness projection coefficient vector to obtain the projection observation column vector.

[0062] Specifically, the strain-stiffness projection coefficient vector is a column vector obtained through calibration experiments. Its dimension is equal to the total number of nodes in the finite element model. The calibration experiment process is as follows: a known axial tensile load is applied to a specimen of the same specification as the chain being measured, and the overall stiffness value under each load is recorded. At the same time, a finite element model of the specimen is established and the same load is applied to calculate the residual stress value of each node. The regularized least squares method is used to establish a linear mapping relationship between the residual stress value of the node and the overall stiffness change. The coefficient vector of this mapping relationship constitutes the strain-stiffness projection coefficient vector. The matrix product of the reduced-order basis matrix transpose and the coefficient vector yields the projection observation column vector with a dimension equal to the truncation order.

[0063] Transpose the projected observation column vector to obtain the equivalent observation row vector.

[0064] Specifically, the projected observation column vector is a column vector with a truncated order row and a column. The transpose operation transforms this column vector into an equivalent observation row vector with a truncated order column. This row vector is used in recursive least squares identification to map the state vector to the observation space.

[0065] Using the filtered stiffness value sequence as the observation for least squares identification, the equivalent residual stress modal coordinate vector in the residual stress digital fingerprint as the initial state vector for recursive least squares identification, and the equivalent observation row vector as the observation row vector for recursive least squares identification, recursive least squares identification is performed to obtain the residual stress estimation sequence.

[0066] Specifically, the observations identified by recursive least squares are scalar values ​​of the filtered stiffness at each sampling time. The initial state vector is taken as the equivalent residual stress modal coordinate vector in the residual stress digital fingerprint to provide a priori state estimate at the start of the recursion. The state transition matrix is ​​set as an identity matrix to characterize that the residual stress modal coordinates remain constant within a sampling interval. The observation row vector is used to project the current state vector into a stiffness prediction value and compare it with the measured filtered stiffness value to generate a residual. The recursive algorithm iterates step by step according to three equations: covariance matrix update, gain matrix calculation, and state vector update. At each time step, a state estimation vector with the same dimension as the equivalent residual stress modal coordinate vector is output. The physical meaning of each component in this vector is the estimated value of the residual stress modal coordinates corresponding to each reduced-order mode at the current time.

[0067] A moving average filter is applied to the residual stress estimation sequence to obtain a smoothed residual stress estimation sequence.

[0068] Specifically, the moving average filter uses a rectangular window to perform an equal-weighted average of the estimated values ​​at each time point in the residual stress estimation sequence. The window length is set according to the dominant frequency period of the residual stress release process. At each sampling time point, the arithmetic mean of all residual stress estimates within the window is output. The average values ​​at each time point are arranged in ascending order of time to form a smooth residual stress estimation sequence.

[0069] Perform a first-order backward difference operation on the smoothed residual stress estimation sequence, and divide the difference result by the preset sampling period to obtain the residual stress release rate.

[0070] Specifically, the first-order backward difference operation refers to using the difference between the current smoothed residual stress estimate and the previous smoothed residual stress estimate as the numerator. The preset sampling period refers to the hardware timer interrupt period of the edge controller. This period is fixed as the interval between the edge controller performing one complete data acquisition and control law calculation. The difference is divided by this interval to obtain the change in the residual stress modal coordinates per unit time. This change is recorded as the residual stress release rate.

[0071] S4. When the hardware timer of the edge controller reaches the preset servo cycle, the feedforward compensation tension is calculated based on the filter stiffness value. The deviation between the feedforward compensation tension and the observed tension value of the catenary is calculated to obtain the tension tracking error. In this embodiment, when the hardware timer of the edge controller reaches a preset servo cycle, the feedforward compensation tension is calculated based on the filter stiffness value, and the deviation between the feedforward compensation tension and the observed tension value of the catenary is calculated to obtain the tension tracking error, including: When the edge controller's hardware timer reaches the preset servo cycle, it obtains the current filtered stiffness value obtained by performing low-pass filtering on the equivalent stiffness.

[0072] Specifically, the preset servo cycle is a fixed time interval preset by the edge controller's built-in hardware timer based on the real-time requirements of the tension control system. This interval is written into the timer's period matching register to determine the trigger frequency of the timer overflow interrupt. When the timer count reaches the preset cycle value, the interrupt service routine is triggered. The interrupt service routine reads the value of the most recent low-pass filter operation output from the filter stiffness value sequence buffer as the current filter stiffness value.

[0073] Based on the current filter stiffness value, a feedforward compensation calculation is performed on the preset reference tension to obtain the feedforward compensation tension.

[0074] Specifically, the preset reference tension is the target tension value determined offline based on the design tension requirements under no-load and no residual stress release conditions before the chain is put into operation. The calibration process is as follows: after the chain is installed, multiple sets of known tensions are applied and the corresponding equivalent stiffness values ​​are recorded. The average value of each set of tensions and equivalent stiffness is used as the reference point. The preset reference stiffness is a reference stiffness value determined by taking the arithmetic mean of the equivalent stiffness values ​​collected under the calibration conditions. The feedforward compensation operation is as follows: calculate the difference between the current filtered stiffness value and the preset reference stiffness, multiply the difference by the preset compensation gain, and then add it to the preset reference tension. The sum of the two is the feedforward compensation tension. The preset compensation gain is adjusted based on the sensitivity coefficient of the chain stiffness change on the tension.

[0075] The tension tracking error is obtained by calculating the difference between the feedforward compensation tension and the observed tension value of the catenary.

[0076] Specifically, the difference calculation involves subtracting the observed catenary tension value from the feedforward compensation tension. The feedforward compensation tension serves as the desired input to the tension control loop, while the observed catenary tension value is the actual feedback value calculated from the sensor signal. The algebraic value obtained by subtracting the two is the tension tracking error. The sign of this error indicates the direction of deviation of the current actual tension from the desired tension.

[0077] In this embodiment, the step of performing feedforward compensation calculation on the preset reference tension based on the current filter stiffness value to obtain the feedforward compensation tension includes: The difference between the current filter stiffness value and the preset reference filter stiffness value is calculated to obtain the filter stiffness value deviation; the preset reference filter stiffness value is determined by taking the arithmetic mean of the initial filter stiffness values ​​obtained by performing low-pass filtering on the equivalent stiffness under the calibration conditions before the chain is put into operation.

[0078] Specifically, the preset reference filter stiffness value is a stiffness reference determined under calibration conditions before the chain is put into operation. Calibration conditions refer to the stable operating conditions when the chain is unloaded and the tensioning mechanism maintains the designed sag. Under these conditions, equivalent stiffness values ​​are continuously collected for no less than ten sampling periods and then processed by low-pass filtering with the same cutoff frequency to obtain a corresponding number of initial filter stiffness values. The quotient obtained by summing all the initial filter stiffness values ​​and dividing by the total number of collections is the preset reference filter stiffness value. The algebraic value obtained by subtracting the preset reference filter stiffness value from the current filter stiffness value is the filter stiffness value deviation.

[0079] The stiffness compensation amount is obtained by multiplying the filter stiffness value deviation with the preset feedforward compensation gain.

[0080] Specifically, the preset feedforward compensation gain is a proportional coefficient pre-tuned based on the linearized transmission relationship between the chain's axial stiffness change and tension change. This coefficient is obtained by determining its theoretical value through static analysis of the chain and then fine-tuning it through on-site step response experiments. The stiffness compensation amount is equal to the product of the filter stiffness value deviation and the preset feedforward compensation gain. This product represents the required tension correction amplitude caused by stiffness drift.

[0081] The stiffness compensation amount is added to the preset reference tension to obtain the feedforward compensation tension.

[0082] Specifically, the preset reference tension is the tension reference value determined by offline calibration of the scraper conveyor chain before commissioning based on the design tension requirements. This design requirement value comes from the specified parameters of the initial tension state of the chain in the scraper conveyor design document. The addition operation refers to adding the stiffness compensation amount to the preset reference tension, and the sum of the two is the feedforward compensation tension.

[0083] S5. Using the residual stress release rate and tension tracking error as indexes, interpolate the pre-stored fuzzy inference digital lookup table to obtain the PID adjustment coefficient, and calculate the adaptive feedback control quantity based on the PID adjustment coefficient through the feedback control law. In this embodiment, the step of interpolating the PID adjustment coefficients using the residual stress release rate and tension tracking error as indexes to a pre-stored fuzzy inference digital lookup table, and calculating the adaptive feedback control quantity based on the PID adjustment coefficients using a feedback control law, includes: Using the residual stress release rate and tension tracking error as indexes, two-dimensional linear interpolation is performed in a pre-stored fuzzy inference digital lookup table to obtain the PID adjustment coefficients, which include the proportional gain adjustment coefficient, the integral time adjustment coefficient, and the derivative gain adjustment coefficient.

[0084] Specifically, the residual stress release rate is used as the first index variable, and the tension tracking error is used as the second index variable. The two constitute the index coordinate points on the two-dimensional plane. The fuzzy inference digital lookup table is a two-dimensional discrete table generated by offline inference calculation after the fuzzy rule base established based on expert experience before commissioning. The row index of the table corresponds to several fuzzy sub-intervals divided by the residual stress release rate after fuzzification by the membership function, and the column index corresponds to several fuzzy sub-intervals divided by the tension tracking error after fuzzification by the membership function. Each table cell stores a set of proportional gain adjustment coefficients, integral time adjustment coefficients, and differential gain adjustment coefficients calculated by fuzzy inference rules. Two-dimensional linear interpolation refers to the process where, when the index coordinate point falls within the rectangular grid cell enclosed by two adjacent row indices and two adjacent column indices, two temporary values ​​are obtained by linear interpolation of the two pairs of adjacent column nodes along the row direction, and then three sets of coefficients that precisely correspond to the index coordinates are obtained by linear interpolation of the two temporary values ​​along the column direction.

[0085] The proportional gain, integral time, and derivative gain are read from the pre-stored data in the edge controller. The proportional gain, integral time, and derivative gain are obtained offline by digital proportional-integral-derivative feedback control law before the chain is put into operation.

[0086] Specifically, the proportional gain, integral time, and derivative gain pre-stored in the edge controller are obtained by tuning using the critical proportional gain method before the chain is put into operation. The tuning process is as follows: in the closed-loop control system, only proportional control is enabled, and the proportional gain is gradually increased until the system output produces a continuous oscillation with constant amplitude. The critical gain and critical oscillation period at this time are recorded. The initial proportional gain value, integral time value, and derivative gain value are calculated according to the empirical formula of the critical proportional gain method. The three values ​​are written into the non-volatile memory of the edge controller as the initial control parameters for the online operation phase.

[0087] The adaptive proportional gain is obtained by multiplying the proportional gain adjustment coefficient with the proportional gain.

[0088] Specifically, the proportional gain adjustment coefficient is a dimensionless scaling factor, and the product of this factor and the initial proportional gain value is denoted as the adaptive proportional gain.

[0089] The adaptive integral time is obtained by multiplying the integral time adjustment coefficient with the integral time.

[0090] Specifically, the integration time adjustment factor is a dimensionless scaling factor, and the product of this factor and the initial integration time value is denoted as the adaptive integration time.

[0091] The adaptive differential gain is obtained by multiplying the differential gain adjustment coefficient with the differential gain.

[0092] Specifically, the differential gain adjustment coefficient is a dimensionless scaling factor, and the product of this factor and the initial differential gain value is denoted as the adaptive differential gain.

[0093] Based on adaptive proportional gain, adaptive integral time, and adaptive derivative gain, a digital proportional-integral-derivative feedback control law is used to calculate the tension tracking error, resulting in an adaptive feedback control quantity.

[0094] Specifically, the digital proportional-integral-derivative feedback control law is implemented using a positional discrete algorithm. Its expression consists of three parts: a proportional term, an integral term, and a derivative term. The proportional term is equal to the adaptive proportional gain multiplied by the tension tracking error value at the current moment. The integral term is equal to the adaptive proportional gain divided by the adaptive integral time and then multiplied by the sum of the products of the tension tracking error values ​​at each moment and the sampling interval. The derivative term is equal to the adaptive proportional gain multiplied by the adaptive derivative gain and then multiplied by the quotient obtained by dividing the difference between the tension tracking error at the current moment and the tension tracking error at the previous moment by the sampling interval. The sum of the three terms is the adaptive feedback control quantity.

[0095] S6. The edge controller receives the pre-built timing prediction proxy model, and generates the tensioning mechanism advance action timing based on the timing prediction proxy model and the filter stiffness value, sag, residual stress release rate and tension tracking error. Based on the tensioning mechanism advance action timing and the feedforward compensation tension and adaptive feedback control quantity, the tensioning mechanism control command is generated.

[0096] In this embodiment, the edge controller receives a pre-built timing prediction proxy model, and generates an advance action timing sequence for the tensioning mechanism based on the timing prediction proxy model and the filtered stiffness value, sag, residual stress release rate, and tension tracking error. Based on the advance action timing sequence of the tensioning mechanism and the feedforward compensation tension and adaptive feedback control quantity, it generates control commands for the tensioning mechanism, including: The time-series prediction agent model is received from the cloud via an industrial Ethernet gateway.

[0097] Specifically, the industrial Ethernet gateway is the physical interface device between the edge controller and the upper-layer network. It supports industrial communication protocols such as Modbus TCP or PROFINET to achieve data transmission and reception. The time-series predictive agent model is a mathematical model built offline on the cloud server. This model includes three types of parameters: state transition matrix, measurement matrix, and trend transition coefficient. The model is encapsulated in the form of a structured binary file and then sent to the file storage area of ​​the edge controller through the gateway.

[0098] The filter stiffness value, chain sag, residual stress release rate, and tension tracking error within the continuous sampling period are written into the annular thermal data buffer of the edge controller in the order of acquisition time to obtain the edge controller cached data.

[0099] Specifically, the circular thermal data buffer is a circular queue storage area in the edge controller's memory divided into fixed byte lengths. The buffer capacity is set according to the edge controller's memory resources and the data requirements of the prediction algorithm. In each sampling period, the four values ​​of the current moment's filter stiffness value, chain sag, residual stress release rate, and tension tracking error are packaged into a data record and written to the tail of the queue. When the queue is full, the new record overwrites the oldest historical record.

[0100] Using the filtered stiffness value in the edge controller cache data as the measurement input, and the sag, residual stress release rate, and tension tracking error in the edge controller cache data as auxiliary inputs, a linear Kalman filter state estimation is performed through a time-series prediction surrogate model to obtain a state estimation vector; the state estimation vector includes a trend intercept state component, a trend slope state component, and an autoregressive state component.

[0101] Specifically, in the time-series prediction proxy model, the state transition matrix is ​​used to describe the linear recursive relationship of the state vector from the previous time to the current time. The measurement matrix is ​​used to map the state vector to the measurement space to calculate the measurement prediction value. The trend transition coefficient is used to linearly superimpose the auxiliary input vector into the state prediction equation. The linear Kalman filter is recursively executed in two stages: the time update equation and the measurement update equation. In the time update stage, the state prior estimate and its covariance matrix are calculated using the state transition matrix and the trend transition coefficient. In the measurement update stage, the Kalman gain is calculated using the measurement matrix and the measurement input, and the prior estimate is corrected to obtain the state posterior estimate vector. The posterior estimate vector consists of three values: the trend intercept component, the trend slope component, and the autoregressive component.

[0102] The trend intercept state component and the trend slope state component are used as the current trend intercept and the current trend slope, respectively.

[0103] Specifically, the values ​​corresponding to the trend intercept state components are extracted from the state estimation vector and assigned to the current trend intercept variable, and the values ​​corresponding to the trend slope state components are extracted and assigned to the current trend slope variable.

[0104] Construct the current trend line based on the current trend intercept and the current trend slope.

[0105] Specifically, the current trend line is a linear function with the sampling number as the independent variable and the filter stiffness value as the dependent variable. Its function intercept is equal to the current trend intercept, and its function slope is equal to the current trend slope.

[0106] Linear extrapolation is performed on the current trend line according to the preset extrapolation step size to obtain the advance action sequence of the tensioning mechanism, which includes the predicted filtered stiffness values ​​at each extrapolation time point in the future.

[0107] Specifically, the preset extrapolation step size is a positive integer set before commissioning based on the mechanical action delay time of the tensioning mechanism and the advance prediction requirements of the control system. This integer represents the number of sampling cycles that will be advanced into the future from the current sampling time. Linear extrapolation refers to taking points along the positive time axis with the current sampling time as the starting point at preset step size intervals, and calculating the corresponding predicted value of the filter stiffness at each extrapolation point using the current trend line. The predicted values ​​of all extrapolation points are arranged in chronological order to form the advance action sequence of the tensioning mechanism.

[0108] The sum of the feedforward compensation tension and the adaptive feedback control quantity is used as the basic control amplitude at the current moment. Based on the predicted filter stiffness value at each extrapolated time point in the advance action sequence of the tensioning mechanism, the tensioning mechanism is triggered when the predicted filter stiffness value reaches the preset action threshold, and a tensioning mechanism control command with a timestamp is generated.

[0109] Specifically, the basic control amplitude is the sum of the feedforward compensation tension and the adaptive feedback control quantity. This amplitude determines the output torque reference of the tensioning mechanism at the current moment. The preset action threshold includes two values: an upper stiffness threshold and a lower stiffness threshold. This set of thresholds is tuned according to the allowable stiffness range of the chain specified in the scraper conveyor design document. The edge controller scans each predicted value in the advance action sequence at fixed time intervals. When the predicted filtered stiffness value corresponding to a certain extrapolation moment is higher than the upper threshold or lower than the lower threshold, the timestamp of that moment is combined with the basic control amplitude and encoded into a control command message, which is sent to the servo drive of the tensioning mechanism to achieve the advance action.

[0110] In this embodiment, the pre-built time-series prediction proxy model is generated according to the following steps: Obtain historical operating data of the scraper conveyor chain. The historical operating data includes historical filter stiffness value time series, historical sag time series, historical residual stress release rate time series, and historical tension tracking error time series.

[0111] Specifically, the historical operation data comes from the actual operation log of the chain recorded in the flash memory of the edge controller. Each log contains a timestamp of the sampling time and four fields at that time: filter stiffness value, chain sag, residual stress release rate, and tension tracking error. After reading the log records, they are sorted in ascending order according to the sampling time, and the values ​​of each field are extracted to form four independent one-dimensional time series.

[0112] A trend term significance test was performed on the time series of historical filter stiffness values, and the results of the trend term significance test were obtained.

[0113] Specifically, the Mann-Kendall test is used to test the significance of the trend term. The test calculates the sign statistic of the difference between all pairs of consecutive data in the historical filter stiffness value time series. This statistic is then compared with a critical value with a significance level of 5%. If the absolute value of the statistic is greater than the critical value, the test result is significant; otherwise, it is not significant.

[0114] When the trend term significance test result is significant, the trend intercept state component and the trend slope state component are included in the state vector; when the trend term significance test result is not significant, the initial estimates of the trend intercept state component and the trend slope state component are set to zero.

[0115] Specifically, the state vector is the core set of variables used to describe the dynamic behavior of the system in the time series prediction surrogate model. The trend intercept component represents the baseline level of the filter stiffness value, and the trend slope component represents the rate of change of the filter stiffness value per unit time. When the trend is not significant, the initial estimates of both are assigned to zero.

[0116] Autocorrelation and partial autocorrelation functions were performed on the time series of historical filter stiffness values ​​to obtain the autoregression order.

[0117] Specifically, the autocorrelation function is used to calculate the autocorrelation coefficient of a time series at lag orders of one, two, three up to the maximum lag order. The partial autocorrelation function is used to calculate the partial autocorrelation coefficient after removing the influence of intermediate lag terms. It is calculated step by step starting from the lag order of one. When the absolute value of the partial autocorrelation coefficient is first lower than twice the standard deviation, the lag order minus one is the autoregression order.

[0118] The autoregression order is used as the dimension of the autoregressive state components in the state vector.

[0119] Specifically, the autoregressive state components occupy multiple consecutive dimensions in the state vector, with each dimension corresponding to an autoregressive coefficient of a lag order. The total number of dimensions is equal to the autoregressive order.

[0120] A state-space model is constructed using historical sag time series, historical residual stress release rate time series, and historical tension tracking error time series as auxiliary input variables, historical filtered stiffness value time series as measurement output variable, and state vector containing trend intercept state component, trend slope state component, and autoregressive state component as state variable.

[0121] Specifically, the state-space model consists of state equations and measurement equations. The state equations associate the state vector at the current sampling time with the state vector at the next sampling time through a state transition matrix, and simultaneously add auxiliary input variables to the state update term through trend transition coefficients. The measurement equations map the state vector to measurement output variables through a measurement matrix.

[0122] By jointly identifying the state-space model, the state transition matrix, measurement matrix, and trend transition coefficients are obtained.

[0123] Specifically, the joint identification adopts a numerical algorithm based on subspace identification. This algorithm takes the historical filtered stiffness value time series as output data and the historical sag time series, historical residual stress release rate time series, and historical tension tracking error time series as input data. By constructing a data matrix and performing QR decomposition and singular value decomposition, it simultaneously estimates all unknown elements in the three matrices: state transition matrix, measurement matrix, and trend transition coefficient.

[0124] The state transition matrix, measurement matrix, and trend transition coefficients are encapsulated into a time series prediction proxy model.

[0125] Specifically, encapsulation refers to storing the three numerical matrices—state transition matrix, measurement matrix, and trend transition coefficient—continuously in the same file in row-major order. The file header contains the number of rows and columns of the matrix, as well as the model version number. This file forms a time-series prediction proxy model for the edge controller to load and use for Kalman filter recursive calculation.

[0126] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0127] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0128] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for adaptive control of chain tension in a scraper conveyor based on intelligent welding, characterized in that, Includes the following steps: S1. Perform thermo-mechanical coupled finite element simulation on the conveyor chain, extract the residual stress distribution field, and perform model order reduction and compression on the residual stress distribution field to obtain the residual stress digital fingerprint; S2. Obtain the original analog signal from the sensor array, perform analog-to-digital conversion and digital filtering on the original analog signal to obtain the digital signal, perform channel analysis and calculation on the digital signal to obtain the equivalent stiffness, sag and catenary observation tension value; S3. Perform low-pass filtering on the equivalent stiffness through the edge controller to obtain the filtered stiffness value; perform recursive least squares identification on the filtered stiffness value and the digital fingerprint of residual stress to obtain the residual stress release rate. S4. When the hardware timer of the edge controller reaches the preset servo cycle, the feedforward compensation tension is calculated based on the filter stiffness value. The deviation between the feedforward compensation tension and the observed tension value of the catenary is calculated to obtain the tension tracking error. S5. Using the residual stress release rate and tension tracking error as indexes, interpolate the pre-stored fuzzy inference digital lookup table to obtain the PID adjustment coefficient, and calculate the adaptive feedback control quantity based on the PID adjustment coefficient through the feedback control law. S6. The edge controller receives the pre-built timing prediction proxy model, and generates the tensioning mechanism advance action timing based on the timing prediction proxy model and the filter stiffness value, sag, residual stress release rate and tension tracking error. Based on the tensioning mechanism advance action timing and the feedforward compensation tension and adaptive feedback control quantity, the tensioning mechanism control command is generated.

2. The adaptive control method for chain tension of a scraper conveyor based on intelligent welding according to claim 1, characterized in that, The process involves performing thermo-mechanical coupled finite element simulation on the conveyor chain, extracting the residual stress distribution field, and then performing model order reduction and compression on the residual stress distribution field to obtain a digital fingerprint of the residual stress, including: We collect welding current, welding voltage, welding speed, and geometric model of the conveyor chain. The double ellipsoidal heat source model is used as the welding input load. Zero displacement constraint boundary conditions are applied to the non-welding area in the geometric model. Thermo-mechanical coupling simulation is performed to obtain the residual stress distribution field. The von Mises equivalent stress at each node in the residual stress distribution field is extracted as the residual stress value; The residual stress values ​​of all spatial nodes corresponding to each simulation time step constitute a column vector. The column vectors are arranged sequentially according to the time step order of the thermo-mechanical coupling simulation to construct a residual stress snapshot matrix. Singular value decomposition is performed on the residual stress snapshot matrix to obtain the singular value decomposition results; Extract the left singular vector matrix and singular value vector from the singular value decomposition results; The singular values ​​in the singular value vector are arranged in descending order of their numerical values, and the sequence number is the order corresponding to each singular value. The cumulative energy percentage is obtained by calculating the sum of squares of each singular value in the singular value vector and the cumulative percentage of the sum of squares. The smallest order at which the cumulative energy percentage first reaches or exceeds the preset energy threshold is determined as the cutoff order. The reduced-order basis matrix is ​​obtained by truncating the left singular vectors of the first truncation order from the left singular vector matrix. The low-dimensional projection coefficient matrix is ​​obtained by multiplying the transpose of the reduced basis matrix with the residual stress snapshot matrix. Extract the low-dimensional projection coefficient column vector corresponding to the cooling end time step of the thermo-mechanical coupling simulation and use it as the equivalent residual stress modal coordinate vector. The equivalent residual stress modal coordinate vector and the reduced-order basis matrix are encapsulated to obtain the residual stress digital fingerprint.

3. The adaptive control method for chain tension of a scraper conveyor based on intelligent welding according to claim 1, characterized in that, The process involves acquiring the original analog signal from the sensor array, performing analog-to-digital conversion and digital filtering to obtain a digital signal, and then performing channel analysis and calculations on the digital signal to obtain the equivalent stiffness, sag, and observed catenary tension values, including: Raw analog voltage signals are acquired using a sensor array; The original analog voltage signal is pre-filtered by an analog anti-aliasing filter and then converted from analog to digital to obtain the original digital sequence. The original digital sequence is low-pass filtered by a fixed-point infinite impulse response filter to obtain a digital signal; The digital signal is analyzed by channel analysis to obtain the channel analysis results, which include the tension value at the first end of the chain, the tension value at the second end of the chain, and the mid-span displacement value of the chain. The equivalent stiffness is obtained by least squares regression fitting of the tension values ​​at the first and second ends of the chain and the mid-span displacement of the chain. The chain sag is obtained by estimating the catenary sag based on the mid-span displacement value of the chain. Based on the tension values ​​at the first and second ends of the chain, the horizontal tension of the catenary is calculated using the static equilibrium equation of the catenary, thus obtaining the observed tension value of the catenary.

4. The adaptive control method for chain tension of a scraper conveyor based on intelligent welding according to claim 1, characterized in that, The equivalent stiffness is low-pass filtered by the edge controller to obtain the filtered stiffness value; Recursive least-squares identification is performed on the filtered stiffness value and the digital fingerprint of residual stress to obtain the residual stress release rate, including: The equivalent stiffness is low-pass filtered to obtain the filtered stiffness value. The filtered stiffness values ​​are arranged in chronological order to form a sequence of filtered stiffness values. The reduced-order basis matrix in the residual stress digital fingerprint is transposed to obtain the transpose of the reduced-order basis matrix. The reduced basis matrix is ​​transposed and multiplied with the preset strain-stiffness projection coefficient vector to obtain the projection observation column vector; Transpose the projected observation column vector to obtain the equivalent observation row vector; Using the filtered stiffness value sequence as the observation for least squares identification, the equivalent residual stress modal coordinate vector in the residual stress digital fingerprint as the initial state vector for recursive least squares identification, and the equivalent observation row vector as the observation row vector for recursive least squares identification, recursive least squares identification is performed to obtain the residual stress estimation sequence; the strain-stiffness projection coefficient vector is a coefficient vector obtained in advance through calibration experiments, used to map the spatial node stress values ​​to the overall stiffness change of the chain; A moving average filter is applied to the residual stress estimation sequence to obtain a smoothed residual stress estimation sequence. A first-order backward difference operation is performed on the smoothed residual stress estimation sequence, and the difference result is divided by the preset sampling period to obtain the residual stress release rate.

5. The adaptive control method for chain tension of a scraper conveyor based on intelligent welding according to claim 3, characterized in that, When the hardware timer of the edge controller reaches the preset servo cycle, the feedforward compensation tension is calculated based on the filter stiffness value. The deviation between the feedforward compensation tension and the observed tension value of the catenary is calculated to obtain the tension tracking error, including: When the hardware timer of the edge controller reaches the preset servo cycle, it obtains the current filtered stiffness value obtained by performing low-pass filtering on the equivalent stiffness; Based on the current filter stiffness value, a feedforward compensation calculation is performed on the preset reference tension to obtain the feedforward compensation tension. The tension tracking error is obtained by calculating the difference between the feedforward compensation tension and the observed tension value of the catenary.

6. The adaptive control method for chain tension of a scraper conveyor based on intelligent welding according to claim 5, characterized in that, The step of performing feedforward compensation calculation on the preset reference tension based on the current filter stiffness value to obtain the feedforward compensation tension includes: The difference between the current filter stiffness value and the preset reference filter stiffness value is calculated to obtain the filter stiffness value deviation; the preset reference filter stiffness value is determined by taking the arithmetic mean of the initial filter stiffness values ​​obtained by performing low-pass filtering on the equivalent stiffness under the calibration conditions before the chain is put into operation. The stiffness compensation amount is obtained by multiplying the filter stiffness value deviation with the preset feedforward compensation gain. The stiffness compensation amount is added to the preset reference tension to obtain the feedforward compensation tension. The preset reference tension is the tension reference value of the scraper conveyor chain determined offline based on the design tension requirements before it is put into operation.

7. The adaptive control method for chain tension of a scraper conveyor based on intelligent welding according to claim 5, characterized in that, The process of interpolating the PID adjustment coefficients using a pre-stored fuzzy inference digital lookup table with residual stress release rate and tension tracking error as indexes, and then calculating the adaptive feedback control quantity based on the PID adjustment coefficients using a feedback control law, includes: Using the residual stress release rate and tension tracking error as indexes, two-dimensional linear interpolation is performed in a pre-stored fuzzy inference digital lookup table to obtain the PID adjustment coefficients, which include the proportional gain adjustment coefficient, integral time adjustment coefficient, and derivative gain adjustment coefficient. The proportional gain, integral time, and derivative gain are read from the edge controller. The proportional gain, integral time, and derivative gain are obtained offline by digital proportional-integral-derivative feedback control law before the chain is put into operation. The adaptive proportional gain is obtained by multiplying the proportional gain adjustment coefficient with the proportional gain. The adaptive integral time is obtained by multiplying the integral time adjustment coefficient by the integral time. The adaptive differential gain is obtained by multiplying the differential gain adjustment coefficient with the differential gain. Based on adaptive proportional gain, adaptive integral time, and adaptive derivative gain, a digital proportional-integral-derivative feedback control law is used to calculate the tension tracking error, thereby obtaining the adaptive feedback control quantity.

8. The adaptive control method for chain tension of a scraper conveyor based on intelligent welding according to claim 1, characterized in that, The edge controller receives a pre-built timing prediction proxy model, and based on the timing prediction proxy model and the filtered stiffness value, sag, residual stress release rate, and tension tracking error, generates an advance action timing sequence for the tensioning mechanism. Based on the advance action timing sequence of the tensioning mechanism and the feedforward compensation tension and adaptive feedback control quantity, it generates control commands for the tensioning mechanism, including: Receive the time-series prediction agent model sent from the cloud through the industrial Ethernet gateway; The filter stiffness value, chain sag, residual stress release rate and tension tracking error within the continuous sampling period are written into the ring thermal data buffer of the edge controller in the order of acquisition time to obtain the edge controller cache data. Using the filtered stiffness value in the edge controller cache data as the measurement input, and the sag, residual stress release rate and tension tracking error in the edge controller cache data as auxiliary inputs, a linear Kalman filter state estimation is performed through a time-series prediction surrogate model to obtain a state estimation vector; the state estimation vector includes a trend intercept state component, a trend slope state component and an autoregressive state component. The trend intercept state component and the trend slope state component are respectively used as the current trend intercept and the current trend slope. Construct the current trend line based on the current trend intercept and the current trend slope; Perform linear extrapolation on the current trend line according to the preset extrapolation step size to obtain the advance action sequence of the tensioning mechanism, which includes the predicted filtered stiffness values ​​at each extrapolation time point in the future. The sum of the feedforward compensation tension and the adaptive feedback control quantity is used as the basic control amplitude at the current moment. Based on the predicted filter stiffness value at each extrapolated time point in the advance action sequence of the tensioning mechanism, the tensioning mechanism is triggered when the predicted filter stiffness value reaches the preset action threshold, and a tensioning mechanism control command with a timestamp is generated.

9. The adaptive control method for chain tension of a scraper conveyor based on intelligent welding according to claim 8, characterized in that, The pre-built time-series prediction proxy model is generated according to the following steps: Acquire historical operating data of the scraper conveyor chain. The historical operating data includes historical filter stiffness value time series, historical sag time series, historical residual stress release rate time series, and historical tension tracking error time series. A significance test of the trend term was performed on the time series of historical filter stiffness values, and the results of the significance test of the trend term were obtained. When the trend term significance test result is significant, the trend intercept state component and the trend slope state component are included in the state vector. When the significance test result of the trend term is not significant, the initial estimates of the trend intercept state component and the trend slope state component are set to zero. Autocorrelation and partial autocorrelation functions were performed on the time series of historical filter stiffness values ​​to obtain the autoregression order. The autoregression order is used as the dimension of the autoregressive state components in the state vector; A state-space model is constructed using historical sag time series, historical residual stress release rate time series and historical tension tracking error time series as auxiliary input variables, historical filtered stiffness value time series as measurement output variable, and state vector containing trend intercept state component, trend slope state component and autoregressive state component as state variable. Joint identification of the state-space model yields the state transition matrix, measurement matrix, and trend transition coefficients; The state transition matrix, measurement matrix, and trend transition coefficients are encapsulated into a time series prediction proxy model.

Images