Automatic tension control system and method for chain conveyor belt chain of dump semi-trailer

Abstract

The application discloses a chain type conveying belt chain automatic tensioning control system and method for a self-unloading semitrailer, belongs to the technical field of automatic control of self-unloading semitrailers and material conveying, and comprises a multi-dimensional sensing data acquisition and analysis module, a multi-parameter state feature construction module, a time and space perception model calling and instruction generation module, and a resonance prevention instruction execution and driving module. The system acquires multi-source sensing data and extracts real-time parameters, performs compensation calculation based on an inclination angle and temperature to construct a dynamic relaxation state feature matrix. The feature matrix is input into a time sequence model to generate a target instruction. A tensioning compensation driving signal control system with damping buffer constraints is used to control a tensioning execution mechanism, so that the chain tension of the chain type conveying belt can be automatically and safely changed.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology for dump semi-trailers and material conveying, and in particular to an automatic tensioning control system and method for chain conveyor belts used in dump semi-trailers. Background Technology

[0002] Dump semi-trailers play a central role in modern logistics and bulk material transportation. Their cargo boxes are typically equipped with chain conveyor belts for rapid material unloading. In actual operation, due to frequent changes in the weight of the load, dynamic alternation of the cargo box's lifting angle, and fluctuations in ambient temperature, the conveyor chain is prone to nonlinear deformation and dynamic slack. If proper tension cannot be maintained in real time, serious problems such as chain slippage, severe vibration, and even derailment and breakage can occur.

[0003] Traditional methods for addressing chain slack primarily rely on mechanical spring-type passive tensioning devices or manual adjustments during regular inspections. Maintenance personnel typically rely on their experience to manually tighten tension bolts while the vehicle is stationary to restore the chain's initial tension; or they use spring assemblies with fixed stiffness to provide some compensating tension through the natural rebound of the springs when the chain slacks.

[0004] Traditional manual adjustments are time-consuming, labor-intensive, and inefficient, and cannot provide dynamic intervention during operation, resulting in a lag. Passive mechanical spring devices cannot sense the combined effects of carriage tilt angle, material weight distribution, and thermal expansion and contraction, and the compensation force they provide often deviates from actual operating conditions. Furthermore, traditional devices lack damping and anti-resonance design when performing tensioning actions, easily leading to mechanical resonance, which accelerates chain fatigue and equipment wear. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides an automatic tensioning control system and method for chain conveyor belts used in dump semi-trailers. The system generates target commands using multi-dimensional sensor data analysis and a time-series tension evaluation model, and is further driven by damping buffer constraints. This enables intelligent, precise, and adaptive adjustment of chain tension under complex working conditions.

[0006] The above objectives can be achieved through the following approach:

[0007] The automatic tension control system for chain conveyor belts used in dump semi-trailers includes a multi-dimensional sensor data acquisition and analysis module, used to acquire multi-source sensing data of the operating conditions of the chain conveyor belt mounted on the dump semi-trailer, and analyze the multi-source sensing data of the operating conditions to extract real-time chain tension values, current load material weight values, real-time lifting angle values ​​of the trailer, and real-time operating environment temperature values; a multi-parameter state feature construction module, used to perform compensation calculations on the real-time chain tension values ​​and the current load material weight values ​​based on the real-time lifting angle values ​​of the trailer and the real-time operating environment temperature values, and construct a dynamic relaxation state feature matrix; a spatiotemporal sensing model invocation and instruction generation module, used to invoke a preset temporal state tension evaluation model, input the dynamic relaxation state feature matrix into the preset temporal state tension evaluation model, and generate a target tension adjustment instruction; and an anti-resonance instruction execution and drive module, used to execute the target tension adjustment instruction, outputting a tension compensation drive signal with damping buffer constraints to the system tension execution mechanism to change the chain tension of the chain conveyor belt.

[0008] Optionally, the multi-dimensional sensing data acquisition and analysis module includes: a multi-source sensing signal receiving unit, used to receive the original multi-modal analog sensing electrical signals transmitted in real time by the monitoring node array deployed on the chain conveyor belt; a signal filtering and conversion unit, used to sequentially perform photoelectric isolation and adaptive Kalman filtering noise reduction operations on the original multi-modal analog sensing electrical signals to generate a standard digital sensing signal sequence as the multi-source sensing data of the operating conditions; and a multi-modal data interpretation unit, used to interpret the data packet header identifier of the standard digital sensing signal sequence, and separate the real-time tension value of the chain representing the tensile physical state, the weight value of the currently carried material representing the gravity load state, the real-time lifting tilt angle value of the carriage representing the spatial attitude, and the real-time temperature value of the operating environment representing the thermodynamic state.

[0009] Optionally, the multi-parameter state feature construction module includes: an inclination gravity component compensation unit, used to perform trigonometric function decomposition calculations by combining the current weight value of the carried material with the real-time lifting inclination angle value of the carriage, to generate an axial tensile force compensation value under the action of the inclination angle; a thermal expansion and contraction deformation compensation unit, used to extract the metal thermal expansion coefficient of the chain contained in the chain conveyor belt, and generate a temperature-induced deformation compensation value based on the difference between the real-time temperature value of the operating environment and the preset standard reference temperature value and the metal thermal expansion coefficient; and a feature matrix fusion and splicing unit, used to fuse the axial tensile force compensation value and the temperature-induced deformation compensation value into the real-time tension value of the chain, generate a corrected true tension value, and splice the corrected true tension value, the current weight value of the carried material, and the real-time lifting inclination angle value of the carriage according to a preset time-series data dimension to construct the dynamic relaxation state feature matrix.

[0010] Optionally, the spatiotemporal perception model invocation and instruction generation module includes: a network structure loading unit, used to load a long short-term memory neural network structure containing an input feature gating layer, a historical state memory layer, and a nonlinear regression decision layer as the preset temporal state tension evaluation model; a temporal feature forward propagation unit, used to load the dynamic relaxation state feature matrix into the input feature gating layer, and fuse the forgetting control parameters of the previous sampling period through the historical state memory layer to output a deep hidden response feature with temporal dependence; and a regression mapping calculation unit, used to perform regression mapping calculation on the deep hidden response feature through the nonlinear regression decision layer to generate the target tension adjustment instruction containing the target compensation displacement value and the baseline execution speed value.

[0011] Optionally, the temporal feature forward propagation unit includes: a parameter extraction subunit, used to extract the fixed-configuration forget gate weight parameter matrix, input gate weight parameter matrix, and state bias constant vector in the historical state memory layer; a gated activation calculation subunit, used to perform matrix multiplication on the dynamic relaxed state feature matrix based on the forget gate weight parameter matrix and the input gate weight parameter matrix respectively, and process the matrix multiplication result through a preset logical nonlinear function to generate an information retention ratio matrix and an information update ratio matrix; and a memory state update subunit, used to perform Hadamard product operation on the information retention ratio matrix and the information update ratio matrix to generate a product update result, and sum and fuse the product update result with the state bias constant vector to output the deep hidden response feature.

[0012] Optionally, the anti-resonance command execution and drive module includes: a command value extraction unit, used to parse the target tension adjustment command and extract the target compensation displacement value and the reference execution speed value contained in the target tension adjustment command; a damping control parameter matching unit, used to extract the mechanical resonance frequency spectrum pre-stored in the control system and compare the reference execution speed value with the mechanical resonance frequency spectrum to generate damping speed limiting control parameters for avoiding the mechanical resonance range; and a buffer drive signal sending unit, used to generate the tension compensation drive signal with damping buffer constraint based on the target compensation displacement value and the damping speed limiting control parameters, and send the tension compensation drive signal with damping buffer constraint to the power controller of the system tension actuator to drive the system tension actuator to extend or retract the target compensation displacement value.

[0013] Optionally, the system further includes: a post-adjustment state retest module, used to reacquire the actual tension value of the chain conveyor belt after the tensioning actuator of the system executes the tension compensation drive signal with damping buffer constraint to complete the chain tension change action; a dynamic benchmark comparison module, used to extract the dynamic safety tension lower limit value corresponding to the current real-time lifting angle value of the carriage, and compare the actual tension value after adjustment with the dynamic safety tension lower limit value; and a forced intervention trigger module, used to generate a secondary forced intervention tensioning trigger event when the actual tension value after adjustment is less than the dynamic safety tension lower limit value.

[0014] Optionally, the system further includes: a residual deviation calculation module, used to calculate the residual deviation between the dynamic safety tension lower limit value and the adjusted actual tension value based on the secondary forced intervention tension trigger event; a feedback sample construction module, used to fuse the residual deviation value with the dynamic relaxation state feature matrix before the occurrence of the secondary forced intervention tension trigger event to construct a feedback training sample with self-supervised labels; and a model parameter correction module, used to input the feedback training sample with self-supervised labels into the preset time-series tension evaluation model to correct the network node weight parameter matrix in the preset time-series tension evaluation model.

[0015] Optionally, the system further includes: a service status extraction module, used to extract the cumulative tensile action frequency value of the chain conveyor belt from its initial commissioning to the current time; an aging and wear determination module, used to calculate the chain metal fatigue stress integral value by combining the cumulative tensile action frequency value with a preset tension impact attenuation coefficient, and determine whether the chain metal fatigue stress integral value is greater than a preset physical material yield warning limit value, and at the same time determine whether the residual deviation value continuously exceeds a preset safety redundancy error limit; and an extreme wear warning module, used to construct a chain extreme wear warning data package when the chain metal fatigue stress integral value is greater than the preset physical material yield warning limit value, and the residual deviation value continuously exceeds the preset safety redundancy error limit, and convert the chain extreme wear warning data package into a host alarm indication signal for output.

[0016] Based on the same inventive concept, this invention also provides an automatic tension control method for chain conveyor belts used in dump semi-trailers. The method includes: acquiring multi-source sensing data of the operating conditions of the chain conveyor belt mounted on the dump semi-trailer, and analyzing the multi-source sensing data to extract real-time chain tension values, current load material weight values, real-time lifting angle values ​​of the trailer, and real-time ambient temperature values; performing compensation calculations on the real-time chain tension values ​​and the current load material weight values ​​based on the real-time lifting angle values ​​of the trailer and the real-time ambient temperature values ​​to construct a dynamic relaxation state feature matrix; calling a preset time-series tension evaluation model and inputting the dynamic relaxation state feature matrix into the preset time-series tension evaluation model to generate a target tension adjustment command; executing the target tension adjustment command and outputting a tension compensation drive signal with damping buffer constraints to the system tension actuator to change the chain tension of the chain conveyor belt.

[0017] Compared with the prior art, the present invention has the following advantages:

[0018] 1. Improved accuracy in chain tension assessment and adjustment under complex working conditions. By analyzing multi-source sensing data such as real-time chain tension, weight of the load, real-time lifting angle of the carriage, and real-time temperature of the operating environment, in-depth compensation calculations can be performed for the gravity component of the tilt angle and thermal expansion and contraction deformation. This restores the true stress state of the chain during dynamic operation and overcomes the errors caused by single-dimensional monitoring.

[0019] 2. Intelligent tension decision-making based on historical temporal memory was achieved. A long short-term memory neural network structure was introduced as the temporal state tension evaluation model, which can uncover the temporal dependencies in the dynamic relaxation state feature matrix. By combining the residual deviation values ​​between the actual tension and the safety lower limit, it can also autonomously construct feedback training samples to correct the model parameters, endowing the equipment with the ability of continuous self-supervised learning and adaptive optimization.

[0020] 3. Enhanced mechanical stability and safety during the tensioning process. During command execution and drive phases, by comparing the baseline execution speed value with the mechanical resonance frequency spectrum, damped speed-limiting control parameters are specifically generated to avoid the mechanical resonance range. The output drive signal with damped buffer constraints not only changes the chain tension but also avoids the risk of mechanical structure damage due to resonance during tensioning.

[0021] 4. A chain health status monitoring and early warning mechanism is provided throughout its entire lifecycle. The cumulative frequency of tensile actions is extracted in real time and combined with the tension impact attenuation coefficient to quantify the chain's metal fatigue stress integral value. When the fatigue stress integral exceeds the material yield warning limit and the error continues to exceed the limit, a wear warning is automatically triggered, reducing the chain breakage rate.

[0022] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the automatic tensioning control system for the chain conveyor belt of a self-unloading semi-trailer according to an embodiment of the present invention.

[0025] Figure 2 This is a schematic diagram of the dynamic slack state compensation model of the automatic tensioning control system and method for chain conveyor belts used in dump semi-trailers according to an embodiment of the present invention.

[0026] Figure 3 This is a schematic diagram of the chain fatigue wear life cycle evolution model of the automatic tensioning control system and method for chain conveyor belts used in dump semi-trailers according to an embodiment of the present invention.

[0027] Figure 4 This is a flowchart illustrating the automatic tension control method for the chain conveyor belt of a self-unloading semi-trailer according to an embodiment of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0029] Reference Figure 1 One embodiment of the present invention proposes an automatic tension control system for chain conveyor belts used in dump semi-trailers. The system uses multi-dimensional sensor data analysis and a time-series state tension evaluation model to generate target commands, and is driven by damping buffer constraints. This enables intelligent, precise and adaptive adjustment of chain tension under complex working conditions.

[0030] The system described in this embodiment specifically includes:

[0031] The multi-dimensional sensing data acquisition and analysis module is used to acquire multi-source sensing data of the operating conditions of the chain conveyor belt mounted on the dump semi-trailer, and to analyze the multi-source sensing data of the operating conditions to extract the real-time tension value of the chain, the weight value of the current material being carried, the real-time lifting angle value of the truck body, and the real-time temperature value of the operating environment.

[0032] The multi-parameter state feature construction module is used to perform compensation calculations on the real-time tension value of the chain and the weight value of the currently carried material based on the real-time lifting angle value of the carriage and the real-time temperature value of the operating environment, and to construct a dynamic relaxation state feature matrix.

[0033] The spatiotemporal perception model invocation and instruction generation module is used to invoke a preset temporal state tension evaluation model and input the dynamic relaxation state feature matrix into the preset temporal state tension evaluation model to generate a target tension adjustment instruction.

[0034] The anti-resonance command execution and drive module is used to execute the target tension adjustment command and output a tension compensation drive signal with damping buffer constraint to the system tension actuator to change the chain tension of the chain conveyor belt.

[0035] Specifically, the multi-dimensional sensor data acquisition and analysis module is responsible for acquiring operating condition data and extracting four key values: real-time chain tension, weight of the load material, lifting angle of the trailer, and ambient temperature. The multi-parameter state feature construction module uses the tilt angle and temperature values ​​to compensate for the tension and weight values, thereby constructing a dynamic relaxation state feature matrix. The spatiotemporal perception model invocation and command generation module inputs this feature matrix into a preset temporal state tension evaluation model to calculate and generate the target tension adjustment command. The anti-resonance command execution and drive module executes this command, outputting a compensated drive signal with damped buffer constraints to the tension actuator to complete the physical adjustment of the chain tension. This achieves high-precision, intelligent, and stable dynamic tension control of the dump semi-trailer conveyor chain. By extracting tension, weight, tilt angle, and temperature data from multiple dimensions and performing compensation calculations, environmental and operational interference can be eliminated, accurately restoring the true dynamic slack state of the chain and improving the comprehensiveness and accuracy of the control system's data perception. The generation of adjustment commands using a time-series state evaluation model endows the system with intelligent decision-making capabilities. Outputting drive signals with damping and buffer constraints during the execution phase not only ensures the execution of tensioning actions but also prevents destructive mechanical resonance and impact during adjustment, thus enhancing the safety and stability of the entire conveyor belt system.

[0036] Optionally, the multidimensional sensing data acquisition and analysis module includes:

[0037] A multi-source sensing signal receiving unit is used to receive the raw multimodal analog sensing electrical signals transmitted in real time by the monitoring node array deployed on the chain conveyor belt;

[0038] The signal filtering and conversion unit is used to sequentially perform photoelectric isolation and adaptive Kalman filtering noise reduction operations on the original multimodal analog sensing electrical signal to generate a standard digital sensing signal sequence as the multi-source sensing data of the operating conditions;

[0039] The multimodal data interpretation unit is used to interpret the header identifier of the data packet of the standard digital sensing signal sequence, and separate the real-time tension value of the chain representing the tensile physical state, the weight value of the currently carried material representing the gravity load state, the real-time lifting tilt angle value of the carriage representing the spatial attitude, and the real-time temperature value of the operating environment representing the thermodynamic state.

[0040] Specifically, the control board's underlying data acquisition interface receives in real-time raw multimodal analog sensing electrical signals transmitted from the monitoring node array deployed at key stress nodes and environmental monitoring points on the chain conveyor belt. The hardware protection module performs opto-isolation on the raw multimodal analog sensing electrical signals to physically block external common-mode interference and surge impacts. The microprocessor calls an adaptive Kalman filter noise reduction program to process the isolated signals, setting the state transition matrix and observation matrix, updating the prior state estimate and error covariance in real-time, filtering out high-frequency jitter noise by minimizing the mean square error, and outputting a discretized standard digital sensing signal sequence as multi-source sensing data for the operating conditions.

[0041] ;

[0042] ;

[0043] ;

[0044] Let A be the prior error covariance, and A be the state transition matrix. Let Q be the covariance of the estimation error at the previous time step, and let Q be the process noise covariance. R is the Kalman gain, and R is the measurement noise covariance. This is the optimal state estimation vector at the current moment. This is the optimal state estimation vector from the previous time step. This is the measurement vector at the current moment.

[0045] The data processing program parses the header identifier of the standard digital sensing signal sequence. By consulting a pre-set communication protocol identifier mapping dictionary, it separates the real-time chain tension value (representing the tensile physical state), the current weight of the load-bearing material (representing the gravity load state), the real-time lifting tilt angle of the carriage (representing the spatial attitude), and the real-time temperature of the operating environment (representing the thermodynamic state) based on different hexadecimal identifier headers. The monitoring node array is a monitoring network consisting of various acquisition terminals, such as tension sensors and tilt sensors, installed in a distributed topology on the conveyor belt structure. The original multimodal analog sensing electrical signal is a continuous high-low level waveform that is not digitized and mixes mechanical attitude and thermodynamic characteristics. Opto-isolation is a safety protection technology that uses optocouplers to convert electrical signals into optical signals and then back into electrical signals to completely block the electrical circuit. Adaptive Kalman filtering noise reduction is a mathematical statistical algorithm that dynamically adjusts the noise covariance matrix based on the measurement residual to recursively filter out random interference signals under optimal criteria. The data packet header identifier is a special encoded string attached to the beginning of the data frame to indicate the parameter type and data byte length of the current data frame.

[0046] For example, the multi-source sensing signal receiving unit receives the raw multimodal analog sensing electrical signal output by the tension sensor in real time. After entering the signal filtering and conversion unit, opto-isolation is achieved by the optocoupler chip. The microprocessor performs adaptive Kalman filtering noise reduction, presetting the process noise covariance Q to 0.05 and the measurement noise covariance R to 0.1 based on the inherent physical thermal noise characteristics of the tension sensor as calibrated at the factory. Since the tension signal is a one-dimensional feature and maintains linear inertial change within continuous time steps, the state transition matrix A is set to 1. At time k, the microprocessor reads the optimal tension state estimation vector from the previous time step. The value is 5.0. Read the covariance of the estimation error from the previous time step. The value is 0.2, receiving the measurement vector transmitted back by the sensor at the current moment. The value is 5.4. The microprocessor calculates the prior error covariance. Calculate the Kalman gain. Calculate the prior state estimate: 1 × 5.0 = 5.0. Calculate the optimal state estimate vector at the current time step. The value of 5.2856 is the actual tension value after filtering out high-frequency mechanical vibration noise, and it is encoded into a standard digital sensing signal sequence. The multimodal data interpretation unit reads this sequence, identifies the header identifier 0x01 of the first data packet, and finds out from the communication protocol dictionary that 0x01 represents tension data. Thus, it accurately separates the real-time tension value of the chain, 5.2856, which represents the tensile physical state, the weight of the current load material, 30.5, which represents the gravity load state, the real-time lifting angle of the carriage, 45.0, which represents the spatial attitude, and the real-time temperature of the operating environment, 25.5, which represents the thermodynamic state, from the header identifier 0x02.

[0047] Optionally, the multi-parameter state feature construction module includes:

[0048] The tilt angle gravity component compensation unit is used to perform trigonometric function decomposition calculation by combining the current weight value of the loaded material with the real-time lifting tilt angle value of the carriage to generate the axial tensile force compensation value under the tilt angle.

[0049] The thermal expansion and contraction deformation compensation unit is used to extract the metal thermal expansion coefficient of the chain included in the chain conveyor belt, and generate temperature-induced deformation compensation values ​​based on the difference between the real-time temperature value of the operating environment and the preset standard reference temperature value and the metal thermal expansion coefficient.

[0050] The feature matrix fusion and splicing unit is used to fuse the axial tensile force compensation value and the temperature-induced deformation compensation value into the real-time tension value of the chain, generate the corrected real tension value, and splice the corrected real tension value, the current load material weight value and the real-time lifting angle value of the carriage according to the preset time-series data dimension to construct the dynamic relaxation state feature matrix.

[0051] Specifically, the data processing center extracts the current weight of the material being carried and the real-time lifting angle of the conveyor belt. It converts the real-time lifting angle into radians, inputs it into a sine trigonometric function to calculate the projection ratio, and multiplies it by the current weight of the material being carried to calculate the axial tensile force compensation value under the tilt angle. The data processing center reads the pre-configured metal thermal expansion coefficient and standard reference temperature values ​​of the chain constituting the conveyor belt from its built-in memory. It calculates the temperature difference between the real-time operating temperature and the standard reference temperature, multiplies this temperature difference by the metal thermal expansion coefficient and the preset equivalent tensile stiffness coefficient of the chain, and obtains the temperature-induced deformation compensation value. The data processing center uses a linear superposition algorithm to fuse the axial tensile force compensation value and the temperature-induced deformation compensation value as compensation terms into the real-time chain tension value, generating a corrected true tension value that eliminates interference from gravity and thermal deformation. The data processing center establishes a two-dimensional array structure. According to the time sequence of the system sampling cycle, the corrected real tension value of the current and historical cycles, the current weight value of the material being carried, and the real-time lifting angle value of the carriage are concatenated column by column to generate a dynamic relaxation state feature matrix for subsequent time-series network input.

[0052] ;

[0053] ;

[0054] ;

[0055] Here, W represents the axial tensile force compensation value, and W represents the current weight of the material being carried. This provides the real-time lifting and tilting angle value of the carriage. Pi is a constant. This represents the temperature-induced deformation compensation value. This is the equivalent tensile stiffness coefficient of the chain. The coefficient of thermal expansion of the metal is 1. This refers to the real-time temperature value of the operating environment. This is the standard reference temperature value. To correct the actual tension value, This represents the real-time tension value of the chain. For example... Figure 2The diagram illustrates the chain dynamic slack state compensation model. The left figure shows the nonlinear sinusoidal mapping relationship between the axial tension compensation value and the car lifting tilt angle; the right figure shows the linear mapping relationship between the temperature-induced deformation compensation value and the operating environment temperature. The figures highlight the compensation feature points at a tilt angle of 45 degrees and a temperature of 25.5 degrees Celsius in a specific embodiment, visually verifying the mathematical process of multi-parameter feature construction and compensation. The axial tension compensation value is the additional tensile force component generated by the weight of the load-bearing material along the chain's movement direction under the tilted state of the car, which needs to be removed from the total sensor readings. The coefficient of thermal expansion of the metal is a physical property constant characterizing the relative change in length of a specific metal material when the temperature increases by one degree Celsius. The temperature-induced deformation compensation value is the equivalent tension fluctuation caused by the chain's thermal expansion and contraction due to changes in ambient temperature. The dynamic slack state feature matrix is ​​a two-dimensional data set that aggregates multi-dimensional compensated physical parameters according to a time series and can completely map the slack evolution law of the system within a continuous time segment.

[0056] For example, given that the current weight of the material being carried, W, is 30.5 kN, the real-time lifting angle of the carriage, theta, is 45.0 degrees, the real-time temperature of the operating environment, Tr, is 25.5 degrees Celsius, and the real-time chain tension, Freal, is 5.2856 kN, the constant pi is used to convert the angle into radians that can be calculated using a standard mathematical library, based on the theory of mathematical constants. The value is 3.14159. The data processing center substitutes the values ​​into the formula to calculate the axial tensile force compensation value under the tilt angle: Finc = 30.5 × sin[45.0 × 3.14159 / 180] = 30.5 × 0.7071 = 21.5665 kN. Based on the material properties of carbon steel and the thermodynamic theory of materials, the coefficient of thermal expansion of the metal is preset. The value is 0.000012. The standard reference temperature Ts is set to 20.0 degrees Celsius according to the industrial equipment factory calibration specifications. Based on the mechanical principle of the product of chain cross-sectional area and Young's modulus, the equivalent tensile stiffness coefficient K of the chain is set to 60000 kN. The data processing center uses the formula to calculate the temperature-induced deformation compensation value Ftem = 60000 × 0.000012 × [25.5 - 20.0] = 0.72 × 5.5 = 3.9600 kN. The corrected true tension value is calculated based on the force balance superposition algorithm. The calculation is based on the fact that the gravity slip component is additionally superimposed on the sensor reading, so it needs to be subtracted. Thermal expansion will cause the tension reading to be falsely low, so temperature compensation needs to be added. Both need to be restored to the reference state. Substituting into the formula yields... In this context, a negative value (kN) physically represents the current chain losing its basic preload and being in an extremely relaxed state. The data processing center retrieves the corrected true tension value of -10.5000 kN, the weight value of 30.5 kN, and the tilt angle value of 40.0 degrees stored in the previous sampling period. This value is then merged with the current period's values ​​column-wise to generate a dynamic relaxation state feature matrix Mfeat. The first column of the matrix contains data of -10.5000, 30.5, and 40.0, while the second column contains data of -12.3209, 30.5, and 45.0. This constructs a complete multidimensional array containing spatiotemporal features for subsequent neural network calls.

[0057] Optionally, the spatiotemporal perception model invocation and instruction generation module includes:

[0058] The network structure loading unit is used to load a long short-term memory neural network structure containing an input feature gating layer, a historical state memory layer, and a nonlinear regression decision layer as the preset temporal state tension evaluation model.

[0059] The temporal feature forward propagation unit is used to load the dynamic relaxed state feature matrix into the input feature gating layer, and fuse the forgetting control parameters of the previous sampling period through the historical state memory layer to output a deep hidden response feature with temporal dependence.

[0060] The regression mapping calculation unit is used to perform regression mapping calculation on the deep hidden response features through the nonlinear regression decision layer to generate the target tension adjustment command containing the target compensation displacement value and the benchmark execution speed value.

[0061] Specifically, the central processing unit (CPU) calls and loads a pre-trained and solidified long short-term memory (LSTM) neural network structure, containing an input feature gating layer, a historical state memory layer, and a nonlinear regression decision layer, from the storage medium via the bus interface as a preset temporal state tension evaluation model. In the temporal feature forward propagation unit, the CPU flattens the dynamic relaxation state feature matrix and inputs it to the input feature gating layer, extracts the hidden state parameters from the previous sampling period as forgetting control weights, and deeply fuses the multi-dimensional states through weighted summation and nonlinear function activation processing of the historical state memory layer, calculating and outputting deep hidden response features with temporal dependence. The regression mapping calculation unit receives the deep hidden response features, performs regression mapping calculations through the fully connected network of the nonlinear regression decision layer, substitutes the displacement mapping weights and velocity mapping weights respectively, and generates a target tension adjustment command containing the target compensation displacement value and the baseline execution velocity value.

[0062] ;

[0063] ;

[0064] ;

[0065] For deep hidden response features, For the feature fusion weights of the hidden layer of the network, This refers to the data of each state element within the dynamic relaxation state feature matrix. This is the hidden layer bias constant. To compensate for the displacement value of the target. These are the weighting coefficients for displacement regression mapping. This is the displacement regression bias constant. The baseline execution speed value, For velocity regression mapping weight coefficients, For velocity regression bias constant, The activation function is hyperbolic tangent. The Long Short-Term Memory (LSTM) neural network structure is a deep learning mathematical model with a gating mechanism that selectively memorizes and forgets long-sequence time data and prevents gradient vanishing. The input feature gating layer is the initial network topology responsible for receiving external multidimensional state data and controlling the openness of the data flow into the memory cell channels. The historical state memory layer is a set of hidden computation nodes within the recurrent network used to store historical evolution features of previous time nodes and transmit global state parameters across cycles. The nonlinear regression decision layer is the terminal computation network that maps high-dimensional abstract feature vectors through a fully connected matrix and outputs specific control parameters in a continuous space. The deep hidden response features are abstract high-dimensional values ​​representing the chain relaxation evolution trend extracted after multi-layer network gating and nonlinear activation. The target compensation displacement value is the absolute length physical reference base calculated by the control system to require the actuator to extend or retract to offset the current chain relaxation. The reference execution speed value is the reference quantity of the basic operating rate of the actuator calculated and planned to ensure a smooth transition of the mechanical structure.

[0066] For example, the central processing unit extracts six state elements from the matrix: -10.5000, 30.5, 40.0, -12.3209, 30.5, and 45.0. Based on the parameters solidified during the model deployment phase training, the hidden layer feature fusion weight Wh is set to 0.01, and the hidden layer bias constant Bh is set to 0.5. The central processing unit performs a matrix element summation operation, calculating the sum as -10.5000 + 30.5 + 40.0 - 12.3209 + 30.5 + 45.0 = 123.1791. This summation is then substituted into the forward propagation formula to calculate the deep hidden response features. The central processing unit (CPU) calculates the current deep hidden response feature Hout to be 0.9400 by consulting its internal mathematical library. Entering the regression mapping calculation stage, based on mechanical design safety specifications and the convergence results of the deep learning model's regression training, the displacement regression mapping weight coefficient Wd for generating displacement adjustment is set to 20.0, and the displacement regression bias constant Bd is set to 10.0; the velocity regression mapping weight coefficient Ws for matching the hydraulic tensioning mechanism's motion characteristics is set to 5.0, and the velocity regression bias constant Bs is set to 2.0. The CPU substitutes these values ​​into the formula to calculate the target compensation displacement value Dtarget = 20.0 × 0.9400 + 10.0 = 18.8000 + 10.0 = 28.8000 mm. It then substitutes these values ​​into the formula to calculate the baseline execution speed value Sbase = 5.0 × 0.9400 + 2.0 = 4.7000 + 2.0 = 6.7000 mm / s. The final output generates a target tension adjustment command containing a displacement of 28.8000 mm and a speed of 6.7000 mm per second, providing a precise and stable control reference for subsequent mechanism execution.

[0067] Optionally, the temporal feature forward propagation unit includes:

[0068] The parameter extraction subunit is used to extract the fixed-configuration forget gate weight parameter matrix, input gate weight parameter matrix and state bias constant vector in the historical state memory layer;

[0069] The gated activation calculation subunit is used to perform matrix multiplication on the dynamic relaxation state feature matrix based on the forget gate weight parameter matrix and the input gate weight parameter matrix, respectively, and process the matrix multiplication result through a preset logical nonlinear function to generate an information retention ratio matrix and an information update ratio matrix.

[0070] The memory state update subunit is used to perform a Hadamard product operation on the information retention ratio matrix and the information update ratio matrix to generate a product update result, and then sum and fuse the product update result with the state bias constant vector to output the deep hidden response feature.

[0071] Specifically, the parameter extraction subunit embedded in the central processing unit accesses non-volatile memory, calls and extracts the fixed-configuration forget gate weight parameter matrix, input gate weight parameter matrix, and state bias constant vector from the historical state memory layer. Entering the gating activation calculation subunit, the central processing unit utilizes the underlying tensor operation core to perform matrix multiplication on the dynamic relaxation state feature matrix based on the forget gate weight parameter matrix and the input gate weight parameter matrix, extracting the linear combination mapping value of historical residual information and current input information. A preset logical nonlinear function is invoked to activate the matrix multiplication result, compressing and restricting the unbounded linear combination mapping value to a specific range, thereby independently generating the information retention ratio matrix and the information update ratio matrix. The memory state update subunit performs a Hadamard product operation on the information retention ratio matrix and the information update ratio matrix, extracting the common activation features to generate a product update result. This product update result is then summed and fused with the state bias constant vector to shift and compensate for the feature space distribution, ultimately outputting the deep hidden response features for subsequent regression modules.

[0072] ;

[0073] ;

[0074] ;

[0075] ;

[0076] To preserve the proportion matrix of information, The base of the natural logarithm is a constant. This is the forget gate weight parameter matrix. This is the matrix dot product operator. The feature matrix of the dynamic relaxation state. For information update ratio matrix, The input gate weight parameter matrix, To update the product result, This is the Hadamard product operator. For deep hidden response features, The vector represents the state bias constants. The forget gate weight parameter matrix is ​​a set of neural network training parameters that controls the proportion of historical state information discarded to filter out invalid time series interference. The input gate weight parameter matrix is ​​a set of neural network training parameters that controls the proportion of current input information written into the memory unit. The state bias constant vector is a set of fixed compensation values ​​used to adjust the activation function trigger threshold to prevent the network output from getting stuck in a dead zone. Matrix dot product is a linear algebraic calculation method that multiplies and sums corresponding elements of two matrices of the same dimension to achieve feature projection from high-dimensional space to low-dimensional space. The logistic nonlinear function is a mathematical activation tool that smoothly maps unbounded real-number field inputs to a closed interval with a probability of zero to one. The information retention ratio matrix is ​​a set of gating coefficients that, after processing by the logistic nonlinear function, characterizes the degree to which historical data is effectively retained at the current time step. The information update ratio matrix is ​​a set of gating coefficients that, after processing by the logistic nonlinear function, characterizes the degree to which currently sensed data is allowed to be written into the memory layer. The Hadamard product is a matrix operation method that directly multiplies corresponding elements of two matrices of the same dimension to extract the feature intersection. The product update result is the combined gated state variable obtained by multiplying the retention ratio and the update ratio using the Hadamard product.

[0077] For example, to facilitate vector operations, the dynamic relaxation state feature matrix is ​​expanded sequentially into a single-dimensional feature sequence containing six state elements: -10.5000, 30.5, 40.0, -12.3209, 30.5, and 45.0. The central processing unit extracts the forgetting gate weight parameter matrix, focusing on the tension difference, based on the model's fixed parameters. Given the sequence [0.1, 0, 0, -0.1, 0, 0], we extract the input gate weight parameter matrix, which focuses on the weight characteristics of the material being carried. Given the sequence [0, 0.1, 0, 0, 0.1, 0]. The central processing unit performs matrix multiplication, calculating the forget gate multiplication result as 0.1×(-10.5000)+(-0.1)×(-12.3209)=-1.0500+1.2321=0.1821. The input gate multiplication result is calculated as 0.1×30.5+0.1×30.5=3.0500+3.0500=6.1000. The sigmoid activation function is used to process the above multiplication result. This activation function is chosen because it can smoothly map any real number to the probability interval of zero to one, which fits the physical definition of the information throughput ratio in the gating structure. The natural constant e is taken as 2.718 in the calculation. Substituting into the formula, the information retention ratio matrix is ​​calculated. Substitute the values ​​into the formula to calculate the information update ratio matrix. To achieve the logical intersection and fusion of historical memory retention and current input update, a Hadamard product operation is performed on both, and the product update result Mproduct = 0.5454 × 0.9978 = 0.5442 is calculated using the formula. Based on engineering experience in preventing neuron dormancy in deep learning networks, a pre-set state bias constant vector Bc is extracted as 0.3958, and a summation and fusion operation is performed to calculate the deep hidden response feature Hout = 0.5442 + 0.3958 = 0.9400.

[0078] Optionally, the anti-resonance command execution and driving module includes:

[0079] The instruction value extraction unit is used to parse the target tension adjustment instruction and extract the target compensation displacement value and the reference execution speed value contained in the target tension adjustment instruction;

[0080] The damping control parameter matching unit is used to extract the mechanical resonance frequency spectrum pre-stored in the control system, compare the reference execution speed value with the mechanical resonance frequency spectrum, and generate damping speed limiting control parameters to avoid the mechanical resonance range.

[0081] The buffer drive signal sending unit is used to generate the tension compensation drive signal with damping buffer constraint based on the target compensation displacement value and the damping speed limiting control parameter, and send the tension compensation drive signal with damping buffer constraint to the power controller of the system tension actuator to drive the system tension actuator to extend or retract the target compensation displacement value.

[0082] Specifically, the command value extraction unit within the anti-resonance command execution and drive module parses the target tension adjustment command, extracting the target compensation displacement value and the reference execution speed value contained within it. The damping control parameter matching unit extracts the pre-stored mechanical resonance frequency spectrum of the control system, converts the reference execution speed value into the execution fundamental frequency by combining it with the pitch of the physical structure, and compares the execution fundamental frequency with the resonance dead zone in the mechanical resonance frequency spectrum. Based on this comparison result, damping speed limiting control parameters are generated to avoid the mechanical resonance range. The buffer drive signal sending unit, based on the target compensation displacement value and the damping speed limiting control parameters, generates a tension compensation drive signal with damping buffer constraints through a preset drive conversion algorithm, and sends this damping buffer constraint-constrained tension compensation drive signal to the power controller of the system tensioning actuator to drive the system tensioning actuator to accurately extend and retract the target compensation displacement value.

[0083] ;

[0084] ;

[0085] ;

[0086] To execute the base frequency, The baseline execution speed value, For chain pitch, For damping speed limiting control parameters This is the frequency reduction compensation coefficient. This refers to the control voltage corresponding to the tension compensation drive signal with damping buffer constraint. To drive the conversion coefficient, The target compensation displacement value is defined as follows: The mechanical resonance frequency spectrum is a set of characteristic data recording the dangerous frequency ranges where various mechanical components of the equipment are prone to severe amplitude amplification at different operating speeds. The damping speed limiting control parameter is the final value of the speed adjustment to avoid resonance caused by the excitation frequency coinciding with the natural frequency during tension adjustment, by sacrificing some response speed to apply a buffer constraint. The tension compensation drive signal with damping buffer constraint is the final electrical control voltage that integrates the target displacement requirement and the anti-resonance speed limiting condition, enabling smooth system movement without causing impact oscillations. The system tensioning actuator is a physical actuation device that receives electrical energy from the power controller and converts it into linear mechanical displacement to directly change the chain tension. The power controller is the hardware control center that receives weak electrical signals and amplifies their power to drive the strong electrical actuator. The execution fundamental frequency is the fundamental frequency of periodic mechanical vibration generated by the actuator moving at a specific speed under the action of the chain pitch physical interval. The frequency reduction compensation coefficient is a proportional attenuation constant used to forcibly reduce the target speed to escape the dangerous frequency range when the basic operating parameters fall into the resonance frequency band.

[0087] For example, the instruction value extraction unit parses the target tension adjustment instruction, extracts the target compensation displacement value Dtarget as 28.8000 mm, and the base execution speed value Sbase as 6.7000 mm / s. Based on the factory physical parameters of the semi-trailer chassis chain, the chain pitch Lpitch is set to 2.0 mm. The damping control parameter matching unit substitutes the physical conversion formula to calculate the execution fundamental frequency fexec = 6.7000 / 2.0 = 3.3500 Hz. The pre-stored mechanical resonance frequency spectrum of the control system is extracted. Based on the finite element analysis test report of the vibration modes under alternating empty and fully loaded conditions of the cargo box, the mechanical main resonance dead zone is determined to be 3.0 Hz to 4.0 Hz. Comparing the frequency corresponding to the base execution speed value, it is found that the currently calculated 3.3500 Hz falls within the mechanical resonance dead zone. To generate damping speed-limiting control parameters that avoid the mechanical resonance range and trigger the frequency reduction compensation mechanism, based on the safety criterion in vibration theory that the excitation frequency must be reduced by at least 20% to avoid the main resonance peak, the frequency reduction compensation coefficient Kdamp is set to 0.8. Substituting these values ​​into the formula, the damping speed-limiting control parameter Cdamp = 6.7000 × 0.8 = 5.3600 mm / s is calculated. The buffer drive signal sending unit, based on the flow displacement to control voltage mapping ratio provided in the hydraulic power controller's manufacturer's manual, sets the drive conversion coefficient Kpwm to 0.5 volts per second per square millimeter. The calculated target compensation displacement value and damping speed-limiting control parameter are extracted and substituted into the formula to calculate the control voltage Udrive corresponding to the tension compensation drive signal: Udrive = 0.5 × 5.3600 × 28.8000 = 77.1840 volts. Finally, the buffer drive signal sending unit sends this tension compensation drive signal with damping buffer constraint to the power controller of the system tension actuator, driving the system tension actuator to precisely extend and retract the target compensation displacement value, i.e., 28.8000 mm, in a safety damping state that matches the speed limit.

[0088] Optionally, the system further includes:

[0089] The adjusted state retest module is used to reacquire the actual tension value of the chain conveyor belt after the tensioning actuator of the system executes the tension compensation drive signal with damping buffer constraint to complete the chain tension change action;

[0090] The dynamic benchmark comparison module is used to extract the dynamic safety tension lower limit value corresponding to the current real-time lifting angle value of the carriage, and compare the adjusted actual tension value with the dynamic safety tension lower limit value.

[0091] The forced intervention trigger module is used to generate a secondary forced intervention tension trigger event when the actual tension value after adjustment is less than the lower limit value of the dynamic safety tension.

[0092] Specifically, after the tensioning actuator completes the displacement compensation action for the expansion target, the post-adjustment state retesting module uses the underlying communication interface to query the sensor network again to re-acquire the actual tension value of the chain conveyor belt in a stable state after adjustment. Subsequently, the dynamic benchmark comparison module uses internally stored safety mapping rules to extract the dynamic safety tension lower limit value corresponding to the current real-time lifting angle value of the carriage, and compares the actual tension value after adjustment with the dynamic safety tension lower limit value. The forced intervention triggering module monitors the comparison results in real time. When the actual tension value after adjustment is less than the dynamic safety tension lower limit value, it determines that there is a residual slack risk in the system and triggers the underlying protection logic, automatically generating a secondary forced intervention tensioning trigger event.

[0093] ;

[0094] ;

[0095] This is the lower limit value of dynamic safety tension. For translational reference tension constant, This is the tilt tension sensitivity coefficient. This provides the real-time lifting and tilting angle value of the carriage. This is a state quantity representing the tension difference. This refers to the adjusted actual tension value. The adjusted actual tension value is the absolute reading of the chain force state, remeasured and filtered after the actuator has completed physical compensation and the system oscillation has subsided. The dynamic safety tension lower limit value is the minimum operating tension warning threshold calculated dynamically with the car's lifting angle to ensure the chain does not slip or derail. The secondary forced intervention tension trigger event is a command flag directly issued by the system's highest priority safety logic when conventional closed-loop adjustment fails or dangerous slack exists.

[0096] For example, after the system tensioning actuator completes the expansion and contraction target compensation displacement of 28.8000 mm, it pauses for 0.5 seconds to eliminate mechanical inertial oscillations. Subsequently, the post-adjustment state retest module re-acquires the actual tension value of the chain conveyor belt after adjustment. The value is 11.5000 kN. The dynamic benchmark comparison module retrieves the current real-time lifting tilt angle of the trailer body as 45.0 degrees. Based on the safety test data of material gravity distribution at different tilt angles during semi-trailer unloading, the translational benchmark tension constant is set to 5.0000 kN, and the tilt angle tension sensitivity coefficient, which represents the additional resistance to slippage required for each degree of tilt increase, is set to 0.1500 kN per degree. The dynamic safety tension lower limit value is calculated and extracted by substituting into the formula. The Qianniu dynamic benchmark comparison module compares the actual tension value after adjustment with the dynamic safety tension lower limit value, and substitutes them into the formula to calculate the tension difference state quantity. The forced intervention trigger module determines the tension difference to be negative based on the calculation results, meaning the actual tension value after adjustment (11.5000 kN) is less than the dynamic safety tension lower limit (11.7500 kN). According to the mechanical anti-derailment rigid protection logic, there is a risk of chain slippage in this state, and the forced intervention trigger module immediately generates a secondary forced intervention tensioning trigger event.

[0097] Optionally, the system further includes:

[0098] The residual deviation calculation module is used to calculate the residual deviation between the lower limit of the dynamic safety tension and the actual tension value after adjustment, based on the secondary forced intervention tension triggering event.

[0099] The feedback sample construction module is used to fuse the residual deviation value with the dynamic relaxation state feature matrix before the occurrence of the secondary forced intervention tightening trigger event to construct a feedback training sample with self-supervised labels.

[0100] The model parameter correction module is used to input the feedback training samples with self-supervised labels into the preset temporal state tension evaluation model to correct the network node weight parameter matrix in the preset temporal state tension evaluation model.

[0101] Specifically, the residual deviation calculation module receives the secondary forced intervention tension trigger event signal and activates the calculation thread. It retrieves the dynamic safety tension lower limit value and the actual tension value after adjustment from the memory, calculates the difference between the two to generate the residual deviation value. The feedback sample construction module extracts the dynamic relaxation state feature matrix before the secondary forced intervention tension trigger event, introduces label mapping coefficients to convert the residual deviation value into a dimensionless scalar label, and uses a fusion splicing function to splice this dimensionless scalar label to the end of the dynamic relaxation state feature matrix, thereby constructing a feedback training sample with self-supervised labels. The model parameter correction module inputs the feedback training sample with self-supervised labels into the preset time-series tension evaluation model, calls the backpropagation algorithm engine, and combines the set learning rate constant and the gradient calculation function of the loss function with respect to the weights to iteratively update the uncorrected forget gate weight parameter matrix in the historical state memory layer along the opposite direction of the error gradient. The corrected forget gate weight parameter matrix is ​​then output to overwrite the original parameters.

[0102] ;

[0103] ;

[0104] ;

[0105] ;

[0106] This represents the residual deviation value. This is the lower limit value of dynamic safety tension. The actual tension value after adjustment. For dimensionless scalar labels, For label mapping coefficients, For feedback training samples with self-supervised labels, Concat is the fusion and concatenation function. The feature matrix of the dynamic relaxation state. This is the corrected forget gate weight parameter matrix. This is the forget gate weight parameter matrix before correction. The learning rate is a constant. This is the gradient calculation function of the loss function with respect to the weights. The residual deviation value represents the quantization tension difference between the ideal safety margin and the actual physical execution result caused by system hysteresis or model nonlinearity errors. The dimensionless scalar label is a pure numerical target variable that transforms the difference in physical measurements into a directly processable target variable for backpropagation in the neural network. The label mapping coefficient is a proportional conversion constant that balances the magnitude of the physical error with the convergence speed of the network loss function. The feedback training samples with self-supervised labels are a closed-loop learning data structure that uses the device's own operational deviations as correction targets without manual intervention in labeling. The fusion and splicing function is a data aggregation method that arranges the high-dimensional feature matrix and one-dimensional scalar labels continuously in the memory address space. The learning rate constant is a hyperparameter that controls the adjustment step size of the network parameters during each backpropagation iteration to prevent network oscillation or slow convergence. The gradient calculation function of the loss function with respect to the weights is a mathematical operation that uses the chain rule of calculus to differentiate and obtain the direction and magnitude of the current network parameters' contribution to the output error.

[0107] For example, the residual deviation calculation module, based on the received secondary forced intervention tension trigger event, retrieves the dynamic safety tension lower limit value of 11.7500 kN and the adjusted actual tension value of 11.5000 kN. Substituting these values ​​into the formula, the residual deviation value is calculated as 11.7500 - 11.5000 = 0.2500 kN. The feedback sample construction module extracts the state element sequence -10.5000, 30.5000, 40.0000, -12.3209, 30.5000, and 45.0000 from the dynamic relaxation state feature matrix Mfeat before the trigger event. To map the physical error to a gradient scale suitable for network computation, the label mapping coefficient is set to 10.0000 based on the experience of cross-entropy loss normalization. Substituting these values ​​into the formula, the dimensionless scalar label is calculated as follows: The fusion and splicing function is invoked to construct self-supervised training samples in the sequence -10.5000, 30.5000, 40.0000, -12.3209, 30.5000, 45.0000, 2.5000. The model parameter correction module is then entered, extracting the uncorrected forget gate weight parameter matrix from the historical state memory layer in the sequence 0.1000, 0.0000, 0.0000, -0.1000, 0.0000, 0.0000. To avoid model crashes due to excessively large single parameter update intervals, the learning rate constant is set to 0.0100 based on the deep reinforcement learning parameter smooth update principle. The feedback training samples are input into the model, and the gradient calculation function of the loss function with respect to the weights is executed. Following the chain rule of taking the derivative of the error with respect to the input, the gradient vector for the weights of the tension feature dimension is calculated as 0.5000, 0.0000, 0.0000, -0.5000, 0.0000, 0.0000. Substituting these values ​​into the formula, the first element of the update sequence is... The fourth element of the sequence is updated to -0.1000 minus the product of 0.0100 and -0.5000, resulting in -0.0950. The final output of the corrected forget gate weight parameter matrix is ​​the sequence 0.0950, 0.0000, 0.0000, -0.0950, 0.0000, 0.0000.

[0108] Optionally, the system further includes:

[0109] The service status extraction module is used to extract the cumulative tensile action frequency value of the chain conveyor belt from its initial commissioning to the current moment;

[0110] The aging and wear determination module is used to calculate the fatigue stress integral value of the chain metal by combining the cumulative tensile action frequency value with the preset tension impact attenuation coefficient, and to determine whether the fatigue stress integral value of the chain metal is greater than the preset physical material yield warning limit value, and at the same time to determine whether the residual deviation value continuously exceeds the preset safety redundancy error limit.

[0111] The extreme wear warning module is used to construct an extreme wear warning data packet for the chain when the fatigue stress integral value of the chain metal is greater than the preset physical material yield warning limit value and the residual deviation value continuously exceeds the preset safety redundancy error limit, and convert the extreme wear warning data packet into a host alarm indication signal for output.

[0112] Specifically, the control system's underlying timer and counter work in tandem. The service status extraction module reads the equipment history file recorded in non-volatile memory, extracting the cumulative tensile frequency of the chain conveyor belt from its initial commissioning to the current moment. The aging and wear assessment module is activated, retrieving a preset tension impact attenuation coefficient. The cumulative tensile frequency is multiplied by this coefficient to calculate the chain metal fatigue stress integral, characterizing the degree of internal material damage accumulation. The system further determines whether the chain metal fatigue stress integral exceeds a preset physical material yield warning limit and retrieves historical operating logs, comparing the residual deviation from the previous assessment period to the current period. It counts the consecutive times the residual deviation exceeds the limit and determines whether the residual deviation continuously exceeds a preset safety redundancy error limit. When both conditions are met simultaneously—that is, the chain metal fatigue stress integral exceeds the physical material yield warning limit and the residual deviation continuously exceeds the safety redundancy error limit—the extreme wear warning module is automatically triggered. The extreme wear early warning module, following the industry standard communication protocol format, concatenates and merges the current fatigue stress state, timestamp, and over-limit alarm code to construct an extreme wear early warning data packet for the chain. The communication interface controller performs digital-to-analog conversion and level conversion on the extreme wear early warning data packet, converting it into a host alarm indication signal for output, thereby triggering the action of the host computer or on-site audible and visual alarm devices.

[0113] ;

[0114] This represents the integral value of the chain metal fatigue stress. This is the cumulative number of stretching movements. This is the tension impact attenuation coefficient. For example... Figure 3The figure illustrates the entire lifecycle evolution model of chain fatigue wear. The curves in the figure visually reflect the linear decay process of the chain metal fatigue stress integral value as the cumulative tensile action frequency increases. The cumulative tensile action frequency is a physical statistic representing the total number of tensioning actions performed by the actuator since the equipment was put into operation. The tension impact attenuation coefficient is an empirical physical constant characterizing the microscopic damage to the metal lattice caused by a single tensile action, leading to a decrease in macroscopic strength. The chain metal fatigue stress integral value is a quantitative assessment index comprehensively reflecting the degree of cumulative fatigue damage to the chain under long-term alternating loads. The physical material yield warning threshold value is a reference threshold for the critical fatigue state of the chain, set according to the material's mechanical properties, indicating that the chain is about to undergo irreversible plastic deformation or fracture. The safety redundancy error limit is the highest allowable difference threshold for system control deviation without affecting the overall operational safety of the equipment. The chain extreme wear warning data packet is a standardized communication frame encapsulating fault codes and current fatigue state parameters for easy parsing by the host computer. The host alarm indication signal is a low-level interrupt level or communication command that can directly drive the audible and visual alarm or trigger a highlighted pop-up window on the monitoring interface.

[0115] For example, based on a residual deviation of 0.2500 kN, the service status extraction module is accessed, and the memory file is read to obtain the cumulative tensile action frequency of the chain conveyor belt from its initial commissioning to the current time, which is 150,000 times. The aging and wear determination module retrieves the tension impact attenuation coefficient calibrated according to the fatigue life curve of carbon steel material, which is 0.0050 MPa per cycle. Substituting these values ​​into the formula, the fatigue stress integral value of the chain metal is calculated as follows: The physical material yield strength warning limit is set at 700.0000 MPa based on the standard yield strength limit for steel used in industrial chains. The judgment logic is executed, and 750.0000 MPa is greater than 700.0000 MPa, confirming that fatigue accumulation exceeds the limit. The safety redundancy error limit is set at 0.2000 kN based on the maximum allowable clearance of the equipment track. The current residual deviation value of 0.2500 kN is compared with the safety redundancy error limit of 0.2000 kN; 0.2500 is greater than 0.2000, confirming a single instance of exceeding the limit. Historical logs are retrieved, revealing that the residual deviation value has exceeded 0.2000 kN for five consecutive sampling periods. This consecutive number exceeds the three consecutive exceedance tolerances set based on the system response delay characteristics, thus determining that the residual deviation value has continuously exceeded the safety redundancy error limit. When the extreme wear warning module receives the logical truth value that both of the above conditions are met, it immediately constructs a chain extreme wear warning data packet according to the controller local area network communication protocol. The data bits containing the fatigue value 750.0000 and the difference 0.2500 are converted into a host alarm indication signal with alternating high and low levels and output, driving the buzzer in the cockpit to emit an alarm sound.

[0116] Reference Figure 4Based on the same inventive concept, this invention also provides an automatic tension control method for chain conveyor belts used in dump semi-trailers, the method comprising:

[0117] The system acquires multi-source sensing data on the operating conditions of the chain conveyor belt mounted on the dump semi-trailer, and analyzes the multi-source sensing data to extract the real-time tension value of the chain, the current weight of the material being carried, the real-time lifting angle of the truck bed, and the real-time temperature of the operating environment.

[0118] Based on the real-time lifting angle of the carriage and the real-time temperature of the operating environment, compensation calculations are performed on the real-time tension of the chain and the weight of the currently carried material to construct a dynamic relaxation state feature matrix.

[0119] A preset time-series state tension evaluation model is invoked, and the dynamic relaxation state feature matrix is ​​input into the preset time-series state tension evaluation model to generate a target tension adjustment command.

[0120] The target tension adjustment command is executed, and a tension compensation drive signal with damping buffer constraint is output to the system tension actuator to change the chain tension of the chain conveyor belt.

[0121] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the present invention.

[0122] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. An automatic tensioning control system for a chain conveyor belt used in dump semi-trailers, characterized in that, The system includes: The multi-dimensional sensing data acquisition and analysis module is used to acquire multi-source sensing data of the operating conditions of the chain conveyor belt mounted on the dump semi-trailer, and to analyze the multi-source sensing data of the operating conditions to extract the real-time tension value of the chain, the weight value of the current material being carried, the real-time lifting angle value of the truck body, and the real-time temperature value of the operating environment. The multi-parameter state feature construction module is used to perform compensation calculations on the real-time tension value of the chain and the weight value of the currently carried material based on the real-time lifting angle value of the carriage and the real-time temperature value of the operating environment, and to construct a dynamic relaxation state feature matrix. The spatiotemporal perception model invocation and instruction generation module is used to invoke a preset temporal state tension evaluation model and input the dynamic relaxation state feature matrix into the preset temporal state tension evaluation model to generate a target tension adjustment instruction. The anti-resonance command execution and drive module is used to execute the target tension adjustment command and output a tension compensation drive signal with damping buffer constraint to the system tension actuator to change the chain tension of the chain conveyor belt.

2. The automatic tensioning control system for chain conveyor belts used in dump semi-trailers according to claim 1, characterized in that, The multidimensional sensing data acquisition and analysis module includes: A multi-source sensing signal receiving unit is used to receive the raw multimodal analog sensing electrical signals transmitted in real time by the monitoring node array deployed on the chain conveyor belt; The signal filtering and conversion unit is used to sequentially perform photoelectric isolation and adaptive Kalman filtering noise reduction operations on the original multimodal analog sensing electrical signal to generate a standard digital sensing signal sequence as the multi-source sensing data of the operating conditions; The multimodal data interpretation unit is used to interpret the header identifier of the data packet of the standard digital sensing signal sequence, and separate the real-time tension value of the chain representing the tensile physical state, the weight value of the currently carried material representing the gravity load state, the real-time lifting tilt angle value of the carriage representing the spatial attitude, and the real-time temperature value of the operating environment representing the thermodynamic state.

3. The automatic tensioning control system for chain conveyor belts used in dump semi-trailers according to claim 2, characterized in that, The multi-parameter state feature construction module includes: The tilt angle gravity component compensation unit is used to perform trigonometric function decomposition calculation by combining the current weight value of the loaded material with the real-time lifting tilt angle value of the carriage to generate the axial tensile force compensation value under the tilt angle. The thermal expansion and contraction deformation compensation unit is used to extract the metal thermal expansion coefficient of the chain included in the chain conveyor belt, and generate temperature-induced deformation compensation values ​​based on the difference between the real-time temperature value of the operating environment and the preset standard reference temperature value and the metal thermal expansion coefficient. The feature matrix fusion and splicing unit is used to fuse the axial tensile force compensation value and the temperature-induced deformation compensation value into the real-time tension value of the chain, generate the corrected real tension value, and splice the corrected real tension value, the current load material weight value and the real-time lifting angle value of the carriage according to the preset time-series data dimension to construct the dynamic relaxation state feature matrix.

4. The automatic tensioning control system for chain conveyor belts used in dump semi-trailers according to claim 3, characterized in that, The spatiotemporal perception model invocation and instruction generation module includes: The network structure loading unit is used to load a long short-term memory neural network structure containing an input feature gating layer, a historical state memory layer, and a nonlinear regression decision layer as the preset temporal state tension evaluation model. The temporal feature forward propagation unit is used to load the dynamic relaxed state feature matrix into the input feature gating layer, and fuse the forgetting control parameters of the previous sampling period through the historical state memory layer to output a deep hidden response feature with temporal dependence. The regression mapping calculation unit is used to perform regression mapping calculation on the deep hidden response features through the nonlinear regression decision layer to generate the target tension adjustment command containing the target compensation displacement value and the benchmark execution speed value.

5. The automatic tensioning control system for chain conveyor belts used in dump semi-trailers according to claim 4, characterized in that, The temporal feature forward propagation unit includes: The parameter extraction subunit is used to extract the fixed-configuration forget gate weight parameter matrix, input gate weight parameter matrix and state bias constant vector in the historical state memory layer; The gated activation calculation subunit is used to perform matrix multiplication on the dynamic relaxation state feature matrix based on the forget gate weight parameter matrix and the input gate weight parameter matrix, respectively, and process the matrix multiplication result through a preset logical nonlinear function to generate an information retention ratio matrix and an information update ratio matrix. The memory state update subunit is used to perform a Hadamard product operation on the information retention ratio matrix and the information update ratio matrix to generate a product update result, and then sum and fuse the product update result with the state bias constant vector to output the deep hidden response feature.

6. The automatic tensioning control system for chain conveyor belts used in dump semi-trailers according to claim 5, characterized in that, The anti-resonance command execution and driving module includes: The instruction value extraction unit is used to parse the target tension adjustment instruction and extract the target compensation displacement value and the reference execution speed value contained in the target tension adjustment instruction. The damping control parameter matching unit is used to extract the mechanical resonance frequency spectrum pre-stored in the control system, compare the reference execution speed value with the mechanical resonance frequency spectrum, and generate damping speed limiting control parameters to avoid the mechanical resonance range. The buffer drive signal sending unit is used to generate the tension compensation drive signal with damping buffer constraint based on the target compensation displacement value and the damping speed limiting control parameter, and send the tension compensation drive signal with damping buffer constraint to the power controller of the system tension actuator to drive the system tension actuator to extend or retract the target compensation displacement value.

7. The automatic tensioning control system for chain conveyor belts used in dump semi-trailers according to claim 1, characterized in that, The system also includes: The adjusted state retest module is used to reacquire the actual tension value of the chain conveyor belt after the tensioning actuator of the system executes the tension compensation drive signal with damping buffer constraint to complete the chain tension change action; The dynamic benchmark comparison module is used to extract the dynamic safety tension lower limit value corresponding to the current real-time lifting angle value of the carriage, and compare the adjusted actual tension value with the dynamic safety tension lower limit value. The forced intervention trigger module is used to generate a secondary forced intervention tension trigger event when the actual tension value after adjustment is less than the lower limit value of the dynamic safety tension.

8. The automatic tensioning control system for chain conveyor belts used in dump semi-trailers according to claim 7, characterized in that, The system also includes: The residual deviation calculation module is used to calculate the residual deviation between the lower limit of the dynamic safety tension and the actual tension value after adjustment, based on the secondary forced intervention tension triggering event. The feedback sample construction module is used to fuse the residual deviation value with the dynamic relaxation state feature matrix before the occurrence of the secondary forced intervention tightening trigger event to construct a feedback training sample with self-supervised labels. The model parameter correction module is used to input the feedback training samples with self-supervised labels into the preset temporal state tension evaluation model to correct the network node weight parameter matrix in the preset temporal state tension evaluation model.

9. The automatic tensioning control system for chain conveyor belts used in dump semi-trailers according to claim 8, characterized in that, The system also includes: The service status extraction module is used to extract the cumulative tensile action frequency value of the chain conveyor belt from its initial commissioning to the current moment; The aging and wear determination module is used to calculate the fatigue stress integral value of the chain metal by combining the cumulative tensile action frequency value with the preset tension impact attenuation coefficient, and to determine whether the fatigue stress integral value of the chain metal is greater than the preset physical material yield warning limit value, and at the same time to determine whether the residual deviation value continuously exceeds the preset safety redundancy error limit. The extreme wear warning module is used to construct an extreme wear warning data packet for the chain when the fatigue stress integral value of the chain metal is greater than the preset physical material yield warning limit value and the residual deviation value continuously exceeds the preset safety redundancy error limit, and convert the extreme wear warning data packet into a host alarm indication signal for output.

10. An automatic tension control method for the chain of a chain conveyor belt used in dump semi-trailers, applied to the automatic tension control system for the chain conveyor belt used in dump semi-trailers as described in any one of claims 1-9, characterized in that, The method includes: The system acquires multi-source sensing data on the operating conditions of the chain conveyor belt mounted on the dump semi-trailer, and analyzes the multi-source sensing data to extract the real-time tension value of the chain, the current weight of the material being carried, the real-time lifting angle of the truck bed, and the real-time temperature of the operating environment. Based on the real-time lifting angle of the carriage and the real-time temperature of the operating environment, compensation calculations are performed on the real-time tension of the chain and the weight of the currently carried material to construct a dynamic relaxation state feature matrix. A preset time-series state tension evaluation model is invoked, and the dynamic relaxation state feature matrix is ​​input into the preset time-series state tension evaluation model to generate a target tension adjustment command. The target tension adjustment command is executed, and a tension compensation drive signal with damping buffer constraint is output to the system tension actuator to change the chain tension of the chain conveyor belt.

Images