Method and system for on-line monitoring of round knife die wear

Abstract

The application discloses a round knife die cutting die wear online monitoring method and system, relates to the die cutting processing quality monitoring technical field, and the method is characterized in that: a piezoelectric punching force sensor integrated in a round knife shaft is used to collect punching force signals stroke by stroke, a downstream direct depth measuring device is used to collect actual punching depth deviations every preset punching number of times to obtain a sparse depth deviation sequence; an indirect wear mapping output indirect wear estimation value is established based on the punching force signals; the sparse depth deviation sequence is used as an unbiased anchor point to online correct the indirect wear mapping, to obtain a corrected wear mapping adaptive to material batches and knife shaft temperature rise; the punching force signals are decoupled according to a round knife shaft angle to obtain a tooth number index, and the corrected wear mapping is combined to stroke by stroke output wear amount estimation values and predict residual life; the application decouples indirect observation drift by online correction of unbiased direct measurement, and realizes high-precision online estimation of stroke by stroke wear amount and residual life prediction.

Description

Technical Field

[0001] This invention relates to the field of die-cutting quality monitoring technology, specifically to a method and system for online monitoring of wear on circular die-cutting dies. Background Technology

[0002] Rotary die-cutting is one of the most widely used forming processes in the processing of substrates for consumer electronics accessories, optical films, foam, double-sided tape, and other products. Rotary die-cutting machines use a continuously rotating rotary die shaft to drive the die in a single, multi-stage forming process on the substrate. Compared to flatbed die-cutting, it offers significant advantages in efficiency and continuity, making it the mainstream equipment in high-speed die-cutting production lines. However, during long-term continuous cutting, the die teeth inevitably experience wear: with the increase in the cumulative number of cuts, macroscopic abrasion of the cutting edge causes an increase in the radius of the cutting corner, while microscopic cracking occurs at localized stress concentration points. This wear directly leads to deviations in the cutting depth. When the cutting depth deviation exceeds the process standard of 0.02mm, batch defects such as incomplete cutting, burrs, and substrate damage will occur. Because rotary die-cutting involves multiple teeth distributed along the circumference, and these teeth differ in manufacturing tolerances, assembly stress, and localized forces, their wear processes are not uniform. Often, individual teeth deteriorate first while others remain in the normal wear stage. Simply relying on overall wear assessment makes it difficult to pinpoint the specific deteriorated tooth, further complicating quality control. In engineering practice, traditional methods for addressing die wear mainly fall into two categories: one is forced die replacement based on a fixed number of cuts, but because the replacement threshold needs to be conservatively set based on the most easily deteriorated teeth, a large number of undeteriorated teeth are prematurely discarded, resulting in high replacement costs; the other is judging the timing of die replacement based on operational experience, but this is highly subjective and can easily lead to continued cutting even when the cutting edge is already out of tolerance, causing batch defects. Therefore, accurate online monitoring of die wear is of great significance for quality and cost control in rotary die-cutting production lines.

[0003] For online monitoring of tool wear, existing technologies generally employ indirect signal methods, which involve collecting physical signals related to wear and establishing a mapping relationship between the signal and the amount of wear. Chinese patent application CN104723171B discloses a tool wear monitoring method based on a composite signal of current and acoustic emission. This method uses the current signal from the turning spindle motor and the acoustic emission signal from the cutting edge to form a composite signal. It then adaptively extracts wear features using wavelet packet analysis, correlation analysis, and principal component analysis, and determines the degree of tool wear by analyzing the correlation between the current features and the initial wear state features. However, this approach relies on the assumption that the mapping relationship between the signal and the amount of wear is stable after a single calibration. When the batch of material being cut changes or the tool spindle temperature increases, the same amount of wear will produce significantly different signal responses. This causes the correlation judgment on which the approach relies to become systematically inaccurate after batch changes or long-term continuous operation, making it impossible to achieve high-precision online estimation of wear in continuous circular die-cutting production.

[0004] Chinese patent application CN107186547A discloses an online monitoring method for tool wear in batch CNC turning based on cutting power. This method analyzes the influence of tool wear and machining parameters on cutting power, establishes a regression model between cutting power, wear, and machining parameters using orthogonal experimental design and response surface methodology, and updates the cutting power threshold in real time based on this regression model to achieve online wear monitoring. However, this scheme relies solely on the single indirect signal of cutting power, and its regression model is only adjusted through threshold updates after establishment. It lacks a direct measurement and correction method for the drift in the signal-to-wear mapping caused by material batches and temperature rise. This makes it difficult to maintain the accuracy of the mapping relationship in circular die-cutting scenarios with significant differences in material batches, and it also cannot achieve online estimation of wear per tooth.

[0005] Chinese patent application CN110153801B discloses a method for identifying tool wear state based on multi-feature fusion. This method performs aggregated empirical mode decomposition on the original cutting force signal and vibration signal, extracting time-domain, frequency-domain, fuzzy entropy, approximate entropy, and sample entropy features, which are then fused. Singular value decomposition is then used to optimize the fused feature set, and a genetic algorithm is used to optimize the parameters of a least-squares support vector machine to achieve wear state identification. However, all signals used in this scheme are indirect signals, lacking direct measurements independent of material batches and working conditions as anchor points. The identification results are also affected by the drift in the signal-to-wear mapping. Furthermore, this scheme uses the mixed signal of the entire tool as the identification object, failing to decouple the mixed signal to each tooth on the circular die-cutting die, making it difficult to meet the requirements of tooth-by-tooth wear monitoring.

[0006] In summary, existing online tool wear monitoring technologies generally treat the mapping relationship between indirect signals and wear amount as a stable entity after a single calibration, neglecting the objective fact that this mapping relationship continuously drifts with batch differences in the cut material and the temperature rise of the cutter shaft in continuous circular die-cutting production. This leads to inaccurate wear estimation after batch changes or prolonged operation. Furthermore, existing technologies process the mixed signal of the entire tool, failing to achieve online estimation of wear amount and prediction of remaining life for each tooth of the circular die-cutting die. Therefore, how to achieve high-precision online estimation of tooth-by-tooth wear amount and prediction of remaining life using only sparse direct depth measurement in continuous circular die-cutting production where the mapping relationship between indirect signals and wear amount continuously drifts has become a critical technical problem urgently needing to be solved in this field. Summary of the Invention

[0007] To address the core technical bottleneck in existing online monitoring technologies for wear of circular die-cutting dies, where the mapping from indirect signals to wear amount continuously drifts with material batches and die shaft temperature rises, leading to inaccurate wear estimation, this invention provides an online monitoring method and system for wear of circular die-cutting dies. This method uses the downstream sparse actual punching depth deviation as an unbiased anchor point to perform online correction of the indirect wear mapping and decouples the mixed punching force signal according to tooth number to estimate wear amount tooth by tooth. Without altering the continuous production cycle of the circular die-cutting machine, it achieves high-precision online estimation of wear amount and prediction of remaining life for each tooth of the circular die-cutting die through the principle of online fusion correction of biased indirect observation and unbiased direct measurement.

[0008] The technical solution of this invention is as follows:

[0009] A method for online monitoring of wear on circular die-cutting dies includes the following steps:

[0010] Step S1: The punching force signal is collected sequentially by a piezoelectric punching force sensor integrated into the circular cutter shaft;

[0011] Step S2: The actual punching depth deviation of the substrate is collected every preset number of punches by the downstream direct depth measuring device to obtain a sparse depth deviation sequence.

[0012] Step S3: Establish an indirect wear mapping based on the punching force signal and output the indirect wear estimate;

[0013] Step S4: Using the sparse depth deviation sequence as an unbiased anchor point, the wear indirect mapping is corrected online to obtain a corrected wear mapping that adapts to material batch and tool shaft temperature rise.

[0014] Step S5: Obtain the tooth index based on the angle of the circular cutter shaft, decouple the punching force signal according to the tooth index, and output the wear estimate value tooth by tooth in combination with the corrected wear mapping.

[0015] Step S6: Output the remaining life prediction value based on the estimated wear amount per tooth.

[0016] This invention also provides an online monitoring system for wear of a circular die-cutting die, comprising: a cutting force acquisition module, disposed on the circular die shaft, for acquiring cutting force signals sequentially via a piezoelectric cutting force sensor; a depth deviation acquisition module, for acquiring the actual cutting depth deviation of the substrate every preset number of strokes via a downstream direct depth measurement device, to obtain a sparse depth deviation sequence; an indirect mapping module, for establishing a wear indirect mapping based on the cutting force signals and outputting an indirect wear estimate; an online correction module, for performing online correction of the wear indirect mapping using the sparse depth deviation sequence as an unbiased anchor point, to obtain a corrected wear mapping that adapts to material batches and die shaft temperature rise; a tooth-by-tooth decoupling module, for obtaining a tooth index based on the circular die shaft angle, decoupling the cutting force signals according to the tooth index, and outputting a wear estimate tooth-by-tooth based on the corrected wear mapping; and a remaining life prediction module, for outputting a remaining life prediction value based on the tooth-by-tooth wear estimate.

[0017] The beneficial effects of this invention are as follows:

[0018] First, by using the sparse downstream actual cutting depth deviation as an unbiased anchor point to perform online correction of the wear indirect mapping, high-precision wear estimation is maintained under conditions of material batch differences and cutter shaft temperature rise. The mechanism is as follows: the cutting force signal is essentially the output of multiple coupled factors such as wear, the mechanical properties of the cut material, the thermal state of the cutter shaft, and dynamic contact stiffness. Wear is only one of the hidden variables, thus the cutting force signal is a biased observation of wear. The actual cutting depth deviation, however, is directly determined by the actual cutting depth of the cutting edge, and is an approximately unbiased physical observation of wear. This invention uses the sparse but unbiased actual cutting depth deviation to perform online correction of the wear indirect mapping established by the dense but biased cutting force signal, which is equivalent to continuously guiding the biased observation with an unbiased anchor point, thereby eliminating mapping drift. Compared to methods based on a single calibration correlation judgment and methods relying solely on threshold updates, this solution can still stably control the cutting depth deviation within the process standard of 0.02mm after batch changes, fundamentally overcoming the inaccuracy problem caused by mapping drift in existing technologies.

[0019] Secondly, by obtaining the tooth index based on the angle of the circular cutter shaft and decoupling the cutting force signal according to the tooth index, tooth-by-tooth estimation of the wear amount of each tooth of the circular cutter die is achieved. The mechanism is as follows: the circular cutter die has multiple teeth distributed along its circumference; each tooth cuts sequentially with each revolution of the circular cutter shaft. The collected cutting force signal is a mixed signal contributed by multiple teeth in time sequence. This invention uses the tooth index to phase-align the cutting force signal and performs sparse blind separation, decomposing the mixed signal into individual teeth, thereby obtaining the single-tooth cutting force component corresponding to each tooth. Compared to the approach of using the mixed signal of the entire tool as the identification object, this solution can distinguish the inconsistent wear processes between teeth, providing a basis for accurately locating deteriorated teeth and arranging targeted tool replacement.

[0020] Third, the synergistic effect of unbiased direct measurement online calibration and tooth-by-tooth signal decoupling generates a synergistic effect that exceeds the sum of the individual effects of each feature. Specifically: unbiased direct measurement online calibration alone can only correct the overall mapping deviation, while tooth-by-tooth signal decoupling alone will amplify the drift error proportionally to each tooth under mapping drift conditions. Neither of these can independently achieve high-precision tooth-by-tooth estimation. However, this invention first eliminates mapping drift with unbiased anchor point calibration, and then performs tooth-by-tooth decoupling on the calibrated mapping. This makes the tooth-by-tooth wear estimation have both unbiasedness and tooth-level resolution, thereby leveraging downstream sparse and low-cost direct measurements to drive high-frequency high-precision tooth-by-tooth estimation, achieving a nonlinear gain greater than 2.

[0021] Fourth, by detecting abrupt changes in the estimated wear amount of a single tooth as micro-chipping of the cutting edge and fusing the two mechanisms of gradual erosion and abrupt chipping in a dual-mechanism remaining life model, accurate prediction of the remaining life of each tooth is achieved. The mechanism is as follows: macroscopic erosion manifests as a slow gradual change in the estimated wear amount of each tooth, while micro-chipping of the cutting edge manifests as a sudden step in the estimated wear amount of a single tooth accompanied by a sharp increase in high-frequency acoustic emission energy. The two have distinctly different time-dynamic characteristics. This invention uses chipping as a hard constraint trigger term and erosion gradual change as a soft attenuation term for fusion, upgrading the remaining life prediction model from a purely data-driven statistical model to a model with physical constraints. Compared with the approach of extrapolating only a single gradual trend, it has better prediction accuracy and generalization ability. Attached Figure Description

[0022] Figure 1 This is a flowchart of the online monitoring method for wear of circular die cutting molds according to the present invention;

[0023] Figure 2 This is an architecture diagram of the online wear monitoring system for circular die-cutting dies of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0025] This embodiment uses the die-cutting of a 0.15mm thick double-sided adhesive tape substrate on a high-speed rotary die-cutting production line as an example to describe in detail the online monitoring method for die wear in rotary die-cutting according to the present invention. The die mounted on the rotary die shaft of this production line has 24 teeth evenly distributed along its circumference. The rated punching frequency of the rotary die shaft is 300 punches per minute, and the required punching depth deviation tolerance is 0.02mm. Figure 1 As shown, this method includes steps S1 to S6.

[0026] Step S1: The punching force signal is collected sequentially by a piezoelectric punching force sensor integrated into the circular cutter shaft.

[0027] A piezoelectric cutting force sensor is pre-tightly installed between the bearing housing of the circular cutter shaft and the frame. This allows the cutting force applied to the substrate by the die in each stroke to be transmitted axially through the circular cutter shaft and act on the force-bearing surface of the piezoelectric cutting force sensor. The range of the piezoelectric cutting force sensor is selected to be 0 to 50 kN, with a natural frequency of not less than 30 kHz and a sensitivity of not less than 4 pC / N, to ensure distortion-free capture of the rapid change in cutting force over time in each stroke. During the operation of the circular cutter shaft, the piezoelectric cutting force sensor converts the cutting force on its force-bearing surface into an electric charge signal. This charge signal is converted and amplified into a voltage signal by a charge amplifier, and then converted into a digitized cutting force signal through analog-to-digital conversion. In this embodiment, the sampling frequency of the analog-to-digital conversion is set to 50 kHz, meaning that the cutting force is densely sampled within each stroke, thus completely recording the change in cutting force over time for that stroke.

[0028] To correlate the acquired cutting force signal with the rotational phase of the circular cutter shaft, an incremental angle encoder is coaxially mounted on the shaft. The encoder outputs a tooth synchronization pulse for each tooth pitch rotated and a full-cycle reference pulse for each revolution. The cutting force signal acquisition module, triggered by the tooth synchronization pulse from the angle encoder, divides the continuously acquired cutting force signal into segments representing individual strokes. Each segment corresponds to the complete cutting process of one tooth on the circular cutter die in one stroke. In this embodiment, three temporal features are extracted from each segment: the peak cutting force, the slope of the rising cutting force segment, and the duration of the cutting force action, which are used as inputs for subsequent processing. The peak cutting force reflects the maximum cutting action of the cutting edge on the substrate, the slope of the rising cutting force segment reflects the sharpness of the cutting edge, and the duration of the cutting force action reflects the contact duration between the cutting edge and the substrate. These three temporal features collectively characterize the cutting force signal for that stroke.

[0029] As the cutting edge wears, the radius of the cutting edge fillet gradually increases, and the cutting edge changes from sharp to blunt. Consequently, the required cutting force for the same cutting depth increases, and the slope of the rising segment of the cutting force decreases. Therefore, the aforementioned time-domain characteristics of the cutting force signal carry information about the cutting edge wear. However, since the cutting force signal is also affected by factors such as the hardness of the material being cut and the temperature of the cutting shaft, the cutting force signal alone cannot directly provide an accurate estimate of the wear. An accurate estimate of the wear must be obtained through indirect wear mapping and online correction in subsequent steps.

[0030] Step S2: The actual punching depth deviation of the substrate is collected every preset number of punches by the downstream direct depth measuring device to obtain a sparse depth deviation sequence.

[0031] A direct depth measuring device is installed downstream of the rotary die-cutting machine at the station where the printed material is output after die-cutting. In this embodiment, the direct depth measuring device uses a confocal displacement sensor with a measurement resolution of not less than 0.001 mm and a measurement repeatability of not less than 0.002 mm. The confocal displacement sensor measures the actual punching depth of the die-cut printed material in a non-contact manner. The actual punching depth deviation is obtained by subtracting the measured punching depth from the target punching depth set by the process.

[0032] Considering that measuring the depth of the substrate for every die-cut would significantly reduce production cycle time and increase measurement costs, this embodiment uses a direct depth measurement device to perform a sampling inspection of the substrate every preset number of die-cuts. In this embodiment, the preset number of die-cuts is 10, meaning that the actual die-cutting depth deviation is measured every 10 die-cuts. Therefore, during the continuous operation of the rotary cutter shaft, the direct depth measurement device outputs the actual die-cutting depth deviation in a sparse manner. These actual die-cutting depth deviations, arranged in die-cutting order, are marked according to the cumulative number of die-cuts corresponding to their measurement time, thus obtaining a sparse depth deviation sequence. Each depth deviation value in the sparse depth deviation sequence is directly derived from the physical measurement of the die-cut substrate, unaffected by the batch size of the cut material or the temperature rise of the cutter shaft. It represents an approximately unbiased physical observation of wear, and therefore can serve as an unbiased anchor point for subsequent online correction.

[0033] It should be noted that the preset number of strokes needs to be balanced between measurement cost and timely correction: if the preset number of strokes is too high, the sparse depth deviation sequence will be too sparse, resulting in a lag in the correction of mapping drift; if the preset number of strokes is too low, it will increase measurement cost and affect production cycle time. In this embodiment, the preset number of strokes is set to 10, which can control the impact of sampling quantity on production cycle time within an acceptable range while ensuring timely correction.

[0034] Step S3: Establish an indirect wear mapping based on the punching force signal and output the indirect wear estimate.

[0035] Wear indirect mapping is used to characterize the relationship between the time-domain features of the punching force signal and the wear amount. This embodiment uses linear regression to establish the wear indirect mapping, taking the three time-domain features extracted in step S1—the peak punching force, the slope of the punching force rise segment, and the width of the punching force application time—as independent variables, and the wear amount as the dependent variable. The calibration parameters of the wear indirect mapping are the regression coefficients and constant terms before each time-domain feature. When the circular die-cutting die is first used, the initial wear amount, derived from the time-domain features of the punching force signal of the new die under a known material batch and the corresponding measured punching depth deviation, is used as a sample to initially calibrate the calibration parameters of the wear indirect mapping, thus obtaining the initial wear indirect mapping.

[0036] During the operation of the circular cutter shaft, for each stroke, the time-domain characteristics of the punching force signal of that stroke are organized into a punching force signal feature vector and substituted into the current wear indirect mapping to obtain the indirect wear estimate for that stroke. The indirect wear estimate is the inner product of the punching force signal feature vector and the calibration parameter vector of the wear indirect mapping, that is, it gives the estimate of the current wear amount in the form of a linear weighted sum of each time-domain feature. Since the punching force signal is collected in each stroke, the wear indirect mapping can continuously output the indirect wear estimate at the same high frequency as the punching frequency, forming a dense observation of the wear amount, which cannot be obtained by relying solely on sparse sampling depth. However, since the wear indirect mapping is based on the punching force signal, and the punching force signal is a biased observation of the wear amount, the indirect wear estimate will deviate from the actual wear amount after material batch changes or cutter shaft temperature rise. Therefore, in step S4, the wear indirect mapping needs to be corrected online using the sparse depth deviation sequence as an unbiased anchor point. Therefore, the overall idea of ​​this invention is to provide high-frequency wear tracking using dense but biased indirect wear estimates, and to provide low-frequency wear anchoring using sparse but unbiased actual punching depth deviations. These two methods are then fused through online correction to form a wear estimate that is both dense and unbiased. The regression calculation for establishing the indirect wear mapping and outputting the indirect wear estimate is itself a conventional process in the art, and this embodiment will not elaborate on the details of its regression solution.

[0037] Step S4: Using the sparse depth deviation sequence as an unbiased anchor point, the wear indirect mapping is corrected online to obtain a corrected wear mapping that adapts to material batches and tool shaft temperature rise.

[0038] This step is one of the core steps of this invention. The basic idea is to use the sparse but unbiased actual punching depth deviation as an unbiased anchor point. Whenever a new depth deviation value is added to the sparse depth deviation sequence, the measured wear amount obtained by back-deriving the depth deviation value is compared with the indirect wear estimate value in the same punch interval. The deviation between the two is used to correct the calibration parameters of the wear indirect mapping, so that the indirect wear estimate value output by the corrected wear indirect mapping at the anchor point is close to the measured wear amount, thereby eliminating the mapping drift caused by material batch and cutter shaft temperature rise.

[0039] Specifically, whenever the direct depth measuring device is in the... When a new actual punching depth deviation is added at a sampling inspection time, the actual punching depth deviation is first used to inversely deduce the measured wear amount based on the geometric relationship between the actual cutting depth and the punching depth deviation, and recorded as the anchor point wear amount. The indirect wear estimate for the corresponding punch at that sampling inspection time is then taken, and the difference between the two is the mapping deviation at that anchor point. Driven by this mapping deviation, the calibration parameter vector of the wear indirect mapping is corrected online using the following formula:

[0040] ,in: For the first The calibration parameter vector of the wear indirect mapping after online correction at each sampling time is a column vector with dimension [missing information]. The value range is such that all components are real numbers with dimensionless units, and it is obtained by recursion calculation from this formula, representing the corrected wear mapping at the th order. Regression coefficients and constant terms at each sampling time point; For the first The calibration parameter vector indirectly mapped to the wear at each sampling time is a column vector with dimension [missing information]. The value range is that all components are real numbers, and the unit is dimensionless. It is obtained by recursion from this formula at the previous sampling time or by initial calibration when it is first used on the machine. It represents the regression coefficient and constant term before correction. For the first The adaptive correction gain at each sampling time is a scalar, and its value ranges from... The unit is dimensionless and is determined by the adaptive correction gain calculation formula described in step S4, representing the correction strength of the anchor point to the calibration parameters in this step. For the first The anchor point wear amount, calculated by back-calculating the actual punching depth deviation at each sampling time, is a scalar quantity with a value range of [value missing]. The unit is mm, which is derived from the actual punching depth deviation by reverse calculation based on the cutting edge geometry, and represents the actual wear of the die at the time of the sampling inspection. For the first The punching force signal feature vector corresponding to each sampling time is a column vector with dimension . The first three components are the peak value of the punching force, the slope of the punching force rising segment, and the width of the punching force action time, respectively. The fourth component is always 1 to correspond to the constant term. The value range is that each component is a non-negative real number, and the units are kN, kN / ms, ms and dimensionless, respectively. It is extracted from the punching force signal collected in step S1 and represents the punching force signal of this punch. for transpose, superscript This represents the transpose operation of a vector. This is the estimated indirect wear value at that stroke before correction; This represents the sampling time sequence number, which is a positive integer and its value ranges from 1 to 2. The unit is dimensionless and determined by the order of sampling, representing the nth sampling correction. (Left side of the equation) The components are regression coefficients, and... Same dimension; in the second term on the right side of the equation, The unit is mm. Dimensionless Units of each component The units of corresponding components are reciprocals of each other, such that the units of each component of this term are equal to... The corresponding components are consistent, and the units on both sides are strictly consistent.

[0041] The calibration parameter vector obtained after the above online correction The determined indirect wear mapping is the corrected wear mapping. Since the correction uses the unbiased actual punching depth deviation as the anchor point, the corrected wear mapping can adaptively adjust with changes in material batch and cutter shaft temperature rise, always maintaining the indirect wear estimate close to the actual wear amount.

[0042] The determination of the adaptive correction gain in step S4.

[0043] If a fixed correction gain is used for each anchor point during continuous operation, the fixed gain will be unable to track the mapping drift in time during the rapid temperature rise of the cutter shaft, resulting in correction lag. Therefore, this embodiment employs an adaptive correction gain that varies with the cumulative number of punching operations and increases with the increase in cutter shaft temperature, enabling the corrected wear mapping to track the mapping drift during continuous operation. The adaptive correction gain is calculated using the following formula:

[0044] ,in: For the first The adaptive correction gain at each sampling time point is defined as in the aforementioned online correction formula, and is a scalar with a value range of [value missing]. The unit is dimensionless and is calculated by this formula, representing the correction strength of the anchor point to the calibration parameters. The reference gain coefficient is denoted as , and is a scalar with a range of values ​​of . The unit is dimensionless and is determined by the initial calibration. In this embodiment, it is taken as 0.4, which represents the reference level of the adaptive correction gain. Let be the temperature rise sensitivity coefficient, and be a scalar with a value range of . The unit is 1 / ℃, which is determined by the hot test calibration of the tool shaft. In this embodiment, it is taken as 0.05 to characterize the sensitivity of the adaptive correction gain to the temperature rise of the tool shaft. For the first The tool shaft temperature rise at each sampling time, i.e., the increase in the current tool shaft temperature relative to the initial tool shaft temperature, is a scalar quantity with a value range of [value missing]. The unit is ℃, which is calculated from the temperature signal of the cutter shaft output by the temperature sensor installed on the cutter shaft, and characterizes the current thermal state of the cutter shaft; For the natural constant An exponential function with base 0; For the first The cumulative number of punching operations at each sampling time is a positive integer, and its value ranges from 1 to 2. The unit is dimensionless and is obtained by accumulating the number of punches, representing the cumulative number of punches since the machine was put into operation; Let be the cumulative impulse attenuation constant, which is a positive integer and has a range of values ​​of . The unit is dimensionless and determined by empirical values ​​of die life. In this embodiment, it is taken as... This characterizes how quickly the adaptive correction gain decays with the cumulative number of punches; Let be the lower bound of minimum gain, and be a scalar with a range of values. The unit is dimensionless and determined by the steady-state correction requirements; in this embodiment, it is taken as 0.05, representing the lower limit of the adaptive correction gain to ensure that the correction capability is still present in the steady-state stage. In the first term on the right side of the equation... Dimensionless The unit is (1 / ℃) multiplied by ℃, which equals dimensionless. middle The ratio of dimensionless numbers is therefore dimensionless, the whole is dimensionless, and the addition of dimensionless numbers... Still dimensionless, compared to the left side The dimensionless units are consistent, and the units on both sides are strictly consistent.

[0045] As can be seen from the above formula, when the temperature of the cutter shaft rises... When the value is large, the adaptive correction gain increases accordingly, thereby accelerating the tracking of mapping drift; when the cumulative number of punches... As the exponent increases, the exponent term causes the adaptive correction gain to gradually decrease and tend towards the lower limit of the minimum gain. This ensures the stability of the correction after the mapping reaches a steady state, avoiding fluctuations caused by overcorrection.

[0046] Step S5: Obtain the tooth index based on the angle of the circular cutter shaft, decouple the punching force signal according to the tooth index, and output the wear estimate value tooth by tooth in combination with the corrected wear mapping.

[0047] This step is another core step of the present invention. The circular die-cutting die has 24 teeth distributed along its circumference. Each revolution of the circular die shaft results in the 24 teeth sequentially cutting. The collected cutting force signal is a mixed signal composed of the superposition of the cutting contributions from each tooth. Furthermore, due to the partial overlap of adjacent teeth in time sequence and the slight eccentricity of the circular die shaft, simply extracting the signal according to a time window cannot accurately separate the cutting force components of each tooth. Therefore, in this embodiment, the tooth index is first obtained based on the angle of the circular die shaft, and then phase alignment and sparse blind separation are performed on the cutting force signal to obtain the single-tooth cutting force component corresponding to each tooth.

[0048] Specifically, using the integer reference pulse output by the angle encoder as a reference, the angle encoder tooth synchronization pulses within an integer cycle are sequentially numbered from 1 to 24, thus obtaining the tooth number index corresponding to each tooth. Based on the tooth number index, the punching force segments corresponding to the same tooth number within multiple consecutive integer cycles are stacked in phase alignment, that is, punching force segments of teeth with the same number within different integer cycles are aligned to the same phase starting point. For the phase-aligned multi-stroke punching force signal, sparse blind separation is performed using the following formula to solve for the single-tooth punching force component of each tooth:

[0049] ,in: The optimal solution for the single-tooth punching force component matrix obtained by sparse blind separation is a matrix with dimension O(n). , Number of teeth The number of sampling points for the single-tooth punching force component is given, with each element being a non-negative real number and the unit being kN. It is obtained by optimizing the solution of this formula and represents the single-tooth punching force component that corresponds one-to-one with each tooth. The independent variable represents the value that minimizes the objective function within the curly braces. ; Let be the component matrix of the single-tooth punching force to be solved, and be a matrix with dimension . The value range is that each element is a non-negative real number, the unit is kN, and it is an optimization variable that represents the punching force components of each tooth to be separated. The phase-aligned multi-stroke punching force signal matrix is ​​given by , which is a matrix with dimension . , The integer number of cycles involved in the separation is defined as a range where each element is a non-negative real number, with the unit being kN. It is obtained by phase alignment as described in step S5 and represents the observed mixed punching force signal. Let be a mixture matrix, and be a matrix with dimension . The value range is that all elements are non-negative real numbers, and the unit is dimensionless. It is determined by the overlap relationship between adjacent teeth in time and the eccentricity of the circular cutter shaft, and characterizes the contribution weight of each tooth punching force component to each integer observation signal. Representation matrix With matrix Matrix multiplication; Denotes the Frobenius norm of a matrix. That is, the square of the norm is used to measure the reconstruction error; Let be the sparse regularization coefficient, and let be a scalar with a value range of . The unit is kN, determined by the separation effect calibration. In this embodiment, it is taken as 0.1, which characterizes the constraint strength on the sparsity of the single-tooth punching force component. If the value is too large, it will excessively suppress the effective components; if the value is too small, it will not suppress aliasing. The summation variable is the tooth index, which is a positive integer and takes values ​​in the range of 1. The unit is dimensionless, representing the first... One tooth; The number of teeth in a circular die-cutting die is a positive integer, and in this embodiment it is taken as 24. The unit is dimensionless and is determined by the die structure. It represents the total number of teeth distributed along the circumference of the die. The component matrix of single-tooth punching force The row, is a row vector with dimension . The value range is such that all elements are non-negative real numbers, and the unit is kN, representing the th The single-tooth punching force component of each tooth. Representing vectors The norm, which is the sum of the absolute values ​​of the elements of a vector, is used to characterize sparsity. The first term of the objective function... middle and The units are all in kN, and the square of the Frobenius norm of their difference is in kN squared; the second term middle The unit is kN. The unit of is kN, and the unit of the product is the square of kN, which is consistent with the unit of the first term. The two terms can be added together, and the objective function is dimensionally consistent.

[0050] After solving the above equation to obtain the single-tooth punching force components corresponding to each tooth, for each tooth, three time-domain features are extracted from its single-tooth punching force components: the peak value of the punching force, the slope of the punching force rise segment, and the width of the punching force action time. These features constitute the punching force signal feature vector of that tooth. Substituting this vector into the corrected wear mapping obtained in step S4 yields the estimated tooth-by-tooth wear amount. Since the corrected wear mapping has been corrected online by unbiased anchor points to eliminate mapping drift, the obtained estimated tooth-by-tooth wear amount possesses both unbiasedness and tooth-level resolution, accurately reflecting the inconsistent wear process between teeth.

[0051] Step S6: Output the remaining life prediction value based on the estimated wear amount per tooth.

[0052] This step predicts the remaining life of each tooth based on the tooth-by-tooth wear estimate obtained in step S5. Considering that tooth wear includes two mechanisms: gradual macro-erosion and abrupt micro-fracture of the cutting edge, this embodiment first performs inter-tooth sequence analysis on the tooth-by-tooth wear estimate to detect micro-fracture of the cutting edge, and then uses a dual-mechanism remaining life model to fuse the two mechanisms and output the remaining life prediction value.

[0053] The detection of micro-chipping of the cutting edge is as follows: For each tooth, the estimated wear amount of each tooth in several adjacent strokes is used to construct a wear sequence. The step value of the estimated wear amount of the current stroke relative to the average value of its preceding sliding window is calculated. When this step value exceeds a preset abrupt change threshold, and the high-frequency energy in the 20kHz to 100kHz band output by the acoustic emission sensor in the same stroke exceeds a preset energy threshold, it is determined that the tooth has experienced micro-chipping in the current stroke. The step value of the wear sequence is calculated using the following formula:

[0054] ,in: For the first The tooth in the first The wear sequence step of each stroke is a scalar quantity with a real number range and a unit of mm. It is calculated by this formula and represents the degree of abrupt change in the current wear of the tooth relative to its recent average. For the first The tooth in the first The estimated wear amount per tooth per stroke is a scalar value, and its range is [value missing]. The unit is mm, obtained from step S5, and represents the current wear of the tooth; The tooth index is a positive integer with a value range of 1. The unit is dimensionless, and its definition is the same as in step S5, representing the first... One tooth; The sequence number is a positive integer, and its value range is [value range missing]. The unit is dimensionless, representing the first... Strike rate; The length of the sliding window is a positive integer, and its value range is... The unit is dimensionless and is determined by the mutation detection sensitivity calibration. In this embodiment, it is taken as 20, which represents the number of preceding impulses used to calculate the recent mean. Let be the summation variable within the sliding window, a positive integer, and its value range be 1. The unit is dimensionless, representing the preceding number. One stroke; For the first The tooth in the first The estimated wear amount per tooth per stroke is a scalar value, and its range is [value missing]. The unit is mm, obtained from step S5, and represents the wear amount of the tooth in the preceding strokes. (Left side of equation) The unit is mm, and the first term on the right side of the equation The unit is mm, and the second item is... The arithmetic mean of the estimated wear is still in mm, and the difference between the two is in mm. The units on both sides are strictly consistent.

[0055] when When the frequency energy of the same pulse exceeds a preset mutation threshold and the frequency energy of the same pulse exceeds a preset energy threshold, the first pulse is determined to be... The tooth in the first Micro-chipping of the cutting edge occurs during each stroke. In this embodiment, the preset mutation threshold is set to 0.01 mm, and the preset energy threshold is determined by adding three times the standard deviation to the high-frequency energy baseline during the empty cutting of a new die. The high-frequency energy output by the acoustic emission sensor is obtained by integrating the power spectral density of the acoustic emission signal in the 20 kHz to 100 kHz frequency band, which is a conventional calculation in this field and will not be elaborated here.

[0056] Based on the detection of whether micro-chipping has occurred on the cutting edge of each tooth, a dual-mechanism remaining life model is used to integrate both gradual wear and abrupt chipping mechanisms to output the predicted remaining life value for each tooth. The dual-mechanism remaining life model is calculated using the following formula:

[0057] ,in: For the first The predicted remaining life of each tooth is a scalar value, and its range is [value range missing]. The unit is strokes, which are calculated by this formula and represent the number of times the tooth can still cut from the current stroke to the wear limit. This is the tooth index, defined in step S5, and is a positive integer with a value range of [value range missing]. The unit is dimensionless, representing the first... One tooth; For the first The hard constraint trigger term for tooth chipping is a scalar with a value range of [value range missing]. The unit is dimensionless. When the tooth is determined to have micro-chipping of the cutting edge, it is 1; otherwise, it is 0. It is obtained by the micro-chipping detection of the cutting edge as described in step S6, and it represents whether the tooth has triggered the chipping hard constraint. Here, represents the upper limit of wear, and is a scalar with a range of values. The unit is mm, which is determined by the wear amount corresponding to the punching depth deviation tolerance. In this embodiment, it is taken as 0.2, which represents the maximum allowable wear amount of the cutting edge. For the first The tooth in the first The estimated wear amount per tooth per stroke is defined as in the aforementioned formula for the wear sequence step amount, and is a scalar with a range of values. The unit is mm, representing the current wear of the tooth; For the first The wear rate of each tooth is a scalar quantity, with a range of values ​​of [value missing]. The unit is mm / stroke, which is obtained by the linear fitting slope of the recent tooth wear estimate and characterizes the rate of macro-erosion development of the tooth. Let be the lower limit of the minimum gradual rate of change, and be a scalar with a range of values. The unit is mm / stroke. To prevent the denominator from being zero, this embodiment uses... This represents the lower limit of the wear gradient rate; This indicates selecting the larger of two numbers; Let be the critical remaining life at breakout, and be a scalar with a value range of . The unit is the number of strokes, which is determined by the number of safe strokes that can continue cutting after the chipping. In this embodiment, it is taken as 10, which indicates that the tool should be replaced within the remaining life after the micro-chipping of the cutting edge occurs. Let be the upper limit of the predicted remaining useful life, and be a scalar with a range of values. The unit is strokes, determined by the total die life, and represents the upper limit of the predicted remaining life value. In the first term on the right side of the equation... Dimensionless, molecule The unit is mm, and the denominator is... The unit is mm / stroke. Dividing the numerator by the denominator (unit: mm) equals the number of strokes. In the second term... Dimensionless The unit of the product is strokes, and the unit of the product is strokes; both units are strokes, as shown on the left. The units of each stroke are consistent, and the units on both sides are strictly consistent.

[0058] As can be seen from the above formula, when the first When no micro-chipping occurs on the cutting edge of any tooth Taking 0, the remaining life prediction value is obtained by dividing the difference between the wear upper limit and the current wear amount by the wear gradient rate, reflecting a linear extrapolation of the macroscopic wear gradient mechanism; when the th When a tooth experiences micro-chipping of its cutting edge Setting it to 1 forces the remaining lifetime prediction to be set to a smaller critical remaining lifetime for breakage, reflecting the hard constraint of the breakage abrupt change mechanism. The two are connected through... By achieving integration, the remaining life prediction can reflect both the gradual trend of macroscopic wear and provide timely warnings when micro-fracture occurs on the cutting edge, thereby improving the accuracy and safety of the remaining life prediction.

[0059] Thus, steps S1 to S6 of the method embodiment have been fully executed. For each tooth on the circular die-cutting die, the estimated value of tooth-by-tooth wear after online correction with unbiased anchor points and decoupling of tooth number, as well as the predicted value of remaining life based on the fusion of dual mechanisms, can be obtained.

[0060] Based on the above method embodiments, the present invention may further include the following extensions. First, after obtaining the estimated wear amount per tooth in step S5 and the predicted remaining life in step S6, the estimated wear amount per tooth and the predicted remaining life are rendered into a heat map of tooth number-wear amount-remaining life with tooth number as the horizontal axis and cumulative punching count as the vertical axis, and output. This allows operators to intuitively grasp the wear distribution and remaining life distribution of each tooth and quickly locate deteriorated teeth. Second, the peak order periods of the production schedule are obtained, and the tool change timing is given before the peak order periods based on the predicted remaining life, thereby avoiding line stoppages due to tool changes during peak order periods. The above heat map rendering and tool change timing are applications based on the output results of the aforementioned core steps, and their specific implementation is a conventional method in the field.

[0061] This invention also provides an online monitoring system for wear of circular die-cutting dies, used to implement the online monitoring method for wear of circular die-cutting dies described in the above-described method embodiments. For example... Figure 2 As shown, the system includes a punching force acquisition module, a depth deviation acquisition module, an indirect mapping module, an online correction module, a tooth-by-tooth decoupling module, and a remaining life prediction module. Each module corresponds one-to-one with steps S1 to S6 of the method embodiment.

[0062] The punching force acquisition module is installed on the circular cutter shaft and is used to acquire punching force signals stroke by stroke using a piezoelectric punching force sensor. The specific implementation of this module corresponds to step S1 of the method embodiment: a piezoelectric punching force sensor is pre-tightly installed between the bearing housing of the circular cutter shaft and the frame. The piezoelectric punching force sensor converts the punching force of each stroke into an electrical charge signal. This signal is then amplified and converted from analog to digital to obtain a digitized punching force signal. Triggered by the tooth synchronization pulse output from a coaxially mounted angle encoder, the punching force signal is divided into punching force segments, each representing a single stroke. The punching force signal acquired by this module serves as the input to the subsequent indirect mapping module and the tooth-by-tooth decoupling module.

[0063] The depth deviation acquisition module is used to collect the actual punching depth deviation of the substrate at preset punching times via a downstream direct depth measurement device, obtaining a sparse depth deviation sequence. The specific implementation of this module corresponds to step S2 in the method embodiment: a confocal displacement sensor is installed at the substrate output station downstream of the rotary die-cutting machine. The actual punching depth of the die-cut substrate is sampled and measured at preset punching times in a non-contact manner. The actual punching depth deviation is obtained by subtracting it from the target punching depth, and then marked according to the cumulative punching times corresponding to the measurement time, forming a sparse depth deviation sequence. The sparse depth deviation sequence obtained by this module is the source of the unbiased anchor point for the online correction module.

[0064] The indirect mapping module is used to establish an indirect wear mapping based on the punching force signal and output an indirect wear estimate. The specific implementation of this module corresponds to step S3 in the method embodiment: an indirect wear mapping is established using three time-domain features of the punching force signal—the peak value of the punching force, the slope of the punching force rise segment, and the width of the punching force application time—as independent variables, and the wear amount as the dependent variable. In each punch, the feature vector of the punching force signal from that punch is substituted into the current indirect wear mapping, and the indirect wear estimate for that punch is output. The indirect wear estimate output by this module is the object to be corrected by the online correction module.

[0065] The online calibration module is used to perform online calibration of the wear indirect mapping using a sparse depth deviation sequence as an unbiased anchor point, obtaining a calibrated wear mapping that adapts to material batches and cutter shaft temperature rise. The specific implementation of this module corresponds to step S4 in the method embodiment: whenever the depth deviation acquisition module adds a new actual punching depth deviation, it is back-calculated as the anchor point wear amount, and the mapping deviation is calculated by comparing it with the indirect wear estimate for the same punching interval. The calibration parameters of the wear indirect mapping are then updated with an adaptive calibration gain that increases with the cumulative number of punchings and the cutter shaft temperature rise, resulting in the calibrated wear mapping. The calibrated wear mapping output by this module serves as the basis for the tooth-by-tooth wear estimation by the tooth-by-tooth decoupling module.

[0066] The tooth-by-tooth decoupling module is used to obtain the tooth index based on the circular cutter shaft angle, decouple the punching force signal according to the tooth index, and output the wear estimate tooth by tooth in conjunction with the corrected wear map. The specific implementation of this module corresponds to step S5 of the method embodiment: using the integer reference pulse of the angle encoder as a reference, each tooth is numbered to obtain the tooth index; the punching force segments of the same tooth within multiple consecutive integer cycles are phase-aligned; sparse blind separation is performed on the phase-aligned multi-punch punching force signal to obtain the single-tooth punching force component; then, time-domain features are extracted from the single-tooth punching force component of each tooth and substituted into the corrected wear map to output the wear estimate tooth by tooth. The tooth-by-tooth wear estimate output by this module is the input to the remaining life prediction module.

[0067] The remaining life prediction module is used to output the remaining life prediction value based on the tooth-by-tooth wear estimation value. The specific implementation of this module corresponds to step S6 of the method embodiment: performing inter-tooth sequence analysis on the tooth-by-tooth wear estimation value, combining the 20kHz to 100kHz high-frequency energy output from the acoustic emission sensor to detect micro-fractures on the cutting edge, and fusion outputting the remaining life prediction value of each tooth using a dual-mechanism remaining life model with fracture as a hard constraint trigger term and wear gradient trend as a soft attenuation term.

[0068] The aforementioned modules can be implemented by programs deployed on industrial control computers or edge computing devices. These modules sequentially transmit punching force signals, sparse depth deviation sequences, indirect wear estimates, corrected wear maps, tooth-by-tooth wear estimates, and remaining life predictions via data interfaces, thereby collaboratively completing online monitoring of die wear in circular die-cutting. In a system embodiment, a human-machine interface module may be further included to render and output a heatmap of tooth number, wear amount, and remaining life, and to provide tool replacement timing based on the remaining life prediction and peak order periods in the production schedule. Its specific implementation is consistent with the corresponding extensions in the method embodiment and will not be elaborated here.

[0069] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for online monitoring of wear of circular die-cutting dies, characterized in that, include: The punching force signal is collected sequentially by a piezoelectric punching force sensor integrated into the circular cutter shaft; The actual punching depth deviation of the substrate is collected by the downstream direct depth measuring device at preset punching times to obtain a sparse depth deviation sequence. Based on the shearing force signal, an indirect wear mapping is established, and an indirect wear estimate is output. Using the sparse depth deviation sequence as an unbiased anchor point, the wear indirect mapping is corrected online to obtain a corrected wear mapping that adapts to material batch and tool shaft temperature rise. Based on the angle of the circular cutter shaft, the tooth index is obtained, the punching force signal is decoupled according to the tooth index, and combined with the corrected wear mapping, the tooth-by-tooth wear estimate is output one-to-one with each tooth number; The remaining life prediction value is output based on the estimated wear value per tooth.

2. The method for online monitoring of wear of circular die-cutting dies according to claim 1, characterized in that, The online correction of the wear indirect mapping includes: whenever a new depth deviation value is added to the sparse depth deviation sequence, the depth deviation value is used as a reference value, and the deviation is calculated with the indirect wear estimate in the same stroke interval, and the calibration parameters of the wear indirect mapping are updated with the deviation.

3. The method for online monitoring of wear of circular die-cutting dies according to claim 2, characterized in that, The calibration parameters for updating the wear indirect mapping adopt an adaptive correction gain that varies with the cumulative number of punching operations. The adaptive correction gain increases with the increase of the tool shaft temperature rise, so that the corrected wear mapping tracks the mapping drift during continuous operation.

4. The method for online monitoring of wear of circular die-cutting dies according to claim 3, characterized in that, The step of decoupling the punching force signal according to the tooth number index includes: aligning the punching force signal in phase with the tooth number index, performing sparse blind separation on the phase-aligned multi-punch punching force signal, and obtaining single-tooth punching force components corresponding one-to-one with each tooth number.

5. The method for online monitoring of wear of circular die-cutting dies according to claim 4, characterized in that, Also includes: The tooth wear estimate obtained from the single-tooth punching force component is analyzed by inter-tooth sequence analysis. When the tooth wear estimate of a certain tooth number shows a step change exceeding the preset abrupt change threshold relative to its adjacent punch, the tooth number is determined to have undergone micro-chipping of the cutting edge by combining the high-frequency energy of 20kHz to 100kHz output by the acoustic emission sensor.

6. The method for online monitoring of wear of circular die-cutting dies according to claim 5, characterized in that, The output remaining life prediction value adopts a dual-mechanism remaining life model. The dual-mechanism remaining life model uses the micro-fracture of the cutting edge as a hard constraint trigger term and the gradual trend of the estimated wear amount per tooth as a soft attenuation term. The two are combined to output the remaining life prediction value of each tooth number.

7. The method for online monitoring of wear of circular die-cutting dies according to claim 1, characterized in that, Also includes: Acoustic emission signals are synchronously acquired by an acoustic emission sensor integrated into the circular cutter shaft. The high-frequency energy of the acoustic emission signals in the frequency band of 20kHz to 100kHz is extracted, and the high-frequency energy is used as input to participate in the calculation of the estimated value of tooth wear.

8. The method for online monitoring of wear of circular die-cutting dies according to claim 1, characterized in that, Also includes: Using tooth number as the horizontal axis and cumulative punching count as the vertical axis, the estimated wear amount per tooth and the predicted remaining life are rendered into a tooth number-wear amount-remaining life heatmap and output.

9. The method for online monitoring of wear of circular die-cutting dies according to claim 1, characterized in that, Also includes: Obtain the peak order periods of the production schedule and provide tool change timings before the peak order periods based on the remaining life prediction value.

10. An online monitoring system for wear of circular die-cutting dies, used to implement the online monitoring method for wear of circular die-cutting dies as described in any one of claims 1-9, characterized in that, include: The punching force acquisition module is located on the circular cutter shaft and is configured to acquire punching force signals sequentially through a piezoelectric punching force sensor. The depth deviation acquisition module is configured to acquire the actual punching depth deviation of the substrate every preset number of punches via a downstream direct depth measurement device, and obtain a sparse depth deviation sequence. The indirect mapping module is configured to establish an indirect wear mapping based on the punching force signal and output an indirect wear estimate. The online correction module is configured to perform online correction of the wear indirect mapping using the sparse depth deviation sequence as an unbiased anchor point, so as to obtain a corrected wear mapping that adapts to the material batch and the temperature rise of the tool shaft. The tooth-by-tooth decoupling module is configured to obtain the tooth number index based on the angle of the circular cutter shaft, decouple the punching force signal according to the tooth number index, and combine the corrected wear mapping to output the tooth-by-tooth wear estimate value corresponding to each tooth number. The remaining life prediction module is configured to output a remaining life prediction value based on the estimated wear amount of each tooth.

Images