An Online Monitoring Method for Keel Production Line Based on Industrial Cloud Computing

By using an online monitoring method based on industrial cloud computing, the manufacturing control parameters of the greenhouse keel production line are adjusted in real time, which solves the manufacturing error problem of the greenhouse keel production line when facing changing greenhouse structural conditions, and improves manufacturing consistency and assembly adaptability.

CN123113786APending Publication Date: 2026-11-17JIANGSU JIUCHUN AGRI EQUIP TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611623273.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-20
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

When faced with changing greenhouse structural conditions, existing greenhouse keel production lines are prone to a disconnect between the manufacturing process and actual usage requirements, leading to spatial position deviations, connection misalignments, or the accumulation of local deformations. Furthermore, existing online detection systems cannot effectively identify and promptly adjust errors in the manufacturing process.

Method used

An online monitoring method based on industrial cloud computing is adopted. By acquiring scene constraint data, matching material process attributes, collecting servo execution stroke commands and drive end status, and combining three-dimensional space curve detection, manufacturing control parameters are adjusted in real time to form continuous information feedback and control correction.

Benefits of technology

It improves the manufacturing consistency and assembly adaptability of greenhouse keel frames, reduces the disconnect between manufacturing standards and actual use conditions, lowers the risk of gradual error amplification, and improves the dynamic stability and overall control effect of the production process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123113786A_ABST
    Figure CN123113786A_ABST
Patent Text Reader

Abstract

This invention relates to the field of industrial control technology, specifically to an online monitoring method for a greenhouse keel production line based on industrial cloud computing. The method includes state monitoring and closed-loop adjustment applicable to the continuous manufacturing process of greenhouse keel frames. Addressing the problem that greenhouse keel frames need to adapt to different spatial structures and service conditions in actual construction, while the continuous production process is easily affected by material conditions, equipment operating conditions, and manufacturing process fluctuations, leading to deviations between the final component shape and the target assembly state, this invention continuously monitors the equipment execution state, component output state, and spatial morphology during the continuous manufacturing process. Based on the actual assembly requirements of the greenhouse keel frame, a dynamic control boundary is formed. The online monitoring results are compared with the target control state to generate control correction results for subsequent production processes, achieving closed-loop control of the greenhouse keel frame production process, including monitoring, deviation identification, state assessment, and feedback adjustment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of industrial control technology, specifically to an online monitoring method for a keel production line based on industrial cloud computing. Background Technology

[0002] Existing greenhouse frame production lines typically organize continuous manufacturing around fixed specifications, with production control primarily based on pre-set process parameters and product quality standards. When the greenhouse structural conditions corresponding to the production task change, a disconnect can easily arise between the manufacturing process and actual usage requirements. Although some components manufactured according to uniform standards may meet basic dimensional requirements, problems such as spatial position deviations, connection misalignments, or localized deformation accumulation may still occur after the actual greenhouse frame is assembled. Traditional production lines usually maintain the processing with fixed control parameters, lacking the ability to continuously track such long-term changes, which can easily lead to differences between the manufacturing results in the early and later stages of equipment operation.

[0003] Furthermore, existing online inspection systems primarily focus on whether individual components exceed preset quality ranges, neglecting the correlation between localized component deviations and the overall assembly state of the greenhouse frame. Minor deviations in certain local locations may accumulate after continuous connection and overall assembly, and the impact of deviations at different locations on the overall structure varies. Using a uniform standard for production control can easily lead to insufficient control of critical locations and excessive control of non-critical locations. On the other hand, there is often a significant information lag between inspection results and subsequent production control in current manufacturing systems. This means that even after a deviation occurs in a component, subsequent production continues to use the original control state, making it difficult to promptly suppress the propagation of errors during continuous manufacturing.

[0004] To address this, an online monitoring method for keel production lines based on industrial cloud computing is proposed. Summary of the Invention

[0005] This invention aims to provide an online monitoring method for greenhouse keel production lines based on industrial cloud computing. By continuously linking production monitoring, status assessment, deviation feedback, and control correction, it improves the manufacturing consistency and actual assembly adaptability of greenhouse keel frames.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for online monitoring of a keel production line based on industrial cloud computing, the method comprising: Obtain the scene constraint data corresponding to the target farm, and determine the process configuration type and material compatibility requirements of the production line materials based on the scene constraint data; Obtain the actual material process attributes and mechanical state indication values, wherein the actual material process attributes include material data; match the actual material process attributes with the process configuration type, call the corresponding process disturbance preset parameter set and control execution compensation parameter set, and control and correct the system impedance changes and output response deviations during continuous manufacturing process; Collect servo execution stroke commands and drive end status monitoring data from control nodes at all levels, and determine the cumulative loss status of the execution terminal based on the load impedance changes of control nodes at all levels and the equipment operating cycle; The three-dimensional spatial curves of the offline components are collected, and the measured form and position deviation sequence is extracted along the trajectory vector of the offline components. The industrial cloud platform is used to run simulation calculations based on the scenario constraint data to obtain the target manufacturing control tolerance zone. The collaborative control correction vector for the next stage of manufacturing process is obtained by comparison and calculation.

[0007] Preferably, the scenario constraint data includes terminal receptor factors, system architecture factors, and service environment factors. The terminal receptor factors include receptor type, environmental indicator requirements, and layout method. The system architecture factors include span parameters, bay parameters, height parameters, and support spacing. The service environment factors include construction latitude, azimuth, and external load conditions. Based on the terminal receptor factors, the control performance requirements, global load-bearing requirements, and local node load-bearing requirements corresponding to the target component are determined. Based on the system architecture factors, the span adaptation requirements, spatial manufacturing requirements, and component deformation sensitivity corresponding to the target component are determined. Based on the service environment factors, the corrosion resistance environment requirements, external load requirements, and temperature and humidity service requirements corresponding to the target component are determined in advance.

[0008] Preferably, determining the process configuration type and material compatibility requirements of the production line materials based on the scenario constraint data includes: Based on the corrosion resistance requirements, external load requirements, temperature and humidity service requirements, and manufacturing deformation requirements of the target component, production control adaptation evaluations are performed on different candidate material attributes. The process configuration type is determined based on the evaluation results, and a set of preset process disturbance parameters, a set of control execution compensation parameters, and a mechanical strength adaptation range corresponding to the process configuration type are established. When the actual material process attributes, including material data, obtained from the production line material import end are consistent with the process configuration type, the preset process disturbance parameters and the set of control execution compensation parameters corresponding to the process configuration type are called to correct the manufacturing process control parameters. When the actual material process attributes are inconsistent with the process configuration type, the control strategy for the current batch of manufacturing raw materials is adaptively reconstructed according to the material adaptation requirements.

[0009] Preferably, the cumulative wear status of the execution terminal is determined in the following manner: Adaptive filtering is performed on the servo execution stroke commands and drive end status monitoring data of each level of control node to construct a residual sequence between the actual execution impedance and the nominal execution impedance. The residual sequence is then decomposed into trends based on the historical operating cycle of the production line equipment to extract the terminal load change trend as the equipment operating time increases. The cumulative loss status of the execution terminal is determined based on the terminal load change trend. The cumulative loss status of the execution terminal and the material data are input into the compensation calculation process to correct the system output shape and position deviation caused by material batch changes and execution terminal loss.

[0010] Preferably, the acquisition of the measured form and position deviation sequence and the establishment of the terminal performance projection boundary include: obtaining the three-dimensional spatial curve of the offline component through high-frequency radar ranging, laser contour scanning and machine vision acquisition; acquiring spatial point cloud and performing coordinate unification, outlier removal, spatial slicing and data sorting to form a measured form and position deviation sequence that continuously changes along the component trajectory vector; determining the hourly environmental incident direction according to the construction geographical latitude and azimuth angle corresponding to the target application scenario; acquiring the preset assembly pose information of the offline component in the target application scenario; mapping the measured three-dimensional spatial curve of the offline component to a global three-dimensional virtual model established based on the system architecture factors through coordinate system affine transformation to construct a virtual assembly containing the actual form and position deviation; performing illumination occlusion projection simulation on the virtual assembly based on the hourly environmental incident direction to extract the terminal performance projection boundary corresponding to the offline component in the global assembly state.

[0011] Preferably, the determination of the target manufacturing control tolerance zone includes: mapping the hourly environmental incident direction to the target application scene's area of ​​action at fixed time intervals, calculating the constraint coverage area formed by the terminal performance projection boundary; determining the allowable state discrete range according to the physical performance index requirements of the terminal object; calculating the cumulative global structural deformation and constraint coverage change rate caused by the form and position deviations of different trajectory positions of the offline components in the virtual assembly; under the condition that the constraint coverage change rate satisfies the allowable state discrete range and the cumulative global structural deformation satisfies the assembly interference threshold, inverting to determine the control allowable deviation at the corresponding position, and continuously mapping the control allowable deviations corresponding to each position to form the target manufacturing control tolerance zone.

[0012] Preferably, the calculation of the collaborative control correction vector includes: uploading the measured form and position deviation sequence and the target manufacturing control tolerance zone to the industrial cloud platform, and corresponding them according to the trajectory position to determine the output deviation of each control interval; the industrial cloud platform determines the target control node that needs to be adjusted according to the output deviation, and corrects the system response control coefficient of the corresponding control node according to the material data and the cumulative loss status of the execution terminal; the industrial cloud platform calculates the servo execution stroke correction value of each target control node based on the corrected system response control coefficient, and forms a control command adjustment sequence according to the node order; online closed-loop control of the servo execution stroke of the corresponding target control node, and feeding back the measured form and position status after the next manufacturing stage to the industrial cloud platform, continuously updating the control command adjustment sequence.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention does not merely focus on production control based on fixed specifications, but rather incorporates the structural and service requirements involved in the actual construction of the greenhouse frame into the manufacturing control framework. This allows the manufacturing requirements corresponding to different application scenarios to be transmitted to the production process. When greenhouse construction conditions change, the production system can form corresponding control criteria based on different application needs, reducing the disconnect between manufacturing standards and actual greenhouse frame usage conditions. This makes the produced frame components more compatible with the installation and usage requirements of the corresponding greenhouse structure, improving the scenario adaptability of the frame products.

[0014] 2. This invention can continuously monitor the state changes of production equipment during continuous operation and identify deviation trends in the manufacturing process by combining the actual output results of components. This enables the production system to distinguish between normal fluctuations and continuous state changes during production. Compared to using fixed control parameters for a long time, this invention can adjust subsequent manufacturing control in a timely manner according to changes in production status, reducing the drift of processing results caused by increased equipment operating time. This reduces the risk of errors gradually amplifying during continuous production and improves the consistency of keel forming between different production periods.

[0015] 3. This invention uses greenhouse application requirements as the front-end constraint for manufacturing control, equipment operating status as the basis for production process status, and the actual spatial form after the keel is produced as the result feedback. Furthermore, the detection results are applied to subsequent production control, creating a continuous information transmission relationship between different stages. This allows for simultaneous consideration of production tasks, current equipment status, component manufacturing status, and subsequent adjustments, avoiding reliance on a single stage for quality judgment. It achieves synergistic linkage between error identification, impact analysis, and control correction, improving dynamic stability and overall control effectiveness in continuous production processes. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of an online monitoring method for a keel production line based on industrial cloud computing, as described in this invention. Figure 2 This is a schematic diagram of the process for extracting form and position deviations and providing cloud-based tolerance zone feedback in this invention. Figure 3 This is a schematic diagram of the cloud-based tolerance zone generation and closed-loop collaborative control process of the present 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 to 3 This invention provides an online monitoring method for a keel production line based on industrial cloud computing, referring to... Figure 1 Step-by-step flowchart Figure 2 A flowchart illustrating the extraction of form and position deviations and cloud-based tolerance zone feedback, and Figure 3 A schematic diagram of the cloud-based tolerance band generation and closed-loop collaborative control process; the technical solution is as follows: Obtain the scene constraint data corresponding to the target farm, and determine the process configuration type and material compatibility requirements of the production line materials based on the scene constraint data; Obtain the actual material process attributes and mechanical state indication values, wherein the actual material process attributes include material data; match the actual material process attributes with the process configuration type, call the corresponding process disturbance preset parameter set and control execution compensation parameter set, and control and correct the system impedance changes and output response deviations during continuous manufacturing process; Collect servo execution stroke commands and drive end status monitoring data from control nodes at all levels, and determine the cumulative loss status of the execution terminal based on the load impedance changes of control nodes at all levels and the equipment operating cycle; The three-dimensional spatial curves of the offline components are collected, and the measured form and position deviation sequence is extracted along the trajectory vector of the offline components. The industrial cloud platform is used to run simulation calculations based on the scenario constraint data to obtain the target manufacturing control tolerance zone. The collaborative control correction vector for the next stage of manufacturing process is obtained by comparison and calculation.

[0019] Example 1: This embodiment focuses on five aspects of continuous manufacturing: scenario requirement acquisition, material process attribute matching, execution terminal status monitoring, and off-line component shape and position detection and collaborative control correction. By transforming the scenario constraint data corresponding to the target farm into quantifiable manufacturing control basis, and combining the production line equipment operating status and the measured shape and position data of the components for closed-loop feedback, it realizes online identification and correction of output deviations in continuous manufacturing.

[0020] First, the scenario constraint data includes terminal receptor factors, system architecture factors, and service environment factors. The terminal receptor factors include receptor type, environmental indicator requirements, and layout method. The system architecture factors include span parameters, bay parameters, height parameters, and support spacing. The service environment factors include construction latitude, azimuth, and external load conditions. Based on the terminal receptor factors, the control performance requirements, global load-bearing requirements, and local node load-bearing requirements corresponding to the target component are determined. Based on the system architecture factors, the span adaptation requirements, spatial manufacturing requirements, and component deformation sensitivity corresponding to the target component are determined. Based on the service environment factors, the corrosion resistance environment requirements, external load requirements, and temperature and humidity service requirements corresponding to the target component are determined in advance.

[0021] Specifically, in the scene constraint data collection stage, the industrial cloud platform receives the original scene description information submitted by the target farm through the scene configuration interface of the production line host computer. This original scene description information is entered in the form of a structured form, covering ten basic fields, including recipient type, environmental indicator requirements, layout method, span parameters, bay parameters, height parameters, support spacing, construction geographical latitude, azimuth angle, and external load conditions. The platform performs data type verification and value range verification on these ten fields respectively. Fields that fail the verification are marked as pending data entry and a prompt message is returned to the data entry end.

[0022] Specifically, in the terminal receptor factor analysis stage, the platform calls a pre-established receptor attribute mapping table, compares and matches the entered receptor type field with the category entries in the mapping table. Under the same receptor type entry, there are three pre-associated attributes: control performance level, global load-bearing coefficient range, and local node load-bearing coefficient range. The platform weights and corrects the environmental indicator requirement field with the control performance level to obtain the control performance requirement corresponding to the target component. It substitutes the support point distribution density indicated by the layout method field into the global load-bearing coefficient range to obtain the global load-bearing requirement corresponding to the target component. For the node positions marked as suspended or partially loaded in the layout method, the platform extracts the corresponding values ​​from the local node load-bearing coefficient range to form the local node load-bearing requirement corresponding to the target component. For example, when the receptor type is entered as "suspended light load receptor", the control performance level is set to level two, the global load-bearing coefficient range is set to 0.8 to 1.2, and the local node load-bearing requirement is calculated point by point with an additional 15% load margin for each suspended node.

[0023] Specifically, in the system architecture factor analysis stage, the platform substitutes four parameters—span, bay, height, and support spacing—into a pre-established set of spatial structure adaptation rules. This set is divided into several levels based on span ranges, with each level corresponding to a set of span adaptation requirements. The platform determines the span adaptation requirements of the target component based on the level the span parameter falls into, the spatial manufacturing requirements based on the ratio of the bay to height parameters, and the component deformation sensitivity based on the ratio of the support spacing to the span parameter. A higher ratio indicates a higher level of component deformation sensitivity. For example, when the span is 8 meters and the support spacing is 1 meter, the ratio is 0.125, falling into the preset middle level range, and the corresponding component deformation sensitivity is labeled as moderately sensitive. When the support spacing decreases to 0.6 meters and the ratio drops to 0.075, the component deformation sensitivity is labeled as low sensitive.

[0024] Specifically, in the service environment factor analysis stage, the platform pre-establishes a correspondence table between geographical latitude and annual average humidity and heat distribution, and seasonal temperature difference range, as well as a correspondence table between azimuth and prevailing wind direction angle, and between external load conditions and regional wind and snow load standard values. The platform substitutes the construction geographical latitude into the first correspondence table to obtain the temperature and humidity service requirements corresponding to the target component, and substitutes the azimuth and external load conditions into the second correspondence table to obtain the corrosion resistance environment requirements and external load requirements corresponding to the target component. For example, when the construction geographical latitude is 35 degrees North, the corresponding annual average humidity and heat distribution level is marked as level three, and the seasonal temperature difference range is marked as a range of 16 to 20 degrees Celsius. The temperature and humidity service requirements are thus set to a medium level of moisture-proof and anti-condensation. The analysis of the above three types of service environment requirements is pre-executed after the scenario constraint data is received. The analysis results are temporarily stored in the scenario requirement library of the cloud platform for subsequent material adaptation and manufacturing control stages to call upon as needed, without being recalculated with the start or end of a single production task.

[0025] By transforming the three original scenario fields—receptor, structure, and environment—into quantitative demand indicators that can be directly called upon in subsequent manufacturing control processes, the workload of manually setting control parameters repeatedly under different scenario conditions is reduced, and the clarity of the correspondence between the scenario demand analysis results and the actual manufacturing requirements of the target component is improved.

[0026] Furthermore, determining the process configuration type and material adaptation requirements of production line materials based on the scenario constraint data includes: evaluating the production control adaptation of different candidate material attributes based on the target component's corrosion resistance requirements, external load requirements, temperature and humidity service requirements, and manufacturing deformation requirements; determining the process configuration type based on the production control adaptation evaluation results, and establishing a process disturbance preset parameter set, control execution compensation parameter set, and mechanical strength adaptation range corresponding to the process configuration type; when the actual material process attributes, including material data, obtained at the production line material import end are consistent with the process configuration type, the process disturbance preset parameter set and control execution compensation parameter set corresponding to the process configuration type are called to correct the manufacturing process control parameters; when the actual material process attributes are inconsistent with the process configuration type, the control strategy adaptive reconstruction judgment is performed on the current batch of manufacturing raw materials according to the material adaptation requirements.

[0027] Specifically, in the production control adaptation evaluation stage, the industrial cloud platform retrieves three indicators—corrosion resistance environment requirements, external load requirements, and temperature and humidity service requirements—from the scenario requirement database, which were generated during the analysis stage. It then combines these with the component deformation sensitivity obtained from system architecture factor analysis to calculate the manufacturing deformation requirements. The platform pre-establishes a candidate material attribute database, where each candidate material attribute entry records four basic parameters: corrosion resistance level, mechanical strength range, ductility coefficient, and forming springback coefficient. The platform evaluates each candidate material attribute entry one by one, matching and scoring the corrosion resistance level with the corrosion resistance environment requirements, the mechanical strength range with the external load requirements, the ductility coefficient with the temperature and humidity service requirements, and the forming springback coefficient with the manufacturing deformation requirements. These four scores are then summed according to preset weights to obtain the comprehensive adaptation score for the candidate material attribute entry. For example, if a candidate material attribute entry scores 82, 76, 88, and 79 points in the four scores respectively, the comprehensive adaptation score after weighted summation is 81 points.

[0028] Specifically, in the process configuration type determination stage, the platform sorts the comprehensive adaptation scores of each item in the candidate material attribute library, and determines the category corresponding to the candidate material attribute item with the highest score that is not lower than the preset adaptation threshold as the process configuration type of the target component. If the highest score after sorting is still lower than the preset adaptation threshold, its setting is based on the statistical regression model of material mechanical fluctuations and final product yield in past production batches. This threshold corresponds to the minimum material mechanical property confidence level required for the system to pull the final form and position tolerance back to the qualified range through adaptive response under the maximum compensation capability limit of the current production line servo mechanism. The platform marks the generated evaluation results as pending manual review and suspends automatic type determination. After the process configuration type is determined, the platform immediately extracts three types of data bound to the process configuration type from the pre-established parameter association library: the preset set of process disturbance parameters, the set of control execution compensation parameters, and the mechanical strength adaptation range. These three types of data are written into the configuration file of this production task for direct use in subsequent manufacturing control stages. For example, if the preset adaptation threshold is set to 75 points, and the candidate material attribute entries with a comprehensive adaptation score of 81 points are higher than this threshold, their corresponding category is determined as the process configuration type for this production task.

[0029] Specifically, in the process of acquiring and determining the consistency of actual material process attributes, an online detection device is installed at the material inlet of the production line. This device performs online detection on each roll of raw material entering the production line, obtaining the actual material process attributes, including material data. The material data specifically includes three sub-data items: alloy composition percentage, measured yield strength, and surface treatment type. The platform compares these actual material process attributes with the candidate material attribute items corresponding to the determined process configuration type item by item. If all three sub-data comparison results fall within the preset tolerance range (the tolerance range of the actual material directly refers to the national or industry standards for allowable tolerance zones of raw materials such as hot-dip galvanized steel strip), it is considered consistent; if any item exceeds the preset tolerance range, it is considered inconsistent. When the determination result is consistent, the platform calls the preset parameter set for process disturbance and the control execution compensation parameter set corresponding to the process configuration type, corrects the manufacturing process control parameters, and then sends them to the production line execution end. When the determination result is inconsistent, the platform adaptively reconstructs the determination of the control strategy for the current batch of manufacturing raw materials according to the material adaptation requirements. The reconstruction determination process includes re-searching in the candidate material attribute library for the candidate material attribute entry closest to the actual material's process attribute, and interpolating and correcting the parameter set bound to the adjacent entry to form a transitional parameter set dedicated to this batch, which is temporarily called by the manufacturing control link. At the same time, the batch is marked as an unconventional batch and recorded in the process traceability file. For example, if the tolerance range of the measured yield strength of a batch exceeds the range corresponding to the process configuration type, and it is determined to be inconsistent, the platform retrieves the nearest entry in the candidate library and generates a transitional parameter set for the manufacturing control of this batch.

[0030] The actual material process attributes and yield strength indication value are obtained. The actual material process attributes include material data. The yield strength indication value is obtained by the online detection device at the material inlet of the production line for each roll of raw material entering the production line. Together with the measured yield strength value in the material data, it constitutes the basis for determining the mechanical state of the material. The actual material process attributes are matched with the process configuration type, and the corresponding process disturbance preset parameter set and control execution compensation parameter set are called to control and correct the system impedance changes and output response deviations in the continuous manufacturing process. By evaluating the correspondence between scenario requirements and candidate material attributes, and determining the matching status in conjunction with preset judgment thresholds, the material adaptation process generates a judgment result with quantitative basis, while retaining the corresponding evaluation process and judgment records. When it is detected that the actual material attributes have deviated or exceeded the applicable range of the current configuration, the reconstruction and configuration adjustment of the control parameter set are triggered, so that subsequent manufacturing control can be configured according to the updated material status.

[0031] Furthermore, to address the limitation that a single environmental field cannot be accurately mapped to continuous manufacturing material parameters, an adaptive material process reconstruction mechanism based on multimodal scenario knowledge graphs and multi-objective constraint optimization is constructed.

[0032] This mechanism achieves precise quantitative transformation from macroscopic scenario requirements to microscopic process configuration by establishing an implicit nonlinear mapping between the service environment, terminal structure, and material mechanical properties. Specifically, it first extracts the feature vectors of the target farm's terminal receptor, system architecture, and service environment, and inputs them as entity nodes into a pre-trained greenhouse scenario knowledge graph for correlation inference, outputting the time-varying load boundary and weather resistance index of the target component during its actual service life. Subsequently, in the material configuration stage, the system no longer relies on simple linear weight scoring, but establishes a multivariate nonlinear optimization equation with the goal of maximizing overall assembly yield and minimizing manufacturing response deviation. The alloy composition ratio and actual yield strength measured values ​​collected online at the material inlet of the production line are used as dynamic constraints and substituted into the above performance objective function. Particle swarm optimization or genetic algorithm is used to perform global optimization within the feasible process parameter space to solve for the optimal preset set of process disturbance parameters that match the characteristics of the current batch of materials. If the optimization result shows that the current batch of materials cannot converge within the safety threshold, the boundary conditions of the safety threshold are the ultimate fatigue load of the continuous roll forming equipment and the rated torque of the motor, used to prevent the servo system from triggering hardware overload shutdown due to excessive correction. The platform will automatically trigger control strategy reconfiguration decisions and issue constraints to edge terminals regarding material degradation or operational restrictions. The entire mapping and optimization process is performed in parallel in the cloud using a digital twin framework, ensuring that adjustments to on-site process parameters have strict physical and mathematical basis. This enables the processing technology for each batch of materials to achieve optimal alignment with the actual service scenario at the terminal, significantly improving the control system's adaptability under complex raw material disturbances.

[0033] Furthermore, the cumulative loss status of the execution terminal is determined as follows: adaptive filtering is performed on the servo execution stroke commands and drive end status monitoring data of each level of control node to construct a residual sequence between the actual execution impedance and the nominal execution impedance; trend decomposition is performed on the residual sequence in combination with the historical operating cycle of the production line equipment to extract the terminal load change trend as the equipment operating time increases, and the cumulative loss status of the execution terminal is determined according to the terminal load change trend; the cumulative loss status of the execution terminal and the material data are jointly input into the compensation calculation process to correct the system output form and position deviation caused by material batch changes and execution terminal loss.

[0034] Specifically, in the data acquisition and preprocessing stage, each control node at each level of the production line is equipped with a servo drive status acquisition unit. This acquisition unit reads the servo execution stroke command value and the current and voltage monitoring data of the drive end according to a fixed sampling period, forming a set of multi-channel timing data containing stroke command channel and drive end status channel. After the multi-channel timing data is converted into digital signal by the analog-to-digital conversion module of the local controller, it is uploaded to the edge computing node through the production line fieldbus. The edge computing node first performs noise reduction preprocessing on the uploaded data to remove abnormal jump points caused by communication jitter, and interpolates the missing sampling points according to the nearest value to form a complete input data sequence.

[0035] Specifically, in the adaptive filtering stage, the edge computing node calls a pre-deployed adaptive filtering module. This module takes the complete input data sequence as input and dynamically adjusts the filtering coefficients based on the fluctuation amplitude of data within several adjacent sampling windows. Windows with larger fluctuation amplitudes use smaller filtering coefficients to retain load change information, while windows with smaller fluctuation amplitudes use larger filtering coefficients to suppress random noise. Simultaneously, the filtering module reads the pre-calibrated nominal execution impedance reference value of the control node and subtracts the filtered actual execution impedance from the nominal execution impedance reference value point by point, outputting a set of residual sequences arranged continuously along the sampling time. The physical meaning of this residual sequence is the degree of deviation between the actual load response of the execution terminal after removing random noise and the designed load response. For example, if the original residual fluctuation range of a control node within 500 consecutive sampling points is -8 units to +9 units, after adaptive filtering, the fluctuation range narrows to -3 units to +3 units. The sampling window length is set to 20 consecutive sampling points, and the window sliding step size is set to 5 sampling points. The filter coefficient ranges from 0.1 to 0.9. The module pre-establishes a correspondence table between fluctuation amplitude ranges and filter coefficient values. The difference between the maximum and minimum values ​​of the data in the current window is taken as the fluctuation amplitude of the window. The filter coefficient of the window is read from the correspondence table according to the range in which the fluctuation amplitude falls. The fluctuation amplitude range and the filter coefficient value are negatively correlated, that is, the larger the range value, the smaller the corresponding filter coefficient value.

[0036] Specifically, in the trend decomposition stage, the edge computing node inputs the residual sequence along with the historical operating cycle data of the production line equipment corresponding to that control node into the trend decomposition module. The trend decomposition module has two parallel processing branches: a periodic decomposition channel and a trend decomposition channel. The periodic decomposition channel identifies and removes periodic fluctuation components synchronized with the production line's cycle time, while the trend decomposition channel extracts low-frequency components in the residual sequence that monotonically increase or slowly drift with the cumulative operating cycle of the equipment. The intermediate results from the two branches are synthesized and verified at the end of the module. The remaining sequence after removing the periodic components represents the terminal load change trend. The physical meaning of this trend data is the load response drift caused by the continuous operation of the execution terminal. Based on the slope and amplitude of the terminal load change trend, the edge computing node determines the cumulative loss status of the execution terminal by referring to a pre-defined loss classification table. The loss classification table is divided into several loss levels according to the trend slope range. For example, if the load change trend slope of a certain execution terminal shows a continuous upward trend over three consecutive production cycles, it is determined to be in a level two loss state according to the loss classification table.

[0037] The periodic decomposition channel uses the pre-known production line cycle length as the decomposition cycle parameter. It uses the moving average method to calculate the periodic mean of the residual sequence within the cycle length and removes it as a periodic fluctuation component. The trend decomposition channel performs a moving average process along the cumulative operating cycle of the equipment on the remaining sequence after removing the periodic fluctuation component, and extracts the low-frequency components that monotonically increase or slowly drift with the operating cycle. After the two branches are processed, the module subtracts the low-frequency components output by the trend decomposition channel from the original value of the residual sequence point by point to obtain the verification residual. When the mean square value of the verification residual does not exceed the preset verification threshold, the synthesis verification is determined to be passed, and the low-frequency component is output as the terminal load change trend. If the verification fails, the moving average window length is adjusted by a preset step size and the decomposition is re-executed until the verification is passed or the number of adjustments reaches the preset upper limit.

[0038] The preset verification threshold depends on the background electrical noise of the production line servo drive and the system signal-to-noise ratio of the displacement sensor. Specifically, it is obtained by collecting the residual current of the drive end for multiple cycles under no-load and nominal operating conditions, and then weighting it by taking the upper limit of the variance of its high-frequency noise component. This effectively filters out random disturbances in trend decomposition.

[0039] Specifically, in the compensation calculation stage, the edge computing node inputs the cumulative loss status of the execution terminal and material data into the compensation calculation module. The material data is input in the form of structured numerical values ​​such as alloy composition ratio and measured yield strength. The compensation calculation module extracts a set of basic compensation coefficients from the preset compensation coefficient table according to the loss level corresponding to the cumulative loss status, and then corrects the basic compensation coefficients according to the material data to form the corrected compensation coefficients for the subsequent manufacturing process of the control node. The calculation process of the compensation calculation module is deployed on the edge computing node of the production line. The input data and intermediate results are stored in the local time-series database of the edge computing node. The finally generated corrected compensation coefficients are sent to the servo driver of the corresponding control node via the production line fieldbus to correct the system output shape and position deviation caused by the changes in material batches and the loss of the execution terminal. The calculation parameters of this compensation module do not rely on a single fixed threshold. Internally, it updates and trains coefficients according to a preset iteration cycle and new production line operation data. The training data comes from the historical residual sequence accumulated locally by the edge computing node and the paired samples of the corresponding batch material data. The training objective is to minimize the system output shape and position deviation under the action of the corrected compensation coefficient. The learning step size during the training process decreases round by round according to the preset decay rule.

[0040] The coefficient correction part of the compensation calculation module is implemented using a linear regression model. This model uses the mean of the filtered residuals corresponding to the historical residual sequence, as well as the alloy composition ratio and measured yield strength of the paired material data as input features, and the actual corrected compensation coefficient generated in the corresponding time period as the fitting target. The model loss function is the sum of squares of the differences between the predicted compensation coefficient and the fitting target. During the training process, the newly added and accumulated paired samples of the edge computing node are read according to the preset iteration cycle. The regression coefficients are updated round by round using the gradient descent method. The learning step size is set according to the exponential decay rule, that is, the learning step size of the current round is equal to the product of the initial learning step size and the preset decay coefficient raised to the power of the round. The decay coefficient ranges from 0.9 to 0.99.

[0041] The algorithm processing chain distinguishes between random noise and continuous loss trends in equipment operation, and combines the distinction results with material data for compensation calculations in the manufacturing process. This reduces the possibility of misjudging normal fluctuations as equipment abnormalities due to reliance solely on instantaneous monitoring data, and improves the accuracy of the correspondence between terminal loss judgment and manufacturing control correction.

[0042] Furthermore, the acquisition of the measured form and position deviation sequence and the establishment of the terminal performance projection boundary include: obtaining the three-dimensional spatial curve of the offline component through high-frequency radar ranging, laser contour scanning and machine vision acquisition; acquiring spatial point clouds and performing coordinate unification, outlier removal, spatial slicing and data sorting to form a measured form and position deviation sequence that continuously changes along the component trajectory vector; determining the hourly environmental incident direction based on the construction geographical latitude and azimuth angle corresponding to the target application scenario; acquiring the preset assembly pose information of the offline component in the target application scenario; mapping the measured three-dimensional spatial curve of the offline component to a global three-dimensional virtual model established based on the system architecture factors through coordinate system affine transformation to construct a virtual assembly containing the actual form and position deviation; performing illumination occlusion projection simulation on the virtual assembly based on the hourly environmental incident direction to extract the terminal performance projection boundary corresponding to the offline component in the global assembly state.

[0043] Based on the hourly environmental incident direction, the virtual assembly is subjected to illumination occlusion projection simulation. The projection paths and mutual occlusion relationships formed by the environmental incident direction on the surfaces of each component in the virtual assembly are calculated hourly. The terminal performance projection boundary corresponding to the offline component in the global assembly state is extracted. Specifically, the occlusion relationship calculation channel performs ray projection calculations on the discrete coordinate points on the surfaces of each component in the virtual assembly along the incident direction vector for each hour. It determines whether there are other components occluding the coordinate point along the projection path of the incident direction. Occluded coordinate points are marked as projection shadow points, and unoccluded coordinate points are marked as projection illumination points. The boundary extraction channel, along the boundary contour of the target application scene's area, statistically analyzes the distribution boundaries of projection shadow points and projection illumination points in adjacent areas, and connects the coordinate sequences of the boundary positions in a continuous trajectory order to form the projection boundary contour line corresponding to that hourly period.

[0044] Specifically, in the 3D data acquisition stage of the off-line components, three types of sensing devices are sequentially arranged along the component conveying direction at the production line off-line workstation: a high-frequency radar ranging unit, a laser contour scanning unit, and a machine vision acquisition unit. The high-frequency radar ranging unit is used to obtain the coarse positioning coordinates of the overall contour of the component. The laser contour scanning unit collects the cross-sectional contour data segment by segment along the component trajectory vector. The machine vision acquisition unit assists in correcting the contour loss of occluded areas through multi-angle image acquisition. The raw data collected by the three types of sensing devices are aligned by the local industrial control computer according to a unified timestamp and then fused to form spatial point cloud data covering the entire length of the component. This spatial point cloud data is stored in the format of a discrete coordinate point set.

[0045] Specifically, in the point cloud processing stage, the local industrial control computer first performs coordinate unification processing on the fused spatial point cloud, transforming the local coordinate systems of the three types of sensing devices into the production line coordinate system with the component feeding end as the origin; then, it performs outlier removal, identifying discrete points that deviate from the center of the neighboring point set beyond a preset range as outliers and removing them. For example, if a segment of the point cloud originally had 12,000 points, after outlier removal, 11,760 valid points are retained; then, it performs spatial slicing along the component trajectory vector direction, dividing the continuous point cloud into several cross-sectional slices at a fixed slice interval, and sorting the slice results according to the trajectory vector order, finally forming a measured form and position deviation sequence that continuously changes along the component trajectory vector. The physical meaning of this sequence is the deviation of the actual spatial coordinates at each trajectory position of the component from the design reference coordinates. The sequence data is stored in the local database of the production line and simultaneously uploaded to the industrial cloud platform.

[0046] Specifically, in the environmental incident direction determination stage, the industrial cloud platform retrieves the construction geographical latitude and azimuth data obtained in the analysis stage from the scenario requirement library. These two data are then input into the hourly incident direction calculation module. This module iterates through the typical time periods of the target application scenario throughout the year at fixed time intervals, outputting the corresponding environmental incident direction vector for each time period, forming a set of hourly environmental incident direction data arranged in time sequence. The physical meaning of this data is the directional distribution of the external environment acting on the target application scenario at different time periods. The calculation results are stored in the scenario requirement library of the industrial cloud platform for subsequent steps to call.

[0047] Specifically, in the virtual assembly construction phase, the industrial cloud platform pre-establishes a global 3D virtual model covering the overall skeleton of the target application scenario based on the span, bay, height, and support spacing parameters in the system architecture. This global 3D virtual model is stored as a set of coordinates of each support node and component design reference curves. Simultaneously, the platform acquires the preset assembly pose information of the offline components in the target application scenario. This preset assembly pose information includes the component's corresponding installation position number and initial installation angle in the global 3D virtual model. The platform then calls the coordinate system affine transformation module to compare the measured form and position deviation sequence with the preset assembly pose information. Using the information as input, the affine transformation module sequentially executes three consecutive sub-steps: translation alignment, rotation alignment, and scaling correction. This gradually maps the production line coordinate system of the measured 3D spatial curve to the scene coordinate system used by the global 3D virtual model. After mapping, the platform replaces the design reference curve data for the corresponding installation position in the global 3D virtual model with the transformed measured 3D spatial curve data, forming a virtual assembly containing actual shape and position deviations. The physical meaning of this virtual assembly is the digital representation of the offline components after they are installed into the overall skeleton according to their actual shape and position. The virtual assembly data is stored in the scene simulation database of the industrial cloud platform.

[0048] Specifically, in the terminal performance projection boundary extraction stage, the industrial cloud platform calls the illumination occlusion projection simulation module, using virtual assembly data and hourly environmental incident direction data as input. The illumination occlusion projection simulation module has two continuous processing stages: an occlusion relationship calculation channel and a boundary extraction channel. The occlusion relationship calculation channel calculates the projection path and mutual occlusion relationship generated by the environmental incident direction on the surface of each component of the virtual assembly hourly. Based on the occlusion relationship calculation results, the boundary extraction channel extracts the projection boundary contour line along the target application scene's area of ​​action. The simulation calculation is deployed on the server cluster of the industrial cloud platform. The input data is read from the scene simulation database, the intermediate calculation results are temporarily stored in the server memory, and the final output terminal performance projection boundary data is written back to the scene simulation database for use in the subsequent stage of determining the target manufacturing control tolerance zone. The physical meaning of this output data is the boundary influence range of the offline component on the target application scene's area of ​​action under its actual assembly state.

[0049] This data flow closed-loop processing method places the measured shape and position data of a single offline component into its actual overall assembly environment for simulation calculation. This avoids the situation of evaluating the shape and position deviation of a single component in isolation and ignoring its actual effect after assembly, thus improving the accuracy of the correspondence between the terminal performance projection boundary and the actual use state of the component.

[0050] Further, the determination of the target manufacturing control tolerance zone includes: mapping the hourly environmental incident direction to the target application scene's area of ​​action at fixed time intervals, calculating the constraint coverage area formed by the terminal performance projection boundary; determining the allowable state discrete range based on the physical performance index requirements of the terminal object; calculating the cumulative global structural deformation and constraint coverage change rate caused by the form and position deviations of different trajectory positions of the offline components in the virtual assembly; under the condition that the constraint coverage change rate satisfies the allowable state discrete range and the cumulative global structural deformation satisfies the assembly interference threshold, inverting to determine the control allowable deviation at the corresponding position, and continuously mapping the control allowable deviations corresponding to each position to form the target manufacturing control tolerance zone. The assembly interference threshold is based on the geometrical allowable clearance of the greenhouse keel node connectors (such as clamps and bolt holes) and the elastic limit of the components. Specifically, through three-dimensional finite element assembly simulation, without causing plastic yielding of the connectors or shear failure of the fasteners, the limit relative slip of each node is extracted as the basic value of this threshold.

[0051] Specifically, in the constraint coverage area calculation stage, the industrial cloud platform reads the terminal performance projection boundary data and hourly environmental incident direction data generated in the simulation stage from the scene simulation database. The coverage area calculation module uses these two data as input, traverses each time period in the hourly environmental incident direction sequence at fixed time intervals, and superimposes the terminal performance projection boundary corresponding to the time period onto the target application scene's action area plane to form the instantaneous coverage area for that time period. The module accumulates and synthesizes the instantaneous coverage areas of all time periods in chronological order, and outputs the constraint coverage area covering the target application scene's action area within a preset time range. This constraint coverage area data is stored in the form of area identifiers on the action area plane and corresponding coverage durations, and is temporarily stored in the industrial cloud platform's scene simulation database.

[0052] Specifically, in the stage of determining the allowable state discrete range, the platform retrieves the physical performance index requirements of the terminal object from the scenario requirement library. These physical performance index requirements include the target coverage duration interval for each marked position in the action area. The platform combines this target coverage duration interval with the preset allowable fluctuation range to calculate the allowable state discrete range corresponding to each marked position in the action area. The physical meaning of this allowable state discrete range is the fluctuation range that the constrained coverage area can tolerate under the premise of meeting the normal use requirements of the terminal object. The calculation result is stored in the scenario requirement library for subsequent judgment steps in this stage.

[0053] Specifically, in the calculation of global structural deformation accumulation and constraint coverage change rate, the platform calls the deformation accumulation calculation module and the coverage change rate calculation module. These two modules run in parallel and share virtual assembly data as input. The deformation accumulation calculation module reads the measured form and position deviation data of the component position by position along the trajectory vector of the lower-line component. Combining this with the connection constraint relationships between adjacent components in the virtual assembly, it calculates the structural deformation caused by the form and position deviation transmitted to the rest of the global skeleton through the connection nodes, and outputs a sequence of global structural deformation accumulation distributed along the trajectory vector. The coverage change rate calculation module, based on the constraint coverage area data, calculates the change in the constraint coverage area coverage time for each preset unit change in the component's form and position deviation, and outputs a sequence of constraint coverage change rates distributed along the trajectory vector. For example, at a certain trajectory position, after the form and position deviation increases by a preset unit, the corresponding constraint coverage change rate value increases from 0.6 to 0.9, and the global structural deformation accumulation at the same position increases from 1.1 mm to 1.8 mm.

[0054] Specifically, the coverage change rate calculation module, for each position on the trajectory vector, substitutes the coordinates of the component at that position before and after the change in shape and position deviation into the illumination occlusion projection simulation, and recalculates the projection occlusion range formed by that position in the target application scene area under the hourly incident direction of the environment. It also calculates the difference in the coverage duration of the constraint coverage area before and after the change in shape and position deviation, and uses the ratio of the difference in coverage duration to the preset unit shape and position deviation as the constraint coverage change rate at that position. This ensures that the calculation result of the constraint coverage change rate is based on the same illumination occlusion projection simulation principle and is consistent with the method of determining the projection boundary of the terminal performance.

[0055] Specifically, in the control allowable deviation inversion stage, the platform calls a dual-condition judgment module, taking the constraint coverage change rate sequence, allowable state discrete range, global structural deformation accumulation sequence, and preset assembly interference threshold as input. This module performs judgments position by position along the trajectory vector. Only when both conditions are met simultaneously—the constraint coverage change rate at that position falling within the allowable state discrete range and the global structural deformation accumulation at that position not exceeding the assembly interference threshold—is the form and position deviation value corresponding to the current judgment recorded as the control allowable deviation for that position. If a position cannot simultaneously meet both conditions within the preset deviation value range, that position is marked as awaiting manual review. After the judgment module completes the inversion position by position along the trajectory vector, it continuously maps the control allowable deviations corresponding to each position according to the trajectory vector order, outputting the target manufacturing control tolerance band. This tolerance band data is written back to the scene simulation database for subsequent manufacturing control stages.

[0056] The inversion process is implemented using a traversal search method. The dual-condition judgment module generates candidate shape and position deviation values ​​sequentially within a preset deviation value range and at a preset step size. For each candidate value, the coverage change rate calculation module and the deformation accumulation calculation module are called to recalculate the corresponding constraint coverage change rate and global structural deformation accumulation, and to determine whether the calculation result satisfies the aforementioned two conditions simultaneously. The judgment is performed in ascending order of candidate values, and the first candidate value that satisfies both conditions simultaneously is recorded as the control allowable deviation for that position. If no value that satisfies both conditions simultaneously is found after traversing all candidate values ​​within the preset deviation value range, the position is marked as pending manual review.

[0057] This inversion process incorporates both the constraint coverage change scenario and the global structural deformation accumulation scenario into the judgment conditions, avoiding the situation where only the single performance coverage requirement is met while ignoring the actual assembly feasibility after component assembly. This ensures that the formed target manufacturing control tolerance zone takes into account both the end-use effect and the dual constraints of the overall structural assembly.

[0058] Furthermore, the segmentation determination of the target manufacturing control tolerance zone includes: dividing the component into manufacturing control intervals based on the actual assembly position of the component in the global three-dimensional virtual model of the target application scenario; defining the interval located directly above the main action area and whose sensitivity to cumulative deformation of the global structure due to shape and position changes is higher than a preset threshold as a high-sensitivity control interval; defining the interval located above the secondary action area as a low-sensitivity control interval; setting the control allowable deviation as a first control tolerance threshold for the high-sensitivity control interval; setting the control allowable deviation as a second control tolerance threshold for the low-sensitivity control interval, wherein the first control tolerance threshold is less than the second control tolerance threshold; and setting a smooth transition interval between the high-sensitivity control interval and the low-sensitivity control interval, using a preset weighted smoothing function to ensure a continuous transition of the control allowable deviation between the first control tolerance threshold and the second control tolerance threshold. The sensitivity threshold is obtained by calculating the partial derivative of the unit local shape and position deviation with respect to the deflection of key nodes in the global skeleton, and is determined by back-calculation from the maximum allowable deflection-to-span ratio in the structural design specifications.

[0059] Specifically, in the manufacturing control zone segmentation stage, the industrial cloud platform reads the actual assembly position information corresponding to each offline component in the virtual assembly from the scene simulation database. The zone segmentation module uses this assembly position information and the calculated global structural deformation cumulative sequence as input. The module compares the global structural deformation cumulative sensitivity value of each position along the component trajectory vector with a preset threshold. The set of continuous positions with sensitivity values ​​higher than the preset threshold and located directly above the main action area is defined as a high-sensitivity control zone, and the set of continuous positions located above the secondary action area and with sensitivity values ​​not exceeding the preset threshold is defined as a low-sensitivity control zone. The segmentation result is output in the form of the start and end trajectory coordinates of the zone and stored in the scene simulation database. For example, the start and end coordinates of the high-sensitivity control zone divided along the entire length of a component are from 1.2 meters to 3.6 meters, and the intervals corresponding to the other trajectory coordinates are defined as low-sensitivity control zones.

[0060] Specifically, in the tiered tolerance threshold setting stage, the platform sets the allowable control deviation as the first control tolerance threshold for the defined high-sensitivity control zone and the second control tolerance threshold for the defined low-sensitivity control zone. The first control tolerance threshold is smaller than the second control tolerance threshold. Both threshold data are extracted from the corresponding positions of the target manufacturing control tolerance zones obtained through inversion. For example, the first control tolerance threshold for the high-sensitivity control zone is set to 0.8 mm, and the second control tolerance threshold for the low-sensitivity control zone is set to 2.5 mm. The values ​​of these two thresholds are based on the optical transmittance requirements and allowable deformation of the edge seal of the greenhouse covering material (such as PC board, glass, or film). The first threshold for the high-sensitivity zone is strictly constrained by the minimum assembly gap for windproof and leakproof covering material, while the second threshold for the low-sensitivity zone is relaxed to the limit deformation that does not affect the overall windproof load-bearing safety and visual flatness of the greenhouse.

[0061] Specifically, regarding the necessity of setting a smooth transition interval, if the allowable control deviation jumps directly from the first control tolerance threshold to the second control tolerance threshold at the boundary trajectory coordinates between the high-sensitivity control interval and the low-sensitivity control interval, the servo execution stroke command generated by the subsequent manufacturing control link will exhibit a command step at this boundary coordinate, thereby causing acceleration and deceleration shocks in the servo drive device within a short period of time. This shock can easily lead to surface ripples or local stress concentrations in the component near the boundary position. Therefore, the platform delineates a smooth transition interval on both sides of the boundary trajectory coordinates between the high-sensitivity control interval and the low-sensitivity control interval. The length of the transition interval is preset according to the component forming speed and servo response capability. For example, the length of the smooth transition interval is set to 0.3 meters, symmetrically distributed on both sides of the boundary trajectory coordinates.

[0062] Specifically, in the weighted smoothing function processing stage, the platform calls the weighted smoothing processing module, using the start and end trajectory coordinates of the smooth transition interval, the first control tolerance threshold, and the second control tolerance threshold as input. This module calculates the relative distance ratio between the current position and the boundary of the high-sensitivity control interval and the boundary of the low-sensitivity control interval point by point along the trajectory coordinates within the smooth transition interval. This relative distance ratio is then used as a weight to perform a weighted summation of the first and second control tolerance thresholds. Positions closer to the high-sensitivity control interval have a greater weight for the first control tolerance threshold, and positions closer to the low-sensitivity control interval have a greater weight for the second control tolerance threshold. The weighted summation result is continuously output along the trajectory coordinates, forming the control allowable deviation value that varies point by point within the smooth transition interval. For example, if the relative distance ratio between a certain coordinate point within the smooth transition interval and the boundary of the high-sensitivity control interval is 0.3, the corresponding control allowable deviation value is between 0.8 mm and 2.5 mm, closer to the 0.8 mm side. After the weighted smoothing module completes its calculations, it splices the control allowable deviation values ​​of the three segments—the high-sensitivity control interval, the smooth transition interval, and the low-sensitivity control interval—in order of trajectory coordinates to form a segmented continuous target manufacturing control tolerance band. This splicing result is written back to the scenario simulation database for use in the subsequent calculation of the collaborative control correction vector.

[0063] This segmented processing method ensures strict control accuracy at key assembly locations while avoiding abrupt changes in allowable control deviations at interval boundaries through continuous transition processing between intervals. This reduces the impact on component forming quality caused by command step changes in the servo actuator.

[0064] Furthermore, the calculation of the collaborative control correction vector includes: uploading the measured form and position deviation sequence and the target manufacturing control tolerance zone to the industrial cloud platform, and corresponding them according to the trajectory position to determine the output deviation of each control interval; the industrial cloud platform determines the target control node that needs to be adjusted based on the output deviation, and corrects the system response control coefficient of the corresponding control node based on the material data and the cumulative loss status of the execution terminal; the industrial cloud platform calculates the servo execution stroke correction value of each target control node based on the corrected system response control coefficient, and forms a control command adjustment sequence according to the node order; online closed-loop control of the servo execution stroke of the corresponding target control node, and feeding back the measured form and position status after the next manufacturing stage to the industrial cloud platform, continuously updating the control command adjustment sequence.

[0065] Specifically, in the output deviation determination stage, the production line control unit packages the measured form and position deviation sequence generated in this stage and the target manufacturing control tolerance zone generated in this stage into structured data files with trajectory coordinate labels. These files are then uploaded to the industrial cloud platform via the communication link between the production line and the cloud. The platform calls the deviation comparison module, using the two structured data files as input. This module aligns the two data files point by point according to the trajectory coordinate labels. After alignment, it calculates the difference between the measured form and position deviation value and the control allowable deviation value corresponding to the coordinate point by point. The differences are then merged and summarized according to a preset control interval division method, and the output deviation sequence distributed along each control interval is output. The physical meaning of this sequence is the degree to which the actual manufacturing result in each control interval deviates from the control allowable range. The comparison results are stored in the scenario simulation database of the industrial cloud platform. For example, the output deviation after summarizing the point-by-point differences in a certain control interval is 1.1 mm, and the output deviation in another adjacent control interval is 0.3 mm.

[0066] Specifically, in the target control node determination stage, the platform invokes the node mapping module, taking the output deviation sequence and a pre-established table of correspondence between control intervals and control nodes as input. This module maps control intervals whose output deviation exceeds a preset trigger threshold to the control nodes responsible for forming and processing those intervals, forming a list of target control nodes that need adjustment. The list data is output in the form of a list of node numbers and corresponding deviations, for use in subsequent coefficient correction stages. For example, if the control interval with an output deviation of 1.1 mm exceeds the preset trigger threshold, the mapped target control node is included in the list. The adjacent interval with an output deviation of 0.3 mm does not exceed the preset trigger threshold, and the corresponding control node is not included in the adjustment list for this round.

[0067] Specifically, in the system response control coefficient correction stage, the platform invokes the coefficient correction module, taking the target control node list, the material data determined in the stage, and the cumulative loss status of the execution terminal output by the stage as input. For each target control node in the list, this module first extracts the basic system response control coefficient corresponding to that node from the preset response coefficient library, and then sequentially superimposes the material data correction item and the cumulative loss status correction item. The material data correction item is determined based on the deviation of the current batch of material data from the calibration benchmark material data, and the cumulative loss status correction item is determined based on the loss level of the execution terminal corresponding to that node. After superimposing the two correction items, the corrected system response control coefficient of the target control node is output, and the correction result is stored in the scenario simulation database of the industrial cloud platform. For example, if the basic system response control coefficient of a certain target control node is 1.0, after superimposing the material data correction item 0.05 and the loss correction item 0.08, the corrected system response control coefficient is 1.13.

[0068] The material data correction item is calculated by multiplying the relative deviation percentage between the measured yield strength of the current batch of material data and the benchmark yield strength of the calibration benchmark material data by a preset material correction ratio coefficient. For every 1 percentage point increase in the relative deviation percentage, the material data correction item increases by 0.01. The cumulative loss status correction item is determined according to a preset loss level correction table. In this table, the first, second, and third loss statuses correspond to correction item values ​​of 0.04, 0.08, and 0.12, respectively. The higher the loss level, the larger the corresponding correction item value.

[0069] Specifically, in the servo execution stroke correction value calculation stage, the platform calls the stroke correction calculation module, using the corrected system response control coefficient and the output deviation of the corresponding control interval as inputs. For each target control node in the list, this module multiplies the output deviation of that node with the corrected system response control coefficient to obtain the servo execution stroke correction value for that node. The physical meaning of this correction value is the increment of servo execution stroke that the control node needs to adjust to reduce the current output deviation. The module sequentially concatenates the servo execution stroke correction values ​​of each node according to their arrangement on the production line, forming a control command adjustment sequence. This sequence data is then sent to the production line field controller via the production line-cloud communication link. For example, multiplying the output deviation of the aforementioned target control node (1.1 mm) by the corrected system response control coefficient (1.13) yields a servo execution stroke correction value of approximately 1.24 mm.

[0070] Specifically, in the online closed-loop control process, after receiving the control command adjustment sequence, the production line field controller writes the correction value of each servo execution stroke into the servo driver stroke register of the corresponding control node according to the node sequence. The driver then executes the corrected servo execution stroke in the next manufacturing process. After the next manufacturing process is completed, the sensing device at the off-line station re-acquires the three-dimensional spatial curve of the component in the manner described in the process and generates new measured shape and position status data. This data is fed back to the industrial cloud platform via the communication link. The platform re-substitutes the new feedback data into the deviation comparison module and aligns it with the target manufacturing control tolerance zone again. Based on this, the control command adjustment sequence is continuously updated, forming a closed loop process between manufacturing and feedback.

[0071] This collaborative control correction process sequentially links deviation location, control node matching, coefficient correction, and stroke adjustment, and incorporates the measured results after each manufacturing stage into the next round of comparison. This allows the control commands in the continuous manufacturing process to be continuously updated with the measured feedback, reducing the possibility of fixed control parameters gradually deviating from the target state during long-term continuous production.

[0072] Example 2: Furthermore, when there are two or more adjacent control nodes in the target control node list in the production line arrangement order, after the industrial cloud platform completes the preliminary calculation of the servo execution stroke correction value of each node, it also performs coupling verification and coordinated adjustment of the correction instructions between adjacent target control nodes to avoid new output deviations in the boundary trajectory segment after the correction strokes of adjacent nodes are superimposed.

[0073] Specifically, in the adjacent node coupling relationship identification stage, the industrial cloud platform calls the coupling identification module, using both the target control node list and the production line control node arrangement sequence table as input. This module traverses each target control node in the list, comparing the positional interval of adjacent nodes in the production line arrangement sequence table one by one. Node pairs with a positional interval less than a preset coupling distance threshold are marked as coupled node pairs, and a list of coupled node pairs is output. The list data is stored in the form of node number pairings for subsequent steps. The coupling distance threshold is determined by calculating the stress distribution attenuation length based on the geometric moment of inertia and material elastic modulus of the target keel cross-section using Saint-Venant's principle, i.e., the spatial distance boundary where mechanical stress is transmitted to a negligible distance. For example, if the target control node list of a certain component contains three nodes, two of which have a positional interval in the arrangement sequence table less than the preset coupling distance threshold and are marked as a coupled node pair, the third node is not included in the coupled node pair because the interval exceeds the threshold.

[0074] Specifically, in the boundary trajectory segment impact calculation stage, the platform calls the boundary impact calculation module. The module takes as input a list of coupled node pairs, the servo execution stroke correction values ​​initially calculated for each node, and the boundary trajectory coordinate interval between the two nodes. For each coupled node pair, the module converts the servo execution stroke correction values ​​of each node to the boundary trajectory coordinate interval using a preset stroke transfer coefficient. The two conversion results are then superimposed to obtain the expected superimposed deviation of the boundary trajectory coordinate interval after both nodes execute according to the initial correction values. The physical meaning of this expected superimposed deviation is the potential new form and position deviation of the boundary trajectory segment if no coordinated adjustment is made and the two nodes are corrected independently. For example, if the two conversion results for a certain coupled node pair are 0.4 mm and 0.5 mm respectively, the expected superimposed deviation of the boundary trajectory coordinate interval is 0.9 mm.

[0075] Specifically, in the collaborative adjustment phase, the platform invokes the collaborative allocation module, taking the expected superimposed deviation, the control allowable deviation value corresponding to the boundary trajectory coordinate interval, and the initial servo execution stroke correction values ​​of each of the two nodes as inputs. When the expected superimposed deviation exceeds the control allowable deviation value corresponding to the boundary trajectory coordinate interval, the module applies a reduction to the two initial correction values ​​according to the proportional relationship between the initial correction values ​​of the two nodes. After the reduction, the expected superimposed deviation of the boundary trajectory coordinate interval is recalculated, and the reduction and recalculation process is repeated until the expected superimposed deviation does not exceed the control allowable deviation value corresponding to the interval, or the number of reductions reaches a preset upper limit. After the reduction is completed, the module outputs the collaboratively corrected servo execution stroke correction values ​​of each node in the coupled node pair, replacing the original initial correction values ​​of the corresponding nodes in the generated control instruction adjustment sequence. For example, the aforementioned expected superposition deviation of 0.9 mm exceeds the control allowable deviation of 0.7 mm corresponding to the boundary trajectory coordinate interval. After two rounds of reduction, the servo execution stroke correction values ​​of the two nodes are adjusted from 0.4 mm and 0.5 mm to 0.32 mm and 0.4 mm, respectively. At this time, the expected superposition deviation of the boundary trajectory coordinate interval is reduced to 0.65 mm, which does not exceed the control allowable deviation value.

[0076] Specifically, in the collaborative correction instruction issuance stage, the industrial cloud platform sends the adjusted sequence of control instructions after collaborative adjustment to the production line field controller via the communication link. The field controller writes the collaboratively corrected servo execution stroke correction value of each node into the corresponding servo driver according to the node sequence for use in the next manufacturing process. After the next manufacturing process is completed, the measured shape and position status of the boundary trajectory coordinate interval is also sent back to the industrial cloud platform in the same feedback method for reference and call in subsequent rounds of coupling verification, continuously correcting the preset stroke transmission coefficient value.

[0077] This collaborative adjustment method incorporates the superimposed impact of the correction travel of each node on the boundary trajectory segment into the calculation scope and performs collaborative reduction when multiple adjacent target control nodes need to be corrected simultaneously. This reduces the situation where new deviations occur in the boundary trajectory segment after independent correction of adjacent nodes, and improves the consistency of the overall shape and position control results of components in multi-node joint correction scenarios.

[0078] Furthermore, in order to address the technical bottleneck of interference and superimposed oscillations that easily occur when multi-level control nodes are coordinated for correction, a multi-actuator decoupling control law based on model predictive control and constrained least squares method is introduced.

[0079] This strategy abandons simple proportional reduction lacking safety boundaries and reconstructs the underlying multi-servo collaborative closed loop while considering cloud-edge communication latency and system saturation characteristics. Specifically, the field edge controller first identifies the cross-coupling control matrix between multi-level roller stations online based on historical operating data, thereby quantifying the strength of geometric interference generated by adjacent control nodes at trajectory boundaries. When the cloud platform issues the initial output deviation, the edge controller does not directly perform scalar multiplication calculations, but instead constructs a constrained least squares optimization problem that includes the actuator's maximum stroke, stroke change rate limit, and anti-saturation dead zone. During the control law solution process, the system uses the worst-case latency of the industrial communication network as the state variable of the model predictive control for feedforward compensation, predicting the error evolution trend of multiple future control cycles. The constructed cross-coupling control matrix is ​​used to jointly solve the correction commands of each target control node, ensuring that the servo execution stroke correction value output by the multi-node collaboration reaches the optimal physical allocation under global constraints. Once an edge computing node detects a solution divergence caused by a cloud network outage, excessive packet loss, or actuator deep saturation, the underlying security process instantly takes over, executing a smooth shutdown or automatically rolling back to the previous stable instruction set according to the fault-tolerant contingency plan. This paradigm shift from independent feedback from a single node to decoupled control of multiple variables enables the control command adjustment sequence to truly possess determinism for handling complex industrial scenarios. This closed-loop control structure, with communication delay compensation and multi-node decoupling, avoids secondary positional errors caused by the mechanical superposition of correction commands from adjacent nodes.

[0080] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for online monitoring of a keel production line based on industrial cloud computing, characterized in that, The method includes: Obtain the scene constraint data corresponding to the target farm, and determine the process configuration type and material compatibility requirements of the production line materials based on the scene constraint data; Obtain the actual material process attributes and mechanical state indication values, wherein the actual material process attributes include material data; match the actual material process attributes with the process configuration type, call the corresponding process disturbance preset parameter set and control execution compensation parameter set, and control and correct the system impedance changes and output response deviations during continuous manufacturing process; Collect servo execution stroke commands and drive end status monitoring data from control nodes at all levels, and determine the cumulative loss status of the execution terminal based on the load impedance changes of control nodes at all levels and the equipment operating cycle; The three-dimensional spatial curves of the offline components are collected, and the measured form and position deviation sequence is extracted along the trajectory vector of the offline components. The industrial cloud platform is used to run simulation calculations based on the scenario constraint data to obtain the target manufacturing control tolerance zone. The collaborative control correction vector for the next stage of manufacturing process is obtained by comparison and calculation.

2. The online monitoring method for a keel production line based on industrial cloud computing according to claim 1, characterized in that, The scenario constraint data includes terminal receptor factors, system architecture factors, and service environment factors. The terminal receptor factors include receptor type, environmental indicator requirements, and layout method. The system architecture factors include span parameters, bay parameters, height parameters, and support spacing. The service environment factors include construction latitude, azimuth, and external load conditions. Based on the terminal receptor factors, the control performance requirements, global load-bearing requirements, and local node load-bearing requirements corresponding to the target component are determined. Based on the system architecture factors, the span adaptation requirements, spatial manufacturing requirements, and component deformation sensitivity corresponding to the target component are determined. Based on the service environment factors, the corrosion resistance environment requirements, external load requirements, and temperature and humidity service requirements corresponding to the target component are determined in advance.

3. The online monitoring method for a keel production line based on industrial cloud computing according to claim 1, characterized in that, Determining the process configuration type and material compatibility requirements of production line materials based on the scenario constraint data includes: Based on the corrosion resistance requirements, external load requirements, temperature and humidity service requirements, and manufacturing deformation requirements of the target component, production control adaptation evaluations are performed on different candidate material attributes. The process configuration type is determined based on the evaluation results, and a set of preset process disturbance parameters, a set of control execution compensation parameters, and a mechanical strength adaptation range corresponding to the process configuration type are established. When the actual material process attributes, including material data, obtained from the production line material import end are consistent with the process configuration type, the preset process disturbance parameters and the set of control execution compensation parameters corresponding to the process configuration type are called to correct the manufacturing process control parameters. When the actual material process attributes are inconsistent with the process configuration type, the control strategy for the current batch of manufacturing raw materials is adaptively reconstructed according to the material adaptation requirements.

4. The online monitoring method for a keel production line based on industrial cloud computing according to claim 1, characterized in that, The cumulative wear status of the execution terminal is determined in the following manner: Adaptive filtering is performed on the servo execution stroke commands and drive end status monitoring data of each level of control node to construct a residual sequence between the actual execution impedance and the nominal execution impedance; the residual sequence is decomposed in combination with the historical operating cycle of the production line equipment to extract the terminal load change trend as the equipment operating time increases, and the cumulative loss status quantity of the execution terminal is determined based on the terminal load change trend. The cumulative loss status of the execution terminal and the material data are input into the compensation calculation process to correct the system output shape and position deviation caused by material batch changes and execution terminal losses.

5. The online monitoring method for a keel production line based on industrial cloud computing according to claim 1, characterized in that, The acquisition of the measured form and position deviation sequence and the establishment of the terminal performance projection boundary include: The three-dimensional spatial curves of the offline components are obtained through high-frequency radar ranging, laser contour scanning, and machine vision acquisition. Spatial point clouds are acquired and subjected to coordinate unification, outlier removal, spatial slicing, and data sorting to form a measured shape and position deviation sequence that continuously changes along the component trajectory vector. The hourly environmental incident direction is determined based on the construction latitude and azimuth of the target application scenario. The preset assembly pose information of the offline components in the target application scenario is obtained. Through coordinate system affine transformation, the measured three-dimensional spatial curves of the offline components are mapped to a global three-dimensional virtual model established based on system architecture factors to construct a virtual assembly containing actual shape and position deviations. Based on the hourly environmental incident direction, the virtual assembly is subjected to illumination occlusion projection simulation to extract the terminal performance projection boundary corresponding to the offline components in the global assembly state.

6. The online monitoring method for a keel production line based on industrial cloud computing according to claim 5, characterized in that, The determination of the target manufacturing control tolerance zone includes: The hourly environmental incident direction is mapped to the target application scene's area of ​​action at fixed time intervals, and the constraint coverage area formed by the terminal performance projection boundary is calculated. The allowable state discrete range is determined according to the physical performance index requirements of the terminal object. The cumulative amount of global structural deformation and the constraint coverage change rate caused by the form and position deviations of different trajectory positions of the offline components in the virtual assembly are calculated. Under the condition that the constraint coverage change rate meets the allowable state discrete range and the cumulative amount of global structural deformation meets the assembly interference threshold, the control allowable deviation at the corresponding position is determined by inversion. The control allowable deviations corresponding to each position are continuously mapped to form the target manufacturing control tolerance zone.

7. The online monitoring method for a keel production line based on industrial cloud computing according to claim 1, characterized in that, The calculation of the cooperative control correction vector includes: The measured form and position deviation sequence and the target manufacturing control tolerance zone are uploaded to the industrial cloud platform and matched according to the trajectory position to determine the output deviation of each control interval. The industrial cloud platform determines the target control node that needs to be adjusted based on the output deviation and corrects the system response control coefficient of the corresponding control node based on the material data and the cumulative loss status of the execution terminal. The industrial cloud platform calculates the servo execution stroke correction value of each target control node based on the corrected system response control coefficient and forms a control command adjustment sequence according to the node order. The servo execution stroke of the corresponding target control node is controlled online in a closed loop, and the measured form and position status after the next manufacturing stage is fed back to the industrial cloud platform to continuously update the control command adjustment sequence.