Tension measurement monitoring management analysis system control method for tension control index

Abstract

The application relates to the fields of mechanical engineering and automatic control technology, and discloses a control method of a tension control index of a tension measurement monitoring management analysis system, which comprises the following steps: collecting multi-measurement-point tension time sequence data in real time through a high-precision tension sensor array; performing millisecond-level abnormality identification and reporting based on an LSTM model by an edge computing node; matching a control strategy according to a process rule library, and dynamically adjusting the rotating speed and the torque of a main shaft by a central control unit; rapidly responding by an executing mechanism, so that the tension returns to a target interval within a threshold time; meanwhile, abnormal context data is stored into a time sequence database, and a cloud management analysis module uses a graph neural network to mine a root cause, generates an optimization suggestion, and updates the rule library. Through the establishment of a full-link closed-loop control system from tension sensing, edge intelligent judgment, central decision, device execution to cloud analysis, the data flow breakage between a monitoring system and a management analysis module is completely eliminated.

Description

Technical Field

[0001] This invention belongs to the field of mechanical engineering and automatic control technology, specifically relating to a method for controlling tension control indicators in a tension measurement, monitoring, management and analysis system. Background Technology

[0002] In industrial manufacturing, materials processing, and automated production lines, tension control is a crucial step in ensuring product quality and process stability. Tension measurement and monitoring systems use sensors to collect real-time tension data from media such as rolled materials, films, or fibers during operation, providing fundamental feedback to process equipment. However, most current systems only achieve one-way data acquisition and visualization, lacking effective linkage with upper-level management and analysis modules.

[0003] This architectural fragmentation prevents the timely parsing and conversion of abnormal tension events (such as instantaneous changes, continuous over-limits, or periodic fluctuations) captured by the monitoring layer into executable control commands, thus missing critical opportunities to intervene in process parameters (such as winding speed, traction force, or unwinding tension settings) within millisecond to second time windows.

[0004] Dynamic adjustment of tension control parameters relies on a comprehensive analysis of historical trends, real-time status, and process rules. Existing technologies typically separate monitoring, analysis, and control functions into independent subsystems, with communication between modules via low-frequency, unstructured data interfaces, resulting in a significant "perception-decision-execution" delay. When tension anomalies occur, the management and analysis module often only identifies the root cause during the post-event batch processing phase. By this time, the production line has already produced a large number of non-conforming products, leading to raw material waste, equipment downtime, and delivery delays.

[0005] Tension control systems generally suffer from three major drawbacks: data flow interruption, response lag, and lack of closed-loop control. Monitoring data is not injected into the analysis engine in real time, leading to anomaly detection relying on manual experience or static thresholds. Analysis results cannot directly drive the underlying controller, requiring multiple layers of manual confirmation or intermediate system conversion. Control strategies have long update cycles, making it difficult to adapt to dynamic tension requirements under different material properties, operating speeds, or environmental temperature and humidity variations. These problems are significantly amplified, especially in high-speed continuous production scenarios, resulting in persistently high tension-related scrap rates and severely restricting the core requirements of intelligent manufacturing systems for high precision, high efficiency, and high reliability. Summary of the Invention

[0006] This invention provides a method for controlling tension control indicators in a tension measurement, monitoring, management and analysis system. By constructing an end-to-end closed-loop data flow architecture, it integrates real-time tension monitoring, abnormal event identification, process parameter linkage control and historical data analysis, eliminating information silos between the monitoring system and the management and analysis module. This ensures that dynamic adjustments to the operating parameters of the winding equipment can be triggered immediately when an abnormal tension event is detected, thereby avoiding batch product scrapping due to response lag.

[0007] This invention provides a method for controlling tension control indicators in a tension measurement, monitoring, management, and analysis system, comprising:

[0008] The tension time-series data of each key station during the winding process is collected in real time by a tension sensor array. The tension time-series data includes the instantaneous tension value, tension change rate and tension fluctuation spectrum characteristics of multiple measuring points distributed along the material travel direction.

[0009] The tension time-series data is input to the edge computing node, which performs real-time tension status assessment and generates a tension status label sequence containing three levels of status identifiers: normal, warning, and abnormal.

[0010] When an abnormal status identifier appears in the tension status label sequence, the edge computing node immediately sends an abnormal event message to the central control unit, which includes the time of the abnormality, the coordinates of the abnormal location, the abnormality type code, and the current tension value.

[0011] After receiving the abnormal event message, the central control unit matches the corresponding control strategy according to the preset process rule library and generates a set of control instructions for the winding spindle motor speed and unwinding braking torque.

[0012] The control instruction set is sent to the actuator controller, which drives the winding equipment to perform corresponding parameter adjustment actions so that the material tension returns to the target control range within the preset recovery time threshold.

[0013] Simultaneously, the central control unit packages the complete context data of this abnormal event, including tension timing data, equipment operating parameters, ambient temperature and humidity data, and control command execution logs within the preset time window before the abnormality, and stores them in the historical database, and triggers the management analysis module to start the root cause analysis process.

[0014] The management and analysis module uses multi-dimensional correlation data in the historical database and a causal reasoning model based on graph neural networks to identify potential process factors that cause abnormal tension and generate process optimization suggestions. These suggestions are automatically injected into the process rule base for updating subsequent control strategies.

[0015] In one embodiment of the present invention, the tension sensor array is a high-precision strain gauge tension sensor with a sampling frequency of not less than 1000 Hz, a range covering 0 to 500 Newtons, and a nonlinear error of less than 0.1% of the full scale. It is installed at the bearing seats of the unwinding roller, traction roller, and take-up roller of the winding machine, and is fastened by bolts and connected to the analog signal input port of the edge computing node via a shielded twisted pair cable.

[0016] In one embodiment of the present invention, the edge computing node is an embedded industrial computer, which is internally deployed with a tension state evaluation model. The tension state evaluation model is a time-series classifier based on a long short-term memory network. The input layer of the classifier receives tension time-series data for 100 consecutive sampling periods. The hidden layer contains two layers of long short-term memory units, each with 128 units. The output layer generates a three-level state probability distribution through a fully connected layer and a Softmax function. When the probability of an abnormal state exceeds a preset threshold of 0.9, it is determined to be an abnormal state.

[0017] As one embodiment of the present invention, the abnormality type code includes four categories: tension drop, tension rise, high-frequency tension oscillation, and continuous tension deviation. Each type of abnormality corresponds to a unique eight-bit binary code. Tension drop is defined as the tension value decreasing by more than 30% of the set reference value within 10 sampling periods. Tension rise is defined as the tension value increasing by more than 25% of the set reference value within 10 sampling periods. High-frequency tension oscillation is defined as the energy proportion of the tension signal in the 20 Hz to 50 Hz frequency band exceeding 40% of the total signal energy. Continuous tension deviation is defined as the tension value deviating from the center value of the target control range by more than ±10% for more than 100 consecutive sampling periods.

[0018] In one embodiment of the present invention, the process rule library is stored in the non-volatile memory of the central control unit. It is organized in the form of key-value pairs, where the key is an exception type code and the value is a set of control instructions. The set of control instructions includes two parameters: the percentage reduction of the winding spindle motor speed and the increase of the unwinding braking torque. The percentage reduction of the winding spindle motor speed ranges from 5% to 20%, and the increase of the unwinding braking torque ranges from 0.5 Nm to 3 Nm.

[0019] In one embodiment of the present invention, the actuator controller communicates with the central control unit via industrial Ethernet, adopts a real-time transmission protocol, and has a communication cycle of 1 millisecond. It has an internal instruction verification and execution feedback mechanism to ensure that the control instruction is parsed and output to the servo driver within 5 milliseconds after being received, and at the same time, the execution result is sent back to the central control unit.

[0020] As one embodiment of the present invention, the historical database adopts a time-series database architecture and is stored in time-series partitions. Each abnormal event record includes a unified timestamp, a unique device identifier, a tension time-series data block, a snapshot of device operating parameters, environmental data, and control logs. The data retention period is no less than 360 days.

[0021] In one embodiment of the present invention, the management and analysis module is deployed on a cloud server. It periodically pulls new abnormal event records from the historical database, constructs a heterogeneous graph structure with abnormal events as nodes and process parameters as edges, uses a graph attention network to learn the dependencies between nodes, and outputs the ranking of the contribution of each process parameter to the occurrence of the abnormality. Parameters with a contribution higher than a preset threshold of 0.7 are marked as key factors. The process optimization suggestion is to set stricter control tolerances for key factors or adjust their baseline settings.

[0022] This invention provides a control device for tension control indicators in a tension measurement, monitoring, management, and analysis system, comprising:

[0023] The tension data acquisition unit is used to acquire tension timing data of each key station during the winding process in real time through a tension sensor array;

[0024] An edge state evaluation unit is used to input the tension time series data into the edge computing node, perform real-time tension state evaluation, and generate a tension state label sequence.

[0025] An abnormal event reporting unit is used to send an abnormal event message to the central control unit when an abnormal state identifier appears in the tension state tag sequence.

[0026] The control strategy matching unit is used to match the corresponding control strategy according to the process rule library by the central control unit and generate a control instruction set.

[0027] The execution instruction issuing unit is used to issue the control instruction set to the actuator controller to drive the winding equipment to adjust the execution parameters;

[0028] The data archiving and analysis triggering unit is used to store the context data of abnormal events into the historical database and trigger the management and analysis module to start the root cause analysis process.

[0029] The process rule update unit is used to generate process optimization suggestions from the management and analysis module and inject them into the process rule library.

[0030] In one embodiment of the present invention, the tension sensor array in the tension data acquisition unit is connected to the analog input interface of the edge computing node through a shielded twisted pair cable. The interface supports 16-bit analog-to-digital conversion, and the sampling synchronization is achieved by a hardware trigger signal to ensure that the time alignment error of the data at each measuring point is less than 1 microsecond.

[0031] As one embodiment of the present invention, the tension state assessment model deployed by the edge state assessment unit loads the latest version from the central control unit each time the system starts, and performs online hot replacement after receiving the model update package pushed by the management and analysis module. The update process does not affect the execution of the real-time assessment task.

[0032] In one embodiment of the present invention, the control strategy matching unit considers the category code of the current winding material when matching the control strategy. The category code is provided by the material identification system. Different category codes correspond to different process rule sub-libraries to ensure that the control strategy is compatible with the material characteristics.

[0033] As one embodiment of the present invention, the execution instruction issuing unit performs instruction legality verification before issuing the control instruction set, including parameter range checking, instruction conflict detection and device status confirmation, and only allows the instruction to be issued after the verification is passed.

[0034] As one embodiment of the present invention, the data archiving and analysis triggering unit automatically adds operator identification and shift information when packaging abnormal event context data for subsequent human factors analysis.

[0035] As one embodiment of the present invention, before injecting process optimization suggestions, the process rule update unit needs to conduct a virtual trial run through the simulation verification module built into the central control unit to verify that the suggestions will not cause new process conflicts in the digital twin model. Only after the verification is passed can the suggestions be written into the process rule library.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0037] 1. This invention completely eliminates the data flow interruption between the monitoring system and the management and analysis module by establishing a closed-loop control system that extends from tension sensing, edge intelligent judgment, central decision-making, equipment execution to cloud analysis.

[0038] 2. Within milliseconds of an abnormal tension event, the system can complete the entire process of identification, decision-making, and execution, ensuring that key process parameters such as winding speed are adjusted in a timely manner. This effectively prevents material breakage, wrinkling, or interlayer misalignment caused by uncontrolled tension, and significantly reduces the scrap rate of batch products.

[0039] 3. By feeding back the handling results of each abnormal event to the management and analysis module and using graph neural networks for root cause analysis, continuous self-optimization of the process rule base is achieved, enabling the system to have long-term evolution capabilities. Furthermore, the collaborative design of edge computing nodes and the central control unit reduces the computational burden on the central system while ensuring real-time performance, improving the overall architecture's reliability and scalability. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the overall technical solution architecture of the tension measurement, monitoring, management and analysis system proposed in this invention for controlling tension control indicators;

[0041] Figure 2 This is a schematic diagram of the core principle framework of the closed-loop response mechanism for tension anomaly events based on the collaboration between edge computing nodes and the central control unit in this invention.

[0042] Figure 3 This is a logical flowchart of the tension timing data acquisition, state assessment, and abnormal event message generation in this invention.

[0043] Figure 4 This is a flowchart illustrating the logical flow of the central control unit in this invention, which matches control strategies based on the process rule library and issues execution instructions.

[0044] Figure 5 This is a logical flowchart of the root cause reasoning process of the abnormal event context data archiving and cloud management analysis module in this invention;

[0045] Figure 6 This is a schematic diagram of the multi-level interaction relationship and data flow between the terminal device, edge computing node, central control unit and cloud management and analysis module in this invention. Detailed Implementation

[0046] Please refer to Figures 1 to 6 This invention provides a method for controlling tension control indicators using a tension measurement, monitoring, management, and analysis system, aiming to solve the problem of data flow disconnect between the monitoring system and the management and analysis module in existing winding processes. In traditional solutions, although tension anomalies can be detected by sensors, the lack of a closed-loop linkage mechanism prevents timely dynamic adjustments to the winding equipment's operating parameters. This results in the material continuing to operate under uncontrolled tension, leading to defects such as breakage, wrinkles, or interlayer misalignment, causing batch product scrapping. To overcome these shortcomings, this invention constructs an end-to-end closed-loop data flow architecture, integrating real-time tension monitoring, anomaly event identification, process parameter linkage control, and historical data analysis to ensure seamless connection throughout the entire chain from sensing to execution to optimization.

[0047] The method includes the following steps:

[0048] S1, through a tension sensor array, collects tension timing data of each key station in real time during the winding process;

[0049] S2, The tension time series data is input to the edge computing node, and the edge computing node performs real-time tension status evaluation to generate a tension status label sequence containing three levels of status identifiers: normal, warning, and abnormal.

[0050] S3, when an abnormal status identifier appears in the tension status label sequence, the edge computing node immediately sends an abnormal event message to the central control unit, which includes the time of the abnormality, the coordinates of the abnormal location, the abnormality type code, and the current tension value.

[0051] S4, after receiving the abnormal event message, the central control unit matches the corresponding control strategy according to the preset process rule library and generates a set of control instructions for the winding spindle motor speed and unwinding braking torque.

[0052] S5, the control instruction set is sent to the actuator controller, and the actuator controller drives the winding equipment to perform corresponding parameter adjustment actions so that the material tension returns to the target control range within the preset recovery time threshold;

[0053] S6. Simultaneously, the central control unit packages the complete context data of this abnormal event into the historical database and triggers the management analysis module to start the root cause analysis process.

[0054] S7, the management and analysis module, based on the multi-dimensional correlation data in the historical database, uses a causal reasoning model based on graph neural networks to identify potential process factors that cause abnormal tension and generate process optimization suggestions. The process optimization suggestions are automatically injected into the process rule base for updating subsequent control strategies.

[0055] In step S1, the tension sensor array is a high-precision strain gauge tension sensor with a sampling frequency of not less than 1000 Hz, a measurement range covering 0 to 500 Newtons, and a nonlinearity error of less than 0.1% of full scale. This sensor array is installed at the bearing seats of the unwinding roller, traction roller, and take-up roller of the winding machine, and is rigidly fastened using high-strength bolts to ensure no mechanical loosening or signal drift occurs during high-speed winding. The output of each sensor is connected to the analog signal input port of the edge computing node via a shielded twisted-pair cable, with the shielding layer grounded at a single point to suppress electromagnetic interference.

[0056] All sensors share the same hardware trigger signal for synchronous sampling. This trigger signal is generated by a timer module inside the edge computing node, with a period accurate to 1 millisecond, thus ensuring that the time alignment error of the tension data at each measuring point is less than 1 microsecond. The collected tension time-series data not only includes instantaneous tension values, but also calculates the tension change rate and tension fluctuation spectrum characteristics in real time through the built-in digital signal processing unit.

[0057] The tension change rate is defined as the difference between the tension values ​​of two adjacent sampling periods divided by the sampling interval, in Newtons per millisecond. The tension fluctuation spectrum characteristics are obtained by performing a fast Fourier transform on 256 consecutive sampling points, with a focus on the energy proportion in the 20 Hz to 50 Hz frequency band, which corresponds to the high-frequency disturbances caused by mechanical resonance or unevenness of the roller surface during high-speed movement of the material.

[0058] In step S2, the edge computing node is an embedded industrial computer with a quad-core ARM processor, a clock speed of no less than 1.6 GHz, 4 gigabytes of memory, and a 32-gigabyte industrial-grade solid-state drive for storage. The node internally deploys a tension state assessment model, which is a time-series classifier based on a long short-term memory network. The input layer of this classifier receives tension time-series data from 100 consecutive sampling periods, forming a 100-dimensional input vector; the hidden layer contains two layers of long short-term memory units, each with 128 units, and uses a gating mechanism to model long-term dependencies; the output layer is connected to a Softmax function through a fully connected layer to generate probability distributions for normal, warning, and abnormal states.

[0059] During model training, a historical labeled dataset is used, with anomalous samples covering four typical operating conditions: sudden tension drop, sudden tension increase, high-frequency tension oscillation, and continuous tension shift. During inference, when the probability of an anomalous state exceeds a preset threshold of 0.9, the current state is determined to be anomalous, and a corresponding anomalous state identifier is generated. This evaluation process is completed within 500 microseconds after the end of each sampling period, ensuring real-time state determination. Furthermore, each time the system powers on, the edge computing nodes download the latest version of the tension state evaluation model from the central control unit and load it into memory after verifying its integrity. When the management and analysis module pushes a model update package, the edge computing nodes load the new model in the background and seamlessly switch over after the current evaluation task is completed, achieving online hot-swap without affecting the continuous execution of real-time evaluation tasks.

[0060] In step S3, once an anomaly is marked in the tension status tag sequence, the edge computing node immediately constructs an anomaly event message. This message uses a fixed-length binary format, with a total length of 64 bytes, and includes the following fields: the time of the anomaly (represented by a microsecond-level timestamp, occupying 8 bytes), the coordinates of the anomaly location (represented by the workstation number, such as 01 for the unwinding roller, 02 for the traction roller, and 03 for the winding roller, occupying 2 bytes), the anomaly type code (eight-bit binary code, occupying 1 byte), the current tension value (represented as a floating-point number, occupying 4 bytes), the tension change rate (occupying 4 bytes), the high-frequency energy percentage (occupying 4 bytes), and the remaining bytes are checksum and reserved fields. The exception type codes are defined as follows: Sudden tension drop (code 00000001) is defined as a tension value decreasing by more than 30% of the set reference value within 10 sampling periods; sudden tension increase (code 00000010) is defined as a tension value increasing by more than 25% of the set reference value within 10 sampling periods; high-frequency tension oscillation (code 00000100) is defined as the energy proportion of the tension signal in the 20Hz to 50Hz frequency band exceeding 40% of the total signal energy; continuous tension deviation (code 00001000) is defined as the tension value deviating from the center value of the target control range by more than ±10% for more than 100 consecutive sampling periods. This message is sent to the central control unit via the industrial Ethernet interface in the highest priority queue. The transmission protocol uses a time-sensitive network mechanism to ensure that the message arrives within 100 microseconds.

[0061] In step S4, the central control unit is a high-performance industrial server equipped with redundant power supplies and dual gigabit Ethernet interfaces. It internally stores a process rule base, which is stored in non-volatile memory and organized in key-value pairs. The key is the exception type code, and the value is the control instruction set. The control instruction set contains three core parameters: the reduction ratio of the winding spindle motor speed and the increase in unwinding braking torque. The reduction ratio of the winding spindle motor speed ranges from 5% to 20%, with the specific value dynamically selected based on the severity of the exception; the increase in unwinding braking torque ranges from 0.5 Nm to 3 Nm, used to compensate for tension relaxation caused by the speed reduction.

[0062] Upon receiving an anomaly event message, the central control unit first parses the anomaly type code and then queries the process rule library to obtain the corresponding control instruction set. Simultaneously, the central control unit retrieves the category code of the currently wound material from the material identification system. This code is read by an RFID tag or barcode scanner, and different category codes correspond to different process rule sub-libraries. For example, film materials correspond to sub-library A, paper materials to sub-library B, and metal foil to sub-library C. The control strategy matching process comprehensively considers both the anomaly type and the material category to ensure that the instruction set is compatible with the material's physical properties. After matching is complete, the central control unit generates a complete control instruction set and attaches a unique transaction identifier and timestamp.

[0063] In step S5, the control command set is sent to the actuator controller via industrial Ethernet. This controller is a dedicated motion control module that supports real-time transmission protocols with a communication cycle of 1 millisecond. Upon receiving the command, the actuator controller first performs command validity checks, including parameter range checks (ensuring the speed reduction ratio is between 5% and 20%), command conflict detection (e.g., avoiding simultaneous issuance of large deceleration and large acceleration commands), and equipment status confirmation (confirming that the winding spindle is running and there is no emergency stop signal). Only when all three checks pass is the command allowed to enter the execution queue. The actuator controller completes parsing within 5 milliseconds of receiving the command and outputs the control parameters to the servo drive.

[0064] The servo drive adjusts the motor speed, brake current, and electric push rod position according to instructions, thereby changing the winding spindle speed and unwinding braking torque. After execution, the actuator controller sends the actual execution results (including execution time, parameter effective values, and equipment feedback status) back to the central control unit, forming a closed-loop feedback. The goal of the entire control process is to bring the material tension back to the target control range within a preset recovery time threshold. This threshold is set according to the material type and is typically between 50 milliseconds and 200 milliseconds.

[0065] In step S6, the central control unit initiates the abnormal event context data archiving process simultaneously with issuing control commands. The archived content includes: complete tension time-series data within a preset time window before the abnormality (default 500 milliseconds), snapshots of equipment operating parameters (including spindle speed, unwinding tension setpoint, ambient temperature and humidity, and material category code), control command execution logs (including issuance time, execution time, and actual effective value), and operator identification and shift information. All data is aligned with a unified timestamp and packaged into a single record for writing to the historical database. The historical database adopts a time-series database architecture, based on a columnar storage engine, with data partitioned by day, supporting efficient time-range querying and compressed storage. Each abnormal event record contains a unique equipment identifier, facilitating cross-equipment comparative analysis. The data retention period is no less than 360 days, meeting the requirements for process traceability and quality auditing. After the data is written, the central control unit sends an analysis trigger signal to the management and analysis module to initiate the root cause analysis process.

[0066] In step S7, the management and analysis module is deployed on a cloud server cluster, possessing elastic computing capabilities. This module periodically polls the historical database, retrieving newly added anomaly event records. For each record, the management and analysis module constructs a heterogeneous graph structure: with anomaly events as nodes and process parameters (such as spindle speed, ambient temperature, and material thickness) as edges, the edge weights determined by the correlation coefficient between the parameters and the anomaly. Subsequently, a graph attention network is used to learn this graph. The network aggregates neighbor node information through a multi-head attention mechanism and calculates the contribution of each process parameter to the occurrence of the anomaly. The contribution calculation formula is as follows:

[0067] ;

[0068] in, For the first Attention score for each process parameter For the first Attention score for each process parameter This is a set of all relevant parameters. Parameters whose contribution exceeds a preset threshold of 0.7 are marked as key factors. Based on the set of key factors, the management analysis module generates process optimization suggestions, such as "tighten the ambient temperature control tolerance from ±5 degrees Celsius to ±3 degrees Celsius" or "reduce the baseline tension setting of thin film materials by 5%". Before the suggestions are injected into the process rule base, they must undergo virtual trial operation through the simulation verification module built into the central control unit. This module, based on digital twin technology, constructs a physical simulation model of the winding process. After inputting the optimization suggestions, it simulates the operating state of the next 100 winding cycles to detect whether new tension anomalies or equipment over-limits are caused. Only when the anomaly rate decreases in the simulation results and there are no new conflicts is the suggestion approved for inclusion in the process rule base, completing the strategy update.

[0069] Through the coordinated execution of the above seven steps, this invention achieves fully closed-loop control from tension anomaly detection to process self-optimization. The system completes anomaly response within milliseconds, effectively preventing batch scrap; simultaneously, through cloud-based root cause analysis and rule base self-updating, the system possesses continuous evolution capabilities, adapting to constantly changing production environments and material properties.

Claims

1. A method for controlling tension control indicators in a tension measurement, monitoring, management, and analysis system, characterized in that: include: The tension timing data of each key station during the winding process is collected in real time by a tension sensor array. The tension time-series data is input to an edge computing node, which performs real-time tension status assessment and generates a tension status label sequence containing three levels of status identifiers: normal, warning, and abnormal. Tension time-series data for 100 consecutive sampling periods are input into a time-series classifier based on a long short-term memory network. The time-series classifier contains two layers of long short-term memory units, with 128 units in each layer. The three-level state probability distribution is output through a fully connected layer and a Softmax function. When the probability of an abnormal state exceeds a preset threshold of 0.9, it is determined to be an abnormal state and a corresponding abnormal state identifier is generated. When an abnormal status identifier appears in the tension status label sequence, the edge computing node immediately sends an abnormal event message to the central control unit, which includes the time of the abnormality, the coordinates of the abnormal location, the abnormality type code, and the current tension value. After receiving the abnormal event message, the central control unit matches the corresponding control strategy according to the preset process rule library and generates a set of control instructions for the winding spindle motor speed and unwinding braking torque. The control instruction set is sent to the actuator controller, which drives the winding equipment to perform corresponding parameter adjustment actions so that the material tension returns to the target control range within the preset recovery time threshold. Simultaneously, the central control unit packages the complete context data of this abnormal event, including tension timing data, equipment operating parameters, ambient temperature and humidity data, and control command execution logs within the preset time window before the abnormality, and stores them in the historical database, and triggers the management analysis module to start the root cause analysis process. The management analysis module, based on multi-dimensional correlation data in the historical database, uses a causal reasoning model based on graph neural networks to identify potential process factors causing abnormal tension and generate process optimization suggestions, including: Construct a heterogeneous graph structure with abnormal events as nodes and process parameters as edges, and use a graph attention network to calculate the contribution of each process parameter to the occurrence of the abnormality. Parameters with a contribution rate higher than a preset threshold of 0.7 are marked as key factors, and process optimization suggestions are generated based on the key factors; The process optimization suggestions are automatically injected into the process rule base for updating subsequent control strategies. The process optimization suggestions are automatically injected into the process rule base for updating subsequent control strategies, including: The process optimization suggestions are input into the simulation verification module built into the central control unit, and a virtual trial run is conducted in the digital twin model. The process optimization suggestions will only be written into the process rule base if the simulation results show a decrease in the anomaly rate and there are no new process conflicts.

2. The method for controlling tension control indicators in the tension measurement, monitoring, management, and analysis system according to claim 1, characterized in that, The tension time-series data includes the instantaneous tension values, tension change rate, and tension fluctuation spectrum characteristics of multiple measuring points distributed along the material's travel direction.

3. The method for controlling tension control indicators in the tension measurement, monitoring, management, and analysis system according to claim 2, characterized in that, The tension timing data of each key station during the winding process is collected in real time through a tension sensor array, including: High-precision strain gauge tension sensors are installed in the bearing seats of the unwinding roller, traction roller, and take-up roller of the winding machine, and are connected by bolts and shielded twisted pair cables to the analog signal input port of the edge computing node. The instantaneous tension values ​​of each measuring point are collected synchronously at a sampling frequency of not less than 1000 Hz, and the tension change rate and tension fluctuation spectrum characteristics are calculated in real time based on the continuous sampling data. The tension fluctuation spectrum characteristics are obtained by performing a fast Fourier transform on 256 consecutive sampling points, with a focus on the energy proportion in the 20 Hz to 50 Hz frequency band.

4. The method for controlling tension control indicators in the tension measurement, monitoring, management, and analysis system according to claim 3, characterized in that, When an abnormal status identifier appears in the tension status label sequence, the edge computing node immediately sends an abnormal event message to the central control unit, including the time of the abnormality, the coordinates of the abnormal location, the abnormality type code, and the current tension value, including: Construct a fixed-length 64-byte binary message, where the exception type code is an eight-bit binary code, with 00000001 corresponding to a sudden drop in tension, 00000010 corresponding to a sudden increase in tension, 00000100 corresponding to high-frequency oscillation of tension, and 00001000 corresponding to continuous tension offset. The abnormal event message is sent to the central control unit within 100 microseconds via Industrial Ethernet using a time-sensitive networking mechanism.

5. The method for controlling tension control indicators in the tension measurement, monitoring, management, and analysis system according to claim 4, characterized in that, After receiving the abnormal event message, the central control unit matches the corresponding control strategy according to the preset process rule library, and generates a set of control instructions for the winding spindle motor speed and unwinding braking torque, including: Parse the exception type code and query the process rule library to obtain the corresponding control instruction set, which includes the ratio of speed reduction of the winding spindle motor and the amount of increase of unwinding braking torque. By combining the current winding material category code provided by the material identification system, a control strategy that adapts to the material characteristics is matched from the corresponding process rule sub-library.

6. The method for controlling tension control indicators in the tension measurement, monitoring, management, and analysis system according to claim 5, characterized in that, The control command set is sent to the actuator controller, which then drives the winding equipment to perform corresponding parameter adjustment actions, including: Before issuing the command, a legality check is performed, which includes parameter range checking, command conflict detection, and device status confirmation. The verified control instruction set is sent to the actuator controller via industrial Ethernet with a communication cycle of 1 millisecond, and the actuator completes the parsing and outputs it to the servo drive within 5 milliseconds.

7. The method for controlling tension control indicators in the tension measurement, monitoring, management, and analysis system according to claim 6, characterized in that, The central control unit packages the complete context data of this abnormal event, including tension timing data, equipment operating parameters, ambient temperature and humidity data, and control command execution logs within a preset time window before the abnormality, and stores it in the historical database. It then triggers the management analysis module to initiate the root cause analysis process, including: Pack the tension timing data, equipment operating parameter snapshots, ambient temperature and humidity data, control command execution logs, operator identification and shift information within 500 milliseconds prior to the anomaly into a unified timestamp. The packaged data is written to a time-series database based on a columnar storage engine, stored in daily partitions, and retained for at least 360 days.

Images