On-line detection, sorting and control method and system for pole piece die-cutting defects

Abstract

The application discloses a method and system for online detection and sorting control of pole piece die cutting defects, and particularly relates to the technical field of online detection and sorting control of pole piece die cutting, and is used for solving the problems of event landing uncertainty and sorting error caused by time base asynchronization, coordinate mapping drift and late disorder under the condition of multi-station dynamic rhythm. Through the establishment of double-anchor time base and idempotent key, on the basis of water line promotion and disorder window playback as the order under the unified time axis, combined with online self-proving of coordinate transformation of the same layer and time calibration and space calibration in the topological graph, unified decision is implemented at the merging moment and two-section interlocking is executed, so that the pole piece detection event is established in a determined and reviewable piece-level corresponding relationship with the punching frequency, die cavity and discharge port under the condition of multi-station dynamic rhythm, and the consistency of sorting decision is maintained and the risk of misclassification and missed classification is reduced in the face of late arrival, disorder and state transition.

Description

Technical Field

[0001] This invention relates to the field of online detection and sorting control technology for electrode die-cutting, specifically to an online detection and sorting control method and system for electrode die-cutting defects. Background Technology

[0002] Existing electrode die-cutting production lines mostly employ online visual defect identification and downstream sorting. A common practice is to deploy linear or area array vision units on the roll material, using encoder displacement and fixed delay to convert defect locations to stroke counts or meter increments, which are then rejected by the controller at a designated outlet. For ease of implementation, triggering and timing are typically based on a single time reference, with coordinate calibration usually completed offline after startup or tool change, and the mapping relationship maintained subsequently using a fixed ratio or empirical compensation. In scenarios with multiple stations, multiple cavities, and parallel output, simple matching is often achieved through cumulative stroke counts, frame sequence advancement, or first-in-first-out methods. The execution side is often single-trigger, without retaining complete decision paths and caliber versions.

[0003] In actual production, the above methods have several structural limitations. First, the detection and execution links are independent of each other, and the timestamp sources are not unified. When encountering state transitions such as acceleration, deceleration, short stop and restart, tension and temperature fluctuations, and tool changes, a single time reference and static calibration are prone to phase deviation, and the mapping relationship drifts accordingly, resulting in inaccurate judgment of arrival stroke. Second, event mapping under complex topologies relies on the calculation of the number of strokes and the assumption of frame order, lacking bounded tolerance and unified adjudication for late and out-of-order records. This easily leads to duplicate records, out-of-order writing, or the same event being competed for by different outlets, thus causing missorting or missed sorting. Third, current solutions mostly use thresholds and templates for judgment, but lack systematic recording of versions, standards, and adjudication processes. Once the quality of the production line fluctuates, it is difficult to trace the judgment basis and execution conditions at that time, resulting in high audit costs and difficulty in ensuring consistency of verification.

[0004] Based on the above situation, under the dynamic cycle conditions of multi-station electrode die cutting, how can we establish a definite and verifiable sheet-level correspondence between upstream detection events and downstream station topology, die cavity and discharge port within a unified time axis and unified coordinate caliber, and form a consistent decision and sorting execution order in the presence of delays, disorder and state transitions, so as to reduce missorting and omissions caused by time drift, mapping conflict and inconsistency of caliber? Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] To address the shortcomings of existing technologies, this invention provides an online detection, sorting, and control method and system for electrode die-cutting defects, thereby solving the problems mentioned in the background section.

[0007] (II) Technical Solution

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] Online detection and sorting control method for electrode die-cutting defects includes:

[0010] S1. Establish a dual-anchor time base for mechanical anchors and communication anchors, generate idempotent keys, and set water level lines and disordered windows as the basis for adjudication and evidence preservation.

[0011] S2. Based on tension, speed and temperature observations, maintain homogeneous coordinate transformation. Trigger perturbation identification and update transformation parameters when speeding up, decelerating or tool changing occurs.

[0012] S3. Map the event snapshots generated by the detection to time-based machine coordinates according to coordinate transformation, and bind and register them with idempotent keys;

[0013] S4. Construct a topology map of the workstation, mold cavity, and discharge port; generate tokens for events; predict the arrival number of strokes according to the cycle time and determine the mapping; and perform time and space checks.

[0014] S5. Make a unified decision when the water level reaches the merging time. Late events are replayed and corrected within the disordered window, and out-of-window records are isolated.

[0015] S6. Implement two-stage interlocking triggering based on the mapping. If the interlocking conditions are not met, enable the wrong period protection. At the same time, solidify the evidence chain and record the anchor point, template, threshold, mapping and time stamp.

[0016] Furthermore, S1 includes:

[0017] Align the mechanical anchor and the communication anchor to establish a dual-anchor time base and form a unified time axis;

[0018] Set the water level line for sequential advancement and the disordered window;

[0019] The idempotent key is used as the unique identifier for the event. The idempotent key consists of the device identifier, anchor identifier, window number, template version, and fingerprint digest.

[0020] Create a time base record table with an idempotent key as the primary index, and write the water level position, out-of-order window parameters, and clock difference monitoring values ​​into the table.

[0021] Set isolation flags and prevent writing to records that are earlier than the waterline position and later than the upper limit of the out-of-order window;

[0022] For duplicate records with the same idempotent key, retain the first valid record.

[0023] When either the anchor source is continuously missing or the clock error exceeds the limit, the operation will be downgraded to a single anchor working state and the water level line will be maintained for advancement.

[0024] Changes involving time base, out-of-order windows, template versions, and idempotent keys are implemented with version locking and written into the evidence chain.

[0025] Furthermore, S2 includes:

[0026] Establish a homogenized coordinate transformation under a dual-anchor time base and implement online self-verification;

[0027] The affine mapping from pixel coordinates to machine coordinates is continuously corrected based on observations of tension meter, spindle speed, and frame temperature.

[0028] The perturbation excitation, constrained by the safety budget, is applied to identify the events that occur during acceleration, deceleration, and tool change.

[0029] When the convergence criterion is met within the observation window, the transformation parameters are updated, a transformation version is generated, written to the version library and locked in a read-only state, and version switching is performed using atomic replacement.

[0030] Time and space verification call the current transformation version;

[0031] Records are linked one-to-one with idempotent keys and written into the evidence chain.

[0032] Furthermore, when the tension channel loses connection, the virtual reference grid is switched and a degraded annotation state is set;

[0033] If the time verification fails, roll back to the previous stable transformation version and hold until the next observation window converges;

[0034] If the spatial verification fails, roll back to the previous stable transformation version and hold until the next observation window converges;

[0035] When the concurrent flow reaches the quota limit, flow restriction is implemented according to the water level advancement order. Late records are entered into the disordered window and corrected at the next merging time.

[0036] The rule table implements version locking for thresholds, number of attempts, priority, merge cycle, out-of-order window, and retry limit, and is fixed in sync with the changed version.

[0037] Furthermore, S3 includes:

[0038] When the dual-anchor time base and homogenized coordinate transformation are in an authorized state, the event snapshot is de-jittered, time-corrected, and missing data is filled in according to the template in the authorized state, and then verified by the template.

[0039] The centroid is mapped to mechanical coordinates according to the homogenization coordinate transformation, and the centroid deviation is determined by the displacement of the centroid relative to the geometric center of the bounding box.

[0040] Registered with an idempotent key containing device identifier, anchor point identifier, window sequence number, template version and fingerprint digest, generating binding records containing mechanical coordinates, centroid deviation, confidence level, unified timestamp, template version and transformation version, and writing them to the event table in an append-only manner. The event table uses the idempotent key as the primary index and the unified timestamp as the secondary index.

[0041] Within the same water level advance cycle, only the earliest recorded record is retained according to the water level advance order. The earliest record with the unified time stamp is used as the standard. If the unified time stamps are the same, the records are compared sequentially according to the fixed order of the idempotent key fields to determine the earliest record. The remaining records are marked as duplicates.

[0042] Furthermore, S4 includes:

[0043] After the dual-anchor time base, homogenized coordinate transformation and event binding are completed, a directed topology graph consisting of workstations, mold cavities and discharge ports is constructed, where edges are used to record the number of transmission steps and nominal delay.

[0044] For events identified by idempotent keys, generate tokens, predict the arrival number of strokes based on the cycle time and transmission number, and determine the target cavity and target discharge port;

[0045] Perform time verification and space verification. The passing condition for time verification is that the mechanical anchor phase and the communication anchor phase are consistent. The passing condition for space verification is that the mechanical coordinates fall into the feasible region of the mold cavity mask and the boundary allowance meets the template version requirements.

[0046] Only the first mapping record that passes both time and space verification is retained. The mapping record, along with the time stamp, topology version, and template version, is written into the mapping table. A composite index is created using the idempotent key and the time stamp. At the time of merging, the mapping table is read based on the composite index to execute the decision.

[0047] Furthermore, S5 includes:

[0048] Under a unified timeline, a one-time ruling is made on records of the same work station at the moment the water level reaches the merging point.

[0049] Only mapping records that have passed time and space verification are included; late records in the out-of-order window are replayed and corrected using the idempotent key.

[0050] The previous conclusion is revoked; items exceeding the out-of-order window are assigned an isolation marker; competition within the same stroke is adjudicated according to the fixed priority order of the quality quadruple.

[0051] Competition at the same discharge port is adjudicated in a fixed order of consistency of discharge diameter, freshness, integrity and credibility. The adjudication record contains the idempotent key, adjudication conclusion, threshold version, mapping version, timestamp and cause code, and is written into the adjudication table and the chain of evidence.

[0052] The decision submission is serialized and retried within the decision retry limit.

[0053] Furthermore, S6 includes:

[0054] Establish a two-stage interlocking triggering order;

[0055] During the pre-arm phase, the adjudication record and mapping record are read from the adjudication table based on the idempotent key within the pre-arm window. The consistency of the threshold version, topology version, coordinate transformation version, and template version is checked, and the execution buffer is registered with time markers, resource occupancy, and interlock channel status.

[0056] During the submission phase, within the submission slot of the captured image, an action command is issued only when the mapping is valid, the position interlock is established, and the playback conflict is eliminated simultaneously. If any of these conditions are not met, the error cycle protection is triggered and the evidence chain is solidified, prohibiting overshoot compensation.

[0057] Furthermore, link latency and concurrency are capped by a resource quota table;

[0058] If no confirmation is received, retry will be performed according to the preset backoff sequence within the submission time slot. If the retry limit is reached and the failure is still not achieved, error cycle protection will be maintained.

[0059] The interface field is written to the audit database with an idempotent key as the unique index, and the action flag, device feedback, and reason code are registered.

[0060] Parameter and rule modifications take effect at the time of water level merging, are rolled back according to the version dictionary and leave a trace with the summary verification value, and cross-network communication uses an encrypted channel.

[0061] On the other hand, the present invention provides an online detection and sorting control system for electrode die-cutting defects, including:

[0062] Time base anchoring and keying module: used to establish a dual-anchor time base for mechanical anchors and communication anchors, generate idempotent keys, set water level lines and out-of-order windows, and serve as a reference for adjudication and evidence storage;

[0063] Coordinate transformation self-verification module: It is used to maintain the homogeneous coordinate transformation based on tension, speed and temperature observations, and triggers perturbation identification and updates parameters during speed increase, deceleration and tool change to form alignment convergence constraints.

[0064] Event binding module: Used to receive snapshots of detected events, map them to time-based machine coordinates according to coordinate transformation, and complete the binding registration with idempotent keys to ensure the uniqueness of events;

[0065] Topology mapping verification module: used to construct a topology map of the workstation, mold cavity, and discharge port, generate tokens for events, predict the arrival number of strokes and discharge port according to the cycle time, and perform time and space verification;

[0066] Decision replay module: Used to execute a unified decision when the water level reaches the merging time. Late events are replayed and corrected within the out-of-order window, and isolated events are executed outside the window to maintain decision consistency.

[0067] The interlock audit module is used to implement two-stage interlock triggering based on mapping. When the interlock conditions are not met, the error cycle protection is activated. At the same time, the evidence chain is solidified, and anchor points, templates, thresholds, mappings and time stamps are recorded.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] 1. By establishing a dual-anchor time base and idempotent key, and using water level advancement and disordered window playback as the order basis under a unified time axis, combined with online self-verification of homogenized coordinate transformation and time and space verification in the topology graph, a unified decision is implemented at the merging moment and executed in a two-stage interlocking manner. This achieves a definite and verifiable piece-level correspondence between detection events and strokes, mold cavities, and discharge ports under multi-station dynamic cycle time. It maintains consistency in sorting decisions and reduces the risk of missorting and omissions even in the face of lateness, disorder, and state transitions.

[0070] 2. By integrating perturbation identification-driven coordinate transformation version locking, quality rules and cause code dictionary management, resource quotas and retry backoff strategies, and the whole-process solidification and rollback mechanism of the evidence chain, the system can maintain alignment stability and consistency under field conditions such as speed changes, tool changes, tension and communication fluctuations, support rapid traceability and compliance auditing, and improve the continuous availability and maintenance efficiency of the production line. Attached Figure Description

[0071] Figure 1 This is a flowchart illustrating the online detection, sorting, and control method for electrode die-cutting defects of the present invention.

[0072] Figure 2 This is a schematic diagram of the online detection and sorting control system for electrode die-cutting defects of the present invention. Detailed Implementation

[0073] 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.

[0074] Example 1: Figure 1 A flowchart illustrating the online detection, sorting, and control method for electrode die-cutting defects of the present invention is provided. The online detection, sorting, and control method and system for electrode die-cutting defects include:

[0075] S1. Establish a dual-anchor time base for mechanical anchors and communication anchors, generate idempotent keys, and set water level lines and disordered windows as the basis for adjudication and evidence preservation.

[0076] S2. Based on tension, speed and temperature observations, maintain homogeneous coordinate transformation. Trigger perturbation identification and update transformation parameters when speeding up, decelerating or tool changing occurs.

[0077] S3. Map the event snapshots generated by the detection to time-based machine coordinates according to coordinate transformation, and bind and register them with idempotent keys;

[0078] S4. Construct a topology map of the workstation, mold cavity, and discharge port; generate tokens for events; predict the arrival number of strokes according to the cycle time and determine the mapping; and perform time and space verification.

[0079] S5. Make a unified decision when the water level reaches the merging time. Late events are replayed and corrected within the disordered window, and out-of-window records are isolated.

[0080] S6. Implement two-stage interlocking triggering based on the mapping. If the interlocking conditions are not met, enable the wrong period protection. At the same time, solidify the evidence chain and record the anchor point, template, threshold, mapping and time stamp.

[0081] The technical connections and implementation logic of the six steps are as follows:

[0082] S1 first establishes a dual-anchor time base using mechanical and communication anchors, generating idempotent keys and setting water level lines and disordered windows as unified time and identity benchmarks for subsequent stages; S2 maintains homogenized coordinate transformation based on tension, speed, and temperature observations under this time base constraint, triggering perturbation identification and parameter updates upon encountering acceleration, deceleration, or tool change to ensure stable alignment; S3 maps the detected event snapshots to time-based mechanical coordinates according to the current coordinate transformation, and completes binding registration using idempotent keys, making the events traceable and controllable objects; S4 constructs a topology map based on workstations, mold cavities, and discharge ports, providing a unified time and identity reference for the bound events. The process involves generating tokens, predicting arrival times based on the beat, determining mappings, and simultaneously performing time and space checks to form mapping records that can be used for adjudication. In step S5, when the water level reaches the merging time, a unified adjudication is performed on the mapping records of the same workstation. Late records are replayed and corrected within the out-of-order window, while records exceeding the window are isolated and marked, and adjudication records are generated. In step S6, a two-stage interlocking trigger is implemented based on the adjudication and mapping. The action is submitted when the mapping is valid, the position interlock is established, and the replay conflict is eliminated. If any of these conditions are not met, the error cycle protection is enabled. At the same time, the evidence chain is solidified, and anchor points, templates, thresholds, mappings, and time markers are recorded.

[0083] S1. Establish a dual-anchor time base consisting of mechanical and communication anchors, generate idempotent keys, and set water level lines and out-of-order windows as the basis for adjudication and evidence preservation. The specific implementation is as follows:

[0084] A unified timeline is established on the roll die-cutting line for the controller and field equipment to follow, serving as the basis for subsequent adjudication, sorting, and traceability. The mechanical anchor is a time reference formed by the spindle encoder zero position, stroke signal, and alignment photoelectric sensor; the communication anchor is a time reference formed by the camera frame time marker and the control cycle time marker. The dual-anchor timeline is a unified timeline established by aligning the two types of time markers and continuously monitoring drift under the time calibration procedure. The term "time marker" is used consistently throughout the document. Shifts are continuous production time periods divided by the production schedule. When switching shifts, the window sequence number is reset from its initial value, and the template version is locked within the shift. Preferably, a subordinate relationship is adopted with the mechanical anchor as the master and the communication anchor as the slave. The communication anchor aligns with the mechanical anchor at a fixed stroke or fixed time interval. When drift is observed to exceed a set threshold, a forced alignment is immediately performed, and an alarm record is generated.

[0085] On-site data acquisition is measured in milliseconds, with the rhythm based on strokes, ideally controlling clock errors within the millisecond range. To ensure consistency, three preprocessing steps are performed: jitter removal, time alignment, and missing data completion. Jitter removal filters out trigger edge jitter, time alignment eliminates the phase difference between the control cycle and camera trigger, and missing data completion fills in the boundary according to the neighborhood rhythm in case of single frame loss or isolated pulse gaps. All time stamps use the production line's local time zone and record time zone offsets, with continuous advancement across days without looping. The water level line serves as a decision indicator, monotonically advancing along a unified time axis, with its advancement step size tied to the stroke rhythm. The window sequence number is the sequence number of the merged windows formed by the water level line advancement, monotonically increasing along the unified time axis. The merging time is the trigger point when the water level line coincides with the target workstation's beat, forming a one-time decision; only at this time point are unified decisions formed for the records of the same workstation.

[0086] The out-of-order window is a late arrival tolerance zone set around the water level line. It is used to accept late records within the permitted range and perform playback corrections. Records outside the range are marked as isolated. The out-of-order window and the advance step can be set with minimum, maximum, and default values, and are locked at the beginning of the shift. Adjustments during the shift require authorization and preferably take effect at the next merge time. The idempotent key is a unique identifier for the same event in each stage of the system. It includes five types of fields: device identifier, anchor point identifier, window sequence number, template version, and fingerprint digest. These fields are encoded in a fixed field order and the character set and field length are unified in the parameter table. The uniqueness of the key is guaranteed by key space conflict checking and duplicate write interception. When a conflict occurs, the priority is determined by the order of more recent time, higher template version, and source priority. If it is still impossible to distinguish, the key is put into a verification state and subsequent writes are frozen.

[0087] The template version is a configuration version used to interpret events and time-series records. It includes a set of fields, value criteria, threshold configuration, and loading time stamp. The version number and version summary are written into the evidence chain during loading and locked within the shift. The fingerprint summary is an irreversible summary value generated based on field concatenation and records the summary algorithm identifier and algorithm version for integrity verification. It does not contain information that can restore the original text. To support ordered cross-module calls, a time-based record table is established. It uses an idempotent key as the main index to store the double anchor status, water level position, out-of-order window parameters, window sequence number, and clock difference monitoring value. It is preferably stored in the controller's local persistent area with a cyclic write strategy to control resource consumption. The retention period of the time-based record table can be set to no less than the complete records of the most recent three shifts, preferably no less than the most recent five shifts. Records exceeding the retention period are discarded in a first-in-first-out manner, but the evidence chain summary is retained until the shortest audit period expires. Downstream modules only access the time-series status of the same event through the idempotent key to avoid ambiguity caused by inference based on relative time or frame number.

[0088] The communication method can be set to fieldbus or industrial Ethernet, with a link latency upper limit set and maintained within milliseconds. A link is considered disconnected when the continuous observed communication interval exceeds a set multiple of the link latency upper limit or the cumulative packet loss rate exceeds a threshold. The multiple and threshold are locked in the parameter table, preferably using a ratio of two times and one percent. In disconnection scenarios, retrying employs a backoff strategy with increasing time intervals. The backoff interval and retry upper limit are locked in the parameter table. After exceeding the limit, the link is transferred to a degradation queue. When the link recovers, retrying is performed in the order of the water level lines. The order is solely based on the water level line progression; any write crossing the water level line is treated as a violation and the reason is recorded. Out-of-bounds writes refer to write requests whose timestamps are earlier than the current water level line position or later than the allowed out-of-order window upper limit. Deduplication is based on idempotent keys; for any duplicate arrivals, the first one to arrive is valid, and the ruling is recorded. Time and computing resources are subject to quota control, and clock error monitoring and key generation are completed within a limited period. When the clock error exceeds the allowable tolerance or any anchor source is continuously missing, the system is downgraded to single anchor operation and retains the water level advance record, while generating a cause code.

[0089] The safety and compliance boundaries are clearly defined: parameter modifications require authorization review and are logged; cross-network communication uses encrypted channels; time base records and evidence chains are used only for process traceability and are stored according to the principle of minimum necessity; if any of the safety gate, area photoelectric sensor, or emergency stop circuit is disconnected, the water level advance is immediately suspended and the out-of-order window timing is frozen; after restoration, the advance continues in the order of the water level and a cause code and time stamp are written. The interface standard is unified as a field set mode, covering equipment identifier, anchor point identifier, window sequence number, clock difference value, water level position, window parameters, template version, fingerprint summary, and time stamp; error codes and cause codes are managed using the same coding table, and the codes, meanings, and handling suggestions are aligned in the parameter table and fixed with the version. The error code set includes at least key duplication, clock difference exceeding limit, channel disconnection, and out-of-bounds writing for rapid interaction and handling between upstream and downstream; the cause code can be set as follows: No. 1 key duplication, No. 2 clock difference exceeding limit, No. 3 channel disconnection, and No. 4 out-of-bounds writing for attribution and statistics.

[0090] All additions or modifications involving time bases, windows, templates, and keys are version-locked. Version numbers and summary fingerprints are recorded and written into the evidence chain, with idempotent keys connecting each stage. Parameter changes are managed in two phases: pre-installation and switching, preferably taking effect only at the merge time. In case of continuous alarms, a rollback to the previous stable version is possible. Through the disclosure of the above capabilities and scope, the mechanisms for establishing, advancing, monitoring, downgrading, and replaying dual-anchor time bases, the setting and effective boundaries of water level lines and out-of-order windows, the encoding and conflict adjudication of idempotent keys, the storage, retention, and indexing of time base record tables, communication, ordering, and deduplication strategies, disconnection judgment and backoff paths, version locking and evidence chain traceability, as well as security and compliance boundaries have all been clearly disclosed. Based on this, a stable and auditable time base can be formed on-site, providing consistent upstream conditions for subsequent coordinate transformations, topology mapping, unified adjudication, and interlocked execution.

[0091] S2. Based on tension, speed, and temperature observations, maintain the homogenized coordinate transformation. When speed increases, decelerations, or tool changes occur, trigger perturbation identification and update the transformation parameters. The specific implementation is as follows:

[0092] To continuously obtain consistent pixel coordinates to machine coordinates under multi-station cycle fluctuation conditions, a homogeneous coordinate transformation is established and online self-verification is implemented. Homogeneous coordinate transformation refers to the affine relationship that maps pixel coordinates to machine coordinates over time, covering scaling, rotation, shearing, and translation. Online self-verification refers to continuously correcting this relationship based on observable measurements on site and controlled micro-amplitude excitation, and achieving convergence within a specified observation window. The dual-anchor time base is a unified time axis formed by the mechanical anchor and the communication anchor through time synchronization. The mechanical anchor includes the spindle encoder zero position, stroke signal, and alignment photoelectric sensor, while the communication anchor includes the camera frame time marker and the control cycle time marker. The water level line is a decision indicator that advances monotonically along the unified time axis. The merging time is the unified decision point when the water level line reaches the designated workstation, which is used as the timing reference for flow limiting sorting and version switching. The out-of-order window is the time range within which late records are allowed to be corrected in the next merging time, measured in strokes. It is set by the rule table and fixed with each version. The window number is a discrete window number generated as the water level line advances, used for consistency identification of idempotent key deduplication and sequential decision.

[0093] The idempotent key is a unique identifier composed of device identifier, anchor point identifier, window sequence number, template version, and fingerprint digest. The fingerprint digest is a verification value composed of a feature summary of the event snapshot and a time stamp. All records and evidence chains in this stage are linked one-to-one using idempotent keys to ensure traceability. Time verification checks the consistency of mechanical and communication phases under a unified time axis, while spatial verification checks whether the mapped coordinates fall within the feasible region of the target cavity mask. Both are used in subsequent topology mapping stages, where the latest transformation version of this section is called. On-site observations are sourced from tension gauges, spindle speed, and frame temperature, in Newtons, revolutions per minute, and degrees Celsius, respectively. Sampling rhythm, tolerance, and default values ​​are fixed in the parameter table, which includes the observation name, unit, sampling rhythm, tolerance, default value, and version number. To suppress mismatches caused by noise and abrupt changes, smoothing, boundary checks, and short-window interpolation are performed on the above observations. Short-window interpolation is used to interpolate and fill in missing measurement points within a range not exceeding the length of the observation window. The observation window covers a time period of several pulses and is used for continuous determination of alignment errors and convergence confirmation.

[0094] When acceleration, deceleration, or tool change occurs, perturbation identification is initiated. Perturbation identification applies a small excitation within a safety budget constraint to estimate transmission slippage and phase deviation. The safety budget is a constraint on the excitation amplitude and duration, consisting of allowable roll tension, frame temperature rise, and displacement limits. Preferably, it is given by the machine calibration table and does not change the continuous roll quality and thermal balance. Within an observation window covering several strokes, convergence is considered achieved when the alignment error sequence continuously falls within the consistency threshold and reaches the number specified in the rule table. The consistency threshold is the threshold range within which the alignment error must continuously fall. The rule table includes the threshold, number of attempts, priority, out-of-order window, merging period, retry limit, and version number. After convergence, the homogenization coordinate transformation parameters are updated and a transformation version is generated. The transformation version includes the version number, applicable time range, participating observation scope, convergence criterion, and stimulus summary. It is written to the version repository and locked as read-only. The version repository is a local persistent storage of the controller and is used for subsequent mapping to retrieve by version number. Version switching adopts atomic replacement. Atomic replacement refers to switching the transformation version pointer within a single indivisible operation to ensure read and write consistency.

[0095] To enhance robustness, a virtual reference grid is established as a secondary reference. This virtual reference grid consists of a set of reference points formed by electrode edge textures, coating edges, and several positioning holes. When the tension channel loses connection, the system switches to the virtual reference grid and performs a degradation annotation. This degradation annotation sets a status bit in the record when a specified observation loses connection or reliability decreases; the lost record is marked with cause code number three. To meet cycle time requirements, an upper limit is set for the time overhead of single-event calculation and version switching, preferably not exceeding a fixed proportion of the single-stroke interval. When concurrency approaches the upper limit, flow restriction is performed according to the water level advancement order. Flow restriction limits the number of records that can be processed within a single window based on the water level order to meet the delay upper limit. Late records exceeding the threshold are transferred to the out-of-order window and corrected at the next merging time. The bound mechanical coordinates output in this stage are time-verified and space-verified by subsequent topology mapping. If any verification fails, the system backtracks to the previous stable transformation version based on the evidence chain and maintains that version until convergence in the next observation window to avoid cascading offsets.

[0096] The evidence chain is a set of judgment paths and version traces from discovery to execution, including anchor point identifiers, window sequence numbers, template versions, transformation versions, parameter table version numbers, rule table version numbers, and time calibration markers. The time calibration markers are unified time stamps generated under a dual-anchor time base, used for cross-record association and sequential adjudication. The template version is the version number of the detection-side image judgment template and threshold set, used to verify the judgment criteria at the time. Cause codes are managed using a unified table and are fixed with each version, specifically: 1. Key duplication; 2. Clock difference exceeding limits; 3. Tension channel disconnection; 4. Centroid exceeding limits; 5. Time verification failure; 6. Spatial verification failure; 7. Window timeout; 8. Replay conflict unresolved; 9. Interlock not established; 10. Incorrect cycle trigger. Boundary conditions include: alignment error can be set to no more than a certain proportion of the edge safety distance, where the edge safety distance is the minimum safety boundary from the electrode edge to the critical structure; temperature drift must not exceed the stable range specified in the equipment manual; the amplitude and duration of micro-excitation must not cause a decrease in slitting quality.

[0097] Thresholds, number of attempts, default values, and priorities are determined by the rule table and fixed with each version; specific values ​​are listed separately in the preferred embodiment and are not limited in this section. The order is based on the water level as the sole decision sequence, and deduplication is based on the idempotent key; when the observation continuously exceeds the limit, the update is frozen, a busy flag is set, and the reason code is recorded; when the temperature drift approaches the upper limit, the micro-excitation is postponed and the observation window is shortened; when the micro-excitation does not reach the safety budget or the continuous convergence does not meet the number of attempts, the update is skipped and tried in the next cycle; after the retries reach the upper limit, the old version is maintained and the record is marked as low confidence. Low confidence is indicated by a flag in the evidence chain to suggest that subsequent decisions need to be checked again. An alternative approach is to use fixed positioning holes as long-term references in production lines where the visibility of positioning holes is stable in the long term to replace the virtual reference grid. This substitution does not change the causal order of dual-anchor time base, water level advancement, observation window convergence, version change fixing, and evidence chain recording. In summary, without altering the existing cycle time and safety protocols, the aforementioned capabilities and scope enable continuous online self-verification of homogeneous coordinate transformation. This allows for the restoration of alignment stability within a limited number of chips after rate jumps, restarts, and tool changes. Furthermore, versioned traceability ensures verifiability and reproducibility, meeting the control requirements regarding latency limits, sequential consistency, and idempotent strategies.

[0098] S3. Map the event snapshots generated by the detection to time-based machine coordinates according to coordinate transformation, and bind and register them with idempotent keys. The specific implementation is as follows:

[0099] During the event snapshot binding phase, provided that the dual-anchor time base and homogenized coordinate transformation have been established and are in an authorized state, the defect description from the event source is transformed into a binding record that can be directly invoked by the topology mapping. The event source is the upstream detection unit, whose output is constrained by a unified template to form an event snapshot. The event snapshot consists of defect category, bounding box, centroid, confidence level, frame time stamp, and texture fingerprint. The template is a rule set of field names, units, rhythms, tolerances, value ranges, and mandatory relationships, and has an authorized state and version number, which is recorded along with the evidence chain. The mechanical coordinate system is in millimeters, with the origin and axis fixed according to the production line reference, and the orientation marked in the process drawing. The unified time stamp is a millisecond-level time scale after being calibrated by the mechanical anchor and communication anchor, used as the order reference for registration and adjudication.

[0100] To ensure temporal consistency, the water level is defined as an indication of the adjudication moment that progresses monotonically according to a unified time marker, with the unit being milliseconds; the observation window is defined as the effective time range for backtracking from the current water level; and the out-of-order window is defined as the time constraint that allows late records to be replayed and corrected within a limited time range. On-site, the event snapshot is first de-jittered, time-synchronized, and missing data filled according to the template. Verification items include field existence, consistency of units and value ranges, and compliance of rhythm and tolerance. When the template is authorized and passes verification, the template version and template summary are written into the evidence chain; those that fail are not bound and their reason codes are recorded. Subsequently, based on the current homogenized coordinate transformation, the centroid is projected onto the machine coordinate system, and the centroid deviation is calculated. The centroid deviation is the distance between the projected centroid and the geometric center of the bounding box, with the same dimensions as the machine coordinate system.

[0101] The system uses idempotent keys as unique identifiers to register events. Idempotent keys consist of device identifier, anchor point identifier, window sequence number, template version, and fingerprint digest, forming a bound record containing machine coordinates, centroid deviation, confidence level, unified timestamp, template version, and transformation version, which is then written to the event table. The event table is an append-only persistent container, using the idempotent key as the primary index and the unified timestamp as the secondary index for downstream retrieval. In concurrent scenarios, only the first valid record is accepted; subsequent records are marked as duplicates and recorded with reason code one. Sequence is solely based on the waterline progression; any record earlier than the lower boundary of the observation window is not entered into the event table and is discarded according to the unified timestamp and idempotent key. Centroid out-of-bounds is determined when the mapped coordinates fall outside the feasible region of the target workstation or exceed the outer safety zone of the boundary frame, resulting in a centroid out-of-bounds determination and recording reason code four. The feasible region of the target workstation is a closed polygon mask generated based on the process drawing; the width of the outer safety zone is used as the parameter "safety zone width" in millimeters, locked and recorded along with the version number.

[0102] The confidence threshold is set based on the alignment results of historical high-quantile statistics and on-site quality inspection consistency assessment. The threshold is locked as a parameter version and can be set as an independent value at the shift level for shifts experiencing changes in ambient light or dust levels, with the adjustment reason recorded. Shifts with confidence levels below the threshold enter an isolation zone awaiting review and do not participate in topology mapping. The isolation zone is a read-only storage domain, retaining a unified timestamp, idempotent key, reason code, and snapshot summary. The retention period is set as a parameter and locked as a version. The default settings for the observation window and out-of-order window can be set to the high-quantile value of the link end-to-end delay plus the clock jitter margin, and must not exceed the on-site buffer capacity. When the window width is adjusted, the parameter version and adjustment reason are recorded synchronously and written into the evidence chain. To constrain the latency of the registered link, the upper limit of latency is set to the sum of the control cycle multiple and the handshake round-trip latency. When the latency upper limit is reached, a limited number of retries are performed without changing the waterline order. If the retries still fail, a degradation record is stored with a unified timestamp and idempotent key, and the upstream is notified, while the degradation reason is retained.

[0103] The event table provides controlled query methods based on idempotent keys and unified timestamps. Within the same water level advancement cycle, the same record is returned for the same idempotent key, and out-of-order reading is prohibited. Cross-session queries do not produce side effects, and consistency is ensured by the idempotent key. Returned fields include machine coordinates, centroid deviation, confidence level, template version, transformation version, and unified timestamp. After downstream topology mapping is read, the original record must not be modified; only the read time and reader identifier are appended to form a traceable call chain. The reader identifier consists of the system identifier, station identifier, and software version, and is written into the evidence chain. The communication channel uses fieldbus or industrial Ethernet with handshake confirmation enabled. Handshake confirmation includes three states: request, confirmation, and completion. Timeout triggers a retry, and the number of retries can be set to a limit. After the retry limit is exceeded, a read-only alarm record is generated using the channel identifier, unified timestamp, and reason code and added to the evidence chain.

[0104] The security boundary stipulates that binding records can only be registered when the template is in an authorized state. The template authorization state is controlled by the permission policy table, which records the start and end times of its effectiveness. New registration is immediately blocked after the template's authorization is revoked. Existing records in the event table remain unmodifiable and only allow the addition of evidence chains for record keeping. When an area involving personal safety is occupied, the system does not trigger any subsequent processes related to the execution point. The system verifies the binding success rate, centroid deviation distribution, and field completeness rate. The sample size can be set to cover one production batch. The statistical scope of the production batch is fixed by the template, determined by shift or volume length, with the first to be reached being the standard and locked with the version. The statistical process and scope are recorded in the evidence chain with parameter versions and start and end times. Version locking covers the template version, template summary, transformation version, observation window parameter version, out-of-order window parameter version, and cause code set version. The cause code set is a public table with fixed numbers and meaning mappings, version-locked, and recorded with the evidence chain. The meanings of numbers one and four mentioned in this section are recorded along with the version number.

[0105] Texture fingerprints are fixed-length digests, which can be set to a fixed number of bits. When digest conflicts occur, remapping is performed by appending ordinal bits with idempotent keys, and the reason for remapping is recorded. In shifts with increased dust or decreased light intensity, a confidence threshold can be set and parameter versions can be locked to suppress false triggers. Through the above arrangements, event snapshots are bound within capabilities and scope to transform upstream discoveries into adjudicable objects using a unified time stamp and a unified coordinate system. Idempotent keys, version locking, and evidence chains ensure the order and traceability of calls, providing consistent, robust, and auditable preconditions for topology mapping, unified adjudication, and execution interlocking.

[0106] S4. Construct a topology map of the workstation, mold cavity, and discharge port; generate tokens for events; predict the arrival stroke according to the cycle time and determine the mapping; perform time verification and space verification. The specific implementation is as follows:

[0107] After completing the self-verification of dual-anchor time base, homogenized coordinate transformation, and event snapshot binding, topology mapping and verification are carried out. The goal is to establish a deterministic correspondence between each registered event and the stroke, die cavity, and discharge port under multi-station die-cutting cycle time, and to form an auditable record within a unified time axis. The topology graph is a set of directed relationships, with nodes representing stations, cavities, and discharge ports. Edges record the number of transmission steps and nominal delay from upstream to downstream nodes. The cycle time is measured in strokes, and the nominal delay is the time migration amount under that rhythm, preferably determined by the moving average of the most recent several merging cycles. The sliding window and smoothing method are locked in the general parameter table of the entire solution using the version number. The origin of the mechanical coordinate system is zeroed out by the alignment photoelectric reference point, the lateral orientation is close to the strip feed direction, and the longitudinal orientation is perpendicular to the feed direction, with the unit being millimeters. The orientation is locked in the template version.

[0108] Events progress in the graph in the form of tokens. Tokens are placeholders bound to idempotent keys, used to carry state transitions under a unified timeline. Idempotent keys consist of device identifier, anchor point identifier, window sequence number, template version, and fingerprint digest, used to uniquely identify events. Mapping is based on two types of checks: time checks are performed to verify the consistency of mechanical anchor phases and communication anchor phases. The mechanical anchor phases are derived from the mechanical timescales composed of the spindle encoder zero position, impulse signal, and alignment photoelectric sensor; the communication anchor phases are derived from the communication timescales composed of camera frame timemarks and control cycle timemarks. Spatial checks verify whether the mapped coordinates fall within the feasible region of the target cavity mask. The mask is expressed as a closed boundary sequence in the mechanical coordinate system. Boundary allowances and tolerances are given by the template version in millimeters and can be set to allow independent allowances for each side.

[0109] To ensure consistency, all timescales are in milliseconds, all rhythms are measured in strokes, transmission beats are measured in integer beats, and nominal delays are measured in milliseconds. Clock tolerance is the maximum allowable phase difference on a unified time axis, in milliseconds, and can be set to a range of two digits. It is distributed with the universal parameter table and entered into the version library by default. The upstream provided fields are idempotent key, machine coordinates, centroid deviation, timescale, and template version. In this step, a token is first generated for the idempotent key on the topology diagram. Based on the current beat and transmission beats, the arrival stroke and the corresponding mold cavity and discharge port are predicted, and time and space checks are immediately performed. Time check requires that the mechanical anchor phase and the communication anchor phase are consistent within the clock tolerance, and space check requires that the machine coordinates fall within the feasible region of the mask and that the boundary allowance meets the template requirements.

[0110] The merging time is the unified decision point when the water level reaches the current workstation. Its sequence number monotonically increases on the unified time axis and is used as the source of the time stamp for the mapping record. The merging period is the time interval between two adjacent merging times, in milliseconds. It is determined by the water level advancement rhythm and locked in the general parameter table of the whole case with the version number. The unified decision will batch pull the mapping table at the merging time according to the idempotent key and the time stamp. The pull adopts the request-response method and returns a consistency verification digest. The consistency verification digest is calculated by the aggregation of the idempotent key, time stamp, topology version, template version, coordinate transformation version, and dual-core conclusion. It is used as the tamper detection mark after the pull is returned and is written to the audit database together with the mapping record.

[0111] The single-event calculation latency is preferably controlled within a few milliseconds to tens of milliseconds, and can be set to complete within one cycle. If the verification fails, the process stops and the device is added to the verification queue. If the time verification fails, the fifth reason code is used; if the spatial verification fails, the sixth reason code is used. The verification queue is processed in ascending order of time markers. The maximum number of re-verification attempts for a token in the queue can be set to one to three, and the re-verification interval should not cross the current merging cycle. If coordinate transformation or topology version is missing, the token is suspended and marked with the reason for the version missing, and it does not participate in the current merging. When multiple records exist for the same idempotent key, only the first mapping record that passes the double verification is retained; subsequent records are recorded as duplicates and archived with the time marker. After verification, a mapping record is generated and written to a mapping table. The record fields include idempotent key, stroke count, mold cavity, discharge port, time verification conclusion, spatial verification conclusion, time marker, topology version, and template version. The mapping table is stored in the persistent area on the control side, and a composite index is established based on the idempotent key and time marker for batch reading during subsequent unified adjudication at the merging time.

[0112] Communication utilizes fieldbus or industrial Ethernet, supporting a limited number of retransmissions. Retransmissions employ a fixed number of retransmissions and a linear backoff strategy. The backoff step size and maximum number of retransmissions are provided in millisecond ranges and upper limits in the overall general parameter table. The backoff interval does not cross the current arrival beat; exceeding this limit freezes the token and places it in a review queue. Resource constraints ensure that mapping batches are completed within a single merging cycle, and link delays do not cross arrival beats. The arrival beat is the specified beat number predicted by the token in the target cavity according to the topology diagram; it is an integer beat number and does not change with communication retransmissions. To improve robustness in beat fluctuation scenarios, buffer nodes can be configured in the topology diagram to absorb small beat drifts. Buffer nodes are defined in the topology version with three parameters: location, allowed number of excessive beats, and start / stop thresholds. The start and stop thresholds are expressed in beat units, with a sampling granularity of single beats. The allowed number of excessive beats can be set to a range from single digits to two digits. Buffer nodes take effect when the cumulative deviation of consecutive beats reaches the start threshold and exit when the deviation falls back to the stop threshold.

[0113] The verification criteria include mapping hit rate and verification pass rate. Hit rate is the proportion of tokens that successfully reach the target cavity and pass the double verification. Verification pass rate is the proportion of tokens that complete both time verification and spatial verification and pass simultaneously. Preferably, the hit rate and pass rate are stable above a predetermined threshold. The threshold is loaded into the quality configuration, which carries the consistency, freshness, completeness, and credibility of the criteria in the quality quadruple. The versioned parameters for the threshold and weights are published by the quality configuration module and entered into the version repository. The evidence chain generates fragments synchronously when the mapping record is written. The fragments include idempotent keys, topology versions, template versions, coordinate transformation versions, timestamps, verification conclusions, and cause code summaries. The summary value is calculated as a tamper detection marker, which is then concatenated with the adjudication record and execution record in the audit database using idempotent keys. The audit database is a control-side add-only storage domain that supports composite retrieval using idempotent keys and timestamps. Once an entry is written, it can only be appended and cannot be overwritten or deleted.

[0114] The cause codes are globally enumerated, with their numbering and semantics maintained uniformly in the cause code table. This table is a versioned dictionary and incorporated into a universal parameter table. Time-based and spatial-based verification failures use fixed numbers, while other scenarios maintain consistent numbering for multi-segment sharing. The applicable boundary is multi-station die-cutting and multi-channel sorting of continuous roll materials, provided that the dual-anchor time base is calibrated, the homogenized coordinate transformation is in an available version, and the template version has been released and includes masks and tolerances. In terms of capabilities, it can cover topological relationships with multiple stations, multiple cavities, and multiple discharge ports. In terms of scope, it can cover multi-level transmission relationships and various cycle modes, maintaining mapping continuity under cycle step, short-term stop-start, and small tension fluctuation conditions. Safety boundaries require that the addition of topology versions, the start / stop of buffer nodes, and the adjustment of mask boundaries be authorized and traced. Any modifications take effect outside the merging cycle to avoid changing the mapping logic during the merging time. Parameter modifications are written to the version repository and a summary is generated. The summary and time stamp are jointly written into the evidence chain to ensure version locking.

[0115] S5. A unified decision is made when the water level reaches the merging point. Late events are replayed and corrected within the disordered window, and out-of-order records are isolated. The specific implementation is as follows:

[0116] To establish consistent and auditable control conclusions under the cycle conditions of multi-station die cutting, a dual-anchor time base is first established using mechanical anchors and communication anchors as a unified time axis. The mechanical anchor consists of the spindle encoder zero position, stroke signal, and alignment photoelectric sensor, while the communication anchor consists of camera frame time markers and control cycle time markers. A water level line is set on the unified time axis as an indicator of the adjudication time, and a merging time and merging period are defined. The merging time is a fixed point when the water level line reaches the reference phase of the workstation. Before production, the reference phase is calibrated by taking the median value of the stable phase of the alignment photoelectric sensor within a certain number of strokes and recording the drift range. The reference phase is then locked together with the merging period. The merging period is measured in strokes, preferably one to three strokes.

[0117] Records included in the merge set are limited to those that have completed time and space verification. Time verification is based on the consistency of the mechanical anchor phase and the communication anchor phase. Space verification is based on the mapping coordinates falling into the feasible region of the target cavity mask. The feasible region of the cavity mask is formed by shrinking the cavity geometric boundary by a fixed safety distance. The safety distance is a fixed proportional distance band based on the critical distance of the edge, in millimeters, and its proportional range is given in the template version. The mask can be generated by teaching positioning or drawing parameters. Teaching positioning is the process of collecting the cavity boundary and the station reference phase through teaching steps before production. Its step sequence, sampling number, and acceptance range are fixed in the template version and the version is locked.

[0118] All records undergo de-jittering, alignment, boundary padding, and integrity checks before entering the merge set. Timescales are uniformly measured in milliseconds, and rhythms are uniformly measured in strokes. To handle late arrivals and out-of-order events, an out-of-order window is set on a unified timeline as the time range within which playback corrections are allowed. The out-of-order window is measured in strokes, preferably two to five strokes. Whether a record falls into the window is determined by the stroke difference between the timestamp and the target merge time not exceeding the window width. Within the merge cycle, a window sequence number is assigned to each record. The window sequence number monotonically increases by strokes from the previous merge time and is written to an idempotent key for cross-module deduplication and order determination. The idempotent key serves as a unique identifier across the entire link, generated by combining the device identifier, anchor point identifier, window sequence number, template version, and fingerprint digest. The fingerprint digest is a short digest generated by a digest rule based on the local texture and geometric description of the event in a fixed field order. The digest rule is locked with the template version. Deduplication is performed by the idempotent key before writing. If a record with the same idempotent key and an older timestamp exists, the newer record is used to replace it, and a replacement marker is added to the evidence chain.

[0119] When the water level reaches the merging time, a one-time adjudication is performed on all valid records of the same workstation within the merging cycle, forming a unique conclusion and solidifying it. Late records within the out-of-order window enter the replay correction process based on their idempotent keys, revoking previous temporary conclusions and restating the adjudication within the merging set, with a maximum of one replay. Records exceeding the out-of-order window are marked with an isolation flag and enter the isolation queue. Isolated records do not enter the execution chain and are only used for auditing and spot checks. Removing isolation requires authorization and the generation of a review record. Each merging generates an adjudication record, which includes an idempotent key, adjudication conclusion, threshold version, mapping version, timestamp, and reason code, and is written to the adjudication table. The adjudication conclusion is limited to one of three categories: execution, release, or isolation. The threshold version is used to lock the quality criteria, the mapping version is used to lock the topology mapping and coordinate transformation, and the timestamp is the merging time measured by a unified time axis.

[0120] To handle competition within the same stroke or discharge port, a quality quadruple is established with a fixed priority order. The quality quadruple includes consistency, freshness, completeness, and reliability. Consistency is based on the consistency ratio of template fields, freshness is based on the time distance from the record to the merging time, completeness is based on the completeness rate of required fields, and reliability is based on the detection confidence level combined with the time and space verification pass status. The priority order is fixed as consistency, freshness, completeness, and reliability, and is locked according to the threshold version. If the records cannot be uniquely sorted according to this order within the disordered window, or if there is an irresolvable overlap in resource usage with the submitted records after replay, it is judged as an unresolved conflict.

[0121] To control resources and latency, batch decisions are implemented on the record set covered by a single merge, with a maximum batch decision latency set, preferably five to twenty milliseconds. The decision table is stored in partitions based on workstation identifiers and timestamps. The execution stage reads data in a pull manner according to the workstation rhythm. Communication uses fieldbus or industrial Ethernet, with a maximum link latency set, preferably ten milliseconds. The pull order is ascending by timestamp within the workstation partition, and then idempotent key lexicographical order within the same timestamp. Lexicographical order is a sorting rule where idempotent key fields are compared in a fixed order. The field order and comparison method are locked in the threshold version. Records with incomplete fields are not enqueued, and a reason code is retained. To reduce congestion at the end of the window, a secondary merge time can be set in the out-of-order window as an alternative. Enabling and disabling this is controlled by the threshold version and recorded in the evidence chain.

[0122] To ensure safety, merging and adjudication at a workstation are suspended when the safety interlock is not closed or personnel are in a hazardous area. The safety interlock consists of an access control switch and a light-emitting diode link; merging submissions are prohibited when the interlock is not closed. After closure, the queue is restored in ascending order of time stamps, and skipping to higher time stamps is not allowed. Reason codes are managed using closed sets and are version-locked. Window timeout is number seven, unresolved conflict is number eight, safety pause is number eleven, and the maximum number of adjudication retryes is twelve. The number and interpretation are written into the reason code dictionary and locked with the version. Adjudication transactions are committed serially. If a submission fails, it is retried within a limited number of times, preferably one to three times. If the retry still fails, reason code twelve is recorded and fixed to the evidence chain.

[0123] The evidence chain serves as a traceable path from discovery to execution, including idempotent keys, coordinate transformation versions, topology versions, threshold versions, adjudication records, execution records, cause code sequences, and replacement markers. It is stored in an audit database with versioning and tamper-proof capabilities and a composite index is built based on idempotent keys and time stamps. Periodic snapshots and verification are preferably enabled. Statistical verification uses the adjudication table and execution records as the criteria, and sets three indicators for periodic evaluation: missorting rate, missed sorting rate, and playback success rate. The statistical window can be set to one shift or one roll of material. The audit end can perform random checks and recalculations according to the evidence chain to verify statistical consistency.

[0124] To ensure consistency, the merge set only includes records that pass time and space verification. Template version, threshold version, mapped version, reason code dictionary, merge cycle, out-of-order window, and secondary merge time are all locked before activation. Any changes must be authorized and recorded. The recorded content includes the modifier, modification time, version difference summary hash, and automatic association with the current evidence chain. The execution phase only adds records to the queue when all fields of the adjudication record are complete. Within the latency and concurrency constraints that on-site resources can tolerate, the unified adjudication has the capabilities of single-point merging, full consistency, window replay, version recording, idempotent deduplication, competitive adjudication, safe pause and sequential recovery, serialized submission and timed retries, statistical verification and audit trail. Its applicable scope covers working conditions such as speed jumps, stop and start, tool changes, and tension fluctuations, and it is preferably designed to maintain stable performance in large-scale continuous production.

[0125] S6. Implement two-stage interlocking triggering based on the mapping. If the interlocking conditions are not met, activate the mis-period protection, and simultaneously solidify the evidence chain, recording anchor points, templates, thresholds, mappings, and time stamps. The specific implementation is as follows:

[0126] Two-stage interlocking triggering serves as a control order for execution safety and timing consistency, running through both the pre-arm and submission phases. It uses the arrival time marker as the sole time reference, the idempotent key as the sole index, and the water level advancement as the sole adjudication sequence, achieving deterministic actions and verifiable traceability of events without disrupting the rhythm. The control cycle refers to the time rhythm of the controller task, measured in milliseconds; the arrival time marker reserved slot refers to the remaining time interval from the current evaluation time to the arrival time marker boundary, preferably not less than one control cycle.

[0127] The time synchronization source can be set to a unified hardware clock or a certified time synchronization service. The time synchronization period is preferably a number of control cycles. If time synchronization fails, the system enters a protection state and generates a reason code. The pre-arm window, submission time slot, retry interval, and retry limit can be set to be determined by a joint function of the clock cycle, link delay budget, and execution device response capability, preferably not exceeding the reserved time slot for the arrival clock cycle. The retry interval can be set to one to three times the control cycle. The retry limit is jointly constrained by the reserved time slot and the communication quota and does not exceed the maximum number of times allowed by the quota. Retry can be set to linear backoff or mild backoff. The backoff step size does not exceed the upper limit of the control cycle and is constrained by the reserved time slot.

[0128] During the pre-arm phase, the execution side retrieves the latest adjudication record and mapping record from the adjudication table using the idempotent key, verifies the consistency of the threshold version, topology version, coordinate transformation version, and template version, registers the time stamp, resource occupancy, and interlock channel status in the execution buffer, and periodically reviews them within the pre-arm window; the interlock channel performs self-checks according to a fixed rhythm, and if the self-check fails, it remains in a rejected state and submission is prohibited until the self-check recovers and leaves a trace; the safety gate serves as a high-priority criterion for the interlock channel, and if the health is abnormal, it directly rejects the submission and triggers a dedicated reason code and alarm.

[0129] During the submission phase, within the submission time slot of the arrival shot, an action command is issued only when all three conditions are met simultaneously: mapping validity, position interlock establishment, and playback conflict resolution. This command is transmitted to the actuator via fieldbus or industrial Ethernet and the submission time, interlock status, device feedback code, and action flag are recorded. Mapping validity means that the mapping record passes time and space verification within the topology diagram and references the current topology version and coordinate transformation version. Position interlock establishment means that the interlock channel formed by the cavity positioning measurement, proximity sensing, and safety door signal is in a closed state. Interlock degradation judgment can be set to the interlock channel failing to close for several consecutive control cycles, preferably with a threshold of three to five control cycles. When the threshold is reached, the independent hardware interlock takes over the position judgment criterion, and the software interlock is downgraded to consistency verification and logging. Playback conflict resolution means that the correction process within the out-of-order window has been completed and the event has not been canceled or reassigned.

[0130] If any precondition is not met, the mis-cycle protection is triggered, canceling the current action and preserving the state snapshot and reason explanation. The link latency is the transmission and device execution time between submission and confirmation, limited by the resource quota table and must not cross the submission time slot. If no confirmation is received, execution is performed according to the retry interval until the retry limit is reached. If it still fails to reach the limit, the mis-cycle protection is maintained and over-time compensation is prohibited. The communication link can tolerate short-term packet loss and is compensated according to the retry strategy. The tolerance time for continuous packet loss should not exceed the reserved time slot for the arrival time slot. The resource quota table uniformly manages the communication occupancy limit, computing load limit, execution port concurrency limit, and retry budget. If any quota item reaches the limit, it enters the protection state and generates a reason code. The protection state only allows recording and tracekeeping and does not trigger submission. Concurrent conflicts are serialized according to time stamp order and quality priority, maintaining the uniqueness of idempotent keys and the consistency of adjudication. The quality priority order is consistent with the priority caliber of the quality quadruple on the adjudication side.

[0131] Parameter and rule changes take effect uniformly at the water level merging time, without spanning the current arrival time. The version dictionary implements version locking for thresholds, topology, coordinate transformations, interlocking logic, and reason code dictionaries. Each change generates a summary verification value and associates it with an idempotent key. When the version dictionary is inconsistent with the time base record or the topology map enters the effective window for switching, a rollback can be triggered. The rolled-back version number and timestamp string are connected to the evidence chain with the same idempotent key to maintain path continuity. The execution record includes idempotent key, execution port identifier, pre-arm time, submission time, interlock status, action flag, reason code, device feedback, version reference, and timestamp, and is jointly generated with the time base record, transformation record, mapping record, and adjudication record to generate an evidence chain summary. The action flag can be set to trigger locally, cancel, replayed, isolated, and security gate intercepted, and is managed separately from the device feedback code. The device feedback code can be set to execute, reject, busy, hard fault, soft fault, and unknown, and is associated with the error cycle protection strategy.

[0132] The minimum set of evidence can be set as an idempotent key, timestamp, version number set, reason code, action flag, device feedback, and a summary check value based on the sequence relationship. The summary check value is a check flag obtained by performing a stable summary on the minimum set. Field fingerprints are stable summaries generated according to a fixed field order and field values. The above summaries are used for consistency verification and are not limited to specific implementation methods. The evidence chain is stored in a named partition of the audit database. The named partition is organized by work position, date, and batch. The composite index is constructed using idempotent keys and timestamps to achieve fast traceability. Audit database records are retained within the compliance period, and sensitive fields are de-identified and stored according to the principle of minimum necessity.

[0133] Interface fields are uniformly structured key-value sets. The execution side extracts them from the adjudication table using idempotent keys and writes them to the audit database. Confirmation return fields and retry counts are updated along with the submission side records. Cross-network communication uses certified encrypted channels. Parameter and rule modifications must be authorized and logged before they can take effect in the version dictionary. Manual resets require authorization level and double confirmation. Reset times must not fall within the submission time slot. Reset operations are written into the evidence chain and generate corresponding reason codes. Reason codes use a unified dictionary and can be set to key duplication, clock difference exceeding limits, channel disconnection, time verification failure, space verification failure, window timeout, version missing, resource conflict, interlock failure, incorrect period triggering, device refusal to operate, communication congestion, and unresolved replay conflicts.

[0134] The inspection criteria include the occurrence rate of error cycles and the completeness rate of the evidence chain. The sample size covers multiple batches, the entire speed range, and a set of operating conditions involving tool changes and start-stop operations. Preferably, the above indicators meet the established quality objectives and process constraints. Through the above-mentioned unified terminology and criterion design, the two-stage interlocking trigger completes the pre-arm positioning, condition verification, submission triggering, backoff retry, error cycle protection, and evidence solidification under limited time and resource conditions. It also maintains a system-level closed loop that is related to the water level adjudication, topology mapping, and homogenization coordinate transformation, forming an implementable, auditable, measurable execution mechanism that meets the safety and compliance boundaries.

[0135] Example 2: Figure 2 A schematic diagram of the online detection and sorting control system for electrode die-cutting defects of the present invention is provided. The online detection and sorting control system for electrode die-cutting defects includes:

[0136] Time base anchoring and keying module: used to establish a dual-anchor time base for mechanical anchors and communication anchors, generate idempotent keys, set water level lines and out-of-order windows, and serve as a reference for adjudication and evidence storage;

[0137] Coordinate transformation self-verification module: It is used to maintain the homogeneous coordinate transformation based on tension, speed and temperature observations, and triggers perturbation identification and updates parameters during speed increase, deceleration and tool change to form alignment convergence constraints.

[0138] Event binding module: Used to receive snapshots of detected events, map them to time-based machine coordinates according to coordinate transformation, and complete the binding registration with idempotent keys to ensure the uniqueness of events;

[0139] Topology mapping verification module: used to construct a topology map of the workstation, mold cavity, and discharge port, generate tokens for events, predict the arrival number of strokes and discharge port according to the cycle time, and perform time and space verification;

[0140] Decision replay module: Used to execute a unified decision when the water level reaches the merging time. Late events are replayed and corrected within the out-of-order window, and isolated events are executed outside the window to maintain decision consistency.

[0141] The interlock audit module is used to implement two-stage interlock triggering based on mapping. When the interlock conditions are not met, the error cycle protection is activated. At the same time, the evidence chain is solidified, and anchor points, templates, thresholds, mappings and time stamps are recorded.

[0142] The technical connections and implementation logic of the six modules are as follows:

[0143] The time-base anchoring and keying module establishes a dual-anchor time base consisting of mechanical and communication anchors, generates idempotent keys, sets water level lines and disordered windows, and forms a benchmark for adjudication and evidence storage, serving as a unified time and identity constraint for subsequent modules. The coordinate transformation self-verification module maintains homogenized coordinate transformation based on tension, speed, and temperature observations, triggering perturbation identification and parameter updates during acceleration, deceleration, and tool changes, stably mapping pixel coordinates under a unified time axis to mechanical coordinates. The event binding module receives snapshots of detected events, transforms them according to the current coordinates to obtain time-base-based mechanical coordinates, and completes binding registration with idempotent keys, ensuring that each event enters the control chain under a unified time axis and unique identity. The topology mapping verification module constructs a topology map based on the workstation, mold cavity, and discharge port, generates tokens for bound events, predicts the arrival stroke and discharge port according to the cycle time, and executes the verification. Inter-space and spatial verification are performed to generate mapping records and write back version references as adjudication readings. The adjudication playback module performs unified adjudication on the mapping records of the same workstation at the time the water level reaches the merging point. Late records are replayed and corrected within the out-of-order window, isolated by the out-of-window marker, and the adjudication record is output and the reason code and version number are fixed to ensure the consistency of adjudication in terms of timing and caliber. The execution interlock audit module implements two-stage interlock triggering according to adjudication and mapping. When the conditions of mapping validity, position interlock establishment, and playback conflict elimination are met, the action is submitted. When the conditions are not met, the error cycle protection is activated. At the same time, the anchor point, template, threshold, mapping, time stamp, and device feedback are merged into an evidence chain and written into the audit library. The evidence chain and version dictionary inversely constrain the time base, transformation, and mapping caliber to form a closed-loop control and verifiable traceability logic from discovery to execution.

[0144] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0145] It should be noted that this invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0146] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wireless or wired transmission; wired transmission methods include optical fiber, twisted pair, coaxial cable, etc.; wireless transmission includes infrared, microwave, etc. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center containing one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0147] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0148] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0149] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0150] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0151] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0152] 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.

[0153] In conclusion, 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, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An online detection, sorting, and control method for electrode die-cutting defects, characterized in that, include: S1. Establish a dual-anchor time base for mechanical anchors and communication anchors, generate idempotent keys, and set water level lines and disordered windows as the basis for adjudication and evidence preservation. S2. Based on tension, speed and temperature observations, maintain homogenized coordinate transformation. Trigger perturbation identification and update transformation parameters when speeding up, decelerating or tool changing occurs. S3. Map the event snapshots generated by the detection to time-based machine coordinates according to coordinate transformation, and bind and register them with idempotent keys; S4. Construct a topology map of the workstation, mold cavity, and discharge port; generate tokens for events; predict the arrival number of strokes according to the cycle time and determine the mapping; and perform time and space checks. S5. Make a unified decision when the water level reaches the merging time. Late events are replayed and corrected within the disordered window, and out-of-window records are isolated. S6. Implement two-stage interlocking triggering based on the mapping. If the interlocking conditions are not met, enable the wrong cycle protection. At the same time, solidify the evidence chain and record the anchor point, template, threshold, mapping and time mark. S1 includes: Align the mechanical anchor and the communication anchor to establish a dual-anchor time base and form a unified time axis; Set the water level line for sequential advancement and the disordered window; The idempotent key is used as the unique identifier for the event. The idempotent key consists of the device identifier, anchor identifier, window number, template version, and fingerprint digest. Establish a time base record table with an idempotent key as the primary index, and write the water level position, out-of-order window parameters, and clock difference monitoring values ​​into the table. Set isolation flags and prevent writing to records that are earlier than the waterline position and later than the upper limit of the out-of-order window; For duplicate records with the same idempotent key, retain the first valid record. When either the anchor source is continuously missing or the clock error exceeds the limit, the operation will be downgraded to a single anchor working state and the water level line will be maintained for advancement. Changes involving time base, out-of-order windows, template versions, and idempotent keys are implemented with version locking and written into the evidence chain.

2. The online detection, sorting, and control method for electrode die-cutting defects according to claim 1, characterized in that, S2 include: Establish a homogenized coordinate transformation under a dual-anchor time base and implement online self-verification; The affine mapping from pixel coordinates to machine coordinates is continuously corrected based on observations of tension meter, spindle speed, and frame temperature. The perturbation excitation, constrained by the safety budget, is applied to identify the events that occur during acceleration, deceleration, and tool change. When the convergence criterion is met within the observation window, the transformation parameters are updated, a transformation version is generated, written to the version repository and locked in a read-only state, and version switching is performed using atomic replacement. Time and space verification call the current transformation version; Records are linked one-to-one with idempotent keys and written into the evidence chain.

3. The online detection, sorting, and control method for electrode die-cutting defects according to claim 2, characterized in that: When the tension channel loses connection, switch the virtual reference grid and set the degraded annotation status; If the time verification fails, roll back to the previous stable transformation version and hold until the next observation window converges; If the spatial verification fails, roll back to the previous stable transformation version and hold until the next observation window converges; When the concurrent flow reaches the quota limit, flow restriction is implemented according to the water level advancement order. Late records are entered into the disordered window and corrected at the next merging time. The rule table implements version locking for thresholds, number of attempts, priority, merge cycle, out-of-order window, and retry limit, and is fixed in sync with the changed version.

4. The online detection, sorting, and control method for electrode die-cutting defects according to claim 1, characterized in that, S3 include: When the dual-anchor time base and homogenized coordinate transformation are in an authorized state, the event snapshot is de-jittered, time-corrected, and missing data is filled in according to the template in the authorized state, and then verified by the template. The centroid is mapped to mechanical coordinates according to the homogenization coordinate transformation, and the centroid deviation is determined by the displacement of the centroid relative to the geometric center of the bounding box. Registered with an idempotent key containing device identifier, anchor point identifier, window sequence number, template version and fingerprint digest, generating binding records containing mechanical coordinates, centroid deviation, confidence level, unified timestamp, template version and transformation version, and writing them to the event table in an append-only manner. The event table uses the idempotent key as the primary index and the unified timestamp as the secondary index. Within the same water level advance cycle, only the earliest recorded record is retained according to the water level advance order. The earliest record with the unified time stamp is used as the standard. If the unified time stamps are the same, the records are compared sequentially according to the fixed order of the idempotent key fields to determine the earliest record. The remaining records are marked as duplicates.

5. The online detection and sorting control method for electrode die-cutting defects according to claim 1, characterized in that, S4 include: After the dual-anchor time base, homogenized coordinate transformation and event binding are completed, a directed topology graph consisting of workstations, mold cavities and discharge ports is constructed, where the edges are used to record the number of transmission steps and the nominal delay. For events identified by idempotent keys, generate tokens, predict the arrival number of strokes based on the cycle time and transmission number, and determine the target cavity and target discharge port; Perform time verification and space verification. The passing condition for time verification is that the mechanical anchor phase and the communication anchor phase are consistent. The passing condition for space verification is that the mechanical coordinates fall into the feasible region of the mold cavity mask and the boundary allowance meets the template version requirements. Only the first mapping record that passes both time and space verification is retained. The mapping record, along with the time stamp, topology version, and template version, is written into the mapping table. A composite index is created using the idempotent key and the time stamp. At the time of merging, the mapping table is read based on the composite index to execute the decision.

6. The online detection, sorting, and control method for electrode die-cutting defects according to claim 1, characterized in that, S5 include: Under a unified timeline, a one-time ruling is made on records of the same work station at the moment the water level reaches the merging point. Only mapping records that have passed time and space verification are included; late records in the out-of-order window are replayed and corrected using the idempotent key. The previous conclusion is revoked; items exceeding the out-of-order window are assigned an isolation marker; competition within the same stroke is adjudicated according to the fixed priority order of the quality quadruple. Competition at the same discharge port is adjudicated in a fixed order of consistency of discharge diameter, freshness, integrity and credibility. The adjudication record contains the idempotent key, adjudication conclusion, threshold version, mapping version, timestamp and cause code, and is written into the adjudication table and the chain of evidence. The decision submission is serialized and retried within the decision retry limit.

7. The online detection, sorting, and control method for electrode die-cutting defects according to claim 1, characterized in that, S6 include: Establish a two-stage interlocking triggering order; During the pre-arm phase, the adjudication record and mapping record are read from the adjudication table based on the idempotent key within the pre-arm window. The consistency of the threshold version, topology version, coordinate transformation version, and template version is checked, and the execution buffer is registered with time markers, resource occupancy, and interlock channel status. During the submission phase, within the submission slot of the captured image, an action command is issued only when the mapping is valid, the position interlock is established, and the playback conflict is eliminated simultaneously. If any of these conditions are not met, the error cycle protection is triggered and the evidence chain is solidified, prohibiting overshoot compensation.

8. The online detection, sorting, and control method for electrode die-cutting defects according to claim 1, characterized in that: Link latency and concurrency are capped by a resource quota table; If no confirmation is received, retry will be performed according to the preset backoff sequence within the submission time slot. If the retry limit is reached and the failure is still not achieved, error cycle protection will be maintained. The interface field is written to the audit database with an idempotent key as the unique index, and the action flag, device feedback, and reason code are registered. Parameter and rule modifications take effect at the time of water level merging, are rolled back according to the version dictionary and leave a trace with the summary verification value, and cross-network communication uses an encrypted channel.

9. An online detection and sorting control system for electrode die-cutting defects, used to implement the online detection and sorting control method for electrode die-cutting defects as described in any one of claims 1-8, characterized in that, include: Time base anchoring and keying module: used to establish a dual-anchor time base for mechanical anchors and communication anchors, generate idempotent keys, set water level lines and out-of-order windows, and serve as a reference for adjudication and evidence storage; Coordinate transformation self-verification module: It is used to maintain the homogeneous coordinate transformation based on tension, speed and temperature observations, and triggers perturbation identification and updates parameters during speed increase, deceleration and tool change to form alignment convergence constraints. Event binding module: Used to receive snapshots of detected events, map them to time-based machine coordinates according to coordinate transformation, and complete the binding registration with idempotent keys to ensure the uniqueness of events; Topology mapping verification module: used to construct a topology map of the workstation, mold cavity, and discharge port, generate tokens for events, predict the arrival number of strokes and discharge port according to the cycle time, and perform time and space verification; Decision replay module: Used to execute a unified decision when the water level reaches the merging time. Late events are replayed and corrected within the out-of-order window, and isolated events are executed outside the window to maintain decision consistency. The interlock audit module is used to implement two-stage interlock triggering based on mapping. When the interlock conditions are not met, the error cycle protection is activated. At the same time, the evidence chain is solidified, and anchor points, templates, thresholds, mappings and time stamps are recorded.

Images