A method and system for identifying yarn parameters in knitted loop digital twins

By adding a downstream tension sensor to the weft knitting machine and using reference work orders as anchor points, the problem of identifying yarn mechanical parameters was solved, enabling precise setting of yarn feed amount for weft knitting digital twins, reducing the number of sampling attempts, and improving the production efficiency and finished product quality of weft knitting loops.

CN123113744APending Publication Date: 2026-11-17SUZHOU DAYI TEXTILE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611326088.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-29
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

The yarn mechanical parameters in the weft knitting digital twin can only be taken as nominal values, which do not match the actual batch; the original tension sensor of the machine cannot obtain yarn batch information because it is located at the feedback point of the tension closed loop of the yarn feeder; the dry relaxation shrinkage model of the twin itself has calibration mismatch, and directly substituting accurate parameters will amplify the error; as a result, new varieties still need to be sampled repeatedly before they are put into use.

Method used

For each loop formation path, a downstream tension sensor is added between the yarn guide and the loop formation area of ​​the needle cylinder. Using the most recent qualified production work order of the product to be woven as a reference, the friction coefficient and equivalent stiffness of the yarn are calculated by the idle running section and multi-stage loop formation detection, respectively. The dry relaxation shrinkage rate is then calculated, and the yarn feed is adjusted using this as an anchor point to avoid amplification of model errors.

Benefits of technology

It significantly reduces the number of sampling attempts for new varieties, increases the first-round pass rate, reduces errors, and enhances the accuracy of yarn batch information and the precision of the model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123113744A_ABST
    Figure CN123113744A_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for identifying yarn parameters in a digital twin for knitting loop formation, belonging to the field of digital control technology in knitting production. A tension sensor is added between the yarn guide and the loop-forming zone, downstream of the tension closed-loop feedback point of the yarn-storage active yarn feeder. When the knitting needle does not enter the loop-forming zone, it idles. The yarn friction coefficient is obtained by dividing the natural logarithm of the ratio of the sensor reading to the yarn feeder tension setpoint by the cumulative wrap angle. After the knitting needle enters the loop-forming zone, it knits several rows at at least two loop depths. The peak-to-valley difference of the synchronous average waveform of the loop-forming cycle is linearly fitted to the loop depth, and the equivalent yarn stiffness is obtained from the slope. Using the previous qualified work order of the same variety as the anchor point, the actual shrinkage rate is calculated from the measured longitudinal density. The twin model only handles the shrinkage rate increment, and the yarn feed amount is calculated and sent to the yarn feeder transmission ratio control interface. This invention increases the first-round pass rate of new varieties from 32% to 94.5%.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of digital control technology in knitting production, specifically to a method and system for identifying yarn parameters in knitting loop digital twins. Background Technology

[0002] The weft-knitting circular knitting machine is the main equipment for knitted fabric production. During weaving, the yarn is fed in quantitatively by a yarn-storage type positive yarn feeder according to the set feed amount. The knitting needles bend the yarn into loops under the action of the yarn bending cam, and the greige fabric is drawn out by the drafting and take-up mechanism. After the greige fabric comes off the machine, it undergoes a period of dry relaxation. The internal stress in the loops is gradually released, and the fabric shrinks. The final stable warp density of the finished product is the density specification delivered to the customer. When producing a new variety, the process engineer needs to first set the yarn feed amount based on experience and knit a trial section. After dry relaxation, the warp density of the finished product is measured. If it does not meet the standard, the yarn feed amount is adjusted and the trial is repeated until it falls within the density tolerance zone specified in the contract. This trial and error process is called prototyping, and each prototyping consumes raw materials and machine time.

[0003] In recent years, digital twin technology has been introduced into the textile industry. The idea is to create a virtual machine that runs on a computer, where process parameters are pre-calculated to replace physical trial and error. However, in the scenario of weft knitting loops, existing solutions have not achieved the accuracy required to replace sampling, for the following reasons.

[0004] Firstly, most existing textile digital twin solutions remain at the system architecture level. These solutions focus on how to collect data, how to reconstruct the machine's shape in a 3D engine, how to synchronize real-time data with the virtual object, and how to trigger alarms when deviations occur. However, they avoid a more fundamental question: where do the parameters describing the yarn's mechanical behavior within the virtual object actually come from? These parameters are invariably taken from yarn specification tables or historical experience values, so the results calculated by the virtual object naturally carry a fixed deviation from the actual product.

[0005] Secondly, even for yarns of the same specification, there are significant differences in mechanical properties between different batches. Taking 75-denier, 72-well polyester low-elasticity yarn as an example, the equivalent stiffness difference between different batches can reach over 15%, and the coefficient of friction between the yarn and the metal guide surface can differ by over 20%. These differences are not reflected in label information such as yarn count, hole count, or raw materials, but they precisely determine the amount of shrinkage after the fabric is removed from the machine. If twins calculate according to the nominal parameters, they will give the same yarn feed recommendation for different batches of yarn, but the actual finished product density will be different.

[0006] Thirdly, a natural idea is to utilize existing sensor data on the machine to calibrate the twin. Modern circular knitting machines are equipped with a yarn feeder with a storage-type active yarn feeder on each loop, and a tension sensor is located at its outlet. However, the sensor's location means it cannot obtain the necessary information. This sensor is precisely the feedback point of the yarn feeder's tension closed loop. The closed loop's function is to maintain the tension at this point at the set value. Therefore, regardless of the amount of yarn friction, the reading at this point always returns to near the set value. Simultaneously, the periodic tension pulsations generated during the loop formation process are significantly attenuated by the yarn loops wound on the storage device through mechanical low-pass transmission before reaching this point. In other words, information about batch-to-batch yarn differences is erased at this measurement point by the closed loop and the yarn loops working together. This is a structural deficiency caused by the measurement point's location, which cannot be compensated for by increasing the sampling rate, increasing the sample size, or changing the algorithm.

[0007] Fourth, even if accurate yarn parameters are obtained, it may not improve predictions. The dry relaxation shrinkage model within the twin is calibrated from historical work orders, and its coefficients carry calibration errors. Substituting accurate yarn parameters into a model with mismatched coefficients will amplify the error along the model, potentially resulting in worse performance than simply using the nominal parameters. This point is easily overlooked in practice: people tend to assume that the more accurate the input, the more accurate the output, while neglecting to guard against model mismatches.

[0008] Fifth, another objective obstacle has prevented this problem from being solved for a long time: dry relaxation shrinkage occurs after weaving is completed, and is a quantity that can only be observed afterward, making it impossible to directly measure and close the loop during the weaving process. Therefore, any control scheme that relies on online feedback is powerless against it; and to predict it before weaving, the mechanical parameters of the current batch of yarn must be known first, which brings us back to the dilemma mentioned in the third point above.

[0009] Therefore, there is a need in the field for a method and system that can directly identify the mechanical parameters of a batch of yarn on the machine before formal production, and allow these parameters to be included in the prediction without being amplified by the mismatch of the twin model itself. Summary of the Invention

[0010] The technical problem this invention aims to solve is that the yarn mechanical parameters in the weft knitting digital twin can only be taken as nominal values, which do not match the actual batch; the original tension sensor of the machine cannot obtain yarn batch information because it is located at the feedback point of the tension closed loop of the yarn feeder; the dry relaxation shrinkage model of the twin itself has calibration mismatch, and directly substituting accurate parameters will amplify the error; as a result, new varieties still need to be sampled repeatedly before being put into operation.

[0011] The technical concept of this invention is based on three physical facts. First, from the outlet of the yarn feeder to the looping zone of the needle cylinder, the yarn passes through a series of guiding components such as the yarn guide. Its tension increases with the cumulative wrap angle and the coefficient of friction according to the relationship between wrapping friction and yarn tension. As long as the measuring point is downstream of the feedback point of the yarn feeder tension closed loop, the tension at that point simultaneously contains information on the yarn feeder setpoint and yarn friction, and the closed loop has no constraint on it. Second, after the needle enters the bending zone, the yarn is stretched and bent into loops in each looping cycle. The peak tension at the looping moment is higher than the trough value during the non-looping period. The difference between the two increases linearly with the bending depth, and its growth slope is proportional to the equivalent stiffness of the yarn, but is independent of the baseline rise caused by friction. Therefore, by assigning friction and stiffness to two uncoupled detection programs, they can be calculated independently. Third, the dry relaxation shrinkage of the fabric is driven by the bending elastic recovery of the loops and hindered by the frictional locking between the yarns. Therefore, the shrinkage rate is simultaneously governed by the equivalent stiffness and the coefficient of friction.

[0012] Most importantly, this invention prevents the identified parameters from being input into the twin model as absolute quantities. For the most recently inspected and approved production order of the same yarn variety, the setpoint for yarn feed and the actual measured yarn density are readily available figures in the ledger. These two figures can be used to calculate the actual shrinkage rate of that production run, a value free of any model error. This invention uses this production order as an anchor point, allowing the twin model to only account for the incremental shrinkage rate of this batch of yarn relative to the reference batch. In this way, the constant term mismatch in the shrinkage model is eliminated entirely, and its coefficient mismatch only affects this small incremental value, reducing the error by an order of magnitude. This is the key difference between this invention and conventional methods: conventional methods treat the twin as an absolute calculator, while this invention treats it as a comparator only responsible for calculating differences.

[0013] To solve the above-mentioned technical problems, the present invention provides a method and system for identifying yarn parameters in a knitting loop digital twin. The method for identifying yarn parameters in a knitting loop digital twin includes: S1, adding a downstream tension sensor between the yarn guide and the looping area of ​​the needle cylinder, and downstream of the tension closed-loop feedback point of the yarn feeder, for each looping path; S2, using the most recently inspected and approved production work order of the product to be knitted as a reference work order, reading its yarn feed setting value, measured finished product density, reference batch friction coefficient, and reference batch equivalent stiffness; S3, idle running detection, with the needles not entering the yarn bending area, the machine idles at a production speed for no less than 30 rows, and the friction coefficient of the yarn in this batch is obtained by taking the natural logarithm of the ratio of the arithmetic mean of the downstream tension sensor signal to the yarn feeder tension setting value, and then dividing by the cumulative wrap angle; S4, multi-level looping detection, knitting at no less than two levels of yarn bending depth, each for no less than 30... For each row, the peak-to-valley difference of the synchronous average waveform of the looping cycle is taken for each grade. The peak-to-valley difference is used to perform a univariate linear fitting on the bending depth, and the equivalent stiffness of this batch of yarn is obtained from the fitting slope; S5, the actual dry relaxation shrinkage rate is calculated by back-calculating the yarn feeding amount setting value of the reference work order and the actual finished product longitudinal density; S6, the two sets of parameters of this batch and the reference batch are respectively substituted into the dry relaxation shrinkage model, and the difference between the two outputs is taken as the shrinkage rate increment. This is added to the actual dry relaxation shrinkage rate to obtain the predicted value of the dry relaxation shrinkage rate of this batch of yarn; S7, the yarn feeding amount setting value required to achieve the target finished product longitudinal density is calculated in reverse, and sent to the transmission ratio control interface of the yarn storage type active yarn feeder via fieldbus for execution, and put into formal production.

[0014] Furthermore, the number of bend depth settings can be 3 to 6, symmetrically distributed around the process bend depth. When the root mean square of the residual exceeds one-tenth of the peak-to-valley difference, the lowest setting is determined to be in the yarn relaxation zone and is discarded. The needle height of the idle section is 1.5 mm to 3.0 mm above the lowest point of the bend triangle, and the first 5 rows of idle runs are discarded. The looping cycle is triggered by each needle pulse of the main shaft encoder, and the rows containing joints with tension differences exceeding three times the median are discarded before synchronizing and averaging. When the equivalent stiffness variation coefficient recorded in previous work orders for this product is less than 10%, the identification is skipped and the yarn feed amount of the reference work order is directly used. Each looping path is identified and the median is taken. Paths with deviations exceeding 20% ​​of the median are marked as abnormal and a maintenance prompt is output. After the actual measurement of the fabric off the machine is qualified, the work order database is written back to update the anchor point. If it is unqualified, the yarn feed amount is corrected according to the partial derivative of the twin at the current working point and the fabric is re-woven.

[0015] This invention also provides a corresponding system, including a downstream tension sensor, a data acquisition unit, a work order database, a parameter identification module, a twin server, and a transmission ratio control interface. This invention also provides a computer-readable storage medium storing the corresponding computer program.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0017] First, the tension measuring point is moved from the feedback point of the yarn feeder tension closed loop to its downstream location, preventing the closed loop from erasing yarn batch information. Numerical simulations show that when the true value of the yarn friction coefficient changes from 0.14 to 0.36, the reading at the original measuring point on the machine remains constant at 3.421 centinenewtons, while the readings at the downstream measuring point added in this invention are 4.278, 5.020, and 6.083 centinenewtons, respectively. The friction coefficients obtained from this are 0.126, 0.225, and 0.345, respectively.

[0018] Second, the friction coefficient and equivalent stiffness are assigned to two uncoupled detection programs for the idle run and multi-stage looping sections, avoiding the inherent numerical ill-conditioning of simultaneously resolving two parameters from a single waveform. Numerical simulations show that the relative error of equivalent stiffness identification is 1.1%, and that of friction coefficient is 4.7%, which are improvements of approximately 13 times and 5 times respectively compared to 14.4% and 23.7% without identification.

[0019] Third, by using the reference work order as an anchor point and having the twin model only bear the incremental shrinkage rate, the calibration mismatch of the model itself is no longer amplified. Numerical simulations show that, under the condition that the twin model calibration mismatch is 20%, the first-round compliance rate of the scheme that directly substitutes the identified data into the absolute model is only 36.5%, while the present invention, which uses anchor point increments, achieves 94.5%. When the mismatch increases to 35%, the former drops to 4%, lower than the 30% achieved by not identifying the data at all, while the latter remains at 82.7%.

[0020] Fourth, the number of sampling attempts for new varieties has been significantly reduced. Numerical simulations show that the proportion of finished products that achieve the target longitudinal density tolerance zone in the first round has increased from 32% in the current practice to 94.5%, and the number of trial weaving attempts required to achieve the standard has decreased from 1.68 to 1.05.

[0021] Fifth, this invention clearly defines its applicable conditions and incorporates them into the scheme. Numerical simulations show that when the coefficient of variation of the yarn batch is zero, the first-round compliance rate of this invention is 99.3%, slightly lower than the 100% of the current practice; when the coefficient of variation reaches 10%, the compliance rate of this invention is 99.3%, while the current practice is 54%. Accordingly, the claims include an applicability determination based on a coefficient of variation of 10%, avoiding unnecessary additional detection steps. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the system composition and sensor arrangement according to an embodiment of the present invention;

[0023] Figure 2This is a flowchart of a method according to an embodiment of the present invention;

[0024] Figure 3 This is a schematic diagram illustrating the tension signal and parameter inversion principle of the two-stage detection in the empty walking segment and the multi-stage circular segment in an embodiment of the present invention;

[0025] Figure 4 The figures show a comparison of the numerical simulation results of embodiments of the present invention. Detailed Implementation

[0026] The present invention will be further described below with reference to the accompanying drawings and embodiments. The following embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. The present invention is applicable to various weft knitting machines equipped with a yarn storage type active yarn feeder and an electric yarn bending depth adjustment mechanism, including single-sided and double-sided circular knitting machines, rib knitting machines and cotton knitting machines, without limitation on cylinder diameter, machine size and number of loops.

[0027] Note: All effect data presented in this specification are numerical simulation results, not actual bench test values. The modeling methods, parameter values, and control settings used in the simulation are provided at the end of this section to enable those skilled in the art to reproduce the results.

[0028] I. System Composition

[0029] This embodiment uses a 34-inch, 24 needles per inch, 72-way double-sided weft knitting circular knitting machine as the subject of this invention. The system composition is as follows: Figure 1 As shown.

[0030] The cylinder 1 is equipped with knitting needles 2. The knitting needles 2 complete the loop-forming motion under the action of the yarn bending cam 3. The height of the yarn bending cam 3 is driven by an electric yarn bending depth adjustment mechanism 4, which is a lead screw mechanism driven by a stepper motor. The adjustment resolution of the yarn bending depth is 0.01 mm, and the stroke is 2.0 mm to 5.0 mm. Each loop-forming path is equipped with a yarn storage type active yarn feeder 5. Its outlet is equipped with the machine's original upstream tension sensor 6. The reading of the upstream tension sensor 6 is fed back to the tension adjustment ring of the yarn storage type active yarn feeder 5, forming the feedback point of the tension closed loop. In this embodiment, the tension setting value of this closed loop is 3.5 centinewtons. After being drawn out from the yarn storage type active yarn feeder 5, the yarn enters the loop-forming area 9 via the yarn guide 7.

[0031] This invention adds a downstream tension sensor 8 between the yarn guide 7 and the looping area 9. The downstream tension sensor 8 is located downstream of the feedback point formed by the upstream tension sensor 6. This is a key arrangement that distinguishes this invention from existing technologies: the tension closed loop of the yarn storage type active yarn feeder 5 can only constrain the tension at the upstream tension sensor 6, and has no constraint on the tension at the downstream tension sensor 8. Therefore, the latter's reading simultaneously includes the yarn feeder tension setpoint and the amplification of yarn friction along the yarn guide path. The downstream tension sensor 8 is a piezoelectric tension sensor with a range of 0 to 40 centinewtons, a sampling rate of 50 kHz, and 12-bit analog-to-digital conversion. It is installed on the yarn exit side of the yarn guide 7, with the yarn wrap angle on its detection wheel being 15 to 30 degrees. The cumulative wrap angle of the yarn path from the upstream tension sensor 6 to the downstream tension sensor 8 is the structural constant of the machine; in this embodiment, it is 1.60 radians. This value is determined once by the geometric position of the yarn guide and written to the twin server 12.

[0032] The acquisition unit 10 is connected to the downstream tension sensor 8, the upstream tension sensor 6, and the spindle encoder 19. It synchronously acquires tension signals using each pulse from the spindle encoder 19 as a trigger signal and uploads them via the fieldbus 11. In this embodiment, the fieldbus 11 uses an industrial Ethernet network.

[0033] The twin server 12 includes a parameter identification module 13, a dry relaxation shrinkage model 14, and a yarn feed inverse calculation module 15, and is connected to the work order database 16. The work order database 16 stores the yarn feed setting values, measured finished product density, yarn friction coefficient, and equivalent yarn stiffness for each product's production work orders over the years. The yarn feed setting value obtained by the twin server 12 is sent to the yarn storage type active yarn feeder 5 via the transmission ratio control interface 17, which adjusts the yarn feed for each loop. The traction and take-up mechanism 18 leads out the greige fabric in a conventional manner. The human-machine terminal 20 is connected to the twin server 12, displaying various parameters, predicted shrinkage rate values, yarn feed setting values, and the numbers and maintenance prompts for abnormal loops.

[0034] It should be noted that the imaging and signal acquisition in this embodiment are both directed towards the yarn and fabric, and do not include personnel within the field of view. The data collected and stored are only the equipment operating parameters and process parameters, and do not involve personal information.

[0035] II. Method Implementation

[0036] The method flow of this embodiment is as follows: Figure 2 As shown, the fabric is a double-sided plain knitted fabric made of 75 denier 72-hole polyester low-elasticity yarn, with a target finished product density of 5.65 rows per centimeter, and a contractual allowable deviation of ±0.05 rows per centimeter.

[0037] S1. Sensor Arrangement. Following the steps outlined in Section 1 above, add downstream tension sensors 8 to each loop path to measure and record the cumulative wrap angle of 1.60 radians. This step is a one-time installation and will not be repeated in subsequent production runs on the same machine.

[0038] S2. Read the anchor point. Retrieve the most recent production work order that passed the finished product inspection for this variety from the work order database 16 as a reference work order. Read its yarn feed setting value of 8.09 mm per loop, the actual measured finished product density after drying and relaxation of the yarn after the machine is 5.65 mm per column, and the reference batch friction coefficient of 0.258 and the equivalent stiffness of 289 centi-Newtons obtained by S3 and S4 for this work order.

[0039] S3. Idle Section Detection. The electric yarn bending depth adjustment mechanism 4 raises the yarn bending triangle 3, causing the knitting needle 2 to stop 2.0 mm above its lowest point on the yarn bending triangle. At this height, the needle hook of the knitting needle 2 does not hook the yarn, and the yarn only passes through the guide path. The machine is idled at a production speed of 25 rpm for 240 rows, and the signal from the downstream tension sensor 8 is collected. After excluding the first 5 rows of idle operation, the arithmetic mean of the remaining sections is taken, which is 5.020 centinewtons in this example. At the same time, the reading of the upstream tension sensor 6 is read, which is 3.472 centinewtons in this example. The deviation from the tension set value of 3.5 centinewtons is 0.8 percent, which does not exceed 5 percent, and the tension closed loop of the yarn feeder is judged to be normal. Taking the natural logarithm of the ratio of 5.020 to 3.5 gives 0.3597, which is then divided by the cumulative wrap angle of 1.60° to obtain the coefficient of friction of this batch of yarn, which is 0.225.

[0040] The mechanism here is as follows Figure 3 As shown in the upper part, the tension of the yarn increases exponentially with the cumulative wrap angle as it passes through the yarn guide. The higher the coefficient of friction, the greater the increase in tension at the downstream tension sensor 8 relative to the set value of the yarn feeder. The key point is that the tension at the upstream tension sensor 6 is locked at the set value in a closed loop. When the coefficient of friction changes from 0.14 to 0.36, the reading at this point remains constant at 3.421 centinenewtons, indicating that friction information is structurally absent at this point. Under the same conditions, the readings of the downstream tension sensor 8 are 4.278, 5.020, and 6.083 centinenewtons respectively, and the deduced coefficients of friction are 0.126, 0.225, and 0.345 respectively. This is why the present invention must include a downstream measuring point.

[0041] S4. Multi-level loop detection. Reset the yarn bending triangle 3 to allow the knitting needle 2 to enter the yarn bending area. The electric yarn bending depth adjustment mechanism 4 sequentially sets three yarn bending depths of 3.30, 3.40, and 3.50 mm, with a difference of 0.10 mm between adjacent levels. Each level weaves 240 rows. For each gear, the signal from the downstream tension sensor 8 is segmented according to the looping cycle using each pulse of the main spindle encoder 19 as the trigger signal. The number of sampling points extracted in each looping cycle is the integer part of the quotient of the looping frequency of 1068 Hz and the sampling rate of 50 kHz, which is 46 points. After removing the cross-columns containing joints whose tension range exceeds three times the median of the tension range of all cross-columns in that gear, the arithmetic mean of the sampling values ​​of each cross-column on the same phase is taken to obtain the synchronous average waveform of the looping cycle of that gear. The difference between the maximum and minimum values ​​of this waveform is taken as the peak-valley difference of the looping tension of that gear. In this example, the three gears are 5.66, 13.19, and 20.71 centinewtons, respectively.

[0042] A univariate linear fit was performed with the bending depth as the independent variable and the peak-to-valley difference as the dependent variable, yielding a slope of 75.26 centinewtons per millimeter. Multiplying this slope by the current yarn feed setting of 8.09 millimeters and then dividing by the geometric sensitivity of the yarn demand per loop of this machine to the bending depth (2.05), the equivalent stiffness of this batch of yarn was obtained as 297 centinewtons. The geometric sensitivity is a machine constant determined by the geometric positions of the needle hook and sinker, expressed in millimeters of yarn length per millimeter of bending depth, provided by the machine manufacturer or measured during a single calibration.

[0043] The mechanism here is as follows Figure 3 As shown in the lower half, during loop formation, the yarn is stretched and bent, and its peak tension rises from the trough value during the non-loop formation period. The amount of rise is equal to the product of the equivalent stiffness and the yarn strain at the moment of loop formation. Since the strain increases linearly with the bending depth, the peak-to-trough difference also increases linearly with the bending depth, and the slope is only related to the equivalent stiffness. The baseline rise caused by friction occurs simultaneously at both the peak and trough values ​​and is eliminated when the difference is calculated. Therefore, the slope is not affected by the coefficient of friction, which is why the two parameters can be separated.

[0044] Special attention needs to be paid to the value of the adjacent grade difference. When the grade difference is too small, the change in peak-to-valley difference approaches the measurement resolution, increasing the uncertainty of the fitting slope; when the grade difference is too large, the yarn demand for a single loop corresponding to the lowest grade will be less than the set yarn feed amount, the yarn will no longer be stretched during loop formation and will be in a relaxed state, the loop tension will rise to zero, and the linear relationship between peak-to-valley difference and yarn bending depth will be disrupted. Numerical simulations show that when the grade difference is ±0.03 mm, the first-round pass rate is 90%, and when it is ±0.06 to ±0.15 mm, it is 94% to 98.7%. Therefore, this invention limits the grade difference to 0.05 mm to 0.15 mm.

[0045] Regarding the number of gears, numerical simulations show that gear 2 accounts for 96%, gear 3 for 95.3%, gear 4 for 96%, and gear 6 for 94.7%, with no significant difference between them. This aligns with the geometric fact that a straight line can be determined by two points. Therefore, the independent claim of this invention is limited to at least two gears. The significance of increasing the number of gears lies solely in providing fitting residuals to determine whether any gear falls into the relaxation region. This is described as a preferred embodiment in the dependent claims.

[0046] Regarding the number of rows per gear, numerical simulations show that the first-round pass rate is 90.7% with 10 rows, 95.3% with 30 rows, and still 95.3% with 240 rows. It is clear that 30 rows are sufficient, and adding more does not yield any benefit; therefore, this invention uses no fewer than 30 rows. Correspondingly, the entire detection procedure, including 3 gear detections and 1 idle run, takes no more than 15 seconds at the machine speed of this embodiment, which is negligible compared to the machine time of a complete sample production run.

[0047] S5. Anchor point shrinkage rate calculation. The proportional constant between the loop length and the finished warp density of this fabric is 42.0, which is calibrated once from the historical work orders of this product. The yarn feed setting value of 8.09 mm in the reference work order is used to calculate the loop length during weaving using the conversion function from yarn feed to loop length during weaving. Then, the proportional constant is divided by the product of this loop length and the actual finished warp density of the reference work order. Subtracting the quotient from 1 yields the actual dry relaxation shrinkage rate of the reference work order, which is 0.0832. This value is calculated entirely from two measured figures in the ledger, without any model error. This is the foundation of the anchor point mechanism.

[0048] S6. Incremental Retraction Rate Prediction. The form of the dry relaxation shrinkage model 14 is: the shrinkage rate equals the baseline shrinkage rate, plus the product of the stiffness influence coefficient and the relative deviation of the equivalent stiffness from its nominal value, and then subtracts the product of the friction influence coefficient and the relative deviation of the friction coefficient from its nominal value. In this embodiment, the baseline shrinkage rate is taken as 0.090, the stiffness influence coefficient as 0.050, the friction influence coefficient as 0.035, the nominal value of the equivalent stiffness as 300 centinewtons, and the nominal value of the friction coefficient as 0.24. Each value is calibrated by regression from the historical work orders of this machine. Substituting the parameters of this batch, 297 centinewtons and 0.225, into the equation yields 0.0917. Substituting the parameters of the reference batch, 289 centinewtons and 0.258, yields 0.0855. The difference between the two, i.e., the increment of the shrinkage rate, is positive 0.0061. Adding the actual dry relaxation shrinkage rate of 0.0832 from the reference work order to this increment, we obtain the predicted dry relaxation shrinkage rate of this batch of yarn as 0.0893.

[0049] This is where our invention differs from conventional methods. Conventional methods directly substitute 297 and 0.225 into the model to obtain 0.0917 and calculate accordingly. In this case, the calibration errors of all three model coefficients are included in the result. Our invention only takes the difference between the two outputs, completely eliminating the baseline shrinkage rate. The calibration errors of the stiffness and friction influence coefficients also only affect a small amount, positive 0.0061. Numerical simulations show that, under a 20% model calibration mismatch, the absolute error of the shrinkage rate prediction decreases from 0.91 percentage points (from direct substitution) to 0.31 percentage points.

[0050] S7. Yarn Feed Quantity Calculation and Distribution. The yarn feed quantity calculation module 15 uses the predicted shrinkage rate of 0.0893 as a parameter in the twin to calculate the yarn feed quantity setting value that makes the finished product's longitudinal density equal to the target value of 5.65 per centimeter per row. In this example, it yields 8.15 mm per loop. The shrinkage rate of this batch of yarn is higher than that of the reference batch, so the required yarn feed quantity is correspondingly greater than the 8.09 mm per loop in the reference work order, using a longer loop length to compensate for the excess shrinkage. This value is distributed to the yarn storage type active yarn feeder 5 via the fieldbus 11 and the transmission ratio control interface 17. The yarn storage type active yarn feeder 5 adjusts the transmission ratio of its transmission wheel accordingly, so that the yarn feed quantity for each loop becomes 8.15 mm per loop, and the machine is then put into formal production.

[0051] III. Further Implementation Methods

[0052] Applicability Determination. The coefficient of variation of the equivalent stiffness recorded in previous production orders for this product is taken. When it is below 10%, steps S3 to S6 are skipped, and the yarn feed setting value of the reference order is directly used as the input for S7. This determination is based on the negative results given by numerical simulation: when the yarn batch coefficient of variation is zero, the first-round compliance rate of this invention is 99.3%, slightly lower than the 100% compliance rate of directly using the reference order; when the coefficient of variation is 5%, the rates are 99.3% and 90.7%; and when it reaches 10%, the rates are 99.3% and 54%. It is evident that when the batch difference is sufficiently small, the detection program is of no benefit. This invention accurately records this boundary and sets the skip condition accordingly.

[0053] Individual path identification and abnormal path recognition. S3 and S4 are performed on all 72 looping paths separately, and the median of the friction coefficient and the median of the equivalent stiffness obtained from each path are taken as the parameters of this batch of yarn. The median is used instead of the arithmetic mean because wear or dust accumulation on the yarn guide of individual paths can cause the reading of that path to deviate significantly, and the median is not sensitive to such outliers. When the deviation of the equivalent stiffness obtained from any path from the median exceeds 20% of the median, the path is marked as an abnormal path and its number and maintenance prompt are output by the human-machine terminal 20, prompting the machine operator to check the yarn guide and yarn feeder of that path. This function is an additional benefit brought by the sensor arrangement of this invention: the original measuring points of the machine were fixed by the closed loop and could not distinguish the differences between the paths.

[0054] Anchor point update. After the fabric produced in this batch is dried and relaxed on the machine, the measured warp density of the finished product is updated. If the deviation from the target value is less than one percent of the target value, the current yarn feed setting, the measured warp density, the coefficient of friction, and the equivalent stiffness are written into the work order database 16 as a reference work order for the next production of this variety. If the deviation exceeds one percent, the yarn feed setting is corrected based on the difference between the measured value and the target value, and the partial derivative of the warp density of the twin at the current yarn feed with respect to the yarn feed. The correction is then sent through the transmission ratio control interface 17 and the fabric is rewoven until it falls within the tolerance zone. The anchor points are updated on a rolling basis with each qualified production run, and can simultaneously track slowly changing factors such as machine wear.

[0055] IV. Modeling Methods and Results of Numerical Simulation

[0056] This section provides the sources for all the above-mentioned effect data. It should be noted again that the following data are all numerical simulation results, not actual measured values ​​from bench tests.

[0057] Modeling Method. The forward model synthesizes the signal that the sensor should see according to physical laws, while the inverse kinematics algorithm can only read this signal and cannot read the true parameters; the two are completely separated. In the forward model: the instantaneous yarn strain during loop formation is calculated by subtracting the yarn feed rate from the yarn demand of a single loop and dividing by the yarn feed rate. The demand rate varies linearly with the bending depth according to a geometric constant. When the yarn feed rate is excessive, the strain is taken as zero. The peak tension is equal to the exponent of the product of the yarn feeder tension setpoint, the friction coefficient, and the cumulative wrap angle of the loop formation section, plus the product of the equivalent stiffness and the strain. The valley tension is calculated using the same formula but takes the cumulative wrap angle of the idle section. The signal from the upstream measuring point is compressed using closed-loop gain to reduce the residual and attenuated according to the loop transmission rate to reduce the loop formation pulsation. Interference factors include: tension sensor noise of 0.09 centine root mean square (RMS), 12-bit quantization, 50 Hz power frequency crosstalk, yarn linear density unevenness coefficient of variation of 2.4%, machine speed fluctuation of ±0.8%, yarn feeder closed-loop steady-state error of ±2%, and 20% calibration mismatch of the twin shrinkage model itself. The yarn batch true values ​​are randomly selected based on an equivalent stiffness of 300 centine (RMS) and a coefficient of variation of 18%, and a friction coefficient of 0.24 (RMS) and a coefficient of variation of 25%. A total of 200 independent yarn batches were simulated, with each scheme addressing the same batch noise level. The results are the mean and standard deviation.

[0058] Comparison settings. In addition to the present invention, four other comparison groups are set up: Group A is the current practice, which does not perform any identification and directly copies the yarn feeding amount of the previous qualified work order of the same product; Group B is the modification method that is most easily thought of by those skilled in the art, that is, using the original upstream tension sensor of the machine to perform the same identification process; Group C adopts the sensor arrangement and two-stage detection of the present invention, but directly substitutes the identified parameters into the absolute retraction model without using anchor points; Groups D2 and D3 are partial solutions that remove the idle running section and remove the multi-stage detection, respectively.

[0059] Simulation results are as follows Figure 4 As shown, the specific values ​​for each scheme are as follows.

[0060] The current practice of Group A is as follows: the equivalent stiffness identification error is 14.4% ± 14.8%, the friction coefficient identification error is 23.7% ± 25.0%, the shrinkage rate prediction error is 1.41% ± 0.99%, the first-round pass rate is 32.0%, and the number of trial weavings required to pass the standard is 1.68% ± 0.47.

[0061] Group B uses the original sensors on the machine: the equivalent stiffness identification error is 47.5% ± 10.0%, the friction coefficient identification error is 56.0% ± 14.1%, the shrinkage rate prediction error is 1.52% ± 1.01 percentage points, the first-round pass rate is 28.0%, and the number of trial weavings required to reach the standard is 1.72% ± 0.45.

[0062] Group C identification without anchor points: equivalent stiffness identification error is 1.1% ± 0.9%, friction coefficient identification error is 4.7% ± 3.9%, shrinkage rate prediction error is 0.91% ± 0.30%, the first-round pass rate is 36.5%, and the number of trial weavings required to pass is 1.64% ± 0.48.

[0063] Group D2 (without the empty running section): the equivalent stiffness identification error is 1.1% ± 0.8%, the friction coefficient identification error is 23.7% ± 25.0%, the shrinkage rate prediction error is 0.78% ± 0.57%, the first round of compliance rate is 61.5%, and the number of trial weavings required to meet the standard is 1.39% ± 0.49.

[0064] Group D3 removed multiple detections: the equivalent stiffness identification error was 14.4% ± 14.8%, the friction coefficient identification error was 4.7% ± 3.9%, the shrinkage rate prediction error was 0.74% ± 0.59%, the first-round pass rate was 65.0%, and the number of trial weavings required to pass the standard was 1.35% ± 0.48.

[0065] The present invention has the following characteristics: the equivalent stiffness identification error is 1.1% ± 0.9%, the friction coefficient identification error is 4.7% ± 3.9%, the shrinkage rate prediction error is 0.31% ± 0.25%, the first-round pass rate is 94.5%, and the number of trial weavings required to pass the test is 1.05 ± 0.23.

[0066] Three points are evident from the above figures. First, the first-round pass rate of Group B (28%) is lower than that of Group A (32%), which did not perform any identification. This indicates that using the existing sensors on the machine is not only ineffective but also harmful, the reasons for which have been explained in S3 and the background art. Second, Group C's identification accuracy is exactly the same as that of this invention, but its first-round pass rate is only 36.5%, indicating that accurate parameters, if used in absolute terms in a mismatched model, do not bring benefits. Third, removing the idle run section or removing multiple detection stages reduces the pass rate to 61.5% and 65% respectively, demonstrating that the four elements of this invention constitute an inseparable whole, rather than a simple superposition of features.

[0067] Further simulation results regarding the mismatch in the twin models are as follows: at a mismatch of 5%, Group C achieves 100% and the present invention achieves 99.3%; at a mismatch of 12%, the results are 85.3% and 98.7%; at a mismatch of 20%, the results are 34.7% and 95.3%; at a mismatch of 35%, the results are 4% and 82.7%; and at a mismatch of 55%, the results are 0.7% and 64.7%. It is evident that when the model mismatch is very small, anchor points are not required, and the value of anchor points becomes apparent as the mismatch increases. This boundary is accurately recorded here.

[0068] The sensor selection, parameter ranges, and thresholds in the above embodiments are merely examples. Those skilled in the art can adjust them within the scope of this invention based on the machine model, yarn type, and fabric structure. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the protection scope of this invention.

Claims

1. A method for identifying yarn parameters in a knitted loop digital twin, characterized in that, This method is applied to a weft knitting machine. Each loop of the weft knitting machine is equipped with a yarn-storing active yarn feeder and an upstream tension sensor located at the outlet of the yarn-storing active yarn feeder. The upstream tension sensor constitutes the feedback point of the tension closed loop of the yarn-storing active yarn feeder. The weft knitting machine also has an electric yarn bending depth adjustment mechanism. The method includes the following steps: S1. On each loop, a downstream tension sensor is added between the yarn guide and the looping area of ​​the needle cylinder, downstream of the feedback point. The cumulative wrap angle of the yarn path between the downstream tension sensor and the feedback point is a known quantity; S2. Using the most recent production order that has passed inspection for the product to be knitted as a reference order, the yarn feed setting value of the reference order, the measured longitudinal density of the finished product after drying and relaxation, and the friction coefficient and equivalent stiffness of the reference batch obtained according to steps S3 and S4 of the reference order; S3. Idle running detection, causing the knitting needles to stop at a height that does not enter the yarn bending area, allowing the machine to idle at a production speed of no less than 30... In each row, signals from the downstream tension sensor are collected and their arithmetic mean is taken. The ratio of this arithmetic mean to the tension setting value of the yarn storage type active yarn feeder is divided by the natural logarithm and then by the cumulative wrap angle to obtain the friction coefficient of this batch of yarn; S4, multi-stage loop detection, so that the knitting needle enters the yarn bending area, and the electric yarn bending depth adjustment mechanism sequentially sets no less than two different yarn bending depths, with the difference between two adjacent yarn bending depths being 0.05 mm to 0.15 mm, and each stage weaving no less than 30 1. For each gear, the signal from the downstream tension sensor is segmented according to the looping cycle and synchronously averaged point by point. The difference between the maximum and minimum values ​​of the synchronously averaged waveform is taken as the looping tension peak-valley difference for that gear. A univariate linear fit is performed with the bending depth of each gear as the independent variable and the corresponding looping tension peak-valley difference as the dependent variable. The slope of the fitted line is multiplied by the current yarn feed setting value and then divided by the geometric sensitivity of the yarn demand of a single loop of the machine to the bending depth to obtain the equivalent stiffness of the yarn in this batch. S5. According to the proportional constant between the loop length and the finished product longitudinal density of the structure, and the yarn feed setting value of the reference work order and the measured finished product longitudinal density, the actual dry relaxation shrinkage rate of the reference work order is calculated. S6. The friction coefficient and equivalent stiffness of this batch of yarn, as well as the friction coefficient and equivalent stiffness of the reference batch, are substituted into the dry relaxation shrinkage model of the digital twin. The difference between the two outputs is taken as the shrinkage rate increment. The sum of the actual dry relaxation shrinkage rate of the reference work order and the shrinkage rate increment is taken as the predicted dry relaxation shrinkage rate of this batch of yarn. S7. In the digital twin, the yarn feeding amount setting value that makes the finished product longitudinal density equal to the target finished product longitudinal density is calculated in reverse using the predicted dry relaxation shrinkage rate value. The yarn feeding amount setting value is sent to the transmission ratio control interface of the yarn storage type active yarn feeder via the fieldbus. The yarn storage type active yarn feeder adjusts the yarn feeding amount of each loop according to the setting value and then puts it into formal production.

2. The method for identifying yarn parameters in a knitted loop digital twin according to claim 1, characterized in that, In step S4, the number of yarn bending depth levels is 3 to 6. Each level is symmetrically distributed around the yarn bending depth of the fabric to be woven. The yarn requirement of a single loop corresponding to the lowest yarn bending depth is not less than the current yarn feed setting. When the root mean square of the residual of the loop tension peak-valley difference of each level relative to the fitted straight line exceeds one-tenth of the range of the peak-valley difference, it is determined that the lowest level has entered the yarn relaxation zone and the level is removed and refitted.

3. The method for identifying yarn parameters in a knitted loop digital twin according to claim 1, characterized in that, In step S3, the height at which the knitting needle stops is taken as 1.5 mm to 3.0 mm above the lowest point of the yarn bending triangle. The arithmetic mean of the downstream tension sensor signal is calculated in the segment after the start of idling for 5 rows. At the same time, the reading of the upstream tension sensor is collected as the actual value of the tension setting value of the yarn storage type active yarn feeder. When the deviation between the reading and the tension setting value exceeds 5%, the tension closed loop of the yarn feeder is determined to be abnormal and the identification is terminated.

4. The method for identifying yarn parameters in a knitted loop digital twin according to claim 1, characterized in that, In step S4, the segmentation by looping cycle is triggered by each needle pulse of the spindle encoder. The number of sampling points intercepted in each looping cycle is the integer part of the quotient of the looping frequency of that path and the sampling rate of the downstream tension sensor. The point-by-point synchronous averaging takes the arithmetic mean of the sampled values ​​of each column on the same phase, and columns containing yarn splice signals are removed before synchronous averaging. The yarn splice signal is judged by the tension range of a single column exceeding three times the median of the tension range of all columns in that range.

5. The method for identifying yarn parameters in a knitted loop digital twin according to claim 1, characterized in that, The method further includes applicability determination, taking the coefficient of variation of the equivalent stiffness recorded in each production work order of the variety to be woven, skipping steps S3 to S6 when the coefficient of variation is less than 10%, and directly using the yarn feed setting value of the reference work order as the yarn feed setting value of step S7; and executing steps S3 to S6 when the coefficient of variation is not less than 10%.

6. The method for identifying yarn parameters in a knitted loop digital twin according to claim 1, characterized in that, Steps S3 and S4 are performed on all looping paths separately. The median of the friction coefficient and the median of the equivalent stiffness obtained from each path are taken as the friction coefficient and equivalent stiffness of this batch of yarn. Any looping path whose equivalent stiffness deviates from the median by more than 20 percent of the median is marked as an abnormal path. The abnormal path number and maintenance prompt are output through the human-machine terminal.

7. The method for identifying yarn parameters in a knitted loop digital twin according to claim 1, characterized in that, The method also includes anchor point updating. After the fabric is dried and relaxed on the machine in this production, the finished longitudinal density is measured. When the deviation between the measured finished longitudinal density and the target finished longitudinal density does not exceed one percent of the target finished longitudinal density, the yarn feed setting value, the measured finished longitudinal density, the friction coefficient and equivalent stiffness of the yarn in this batch are written into the work order database as a reference work order for the next production of this variety. When the deviation exceeds one percent, the yarn feed setting value is corrected according to the difference between the measured finished longitudinal density and the target finished longitudinal density, and the partial derivative of the finished longitudinal density with respect to the yarn feed value at the current yarn feed setting value of the digital twin. The yarn feed setting value is then sent through the transmission ratio control interface and the fabric is rewoven.

8. A yarn parameter identification system for knitted loop digital twins, characterized in that, include: A downstream tension sensor is installed between the yarn guide and the loop-forming area of ​​the needle cylinder in each loop-forming path of the weft knitting machine, and is located downstream of the feedback point of the tension closed loop of the yarn storage type active yarn feeder in that loop-forming path; a data acquisition unit is connected to the downstream tension sensor and the main shaft encoder of the weft knitting machine, and is used to synchronously acquire the signal of the downstream tension sensor with each needle pulse of the main shaft encoder as a trigger signal; a work order database is used to store the yarn feed setting value, measured finished product longitudinal density, yarn friction coefficient and equivalent yarn stiffness of each production work order for each variety; The parameter identification module, connected to the acquisition unit, is used to obtain the friction coefficient of the current batch of yarn by taking the natural logarithm of the ratio of the arithmetic mean of the downstream tension sensor signal to the tension setting value of the yarn storage type active yarn feeder during the idle running section where the knitting needle does not enter the yarn bending area, and then dividing it by the known cumulative wrap angle. It is also used to obtain the peak-to-valley difference of the synchronous average waveform of the loop forming period for at least two yarn bending depths after the knitting needle enters the yarn bending area, perform a univariate linear fitting on the yarn bending depth using the peak-to-valley difference, and multiply the slope of the fitted straight line by the current yarn feed setting value and then divide it by the geometric sensitivity to obtain the equivalent stiffness of the current batch of yarn. The twin server, connected to the parameter identification module and the work order database, has a built-in dry relaxation shrinkage model. This model is used to calculate the actual dry relaxation shrinkage rate of the reference work order from the yarn feed setting value and the measured finished product density. The friction coefficient and equivalent stiffness of the current batch of yarn and the reference batch are substituted into the dry relaxation shrinkage model, and the difference between the two outputs is taken as the shrinkage rate increment. The sum of the actual dry relaxation shrinkage rate and the shrinkage rate increment is used as the predicted dry relaxation shrinkage rate of the current batch of yarn, and this is used to calculate the yarn feed setting value that makes the finished product density equal to the target finished product density. A transmission ratio control interface, connected to both the twin server and the yarn storage type active yarn feeder, receives the yarn feed setting value and drives the yarn storage type active yarn feeder to adjust the yarn feed amount for each loop according to the setting value.

9. A yarn parameter identification system for knitted loop digital twins according to claim 8, characterized in that, The downstream tension sensor is a piezoelectric tension sensor with a range of 0 to 40 centinewtons, a sampling rate of not less than 20 kHz, and an analog-to-digital conversion bit depth of not less than 12 bits. It is installed on the yarn exit side of the yarn guide, and the wrap angle of the yarn on its detection wheel is 15 to 30 degrees. The system also includes a human-machine terminal connected to the twin server, used to display the friction coefficient and equivalent stiffness of each loop, the predicted dry relaxation shrinkage rate, the yarn feed setting value, and the number and maintenance prompts of abnormal loops.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the yarn parameter identification method for knitted loop digital twins as described in any one of claims 1 to 7.