A self-learning compensation method for registration error in electronic shaft printing of tobacco materials
By linking registration error, historical compensation output, and post-compensation feedback to the same material coordinate window during the printing process of cigarette paper and performing validity determination, the problem of unstable compensation parameters caused by mismatch of control objects and abnormal samples is solved, and more stable registration control is achieved.
Patent Information
- Application Number
- CN202611142940.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-30
- Publication Date
- 2026-11-10
AI Technical Summary
In the continuous printing process of cigarette paper, the existing technology has difficulty effectively solving the problem of unstable compensation parameters caused by mismatch of control objects and abnormal samples entering the online parameter update process.
By associating the registration error, historical compensation output, and post-compensation feedback to the same material coordinate window and determining their validity, the residual information after nominal compensation is calculated, the compensation parameters corresponding to the current working condition are updated, electronic axis compensation commands are generated, and the update results are verified based on the post-compensation feedback.
This reduces the risk of compensation feedback being misattributed, avoids compensation parameter drift and over-intervention, and improves the stability and accuracy of registration control.
Smart Images

Figure CN123061008A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of industrial automatic control technology, specifically to a self-learning compensation method for registration error in electronic shaft printing of tobacco materials. Background Technology
[0002] In continuous roll printing, especially in multi-color gravure printing, hot stamping, or composite printing of cigarette rolls, electronic axis synchronous control is often required to achieve phase coordination, speed fine-tuning, or trajectory correction between multiple logical printing control channels. The conventional approach typically involves: first, the registration detection unit acquires the current registration error; then, based on this error value, the local controller generates control commands according to a fixed compensation ratio (such as compensation, feedback compensation, model feedforward compensation, and periodic learning compensation) and sends them to the electronic axis synchronous control interface. This approach can complete basic registration adjustment when the material batch is relatively stable, the operating speed does not vary significantly, the detection signal is continuous, and the compensation output does not approach its limit.
[0003] However, in the field conditions of continuous printing of cigarette paper, there are often material transmission delays and control cycle delays in the location of error detection, the location where compensation commands take effect, and the location of feedback after compensation. For example, if the local controller directly updates the parameters based on the registration error at a certain sampling moment and the subsequently acquired compensation feedback, it may incorrectly attribute the feedback results of different controlled material segments to the current error sample. Furthermore, when low confidence is found in image or color mark detection, local dirt spots cause error jumps, the electronic axis compensation output approaches the limit, or material batch changes or speed increases / decreases occur, the relevant samples may not represent a stable control response that can be learned. If these samples are still added to the compensation parameter update link, it may cause phenomena such as compensation parameter drift, reverse update, or excessive intervention. Summary of the Invention
[0004] (a) Technical problems to be solved To address the aforementioned issues of unstable compensation parameters caused by mismatched control objects and abnormal samples entering the online parameter update process, this invention provides a self-learning compensation method for registration errors in electronic shaft printing of tobacco materials. First, the registration error, historical compensation output, and post-compensation feedback are associated with the same controlled material segment or material coordinate window. Then, the associated samples are evaluated for validity. When a sample meets the validity criteria, the residual information after nominal compensation is calculated, and the compensation parameters corresponding to the current operating condition are updated based on this residual information. Subsequently, an electronic shaft compensation command is generated, and the compensation parameters are submitted, frozen, or rolled back based on the post-compensation feedback.
[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: A self-learning compensation method for registration error of electronic shaft printing of tobacco materials, applied to a local controller, includes: acquiring control samples of the logic control channel, wherein the control samples include registration error, operating status, historical compensation output and post-compensation feedback; The material coordinate window is determined based on the control sample, and the registration error, historical compensation output, and post-compensation feedback are associated with the same material coordinate window. The validity of the associated control samples is determined. If the control samples are deemed invalid, the control samples are isolated and the compensation parameter updates are frozen. When the condition is deemed valid, the residual information of the control sample after nominal compensation is calculated, the compensation parameters corresponding to the working condition are updated according to the residual information, and the electronic shaft compensation command is output. Based on the feedback verification and update results after compensation, the compensation parameters are submitted, frozen, or rolled back.
[0006] Furthermore, determining the material coordinate window includes: generating material coordinates from the running speed and control cycle in the electronic axis running state, and determining the window length according to at least one of the material running length, registration detection resolution, and error repetition cycle; The material coordinate window carries subsequent correlation of registration errors, historical compensation outputs, and post-compensation feedback.
[0007] Furthermore, linking the registration error, historical compensation output, and post-compensation feedback to the same material coordinate window includes: For each logic control channel, the equivalent delay is obtained. The equivalent delay is determined by at least one of step response, cross-correlation analysis, and production line calibration. The historical compensation output and the compensation feedback are converted to the material coordinate window where the registration error is located according to the equivalent delay.
[0008] Furthermore, the validity determination includes at least two of the following: detection confidence determination, error jump determination, compensation saturation determination, operating condition determination, and residual trend determination. When all selected criteria are met, a valid flag is generated and the process proceeds to residual information calculation; when any selected criterion is not met, an invalid flag is generated, the update of compensation parameters is blocked, and the process switches to abnormal sample processing.
[0009] Furthermore, when generating an invalid flag, the local controller writes the control sample into the abnormal sample cache unit, maintains the previous stable compensation parameter, and calls the nominal compensation output; When generating a valid flag, the local controller allows the control sample to enter the residual information calculation and prohibits the abnormal sample cache unit from writing to the control sample and triggering the freeze action.
[0010] Furthermore, the validity determination of the associated control samples includes: when the control samples contain detection confidence, comparing the detection confidence with the confidence threshold; When the control sample does not contain the detection confidence level, a pseudo confidence level is generated based on error continuity, missing labels, consistency of adjacent logic control channels, and consistency of residual trends, and the pseudo confidence level is compared with the confidence level threshold.
[0011] A registration error compensation system for electronic shaft printing of cigarette rolls includes: a registration error detection interface, an electronic shaft status interface, a material work order interface, a local controller, an electronic shaft synchronization control interface, a parameter storage unit, and an abnormal sample cache unit; The local controller includes a data acquisition module, a material window association module, a sample gating module, a residual update module, a compensation fusion module, and a feedback verification module; The data acquisition module acquires the registration error, electronic axis operating status, historical compensation output, and post-compensation feedback, and generates control samples. The material window association module associates the registration error, historical compensation output, and post-compensation feedback with the same material coordinate window based on the material coordinates and equivalent delay. The sample gating module determines the validity of the associated control samples. The residual update module calculates the nominally compensated residual information based on the valid control samples and updates the compensation parameters. The compensation fusion module generates electronic axis compensation commands. The feedback verification module submits, freezes, or rolls back the compensation parameters based on the post-compensation feedback.
[0012] An electronic device includes a processor, a memory, a communication bus, a network interface, and a field input / output interface; The memory stores the computer program; the processor is connected to the memory, network interface, and field input / output interface via a communication bus, and executes the computer program to achieve: Obtain control samples during the electronic shaft printing process of cigarette paper; based on material coordinates and equivalent delay, associate registration error, historical compensation output, and post-compensation feedback to the same material coordinate window; The validity of the associated control samples is determined; the residual information after nominal compensation is calculated based on the control samples that pass the validity determination, and the compensation parameters corresponding to the current operating condition are updated. Generate electronic axis compensation instructions, and submit, freeze, or roll back the compensation parameters based on the feedback after compensation.
[0013] A computer-readable storage medium storing a computer program; when executed by a processor, the computer program implements a method for compensating for registration errors in electronic shaft printing of cigarette paper, comprising: Acquire control samples; correlate registration error, historical compensation output, and post-compensation feedback based on the material coordinate window; determine the validity of the correlated control samples; When the control sample meets the validity criteria, calculate the residual information after nominal compensation and update the compensation parameters. Generate electronic axis compensation instructions, and submit, freeze, or roll back the compensation parameters based on the feedback after compensation.
[0014] (III) Beneficial Effects This invention provides a self-learning compensation method for registration error in electronic shaft printing of tobacco materials, which has the following beneficial effects: By linking registration error, historical compensation output, and post-compensation feedback to the same material coordinate window, the data used for parameter updates are directed to the same controlled material segment, avoiding direct splicing of data from different sampling times and thus reducing the risk of incorrect attribution of compensation feedback. A validity check is set before residual updates to block samples with low confidence, error jumps, compensation saturation, or unstable operating conditions, and invalid samples are isolated and compensation parameter updates are frozen, making it difficult for abnormal inputs to contaminate long-term compensation parameters.
[0015] The compensation parameters corresponding to the current working condition are updated with the residual information after nominal compensation, so that the basic compensation law is separated from the residual deviation caused by material batch, speed range, and channel response, thereby reducing the divergence risk caused by learning directly based on the original registration error.
[0016] By integrating residual compensation, nominal compensation, and feedback compensation, and learning intervention strength, amplitude limiting, slope, and dead zone constraints before output, the compensation commands are applied gradually within the executable boundaries of the electronic axis, reducing compensation abrupt changes and reverse oscillations. Post-compensation feedback is used to verify candidate compensation parameters, and parameters are submitted, frozen, or rolled back based on the verification results. This creates a traceable version loop for online updates, preventing a single abnormal learning result from having a long-term impact on the registration control of subsequent roll sections. Attached Figure Description
[0017] Figure 1 This is a schematic diagram illustrating an application scenario for the electronic bushing quasi-control of continuous printing of cigarette paper provided in an embodiment of the present invention. Figure 2 This is a logical structure diagram of the registration error compensation system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the data structure of the control sample and material window provided in an embodiment of the present invention; Figure 4 The main flowchart of the registration error compensation method provided in the embodiments of the present invention is shown below; Figure 5 A timing diagram of the material window attribution for error, compensation output, and post-compensation feedback provided in an embodiment of the present invention; Figure 6 This is a sample validity gating state machine diagram provided in an embodiment of the present invention; Figure 7 This is a flowchart of the nominal compensation residual extraction process provided in an embodiment of the present invention; Figure 8 A data structure diagram of the operating condition curve parameter library provided in the embodiments of the present invention; Figure 9 This is a flowchart of the constrained residual parameter update and compensation fusion process provided in an embodiment of the present invention; Figure 10 A flowchart for the gating threshold and verification window calibration provided in this embodiment of the invention; Figure 11 This is a state machine diagram for verifying anomaly degradation and recovery provided in an embodiment of the present invention; Figure 12 A flowchart illustrating an alternative embodiment for generating pseudo-confidence scores provided by this invention. Figure 13 Flowchart of an alternative embodiment for generating weak condition labels provided by the present invention; Figure 14 Flowchart of an alternative embodiment for low-computational-power lookup table residual compensation provided by this invention; Figure 15 A virtual device module diagram provided for embodiments of the present invention; Figure 16 This is a hardware topology diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] Please see Figures 1-16 This invention provides a self-learning compensation method for registration error in electronic shaft printing of tobacco materials.
[0020] like Figure 1 As shown, in this embodiment, the system generally includes: a registration error detection interface 10; an electronic axis status interface 20; a material work order interface 30; a local controller 40; an electronic axis synchronization control interface 50; a parameter storage unit 60; and an abnormal sample cache unit 70.
[0021] in, Figure 1The printing line, roll material, and detection station shown in the dashed box on the left are only for illustrating the data source and control boundaries under continuous printing conditions of cigarette rolls, and are not limited to specific printing press structures, roll material transmission structures, detector installation positions, or any mechanical assembly relationships. The registration error detection interface 10 sends the registration error information corresponding to the logic control channel to the local controller 40. The electronic axis status interface 20 provides the local controller 40 with operating speed, operating status, historical compensation output, and feedback information after compensation. The material work order interface 30 provides the local controller 40 with material batch, work order number, or equivalent operating condition label. Based on the above inputs, the local controller 40 generates compensation commands, which are then sent through the electronic axis synchronization control interface 50 for phase offset, speed fine-tuning, or compensation trajectory updates. The parameter storage unit 60 saves the compensation parameter version corresponding to the current operating condition, and the abnormal sample cache unit 70 saves samples that have not passed the validity judgment to prevent them from entering the online parameter update chain. Thus, a closed-loop control path is formed: input acquisition - window attribution - gating judgment - residual update - compensation output - feedback verification - abnormal degradation.
[0022] In this embodiment, cigarette paper refers to a strip-shaped substrate used for continuous printing of cigarette materials, which can be, for example, tipping paper, cork paper, or similar cigarette paper bases or composite substrates. This term is used to limit the controlled object in this embodiment to continuously operating strip-shaped materials, rather than single sheets of paper or ordinary back-end management objects.
[0023] In this embodiment, electronic axis synchronization control is a program control method in which multiple logic printed control channels coordinate phase, speed, or compensation trajectory using a virtual synchronization reference. The concept of electronic axis does not represent any physical mechanical axis structure, nor does it impose any restrictions on the installation position and assembly relationship of any mechanical transmission components. In this embodiment, the electronic axis synchronization control interface 50 receives the compensation command output by the local controller 40 and converts it into the corresponding phase offset, speed fine-tuning, or compensation trajectory update of the logic control channel.
[0024] In this embodiment, the logic control channel refers to the i-th registration control object defined by the control program. A logic control channel may correspond to a specific color sequence, a specific registration detection object, or a specific group of control variables; however, this terminology does not limit the spatial location of the physical color group, sensor, or actuator. Different logic control channels may have different equivalent delays, different compensation response characteristics, and different operating parameters.
[0025] In this embodiment, the registration error is the offset between a certain logic control channel and the set reference during the k-th control cycle. The offset can be output by the registration error detection interface 10 based on the results of color mark detection, image detection, or equivalent detection. The registration error should have a direction and a unit, such as a length unit, the number of encoded pulses, or an equivalent unit within the control system. The registration error is not a scrap rate, quality score, or manual evaluation result.
[0026] In this embodiment, the control sample is a unified data record encapsulated by the local controller 40 within a control cycle. A control sample includes at least the sampling time, logic control channel, registration error, electronic axis operating status, historical compensation output, and post-compensation feedback. For example, the control sample may also include detection confidence level, material batch, work order number, speed range, acceleration / deceleration status, or operating condition label. The fields in the control sample may come from different interfaces, but before entering the material window association module 42, the data acquisition module 41 unifies the timestamp, field unit, and missing flag to avoid data from different refresh cycles being directly concatenated into the same learning sample.
[0027] In this embodiment, the material coordinates are coordinate quantities that map the sampling time to the roll running length or equivalent material position. The material coordinates can be obtained by integrating the running speed provided by the electronic axis status interface 20, or by using a virtual synchronization reference or equivalent length counting. The material coordinates are neither the printing plate circumference reference coordinates nor the equipment space coordinates.
[0028] In this embodiment, the material coordinate window is a controlled material segment window obtained by segmenting the material coordinates. Through this window, the local controller 40 can establish a correlation between the registration error occurring on a certain material segment, the historical compensation output acting on that material segment, and the feedback state of that material segment after compensation. The material coordinate window is used to solve the object mismatch problem between error, compensation, and feedback. In this embodiment, the material coordinate window can be simply referred to as the material window, which emphasizes the attribution relationship of the same controlled material segment, rather than an image recognition window or a user interface window.
[0029] The equivalent delay in this embodiment is the time delay between the detection of registration error, control calculation, issuance of compensation command and collection of feedback after compensation within a certain logic control channel, or the delay calculated by the distance difference of material movement. This equivalent delay can be obtained through step response testing, cross-correlation analysis or production line calibration. In this case, the local controller 40 can map the error sample, compensation output and feedback result to the same material coordinate window.
[0030] In this embodiment, validity determination refers to the gating determination performed by the local controller 40 before updating the compensation parameters to determine whether the control sample is suitable for participating in online learning. Validity determination may include one or more of the following: detection confidence determination, error jump determination, compensation saturation determination, operating condition determination, and residual trend determination.
[0031] The validity determination result is used to decide whether the sample enters the residual update path or enters the abnormal sample cache unit 70 and triggers a freeze or downgrade action.
[0032] In this embodiment, nominal compensation refers to the basic compensation amount generated based on existing basic control laws. Nominal compensation can be formed by one or more of feedback compensation, feedforward compensation, gain scheduling, or lookup table compensation. Nominal compensation is used to undertake the basic control function under stable operating conditions, rather than representing the entire compensation result of this embodiment.
[0033] In this implementation, the residual is the feedback deviation that cannot be explained after the nominal compensation. The residual is used to update the compensation corresponding to the current operating condition. The residual is not simply a repetition of the original registration error; after attribution and validity determination in the material coordinate window, it represents the basis for updating parameters in areas not covered by the nominal compensation.
[0034] In some embodiments, the residual can be understood as the deviation between the compensated feedback and the nominal compensation expected response. This deviation is processed by feature vectorization and then enters the residual update module 45 instead of directly using the original registration error as the learning output.
[0035] In this embodiment, the operating condition curve is a record of the compensation parameter version corresponding to the current material, operating speed, logic control channel, residual trend, or work order label. The operating condition curve may include information such as compensation parameters, the number of valid samples, residual improvement indicators, version identifier, and rollback pointer. The operating condition curve is used to bind the compensation parameters to the on-site operating conditions, preventing samples from different batches, different speed ranges, or different channels from being mixed together to update the same parameter. The operating condition curve can be saved by the parameter storage unit 60. After the candidate version passes feedback verification, it becomes the current version. If the verification fails, it can be restored to the previous stable version according to the rollback pointer.
[0036] In this embodiment, freezing refers to stopping the update of the current compensation parameters when the local controller 40 detects an invalid sample, abnormal feedback, or parameter verification failure. Rollback, on the other hand, refers to the local controller 40 restoring the current candidate compensation parameters to the previous stable version stored in the parameter storage unit 60. Both freezing and rollback are control actions, not just alarm displays.
[0037] like Figure 1 As shown, in this embodiment, each part of the registration error compensation system is located on the control side of the continuous printing production line for cigarette paper: the registration error detection interface 10, the electronic axis status interface 20, and the material work order interface 30 are all connected to the local controller 40 via communication links. The local controller 40 can be an industrial controller, an edge computing unit, or an embedded control unit running periodic control tasks. The registration error detection interface 10 can receive error outputs from an image acquisition device, a color mark detector, or an equivalent registration detection program; the electronic axis status interface 20 can receive speed, status, historical compensation, and feedback readback from the electronic axis synchronization control program; the material work order interface 30 can receive material tags from the field work order system or material batch records. The compensation output terminal of the local controller 40 is connected to the electronic axis synchronization control interface 50, and the electronic axis synchronization control interface 50 writes compensation commands into the phase offset field, speed fine-tuning field, or compensation trajectory field of the electronic axis control program.
[0038] In this way, the registration error detection interface 10, the electronic shaft status interface 20, and the material work order interface 30 are no longer independent data sources, but are encapsulated in a control sample inside the local controller 40, thereby reducing the risk of sample mismatch caused by inconsistent refresh frequencies between different data sources.
[0039] exist Figure 1 In the diagram, the electronic axis status interface 20 can be abbreviated as "status interface," and it refers to the same interface as the electronic axis status interface 20 in the main text. The bidirectional connection between the parameter storage unit 60 and the local controller 40 indicates that the local controller 40 can both read verified operating condition curves and write candidate parameter versions or rollback states; the connection between the abnormal sample cache unit 70 and the local controller 40 indicates that invalid samples, abnormal triggering reasons, and parameters before freezing can be recorded for subsequent recovery verification or abnormal tracing.
[0040] like Figure 2As shown, the local controller 40 internally includes a data acquisition module 41, a material window association module 42, a sample gating module 43, a nominal compensation module 44, a residual update module 45, a compensation fusion module 46, and a feedback verification module 47. The data acquisition module 41 connects to the registration error detection interface 10, the electronic axis status interface 20, and the material work order interface 30, and is used to read the registration error, operating status, historical compensation output, and post-compensation feedback within a control cycle. The material window association module 42 receives the control sample encapsulated by the data acquisition module 41 and generates a material coordinate window. The sample gating module 43 performs validity determination before the residual update module 45. The nominal compensation module 44 outputs the basic compensation amount, and the residual update module 45 updates the compensation parameters based on the nominally compensated residual. The compensation fusion module 46 merges the basic compensation, feedback compensation, and residual compensation and outputs the result to the electronic axis synchronization control interface 50. The feedback verification module 47 reads the post-compensation feedback and decides whether to submit, freeze, or roll back the operating condition curve in the parameter storage unit 60. The modules mentioned above can be different software tasks, function units, or microservices in the local controller 40, or they can be implemented by different code segments in the same control program. What they have in common is that they all receive the results of the preceding sequence and output the subsequent actions within the same control closed loop, thereby avoiding the floating of functional modules.
[0041] Combination Figure 2 It can be seen that the sample gating module 43 is set after the material window association module 42 and before the residual update module 45, so that samples that have not completed material window attribution will not directly enter the residual update module 45. At the same time, after the compensation fusion module 46 outputs to the electronic axis synchronization control interface 50, the feedback verification module 47 reads the compensated feedback and updates the parameter storage unit 60, thus forming a closed loop consisting of forward compensation output and backward feedback verification.
[0042] like Figure 3 As shown, in step S101, the data acquisition module 41 reads the alignment error of the i-th logic control channel in the k-th control cycle from the alignment error detection interface 10, reads the running speed, operating status, historical compensation output, and post-compensation feedback in the corresponding control cycle from the electronic shaft status interface 20, and reads the material label or work order label from the material work order interface 30. The data acquisition module 41 writes these fields into the control sample.
[0043] For example, a control sample may include information such as sampling time, logic control channel, registration error, detection confidence, running speed, operating status, historical compensation output, post-compensation feedback, and material label. If a certain field is not refreshed in the current control cycle, the data acquisition module 41 will not replace it with the default valid value. Instead, a missing flag will be written to this field and passed to the sample gating module 43. After that, the residual update module 45 will not mistakenly regard the missing data as a real learnable sample, thus reducing the risk of incorrect updates to compensation parameters due to unknown input sources.
[0044] Figure 3 The fields such as time, channel, error, confidence, speed, operating condition, historical compensation, compensation feedback, and material label in the control sample are exemplary fields; the material window association module 42 maps the control sample to the operating speed, sampling time, or equivalent length count in the control sample. Figure 3 The materials window on the right. If the attached drawing is mistakenly labeled with a different drawing number, this paragraph shall prevail. Figure 3 For the purpose of standardization and consistency, revisions were made.
[0045] like Figure 3 As shown, in one implementation, when the material work order interface 30 does not provide a material batch field, the data acquisition module 41 can generate a weak operating condition label using the operating speed range, mean error, error variance, residual autocorrelation, and compensation response gain, and write this weak operating condition label as a substitute field for the material label into the control sample. In this alternative path, the weak operating condition label does not represent the actual material grade, but is used to distinguish operating segments that differ in control response. This alternative implementation allows older production lines without a complete work order system to still use the operating condition differentiation mechanism, reducing the risk of mixing samples from different operating segments to update the same compensation parameter.
[0046] like Figure 4 and Figure 5 As shown, in step S102, the material window association module 42 generates material coordinates based on the running speed and control cycle in the control sample. For example, the material coordinates can be generated using the following formula: ; in, Indicates the first The material coordinates corresponding to each control cycle can be in millimeters, meters, or equivalent length units within the control system. This indicates the material coordinates corresponding to the previous control cycle; Indicates the first Sampling time of each control cycle; This indicates the material travel speed read by the electronic axis status interface 20; This indicates the control cycle of the local controller 40; Indicates the sequence number of the logic control channel. Indicates the control cycle number.
[0047] This formula is only used to illustrate an exemplary method for generating material coordinates. The material window association module 42 can also generate material coordinates using virtual synchronization reference counting or the cumulative value of the length pulse inside the electronic axis. After generating the material coordinates, the material window association module 42 divides the continuous material coordinates into multiple material coordinate windows. The window length can be determined based on the material running length within a control cycle, the registration detection resolution, or the error repetition cycle. With the material coordinate window, errors, compensation, and feedback are no longer mechanically bound according to a simple time sequence, but are attributed according to their corresponding controlled material segments, thereby reducing object mismatch caused by high-speed roll material transmission delay.
[0048] Combination Figure 5 As shown in windows 1, 2, and 3, the material coordinate window does not require the registration error, historical compensation output, and post-compensation feedback to occur at the same sampling time. Instead, it requires that the three correspond to the same controlled material segment after material coordinate conversion. Therefore, the local controller 40 can avoid mistaking feedback from subsequent material segments as the compensation response of the current material segment.
[0049] In step S103, as Figure 5 As shown: The material window association module 42 also associates the registration error, historical compensation output and compensation feedback to the same material coordinate window based on the equivalent delay corresponding to the logic control channel.
[0050] For example, the first The equivalent delay of each logic control channel can be expressed as: This delay represents the equivalent material travel distance from the detection of the registration error of a certain material segment to the feedback acquisition of the compensation effect of that material segment. This can be obtained through step compensation response, cross-correlation analysis of error and feedback, or production line control cycle calibration. Specifically, the material window association module 42 will be located in the material window. The registration error within, and after The historical compensation output and post-compensation feedback applied to the same material segment after conversion are bound together; among them, Indicates the material coordinate window number. Indicates the first A material coordinate window.
[0051] Additionally, subscripts are used in this embodiment. , and It is only used to distinguish logic control channels, control cycles, and material coordinate windows, and does not limit the number of entity color groups, sensors, or actuators; the above material coordinates, speeds, delays, and window lengths can be expressed using equivalent units within the field control system, but the units should remain consistent within the attribution link of the same material window.
[0052] Therefore, the residual obtained by the residual update module 45 comes from the same controlled material segment, rather than the error splicing between different material segments, thereby reducing the risk of writing error feedback into the parameter storage unit 60.
[0053] like Figure 5 As shown, in one embodiment, if the electronic axis status interface 20 cannot directly provide a speed value that can be used for speed integration, the material window association module 42 can read the pulse accumulation value of the electronic axis virtual synchronization reference and use the pulse accumulation increment of adjacent control cycles as the equivalent material coordinate increment. If the production line only provides a material running segment count, the material window association module 42 can also map the count to a material coordinate window according to a preset length ratio.
[0054] The underlying mechanism of the above alternative methods is the same: to transform samples in the time domain into controlled material segment samples in order to avoid updating parameters simply according to the sampling time.
[0055] like Figure 6 In step S104, the sample gating module 43 will determine the validity of the associated control samples. The determination may include at least two of the following: detection confidence determination, error jump determination, compensation saturation determination, operating condition determination, and residual trend determination. For example, when the registration error detection interface 10 outputs a detection confidence level, the sample gating module 43 compares the detection confidence level with a confidence threshold. This confidence threshold is determined by the low quantile of the detection confidence level distribution within the stable production segment and is not arbitrarily set. If the detection confidence level is less than this threshold, the sample gating module 43 writes the sample to the abnormal sample cache unit 70 and blocks the sample from entering the residual update module 45. The instruction compensation fusion module 46 then compensates using a conservative nominal method. This feature reduces the risk of parameter contamination caused by local blemishes, recognition failures, or discontinuous detection by blocking low-confidence detection samples from entering the parameter update chain.
[0056] like Figure 6The sample gating module 43 shown has several states: learnable, frozen, verified, submitted, rolled back, and abnormal. When it is in the learnable state, it means that the sample has passed the validity judgment and can enter the residual update path. In the frozen state, it means that the update operation of the current parameter is paused. In the verified state, it is necessary to compensate before feedback confirmation. In the submitted state, it means that the candidate parameter can be written into the parameter storage unit 60. In the rolled back state, it restores the previous stable version, and in case of an anomaly, it records the sample or feedback into the abnormal sample cache unit 70.
[0057] like Figure 6 As shown, the sample gating module 43 can also determine the registration error jump. For example, the sample gating module 43 can compare the difference between the current registration error and the registration error of the previous control cycle to see if it exceeds a jump threshold determined by the error difference distribution of the stable production section. If a jump exists, the compensation parameter update operation is not performed on this sample, but it is recorded in the abnormal sample cache unit 70. This is not a simple alarm behavior, but directly prevents the parameter writing action of the residual update module 45, thereby avoiding the learning of local printing defects or detection misidentifications into long-term compensation rules. The jump threshold can be determined based on the median absolute deviation or quantile of the error difference in the stable production section.
[0058] like Figure 6 As shown, the sample gating module 43 can also read the state of the compensation output limiting from the electronic axis synchronization control interface 50. When the historical compensation output is close to the maximum phase offset, speed fine-tuning limit, or compensation trajectory slope limit allowed by the electronic axis, the sample gating module 43 determines that the feedback at this time is likely affected by execution saturation and does not use this sample for residual parameter updates. The significance of this feature is: to distinguish between the nominal model not covering the real residual and the residual caused by the execution end being unable to continue responding; freezing the residual update under the compensation saturation condition can reduce the possibility of mistaking execution limitation for model mismatch.
[0059] In another implementation (see...) Figure 6 If the registration error detection interface 10 does not output a detection confidence level, the sample gating module 43 can also generate a pseudo-confidence level. This pseudo-confidence level can be calculated based on error continuity, missing marker flags, consistency between adjacent logic control channels, and trend consistency of residuals. For example, if there are missing markers for several consecutive control cycles and there is no synchronization trend between adjacent logic control channels, the pseudo-confidence level is reduced. If the pseudo-confidence level is less than the corresponding threshold, the sample gating module 43 also freezes residual updates.
[0060] This alternative approach enables production lines that lack confidence output for their testing interfaces to determine sample validity, thus expanding the application scope of the solution.
[0061] like Figure 7 As shown, in step S105, after the sample passes the validity determination, the nominal compensation module 44 generates nominal compensation based on the current registration error, error change trend, running speed, and operating condition. Nominal compensation can be formed by one or more of feedback compensation, model feedforward, gain scheduling, or lookup table compensation.
[0062] For example, the nominal compensation module 44 can output a basic phase offset based on the direction of the current error and select different basic gains according to the operating speed range. Nominal compensation is responsible for basic error correction under stable operating conditions, but the residual update module 45 does not learn the original registration error and calculates the unexplained residuals after nominal compensation. In this way, the basic control law and the systematic deviations caused by material batches or speed ranges are separated, reducing the risk of compensation divergence caused by direct black-box learning.
[0063] Combination Figure 7 As can be seen, residual extraction can take valid samples, current error feedback, basic compensation, error trend, operating speed, operating condition, and historical response as inputs, and output a feature vector for the residual update module 45. This feature vector is used to describe the sources of deviation not covered by the nominal compensation, rather than replacing the basic feedback control of the local controller 40.
[0064] like Figure 7 As shown, the residual update module 45 can construct a residual feature vector. For example, the residual feature vector may include registration error, error variation trend, operating speed range, operating condition, historical compensation response, and material label. The residual update module 45 calculates the residual compensation amount based on this feature vector and the compensation parameters in the current operating condition curve. This calculation can employ lookup table mapping, recursive least squares, Kalman estimation, low-order autoregressive exogenous models, or restricted gradient updates. Regardless of the implementation method, the algorithm output must be transformed into a compensation amount usable by the compensation fusion module 46, rather than simply outputting an abstract score.
[0065] In an optional parameterized description, the residual eigenvector and residual compensation amount can be expressed by the following formula: ; in, Indicates the first The logic control channel in the first The residual eigenvector of each control cycle; This indicates the corresponding registration error; Indicates the amount of error change or the trend of error change; Indicates the operating condition; Indicates historical compensation response; This indicates a material label or a weak condition label; Indicates the operating condition curve; Represents the operating condition curve and logic control channels The corresponding compensation parameters; This represents residual compensation mapping relationships such as lookup table mapping, recursive estimation, or restricted gradient update; This represents the residual compensation amount. This parameterized expression is only used to illustrate the input, parameter, and output relationship of the residual update module 45, and does not limit the specific algorithm model.
[0066] like Figure 7 As shown, in one implementation, if the control cycle budget of the local controller 40 is small, the residual update module 45 can use a lookup table for residual compensation: the lookup table entries can be indexed by the operating condition curve, logic control channel, and speed range, outputting the residual bias gain. This lookup table method reduces the amount of online matrix operations, allowing residual compensation to be completed within a shorter control cycle; it can also reduce the risk of compensation instruction lag during computing power congestion. Once the control cycle returns to within the allowable range, the residual update module 45 can switch back to a more refined parameter update method.
[0067] like Figure 8 As shown, the operating condition curves are stored in the parameter storage unit 60. These operating condition curves may include information such as: operating condition identifier, compensation parameters, number of valid samples, residual improvement index, version identifier, and rollback pointer. The operating condition identifier can originate from material batches, speed ranges, logic control channels, error trends, or even weak operating condition labels, etc. The candidate compensation parameters updated by the residual update module 45 do not immediately become long-term parameters; instead, they are first written into the candidate version. This ensures that the updates of the compensation parameters have clear version boundaries, thus preventing the situation where abnormal samples contaminate long-term parameters all at once.
[0068] like Figure 8 As shown, the parameter storage unit 60 can store the current version and the candidate version at the same time. After the candidate version is generated by the residual update module 45, it will enter the verification process. Only after the verification by the feedback verification module 47 is passed will it be replaced or updated to the current version. If the verification fails, it will return to the previous stable version according to the rollback pointer.
[0069] like Figure 8One implementation method is shown: if the material work order interface 30 cannot provide the material batch or material type, the local controller 40 can generate a working condition curve through weak working condition labels; wherein, the weak working condition label can be composed of speed range, mean error, error variance, residual autocorrelation, and compensation response gain. This alternative method, even without a complete material database, can separate significantly different control response segments into different curves, thereby reducing the risk of parameter drift caused by cross-working condition mixed updates.
[0070] like Figure 9 In step S106, the compensation fusion module 46 fuses nominal compensation, feedback compensation, and residual compensation into a final compensation output. For example, the final compensation output can be one or more of the following: phase offset, velocity fine-tuning, or compensation trajectory parameters. The compensation fusion module 46 can also incorporate learning confidence to determine the proportion of residual compensation intervention. Learning confidence can be determined by the number of effective samples, the gating pass rate, and the degree of residual improvement. If the number of effective samples for the current operating condition is small or a batch change has just occurred, the learning confidence is low, and residual compensation intervenes only at a small proportion. As continuous verification passes, the learning confidence gradually increases. This limits the instantaneous intervention intensity of residual compensation, reducing the risk of compensation overshoot or oscillation.
[0071] For example, the final compensation output of the compensation fusion module 46 can be expressed by the following formula: ; in, Indicates the first The logic control channel in the first The final compensation amount is output to the electronic shaft synchronous control interface 50 for each control cycle; Indicates the nominal compensation amount; Indicates the amount of feedback compensation; Indicates the amount of residual compensation; This indicates the learning credibility or the proportion of residual compensation intervention; This represents a combined operator for amplitude limiting, slope limiting, and dead zone handling. Indicates the maximum allowable compensation output boundary of the electronic axis. This indicates the maximum allowable rate of change of compensation between adjacent control cycles. This represents the dead zone threshold.
[0072] like Figure 9As shown, before outputting the compensation command, the compensation fusion module 46 performs amplitude limiting, slope adjustment, and dead-zone processing. Amplitude limiting prevents the compensation command from exceeding the allowable range of the electronic axis synchronization control interface 50; slope adjustment limits the change in compensation amount within two adjacent control cycles to prevent it from changing too rapidly; and dead-zone adjustment prevents frequent compensation triggering due to minor noise. The specific thresholds for amplitude limiting, slope adjustment, and dead-zone can be determined by the allowable output range of the electronic axis synchronization control interface 50, the historically stable error distribution, and the control cycle. If the compensation output approaches the amplitude limit, the compensation fusion module 46 will feed back a saturation state to the sample gating module 43 and trigger subsequent samples to no longer enter the learning path. This ensures that the output state also participates in the sample admission process, thus forming a closed-loop error prevention mechanism.
[0073] exist Figure 9 In the middle, the saturation state feedback line indicates that: in addition to providing compensation instructions to the electronic axis synchronization control interface 50, the compensation fusion module 46 also returns the saturation state after amplitude / slope or dead zone processing to the sample gating module 43, and then performs compensation saturation judgment on the next sample to prevent the learning intervention of the residual from continuing to amplify when the execution end is already limited.
[0074] like Figure 10 As shown, thresholds are not arbitrarily set for parameters. For example, the confidence threshold can be determined by a stable production section detection confidence distribution; the error jump threshold can be determined by the median absolute deviation or quantile of a stable production section error difference; the compensation saturation threshold can be determined by the compensation limit and slope upper limit allowed by the electronic axis synchronization control interface 50; and the verification window length can be a window covering a stable material coordinate or one cycle of error repetition, etc. Using such calibration sources ensures that gating conditions are consistent with field noise and execution capabilities, reducing the risk of insufficient disclosure due to parameter ambiguity.
[0075] Combination Figure 10 As can be seen, the calibration of the threshold and verification window can be sequentially included as follows: extracting the detection value and difference value from the stable production section; performing statistical analysis on the detection distribution and difference distribution respectively, and determining the confidence threshold and jump threshold based on their distribution characteristics; then verifying the saturation threshold through the output limiting of the electronic shaft synchronization control interface 50; and finally determining the verification window based on the error repetition period or the continuity of the material window. The above process illustrates that the threshold is derived from statistics from the stable production section rather than being fixed solely based on human experience.
[0076] In the description of the gating threshold, the detection confidence can be denoted as: The error jump variable can be denoted as The compensation saturation close quantity can be denoted as: The corresponding thresholds can be denoted as: , and The verification window length can be denoted as... .
[0077] For example, when , or At that time, the corresponding control sample will not enter the residual update path; This is used to limit the number of control cycles or material windows for the continuous comparison of residual improvement, overshoot count, and parameter drift amplitude in the feedback verification module 47.
[0078] like Figure 11 As shown, in step S107, the feedback verification module 47 reads the compensated feedback and compares the improvement of residuals, the number of overshoots, and the parameter drift amplitude in the verification window. If the verification result meets the submission requirements, the feedback verification module 47 submits the candidate parameter to the parameter storage unit 60 as a new version. Otherwise, when the verification result shows that the residuals have increased, the overshoots have increased, or the learning direction has reversed, the feedback verification module 47 instructs the parameter storage unit 60 to restore to the previous stable version according to the rollback pointer. Only candidate compensation parameters that have passed the feedback verification after compensation can be effective for a long time, avoiding the situation where abnormal candidate parameters continuously affect subsequent roll sections.
[0079] like Figure 11 As shown, in a partial degradation implementation: if the feedback verification module 47 does not receive compensated feedback within the verification window or the electronic axis synchronization control interface 50 times out, the local controller 40 will not submit candidate parameters but will maintain the previous stable curve and switch the system to conservative nominal compensation mode; after the feedback signal recovers, the local controller 40 will return to the short window verification state, and only after successful verification will the residual parameters be allowed to be updated again. This degradation path is not simply an alarm, but clearly defines the three engineering actions of freezing, conservative compensation, and recovery verification, which can reduce the possibility of incorrect learning occurring momentarily during communication interruption recovery.
[0080] Figure 11 This is the state machine for the corresponding system-level anomaly degradation and recovery verification. As can be seen in the diagram, it is related to... Figure 6 Connected to the sample-level gated state machine in ( Figure 6 It is mainly used to determine whether one or more control samples enter the residual update path, and Figure 11 When system-level anomalies such as communication interruption, receipt timeout, parameter out-of-bounds, or abnormal reset occur, the local controller 40 will enter a frozen, conservative compensation, recovery verification, submission, or rollback state.
[0081] like Figure 12 As shown, in the first parallel embodiment, the registration error detection interface 10 has the capability to output detection confidence. The local controller 40 can directly read the detection confidence, and then the sample gating module 43 performs low-confidence gating. This parallel embodiment is suitable for production lines where the detection end can provide image quality, color mark integrity, or recognition confidence. In this path, the confidence threshold can be determined by the stable production section detection confidence distribution. If the detection confidence is less than the threshold, the sample gating module 43 will block residual updates and write the sample to the abnormal sample cache unit 70. By controlling sample admission at the source of detection quality, the contamination of compensation parameters by low-quality detection input is reduced.
[0082] In the second parallel embodiment, such as Figure 12 As shown, since the registration error detection interface 10 does not provide a detection confidence level, the local controller 40 uses error continuity, missing label status, consistency of adjacent logic channels, and residual trend consistency to generate a pseudo-confidence level. For example, if a logic control channel experiences a sudden error change, but adjacent logic control channels do not have a corresponding trend, and the detection end also records a missing label, the pseudo-confidence level will be reduced, triggering frozen learning. This parallel embodiment differs from the first parallel embodiment in that it does not rely on the detection end directly outputting a confidence field. Instead, the local controller 40 generates sample quality indicators based on time-series behavior, suitable for older production lines or low-configuration detection systems. This maintains the feasibility of the sample gating mechanism without being affected by the problem of the detection end field being too narrowly limited in the claims.
[0083] Specifically, in Figure 12 In this process, error continuity, missing label state, consistency of adjacent logical channels, and consistency of residual trend are all inputs for generating pseudo-confidence. When the pseudo-confidence is greater than the confidence threshold, the sample gating module 43 allows the sample to pass through the gating and enter the subsequent residual learning path; when the pseudo-confidence is less than the confidence threshold, the sample gating module 43 will trigger frozen learning and write the sample to the abnormal sample cache unit 70, and then the local controller 40 will perform conservative compensation output.
[0084] In the third parallel embodiment (see Figure 13 The material work order interface 30 can provide material batch / work order number or material type; the local controller 40 treats these as part of the operating condition curve, resulting in different compensation parameter versions for different material batches or different work order segments. This solution is suitable for situations where the work order system is complete and material batches are recorded. Binding the curve to material tags reduces the possibility of cross-batch samples mixing and updating the same parameter.
[0085] like Figure 13As shown, in the fourth parallel embodiment, the material order interface 30 cannot provide complete material fields. The local controller 40 generates a weak operating condition label based on the speed range, mean error, error variance, residual autocorrelation, and compensation response gain. Although this weak operating condition label is not a material batch number, it can distinguish operating segments with significantly different control responses. This parallel embodiment enables the system to perform operating condition differentiation even in scenarios without a material batch field, avoiding mixing samples from low-speed segments, high-speed segments, and batch change transition segments into the same compensation parameter update path.
[0086] Figure 13 The speed range, mean error, variance error, residual autocorrelation, and compensation response gain in the figure are example input fields for the weak operating condition label. If local numerical labels are set for these input fields in the attached figure, these local labels are only used for the process description in that figure and do not change the meaning of the global attached figure labels such as 10, 20, 30, 40, 50, 60, and 70 in this specification. To avoid multiple interpretations of labels, preferably, the attached figure... Figure 13 You can retain only the field names without setting global map labels for these fields.
[0087] like Figure 14 As shown, in the fifth parallel embodiment, since the local controller 40 has sufficient computing power, and the residual update module 45 uses recursive least squares or Kalman estimation, or a low-order autoregressive exogenous model to update the residual parameters, this path is suitable for situations where the control cycle budget is relatively wide and a finer residual response is required. The residual update module 45 sends candidate parameters to the feedback verification module 47 after each update, instead of directly submitting the long-term curve, thus ensuring rollback capability.
[0088] In the sixth parallel embodiment, due to the limited computing power of the local controller 40, the residual update module 45 uses lookup table bias compensation (see below). The lookup table index can be a working condition curve, a logic control channel, or a speed range; the table entry output is the residual bias gain. This path does not involve complex online matrix operations, allowing the system to generate compensation instructions within a relatively short control cycle. It also reduces the compensation output lag caused by computing power congestion while still retaining the material window attribution, sample gating, and verification rollback main chain.
[0089] exist Figure 14In the path shown, limited computing power can manifest as insufficient control cycle budget of the local controller 40 to complete complex residual model calculations, or the complex model calculations potentially affecting the periodic distribution of compensation outputs. In this case, the residual update module 45 uses the operating condition curve, logic control channel, and speed range to form a lookup table index, reads the corresponding residual bias gain from the parameter storage unit 60, and then sends this residual bias gain to the compensation fusion module 46. This path only replaces the residual compensation calculation method and does not replace material window attribution, sample validity determination, or feedback verification backtracking to the main chain.
[0090] like Figure 11 As shown, in the event of a communication interruption (e.g., the registration error detection interface 10 fails to send valid data to the data acquisition module 41 for several consecutive control cycles, or the electronic axis status interface 20 fails to send back the compensated feedback), the data acquisition module 41 sets the corresponding field to a missing state and sends the missing state to the sample gating module 43. The sample gating module 43 then prevents the residual update module 45 from updating the curve, the compensation fusion module 46 uses a stable curve or conservative nominal compensation output, and the feedback verification module 47 stops submitting candidate parameters. After communication is restored, the local controller 40 does not immediately resume residual learning but enters a short window verification. Only samples within the verification window that are deemed valid will resume parameter updates. This forms a closed loop of interruption-freeze-conservative compensation-restore verification, preventing abnormal data from being written into the parameter storage unit 60 the instant communication is restored.
[0091] like Figure 6 and Figure 11 As shown, in the absence of sensor feedback or when sensor information cannot be acquired (e.g., registration error detection interface 10 reports missing markers; image acquisition fails; error fields are continuously missing), the data acquisition module 41 writes the missing marker into the control sample. The sample gating module 43 determines that the sample is invalid and records the sample and the time of its occurrence in the abnormal sample cache unit 70. The compensation fusion module 46 does not call residual compensation but directly uses the nominal compensation module 44 to output a conservative compensation amount. If the detection input is restored later, the feedback verification module 47 needs to request verification window confirmation first. This avoids treating the void data generated during sensor failure as valid errors in the learning process.
[0092] like Figure 9 and Figure 11As shown, when the electronic axis synchronization control interface 50 does not receive a compensation command or the command's acknowledgment times out, the local controller 40 cannot read the corresponding compensated feedback. The feedback verification module 47 determines that this situation constitutes an acknowledgment anomaly and prohibits the submission of candidate parameters. At the same time, the compensation fusion module 46 reduces the learning confidence and limits the magnitude of subsequent compensation output changes. If consecutive acknowledgment timeouts occur, the local controller 40 enters a conservative compensation state and writes the anomaly into the anomaly sample cache unit 70 to prevent the risk of overcompensation caused by continuing to increase the residual compensation amount when the execution end state is unclear.
[0093] like Figure 10 and Figure 11 As shown, when parameters go out of bounds (for example, the residual parameter update magnitude is too large and exceeds the allowable drift limit of the historical stable curve), the feedback verification module 47 will prompt the parameter storage unit 60 to revert to the previous stable version and treat the current candidate version as an unsubmittable version. This action is different from simply discarding samples. Instead, it explicitly restores the previous stable compensation parameters, so that the electronic axis synchronization control interface 50 can continue to obtain executable compensation outputs, avoiding an abnormal learning that pushes the compensation state away from the stable range for a long time.
[0094] like Figure 11 As shown, in the event of repeated startups or abnormal resets, the local controller 40, after power-on initialization, will first read the most recently verified curve from the parameter storage unit 60, instead of reading the most recently selected candidate parameter. If the curve version verification fails, the local controller 40 will use the default conservative curve and will only enter the verification window after collecting continuous valid samples. This prevents erroneous candidate parameters from taking effect immediately after an abnormal reset.
[0095] like Figure 11 As shown, in the event of a data frame verification failure (such as a control sample field failing verification on an industrial communication link), the data acquisition module 41 will not pass the sample to the material window association module 42, but will instead write it to the abnormal sample cache unit 70; while the compensation fusion module 46 will keep the previous safety compensation output unchanged. This avoids erroneous compensation due to field misalignment or data corruption.
[0096] like Figure 11As shown, when deviations occur in the process (such as material batch change, speed increase, speed decrease, or pause and recovery), the electronic axis status interface 20 sends the current operating status to the data acquisition module 41. The sample gating module 43 determines that this status is an unstable operating condition, the residual update module 45 reduces the update frequency or freezes the update, and the compensation fusion module 46 reduces the learning reliability. If a similar curve appears after batch change, the local controller 40 can call this curve with a lower weight, and gradually increase its intervention ratio after the verification window is passed. This anomaly strategy reduces the risk of overcompensation caused by the old material parameters directly and strongly affecting the new material segment.
[0097] In the above global anomaly embodiments, communication interruption, no sensor feedback, execution receipt timeout, parameter out-of-bounds, abnormal reset, data frame verification failure, and process deviation do not directly lead to long-term parameter writing; their common processing logic is: first, freeze or limit the current learning path, and then use conservative compensation or the previous stable curve to maintain the executable output of the electronic axis synchronization control interface 50, and the feedback verification module 47 performs recovery verification.
[0098] like Figure 15 As shown, in the virtual device embodiment, the above method can be encapsulated as a registration error compensation device for electronic shaft printing of cigarette paper. This device includes a sample acquisition module 81, a window association module 82, a sample judgment module 83, a residual calculation module 84, a parameter update module 85, a compensation output module 86, and a feedback verification module 87. The sample acquisition module 81 can correspond to the data acquisition thread, field communication callback function, or sampling task in the industrial control program within the local controller 40; it connects to the registration error detection interface 10, the electronic shaft status interface 20, and the material work order interface 30, receiving error, speed, status, and feedback. The window association module 82 can correspond to the spatiotemporal alignment operator or cache management program in the local controller 40, receiving the control samples output by the sample acquisition module 81 and outputting the material coordinate window association result. The sample judgment module 83 can correspond to the gating state machine code, receiving the window association result and outputting a valid sample or abnormal sample flag.
[0099] The residual calculation module 84 corresponds to the nominal compensation and residual extraction functions; the parameter update module 85 corresponds to the curve update function or lookup table update logic; the compensation output module 86 is connected to the electronic axis synchronization control interface 50; and the feedback verification module 87 is connected to the parameter storage unit 60. These virtual modules are not abstract empty names; they have data inputs, code actions, and output destinations within the local controller 40.
[0100] Figure 15 The virtual device modules 81 to 87 shown are used to carry out the aforementioned method steps from the perspective of the device object; Figure 2 The data acquisition module 41 to the feedback verification module 47 shown are used to describe the system structure from the perspective of the internal logic of the local controller 40. A correspondence between steps and functions can be established between the two, but the reference numerals in the accompanying drawings are not interchangeable.
[0101] like Figure 15 In the virtual device embodiment, when the sample determination module 83 detects a low confidence level or an error jump, in addition to issuing an alarm, it also sends a freeze flag to the parameter update module 85 and writes the abnormal sample to the abnormal sample cache unit 70. The compensation output module 86, upon receiving a freeze flag from the parameter update module 85, will invoke nominal compensation or the previous stable curve instead of the residual compensation output. This ensures that the anomaly handling has an engineering closed loop rather than merely remaining at the alert level.
[0102] like Figure 16 As shown, in this embodiment of the electronic device, the electronic device 90 may include a processor 91, a memory 92, a communication bus 93, a network interface 94, and a field input / output interface 95. The processor 91 is connected to the memory 92, the network interface 94, and the field input / output interface 95 via the communication bus 93. The network interface 94 can receive data from the registration error detection interface 10, the electronic axis status interface 20, or the material work order interface 30; the field input / output interface 95 can send compensation commands to the electronic axis synchronization control interface 50 and receive feedback after compensation. The memory 92 stores computer programs, control sample caches, curve version records, and abnormal sample records. After the processor 91 calls the program instructions in the memory 92, it executes the steps of control sample acquisition, material window attribution, validity determination, residual information calculation, compensation parameter update, compensation output, and feedback verification rollback.
[0103] Combination Figure 16 It can be seen that the network interface 94 mainly handles the data input of the registration error detection interface 10, the electronic axis status interface 20, and the material work order interface 30, while the field input / output interface 95 mainly handles the output of compensation commands to the electronic axis synchronization control interface 50 and the reception of field control interactions after compensation feedback. The processor 91 retrieves the program and curve records in the memory 92 through the communication bus 93, enabling the aforementioned method to be executed on the electronic device 90 with a clearly defined hardware carrier.
[0104] like Figure 16As shown, electronic device 90 can be an industrial controller / edge computing device / industrial computer or a control unit integrated in a motion control system; processor 91 is not limited to a specific model and can be a general-purpose processor, industrial control processor or embedded processor capable of periodic control task scheduling; memory 92 can include volatile and non-volatile memory, wherein the volatile memory is used to cache control samples during runtime, while the non-volatile memory is used to store information such as curve version, rollback pointer and abnormal sample record.
[0105] This hardware-coordinated approach gives the software method a clear physical carrier, rather than existing merely as an abstract algorithm.
[0106] In the computer-readable storage medium embodiment, the storage medium can be a non-volatile storage medium or a volatile storage medium. For example, a non-volatile storage medium may include flash memory, EEPROM, solid-state memory, a memory card, or a built-in storage unit of an industrial controller; a volatile storage medium may include random access memory. The storage medium stores a computer program that, when executed by the processor 91, causes the electronic device 90 to perform the electronic shaft printing registration error compensation method for cigarette paper in any of the foregoing embodiments. During program execution, the processor 91 receives control samples from the network interface 94 or the field input / output interface 95, associates the samples with the material coordinate window, performs validity determination, updates compensation parameters based on residual information, and submits, freezes, or rolls back parameters based on the feedback after compensation. This storage medium embodiment can cover control algorithm products delivered in the form of software programs.
[0107] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A self-learning compensation method for registration error in electronic shaft printing of tobacco materials, applied to a local controller, characterized in that, include: Acquire control samples for the logic control channel. The control samples include registration error, operating status, historical compensation output, and post-compensation feedback. The material coordinate window is determined based on the control sample, and the registration error, historical compensation output, and post-compensation feedback are associated with the same material coordinate window. The validity of the associated control samples is determined. If the control samples are deemed invalid, the control samples are isolated and the compensation parameter updates are frozen. When the condition is deemed valid, the residual information of the control sample after nominal compensation is calculated, the compensation parameters corresponding to the working condition are updated according to the residual information, and the electronic shaft compensation command is output. Based on the feedback verification and update results after compensation, the compensation parameters are submitted, frozen, or rolled back.
2. The self-learning compensation method according to claim 1, characterized in that: Determining the material coordinate window includes: generating material coordinates based on the running speed and control cycle in the electronic axis's running state, and determining the window length according to at least one of the following: material running length, registration detection resolution, and error repetition cycle; The material coordinate window carries subsequent associated registration errors, historical compensation outputs, and post-compensation feedback; Linking registration error, historical compensation output, and post-compensation feedback to the same material coordinate window includes: For each logic control channel, the equivalent delay is obtained. The equivalent delay is determined by at least one of step response, cross-correlation analysis, and production line calibration. The historical compensation output and the compensation feedback are converted to the material coordinate window where the registration error is located according to the equivalent delay.
3. The self-learning compensation method according to claim 1, characterized in that: The validity determination includes at least two of the following: detection confidence determination, error jump determination, compensation saturation determination, operating condition determination, and residual trend determination. When all selected criteria are met, a valid flag is generated and the process proceeds to residual information calculation; when any selected criterion is not met, an invalid flag is generated, the update of compensation parameters is blocked, and the process switches to abnormal sample processing.
4. The self-learning compensation method according to claim 3, characterized in that: When an invalid flag is generated, the local controller writes the control sample into the abnormal sample buffer unit, maintains the previous stable compensation parameter, and calls the nominal compensation output; When generating a valid flag, the local controller allows the control sample to enter the residual information calculation and prohibits the abnormal sample cache unit from writing to the control sample and triggering the freeze action.
5. The self-learning compensation method according to claim 1, characterized in that: The validity determination of the associated control samples includes: when the control samples contain the detection confidence level, comparing the detection confidence level with the confidence level threshold; When the control sample does not contain the detection confidence level, a pseudo confidence level is generated based on error continuity, missing labels, consistency of adjacent logic control channels, and consistency of residual trends, and the pseudo confidence level is compared with the confidence level threshold.
6. The self-learning compensation method according to claim 5, characterized in that: When the control sample contains a material label, the operating condition is determined by the material label. When the control sample does not contain material labels, the local controller generates weak operating condition labels based on the speed range, mean error, error variance, residual autocorrelation, and compensation response gain, and uses the weak operating condition labels as components of the operating condition status. The compensation parameters corresponding to the operating conditions are updated based on residual information, including: when the control cycle budget of the local controller does not exceed the calculation boundary, the residual compensation amount is obtained by recursive update; When the control cycle budget exceeds the calculation boundary, the residual bias gain is obtained by looking up a table using the operating condition, logic control channel, and speed range as indexes.
7. The self-learning compensation method according to claim 1, characterized in that: The verification update results based on the compensation feedback include: when compensation feedback is obtained in the verification window and the receipt has not expired, the corresponding processing in submission, freezing, and rollback is performed according to the verification results; If at least one of the following conditions is met in the verification window: no compensation feedback is obtained or the receipt times out, the submission of candidate compensation parameters is prohibited, the previous stable compensation parameter is maintained, and the verification is re-executed after the feedback is restored.
8. The self-learning compensation method according to claim 1, characterized in that: The electronic axis compensation command includes at least one of the following: phase offset field, speed fine-tuning field, and compensation trajectory field; The local controller writes the electronic axis compensation command into the electronic axis synchronous control interface and records the writing time, logic control channel, acknowledgment status, compensation output status and corresponding compensation parameter version in the parameter storage unit to form an externally readable compensation log.
9. The self-learning compensation method according to claim 8, characterized in that: Compensation parameters are stored in the parameter storage unit as operating condition curves. The operating condition curves include operating condition identifiers, compensation parameters, number of valid samples, residual improvement indicators, version identifiers, and rollback pointers. Submission makes the candidate version the current version, and rollback restores the previous stable version according to the rollback pointer and makes it available for the feedback verification module to read.
10. The self-learning compensation method according to claim 9, characterized in that: The records written to the abnormal sample buffer unit include the logical control channel, sampling time, invalidation reason, compensation parameter version before freezing, and corresponding compensation output status; Invalid reasons include at least one of the following: low confidence, error jump, compensation saturation, unstable operating conditions, and inconsistent residual trends, and are saved in correspondence with the frozen compensation parameter update action; The local controller receives registration error, operating status, and material labels through the network interface, and outputs electronic axis compensation commands to the electronic axis synchronization control interface through the field input / output interface and receives the compensation feedback. The input / output records of the network interface and the field input / output interface are saved in correspondence with the control samples for use by the feedback verification module.
