Dynamic size adjustment interface optimization method for down jacket quilting machine

Abstract

The invention provides a dynamic size adjustment interface optimization method for a down jacket quilting machine, and relates to the technical field of monitoring or compensation optimization, acoustic and vibration signals are utilized, fabric tension and down unevenness at the next moment are accurately predicted through a two-channel prediction model, and control is improved from hysteresis response to active intervention; based on a prediction result, before defects are formed, actively homogenizing down feather distribution through a reconstructed electric field, and presetting a compensation amount through a path and pressure cooperative compensation module; and at the sewing moment, in-situ fixing is carried out by a self-adaptive locking electric field, so that three-dimensional cooperative inhibition on dynamic disturbance is formed. According to the method, the microcosmic quality serves as a core drive, stitch deformation and cold spots are eliminated from the source, the control targets of the macroscopic size precision and the microcosmic quality are unified, and the consistency and the high-end value of a final product are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of monitoring or compensation optimization, in particular to a dynamic size adjustment interface optimization method for a down jacket quilting machine. BACKGROUND

[0002] In the field of high-end functional garment manufacturing, especially for the production of down jackets using ultra-thin fabric and high-loft down, based on ultra-thin, high-density nylon fabric, its surface is smooth, low in elasticity but easy to deform; water-repellent eiderdown is extremely fluffy and light, easily displaced by air flow or static electricity; the uniformity of down distribution in each quilting grid and the flatness of the fabric are extremely high, and any "cold spots" or "micro-creases" are defective products; interface refers not only to the user interface (UI), but also to the deeper interaction channel of data and control instructions; that is, the software and hardware protocol, data format and processing logic between external parameters (such as design data, sensor data) and quilting machine motion controllers (such as servo motors, drivers); due to the extreme lightness, smoothness, softness and elasticity of the material, small tension unevenness of the fabric and random displacement of the down during high-speed quilting can cause stitch deformation, local wrinkles and uneven down distribution (forming "cold spots"), which seriously affect the appearance, comfort and warmth of the product; traditional quilting machines face great challenges; existing technologies either optimize the version before production or perform lagging mechanical compensation during production, neither of which can solve these dynamic, random micro-defects from the root.

[0003] In the prior art, a garment pattern making parameter optimization method based on big data is disclosed, with publication number CN118761174A, which includes S1, obtaining original data related to garment pattern making from multiple specific channels; S2, obtaining a set of garment pattern making parameters; S3, constructing a comprehensive data model through the set of garment pattern making parameters; S4, generating virtual samples through a generative adversarial network;

[0004] S5, obtaining the prediction results of the comprehensive data model; S6, based on the prediction results, constructing a dynamic simulation environment to simulate the influence of different pattern making parameters on garment fit and comfort; S7, adjusting the pattern making parameters according to the feedback of dynamic simulation and virtual fitting, applying the optimized parameters to actual production, and performing secondary optimization through continuous monitoring of feedback data; S8, collecting feedback data in the actual production process.

[0005] The existing technology mainly has the following three core technical bottlenecks in the dynamic size adjustment interface optimization for down jacket quilting machines:

[0006] 1. "Uncertainty" of material state and "blindness" of control: the prior art lacks real-time and accurate sensing means for the micro-physical state of the material in the production process. The local tension change of the fabric and the random density distribution of the down filling, which are the two key variables that determine the final quality of the quilting, are in a "black box" state during processing. This leads to the inability of the control system to obtain the necessary input information for accurate adjustment, and the path and pressure instructions executed by the system are based on idealized models rather than real working conditions, resulting in blindness in control and the inability to avoid stitch deformation and "cold spot" problems caused by uneven material state.

[0007] 2. "Dynamic disturbance" of the processing process and "hysteresis" of the response: the dynamic mechanical disturbance generated by the high-speed movement of the quilting machine is instantaneous and nonlinear, while the response mechanism of the prior art to such disturbance is passive and hysteresis. The system usually detects and compensates through sensors after defects such as wrinkles have been formed. This "after-the-fact" control logic is too late for light and sensitive materials and cannot effectively suppress the occurrence of disturbances in the embryonic stage, leading to the accumulation of dynamic defects.

[0008] 3. Contradiction between "macro" control target and "micro" quality demand: the optimization target of the prior art mainly focuses on the macro-geometric indicators such as the overall profile size of the finished product, while ignoring the micro-quality that determines the value of high-end products, such as the down fullness within a single quilting unit and the absolute flatness of the fabric. The control strategy is fragmented, with the path control, pressure control and other subsystems running independently, lacking a unified framework to integrate the macro-size and micro-quality targets, and even sacrificing micro-quality when adjusting macro-size.

[0009] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0010] The purpose of the present application is to provide a dynamic size adjustment interface optimization method for a down jacket quilting machine to solve the problems raised in the background art.

[0011] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0012] A dynamic size adjustment interface optimization method for a down jacket quilting machine, the specific steps comprising:

[0013] Step S1: Collect and process the original signal data of the target fabric to extract the needle puncture acoustic spectrum high-frequency energy proportion strongly related to the fabric tension state and the foot vibration specific frequency band attenuation rate strongly related to the down density state, respectively;

[0014] Step S2: input the needle acupuncture acoustic spectrum high frequency energy proportion and the time sequence of the specific frequency band attenuation rate of the presser foot vibration into the pre-constructed double-channel prediction model respectively, and the double-channel prediction model is used for predicting the predicted tension abnormality degree of the next quilting path node and the predicted down unevenness of the next unit to be quilted;

[0015] Step S3: receiving the predicted tension abnormality degree and sending to the path and pressure cooperative compensation module to calculate the path compensation vector through a nonlinear gain function;

[0016] And based on the predicted tension abnormality degree and the predicted down unevenness, the presser foot pressure adjustment value is calculated through a cooperative weighting model;

[0017] Step S4: based on the predicted down unevenness, secondary analysis is carried out to generate a reconstructed voltage matrix for reconstructing the down distribution in the next unit to be quilted;

[0018] Step S5: when the needle of the quilting machine is about to enter the next unit to be quilted which has been reconstructed, the new reconstructed voltage matrix is switched, and then the adaptive locking voltage matrix is applied to the microelectrode array corresponding to the next unit to be quilted, so as to fix the target fabric in place by using the electrostatic adsorption force, and at the same time, the path and pressure cooperative compensation module is activated.

[0019] Compared with the prior art, the beneficial effects of the present application are:

[0020] The "precise prediction" and "active intervention" of the microstate of the material are realized: by innovatively extracting the needle acupuncture acoustic spectrum high frequency energy proportion and the specific frequency band attenuation rate of the presser foot vibration as key characteristic quantities, and using a double-channel prediction model, the prospective prediction of the fabric tension abnormality and down unevenness at the next moment is realized for the first time. This completely changes the dilemma of "uncertainty", and makes the control system change from "blind execution" to "predicting the future". Based on this prediction, the present application actively adjusts the down distribution by reconstructing the voltage matrix, fundamentally eliminates the inherent uncertainty of the material, and ensures that the initial state of each quilting unit before processing is uniform and ideal.

[0021] The "predictive compensation" and "multi-dimensional collaborative inhibition" of dynamic disturbance are realized: the control logic is improved from "lag response" to "predictive compensation" in the application. Through the path and pressure collaborative compensation module, the path compensation vector and the pressure adjustment value are calculated according to the predicted tension abnormality before the dynamic disturbance occurs. This is equivalent to setting up a "firewall" for the upcoming disturbance. The "firewall" further realizes the final and strong inhibition of dynamic disturbance by adaptively locking the voltage matrix to fix the material in situ at the moment of sewing. The multi-physical field collaboration of "predictive mechanical fine-tuning" and "instantaneous electric field locking" constitutes a double and three-dimensional defense system for dynamic disturbance.

[0022] The "high degree of unity" of "macro" and "micro" quality control objectives is realized: the application constructs a new control framework with micro quality as the core driving. All compensation and adjustment (path, pressure, down distribution) are derived from the prediction of micro state (tension, down density). In this way, dynamic adjustment of macro size naturally serves and unifies under the goal of ensuring micro quality. Finally, the application not only guarantees the macro size accuracy of the finished product, but also greatly improves the micro quality indicators such as down fullness, fabric flatness and stitch beauty of a single quilting unit that determine the value of high-end products, perfectly solving the contradiction between the two. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 It is a schematic diagram of the overall method flow of the application;

[0024] Figure 2 It is a logic block diagram of step S2 of the application;

[0025] Figure 3 It is a logic block diagram of step S3 of the application;

[0026] Figure 4 It is a logic block diagram of step S4 of the application;

[0027] Figure 5 It is a logic block diagram of step S5 of the application;

[0028] Figure 6 It is a schematic diagram of the execution interface of the application. DETAILED DESCRIPTION

[0029] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the specific embodiments of the application will be described in detail below with reference to the drawings of the specification.

[0030] In the following description, a lot of specific details are set forth in order to provide a thorough understanding of the present application, however, the present application can be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the concept of the present application, therefore, the present application is not limited to the specific embodiments disclosed below.

[0031] Embodiment one:

[0032] Please refer to Figures 1 to 6 , the present application provides a technical solution:

[0033] A dynamic size adjustment interface optimization method for a down jacket quilting machine, the specific steps include:

[0034] Step S1: collect and process the original signal data of the target fabric, so as to respectively extract the needle acupuncture acoustic frequency spectrum high frequency energy proportion strongly related to the fabric tension state, and the foot vibration specific frequency band attenuation rate strongly related to the down density state;

[0035] Further description: through the piezoelectric acoustic sensor and the micro accelerometer arranged at the foot of the quilting machine, the original signal data composed of the original acoustic signal and the original vibration signal are synchronously collected with the unified system clock;

[0036] Based on the preset signal processing model, the original signal data is decoupled to extract the needle acupuncture acoustic frequency spectrum high frequency energy proportion representing the local tension state of the fabric, and the foot vibration specific frequency band attenuation rate representing the down density state under the foot; In this embodiment, the needle acupuncture acoustic frequency spectrum high frequency energy proportion is denoted as E hf ; the foot vibration specific frequency band attenuation rate is denoted as R damp ;

[0037] The piezoelectric acoustic sensor is integrated at the bottom of the foot and close to the needle hole position, and the micro accelerometer is installed at the center of mass position of the foot connecting rod; so as to maximize the signal-to-noise ratio and relevance of signal acquisition. Further, the unified system clock is generated by a master control field programmable gate array (FPGA) to ensure nanosecond level synchronization accuracy.

[0038] The specific implementation content of step S1 is divided into the following flows:

[0039] Flow one, quantitative calculation of fabric tension characteristics:

[0040] Data acquisition: in each data acquisition window triggered by the needle position encoder, a length of N discrete original acoustic signal data stream D acoustic,raw (t) is synchronously acquired; wherein t is the index of time point;

[0041] Signal preprocessing: D acoustic,raw(t) Band-pass filtering to filter out low frequency noise below 500 Hz and ultra-high frequency interference above 50 kHz.

[0042] Time-frequency spectrum calculation: Apply short-time Fourier transform (STFT) to the pre-processed signal. The composition of time-frequency spectrum calculation is expressed as: divide the signal into multiple overlapping frames, multiply each frame by a window function, and then perform fast Fourier transform (FFT) on each windowed frame to obtain a time-frequency spectrum matrix P(t, f), where f represents frequency; where the row represents time, the column represents frequency, and the matrix element value is the power spectral density; the window function in this embodiment is a Hamming window;

[0043] Characteristic parameter calculation: identify the specific time point t at which the needle fully penetrates the fabric pierce , and calculate the high-frequency energy proportion E of the needle puncture acoustic spectrum at this time point hf ; The calculation logic composition is expressed as follows:

[0044] Composition I: Calculate the sum of power spectral density in the frequency range from the preset high-frequency start frequency f hf,start to the high-frequency cutoff frequency f hf,end at time point t pierce , to obtain the high-frequency energy integral value S hf .

[0045] Composition II: Calculate the sum of power spectral density in the frequency range from 0 to Nyquist frequency f nyquist at time point t pierce , to obtain the full-band energy integral value S total .

[0046] Composition III: Divide the high-frequency energy integral value S hf by the full-band energy integral value S total , to obtain the final high-frequency energy proportion E hf of the needle puncture acoustic spectrum;

[0047] In this embodiment, the high-frequency start frequency f hf,start is determined by statistical analysis of needle puncture acoustic experiments on fabrics under different tensions to determine the lower limit of the frequency that can most significantly distinguish tension changes. In this embodiment, it is determined to be 15000 Hz.

[0048] The high-frequency cutoff frequency f hf,end is used to determine the upper limit of the frequency that can most significantly distinguish tension changes. In this embodiment, it is determined to be 30000 Hz.

[0049] Flow II, quantification calculation of down density characteristics:

[0050] Data acquisition: Synchronously acquire the original vibration signal data stream, and record the original vibration signal data stream at time point t as Dvibration,raw (t);

[0051] Signal preprocessing: band-pass filter the D vibration,raw (t) to retain the frequency band most relevant to the friction and damping of the presser foot.

[0052] Band decomposition: apply a three-level wavelet packet decomposition to the preprocessed signal. The composition of this calculation is as follows: select a mother wavelet, decompose the signal layer by layer to obtain 8 terminal node coefficient sequences covering different frequency bands; the mother wavelet in this embodiment is Daubechies5;

[0053] Feature parameter calculation: identify the target frequency band coefficient sequence C target,band most relevant to the damping of down and calculate the presser foot vibration specific frequency band attenuation rate R damp . The target frequency band coefficient sequence C target,band in this embodiment is determined by the coefficient sequence corresponding to the 200Hz-500Hz frequency band through experimental calibration; the "target frequency band coefficient sequence" in this embodiment is the coefficient sequence corresponding to the 200Hz-500Hz frequency band through experimental calibration;

[0054] The calculation logic composition of the presser foot vibration specific frequency band attenuation rate R damp is as follows:

[0055] Composition I: calculate the instantaneous energy of the target frequency band coefficient sequence C target,band to obtain the energy envelope sequence E envelope (t) at time point t.

[0056] Composition II: apply the least squares method to the energy envelope sequence E envelope (t) to fit an exponential decay model; the calculation logic of the exponential decay model is: the energy envelope sequence E envelope (t) is equal to the initial energy amplitude A0 multiplied by the natural constant e raised to the power of (the negative presser foot vibration specific frequency band attenuation rate R damp times the time point t);

[0057] Composition III: directly extract the presser foot vibration specific frequency band attenuation rate R damp from the fitting result.

[0058] The initial energy amplitude A0 is a parameter in the fitting process, representing the energy size at the beginning of vibration;

[0059] The technical effect of this step is that through the above accurate calculation process, the original mixed physical signal is successfully decoupled into two quantitative indicators with clear physical meaning and mutual independence: the needle acupuncture acoustic frequency spectrum high frequency energy proportion E hf and the presser foot vibration specific frequency band attenuation rate R dampThis decoupling provides high-quality, unambiguous data input for all subsequent model-based prediction and control strategies, and is the fundamental guarantee for the high-precision control of the entire invention;

[0060] Need to explain: for the calculation method of the needle acoustic spectrum high frequency energy ratio E hf , which comes from the spectrum analysis theory in the field of signal processing. Its physical basis is that the greater the internal stress of the material under high-speed impact (such as needle puncture), the stronger the "brittleness" of the material, and the more high-frequency sound wave components generated;

[0061] Application level parameter substitution and calculation: at a certain time t pierce , the high frequency energy integral value S hf calculated and obtained is 300 units, and the full frequency energy integral value S total is 1000 units. The value of the needle acoustic spectrum high frequency energy ratio E hf is 300 divided by 1000, which is equal to 0.3.

[0062] The calculation formula of the needle acoustic spectrum high frequency energy ratio E hf is the ratio of energy, so it is a dimensionless pure number;

[0063] The value range of the needle acoustic spectrum high frequency energy ratio E hf is limited between [0, 1];

[0064] When the high frequency energy integral value S hf tends to 0, the needle acoustic spectrum high frequency energy ratio E hf tends to 0;

[0065] When all the energy is more concentrated in the high frequency band, the needle acoustic spectrum high frequency energy ratio E hf tends to 1.

[0066] When the needle acoustic spectrum high frequency energy ratio E hf output tends to 0: indicates that the ratio of the high frequency energy integral value S hf to the full frequency energy integral value S total is lower; the sound wave energy generated by the needle puncture is more concentrated in the low frequency band; the fabric is in a relaxed or low tension state, and its deformation is softer, and high frequency oscillation is less.

[0067] When the needle acoustic spectrum high frequency energy ratio E hf output tends to 1: indicates that the high frequency energy integral value S hf is more dominant in the full frequency energy integral value S total ; the sound wave energy is highly concentrated in the high frequency; the fabric is in a highly tight state;

[0068] The foot pressure vibration specific frequency band attenuation rate R damp , whose calculation method is derived from the damping theory in vibration mechanics and the parameter estimation method in signal processing; its physical basis is that the energy of a vibration system will decay exponentially when it is subjected to external damping, and the decay rate is directly proportional to the size of the damping;

[0069] The parameter substitution design at the application level is that the exponential decay model obtained by least squares fitting is: the energy is equal to 150 times e raised to the power of (-0.8 times time). The value of the foot pressure vibration specific frequency band attenuation rate R damp is 0.8. In order to map its value range to the (0, 1) range, the Sigmoid function is applied for transformation: Where k and R0 are scaling and translation parameters determined according to a large amount of experimental data; R' damp is the mapped foot pressure vibration specific frequency band attenuation rate;

[0070] The unit of the foot pressure vibration specific frequency band attenuation rate R damp is the reciprocal of time, representing the proportion of energy decay per second. After the Sigmoid function conversion, the foot pressure vibration specific frequency band attenuation rate R damp is converted to the transformed R' damp , whose value range is limited between (0, 1); the setting of parameters k and R0 ensures that the foot pressure vibration specific frequency band attenuation rate R damp will not output as 0 or 1, but infinitely approaches;

[0071] When the mapped foot pressure vibration specific frequency band attenuation rate R' damp output approaches 0: it indicates that the original foot pressure vibration specific frequency band attenuation rate R damp value is smaller; that is, the fitted exponential decay is slower; the energy of the foot pressure vibration is more difficult to dissipate, and thus the damping provided by the medium down below is smaller; this corresponds to the state that the down is more sparse, the loftiness is insufficient, or the hollow condition is more serious; the area more needs to increase the foot pressure or perform electric field reconstruction.

[0072] When the mapped foot pressure vibration specific frequency band attenuation rate R' damp output approaches 1: it indicates that the original foot pressure vibration specific frequency band attenuation rate R damp value is larger. That is, the exponential decay is more rapid. The energy of the foot pressure vibration is more easily absorbed, and the medium down below provides greater damping. This corresponds to the ideal state that the down is more dense and filled more uniformly and fully; it represents that the down state of the current area is better, and the degree of intervention needed is less.

[0073] Step S2: inputting the time sequence of the needle-punched acoustic spectrum high-frequency energy proportion and the foot vibration specific frequency band attenuation rate into a pre-constructed double-channel prediction model respectively, the double-channel prediction model being used to predict a predicted tension abnormality degree of a next quilting path node and a predicted down unevenness of a next unit to be quilted;

[0074] Further description: the double-channel prediction model includes a first prediction channel and a second prediction channel, the first prediction channel being used to predict the predicted tension abnormality degree of the next quilting path node by giving higher weights to key time points in a historical sequence corresponding to the needle-punched acoustic spectrum high-frequency energy proportion, denoted as δ T,pred ;

[0075] The second prediction channel is used to predict the predicted down unevenness of the next unit to be quilted by giving higher weights to key time points in a historical sequence corresponding to the mapped foot vibration specific frequency band attenuation rate, denoted as δ D,pred ;

[0076] The double-channel prediction model contains an attention mechanism;

[0077] Further, the double-channel prediction model is composed of two parallel long short-term memory networks (LSTM) with the same structure, and a self-attention (Self-Attention) layer is connected to the output end of each LSTM network, which is used to calculate the weights;

[0078] The specific implementation process of step S2 is as follows:

[0079] Process one, for the calculation of the predicted tension abnormality degree δ T,pred :

[0080] Input sequence construction: the needle-punched acoustic spectrum high-frequency energy proportion E hf values collected and calculated in the last L time steps are used to form an input sequence S E =[E hf,t-L+1 ,...,E hf,t ], with a length of L time steps, wherein t is a time point;

[0081] Time sequence feature coding: the input sequence S E =[E hf,t-L+1 ,...,E hf,t ] is sent to a long short-term memory network (LSTM) encoder. The LSTM network processes the sequence one by one to generate a tension hidden state matrix H E containing all the hidden states of the time steps. The historical hidden state h E,i of each column of the matrix encodes the historical information up to time step i;

[0082] Attention weight calculation: Calculate the tension hidden state matrix H E The data is fed into a self-attention layer, where the importance weight of each historical time point for predicting future tension is calculated. The calculation logic is expressed as follows:

[0083] Component 1: For the tension hidden state matrix H E The historical hidden state h at each time step i in the data. E,i Calculate its relationship with the current hidden state h. E,t The dot product similarity is used to obtain the original attention score for time step i. i .

[0084] Component 2: Raw attention scores for all historical time points i Normalization is performed using the Softmax function to obtain the final tension attention weight sequence α. T The tension attention weight α at each time step i in this tension attention weight sequence. T,i The values ​​are all between 0 and 1, and the tension attention weight sequence α T Each tension attention weight is a weight, and the sum of all elements is 1.

[0085] Context vector generation: for tension hidden state matrix H E The historical hidden state h at each time step i E,i According to the tension attention weight sequence α T By performing a weighted summation, we obtain a single tension context vector C that contains all the key historical information. T ;

[0086] Predicted output: The tension context vector C T By using a fully connected layer and applying an activation function Tanh, a scalar value is ultimately generated, which is the predicted tension anomaly δ. T,pred .

[0087] Step 2, Explanation of the calculation for predicting down unevenness:

[0088] Input sequence construction: Map the presser foot vibration attenuation rate R′ of the most recent L time steps to a specific frequency band. damp The values ​​constitute the input sequence S. R =[R′ damp,t-L+1 ,...,R′ damp,t ].

[0089] Temporal feature encoding: S R =[R′ damp,t-L+1 ,...,R′ damp,t The data is fed into another independent LSTM encoder to obtain the down feather hidden state matrix H.R ;

[0090] Attention weight calculation: through the self-attention layer, the down feather attention weight sequence a is calculated D .

[0091] Context vector generation: weighted summation to obtain the down feather context vector C D .

[0092] Predicted value output: the down feather context vector C D is input into the fully connected layer to generate the predicted down feather unevenness δ D,pred ;

[0093] Tension hidden state matrix H E and down feather hidden state matrix H R : is an internal calculation quantity of the model, which represents the encoding of historical information; the original attention score score i is an internal calculation quantity of the model;

[0094] Tension attention weight sequence a T and down feather attention weight sequence a D : an internal calculation quantity of the model, the value of which is learned autonomously by the model during the training process, reflecting the importance of different historical moments;

[0095] Tension context vector C T and down feather context vector C D : an internal calculation quantity of the model, which is a weighted condensation of historical information;

[0096] The technical effect of the embodiment is that by introducing the attention mechanism, the model has the ability to dynamically focus on key information, thereby realizing more accurate and robust prediction than traditional time series models. It no longer treats all historical information equally, but simulates the decision-making process of a human expert: when predicting the tension of a corner, it will pay more attention to the previous several consecutive rising tension values; when predicting the down feather distribution of a region, it will pay more attention to the previously appeared abnormal sparse points.

[0097] For the quantification of the predicted tension anomaly δ T,pred , the following is deduced:

[0098] It combines the long short-term memory network (LSTM) in deep learning and the attention mechanism (Attention-Mechanism) in natural language processing. LSTM is used to capture temporal dependencies, and the attention mechanism is used to optimize information filtering.

[0099] Parameter substitution and calculation in the application layer: the embodiment sets the current input sequence S E = [E hf,t-L+1..., E hf,t ] = [0.2, 0.3, 0.6, 0.7], which represents a rapid and continuous rise in tension. After LSTM encoding, the model calculates the tension attention weight sequence a T = [0.1, 0.2, 0.3, 0.4]. This indicates that the model considers the most recent and highest value of E hf = 0.7 to be the most important for future prediction; subsequently, a tension context vector C T is generated by weighted summation, and finally the predicted tension anomaly degree d T,pred is output, with a value of 0.85.

[0100] The predicted tension anomaly degree d T,pred is the difference between the model's prediction of the future value of the needle-punched acoustic spectrum high-frequency energy proportion E hf and the ideal value, after passing through the Tanh activation function. The predicted tension anomaly degree d T,pred is limited to the range (-1, 1); it is a dimensionless pure number. The output value in this embodiment does not take 0 and 1, but approaches infinity.

[0101] When the predicted tension anomaly degree d T,pred output approaches -1: it indicates that the model predicts that the future fabric tension is lower than the ideal value by a larger margin, and the fabric is more relaxed; this is because the needle-punched acoustic spectrum high-frequency energy proportion E E in the input sequence S hf,t-L+1 = [E hf,t ,..., E hf ] is consistently low or continuously decreasing, and the tension attention weight sequence a T is concentrated on these low values; this further warns of the accumulation and wrinkles caused by fabric relaxation, providing a basis for subsequent control;

[0102] When the predicted tension anomaly degree d T,pred output approaches 1: it indicates that the model predicts that the future fabric tension is higher than the ideal value by a larger margin, and the tightness state is greater; this is because the needle-punched acoustic spectrum high-frequency energy proportion E E in the input sequence S hf,t-L+1 = [E hf,t ,..., E hf ] is consistently high or rapidly rising, and the tension attention weight sequence a T also focuses on these high-risk moments accordingly; this further warns of the risk of excessive tension that is about to occur, which is a key signal to trigger subsequent path compensation control;

[0103] The quantitative deduction for the predicted down unevenness d D,pred is as follows:

[0104] The parameter substitution and calculation at the application level are as follows: the current input sequence S R = [R' damp,t-L+1 ,...,R' damp,t ] is [0.8, 0.7, 0.2, 0.8], representing that the down state suddenly becomes sparse (0.2) at a point and then returns to normal. The model obtains the down attention weight sequence α D = [0.1, 0.1, 0.7, 0.1] through the attention mechanism; the model will focus on the abnormal low value point, indicating an uncompletely filled hole, even if it returns to normal now, there is a possibility of re-occurring problems. The final model output predicts the down unevenness δ D,pred = -0.6, indicating a high risk of predicting that the down in the next area is sparse.

[0105] The predicted down unevenness δ D,pred is the difference between the model's prediction of the specific frequency band attenuation rate value of the mapped foot pressure vibration and the ideal value, and after the Tanh activation function, the value range is also limited to (-1, 1), which is a dimensionless pure number;

[0106] When the predicted down unevenness δ D,pred output tends to -1: it indicates that the model predicts that the future down state is more sparse; specifically, one or more significant low values appear in the input sequence S R = [R' damp,t-L+1 ,...,R' damp,t ], and the down attention weight sequence α D is highly concentrated on the low value point; it is the core basis for triggering subsequent electric field reconstruction or pressure compensation;

[0107] When the predicted down unevenness δ D,pred output tends to 1: it indicates that the model strongly predicts that the future down dense state is larger. Specifically, the values in the input sequence S R = [R' damp,t-L+1 ,...,R' damp,t ] are continuously high, and the down attention weight sequence α D is also concentrated on these high value points. It further warns of the possible sewing difficulties or stiffness problems caused by down accumulation, providing decision support for subsequent control strategies such as electric field dispersion or reduced pressure.

[0108] Step S3: receiving the predicted tension abnormality and sending it to the path and pressure collaborative compensation module to calculate the path compensation vector through a nonlinear gain function;

[0109] and based on the predicted tension abnormality and the predicted down unevenness, calculating the foot pressure adjustment value through a collaborative weighting model;

[0110] Further explanation: the nonlinear gain function is used to compensate only when the predicted tension abnormality exceeds the preset tension compensation trigger threshold; and the nonlinear gain function is an exponential function;

[0111] The path and pressure collaborative compensation module is provided with a collaborative weighting model;

[0112] The collaborative weighting model includes calculating a pressure collaborative weight factor W sync , introducing a basic pressure adjustment demand ΔP base ; multiplying the basic pressure adjustment demand ΔP base and the pressure collaborative weight factor W sync ; and then multiplying the multiplication result by a preset pressure compensation basic gain coefficient γ to obtain a final pressure foot pressure adjustment value ΔP.

[0113] 3.1) The specific implementation of step S3 is as follows:

[0114] Flow 1, calculation of the path compensation vector:

[0115] Input acquisition: acquire the predicted tension abnormality δ T,pred of the current time from step S2;

[0116] Compensation amplitude calculation: calculate the path compensation basic amplitude M path ; the calculation logic composition is expressed as follows:

[0117] Composition 1: compare the predicted tension abnormality δ T,pred and the preset tension compensation trigger threshold θ T , take the larger value between the difference and 0, and obtain the effective tension abnormality δ' T,pred . Only when the predicted tension abnormality δ T,pred exceeds the tension compensation trigger threshold θ T , compensation is performed;

[0118] Composition 2: take the effective tension abnormality δ' T,pred as the base, the preset path compensation nonlinear gain index k p as the index, and perform power operation to obtain the path compensation nonlinear gain factor G path ;

[0119] Composition 3: multiply the path compensation nonlinear gain factor G path by a preset path compensation basic gain coefficient β1 to obtain the final path compensation basic amplitude M path .

[0120] Compensation vector generation: multiply the path compensation basic amplitude M pathThe final path compensation vector V is obtained by multiplying the unit vector in the opposite direction of the current sewing path path ; and adding the path compensation vector V path to the original coordinates of the next target path node;

[0121] 3.2) Process two, calculation of the presser foot pressure adjustment value:

[0122] Input acquisition: acquire the predicted tension abnormality degree δ T,pred and the predicted down unevenness δ D,pred from step S2 at the current time. Let the pressure coordination weight factor be W sync ; let the presser foot pressure adjustment value be ΔP; and let the basic pressure adjustment requirement be ΔP base .

[0123] Coordination weight calculation: calculate the pressure coordination weight factor W sync ; this parameter is used to enhance or weaken the pressure adjustment based on the down state when the fabric tension is abnormal; and its calculation logic composition expression is as follows:

[0124] Take the absolute value of the predicted tension abnormality degree δ T,pred , multiply it by a preset coordination influence coefficient λ, take the negative value of the result, and calculate its natural exponent to obtain the pressure coordination weight factor W sync .

[0125] Calculate the final presser foot pressure adjustment value ΔP; its calculation logic composition expression is as follows:

[0126] Composition one: subtract the predicted down unevenness δ D,pred from the preset down sparsity compensation trigger threshold θ D to obtain the basic pressure adjustment requirement ΔP base .

[0127] Composition two: multiply the basic pressure adjustment requirement ΔP base by the pressure coordination weight factor W sync calculated in the previous step.

[0128] Composition three: multiply the weighted result by a preset pressure compensation basic gain coefficient γ to obtain the final presser foot pressure adjustment value ΔP.

[0129] The tension compensation trigger threshold θ T is determined by experiment calibration to determine the value of the predicted tension abnormality degree δ T,pred , and it is considered that the tension is abnormal and needs to be intervened when the value exceeds this value. The tension compensation trigger threshold θ T in this embodiment is set to 0.6.

[0130] The path compensation nonlinear gain index k pis a value greater than 1, used to control the degree of nonlinearity of compensation. It is determined by analyzing the stress-strain curves of different fabrics. The greater the value, the faster the compensation increases in the high abnormality interval. The path compensation nonlinearity gain index k of the embodiment p is set to 1.5.

[0131] The path compensation base gain coefficient β1 is used to scale the dimensionless compensation factor to the actual physical displacement, which is calibrated by experiment.

[0132] The synergistic effect coefficient λ is used to adjust the influence of the tension state on the pressure compensation; it is determined by experimental analysis of the synergistic effect under different working conditions. The embodiment is set to 2.0.

[0133] The down sparse compensation trigger threshold θ D is calibrated by experiment to determine a predicted down unevenness δ D,pred value, below which it is considered that the down is sparse and needs to be intervened. The down sparse compensation trigger threshold θ D of the embodiment is set to -0.5.

[0134] The pressure compensation base gain coefficient γ is used to scale the dimensionless adjustment demand to the actual pressure unit, which is calibrated by experiment;

[0135] 3.3) The calculation and derivation of the path compensation vector V path , whose calculation logic combines threshold control and nonlinear system theory in control theory. The design of nonlinear gain is inspired by the idea of material nonlinear deformation in physics;

[0136] Application of parameters: In this embodiment, the tension compensation trigger threshold θ T = 0.6, the path compensation nonlinearity gain index k p = 1.5, and the path compensation base gain coefficient β1 = 0.1.

[0137] Case one: the predicted tension abnormality δ T,pred = 0.7. The effective tension abnormality δ' T,pred = 0.7-0.6 = 0.1. The path compensation nonlinearity gain factor G path = 0.1 1.5 ≈0.0316. The path compensation base amplitude M path = 0.0316×0.1 = 0.00316 mm, and the compensation amount is small.

[0138] Case two: the predicted tension abnormality δ T,pred = 0.9; the effective tension abnormality δ' T,pred = 0.9-0.6 = 0.3. The path compensation nonlinearity gain factor G path = 0.31.5 ≈0.1643. Path compensation base amplitude M path = 0.1643 x 0.1 = 0.01643 mm; the abnormality only increased by 0.2, but the compensation increased more than 5 times, which embodies the nonlinear effect;

[0139] Path compensation vector V path The final dimension is consistent with the dimension of length and coordinates; its value range is [0, +∞), which is theoretically unlimited, but limited by the value range of the predicted tension abnormality δ T,pred (-1, 1) and the setting of the gain coefficient, the actual compensation will be controlled within a reasonable physical range;

[0140] The formula ensures that compensation only occurs when the predicted tension abnormality δ T,pred exceeds the tension compensation trigger threshold θ T , avoiding unnecessary fine-tuning.

[0141] Since the path compensation nonlinear gain index k p > 1, the path compensation base amplitude M path increases with the increase of the effective tension abnormality δ' T,pred ; this embodiment makes moderate compensation for slight tension abnormalities, and makes large compensation for serious tension abnormalities to effectively prevent the formation of wrinkles;

[0142] Calculation and derivation of the presser pressure adjustment value ΔP: the calculation core is to introduce a pressure synergy weight factor W sync derived from the exponential decay function;

[0143] Parameter substitution and calculation at the application level: this embodiment sets the synergy influence coefficient λ = 2.0, the down sparse compensation trigger threshold θ D = -0.5, and the pressure compensation base gain coefficient γ = 0.2.

[0144] In the case of normal tension: the predicted tension abnormality δ T,pred = 0.1, the predicted down unevenness δ D,pred = -0.8, which represents down sparseness, the pressure synergy weight factor W sync = e -2.0×|0.1| ≈0.8187. The base pressure adjustment requirement ΔP base = -0.5-(-0.8) = 0.3. The presser pressure adjustment value ΔP = 0.3 x 0.8187 x 0.2 ≈ +0.049 Newton; the pressure synergy weight factor W sync tends to 1, and the pressure adjustment is mainly determined by the down state;

[0145] In the case of tension abnormality: the predicted tension abnormality δ T,pred= 0.9, predicted down unevenness δ D,pred = -0.8. Pressure synergy weight factor W sync = e -2.0×|0.9| ≈ 0.1653. Base pressure adjustment demand ΔP base = 0.3. Pressure foot pressure adjustment value ΔP = 0.3 x 0.1653 x 0.2 ≈ +0.010 Newton, representing a small increase in pressure.

[0146] The final dimension of the pressure foot pressure adjustment value ΔP is Newton, consistent with the unit of the pressure adjustment system;

[0147] The value range of the pressure synergy weight factor W sync is limited between (0, 1]. When the predicted tension abnormality degree δ T,pred tends to 0, the pressure synergy weight factor W sync tends to 1. When the absolute value of the predicted tension abnormality degree δ T,pred tends to 1, the pressure synergy weight factor W sync tends to 0.

[0148] When the target fabric tension is more normal, the pressure synergy weight factor W sync will tend to 0; the pressure synergy weight factor W sync tends to 1, the influence degree of the adjustment range of the pressure adjustment determined by the predicted down unevenness δ D,pred representing the down state is greater, and the system is more focused on solving the down problem;

[0149] When the target fabric tension becomes more abnormal, the absolute value of the predicted tension abnormality degree δ T,pred is greater, the pressure synergy weight factor W sync rapidly decreases in range, and the adjustment range of the pressure foot pressure adjustment value is smaller.

[0150] The technical effect is that in the case of large target fabric tension, further increasing the pressure foot pressure will exacerbate the deformation of the fabric, resulting in more serious wrinkles. Therefore, the embodiment will "sacrifice" a part of the compaction effect on the down to prioritize the flatness of the fabric. This dynamic adjustment priority strategy demonstrates the rationality and advancement of the design of the synergy weighting model.

[0151] In another embodiment, it needs to be noted that:

[0152] This embodiment aims to verify the advancement and effectiveness of the method described in the present invention in dealing with complex and coupled sewing defects; the test equipment is a standard JUKI-DU-1181N industrial flat bed sewing machine, which is modified by installing a three-axis MEMS accelerometer (ADXL355) at the centroid position of the presser foot connecting rod and integrating a wideband MEMS piezoelectric acoustic sensor (SPU0410HR5H-QB) beside the needle hole at the bottom of the presser foot; the head controller is replaced by a master control board equipped with Xilinx-Artix-7-FPGA, which is used to realize nanosecond-level precision synchronous signal acquisition and preliminary processing;

[0153] The test material is a representative challenge combination: the fabric is a 400T high-density nylon fabric, which has low ductility and is prone to tension accumulation and wrinkles during sewing; the filler is 800-loft water-repellent white goose down, which has high loft but is prone to local density unevenness when laid in large areas. The preset quilting path is a complex rectangular grid containing continuous straight line segments and a 90-degree sharp turn.

[0154] To compare the effects, two test groups are set up:

[0155] Group A1 (invention embodiment): the technical solution described in the present invention is adopted, i.e., through step S1, synchronous sensing and feature decoupling are performed to obtain the needle-punching acoustic spectrum high-frequency energy proportion E hf and the normalized presser foot vibration specific frequency band attenuation rate R′ damp ; through the double-channel LSTM model with attention mechanism in step S2, prediction is performed; finally, through the compensation module based on the nonlinear gain and collaborative weighting model in step S3, control instructions are generated.

[0156] Group B1: a simplified scheme representing the current mainstream intelligent control idea is adopted. This scheme also uses the sensor hardware of step S1, but the data processing and control logic are different: 1) the prediction model uses a standard LSTM network without attention mechanism; 2) the control module is two independent PID controllers, one adjusts the path based on the predicted tension abnormality, and the other adjusts the presser foot pressure based on the predicted down unevenness, and there is no information exchange or collaboration between the two controllers.

[0157] The test process is as follows:

[0158] The machine is started and begins to quilt along the preset path. The system continuously executes step S1 at a sampling rate of 51.2kS / s to generate the time series of the needle-punching acoustic spectrum high-frequency energy proportion E hf and the normalized presser foot vibration specific frequency band attenuation rate R′ damp in real time.

[0159] Scenario 1: Attention mechanism validation of the prediction model. When the sewing path approaches a 90-degree corner from a straight segment, the tension starts to accumulate gradually under the pulling of the feeding mechanism. The system collects the needle-punching acoustic spectrum in real time, and the high-frequency energy proportion E hf The sequence shows a clear upward trend. At this time, the attention model of the A1 group, when executing step S2, its internal tension attention weight sequence α T automatically assigns higher weights to the last few values of the sequence, which continuously increase the high-frequency energy proportion E hf of the needle-punching acoustic spectrum. This enables the model to “focus” on this dangerous accumulation trend, thereby outputting an accurate and large-scale predicted tension abnormality degree δ T,pred . In contrast, the standard LSTM model of the B1 group can also perceive the upward trend, but due to the lack of attention mechanism, its processing of historical information is uniform, resulting in a delayed and low-amplitude response of the predicted abnormality degree.

[0160] Scenario 2: Decision validation of the synergistic compensation strategy. The test designs a key test point: a small range of down feather sparse area is artificially created on a small section of the path before the 90-degree sharp corner; when the presser foot passes through this area, both systems of the two test groups detect a sudden drop in the normalized presser foot vibration specific frequency band attenuation rate R′ damp . Then, the machine is about to enter the corner with the highest tension risk. At this time, the systems of the A1 and B1 groups face a dilemma: they need to deal with the down feather sparsity, which theoretically requires increased pressure, and they also need to deal with the upcoming high tension, which theoretically requires avoiding additional pressure.

[0161] The A1 group system executes step S3 and receives two strong prediction signals from step S2: a high positive predicted tension abnormality degree δ T,pred and a low negative predicted down feather unevenness δ D,pred . At this time, the core role of the synergistic weighting model is revealed. Since the absolute value of the predicted tension abnormality degree δ T,pred is large, the calculated pressure synergistic weight factor W sync becomes small. Therefore, although the basic pressure adjustment demand ΔP base indicates that pressure needs to be increased, after being multiplied by this small weight factor, the final generated presser foot pressure adjustment value ΔP is actively and intelligently suppressed to a very small level. At the same time, the path compensation vector V T,pred calculated based on the predicted tension abnormality degree δ path obtains a significant compensation amount due to the nonlinear gain effect, and preferentially performs the path relaxation operation.

[0162] The B1 group system exposes its fundamental defects. Its two independent PID controllers do not affect each other. The tension controller outputs a path compensation instruction according to its lower predicted value. At the same time, the down controller receives the signal of down sparseness and outputs an instruction to greatly increase the presser foot pressure completely independently and without hesitation.

[0163] The following table records the comparison data of the internal parameters of the systems and the quality of the final products of the two test groups at the critical time point of "a moment before the corner" in the sewing path.

[0164] Table 1: Demonstration of the synergistic weighting model in step S3:

[0165]

[0166]

[0167] After sewing, the finished product is checked; the product of the A1 group is smooth at the corner without any visible wrinkles. Touching the area with your hand, you can only feel that the loft of the down in this area is slightly lower than that in other areas, but it is completely within the acceptable quality range; the product of the B1 group has obvious and permanent "dead folds" at the corner. Because it applies extra presser foot pressure at the moment when the fabric tension is the greatest, it causes the fabric to be stretched and fixed too much, causing irreversible damage. This clearly proves the superiority of the synergistic strategy of the present application: it "sacrifices" the secondary indicator, i.e. the extreme compaction of the down, to preserve the fabric flatness, achieving the optimization of the overall quality.

[0168] Step S4: Based on the predicted down unevenness, a secondary analysis is performed to generate a reconstruction voltage matrix for reconstructing the distribution of down in the next unit to be quilted;

[0169] Further explanation: the generation of dynamic electric field driving instructions for reconstructing the distribution of down in the next unit to be quilted further includes: based on the predicted down unevenness δ D,pred , a preset electric potential gradient direct mapping model is used for calculation to generate a reconstruction voltage matrix;

[0170] The reconstruction voltage matrix is applied to the microelectrode array to generate a non-uniform electrostatic field, which pulls the down from the predicted dense areas to the predicted sparse areas by generating a dielectrophoretic force opposite to the direction of the down density gradient;

[0171] Further, the microelectrode array is arranged below the workbench plate of the quilting machine, and the electric potential gradient direct mapping model is a mathematical model that directly converts the predicted down unevenness into a target electric potential gradient and then solves each electrode voltage;

[0172] The microelectrode array integrates corresponding sensing electrodes on the pressure feet; by applying different voltages to the array, a precise three-dimensional electric field can be formed in each cavity. By measuring the capacitance between the electrodes, the real-time value of the cavity dielectric constant can be obtained.

[0173] The specific implementation process of the potential gradient direct mapping model is as follows:

[0174] The core of step S4 is to build an efficient computing pathway to directly and quickly convert the macroscopic down unevenness prediction value into a microscopic voltage distribution command applied to the microelectrode array, so as to achieve active and accurate reconstruction of the down distribution.

[0175] Step 1: Target density gradient generation:

[0176] Input Acquisition: The electric field control module receives the predicted down unevenness δ from step S2. D,pred .

[0177] Gradient transformation: The electric field control module is based on the predicted down unevenness δ D,pred Generate a target down density correction gradient ▽ρ target The algorithm's logical components are expressed as follows:

[0178] First, the scalar value is used to predict the down unevenness δ. D,pred Multiply by a negative gradient transformation coefficient k grad This yields a scalar target density corrected gradient magnitude. The "negative sign" ensures that when the down is sparse, i.e., when predicting down unevenness δ, it is accurate. D,pred The value is negative, and the goal is to generate a positive density gradient.

[0179] Then, the target density correction gradient magnitude is... Multiplying this by the unit vector in the current sewing direction yields the target down density correction gradient. The unevenness in this step mainly accumulates along the sewing direction;

[0180] Step 2, Target Potential Gradient Calculation: The electric field control module uses the principle of dielectric electrophoresis to correct the target down density gradient. Convert to the square gradient of the target potential Its algorithmic logic is expressed as follows:

[0181] Dielectric electrophoresis force is proportional to the square of the potential gradient and in the opposite direction. To generate a mass flow in the same direction as the density gradient, a force in the opposite direction needs to be applied. Therefore, the target down density correction gradient is used. Multiplied by a negative electrodynamic coupling coefficient α ep The square gradient of the target potential is obtained directly.

[0182] Flow three, reconstruction voltage matrix analytical solution:

[0183] Inverse gradient operation: the target electric field control module on the square gradient of the target potential Carry out inverse operation, solve the target voltage of each electrode. Its algorithm logic composition expression as follows:

[0184] Composition one: on the square gradient of the target potential In the next to be quilted unit area within the spatial integral, and superimposed a basic bias voltage V bias The square, get the target potential square value U 2 (x,y) of each position in the region.

[0185] Composition two: calculate the square root of the target potential square value U 2 (x,y) to get the target potential distribution U(x,y) of each position in the region.

[0186] Composition three: extract the value of the target potential distribution U(x,y) corresponding to each electrode center position in the micro electrode array, constitute the final reconstruction voltage matrix, and recorded as V recon .

[0187] Gradient conversion coefficient k grad : for the dimensionless prediction of down unevenness δ D,pred Value, converted to the density gradient (unit: kg / m 4 ) with physical dimension. Through the experimental calibration of the physical properties of down materials to obtain.

[0188] Electrodynamic coupling coefficient α ep : the polarization rate, dielectric constant, electric field frequency and other factors of down particles are integrated, which represents the size of the dielectrophoretic force generated by unit electric potential square gradient. It is determined by combining theoretical calculation and experimental measurement.

[0189] Basic bias voltage V bias : a direct current or alternating current basic voltage applied to all electrodes, used to ensure that all down particles are polarized, so as to respond to the change of electric field gradient. Its value is set according to the dielectric properties of down materials, and the initial setting of this embodiment is 50 volts.

[0190] The technical effect of step S4 is to provide a kind of unprecedented, active, based on physical model direct calculation material distribution preprocessing method.It is not passive to adapt to material defects, but before the influence caused by defect, it is "smoothed" by non-contact electrostatic force field. The direct mapping model of the method ensures that the generation speed of control command is much faster than the physical response speed of material, realizes real-time, predictive reconstruction.

[0191] For target down density correction gradient The generation derivation: The calculation is derived from fluid mechanics and material transport theory, which converts macroscopic unevenness indicators into local gradient fields that drive material flow.

[0192] Parameter substitution and calculation at the application level: This embodiment sets the gradient conversion coefficient k grad = 100 kg / m 4 , and the current sewing direction is the positive direction of the X axis, with a unit vector of (1, 0).

[0193] If the predicted down unevenness δ D,pred = -0.6 (significantly sparse), the target density correction gradient amplitude The target down density correction gradient is (60, 0). This means that the system needs to establish a positive density gradient in the X direction, i.e., to move the down from the back (X negative direction) to the front (X positive direction) to fill the sparse area.

[0194] Physical dimension and value range: The dimension of the target down density correction gradient is density gradient (kg / m 4 ). Its value range is determined by the value range of the predicted down unevenness δ D,pred (-1, 1) and the gradient conversion coefficient k grad .

[0195] For the calculation derivation of the reconstructed voltage matrix V recon : The calculation is derived from the Dielectrophoresis (DEP) theory in electrokinetics. The basic idea is that a non-uniform electric field will produce a force on a polarizable neutral particle (such as down fibers), with the force direction pointing to areas with stronger or weaker electric field intensity, depending on the relative polarizability of the particle and the medium. This principle is used by the invention to drive the movement of down.

[0196] Parameter substitution and calculation: The target down density correction gradient is (60, 0). Set the electrokinetic coupling coefficient α ep = 0.5; unit (V 2 / m 2 ) / (kg / m 4 ), and the basic bias voltage V bias = 50 V.

[0197] The target electric potential square gradient unit V 2 / m 2 ; this means that the square value of the electric potential needs to decrease along the negative direction of the X axis.

[0198] Target electric potential square gradient Integrating the above equation, we get the target electric potential square value U 2 (x,y) = -30x + C1. After superimposing the bias, we have where C1 represents an overall bias or baseline value of the entire electric potential field. It is a value that is freely set by the system designer according to needs. (x,y) represents the horizontal and vertical coordinates in the spatial position;

[0199] Target electric potential distribution

[0200] The micro electrode array in this embodiment has two electrodes in the X direction, located at x1 = 0 meters and x2 = 0.01 meters. Then the reconstructed voltage matrix V recon corresponding to the two voltages are:

[0201] By applying these two different voltages, the required electric field gradient is generated between the electrodes, thereby driving the down feather to move.

[0202] The reconstructed voltage matrix V recon has the dimension of voltage and is consistent with the unit of the electrode driving system. Its value range is jointly determined by the basic bias voltage and the calculated gradient.

[0203] When the predicted down feather unevenness δ D,pred output approaches 0: the target down feather density correction gradient approaches 0, and the target electric potential square gradient also approaches 0. Finally, the voltage of all electrodes in the reconstructed voltage matrix V recon approaches the basic bias voltage V bias . This means technically that when the down feather distribution is uniform, the system only applies a uniform bias electric field and does not generate driving force, maintaining the status quo.

[0204] When the predicted down feather unevenness δ D,pred output deviates from 0 more: the absolute value of the target down feather density correction gradient is larger, resulting in the absolute value of the target electric potential square gradient is also larger. This makes the reconstructed voltage matrix V recon ​​The greater the voltage difference between adjacent electrodes. The greater the voltage difference produces a steeper electric field gradient, thus generating a stronger dielectrophoresis force to drive the down from the enrichment area to the sparse area at a faster speed; It is proved that the direct mapping model can generate appropriate physical driving force in proportion to the predicted unevenness severity, and realize the accurate and active reconstruction of down distribution.

[0205] Step S5: When the needle is about to enter the next to-be-quilted unit that has completed reconstruction on the quilting machine, switch the new reconstruction voltage matrix, and then apply an adaptive locking voltage matrix to the microelectrode array corresponding to the next to-be-quilted unit, to fix the target fabric in place using electrostatic adsorption force, while placing the path and pressure compensation module in an active state.

[0206] Further, the system controller calculates and generates an adaptive locking voltage matrix V damp based on the real-time acquisition of the pressure foot vibration specific frequency band attenuation rate R lock,adp in the to-be-quilted unit, and applies the adaptive locking voltage matrix V lock,adp to the microelectrode array to fix the target fabric in place.

[0207] The system controller determines the needle position through the needle position encoder, and when the distance between the needle and the boundary of the to-be-quilted unit is less than a preset locking trigger distance, the calculation and activation operations are performed.

[0208] Step S5 is implemented as follows:

[0209] The core of this step S5 is to realize seamless switching from "dynamic reconstruction" to "adaptive locking" through an accurate timing control logic, and to activate the final mechanical compensation path at the moment of locking, forming a double-insured, high-robustness quality control closed loop;

[0210] Process one, locking state triggering: the system controller continuously monitors the readings of the needle position encoder and calculates the distance to the entrance boundary of the next to-be-quilted unit;

[0211] When the distance is less than a preset locking trigger distance, the controller immediately triggers the locking process and stops executing the reconstruction instructions in step S4; the locking trigger distance in this embodiment is initially set to 2 mm;

[0212] Process two, calculation of adaptive locking voltage matrix:

[0213] Input acquisition: at the moment of triggering the lock, the system controller acquires the instantaneous value of the pressure foot vibration specific frequency band attenuation rate R damp under the current pressure foot. This value reflects the final density of down in the to-be-quilted unit after reconstruction;

[0214] Process three, calculate the locking voltage: the system controller calculates the locking voltage based on the foot vibration specific band attenuation rate R damp , and calculates an adaptive locking voltage value V lock,scalar that is uniformly applied to all electrodes. Its calculation logic is expressed as follows:

[0215] Composition one: set a minimum locking voltage V min , which is the basic voltage to ensure that the thinnest fabric can be effectively fixed.

[0216] Composition two: multiply the real-time collected foot vibration specific band attenuation rate R damp by a preset locking voltage gain coefficient k lock , to get a locking voltage increment ΔV lock .

[0217] Composition three: add the minimum locking voltage V min and the locking voltage increment ΔV lock to get the final adaptive locking voltage value V lock,scalar .

[0218] The system controller generates a matrix with all elements equal to the adaptive locking voltage value V lock,scalar , as the final adaptive locking voltage matrix V lock,adp , and outputs it to the micro-electrode array;

[0219] Process four, path and pressure collaborative compensation module activation: while applying the adaptive locking voltage matrix V lock,scalar to the micro-electrode array, the system controller sets the running permission of the "path and pressure collaborative compensation module" described in step S3 to the highest priority, so that it enters the active state, and is ready to make the final real-time mechanical fine-tuning to any residual disturbance that occurs during the sewing process.

[0220] Locking trigger distance: a safety distance set according to the needle movement speed and system response time, to ensure that the locking is completed before the needle arrives. Through experimental calibration, the value of this embodiment is 2 millimeters.

[0221] Minimum locking voltage V min : determined through electrostatic adsorption experiments on a series of fabrics with different gram weights and different materials, it is the lowest voltage value that can produce effective adsorption force. The initial setting value of this embodiment is 100 volts.

[0222] Locking voltage gain coefficient k lock : used to convert the change of foot vibration specific band attenuation rate R damp into the change of voltage. Through experimental calibration, the final locking voltage is within a safe and effective range for the thickest filler.

[0223] The technical effect of step S5 is that adaptive locking ensures that both light summer quilts and heavy winter quilts can be fixed with adaptive force, avoiding the problem of excessive electrostatic adsorption force. The synchronization of "locking" and "mechanical compensation activation" forms a perfect double insurance system of "active prevention + passive response".

[0224] The calculation of the adaptive locking voltage matrix V lock,adp : The calculation model is a closed-loop control strategy based on real-time feedback, which is derived from gain scheduling in control engineering, that is, adjusting control parameters according to the current state of the system.

[0225] Parameter substitution and calculation at the application level: This embodiment sets the minimum locking voltage V min = 100V, the locking voltage gain coefficient k lock = 5, unit V / s -1 .

[0226] Case 1: When the thin filler is locked, the measured pressure foot vibration specific band attenuation rate R damp = 10s -1 . The locking voltage increment ΔV lock = 10 × 5 = 50V. The adaptive locking voltage value V lock,adp = 100 + 50 = 150V. The system applies a voltage of 150V for locking.

[0227] Case 2: When the thick filler is locked, the measured pressure foot vibration specific band attenuation rate R damp = 40s -1 . The locking voltage increment ΔV lock = 40 × 5 = 200V. The adaptive locking voltage value V lock,adp = 100 + 200 = 300V. The system applies a higher voltage of 300V to generate a stronger adsorption force to fix the thicker material.

[0228] The elements in the adaptive locking voltage matrix V lock,adp have the dimension of voltage (volt). The lower limit of its value range is the minimum locking voltage V min , and the upper limit is theoretically determined by the maximum value of the pressure foot vibration specific band attenuation rate R damp , but in actual application, it will be limited by hardware protection.

[0229] The parameter variation logic is as follows: when the down in the unit to be quilted is denser and thicker, its attenuation ability to vibration is stronger, resulting in a real-time measured pressure foot vibration specific band attenuation rate R dampThe value is higher. Higher foot vibration specific band attenuation rate R damp The value will linearly result in higher locking voltage increment AV lock , thus generating a higher final adaptive locking voltage value V lock,adp .

[0230] The rationality of the adaptive mechanism is demonstrated: the system automatically applies stronger locking force to thicker and more difficult-to-fix materials, and weaker locking force to light and thin materials that are easy to fix. This "on-demand allocation" strategy not only ensures the locking effect on all types of products, but also maximizes energy efficiency and protects delicate fabrics from unnecessary strong electric fields.

[0231] In another embodiment, it is necessary to note that:

[0232] To verify the technical effects of the "adaptive electric field locking and collaborative control activation method based on real-time feedback" described in the invention under different working conditions, this embodiment is performed. The test aims to quantify and compare the performance differences between the invention method and the traditional fixed voltage locking method in terms of quilting quality, locking energy consumption, and fabric protection.

[0233] The test equipment uses a highly customized "Seiko QX-2000-Pro" high-speed computer quilting machine. The core reform part of the equipment includes: integrating a high-density microelectrode array (resolution of 256x256 electrodes / 10cm 2 ) under the standard workbench plate; equipped with a high-speed multi-channel voltage controller with a response time less than 1 millisecond; integrating a high-frequency (sampling rate 50kHz) piezoelectric vibration sensor on the presser foot assembly; and installing a high-precision (resolution 0.01mm) machine needle position optical encoder. The whole system is scheduled by a central industrial control computer, which runs control software containing the complete algorithm of steps S1 to S5 described in the invention. The test parameters are preset as follows: the locking trigger distance is set to 2.0 millimeters, the minimum locking voltage V min is set to 100.0 volts, and the locking voltage gain coefficient k lock is set to 5.0 volts·second.

[0234] Six representative samples are selected for the test, covering a range of common products from ultra-thin to high-filled, to fully evaluate the adaptability of the invention method. Two copies of each sample are prepared, one for testing the invention method (test group), and the other for testing the traditional fixed voltage locking method (comparative example). The comparative example adopts a common strategy in the industry, that is, regardless of the material, a fixed and empirical 250.0 volt locking voltage is applied.

[0235] The test procedure is strictly in accordance with the complete logic of the present application. Taking the "thick winter quilt core" sample as an example, its detailed implementation process is as follows:

[0236] 1. Start and pretreatment: Place the sample on the quilting machine workbench. After starting the program, the system begins to perform sewing. During the movement of the machine needle to the first to-be-quilted unit, the presser foot vibration sensor and the high-definition camera continuously collect data, and the system controller performs real-time feature extraction and normalization in step S1, and LSTM model prediction in step S2, to obtain the predicted tension abnormality δ T,pred and the predicted down unevenness δ D,pred .

[0237] 2. Dynamic reconstruction and compensation: Based on the prediction results, the path and pressure collaborative compensation module (step S3) is in standby state. At the same time, the electric field control module (step S4) calculates the reconstruction voltage matrix V recon according to the predicted down unevenness δ D,pred through the electric potential gradient direct mapping model, and applies it to the microelectrode array. At this time, through high-speed photography, it can be observed that the down in the unit cell migrates from the local dense area to the sparse area under the action of invisible electrostatic force, and the down distribution of the whole unit tends to be uniform.

[0238] 3. Adaptive locking trigger and execution: When the reading feedback by the needle position encoder shows that the needle tip is 1.99 mm away from the entrance boundary of the current to-be-quilted unit, the system controller immediately triggers the locking program in step S5. First, the controller stops outputting the reconstruction voltage matrix V recon . Then, the controller instructs the presser foot vibration sensor to perform a transient high-frequency sampling, and quickly calculates the original physical value of the specific frequency band attenuation rate R damp of the presser foot vibration of the unit after reconstruction.

[0239] 4. Voltage calculation and application: The controller substitutes the measured presser foot vibration specific frequency band attenuation rate R damp original value (for example, for the thick winter quilt core, the measured value is 38.5 s -1 ) into the adaptive locking voltage calculation model, that is, the adaptive locking voltage value V lock,scalar = 100.0 V + 5.0 V·s × 38.5 s -1 = 292.5 V. Subsequently, the controller generates an adaptive locking voltage matrix V lock,adp with all elements being 292.5 V, and applies it to the microelectrode array corresponding to the unit. The strong and uniform electrostatic adsorption force instantly "freezes" the already homogenized down and the upper and lower layers of fabric in place.

[0240] 5. Collaborative activation and sewing: After applying the adaptive locking voltage matrix V lock,adpAt the same time, the controller raises the running permission of the path and pressure synergistic compensation module to the highest, making it fully activated. Then, the needle enters the unit for high-speed quilting. During this process, even if there is a small disturbance that is not completely suppressed by the electric field (such as the slight fabric drag caused by the needle puncture), the module can still make the final mechanical path and presser foot pressure fine-tuning.

[0241] 6. Data recording: After the completion of the unit sewing, the system automatically records the locked voltage value, the single-unit locking energy calculated by the voltage and current integration of this time. After sewing is completed, a special fabric light transmittance scanner is used to detect the finished product, quantify the post-sewing down unevenness, and the quality inspector evaluates whether the fabric has static damage through a microscope.

[0242] The above process is repeated for all six samples and their comparative examples, and the key performance indicators are recorded in the following table.

[0243] Table 2 Study on adaptive electric field locking and synergistic control activation method based on real-time feedback:

[0244]

[0245]

[0246]

[0247] From the above table data, the significant advantages of the method of the present application can be clearly seen.

[0248] 1. Universal quality improvement and stability: The "post-sewing down unevenness" of the method of the present embodiment is always stable at an extremely low level of 2.1% or less. The comparative examples perform poorly when dealing with materials with two polarizations (such as "light summer quilt core" and "thick winter quilt core"), with unevenness of 4.5% and 5.2% respectively. This shows that the fixed voltage of the comparative example may be too strong for light materials, causing disturbance, and not strong enough for heavy materials, while the adaptive mechanism of the present application can provide just the right amount of locking force for each material, ensuring universal and high-quality quilting results.

[0249] 2. Significant energy efficiency: When dealing with light materials, the energy consumption of the present embodiment is much lower than that of the comparative example (for example, the energy consumption of "light summer quilt core" is only 31.7% of that of the comparative example). Although the energy consumption may exceed the fixed value of the comparative example when dealing with heavy materials, in terms of the average energy consumption of the six samples as a whole, the present embodiment (average 30.3 mJ) saves more than 22% of energy compared to the comparative example (39.1 mJ). This proves the great value of the "on-demand allocation" strategy in energy efficiency optimization.

[0250] 3. Excellent material protection capability: the 250.0V fixed voltage of the comparative example has overvoltage risk for ultra-thin fabrics such as 15D and 20D, and is evaluated as "slight breakdown risk". The locking voltage (141.0V) of the embodiment is automatically reduced according to the actual situation of the material, which fundamentally avoids the possibility of static damage and realizes effective protection of delicate fabrics.

[0251] In summary, the embodiment fully proves the rationality and advancement of the adaptive electric field locking mechanism described in step S5 of the application through objective and quantifiable data. It can automatically match the optimal locking force according to the real-time physical state of the material, thereby achieving breakthrough progress compared to traditional technology in three dimensions of improving product quality, reducing production energy consumption and protecting raw materials.

[0252] It should be noted that all calculation formulas in the present application file use regression analysis in machine learning algorithms, including but not limited to machine learning algorithms, to analyze the collected relevant parameters in depth, identify their natural trends and mutual relationships. Professional software such as Python's Scikit-learn library or R language is used to automatically generate mathematical models that match the data. Then, the model performance is objectively evaluated through cross-validation and other methods, and combined with continuous feedback and optimization to ensure that the created formula truly reflects the inherent law of the data, thereby ensuring its effectiveness and accuracy. In all calculation formulas of the present application, the parameters in each formula are processed by consistent range of dimensionless to ensure that different physical quantities are compared on the same scale; Dimensionless techniques include but are not limited to Min-Max-Normalization, Z-Score standardization;

[0253] The technical solutions of the present application can be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), FLASH, hard disk or optical disk, etc., including a number of instructions to make a computer device (which can be a personal computer, server, or network device, etc.) execute the method of each embodiment of the present application.

[0254] The logic and / or steps represented in the flow diagrams, or otherwise described herein, for example, can be considered as a sequence of executable instructions, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For purposes of this specification, a "computer-readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be, for example but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium include an electrical connection, hard-wired

[0255] It should be noted that the above-mentioned embodiments are only used to illustrate but not to limit the technical solutions of the present application, and although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A method for optimizing the dynamic size adjustment interface of a down jacket quilting machine, characterized in that, The specific steps include: Step S1: Collect and process the raw signal data of the target fabric to extract the proportion of high-frequency energy in the needle-punching acoustic spectrum that is strongly correlated with the fabric tension state, and the attenuation rate of the presser foot vibration in a specific frequency band that is strongly correlated with the down density state. Step S2: Input the time series of the high-frequency energy ratio of the needle-punched acoustic spectrum and the attenuation rate of the specific frequency band of the presser foot vibration into the pre-constructed dual-channel prediction model. The dual-channel prediction model is used to predict the predicted tension anomaly of the next quilting path node and the predicted down unevenness of the next quilting unit. Step S3: Receive the predicted tension anomaly and send it to the path and pressure co-compensation module to calculate the path compensation vector through a nonlinear gain function; Based on the predicted tension anomaly and the predicted down unevenness, the presser foot pressure adjustment value is calculated through a collaborative weighted model. Step S4: Perform secondary analysis based on the predicted down unevenness to generate a reconstruction voltage matrix for reconstructing the down distribution in the next quilting unit; Step S5: When the needle on the quilting machine is about to enter the next quilting unit that has been reconstructed, switch to a new reconstruction voltage matrix, and then apply an adaptive locking voltage matrix to the microelectrode array corresponding to the next quilting unit to use electrostatic adsorption force to fix the target fabric in place, while putting the path and pressure co-compensation module into an active state.

2. The method for optimizing the dynamic size adjustment interface of a down jacket quilting machine according to claim 1, characterized in that: The system uses a piezo-acoustic sensor and a miniature accelerometer located at the presser foot of the quilting machine to synchronously collect raw signal data consisting of raw acoustic signals and raw vibration signals using a unified system clock. The original signal data is decoupled based on a preset signal processing model to extract the high-frequency energy ratio of the needle-punched acoustic spectrum, which represents the local tension state of the fabric, and the specific frequency band attenuation rate of the presser foot vibration, which represents the density of down under the presser foot. The value range of the high-frequency energy ratio of the needle-punched acoustic spectrum is limited to the range of [0,1]. When the high-frequency energy ratio of the acupuncture acoustic spectrum output approaches 0, it indicates that the proportion of the high-frequency energy integral value relative to the full-frequency energy integral value is lower. When the proportion of high-frequency energy in the acoustic spectrum of acupuncture approaches 1, it indicates that the high-frequency energy integral value dominates the overall energy integral value across the entire frequency band. The range of the attenuation rate of the presser foot vibration in a specific frequency band after nonlinear mapping by the Sigmoid function is limited to (0,1); When the output of the attenuation rate of the presser foot vibration in a specific frequency band after nonlinear mapping is closer to 0, it indicates that the original attenuation rate value of the presser foot vibration in that specific frequency band is smaller. When the output of the attenuation rate of the presser foot vibration in a specific frequency band after nonlinear mapping is closer to 1, it indicates that the original attenuation rate value of the presser foot vibration in a specific frequency band is larger.

3. The method for optimizing the dynamic size adjustment interface of a down jacket quilting machine according to claim 2, characterized in that: The dual-channel prediction model includes a first prediction channel and a second prediction channel. The first prediction channel predicts the predicted tension anomaly of the next quilting path node by assigning higher weights to key time points in the historical sequence corresponding to the proportion of high-frequency energy in the acupuncture acoustic spectrum. The second prediction channel predicts the down unevenness of the next quilting unit by assigning higher weights to key time points in the historical sequence corresponding to the attenuation rate of a specific frequency band of the mapped presser foot vibration. The dual-channel prediction model incorporates an attention mechanism; the dual-channel prediction model consists of two parallel, structurally identical long short-term memory networks, with a self-attention layer connected to the output of each LSTM network for calculating weights; The predicted range of tension anomaly is limited to (-1, 1); When the predicted tension anomaly output is closer to -1: the larger the predicted fabric tension is below the ideal value, the more relaxed the fabric is. When the predicted tension anomaly output is closer to 1, it means that the model predicts that the future fabric tension will be greater than the ideal value, and the tighter the state will be. When the predicted down unevenness output is closer to -1, it indicates that the model predicts a sparser future down state. When the predicted down unevenness output is closer to 1, it indicates that the model predicts a greater degree of down density in the future.

4. The method for optimizing the dynamic size adjustment interface of a down jacket quilting machine according to claim 3, characterized in that: The nonlinear gain function is used to compensate only when the predicted tension anomaly exceeds the preset tension compensation trigger threshold; and the nonlinear gain function is an exponential function; the path and pressure collaborative compensation module is equipped with a collaborative weighted model; The collaborative weighted model includes calculating the pressure collaborative weighting factor, introducing the basic pressure adjustment demand, and multiplying the basic pressure adjustment demand by the pressure collaborative weighting factor. The result of the multiplication is then multiplied by a preset pressure compensation base gain coefficient to obtain the final presser foot pressure adjustment value. The more abnormal the tension of the target fabric becomes, the greater the absolute value of the predicted tension abnormality, the more rapidly the pressure synergy weight factor decreases, and the smaller the adjustment range of the presser foot pressure adjustment value.

5. The method for optimizing the dynamic size adjustment interface of a down jacket quilting machine according to claim 4, characterized in that: A dynamic electric field driving command is generated for reconstructing the down distribution in the next quilting unit; further comprising: the electric field control module calculates the reconstructed voltage matrix based on the predicted down unevenness through a preset potential gradient direct mapping model. The reconstructed voltage matrix is ​​applied to the microelectrode array to generate a non-uniform electrostatic field. This non-uniform electrostatic field pulls down in the target fabric from the predicted dense region to the predicted sparse region by generating a dielectric electrophoretic force opposite to the direction of the down density gradient.

6. The method for optimizing the dynamic size adjustment interface of a down jacket quilting machine according to claim 5, characterized in that: The microelectrode array is set below the worktable of the quilting machine. The potential gradient direct mapping model is a mathematical model that directly converts the predicted down unevenness into the target potential gradient, and then analytically solves the voltage of each electrode. The more the predicted down unevenness output deviates from 0, the stronger the dielectric electrophoretic force needs to be generated by the microelectrode array.

7. The method for optimizing the dynamic size adjustment interface of a down jacket quilting machine according to claim 6, characterized in that: Based on the attenuation rate of a specific frequency band of presser foot vibration collected in real time within the quilting unit, an adaptive locking voltage matrix is ​​calculated and generated, and the adaptive locking voltage matrix is ​​applied to the microelectrode array to fix the target fabric in situ. The needle position is determined by the needle position encoder. When the distance between the needle and the boundary of the unit to be quilted is less than the preset locking trigger distance, the "adaptive locking voltage matrix" calculation and the "path and pressure collaborative compensation module" activation operation are performed.

Images