An automatic zipper sewing control system and its device

Abstract

The present invention relates to the technical field of automatic zipper sewing control systems, and specifically to an automatic zipper sewing control system, which includes a sensor array unit, a real-time data acquisition and processing module, a fault detection and diagnosis module, an adaptive control algorithm, a fault prediction and alarm unit, and a control unit; wherein the sensor array unit includes a current sensor applied to motor overload detection, a temperature sensor for monitoring the temperature of the motor and the control system, a tension sensor for detecting the tension of the zipper fabric, and a displacement sensor for monitoring the movement of the zipper; among which the real-time data acquisition and processing module uses a real-time operating system to collect and analyze various sensor data. It can perform fault detection efficiently and accurately, especially can timely detect some potential faults, such as motor overload, sensor accuracy drift, internal logic errors of the control system, etc., thereby reducing fault expansion, maintenance costs and downtime.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic zipper sewing control systems, and in particular to an automatic zipper sewing control system and its device. Background Art

[0002] In the prior art, when sewing a clothing zipper onto fabric, multiple processes are required. First, it is necessary to manually sew the zipper and the lower fabric together as the first sewing process. Then, sew the lower fabric with the zipper onto the upper fabric as the second sewing process. The zipper is located between the lower fabric and the upper fabric, and the third and fourth sewing processes are required for the upper fabric, the lower fabric, and the zipper pull. Specifically, turn the lower fabric outward and press the line towards the zipper, then perform the third sewing process of the zipper. Then, turn the upper fabric outward, position, and press the line towards the zipper to perform the fourth sewing process of the zipper. Finally, the edge of the zipper is left between the upper fabric and the lower fabric.

[0003] Therefore, the invention with the publication number CN106930009A specifically discloses an automatic zipper sewing device, a sewing machine, and a sewing method, belonging to the field of industrial sewing automation. In the present invention, the upper fabric, the lower fabric, and the zipper to be sewn are placed on the sewing substrate. The sliding frame is lifted away from the sewing substrate by the tilting cylinder, and the movable large plate is moved to the position of sewing the zipper through the pushing component to achieve the purpose of feeding. The fixing of the zipper and the hemming of the upper fabric and the lower fabric are completed through the air suction component, the zipper positioning groove, the zipper fixing component, and the air blowing component, solving the problems of low production efficiency and unstable quality caused by the inability to complete multiple sewing processes at one time.

[0004] In the prior art, the automatic zipper sewing control system is a system applied to industries such as clothing manufacturing and luggage production, specifically for automating the control of the zipper sewing process. However, there are some potential faults in the actual application process of the traditional automatic zipper sewing control system, such as motor overload, sensor accuracy drift, internal logic errors in the control system, etc., resulting in increased enterprise maintenance costs and downtime in the actual production process, affecting the processing efficiency.

[0005] Therefore, an automatic zipper sewing control system and its device are proposed. Summary of the Invention

[0006] The purpose of the present invention is to provide an automatic zipper sewing control system and its device to solve the problems raised in the above background art.

[0007] To achieve the above purpose, the present invention provides the following technical solution: An automatic zipper sewing control system includes a sensor array unit, a real-time data acquisition and processing module, a fault detection and diagnosis module, an adaptive control algorithm, a fault prediction and alarm unit, and a control unit.

[0008] The sensor array unit includes a current sensor applied to motor overload detection, a temperature sensor for monitoring the temperature of the motor and control system, a tension sensor for detecting the pulling force of the zipper fabric, and a displacement sensor for monitoring the movement of the zipper;

[0009] The real-time data acquisition and processing module uses a real-time operating system to collect and analyze various sensor data, including motor current, sensor accuracy, and zipper movement status, and feeds back to the control unit in real time;

[0010] The fault detection and diagnosis module performs intelligent fault analysis on the system, predicts potential faults and alarms in a timely manner. This module has an adaptive ability and automatically adjusts the diagnostic model according to different production conditions and equipment status;

[0011] The adaptive control algorithm is an adaptive control algorithm based on the combination of PID control and fuzzy control, which ensures that the tension between the zipper and the fabric during the sewing process remains stable and can automatically compensate for minor motor overload abnormalities;

[0012] The fault prediction and alarm unit uses historical data and sensor information to predict faults through a prediction model. During the operation of the system, it monitors the feedback of each sensor in real time. If abnormal motor current, voltage, and temperature parameters are detected, an alarm is triggered.

[0013] Preferably, the specific processing steps of the fault prediction model are as follows:

[0014] Step 1: Data collection, collect various sensor data from the automatic zipper sewing control system, including current, voltage, temperature, tension, and motor speed, and preprocess the collected data;

[0015] Data preprocessing, remove noise through low-pass filtering and moving average filtering methods. For time-series data, the filtering methods used are moving average and median filtering;

[0016] The moving average formula: where k is the size of the moving window, is the original sensor value at time i, is the value after moving average; [[ID=3i]]

[0017] Normalization: To improve the efficiency of model training, data is usually normalized or standardized so that the data mean is 0 and the variance is 1: where X is the original data, is the mean, is the standard deviation;

[0018] Step 2: Build a time-series prediction model:

[0019] Model design: The LSTM model is adopted to learn the long-term dependencies in time series data by introducing forget gates, input gates, and output gates, and then the model is trained. The mean squared error is used as the loss function:

[0020] Among them, is the true value, is the predicted value, and N is the number of samples;

[0021] Step 3: Autoencoder anomaly detection:

[0022] In the autoencoder, the encoder maps the input data, i.e., sensor data, to a low-dimensional space: ,

[0023] Among them, and are the weight matrix and bias term of the encoder, is the activation function, i.e., Sigmoid, ReLU;

[0024] The decoder is used to reconstruct the low-dimensional feature z into the original input data: , and then calculate the error between the original input data and the reconstructed data: ,

[0025] Among them, N is the number of samples, and respectively represent the th input sample and its reconstructed output;

[0026] When the reconstruction error exceeds the preset threshold, mark this moment as an anomaly, and the threshold is set through the error distribution of historical data.

[0027] Step 4: Hybrid model based on clustering and prediction: Use the clustering algorithm to cluster historical data and identify different working states of the device.

[0028] K-means algorithm: Among them, is the probability that sample i belongs to cluster k, is the data point, is the center of cluster k, and K is the number of clusters. Combining the clustering analysis results with the LSTM prediction, if a certain working state deviates significantly and the prediction error of the LSTM model exceeds the threshold, it is determined as a fault;

[0029] Step 5: Ensemble learning and reinforcement learning optimization: Integrate multiple fault prediction models, i.e., LSTM autoencoders, and use a weighted average or voting mechanism to determine the final prediction result:

[0030] Weighted average:

[0031] Among them, is the weight of each model, is the prediction result of the i-th model, M is the number of models. The control strategy of the system is optimized using reinforcement learning. In the actual production process, the system will select different control actions according to the results of fault prediction and gradually optimize the control strategy through a reward mechanism, specifically including giving a positive reward to the system for correct fault prediction and timely response; giving a negative reward for failing to predict a fault.

[0032] Preferably, the fifth step, ensemble learning and reinforcement learning optimization, specifically further includes two parts: model weighting strategy optimization and further optimization of ensemble learning. Among them, the model weighting strategy optimization is divided into model weighting strategy optimization and adaptive voting mechanism;

[0033] The model weighting strategy optimization dynamically adjusts the weight of each model based on the prediction performance of each sub-model in different environments, and uses the accuracy evaluation indicators of the model, namely AUC, F1-score, and mean square error, to adjust the weight:

[0034] Among them, is the prediction accuracy of the i-th model in a certain time period or working condition, M is the total number of the ensemble model. The weight is dynamically adjusted according to the current fault prediction effect within different time windows. When the model successfully predicts multiple faults in a short time, the system will give it a preferential weight;

[0035] The adaptive voting mechanism makes predictions based on the model weights by introducing a weighted voting mechanism:

[0036] Among them, is the weight of the model, is the prediction result of the i-th model. The model with a larger weight will occupy a larger share in the voting.

[0037] Preferably, the further optimization of the ensemble learning is specifically divided into the expansion of the state space and action space in reinforcement learning and the optimization of the reward function;

[0038] In the expansion of the state space and action space in reinforcement learning, historical data of sensors and the evolution of fault modes are used to expand the state space;

[0039] For the optimization of the reward function, for time series data, the reward is designed according to the duration of the system state. By increasing the penalty for failing to predict a fault in time, the learning ability of the model is strengthened to avoid missing potential fault modes.

[0040] An automatic zipper sewing device, comprising a laser bag-opening platform, a laser cutting mechanism, a fabric grabbing and feeding mechanism, a fabric placement platform and a fabric sewing machine;

[0041] On the upper end surface of the laser bag-opening platform, there are successively distributed from left to right a fabric placement platform, a fabric grabbing and feeding mechanism, a laser cutting mechanism and a fabric sewing machine;

[0042] The fabric is placed on the fabric placement platform, and the fabric grabbing and feeding mechanism grabs the fabric to the laser cutting mechanism. The lower fabric is laser cut by the laser cutting mechanism, and then the cut fabric is sewn by the fabric sewing machine.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] Design a new type of automatic zipper sewing control system, which can perform fault detection efficiently and accurately, especially can timely detect some potential faults, such as motor overload, sensor accuracy drift, internal logic error of the control system, etc., so as to reduce fault expansion, maintenance cost and downtime. This system combines advanced fault detection technology, intelligent control algorithms and multi-level monitoring and feedback mechanisms. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0046] Figure 1 It is a block diagram of the automatic zipper sewing control system of the present invention;

[0047] Figure 2 It is a flowchart of the specific processing steps of the fault prediction model of the present invention;

[0048] Figure 3 It is a diagram of the automatic zipper sewing device of the present invention;

[0049] Figure 4 It is a bottom view of the automatic zipper sewing device of the present invention.

[0050] Component labels:

[0051] 1. Laser bag-opening platform; 2. Laser cutting mechanism; 3. Fabric grabbing and feeding mechanism; 4. Fabric placement platform; 5. Fabric sewing machine. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0053] Please refer to Figures 1 to 4 , the present invention provides a technical solution:

[0054] An automatic zipper sewing control system includes a sensor array unit, a real-time data acquisition and processing module, a fault detection and diagnosis module, an adaptive control algorithm, a fault prediction and alarm unit, and a control unit. A high-performance control unit designed based on an embedded system is responsible for real-time calculation and processing of various sensor data and adjusting sewing parameters according to feedback to avoid faults.

[0055] Among them, the sensor array unit includes a current sensor applied to motor overload detection, a temperature sensor for monitoring the temperature of the motor and the control system, a tension sensor for detecting the tension of the zipper fabric, and a displacement sensor for monitoring the movement of the zipper.

[0056] Among them, the real-time data acquisition and processing module uses a real-time operating system to collect and analyze various sensor data, including motor current, sensor accuracy, and zipper movement status, and real-time feedback to the control unit.

[0057] Among them, the fault detection and diagnosis module conducts intelligent fault analysis on the system, predicts potential faults and gives timely alarms. This module has an adaptive ability and automatically adjusts the diagnosis model according to different production conditions and equipment states.

[0058] Among them, the adaptive control algorithm is an adaptive control algorithm based on the combination of PID control and fuzzy control, which ensures that the tension between the zipper and the fabric during the sewing process remains stable and can automatically compensate for minor motor overload abnormalities.

[0059] The adaptive control algorithm uses fuzzy logic to dynamically adjust the PID parameters to achieve tension stability and motor overload compensation. The specific structure is as follows:

[0060]

[0061] Among them, is the tension error, and Kp, Ki, and Kd are parameters to be adjusted.

[0062] Fuzzy logic adjustment rules:

[0063] Error and its change rate ;

[0064] The membership functions are divided into: Negative Big (NB), Negative Small (NS), Zero (ZO), Positive Small (PS), Positive Big (PB).

[0065] Output variable: (Adjustment amount of PID parameters);

[0066] When the current sensor detects a slight overload (such as I > I rated, but the alarm threshold is not triggered), the fuzzy PID automatically reduces and increases , in order to smooth the output torque: , where α and β are empirical coefficients, calibrated through experiments.

[0067] Among them, the fault prediction and alarm unit uses historical data and sensor information to predict faults through a prediction model. During the operation of the system, it monitors the feedback of each sensor in real time. If abnormal parameters of motor current, voltage, and temperature are detected, an alarm is triggered.

[0068] Motor overload and sensor accuracy detection: The judgment formula for motor overload: P = U × I;

[0069] Among them, P is the motor power, U is the voltage, and I is the current. If the power exceeds the set threshold, the system will automatically judge it as overload and adjust the motor load to avoid overload damage.

[0070] Sensor accuracy drift monitoring: Based on the Kalman filtering algorithm, real-time data smoothing and correction are performed to detect abnormal fluctuations or drifts in sensor data.

[0071] Kalman filter model:

[0072] State equation (system dynamic model):

[0073]

[0074] Among them, xt is the system state (such as the real temperature), ut is the control input, wt is the process noise, A is the state transition matrix, and B is the input matrix.

[0075] Observation equation (sensor model):

[0076]

[0077] Among them, zt is the sensor measurement value, vt is the observation noise, and H is the observation matrix.

[0078] Filtering steps:

[0079] 1. Prediction:

[0080] State prediction:

[0081] Error covariance prediction:

[0082] 2. Update:

[0083] Kalman gain calculation: ;

[0084] Status correction: ;

[0085] Covariance update: ;

[0086] Assume that the actual temperature of the motor changes slowly (state equation A=1, B=0), the sensor noise R=0.5, and the process noise Q=0.1.

[0087] If the sensor reading suddenly increases at a certain moment (such as jumping from 50°C to 55°C), the Kalman filter smoothes the historical data and outputs a corrected value (such as 51.2°C) to suppress abnormal fluctuations.

[0088] Zipper sewing tension monitoring: Zipper tension monitoring formula: Where T is the tension, k is the elastic constant of the fabric, L is the current fabric length, and L0 is the fabric length at the starting position of the zipper. If the tension is abnormal, the system will automatically adjust the sewing speed.

[0089] System operation process: Data acquisition: Each sensor collects data (current, voltage, temperature, tension) in real time.

[0090] Data processing: The control unit performs real-time analysis based on sensor data to determine whether there is a potential fault.

[0091] Fault diagnosis and early warning: If a fault is detected, the system will automatically alarm and adjust operating parameters, such as reducing motor load or adjusting the zipper sewing speed.

[0092] Control output: Adjust the sewing process according to the fault diagnosis results to avoid the expansion of the fault and ensure the continuous and efficient operation of the production line.

[0093] Relationship between temperature and motor operating status:

[0094] Use a temperature sensor to monitor the motor's operating temperature and calculate the motor load using the empirical formula:

[0095]

[0096] Among them, Te is the motor temperature, T0 is the normal operating temperature, It is the temperature difference caused by the increase in load.

[0097] Example 1: Motor Overload Detection

[0098] Background: Assume the rated power of the motor is 200W. When the current exceeds 1A, the motor power will exceed the rated value.

[0099] Experimental data: Voltage U = 220V;

[0100] Current I = 1.2A;

[0101] Calculated power: P = 220 × 1.2 = 264P = 220*1.2 = 264W (exceeding the rated power of 200W)

[0102] Result: The system detects that the power exceeds the threshold, automatically adjusts the motor load and triggers an alarm, mainly implemented using the load balancing algorithm.

[0103] Example 2: Sensor Accuracy Drift Detection

[0104] Background: The initial error of the sensor is ±0.5%. After long-term use, the error may gradually increase.

[0105] Experimental data:

[0106] Initial sensor accuracy: ±0.5%

[0107] After a period of use, the error increases to ±1.2%.

[0108] Result: The error is corrected in real time through the Kalman filtering algorithm to avoid failures caused by sensor accuracy drift

[0109] The specific processing steps of the fault prediction model are as follows:

[0110] Step 1: Data collection. Collect various sensor data from the automatic zipper sewing control system, including current, voltage, temperature, tension, and motor speed, and preprocess the collected data;

[0111] Data preprocessing. Remove noise through low-pass filtering and moving average filtering methods. For time series data, the filtering methods used are moving average and median filtering;

[0112] Among them, the moving average formula: Among them, k is the size of the moving window, is the original sensor value at time i, is the value after moving average;

[0113] Among them, standardization: To improve the efficiency of model training, the data is usually standardized or normalized so that the data mean is 0 and the variance is 1: Among them, X is the original data, is the mean, is the standard deviation;

[0114] Step 2: Construct a time series prediction model:

[0115] Model design: Use the LSTM model to learn the long-term dependencies in time series data by introducing forget gates, input gates, and output gates;

[0116] The core formula of the LSTM model is as follows:

[0117] Forget gate:

[0118] Input gate:

[0119] Candidate memory cell:

[0120] Updated memory cell:

[0121] Output gate:

[0122] Hidden state:

[0123] Among them, is the input data, is the hidden state, is the memory cell, , , , is the weight matrix, , , , is the bias term, is the activation function

[0124] Then perform model training, where the loss function uses the mean squared error:

[0125] Among them, is the true value, is the predicted value, and N is the number of samples;

[0126] Step 3: Autoencoder anomaly detection:

[0127] In the autoencoder, the encoder maps the input data, i.e., the sensor data, to a low-dimensional space: ,

[0128] Among them, and are the weight matrix and bias term of the encoder, is the activation function, i.e., Sigmoid, ReLU;

[0129] Use a decoder to reconstruct the low-dimensional feature z into the original input data: , and then calculate the error between the original input data and the reconstructed data: ,

[0130] where N is the number of samples, and respectively represent the -th input sample and its reconstructed output;

[0131] When the reconstruction error exceeds the preset threshold, mark this moment as abnormal, and the threshold is set according to the error distribution of historical data.

[0132] Step 4: Hybrid model based on clustering and prediction: Use a clustering algorithm to cluster historical data and identify different working states of the device.

[0133] K-means algorithm: where, is the probability that sample i belongs to cluster k, is the data point, is the center of cluster k, K is the number of clusters. Combining the clustering analysis results with LSTM prediction, if a certain working state deviates greatly and the prediction error of the LSTM model exceeds the threshold, it is determined as a fault;

[0134] Step 5: Ensemble learning and reinforcement learning optimization: Integrate multiple fault prediction models, namely LSTM autoencoders, and use a weighted average or voting mechanism to determine the final prediction result:

[0135] Weighted average:

[0136] where, is the weight of each model, is the prediction result of the i-th model, M is the number of models. Use reinforcement learning to optimize the control strategy of the system. In the actual production process, the system will select different control actions according to the fault prediction results and gradually optimize the control strategy through a reward mechanism, specifically including giving the system a positive reward for correct fault prediction and timely response; giving a negative reward for failure to predict a fault.

[0137] The above-mentioned Step 5: Ensemble learning and reinforcement learning optimization specifically also includes two parts: model weighted strategy optimization and further optimization of ensemble learning, where model weighted strategy optimization is divided into model weighted strategy optimization and adaptive voting mechanism;

[0138] The optimization of the model weighting strategy is based on the prediction performance of each sub-model in different environments, dynamically adjusts the weight of each model, and uses the accuracy evaluation metrics of the model, namely AUC, F1-score, and mean squared error, to adjust the weight:

[0139] Among them, is the prediction accuracy of the i-th model in a certain time period or working condition, M is the total number of integrated models, and the weight is dynamically adjusted according to the current fault prediction effect within different time windows. When the model successfully predicts multiple faults in a short time, the system will give it a preferential weight; different models have different advantages and disadvantages. By learning the complementary characteristics of different models, more efficient prediction can be achieved under various working conditions. For example, the LSTM model is good at processing time series data, while the SVM may perform excellently in small sample learning. Combining their advantages can further improve the reliability of fault prediction;

[0140] The adaptive voting mechanism makes predictions by introducing a weighted voting mechanism based on the weights of the models:

[0141] Among them, is the weight of the model, is the prediction result of the i-th model. The model with a larger weight will occupy a larger share in the voting.

[0142] Specifically, the further optimization of the ensemble learning is specifically divided into the expansion of the state space and action space in reinforcement learning and the optimization of the reward function;

[0143] In reinforcement learning, the design of the state space and action space directly determines the learning ability of the model. For the automatic zipper sewing control system, the state is not only the traditional sensor data, but also more dynamic environmental factors should be considered. In the expansion of the state space and action space in reinforcement learning, historical sensor data and the evolution of fault modes are used to expand the state space; in traditional reinforcement learning, actions may be discrete (such as "start" "stop"). However, in complex systems, the action space should be more continuous. For certain specific fault modes, continuous adjustment parameters (such as adjusting the motor speed, regulating the pressure, etc.) can be selected. For each fault prediction result, the reinforcement learning model can select a combination of multiple actions to form a multi-level response strategy. For example, if a motor overload fault is predicted, a combination of three actions "decelerate", "enable standby motor", and "warn" can be selected;

[0144] For the optimization of the reward function, for time series data, the reward is designed according to the duration of the system state. By increasing the penalty for not predicting faults in time, the learning ability of the model is strengthened to avoid missing potential fault modes. Assuming that a fault occurs in the system, the reward function can be designed as: Among them, represents the severity of the fault, represents the downtime, and are weight coefficients of 1.0 and 0.5 respectively, controlling the influence of fault severity and downtime. By increasing the penalty for failure to predict faults in a timely manner, the learning ability of the model can be strengthened, and potential fault modes can be avoided from being missed. If the system fails to effectively predict a fault before it occurs, the learning effect of the model can be strengthened by increasing the penalty (negative reward).

[0145] Design multi-objective rewards based on fault prediction accuracy and system response efficiency:

[0146]

[0147] SuccessReward: Reward for successfully predicting a fault (such as +10);

[0148] FaultSeverity: Fault severity (calculated based on the over-limit ratio of current and temperature).

[0149] ResponseDelay: Delay time (in seconds) from the occurrence of a fault to the system response.

[0150] Weight coefficient .

[0151] Action space and policy optimization, action space: Discrete actions: deceleration, switching to standby motor, alarm.

[0152] Continuous actions: Fine-tuning of PID parameters ( )

[0153] Policy optimization (PPO algorithm):

[0154] Objective function:

[0155]

[0156] Among them, is the estimated advantage function at time step t, is the clipping coefficient, a represents the action, s represents the state, and clip is the clipping function.

[0157] Motor overload response

[0158] States: Continuous over-limit of current, accelerating rate of temperature rise.

[0159] Actions: Trigger "deceleration" + "adjust Kp".

[0160] Reward calculation:

[0161] If overload is successfully avoided, .

[0162] If downtime is caused by failure to respond in time, .

[0163] An automatic zipper sewing device, comprising a laser bag opening platform 1, a laser cutting mechanism 2, a fabric grabbing and feeding mechanism 3, a fabric placing platform 4 and a fabric sewing machine 5;

[0164] Wherein, on the upper end surface of the laser bag opening platform 1, there are successively distributed a fabric placing platform 4, a fabric grabbing and feeding mechanism 3, a laser cutting mechanism 2 and a fabric sewing machine 5 from left to right;

[0165] The fabric is placed on the fabric placing platform 4, and the fabric grabbing and feeding mechanism 3 grabs the fabric to the laser cutting mechanism 2. The fabric below is laser cut by the laser cutting mechanism 2, and then the cut fabric is sewn by the fabric sewing machine 5;

[0166] The fabric grabbing and feeding mechanism 3, the laser cutting mechanism 2 and the fabric sewing machine 5 are internally provided with a plurality of drive motors.

[0167] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An automatic zipper sewing control system, characterized in that, It includes a sensor array unit, a real-time data acquisition and processing module, a fault detection and diagnosis module, an adaptive control algorithm, a fault prediction and alarm unit, and a control unit; Among them, the sensor array unit includes a current sensor applied to motor overload detection, a temperature sensor for monitoring the temperature of the motor and the control system, a tension sensor for detecting the tension of the zipper fabric, and a displacement sensor for monitoring the movement of the zipper; Among them, the real-time data acquisition and processing module uses a real-time operating system to collect and analyze various sensor data, including motor current, sensor accuracy, and zipper movement status, and provides real-time feedback to the control unit; Among them, the fault detection and diagnosis module conducts intelligent fault analysis on the system, predicts potential faults and gives timely alarms. This module has an adaptive ability and can automatically adjust the diagnosis model according to different production conditions and equipment status; Among them, the adaptive control algorithm is an adaptive control algorithm based on the combination of PID control and fuzzy control, which ensures that the tension between the zipper and the fabric during the sewing process remains stable and can automatically compensate for slight motor overload abnormalities; Among them, the fault prediction and alarm unit uses historical data and sensor information to predict faults through a fault prediction model. During the operation of the system, it monitors the feedback of each sensor in real time. If abnormal parameters such as motor current, voltage, and temperature are detected, an alarm is triggered; The specific processing steps of the fault prediction model are as follows: Step 1: Data collection; Step 2: Construct a time series prediction model; Model design: Use the LSTM model, introduce forget gates, input gates, and output gates to learn long-term dependencies in time series data, and then perform model training. The loss function uses the mean squared error; Step 3: Autoencoder anomaly detection; Step 4: A hybrid model based on clustering and prediction; Use the K-means clustering algorithm to cluster historical data, identify different working states of the equipment, and combine the clustering analysis results with LSTM prediction. If a certain working state deviates greatly and the prediction error of the LSTM model exceeds the threshold, it is determined as a fault; Step 5. Optimization of ensemble learning and reinforcement learning: Integrate multiple fault prediction models, namely LSTM autoencoders, and use a weighted average or voting mechanism to determine the final prediction result. Weighted average: ; Among them, is the weight of each model, is the prediction result of the i-th model, M is the number of models. The control strategy of the system is optimized using reinforcement learning. During the actual production process, the system will select different control actions according to the results of fault prediction, and gradually optimize the control strategy through a reward mechanism, specifically including giving the system a positive reward for correct fault prediction and timely response, and giving a negative reward for failure to predict faults.

2. The automatic zipper sewing control system according to claim 1, characterized in that: In Step 1: Data collection, collect various sensor data from the automatic zipper sewing control system, including current, voltage, temperature, tension, and motor speed, and preprocess the collected data; Data preprocessing: Remove noise through low-pass filtering and moving average filtering methods. For time series data, the filtering methods used are moving average and median filtering; Among them, the moving average formula: ； where k is the size of the sliding window, is the original sensor value at time i, is the value after moving average; Among them, standardization: To improve the efficiency of model training, data is usually standardized or normalized so that the data mean is 0 and the variance is 1: ； Among them, X is the original data, is the mean value, is the standard deviation.

3. An automatic zipper sewing control system according to claim 1, characterized in that: Step 2: Construct a time series prediction model: The loss function uses the mean squared error: ； Among them, is the true value, is the predicted value, and N is the number of samples.

4. An automatic zipper sewing control system according to claim 1, characterized in that: Step 3: Autoencoder anomaly detection: In the autoencoder, the encoder maps the input data, that is, sensor data, to a low-dimensional space: ； Among them, and are the weight matrix and bias term of the encoder, is the activation function, namely Sigmoid, ReLU; The decoder is used to reconstruct the low-dimensional feature z into the original input data: , and then the error between the original input data and the reconstructed data is calculated: ; where N is the number of samples, and respectively represent the th input sample and its reconstructed output; When the reconstruction error exceeds a preset threshold, mark this moment as abnormal, and the threshold is set through the error distribution of historical data.

5. An automatic zipper sewing control system according to claim 1, characterized in that: The K-means algorithm: ； wherein, is the sample is the probability of belonging to cluster k, is the data point, is the center of cluster k, and K is the number of clusters.

6. The automatic zipper sewing control system according to claim 1, wherein: In Step 5: Ensemble learning and reinforcement learning optimization, it specifically includes two parts: model weighting strategy optimization and further optimization of ensemble learning. Among them, model weighting strategy optimization is divided into model weighting strategy optimization and adaptive voting mechanism; The optimization of the model weighting strategy is based on the prediction performance of each sub-model in different environments, dynamically adjusts the weight of each model, and uses the accuracy evaluation metrics of the model, namely AUC, F1-score, and mean squared error, to adjust the weight: ； Among them, is the prediction accuracy of the i-th model in a certain time period or working condition. M is the total number of integrated models. The weights are dynamically adjusted according to the current fault prediction effect within different time windows. When a model successfully predicts multiple faults in a short time, the system will give it a preferential weight; The adaptive voting mechanism makes predictions according to the weights of the models by introducing a weighted voting mechanism: ； Among them, are the weights of the models, is the prediction result of the i-th model. The model with a larger weight will occupy a larger share in the voting.

7. An automatic zipper sewing control system according to claim 6, characterized in that: The further optimization of the ensemble learning is specifically divided into the expansion of the state space and action space in reinforcement learning and the optimization of the reward function; In the expansion of the state space and action space in the reinforcement learning, historical data of sensors and the evolution of fault modes are used to expand the state space; For the optimization of the reward function, for time series data, the reward is designed according to the duration of the system state. By increasing the penalty for not predicting faults in time, the learning ability of the model is strengthened, and potential fault modes are avoided from being missed.

8. An apparatus for an automatic zipper sewing control system according to any one of claims 1-7, characterized in that: It includes a laser bag-opening platform (1), a laser cutting mechanism (2), a fabric grabbing and feeding mechanism (3), a fabric placement platform (4), and a fabric sewing machine (5); Among them, on the upper end surface of the laser bag-opening platform (1), a fabric placement platform (4), a fabric grabbing and feeding mechanism (3), a laser cutting mechanism (2), and a fabric sewing machine (5) are sequentially distributed from left to right; The fabric is placed on the fabric placement platform (4). The fabric grabbing and feeding mechanism (3) grabs the fabric to the laser cutting mechanism (2). The lower fabric is laser-cut by the laser cutting mechanism (2), and then the cut fabric is sewn by the fabric sewing machine (5).

Images