Electronic selvage control method and system based on FPGA (Field Programmable Gate Array)

Abstract

The invention provides an electronic selvedge control method and system based on an FPGA, and relates to the technical field of electronic selvedge control, and the method comprises the steps: collecting tension data and thickness data of a selvedge fabric, and rotating speed data, torque data and temperature data of an operation motor, and carrying out the preprocessing of the collected data; performing feature extraction on the preprocessed data to obtain key features; quickly fusing the key features; based on a deep learning algorithm, constructing a state recognition model, and classifying and recognizing the fused key features to obtain an operation mode and a fabric state; according to the operation mode and the fabric state, constructing an optimal control decision; according to the optimal control decision, a fault early warning model is constructed, a fault risk coefficient is obtained, and then the electronic selvedge control system is accurately controlled; and according to the fault risk coefficient, a control optimization self-learning algorithm is constructed, and the fault early warning model is optimized, so that safe and stable operation of the electronic selvage control system is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic edge control, and specifically to an FPGA-based electronic edge control method and system. Background Art

[0002] Traditional electronic edge control systems are mainly implemented using or general-purpose processors, which will then result in multi-sensor data processing lag: the data acquisition frequencies of multiple channels such as tension, thickness, and rotational speed are limited, and it is impossible to capture high-frequency fluctuation characteristics in real time; the real-time performance of state recognition is insufficient: traditional algorithms based on rules are difficult to handle complex working conditions, resulting in delays in defect detection such as edge wear and abnormal thickness; the response of control decisions is slow: the delay from data acquisition to control output reaches hundreds of milliseconds, which cannot meet the requirements of high-speed production scenarios; the system scalability is poor: traditional architectures are difficult to integrate deep learning algorithms for model iteration and optimization. Therefore, an FPGA-based electronic edge control method and system are provided. Summary of the Invention

[0003] In order to solve the above technical problems, the purpose of the present invention is to provide an FPGA-based electronic edge control method and system.

[0004] In order to achieve the above purpose, the present invention provides the following technical solution: An FPGA-based electronic edge control method, the method comprising: Based on multi-channel parallel acquisition technology, acquiring tension data, thickness data of the edge-hemmed fabric, rotational speed data, torque data, and temperature data of the running motor; based on data preprocessing technology, preprocessing the acquired data; Performing feature extraction on the preprocessed data to obtain key features of the running state of the electronic edge control system and the quality of the edge-hemmed fabric; based on the parallel processing ability of the FPGA, quickly fusing the key features; Based on a deep learning algorithm, constructing a state recognition model, classifying and recognizing the fused key features to obtain the running mode of the electronic edge control system and the fabric state of the edge-hemmed fabric; Constructing an optimal control decision according to the running mode and the fabric state; according to the optimal control decision, constructing a fault warning model to obtain the fault risk coefficient of the electronic edge control system, and then precisely controlling the electronic edge control system; Constructing a control optimization self-learning algorithm according to the fault risk coefficient; according to the control optimization self-learning algorithm, optimizing the fault warning model, and then ensuring the safe and stable operation of the electronic edge control system.

[0005] Further, the process of performing feature extraction on the preprocessed data to obtain key features of the running state of the electronic edge control system and the quality of the edge-hemmed fabric includes: Obtain the pre - processed tension data, thickness data, rotational speed data, torque data, and temperature data at different acquisition times within the same acquisition cycle; Based on the pre - processed tension data, thickness data, rotational speed data, torque data, and temperature data, obtain the key features of the corresponding tension data, key features of the thickness data, key features of the rotational speed data, key features of the torque data, and key features of the temperature data.

[0006] Furthermore, the process of quickly fusing the key features includes: Read the key features of the tension data, key features of the thickness data, key features of the rotational speed data, key features of the torque data, and key features of the temperature data from the FPGA; Construct feature quantization and standardization codes based on format and Z - score technology, and perform feature quantization and standardization respectively; Concatenate the key features after feature quantization and standardization respectively to obtain a set of feature vectors, including: tension feature vector set, thickness feature vector set, rotational speed feature vector set, torque feature vector set, and temperature feature vector set; Based on 's parallel processing ability, simultaneously process the tension feature vector set, thickness feature vector set, rotational speed feature vector set, torque feature vector set, and temperature feature vector set to obtain the correlation and weight between different feature vector sets.

[0007] Furthermore, the process of obtaining the correlation and weight between different feature vector sets includes: Construct a correlation calculation code applicable to the FPGA, use the tension feature vector set, thickness feature vector set, rotational speed feature vector set, torque feature vector set, and temperature feature vector set as input quantities, and according to the correlation calculation code, calculate the Pearson correlation coefficient between different feature vector sets, denoted as the correlation coefficient, and based on array parallel calculation, generate 's correlation matrix; Construct a dynamic weight code for the correlation coefficient applicable to the FPGA, preset a fault threshold, and obtain the weights of different feature vector sets by using the correlation matrix and the fault threshold as input quantities.

[0008] Furthermore, the process of constructing a state recognition model based on a deep learning algorithm includes: Read the comprehensive correlation features of several acquisition cycles from the FPGA , and perform standardization and format conversion processing; The processed comprehensive correlation features Divided into a training set, a validation set, and a test set; Construct the code of the long short-term memory network model and input it into the FPGA for model compilation to obtain the long short-term memory network model; Optimize the long short-term memory network model and save the long short-term memory network model with the highest accuracy on the validation set; Train the optimized long short-term memory network model according to the training set, validation set, and test set, and record it as the state recognition model.

[0009] Furthermore, the process of parameter tuning for the state recognition model includes: Construct the model optimization strategy code and call the tuner, and initialize the tuner to search for the number of units, rate, and optimizer type of the state recognition model; tune the number of units in the layer from 64 to 128, tune the rate from 0.2 to 0.3, and adopt the optimizer and apply it to the state recognition model.

[0010] Furthermore, the process of classifying and recognizing the fused key features to obtain the operating mode of the electronic edge winding control system and the fabric state of the edge wound fabric, and then constructing the optimal control decision includes: Define the state category names; Map the state category names to the optimized state recognition model, and then output the operating mode confidence level and the fabric state confidence level ; Preset the operating mode confidence threshold and the fabric state confidence threshold ; According to the operating mode confidence , fabric state confidence , operating mode confidence threshold and fabric state confidence threshold make a judgment to obtain the corresponding operating state abnormality and fabric state abnormality; According to the fabric state confidence with operating state abnormality and fabric state abnormality and the operating mode confidence compare the magnitude values, and preferentially execute the control decision corresponding to the larger confidence level, and record the control decisions that will sequentially solve the corresponding abnormal states as the optimal control decision.

[0011] Furthermore, the process of constructing a fault warning model according to the optimal control decision includes: Obtain several groups of optimal control decisions corresponding to historical acquisition cycles in different time stages, and construct a fault warning model based on a temporal convolutional network; According to the fault warning model, obtain a fault risk coefficient; input the operating mode of the actual electronic edge winding control system and the fabric state of the edge-wound fabric into the fault warning model to obtain a true fault risk coefficient label; Calculate the error between the fault risk coefficient and the true fault risk coefficient label through a loss function to obtain a loss value ; According to the optimizer and the loss value update and optimize the parameters of the fault warning model.

[0012] The present invention further provides an FPGA-based electronic edge winding control system to implement the above-mentioned FPGA-based electronic edge winding control method, including: a data acquisition and preprocessing module, a feature extraction and state recognition module, an intelligent control decision-making module, and a self-learning module; The data acquisition and preprocessing module collects the tension data, thickness data of the edge-wound fabric, the rotational speed data, torque data, and temperature data of the running motor, and preprocesses the collected data; The feature extraction and state recognition module extracts features from the preprocessed data to obtain key features of the operating state of the electronic edge winding control system and the quality of the edge-wound fabric, and quickly fuses the key features; constructs a state recognition model to classify and identify the fused key features; The intelligent control decision-making module constructs an optimal control decision according to the operating mode and the fabric state; constructs a fault warning model according to the optimal control decision to obtain the fault risk coefficient of the electronic edge winding control system, and then precisely controls the electronic edge winding control system; The optimization self-learning module constructs a control optimization self-learning algorithm according to the fault risk coefficient; optimizes the fault warning model according to the control optimization self-learning algorithm.

[0013] The present invention further provides a computer-readable storage medium storing a computer program, and the computer program can be executed by a processor to implement the above-mentioned FPGA-based electronic edge winding control method.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. Based on the FPGA parallel acquisition and feature extraction technology, the synchronous processing efficiency of various types of data such as tension and thickness is significantly improved, and the high-frequency feature capture ability is greatly enhanced, enabling the detection of defects such as edge wear to be advanced to early warning before faults occur, and significantly reducing the missed detection rate under high-speed working conditions. Based on the "LSTM model deployment" technology, through model optimization and hardware adaptation, low-latency inference is achieved on the FPGA, and the power consumption is significantly reduced compared with traditional GPU solutions. The state recognition accuracy is significantly improved, and the misjudgment rate of complex states such as abnormal thickness is greatly reduced.

[0015] 2. According to the Pearson correlation coefficient dynamic weight and control optimization self-learning algorithm, the system can dynamically adjust the decision-making priorities of features such as tension and speed, and significantly improve the control response speed. Through self-learning iteration, the fault warning accuracy is significantly improved, the equipment maintenance cycle is extended, and the energy consumption is reduced. Brief Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings.

[0017] Figure 1 It is a schematic diagram of the steps of an FPGA-based electronic edge control method.

[0018] Figure 2 It is a schematic diagram of the modules of an FPGA-based electronic edge control system.

[0019] Figure 3 It is a flow judgment diagram of an FPGA-based electronic edge control system. Detailed Embodiments

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be described in detail below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other implementation manners obtained by those of ordinary skill in the art without creative efforts belong to the scope protected by the present invention.

[0021] As Figure 1 shown, an FPGA-based electronic edge control method, the method includes: Based on the multi-channel parallel acquisition technology, collect the tension data, thickness data of the edge fabric, the speed data, torque data, and temperature data of the running motor; based on the data preprocessing technology, preprocess the collected data; Extract features from the preprocessed data to obtain the key features of the operating state of the electronic edge - winding control system and the quality of the edge - wound fabric; based on the parallel processing ability of FPGA, quickly fuse the key features; Based on deep - learning algorithms, construct a state recognition model to classify and identify the fused key features, and obtain the operating mode of the electronic edge - winding control system and the fabric state of the edge - wound fabric; According to the operating mode and the fabric state, construct an optimal control decision; according to the optimal control decision, construct a fault warning model to obtain the fault risk coefficient of the electronic edge - winding control system, and then precisely control the electronic edge - winding control system; According to the fault risk coefficient, construct a control - optimization self - learning algorithm; according to the control - optimization self - learning algorithm, optimize the fault warning model, and then ensure the safe and stable operation of the electronic edge - winding control system.

[0022] It should be further noted that in the specific implementation process, based on the multi - channel parallel acquisition technology, the specific process of acquiring the tension data, thickness data of the edge - wound fabric, rotational speed data, torque data, and temperature data of the operating motor includes: Set up a data acquisition device, including: a tension data acquisition unit, a thickness data acquisition unit, a rotational speed data acquisition unit, a torque data acquisition unit, and a temperature data acquisition unit; and set an acquisition period, which contains several acquisition moments; The tension data acquisition unit arranges several tension sensors at the tension measurement points of the edge - wound fabric. Based on the multi - channel parallel acquisition technology, multiple tension sensors work simultaneously. Each tension sensor converts the acquired tension data into an electrical signal, converts the electrical signal into a digital signal through an analog - to - digital converter (ADC), and transmits the digital signal to the cache module inside the FPGA through the data bus.

[0023] For example, the tension sensors evenly distributed on the edge - wound fabric simultaneously acquire tension data during the acquisition period. The multi - channel parallel acquisition function of the FPGA ensures the real - time and accuracy of the tension data.

[0024] The thickness data acquisition unit arranges several thickness sensors at the edge of the edge - wound fabric. The thickness sensors convert the thickness data at the edge of the edge - wound fabric into corresponding electrical signals, and also convert the electrical signals into digital signals through an analog - to - digital converter (ADC), and transmit the digital signals to the cache module inside the FPGA through the data bus.

[0025] For example, use capacitive thickness sensors or ultrasonic thickness sensors to simultaneously acquire thickness data at different positions at the edge of the edge - wound fabric to obtain the distribution of the fabric thickness.

[0026] The rotational speed data acquisition unit installs several rotational speed sensors on the shaft of the motor. The pulse signals generated by the rotational speed sensors are proportional to the rotational speed of the motor. Based on multi-channel parallel acquisition technology, it processes the pulse signals of multiple rotational speed sensors simultaneously; by counting and analyzing the pulse signals, it calculates the rotational speed data of the motor and stores it in the internal cache module of the FPGA.

[0027] For example, for multiple operating motors, their rotational speed data is collected simultaneously to comprehensively understand the operating state of the motors.

[0028] The torque data acquisition unit installs several torque sensors on the output shaft of the motor. The torque sensors convert torque data into electrical signals. Based on multi-channel parallel acquisition technology, it ensures the simultaneous acquisition of torque data of different motors and stores it in the internal cache module of the FPGA; It should be further noted that by analyzing the torque data, the load condition of the motor can be judged. The simultaneous acquisition of torque data of multiple motors helps to analyze the load balance situation between the motors.

[0029] The temperature data acquisition unit arranges several temperature sensors on the motor. The temperature sensors convert temperature information into electrical signals and transmit them to the internal cache module of the FPGA through the data bus.

[0030] For example, thermocouples or thermistors can be selected to simultaneously collect the temperature of the motor winding, the temperature of key components, etc., and promptly detect abnormal temperature conditions. Multi-channel parallel acquisition technology can simultaneously collect the data of multiple temperature sensors and monitor the temperature state of the system in real time.

[0031] It should be further noted that multi-channel parallel acquisition technology ensures the synchronous acquisition of thickness data at different positions, improving the acquisition efficiency and accuracy of thickness data.

[0032] It should be further noted that in the specific implementation process, based on data preprocessing technology, the specific process of preprocessing the collected data includes: Implement the filter kernel based on the hardware description language of the FPGA, and based on the multi-channel parallel architecture of the FPGA, filter the tension data, thickness data, rotational speed data, torque data, and temperature data simultaneously, and independently configure filter parameters for each type of data channel, including: filter type, filter structure parameters, cut-off frequency, transition band width, stop band attenuation, sampling frequency, pass band ripple, and phase response.

[0033] It should be further noted that the filter parameters can be dynamically adjusted according to the actual requirements for accuracy.

[0034] Obtain the filtered tension data, thickness data, rotational speed data, torque data, and temperature data according to the channels after configuring the filter parameters.

[0035] Preset standard parts; Calibrate the data acquisition device based on the standard parts, and collect groups of calibration data , where is the standard value; is the output value of the corresponding sensor; Based on the least squares method, fit the calibration curve, which is: ; where is the true value; is the output value of the corresponding sensor; , , are calibration coefficients; Obtain the calibrated true value ; Store the calibration coefficients , , in the memory of the FPGA, and complete the calibration calculation through the hardware multiplier and adder after real-time data acquisition.

[0036] For example, the output deviation of the thickness sensor at 25°C is , and after temperature compensation calibration, the measurement error in the full temperature range .

[0037] Perform multi-channel parallel normalization processing on the calibrated data, including: calculating the and of each group of data, and storing them in the register group. Each channel independently performs the normalization calculation, and uses the parallel processing unit of the FPGA to simultaneously process the data collected by the tension data acquisition unit, thickness data acquisition unit, rotational speed data acquisition unit, torque data acquisition unit, and temperature data acquisition unit to obtain the corresponding normalized data.

[0038] For example, the rotational speed data range of the motor is , and the range of the torque data is . After normalization, both are mapped to to eliminate the influence of dimension.

[0039] It should be further noted that in the specific implementation process, the specific process of extracting features from the preprocessed data to obtain the key features of the operating state of the electronic edge winding control system and the quality of the edge wound fabric includes: Feature extraction is performed on the preprocessed data to obtain the key features of the preprocessed data, specifically including: The specific process of obtaining the key features of the corresponding data of the edge-twisted fabric after preprocessing: The tension data and thickness data at different acquisition times in the same acquisition period are respectively denoted as , , where represents the corresponding acquisition time; Based on the sliding window algorithm, the mean values of the tension data and thickness data are obtained and , standard deviation and , as the basic data support. The specific acquisition process belongs to the conventional process in this field and will not be elaborated; According to the mean value of the tension data and the standard deviation , the skewness , kurtosis and impulse factor of the edge-twisted fabric are obtained; it should be further noted that the skewness represents the distribution symmetry, and the negative value indicates left skew, that is, the risk of sudden drop in tension; the impulse factor represents the impact characteristic of the tension data, detecting sudden changes in tension, that is, the precursor of the fracture of the edge-twisted fabric; the skewness ; where the function represents the mathematical expectation; the kurtosis ; the impulse factor ; where the function represents the maximum value, represents the root mean square value. The specific acquisition process belongs to the conventional process in this field and will not be elaborated; Based on , the fundamental frequency amplitude and harmonic content of the tension data in the frequency domain are obtained; it should be further noted that the fundamental frequency amplitude is used to detect the tension fluctuation caused by power supply interference; the harmonic content is used to evaluate the tension harmonics caused by the nonlinearity of the transmission system, such as the gear meshing frequency; The harmonic content ; where represents the th harmonic amplitude; represents the upper limit of the harmonic order; is the fundamental frequency. The specific acquisition process belongs to the conventional process in this field and will not be elaborated; Based on , the time-frequency spectrum mutation , wavelet detail coefficients and wavelet energy entropy ; It should be further noted that the sudden change in the time-frequency spectrum indicates the sudden enhancement or attenuation of specific frequency components detected in the time-frequency spectrum, that is, the edge breakage of the hemming fabric, such as a sudden increase in frequency noise; the wavelet energy entropy represents the change in the complexity of the tension fluctuation. Excessive or too small values indicate the failure of the control system. The specific acquisition process belongs to the conventional process in this field and will not be elaborated here.

[0040] Based on the time-domain feature extraction technology and according to the mean value and standard deviation of the thickness data, the time-domain features of the hemming fabric thickness data are obtained, including the mean deviation , local standard deviation , gradient and change slope ; It should be further noted that the mean deviation represents the deviation between the actually collected thickness and the set value, which directly affects the compliance of the product specifications; the local standard deviation is used to detect the thickness non-uniformity, such as stripe defects; the gradient represents the thickness change rate to identify sudden thickness changes, such as a sudden drop in the edge thickness; the change slope is used to judge whether the thickness transition region meets the process requirements, such as the gradual change of the coating thickness. The specific acquisition process belongs to the conventional process in this field and will not be elaborated here; Based on the frequency-domain feature extraction technology, the frequency-domain features of the hemming fabric thickness data are obtained, including spectral kurtosis and cut-off frequency; It should be further noted that the spectral kurtosis is used to detect sudden thickness anomalies, such as spikes caused by impurities; the cut-off frequency is used to evaluate the high-frequency components of the thickness fluctuation, reflecting the resolution of the measurement system. The specific acquisition process belongs to the conventional process in this field and will not be elaborated here; The specific process of obtaining the key features of the corresponding data of the motor after preprocessing: The rotational speed data, torque data and temperature data in the acquisition period are respectively denoted as , and , where represents the corresponding acquisition moment; Based on the sliding window algorithm, the mean values , and , standard deviations , and of the rotational speed data, torque data and temperature data are obtained as the basic data support; Based on time-domain feature extraction technology, the time-domain features of rotational speed data, torque data, and temperature data are obtained, which are respectively the rotational speed volatility, overshoot, and rotational speed fluctuation period of the rotational speed data, the torque volatility, peak torque ratio, and torque change rate of the torque data, and the temperature gradient, temperature fluctuation, and thermal time constant of the temperature data; it should be further noted that the excessive fluctuation of the volatility of the rotational speed data leads to uneven fabric tension, the overshoot represents the rotational speed overshoot in the step response, reflecting the damping ratio of the control system; the rotational speed fluctuation period is used to identify periodic load fluctuations, such as mechanical resonance; the excessive fluctuation of the torque volatility of the torque data leads to fatigue of the transmission system, such as wear of the coupling, the peak torque ratio is used to evaluate the degree of impact load, such as the peak torque during loom beating-up, and the torque change rate is used to detect torque mutations, such as the sudden increase in torque caused by fabric jamming. The specific acquisition process belongs to the conventional process in this field and will not be elaborated further.

[0041] Based on frequency-domain feature extraction technology, the frequency-domain features of rotational speed data, torque data, and temperature data are obtained, which are respectively the rotational frequency, rotational frequency sideband, and spectral entropy of the rotational speed data, the gear meshing frequency and torque ripple coefficient of the torque data, and the low-frequency temperature fluctuation, temperature spectral peak, and temperature-load correlation of the temperature data; it should be further noted that the rotational frequency of the rotational speed data is used to determine the motor base frequency; the rotational frequency sideband is used to identify load fluctuations, such as the increase in sideband energy caused by conveyor belt slipping; the spectral entropy represents the change in spectral complexity, and too large or too small indicates bearing failure or uneven load; the gear meshing frequency of the torque data is used to detect gear wear; the torque ripple coefficient is used to evaluate the smoothness of the motor output, such as the ripple coefficient , the influence of motor output vibration; the low-frequency temperature fluctuation of the temperature data is used to identify periodic faults of the cooling system, such as the fluctuation of the fan speed; the temperature spectral peak is used to locate the oscillation of the control system, such as the temperature oscillation caused by inappropriate parameters; the temperature-load correlation is used to evaluate the influence of load changes on temperature, such as a decrease in the temperature-load correlation indicating the failure of the heat dissipation system. The specific acquisition process belongs to the conventional process in this field and will not be elaborated further.

[0042] It should be further noted that in the specific implementation process, based on the parallel processing ability of FPGA, the specific process of quickly fusing key features includes: Reading the key features of tension data, thickness data, rotational speed data, torque data, and temperature data from the FPGA; Constructing feature quantization and standardization codes based on format and Z-score technology, and performing feature quantization and standardization respectively; For example, the local standard deviation of thickness Format: , rounded to , the actual represented value is , if the precision standard is , the actual represented value is , the error is within the precision range of the format, meeting the quantization requirements.

[0043] Perform feature vector splicing on the key features after feature quantization and standardization respectively to obtain a set of feature vectors, including: a set of tension feature vectors, a set of thickness feature vectors, a set of rotational speed feature vectors, a set of torque feature vectors, and a set of temperature feature vectors.

[0044] Based on the parallel processing ability of, simultaneously process the set of tension feature vectors, the set of thickness feature vectors, the set of rotational speed feature vectors, the set of torque feature vectors, and the set of temperature feature vectors to obtain the correlation and weight between different sets of feature vectors. The specific process includes: It should be further noted that the format of the set of tension feature vectors is: ; The format of the set of thickness feature vectors is: [mean deviation, local standard deviation, gradient, change slope, spectral warp, cut-off frequency]; The format of the set of rotational speed feature vectors is: [volatility, overshoot, rotational speed fluctuation period, rotational frequency, rotational frequency sideband, spectral entropy]; The format of the set of torque feature vectors is: [torque volatility, peak torque ratio, torque change rate, gear meshing frequency, torque ripple coefficient]; The format of the set of temperature feature vectors is: [temperature gradient, temperature fluctuation, thermal time constant, low-frequency temperature fluctuation, temperature spectral peak, temperature load correlation].

[0045] Construct a correlation calculation code applicable to FPGA, use the set of tension feature vectors, the set of thickness feature vectors, the set of rotational speed feature vectors, the set of torque feature vectors, and the set of temperature feature vectors as input quantities. According to the correlation calculation code, calculate the Pearson correlation coefficient between different sets of feature vectors, denoted as the correlation coefficient, and according to array parallel calculation, generate the correlation matrix of; For example, represent the correlation matrix in tabular form: As shown in Table 1: Correlation coefficient Tension eigenvector set Thickness eigenvector set Rotational speed eigenvector set Torque eigenvector set Temperature eigenvector set Tension eigenvector set 1.000 0.653 0.127 0.812 0.345 Thickness eigenvector set 0.653 1.000 0.328 0.547 0.189 Rotational speed eigenvector set 0.127 0.328 1.000 0.763 0.621 Torque eigenvector set 0.812 0.547 0.763 1.000 0.476 Temperature eigenvector set 0.345 0.189 0.621 0.476 1.000 Build the dynamic weight code of the correlation coefficient applicable to FPGA, preset the fault threshold, and obtain the weights of different feature vector sets by taking the correlation matrix and the fault threshold as input quantities. It should be further noted that the fault threshold is set according to the actual tension feature vector set, thickness feature vector set, rotation speed feature vector set, torque feature vector set, and temperature feature vector set.

[0046] For example, normal state: tension feature vector set (0.35) + thickness feature vector set (0.25) + rotation speed feature vector set (0.15) + torque feature vector set (0.15) + temperature feature vector set (0.10); Abnormal state: When abnormal tension fluctuation is detected, it is dynamically adjusted to: Tension feature vector set (0.60) + thickness feature vector set (0.15) + rotation speed feature vector set (0.05) + torque feature vector set (0.15) + temperature feature vector set (0.05).

[0047] Based on the weighted average method, quickly fuse the key features to generate the comprehensive correlation feature ; The comprehensive correlation feature ; Among them, is the correlation coefficient; is the product of the weights corresponding to two correlation coefficients; It should be further noted that in the specific implementation process, based on the deep learning algorithm, the specific process of building the state recognition model includes: Read the comprehensive correlation features of several acquisition cycles from the FPGA ; Build the data normalization code and save it to the FPGA. The data processing unit in the FPGA executes the data normalization code to normalize the comprehensive correlation features to ensure the stability of model training.

[0048] Build the format conversion code and save it to the FPGA. The data processing unit in the FPGA executes the format conversion code to convert the comprehensive correlation features into the required time series data format.

[0049] For example, set the time step , and expand the comprehensive correlation features of each acquisition cycle into a sequence containing the first 10 time points to form a three-dimensional array for capturing the time dependence of the data.

[0050] It should be further noted that the format of the three-dimensional array is: [number of samples, time step, number of features].

[0051] The comprehensive correlation features after converting the time series data format are divided into a training set, a validation set, and a test set; Build the code of the long short-term memory network model and input it into the FPGA for model compilation to obtain the long short-term memory network model; According to the long short-term memory network model, configure the callback function. When the validation set loss does not decrease for 5 consecutive rounds, stop training and restore the best weights to avoid overfitting. When the validation set loss stagnates, multiply the learning rate of the long short-term memory network model by 0.2, and the long short-term memory network model jumps out of the local optimum. Save the long short-term memory network model with the highest validation set accuracy to ensure the use of the best parameters during deployment.

[0052] Set the sample size of several comprehensive correlation features to 64 to balance memory occupancy and gradient estimation accuracy; train for 50 rounds, and shuffle the data order through the code instructions to avoid periodic deviation.

[0053] Record the losses and accuracies of the training set and the validation set in real time, generate curves to visualize the training dynamics, and determine whether the trained long short-term memory network model converges or overfits; record the trained long short-term memory network model as the state recognition model.

[0054] It should be further noted that the specific process of generating curves based on the losses and accuracies of the training set and the validation set includes: Create parameters which are used to set the canvas width and height; create parameters for creating a subplot layout, and the subplot layout includes: a left subplot and a right subplot; the left subplot is used to describe the loss curve, and the right subplot is used to describe the accuracy curve; create parameters which are respectively used to determine the line styles of the curves; Based on instructions, automatically display the loss curve and the accuracy curve, and refresh the loss curve and the accuracy curve by adding a timer; it should be further noted that the state judgment of the loss curve: 1. Ideal situation: The training loss and the validation loss decrease synchronously, and the validation loss is slightly higher than the training loss. Eventually, the loss tends to be stable without significant fluctuations; 2. Abnormal situation: Overfitting: The training loss continues to decrease, and the validation loss first decreases and then increases; Underfitting: Both the training loss and the validation loss are higher than expected and decrease slowly. The state judgment of the accuracy curve:

[0055] ​1. Ideal situation: The training accuracy and validation accuracy increase simultaneously, and finally the validation accuracy approaches the training accuracy; 2. Abnormal situations: Data imbalance: The validation accuracy fluctuates greatly and the sample accuracy is significantly low; Model instability: The accuracy shows irregular oscillations in the later stage.

[0056] Parameter tuning is performed on the state recognition model, and the specific process includes: Construct the model optimization strategy code and call the tuner and initialize the tuner to search for the number of units of the state recognition model ( , step size 16), rate ( , step size ), optimizer type ( ), and then obtain the optimal parameters.

[0057] It should be further noted that to maximize the validation set accuracy, use the algorithm to dynamically allocate computing resources and efficiently search for the optimal parameter combination.

[0058] Apply the optimal parameters obtained from tuning, such as the number of units in the layer is tuned from 64 to 128, the rate is tuned from 0.2 to 0.3, and use the optimizer and apply it to the state recognition model to improve the generalization ability.

[0059] It should be further noted that in the layer and the fully connected layer, add regularization to constrain the complexity of the state recognition model; increase the rate to 0.3 to further suppress overfitting and ensure the stability of the model on unknown data.

[0060] It should be further noted that in the specific implementation process, classifying and identifying the fused key features to obtain the operating mode of the electronic edge winding control system and the fabric state of the edge wound fabric, and then the specific process of constructing the optimal control decision includes: Define the state category names, including: The operating modes of the electronic edge winding control system are normal operation, edge fluctuation, high-speed production, and low-speed debugging; The fabric states of the edge wound fabric are qualified state, edge wear, thickness abnormality, and comprehensive defect; Map the state category names to the optimized state recognition model, and then output the confidence of the operating mode for the corresponding acquisition period and the confidence of the fabric state ; The confidence of the operating mode , including: normal operation confidence, edge fluctuation confidence, high-speed production confidence, and low-speed debugging confidence; fabric state confidence , including: qualified state confidence, edge wear confidence, thickness anomaly confidence, and comprehensive defect confidence; Preset operating mode confidence threshold And fabric state confidence threshold ; If the actual operating mode confidence , then it is determined that the operating mode is high confidence, and a green light is responded, and there is no abnormal operating state; If the actual operating mode confidence , then it is determined that the operating mode is medium confidence, and a yellow light is responded, there is a risk of abnormal operating state, and the maintenance personnel are reminded to check; If the actual operating mode confidence , then it is determined that the operating mode is low confidence, and a red light is responded, there is an abnormal operating state, and the maintenance personnel are reminded to implement the corresponding control decision.

[0061] If the actual fabric state confidence , then it is determined that the fabric state is high confidence, and a green light is responded, and there is no abnormal fabric state; If the actual fabric state confidence , then it is determined that the fabric state is medium confidence, and a yellow light is responded, there is a risk of abnormal fabric state, and the maintenance personnel are reminded to check; If the actual fabric state confidence , then it is determined that the fabric state is low confidence, and a red light is responded, there is an abnormal fabric state, and the maintenance personnel are reminded to implement the corresponding control decision; It should be further noted that the higher the operating mode confidence and the fabric state confidence , the corresponding operating mode and fabric state are recorded.

[0062] According to the comparison of the size values of the fabric state confidence and the operating mode confidence with abnormal operating state and abnormal fabric state, the control decision corresponding to the larger confidence is preferentially executed, and the control decisions for solving the corresponding abnormal states will be solved in sequence, which is recorded as the optimal control decision. The control decisions for the corresponding states will not be elaborated.

[0063] For example, if the edge fluctuation confidence level is greater than the high-speed production confidence level, the control decision for edge fluctuation is executed first, followed by the control decision for high-speed production; if the high-speed production confidence level is greater than the edge wear confidence level, the control decision for high-speed production is executed first, followed by the control decision for edge wear; if the edge fluctuation confidence level is equal to the high-speed production confidence level, the control decision for the corresponding state can be executed arbitrarily.

[0064] It should be further noted that in the specific implementation process, the specific process of constructing a fault warning model according to the optimal control decision includes: Obtain several groups of optimal control decisions corresponding to the historical collection period within different time stages; it should be further noted that the optimal control decisions executed in different time stages are universal, reducing the influence of time factors on the model; Group and label the optimal control decisions corresponding to the historical collection period within several groups of different time stages, denoted as is a natural number; Take groups of optimal control decisions corresponding to the historical collection period within different time stages as sample data, and is a natural number less than , denoted as the sample set; Take the remaining several groups of optimal control decisions corresponding to the historical collection period within different time stages as the test set; according to the sample set and the test set, form a training sample set; Based on the temporal convolutional network, construct a standard fault warning model; And input the training sample set into the standard fault warning model to train the standard fault warning model, and denote the completed trained standard fault warning model as the fault warning model.

[0065] It should be further noted that in the specific implementation process, according to the fault risk coefficient, the specific process of constructing a control optimization self-learning algorithm to optimize the fault warning model includes: Based on the cross-entropy loss function, measure the difference between the predicted operation mode of the electronic edge winding control system and the fabric state of the edge-wound fabric and the actual operation mode of the electronic edge winding control system and the fabric state of the edge-wound fabric, and select the optimizer for optimization, and the specific process includes: According to the fault warning model, obtain the fault risk coefficient; input the actual operation mode of the electronic edge winding control system and the fabric state of the edge-wound fabric into the fault warning model to obtain the true fault risk coefficient label; Calculate the error between the fault risk coefficient and the true fault risk coefficient label through the loss function, and the formula is: ; where is the loss value; is the true fault risk coefficient label, and the value is or ; is the probability that the category predicted by the fault warning model is ; is the probability that the category predicted by the fault warning model is ; According to the optimizer updates the parameters of the fault warning model. Based on minimizing the loss function, the update formula is: ; where is the learning rate; are the model parameters of the fault warning model. By updating the fault warning model is optimized, thereby ensuring the safe and stable operation of the electronic edge winding control system.

[0066] As Figure 2 shown, an FPGA-based electronic edge winding control system includes: a data acquisition and preprocessing module, a feature extraction and state recognition module, an intelligent control decision-making module, and a self-learning module; The data acquisition and preprocessing module acquires the tension data, thickness data of the edge winding fabric, the rotational speed data, torque data, and temperature data of the running motor, and preprocesses the acquired data; The feature extraction and state recognition module extracts features from the preprocessed data to obtain the key features of the operating state of the electronic edge winding control system and the quality of the edge winding fabric, and quickly fuses the key features; constructs a state recognition model to classify and recognize the fused key features; The intelligent control decision-making module constructs an optimal control decision according to the operating mode and the fabric state; constructs a fault warning model according to the optimal control decision to obtain the fault risk coefficient of the electronic edge winding control system, and then precisely controls the electronic edge winding control system; The optimization self-learning module constructs a control optimization self-learning algorithm according to the fault risk coefficient; optimizes the fault warning model according to the control optimization self-learning algorithm.

[0067] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An FPGA-based electronic edge control method, characterized in that, The method includes: Based on the multi-channel parallel acquisition technology, acquire the tension data, thickness data of the edge-hemmed fabric, the rotational speed data, torque data and temperature data of the running motor; based on the data preprocessing technology, preprocess the acquired data; Extract features from the preprocessed data to obtain the key features of the operating state of the electronic edge-hem control system and the quality of the edge-hemmed fabric; based on the parallel processing ability of FPGA, quickly fuse the key features; Based on the deep learning algorithm, construct a state recognition model to classify and identify the fused key features, and obtain the operating mode of the electronic edge-hem control system and the fabric state of the edge-hemmed fabric; Construct an optimal control decision according to the operating mode and the fabric state; according to the optimal control decision, construct a fault warning model to obtain the fault risk coefficient of the electronic edge-hem control system, and then precisely control the electronic edge-hem control system; Construct a control optimization self-learning algorithm according to the fault risk coefficient; according to the control optimization self-learning algorithm, optimize the fault warning model, and then ensure the safe and stable operation of the electronic edge-hem control system.

2. The electronic edge control method based on FPGA according to claim 1, wherein The process of extracting features from the preprocessed data to obtain the key features of the operating state of the electronic edge-hem control system and the quality of the edge-hemmed fabric includes: Obtain the preprocessed tension data, thickness data, rotational speed data, torque data and temperature data at different acquisition times in the same acquisition period; According to the preprocessed tension data, thickness data, rotational speed data, torque data and temperature data, obtain the key features of the corresponding tension data, thickness data, rotational speed data, torque data and temperature data.

3. The electronic edge control method based on FPGA according to claim 2, characterized in that, The process of quickly fusing the key features includes: Read the key features of the tension data, thickness data, rotational speed data, torque data and temperature data from the FPGA; Construct based on Format and Z-score technology for feature quantization and standardization code, respectively perform feature quantization and standardization; Respectively splice the key features after feature quantization and standardization into feature vectors to obtain a set of feature vectors, including: a set of tension feature vectors, a set of thickness feature vectors, a set of rotational speed feature vectors, a set of torque feature vectors and a set of temperature feature vectors; Based on parallel processing capabilities, the tension feature vector set, thickness feature vector set, rotational speed feature vector set, torque feature vector set, and temperature feature vector set are processed simultaneously to obtain the correlations and weights between different feature vector sets.

4. The electronic edge control method based on FPGA according to claim 3, wherein The process of obtaining the correlation and weight between different sets of feature vectors includes: Construct the correlation calculation code applicable to FPGA, take the set of tension feature vectors, the set of thickness feature vectors, the set of rotational speed feature vectors, the set of torque feature vectors, and the set of temperature feature vectors as input quantities, and according to the correlation calculation code, calculate the Pearson correlation coefficient between different feature vector sets, denoted as the correlation coefficient, and according to array parallel calculation, generate correlation matrix; Construct a dynamic weight code of the correlation coefficient applicable to the FPGA, preset a fault threshold, and obtain the weights of different sets of feature vectors by using the correlation matrix and the fault threshold as input quantities.

5. A method for electronically controlling the edge hemming based on FPGA according to claim 4, characterized in that, The process of constructing a state recognition model based on the deep learning algorithm includes: Read the comprehensive correlation features of several acquisition cycles from the FPGA , and perform standardization and format conversion processing; The processed comprehensive correlation features are divided into a training set, a validation set, and a test set; Construct a long short-term memory network model code and input it into the FPGA for model compilation to obtain a long short-term memory network model; Optimize the long short-term memory network model and save the long short-term memory network model with the highest accuracy of the validation set; Train the optimized long short-term memory network model according to the training set, validation set and test set, and denote it as the state recognition model.

6. The electronic edge control method based on FPGA according to claim 5, characterized in that, The process of tuning the parameters of the state recognition model includes: Build the model optimization strategy code and call the tuner, and initialize the tuner to search for the number of units, rate, and optimizer type of the state recognition model; set the number of units in the layer from 64 to 128, the rate from 0.2 to 0.3, and adopt the optimizer and apply it to the state recognition model.

7. A method for controlling electronic edge hemming based on FPGA according to claim 6, characterized in that The process of classifying and identifying the fused key features to obtain the operating mode of the electronic edge - winding control system and the fabric state of the edge - wound fabric, and then constructing the optimal control decision includes: Defining the state category names; Map the state category name to the optimized state recognition model, and then output the operation mode confidence of the corresponding acquisition period and the fabric state confidence ; Preset confidence threshold for the operating mode and confidence threshold for the fabric state ; According to the running mode confidence , fabric state confidence , running mode confidence threshold and fabric state confidence threshold to make a judgment, and then obtain the corresponding abnormal running state and abnormal fabric state; According to the fabric state confidence level indicating abnormal operating status and abnormal fabric state and the operating mode confidence level Based on the comparison of the magnitude values, the control decision corresponding to the larger confidence level is preferentially executed, and the control decisions for resolving the corresponding abnormal states will be sequentially recorded as the optimal control decisions.

8. A method for electronically controlling the edge hemming based on FPGA according to claim 7, characterized in that The process of constructing a fault warning model according to the optimal control decision includes: Obtaining several groups of optimal control decisions corresponding to historical acquisition cycles in different time stages, and constructing a fault warning model based on a temporal convolutional network; According to the fault warning model, obtaining a fault risk coefficient; inputting the operating mode of the actual electronic edge - winding control system and the fabric state of the edge - wound fabric into the fault warning model to obtain a true fault risk coefficient label; Calculate the error between the failure risk coefficient and the true failure risk coefficient label through a loss function to obtain a loss value ; According to Optimizer and loss value Update and optimize the parameters of the fault warning model.

9. An FPGA-based electronic edge hemming control system that implements the FPGA-based electronic edge hemming control method described in any one of claims 1-8 above, characterized in that, Including: A data acquisition and pre - processing module, a feature extraction and state recognition module, an intelligent control decision - making module, and a self - learning module; The data acquisition and pre - processing module collects the tension data, thickness data of the edge - wound fabric, the rotational speed data, torque data, and temperature data of the operating motor, and pre - processes the collected data; The feature extraction and state recognition module extracts features from the pre - processed data to obtain the key features of the operating state of the electronic edge - winding control system and the quality of the edge - wound fabric, and quickly fuses the key features; constructs a state recognition model to classify and identify the fused key features; The intelligent control decision - making module constructs an optimal control decision according to the operating mode and the fabric state; constructs a fault warning model according to the optimal control decision, obtains the fault risk coefficient of the electronic edge - winding control system, and then precisely controls the electronic edge - winding control system; The optimized self - learning module constructs a control optimization self - learning algorithm according to the fault risk coefficient; optimizes the fault warning model according to the control optimization self - learning algorithm.

10. A computer-readable storage medium, characterized in that, The computer - readable storage medium stores a computer program, and the computer program can be executed by a processor to implement a method for electronic edge - winding control based on FPGA according to any one of claims 1 - 8 above.

Images