High-speed self-adaptive line tension control method and system

Abstract

The invention discloses a high-speed self-adaptive thread tension control method and system, relates to the technical field of thread tension control, and solves the problem that in the prior art, sewing thread tension control lags behind in the high-speed sewing process. The system comprises a tension detection module, a data processing module, a prediction model module, a control module and an execution module, the tension detection module comprises a high-precision tension sensor, and the high-precision tension sensor is used for detecting the tension value of a sewing thread in the high-speed sewing process in real time; the data processing module is used for receiving the tension value transmitted by the high-precision tension sensor, filtering the tension value by adopting a filtering algorithm, and removing an abnormal value by adopting a sliding window method so as to obtain a stable tension data sequence; the prediction model module constructs an LSTM deep learning model based on a neural network, and is used for calculating and outputting a tension prediction value according to the tension data sequence obtained by the data processing module. The yarn tension control device has the advantage that yarn tension is controlled in a prospective mode.

Description

Technical Field

[0001] This invention relates to the field of line tension control technology, and more specifically, to a high-speed adaptive line tension control method and system. Background Technology

[0002] The stability of sewing thread tension directly determines the quality grade of sewing products, especially in high-speed sewing scenarios where the sewing thread moves at extremely high speeds, with sewing machine speeds typically reaching 3000-6000 rpm. Its tension can fluctuate by several times or even tens of times instantaneously. Currently, the mainstream tension control method relies primarily on the traditional PID feedback control algorithm. This algorithm detects the deviation between the current tension value and the setpoint in real time, thereby driving the actuator to adjust accordingly. However, in high-speed sewing, this closed-loop control mode based on "detection-deviation-adjustment" has significant limitations, such as: From the perspective of response speed, the adjustment action of traditional PID algorithms depends on the tension deviation that has already occurred. From the acquisition of signals by the tension detection element, the calculation by the controller, to the completion of the action by the actuator, there is a delay in the entire process. In high-speed sewing, this delay is sufficient to cause the sewing thread to experience multiple tension peak impacts, thereby causing problems such as thread breakage, stitch loosening, or fabric stretching and deformation. In view of this, we propose a high-speed adaptive thread tension control method and system. Summary of the Invention

[0003] The purpose of this invention is to provide a high-speed adaptive thread tension control method and system to solve the problem of lagging thread tension control in the prior art during high-speed sewing.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a high-speed adaptive line tension control system, which includes a tension detection module, a data processing module, a prediction model module, a control module, and an execution module; The tension detection module includes a high-precision tension sensor, which is used to detect the tension value of the sewing thread in real time during high-speed sewing. The data processing module is used to receive the tension value transmitted by the high-precision tension sensor, filter it using a filtering algorithm, and remove outliers using a sliding window method to obtain a stable tension data sequence. The prediction model module is based on a neural network to construct an LSTM deep learning model, which is used to calculate and output the tension prediction value based on the tension data sequence obtained by the data processing module. The control module uses a built-in tension threshold database to preset multiple sets of upper and lower tension thresholds based on fabric type, sewing thread material, and stitch type. Then, it compares the tension prediction value obtained by the prediction model module with the corresponding threshold in the tension threshold database. When the prediction value exceeds the threshold range, a PWM adjustment signal is generated. The execution module includes a servo motor and a tension adjustment mechanism linked thereto, which is used to control the tension adjustment mechanism through the servo motor according to the PWM adjustment signal generated by the control module, so as to realize continuous adjustment of the sewing thread tension.

[0005] Preferably, when the prediction model module constructs an LSTM deep learning model based on a neural network, it is trained using historical sewing data containing more than 100,000 samples.

[0006] Preferably, the prediction model module further includes an online model update unit. When a new fabric type or new stitch type is detected and the continuous sewing time exceeds 5 minutes, incremental learning is automatically started to fine-tune the model parameters using the newly added real-time data to adapt to the new working conditions.

[0007] Preferably, the tension detection module further includes a temperature compensation unit, which compensates the tension detection value in real time when the ambient temperature changes.

[0008] Preferably, the control module is also equipped with an emergency protection unit. When the sewing thread tension value is detected to momentarily exceed 30N and last for more than 10ms, a stop signal is immediately generated to control the sewing machine to stop running and to issue an alarm signal.

[0009] Preferably, the tension adjustment mechanism of the execution module adopts a double pressure wheel structure, with the two pressure wheels symmetrically distributed and able to adjust the pressure independently. The pressure adjustment range is 0.1-5N, and the differential adjustment is used to adapt to sewing threads of different diameters.

[0010] Preferably, it also includes a human-machine interaction module, which includes a touch screen for displaying the current tension value, predicted tension curve, sewing speed and equipment operating status in real time, and supports operators to manually input tension threshold correction values ​​from the tension threshold database.

[0011] Preferably, the prediction model module further includes a model evaluation unit, which compares the predicted value with the actual detected value every hour, and sends a model warning signal to the control module when there is a deviation between the predicted value and the actual detected value.

[0012] Preferably, the historical sewing data includes continuous sampling values ​​in the sewing speed range of 3000-6000 rpm, more than 20 fabric type parameters, more than 10 stitch type parameters, more than 15 sewing thread material parameters, and corresponding tension time series.

[0013] This invention also provides a high-speed adaptive line tension control method, which includes the following steps: S1. The tension value of the sewing thread is detected in real time during high-speed sewing using a tension detection module; S2. The tension values ​​are filtered using a filtering algorithm by the data processing module, and outliers are removed by the sliding window method to obtain a stable tension data sequence. S3. The prediction model module calculates and outputs the predicted tension value based on the tension data sequence obtained from the data processing module. S4. Using the tension threshold database built into the control module, multiple sets of upper and lower tension thresholds are preset according to the fabric type, sewing thread material and stitch type. Then, the tension prediction value obtained by the prediction model module is compared with the corresponding threshold in the tension threshold database. When the prediction value exceeds the threshold range, a PWM adjustment signal is generated. S5. Based on the PWM adjustment signal generated by the control module, the execution module controls the tension adjustment mechanism through the servo motor to achieve continuous adjustment of the sewing thread tension.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention constructs an LSTM-based deep learning model through a prediction model module, trained using over 100,000 historical sewing data points. This model can accurately predict future tension values ​​based on the processed tension sequence. It selectively retains and updates historical tension information through a cell state update formula, and enhances nonlinear fitting capabilities by combining an improved ReLU activation function. This allows it to capture the temporal and complex nonlinear changes in tension, outputting predicted tension values ​​in advance. The control module compares the predicted values ​​with preset thresholds and generates PWM adjustment signals in advance. The execution module responds promptly through a servo motor and tension adjustment mechanism, avoiding the lag problem of adjustment only after tension changes have occurred. This ensures intervention before tension exceeds the threshold, achieving proactive control.

[0015] 2. In this invention, the data processing module introduces speed disturbance compensation terms and temperature compensation terms through filtering algorithms to reduce the interference of sewing speed fluctuations and temperature changes on tension detection. At the same time, it removes outliers by using the sliding window method to eliminate abnormal data caused by instantaneous interference from the sensor or mechanical vibration. The multiple processing mechanisms make the obtained tension data sequence more stable and closer to reality, providing high-quality input for the prediction model module, improving the model prediction accuracy, and thus ensuring the accuracy of the control module's decision-making.

[0016] 3. The control module of this invention has a built-in tension threshold database, which can preset multiple sets of thresholds according to fabric type, sewing thread material and stitch type. The execution module adopts a double pressure wheel structure, supports 0.1-5N pressure adjustment and differential adjustment, adapts to sewing threads of different diameters, can meet the sewing needs of different fabrics, stitches and thread diameters, and improves versatility. Attached Figure Description

[0017] Figure 1 This is a schematic block diagram of the system structure of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0018] Example 1 like Figure 1 As shown, the present invention relates to a high-speed adaptive line tension control system, which includes a tension detection module, a data processing module, a prediction model module, a control module, and an execution module. The tension detection module includes a high-precision tension sensor, which detects the tension value of the sewing thread in real time during high-speed sewing, providing raw tension data for the entire system. The data processing module is connected to the tension detection module and receives the tension values ​​transmitted by the high-precision tension sensor. It then filters the values ​​using a filtering algorithm and removes outliers using a sliding window method, specifically by setting the window size to [value missing]. Take the current data point as the center. Calculate the mean of several continuous tension values. and standard deviation If the current data point is the same as The absolute value of the difference is greater than If the value is abnormal, it is identified as an outlier and replaced with the mean value within the window. The purpose of this method is to remove abnormal tension data caused by instantaneous interference from the sensor or sudden mechanical vibration, so as to obtain a stable tension data sequence, provide reliable input for subsequent model prediction, and improve the accuracy of model prediction. The filtering algorithm formula is as follows: State prediction: In the formula, Let k be the predicted state value at time k. Here is the state transition matrix. This is the optimal state estimate at time k-1. To control the input matrix, This is the control input at time k-1. This is the velocity disturbance compensation coefficient. This formula represents the sewing speed fluctuation and is based on the fundamental state equations of a linear system. In the formula, x k-1The tension state value at time k-1 represents the tension state at the previous sampling time, w k The disturbance term at time k represents the uncontrollable interference factors affecting the tension state at time k (such as sewing speed fluctuations, uneven fabric thickness, and mechanical vibration). Considering the significant impact of speed fluctuations on tension during high-speed sewing, a speed disturbance compensation term is introduced. This allows for early prediction of tension conditions, reducing prediction errors caused by speed fluctuations and improving the predictive power. Status Update: ,in Let be the optimal state estimate at time k. For Kalman gain, for The observed value at time, For the observation matrix, This is the temperature compensation coefficient. For the temperature change, the basic update formula for this state update is: To correct for the effect of temperature on detection, a temperature compensation term was added. By combining real-time observations and temperature compensation, the tension state estimation is optimized, making the results closer to the actual tension and reducing the interference of temperature changes on the detection. The prediction model module is connected to the data processing module. It constructs an LSTM deep learning model based on a neural network to calculate the output tension prediction value based on the tension data sequence obtained from the data processing module. The formula for building an LSTM deep learning model based on a neural network is as follows: Cell status update: In the formula, C t for Cellular state at any given moment Output for the forget gate. For input gate output, for Cell state at time -1 For candidate cell states, this formula, in order to selectively retain and update historical information, filters historical states after the forgetting gate. New information after input gate filtering The summation yields the current cell state, allowing the model to remember important historical tension information and update new tension features, thus solving the problem that traditional models struggle to handle long sequence dependencies and adapting to the temporal characteristics of tension changes during sewing. Hidden layer output: In the formula, for Hidden layer output at any given time, As the output of the output gate, this formula uses the output of the hidden layer as the output ratio controlled by the output gate to control the cell state, first through... Map the cell state values ​​to [-1, 1], and then output the result via an output gate. Multiplying them together, we can extract key features from the cell state and output them to provide effective feature support for tension prediction in subsequent output layers. The hidden layer uses a modified ReLU activation function: In the formula, Leak coefficient, For the output of the activation function, This formula introduces a leak coefficient to address the problem of neuron failure when the basic ReLU function receives a negative input, serving as the input to the hidden layer neurons. To make the output the same when there is a negative input. Maintaining neuron activity enhances the model's nonlinear fitting ability, prevents neuron failure, and enables the model to better capture the nonlinear variation characteristics of tension, thereby improving its ability to predict complex tension changes. Output layer: In the formula, for Predicted tension value at time (unit: N), This is the output layer weight matrix. for Hidden layer output at any given time, For output layer bias, this formula is applied by the hidden layer output. Mapped to the tension value space through a linear transformation, weights and bias The features extracted from the hidden layer are obtained through training and are transformed into specific tension prediction values, thereby achieving accurate prediction of future tension and providing decision-making basis for the control module to respond to tension changes in advance. The control module is connected to the prediction model module and the execution module respectively. Through the built-in tension threshold database, multiple sets of upper and lower tension thresholds are preset according to the fabric type, sewing thread material and stitch type. Then, the tension prediction value obtained by the prediction model module is compared with the corresponding threshold in the tension threshold database. When the prediction value exceeds the threshold range, a PWM adjustment signal is generated. The execution module includes a servo motor and a tension adjustment mechanism linked to it. Based on the PWM adjustment signal generated by the control module, the servo motor controls the tension adjustment mechanism to continuously adjust the tension of the sewing thread, so that the tension of the sewing thread is stabilized within the target range and the sewing quality is guaranteed.

[0019] Furthermore, when the prediction model module builds an LSTM deep learning model based on a neural network, it is trained using historical sewing data containing more than 100,000 samples. Training with large-scale historical sewing data enables the LSTM deep learning model to learn rich patterns of sewing tension changes, thereby improving the generalization ability and prediction accuracy of the LSTM deep learning model and adapting it to different sewing scenarios.

[0020] Furthermore, the prediction model module also includes an online model update unit. When a new fabric type or new stitch type is detected and the continuous sewing time exceeds 5 minutes, incremental learning is automatically started. The model parameters are fine-tuned using the newly added real-time data. Incremental learning enables the model to quickly adapt to new working conditions, avoids prediction deviations caused by changes in fabric and stitch, and maintains long-term stable prediction performance.

[0021] Furthermore, the tension detection module also includes a temperature compensation unit, which compensates for changes in the ambient temperature in real time. The compensation formula is as follows: , To compensate for the tension (unit: N). For measuring tension (unit: N). For temperature coefficient, The current temperature (unit: °C). The reference temperature is 25℃. Since the deviation has a quadratic function relationship with the temperature difference, a quadratic polynomial fitting is used to obtain the compensation formula. The purpose of temperature compensation is to eliminate the influence of temperature change on tension detection, make the detected value closer to the true tension, improve detection accuracy, and provide reliable data for subsequent processing and control.

[0022] Furthermore, the control module is equipped with an emergency protection unit. When the sewing thread tension value is detected to momentarily exceed 30N and last for more than 10ms, a stop signal is immediately generated to control the sewing machine to stop running and issue an alarm signal to quickly respond to extreme tension abnormalities, avoid sewing thread breakage or equipment damage, and ensure production safety and equipment lifespan.

[0023] Furthermore, the tension adjustment mechanism of the execution module adopts a double pressure wheel structure. The two pressure wheels are symmetrically distributed and can adjust the pressure independently. The pressure adjustment range is 0.1-5N. Differential adjustment is used to adapt to sewing threads of different diameters. The double pressure wheel design and independent pressure adjustment enhance the adaptability to different thread diameters. Differential adjustment makes tension control more flexible and meets diverse sewing needs.

[0024] Furthermore, it also includes a human-machine interaction module, which includes a touch screen for real-time display of the current tension value, predicted tension curve, sewing speed, and equipment operating status. It also supports operators to manually input tension threshold correction values ​​from the tension threshold database, providing a visual interface for operators to easily monitor the system status in real time. The manual correction function allows for fine-tuning of parameters based on actual sewing results, improving the system's ease of use and adaptability.

[0025] Furthermore, the prediction model module also includes a model evaluation unit, which compares the hourly sewing thread tension prediction value with the actual detection value. When the deviation between the prediction value and the actual detection value exceeds ±2N, a model warning signal is sent to the control module to evaluate the model performance in real time, detect model drift or failure problems in a timely manner, provide early warning to maintenance personnel, and ensure the long-term reliable operation of the model.

[0026] Furthermore, the historical sewing data includes continuous sampling values ​​in the sewing speed range of 3000-6000 rpm, parameters for more than 20 fabric types, parameters for more than 10 stitch types, parameters for more than 15 sewing thread materials, and corresponding tension time series. With rich historical sewing data covering a variety of sewing scenarios, the model training is more comprehensive, which can better adapt to different process conditions and improve the versatility of the system.

[0027] A high-speed adaptive line tension control method, comprising the following steps: S1. The tension value of the sewing thread is detected in real time during high-speed sewing using a tension detection module; S2. The tension values ​​are filtered using a filtering algorithm by the data processing module, and outliers are removed by the sliding window method to obtain a stable tension data sequence. S3. The prediction model module calculates and outputs the predicted tension value based on the tension data sequence obtained from the data processing module. S4. Using the tension threshold database built into the control module, multiple sets of upper and lower tension thresholds are preset according to the fabric type, sewing thread material and stitch type. Then, the tension prediction value obtained by the prediction model module is compared with the corresponding threshold in the tension threshold database. When the prediction value exceeds the threshold range, a PWM adjustment signal is generated. S5. Based on the PWM adjustment signal generated by the control module, the execution module controls the tension adjustment mechanism through the servo motor to achieve continuous adjustment of the sewing thread tension.

[0028] The embodiments disclosed in this invention are preferred embodiments, but are not limited thereto. Those skilled in the art can easily understand the spirit of this invention based on the above embodiments and make different extensions and variations, but as long as they do not depart from the spirit of this invention, they are all within the protection scope of this invention.

Claims

1. A high-speed adaptive linear tension control system, characterized in that, The system includes a tension detection module, a data processing module, a prediction model module, a control module, and an execution module; The tension detection module includes a high-precision tension sensor, which is used to detect the tension value of the sewing thread in real time during high-speed sewing. The data processing module is used to receive the tension value transmitted by the high-precision tension sensor, filter it using a filtering algorithm, and remove outliers using a sliding window method to obtain a stable tension data sequence. The prediction model module is based on a neural network to construct an LSTM deep learning model, which is used to calculate and output the tension prediction value based on the tension data sequence obtained by the data processing module. The control module uses a built-in tension threshold database to preset multiple sets of upper and lower tension thresholds based on fabric type, sewing thread material, and stitch type. Then, it compares the tension prediction value obtained by the prediction model module with the corresponding threshold in the tension threshold database. When the prediction value exceeds the threshold range, a PWM adjustment signal is generated. The execution module includes a servo motor and a tension adjustment mechanism linked thereto, which is used to control the tension adjustment mechanism through the servo motor according to the PWM adjustment signal generated by the control module, so as to realize continuous adjustment of the sewing thread tension.

2. The high-speed adaptive linear tension control system according to claim 1, characterized in that, The prediction model module is trained using historical sewing data containing more than 100,000 samples when constructing an LSTM deep learning model based on a neural network.

3. A high-speed adaptive linear tension control system according to claim 1, characterized in that, The prediction model module also includes an online model update unit. When a new fabric type or new stitch type is detected and the continuous sewing time exceeds 5 minutes, incremental learning is automatically started to fine-tune the model parameters using the newly added real-time data to adapt to the new working conditions.

4. A high-speed adaptive linear tension control system according to claim 1, characterized in that, The tension detection module also includes a temperature compensation unit, which compensates the tension detection value in real time when the ambient temperature changes.

5. A high-speed adaptive linear tension control system according to claim 1, characterized in that, The control module is also equipped with an emergency protection unit. When the sewing thread tension value is detected to exceed 30N instantaneously and last for more than 10ms, a stop signal is immediately generated to control the sewing machine to stop running and issue an alarm signal.

6. A high-speed adaptive linear tension control system according to claim 1, characterized in that, The tension adjustment mechanism of the execution module adopts a double pressure wheel structure. The two pressure wheels are symmetrically distributed and can adjust the pressure independently. The pressure adjustment range is 0.1-5N. Differential adjustment is used to adapt to sewing threads of different diameters.

7. A high-speed adaptive linear tension control system according to claim 1, characterized in that, It also includes a human-machine interaction module, which includes a touch screen for displaying the current tension value, predicted tension curve, sewing speed and equipment operating status in real time, and supports operators to manually input tension threshold correction values ​​from the tension threshold database.

8. A high-speed adaptive linear tension control system according to claim 1, characterized in that, The prediction model module also includes a model evaluation unit, which compares the predicted value with the actual detected value every hour. When there is a deviation between the predicted value and the actual detected value, a model warning signal is sent to the control module.

9. A high-speed adaptive linear tension control system according to claim 2, characterized in that, The historical sewing data includes continuous sampling values ​​in the sewing speed range of 3000-6000 rpm, parameters for more than 20 fabric types, parameters for more than 10 stitch types, parameters for more than 15 sewing thread materials, and corresponding tension time series.

10. A high-speed adaptive line tension control method, applicable to the high-speed adaptive line tension control system described in any one of claims 1-9, characterized in that, The method includes the following steps: S1. The tension value of the sewing thread is detected in real time during high-speed sewing using a tension detection module; S2. The tension values ​​are filtered using a filtering algorithm by the data processing module, and outliers are removed by the sliding window method to obtain a stable tension data sequence. S3. The prediction model module calculates and outputs the predicted tension value based on the tension data sequence obtained from the data processing module. S4. Using the tension threshold database built into the control module, multiple sets of upper and lower tension thresholds are preset according to the fabric type, sewing thread material and stitch type. Then, the tension prediction value obtained by the prediction model module is compared with the corresponding threshold in the tension threshold database. When the prediction value exceeds the threshold range, a PWM adjustment signal is generated. S5. Based on the PWM adjustment signal generated by the control module, the execution module controls the tension adjustment mechanism through the servo motor to achieve continuous adjustment of the sewing thread tension.

Images