A linkage execution control method with adaptive time delay compensation

Abstract

The application provides a linkage execution control method with adaptive time delay compensation, and relates to the technical field of automation control. The application receives surface state sensing data stream and real-time production line control data stream, extracts working condition speed extreme value difference and basic defect risk identification; compares the same with limit fluctuation threshold value, dynamically determines visual detection mode and obtains associated linkage execution time delay compensation amount; processes the same by using execution time sequence reference addition, generates dynamic pre-trigger compensation time length, and generates comprehensive defect judgment state in combination with defect identification; and finally, based on execution mechanism response failure limit check result, generates synchronous or global control signaling and issues the same to linkage rejection unit. The application solves the pain point of heterogeneous instruction mapping dislocation in limited variable speed environment through processing reduced frequency and feedforward time delay targeted compensation, and realizes dynamic drift compensation and accurate synchronous rejection of cross-domain time sequence reference.

Description

Technical Field

[0001] This invention relates to the field of automation control technology, specifically to a linkage execution control method with adaptive time delay compensation. Background Technology

[0002] In modern industrial manufacturing, especially in continuous processing fields such as metal rolling, machine vision has become a core supporting technology for quality inspection. With the deep integration of plant-wide automation control and image analysis algorithms, quality control is evolving from static, isolated offline feature extraction to highly integrated, real-time closed-loop collaborative control. In constrained variable-speed rolling production environments, mapping visual inspection features to physical execution nodes faces severe challenges in spatiotemporal decoupling.

[0003] Existing technologies (such as patent document CN121934510A, entitled "A Multi-station Collaborative Intelligent Control System for Synchronous Coating of Multiple Roll Materials") propose a control scheme based on the Dec-POMDP multi-agent collaborative decision-making framework. This scheme introduces a fuzzy sliding mode control strategy and utilizes adaptive fuzzy logic to generate compensation quantities for dynamic disturbances during system operation. While this type of scheme possesses a certain degree of anti-interference capability in macroscopic continuous collaborative operations, its adaptability to dynamic implementation conditions remains significantly limited when facing discrete fixed-point execution (defect linkage elimination) scenarios involving continuously flowing materials.

[0004] Existing disturbance compensation technologies largely rely on system-level macroscopic estimation using adaptive fuzzy logic, failing to accurately decouple and quantify the "variance of high-load data processing time at the front-end sensing nodes" from the "inherent mechanical hysteresis of the end-effector (pneumatic components)" at the physical level. This fuzzy compensation cannot be extracted as a precise absolute time-domain lead.

[0005] Existing technologies generally rely on comparing the execution result with a preset trajectory to generate a deviation value, and then performing real-time corrections (proportional-resonant control). In variable-speed material flow conditions, for the "check and reject" rejection action, the open-loop network transmission delay combined with physical inertia will cause the post-correction to always lag behind the physical execution cycle. Summary of the Invention

[0006] The purpose of this invention is to provide a linkage execution control method with adaptive delay compensation, which constructs a temporal collaborative processing hub between visual features and underlying mechanisms at the application layer. By extracting the extreme difference in operating speed as the core control valve, the image processing variance on the software side is forcibly converged under extreme operating conditions, and feedforward superposition is performed using the linkage execution delay compensation amount to systematically offset the inherent hysteresis across nodes. This addresses the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A linkage execution control method with adaptive time delay compensation includes the following steps:

[0009] Step S1: Receive the surface state perception data stream of the target material in transit and the real-time production line control data stream from the front-end visual perception node; extract the extreme difference of the working speed characteristic of the dynamic acceleration and deceleration limit boundary working conditions of the main drive equipment based on the real-time production line control data stream;

[0010] Extract the basic defect risk identifier from the surface condition sensing data stream; compare the extreme difference of the working condition speed with the set limit fluctuation interception threshold to dynamically determine the detection mode status identifier corresponding to the surface condition sensing data stream; and obtain the linkage execution delay compensation amount associated with the detection mode status identifier.

[0011] Step S2: Using the extreme difference of the working condition speed and the linkage execution delay compensation amount as input, and the detection mode status identifier as the mapping index, call the linkage execution delay compensation amount that matches the detection mode status identifier to perform timing benchmark summation processing, and generate a dynamic advance trigger compensation duration characterizing the physical action advance amount benchmark; and associate the dynamic advance trigger compensation duration with the basic defect risk identifier to generate a comprehensive defect judgment status;

[0012] Step S3: Verify the extreme difference of the working condition speed with the set failure limit of the actuator response, and generate a synchronous control signal or global control command to trigger the linkage elimination control unit based on the comprehensive defect judgment status and the verification result.

[0013] Specifically, the synchronous control signal is sent to the downstream linkage rejection control unit, or the global control command is sent to the upper-level bus, in order to eliminate the coordination delay between the front-end visual perception node and the linkage rejection control unit on the application layer flow control unit side. On this basis, the effective time of the issued synchronous control signal can be dynamically synchronized with the real-time transmission position of the target flow material.

[0014] Compared with the prior art, the beneficial effects of the present invention are:

[0015] By extracting the extreme value difference of the operating speed representing the dynamic acceleration and deceleration limit boundary and comparing it with the limit fluctuation interception threshold, the detection mode status identifier is dynamically determined. This mechanism can proactively intercept the full-element feature traversal comparison path and reduce the dimensionality to shallow feature extraction when the calender acceleration undergoes a sudden change or the physical response hysteresis nonlinearity increases dramatically. This reduces the variance in image processing time under high concurrency loads and improves the robustness of the control system to extreme disturbances.

[0016] Using the extreme difference in operating speed and the amount of compensation for linkage execution delay as input, the timing benchmark is called to sum and generate a dynamic pre-trigger compensation duration, which is then associated with the defect risk identifier to generate a comprehensive judgment status. Network jitter and flow lag are transformed into proactive advance correction benchmarks, absorbing inherent system-level delays in advance through feedforward superposition, effectively eliminating the resistance of the mapping step size lagging behind the physical execution cycle.

[0017] The extreme difference in operating speed and the failure limit of the actuator response are rigidly verified, and synchronous control signals or global blocking commands are distributed accordingly. Under normal conditions, cross-node coordination delays are eliminated to ensure dynamic synchronization between rejection commands and material positions; when physical tolerances are exceeded, blocking signals are decisively issued to force a global flow state degradation. This avoids control timing misalignment conflicts between the front-end visual perception nodes and the linkage rejection control unit. Attached Figure Description

[0018] Figure 1 This is a topology diagram of the technical route of the adaptive and collaborative industrial vision inspection and linkage execution control method of the present invention;

[0019] Figure 2 This is a schematic diagram of the technical logic of steps S1 to S3 of the present invention;

[0020] Figure 3 This is a schematic diagram showing the results of a numerical verification experiment performed on this method;

[0021] Figure 4 A schematic diagram illustrating the deep feature space mapping performed on a surface state-aware data stream. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0023] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various elements, but unless otherwise stated, these elements are not limited by these terms. These terms are used only to distinguish one element from another.

[0024] Example 1:

[0025] Please see Figures 1 to 4 The present invention provides a technical solution:

[0026] A linkage execution control method with adaptive time delay compensation, the method being executed by the application layer flow control unit, includes the following steps:

[0027] Step S1: Receive the surface state perception data stream of the target material in transit and the real-time production line control data stream from the front-end visual perception node; extract the extreme difference of the working speed characteristic of the dynamic acceleration and deceleration limit boundary working conditions of the main drive equipment based on the real-time production line control data stream;

[0028] It should be noted that the surface condition sensing data stream in this embodiment includes the original detection image; the target material being transferred is a continuous aluminum sheet and strip; the real-time production line control data stream is a real-time rolling control data stream; and the main drive equipment is a rolling mill.

[0029] Extract the basic defect risk identifier from the surface condition sensing data stream; compare the extreme difference of the working condition speed with the set limit fluctuation interception threshold to dynamically determine the detection mode status identifier corresponding to the surface condition sensing data stream; and obtain the linkage execution delay compensation amount associated with the detection mode status identifier.

[0030] Step S2: Using the extreme difference of the working condition speed and the linkage execution delay compensation amount as input, and the detection mode status identifier as the mapping index, call the linkage execution delay compensation amount that matches the detection mode status identifier to perform timing benchmark summation processing, and generate a dynamic advance trigger compensation duration characterizing the physical action advance amount benchmark; and associate the dynamic advance trigger compensation duration with the basic defect risk identifier to generate a comprehensive defect judgment status;

[0031] Step S3: Verify the extreme difference of the working condition speed with the set failure limit of the actuator response, and generate a synchronous control signal or global control command to trigger the linkage elimination control unit based on the comprehensive defect judgment status and the verification result.

[0032] Specifically, the synchronous control signal is sent to the downstream linkage rejection control unit, or the global control command is sent to the upper-level bus, in order to eliminate the coordination delay between the front-end visual perception node and the linkage rejection control unit on the application layer flow control unit side. On this basis, the effective time of the issued synchronous control signal can be dynamically synchronized with the real-time transmission position of the target flow material.

[0033] Further defining the steps of extracting the extreme value difference of the working speed characterizing the dynamic acceleration and deceleration limit boundary conditions of the main drive equipment based on the real-time production line control data stream, and extracting the basic defect risk identifier and dynamically determining the detection mode state identifier corresponding to the surface state perception data stream, specifically includes:

[0034] The real-time production line control data stream is analyzed, and the absolute value difference between the macroscopic speed peak parameter and the macroscopic speed trough parameter within the set monitoring period is calculated to generate the extreme value difference of the operating condition speed.

[0035] In response to the extreme difference of the working condition speed exceeding the set limit fluctuation interception threshold, the full feature traversal comparison path is intercepted at the application layer, and the shallow contour risk identifier is extracted as the basic defect risk identifier.

[0036] The detection mode status identifier of the surface state perception data stream is updated to the shallow dimension reduction visual detection mode simultaneously.

[0037] In response to the fact that the extreme difference of the working condition speed does not exceed the set limit fluctuation interception threshold, the full feature traversal comparison path is maintained to extract the standardized multi-dimensional feature cluster vector as the basic defect risk identifier, and is simultaneously updated to the full-volume deep visual detection mode.

[0038] Further specifying, the step of obtaining the linkage execution delay compensation amount associated with the detection mode status identifier specifically includes:

[0039] Read the historical control execution log, extract the pre-judgment time marker and the action confirmation time marker, and calculate and generate the absolute time difference;

[0040] If the comprehensive defect judgment state is not generated or the downstream action confirmation feedback is not received within the set control silence time window, a virtual detection synchronization signal with a terminal physical action shielding identifier is generated and sent to the linkage elimination control unit.

[0041] Extract the round-trip response time of the virtual probe synchronization signal transmitted via the control bus;

[0042] According to the detection mode status identifier corresponding to the extracted absolute time difference or round-trip response time, the absolute time difference and the round-trip response time are isolated and archived together.

[0043] Calculate the average value of the isolated archived data, and generate and dynamically refresh the linkage execution delay compensation amount corresponding to the status identifier of each detection mode.

[0044] Further specifying the steps for performing time-series baseline summation and generating a comprehensive defect determination status, the specific steps include:

[0045] The extracted detection mode status identifier is used as the hierarchical retrieval identifier. The corresponding linkage execution delay compensation amount is matched and called in the global delay configuration table, and it is used as the target execution delay compensation amount.

[0046] The target execution delay compensation amount is used as an advance compensation factor and is added to the current global control coordination benchmark in advance to generate the dynamic advance trigger compensation duration.

[0047] In response to the detection mode status being identified as full-scale deep visual detection mode, based on a preset defect semantic evaluation model, the image feature matrix is ​​extracted according to the basic defect risk identifier and forward reasoning is performed to output the matching target hazard weight.

[0048] In response to the detection mode status being identified as a shallow dimensionality reduction visual detection mode, the basic defect risk identifier is directly converted into an extreme target hazard weight through application layer static mapping bypass. Based on the pre-configured control parameter normalization coefficient, a scale alignment mapping operation is performed on the target hazard weight and the dynamic pre-trigger compensation duration to generate dimensionless priority evaluation parameters and time-series constraint parameters. Based on the process safety tolerance boundary, a cross-threshold comparison is performed on the priority evaluation parameters and time-series constraint parameters to map and generate the comprehensive defect judgment status.

[0049] Further defining the steps of sending the synchronization control signaling to the downstream linkage rejection control unit or sending the global control command to the upper-level bus, specifically includes:

[0050] In response to the fact that the extreme difference of the working condition speed does not exceed the set failure limit of the actuator response, for the current material flow batch corresponding to the target flow material, the start and end time window of the rejection action used to constrain the linkage rejection control unit is locked, and the state synchronization is triggered in the global scheduling unit to map the control scheduling state of the current material flow batch to the high priority interception confirmation state, and the synchronization control signal is generated and issued based on the above state.

[0051] In response to the extreme difference in operating speed exceeding the set failure limit of the actuator response, a blocking signal is generated as the global control command; based on the blocking signal, the dynamic generation process of the dynamic pre-trigger compensation duration is intercepted, the pre-configured system preset static compensation constant is forcibly rolled back to replace the dynamic pre-trigger compensation duration, and the subsequent detection mode status identifier is forcibly downgraded and locked to the shallow dimensionality reduction visual detection mode.

[0052] Further defining the step of triggering state synchronization within the global scheduling unit to map the control scheduling state of the current material flow batch to a high-priority interception confirmation state, specifically includes:

[0053] Read the current concurrent task sequence in the global scheduling unit and extract the scheduling identifier that represents the processing order of each associated collaborative task;

[0054] In response to the generated synchronization control signaling, a queue-jumping control beacon is constructed for the current material flow batch;

[0055] Based on the queue-jumping control beacon, the scheduling identifier corresponding to the current material flow batch is overwritten to prioritize the logical execution of the current material flow batch, thereby mapping and generating the high-priority interception confirmation state.

[0056] Further specifying, the linkage rejection control unit is equipped with a controlled pneumatic actuator; the step of locking the start and end time window of the rejection action of the linkage rejection control unit specifically includes:

[0057] Read the real-time transmission speed parameters of the target material and the pre-configured absolute transmission distance across workstations; and read the pre-configured physical response time compensation.

[0058] Based on the aforementioned basic defect risk identifier, the corresponding vertical physical length of the defect is parsed and derived.

[0059] Based on the real-time transmission speed parameter, the absolute transmission distance across workstations, the longitudinal physical length of the defect, and the exhaust hysteresis constant extracted in advance from the controlled pneumatic actuator, the reference theoretical trigger timestamp is calculated.

[0060] Based on the physical response time compensation, a bidirectional time-domain extension mapping is performed on the reference theoretical trigger timestamp to construct a removal action start and end time window that is presented as an absolute time-domain closed interval.

[0061] Furthermore, the following specific implementation details are provided for the above content: The extreme difference in operating speed characterizes the absolute quantitative deviation of the calender's acceleration and deceleration intensity under transient operating conditions, serving as an objective benchmark for determining whether forced convergence of the variance processed by the visual calculation side is necessary. In this embodiment, limited by the spindle rotational inertia of the heavy-duty calendering equipment, the extreme difference in operating speed is configured within the preferred range of absolute acceleration under normal frequency conversion adjustment as 0 m / s² to 2.0 m / s².

[0062] The threshold for intercepting extreme fluctuations is set at 0.8 m / s². When the acceleration of the calender exceeds 0.8 m / s², it indicates that the production line is in an unstable transitional state of emergency acceleration or emergency stop. At this time, the hysteresis of the exhaust valve of the controlled pneumatic actuator will increase nonlinearly. At this threshold for intercepting extreme fluctuations, feature reduction interception on the visual side is triggered to recover the timing lead.

[0063] The calibration steps for the extreme fluctuation interception threshold are as follows: continuously inject gradient-increasing extreme value test sequences into the sandbox test control node with an equal load, and simultaneously collect response feedback data streams characterizing the effective delay of the controlled pneumatic actuator in performing exhaust actions at the downstream control end; when the extracted action response feedback time exceeds the set system tolerance boundary, record the acceleration extreme value characteristics corresponding to the currently injected test sequence; after obtaining discrete convergence distribution points through multiple rounds of independent offline stress testing, extract the lower limit value of the convergence interval as the extreme fluctuation interception threshold.

[0064] The amount of delay compensation for coordinated execution (denoted as) This serves as a smoothing delay constant benchmark for the corresponding detection mode status identifier after eliminating network jitter and random errors; in this embodiment, the preferred range is... (Unit: ms), the preferred value is the current average value dynamically calculated based on the historical sliding window.

[0065] The target hazard weight, representing the control response priority scalar output by the defect semantic assessment model, determines the extreme value of the interception authority of this batch of aluminum plates in the global control scheduling bus. Its state transition value is {1,2,3,4,5}, where 5 corresponds to the highest level "fatal defect / breakage-level quality hazard".

[0066] Define a first control parameter normalization coefficient and a second control parameter normalization coefficient: the first control parameter normalization coefficient is used to eliminate the subjective discrete dimension of the weights, and the second control parameter normalization coefficient is used to eliminate the physical time dimension of the time parameter. In this embodiment, the preferred range for both is within the interval (0,1].

[0067] The scheduling identifier represents the control word indicating the order of material flow batch processing in the concurrent queue of the global scheduling unit; the queue-jumping control beacon represents the business instruction payload used to forcibly overwrite this order in the concurrent queue of the global scheduling unit. The state transition value of the queue-jumping control beacon is [0x00 (normal), 0xFF (rapid preemption)].

[0068] The system continuously parses the real-time production line control data stream in the form of discrete digital sequences input from the front end, extracts the macroscopic speed peak and trough parameters within the set monitoring period, and performs absolute difference calculation to generate the extreme difference of operating speed. The calculation logic is as follows:

[0069]

[0070] in The extreme difference in operating conditions represents the absolute speed variation caused by the dynamic acceleration and deceleration of the calender within the current calculation cycle. The monitoring period is defined (initially set to 1 second), representing the step size range of the time sliding window used to collect real-time production line control data streams. t represents the continuous or discrete time sampling cycle within the current monitoring period. It characterizes the various macroscopic velocity peak parameters obtained by analyzing the real-time production line control data stream within a given set monitoring period. This characterizes the various macroscopic velocity trough parameters obtained by analyzing the real-time production line control data stream within the same set monitoring period. `max` and `min` represent the business processing operators for extracting extreme value data features within the controlled time domain, respectively. This represents the mathematical boundary operator for extracting the absolute deviation of a scalar, ensuring that the output of the extreme difference is always a non-negative real number representing the amplitude of the fluctuation.

[0071] In this embodiment, "macroscopic velocity peak parameter" and "macroscopic velocity trough parameter" refer to structural extreme values ​​that can characterize the substantial physical acceleration and deceleration motion trend of the calender spindle. The acquisition logic is as follows: An optical encoder coupled to the main drive shaft of the downstream calender continuously acquires angular displacement pulse signals. Based on a preset microsecond-level sampling clock, these signals are converted into discrete raw velocity sequences with physical dimensions (m / s or rad / s). When the dimension of the extracted discrete raw velocity sequence is angular velocity (rad / s), the pre-configured radius constant of the calender drive shaft is extracted and algebraically multiplied to force a uniform conversion to linear velocity dimensions (m / s). This uniformly dimensional sequence constitutes the physical input source of the real-time production line control data stream. A low-pass sliding filter algorithm is used to calculate the sliding weighted average of historical data points within the sliding window to remove micron-level transient high-frequency pulses caused by mechanical transmission backlash, motor electromagnetic noise, or electrical high-frequency interference, and to fit and generate a smooth motion envelope curve. A macroscopic extreme value determination using a hysteresis loop is introduced: within a set monitoring period, a peak-finding algorithm with a hysteresis interval is used to traverse the motion envelope curve in one direction. Set an extreme value significance threshold (preferably a constant between 0.1 m / s and 0.3 m / s) to characterize normal physical fluctuations. A local maximum is locked and designated as a macroscopic velocity peak parameter only if a local maximum is located and the subsequent data retracement exceeds the extreme value significance threshold. A local minimum is designated as a macroscopic velocity trough parameter only if it is located and the subsequent data rebound exceeds the significance threshold. .

[0072] Obtain the absolute time difference extracted passively and the round-trip response time returned by the active virtual probe, write them into the corresponding latency benchmark archiving unit, and perform moving average calculation: construct a one-dimensional dynamic latency buffer sequence. The passively extracted absolute time difference Round trip response time The data sequence is concatenated and integrated according to the timestamp order; then, the arithmetic mean is calculated on the integrated one-dimensional array sequence containing N valid time-delay samples. The calculation logic is as follows: ;in The i-th latency scalar element in the completed unified archive pool sequence represents the current index of the latency data sample being processed. (One-dimensional dynamic latency cache sequence) It manifests as an ordered set of heterogeneous time parameters. For any element in this set Their data sources are all uniquely derived from passively extracted absolute time differences. Or the round-trip response time actively retrieved One of the two. In this step, Characterize the target execution latency compensation amount dynamically calculated and isolated within the corresponding latency benchmark archiving unit for the k-th detection mode status identifier (shallow dimensionality reduction visual detection mode or full-scale deep visual detection mode).

[0073] In this embodiment, the preferred range for the control silence time window is between 500ms and 5000ms. The minimum physical degradation period of the exhaust valve on the controlled pneumatic actuator, which causes nonlinear drift in its inherent response time due to continuous unsteady temperature rise, is read. In this embodiment, it is on the order of 10 seconds. This period is then divided by a set safety conversion factor, preferably in the range of 2 to 20.

[0074] The execution scale alignment mapping operation steps are defined by obtaining the normalized coefficient of the first control parameter matching the target hazard weight and the normalized coefficient of the second control parameter matching the dynamic pre-trigger compensation duration, and then performing product mapping to generate priority evaluation parameters and timing constraint parameters respectively:

[0075]

[0076]

[0077] The priority evaluation parameter is characterized after dimensional normalization mapping. The target hazard weight is represented by the output of the defect semantic evaluation model. The normalization coefficient of the first control parameter is specifically configured as a dimensionality reduction convergence factor used to converge discrete target hazard weights to a unified evaluation domain. The time-series constraint parameters are represented after dimensional normalization mapping. It represents the dynamic pre-triggered compensation duration generated by superimposing the base clock and the deviation. The normalization coefficient of the second control parameter is specifically configured as a numerical scaling factor used to scale the time series parameters proportionally to the numerical values ​​within a unified evaluation domain.

[0078] This method constructs a hardware extreme boundary constrained by the physical actuator of the calender to characterize the visual feature extraction strategy. When the extreme difference in operating speed exceeds the limit fluctuation interception threshold, the full image feature traversal matching process is interrupted, and the time delay variance of the front-end business processing is forced to converge to a controllable static constant to extract shallow contour risk identifiers in a dimensionality-reduced manner. After extracting the target hazard weight and dynamic pre-trigger compensation duration, the normalization coefficients of the first and second control parameters are retrieved to perform scale alignment, generating dimensionless priority evaluation parameters and timing constraint parameters. A unified evaluation domain is constructed, and the configured process safety tolerance boundary limit value is extracted. Based on this process safety tolerance boundary, cross-threshold verification is performed on the priority evaluation parameters and timing constraint parameters. A comprehensive defect judgment state is generated only when both parameter mapping domains are located inside the safety boundary.

[0079] This method is set based on the limitation of "actuator response failure limit"; when the dynamic flow encounters extreme business boundary (the transient extreme value of the working condition exceeds the physical response limit of the exhaust valve of the controlled pneumatic actuator, causing the risk of asynchronous delay failure), the passive disaster degradation flow logic is triggered: the dynamic compensation branch is cut off instantly, the generation and distribution of the queue-jumping control beacon are blocked, and the historical global delay step under the static safety average of the entire operation cycle is forcibly rolled back.

[0080] It should be noted that, under the condition of restricted variable speed rolling, the target material in actual physical flow is preferably a continuous aluminum strip after the aluminum coil is unrolled and is in a high-frequency continuous transmission state. This method needs to introduce a judgment criterion for strip breakage-level quality hazards when performing defect assessment; and when performing the linked rejection action, it is necessary to calculate the maximum value of the absolute time-domain closed interval based on the physical longitudinal length of the defect, in order to achieve accurate stripping and rejection of defective sections on the continuous strip.

[0081] Further explanation of step S1: The system receives discrete surface state sensing data streams of the target flowing material from the front-end visual sensing node, as well as real-time production line control data streams in the form of sampled discrete digital sequences. The continuously input real-time production line control data streams are parsed. A signal preprocessing node configured within the application layer flow control unit performs high-frequency noise reduction processing on the real-time production line control data streams based on a low-pass sliding filter algorithm with a preset step size, filtering out discrete random jumps and spurious fluctuations caused during the sampling process of heterogeneous interface data. In the cleaned and smoothed data sequence, the macroscopic velocity trough and peak parameters within the set monitoring period are located. The absolute difference is calculated to obtain the extreme difference of the operating speed for the current period. The extreme difference of the operating speed, as an absolute quantitative deviation characterizing the severity of transient physical load changes, is constrained by the physical limits of the rotational inertia of the main drive shaft of the currently mounted equipment and the rated torque of the motor.

[0082] This embodiment is deployed in low-to-medium speed control sections such as rewinding inspection lines or cross-cutting sorting lines for aluminum sheet and strip. In such sections, the strip turnover speed is preferably 30 m / min to 180 m / min (0.5 m / s to 3 m / s). Based on this operating speed benchmark and the physical rotational inertia of the equipment, this embodiment, in order to prevent excessive acceleration and deceleration from causing strip tension loss of control or pneumatic mechanism response discontinuity, sets the extreme speed difference within the monitoring cycle to the unit second level (1 s), and its benchmark range is preferably calibrated to 0 m / s to 2 m / s (equivalent to the maximum transient acceleration limit allowed by the physical equipment, preferably 2 m / s). 2 ).

[0083] Further boundary verification is performed by matching the preset extreme fluctuation interception threshold: if the extreme difference in speed under this working condition is determined to exceed the set extreme fluctuation interception threshold, the system feature degradation branch is triggered. The action is to cut off and intercept the full-element feature traversal comparison path with high computational concurrency load, and instead schedule the basic image filtering operator to process the surface state perception data stream, extracting local abnormal edge information only based on the basic pixel contrast rules, and generating a shallow contour risk label as the basic defect risk label; the detection mode status label of the current state perception data stream is forcibly updated and overwritten to the shallow dimensionality reduction visual detection mode representing the extreme speed mode. In response to the determination that the extreme difference in operating speed has not exceeded the set limit for intercepting fluctuations, a high-concurrency computation operator is scheduled to maintain the full-element feature traversal and comparison path to perform deep feature space mapping on the surface state perception data stream. Specifically, spatial clustering is performed on abnormal pixel clusters based on the connected component labeling algorithm, and the minimum bounding rectangle of each connected component is calculated to extract the boundary coordinate dataset representing the two-dimensional spatial occupancy of defects. The Gray-Level Co-occurrence Matrix (GLCM) algorithm is called to extract texture connectivity distribution parameters, and the Sobel edge operator is called to extract deep gray-level gradient features. Through the system's built-in feature cascade layer, the extracted heterogeneous feature matrices are flattened and serialized and spliced ​​into a one-dimensional multi-dimensional visual feature set according to a preset data protocol. Then, based on this multi-dimensional visual feature set, a basic defect risk label is generated.

[0084] Furthermore, this embodiment pre-defines a data protocol definition for dimension elimination and heterogeneous data addressing and splicing to ensure that the downstream defect semantic evaluation model has a deterministic tensor alignment benchmark when receiving high-dimensional fused data. The specific serialization and encapsulation mechanism of this data protocol is as follows: For the extracted morphological boundary size features with objective physical scale, gray-level co-occurrence matrix texture features with statistical probability distribution attributes, and Sobel depth gray-level gradient features with amplitude and direction angle attributes, independent normalization bypasses are called respectively. Specifically, Z-score normalization is used to forcibly compress the above three heterogeneous tensors into the dimensionless unified logical number domain of [-1,1]. In response to the tensor dimension differences generated by each local defect during feature extraction, one-dimensional fixed-length flattening is performed on each normalized feature tensor based on a preset maximum feature step size constant. If the length of the flattened feature sequence is less than the set span of the maximum feature step size constant, zero-value padding is performed at the end of the array. If it exceeds the limit, max pooling is performed to truncate, thereby generating three one-dimensional sub-feature vectors with absolutely constant length.

[0085] A global one-dimensional array representing a standardized multidimensional feature cluster vector is constructed. A global timestamp identifier representing the current deep processing cycle is embedded at the head of the global one-dimensional array; these sequences are then stacked sequentially according to their address order: morphological feature sequence, texture feature sequence, and gradient feature sequence; and a CRC checksum is encapsulated at the tail of the array. The resulting standardized one-dimensional multidimensional feature cluster vector not only achieves lossless fusion of multidimensional visual features internally, but its region addressing reduction also avoids semantic feature misalignment and contamination at the input layer of the deep neural network after feature concatenation.

[0086] This embodiment extracts local abnormal edge information based on basic pixel contrast rules. This action is executed by a lightweight static threshold filtering engine configured in the application layer, aiming to achieve rapid melting of extreme defects at the band break level with low computational overhead. The specific steps are as follows: The surface state perception data stream input from the front end is reduced in dimension and converted into a single-channel grayscale image matrix through a color space conversion channel to remove the non-core color dimension computational load; the grayscale image matrix is ​​traversed based on a local sliding window of a set size (preferably 3×3 or 5×5) to calculate the absolute contrast difference between the grayscale value of the center pixel and the average grayscale value of its neighboring pixels; the static contrast extreme value interception constant configured by the system is read (this constant corresponds to the threshold of the background noise brightness difference at the macroscopic physical structure missing such as normal reflection and broken strip / severe damage of the target circulating material); if and only if the absolute contrast difference is greater than the static contrast extreme value interception constant, the center pixel is anchored as a candidate risk coordinate point of abnormal high brightness or abnormal dark spot; through the 4-connectivity or 8-connectivity verification of the lightweight neighbor connectivity judgment mechanism, isolated single-pixel noise points are eliminated, and the candidate risk coordinate point cluster that meets the minimum physical aggregation area (continuously occupying more than 50 pixel units) is generated into a shallow contour risk label containing only macroscopic boundary coordinates and area scale. This rule relies on comparing pixel scalar differences with static constants to ensure that microsecond-level hard real-time feature output can still be maintained even under degraded computing resources. Based on the allocation strategy of the currently active lightweight filtering engine, a shallow dimensionality reduction visual detection mode matching this processing chain is output as a detection mode status identifier for the current streaming frame.

[0087] Regarding the update mechanism for the latency compensation amount during coordinated execution, this method performs parallel dual-track archiving calculations:

[0088] The first aspect is to read the historical control execution logs, extract the pre-judgment time marker and action confirmation time marker from the logs based on the system's global clock reference, and calculate the absolute time difference representing the time consumption of the entire link communication and hardware execution by algebraically subtracting the timestamps.

[0089] The second aspect involves maintaining a countdown service state machine at the application layer. In response to the state machine determining that no comprehensive defect judgment state has been generated within the set control silence time window, indicating a continued lack of downstream action confirmation feedback for the synchronization control signaling, a test interaction data packet with a physical action masking bit is actively encapsulated based on the idle keep-alive mechanism. This data packet is then converted into a virtual probe synchronization signaling with a terminal physical action masking identifier and sent across domains to the downstream linkage elimination control unit. Subsequently, the round-trip response time of this virtual probe synchronization signaling transmitted via the control bus is extracted. According to the detection mode state identifier corresponding to the extraction of the aforementioned time difference, the absolute time difference and the round-trip response time are imported together into the corresponding latency benchmark archiving unit for logical isolation archiving.

[0090] The data cleanup protocol maintained in the background performs timeliness elimination of historical data samples in the latency benchmark archiving unit according to the first-in-first-out (FIFO) principle. Based on the current valid latency sample set accumulated and cached in the cleaned logical isolation pool, a moving average calculation is performed independently of the main communication flow to generate and dynamically refresh a high-confidence linkage execution latency compensation amount corresponding to the status identifier of each detection mode. This linkage execution latency compensation amount serves as the time-domain smoothing correction benchmark for the feedforward control link, and its effective value range is physically converged to a constant range of 5ms to 120ms. This mechanism can effectively eliminate the contamination of stale samples caused by expired bus communication jitter, ensuring that the compensation amount maintains high sensitivity and characterization accuracy to the current real underlying control link state.

[0091] The specific implementation of step S2 is as follows: The extreme difference in operating speed and the amount of linkage execution delay compensation output from step S1 are used as input parameters to perform timing advance prediction and data encapsulation. The global delay configuration table is represented as a dynamically maintained system-level logical mapping matrix on the application layer flow control unit side. Its construction mechanism is as follows: Structurally, the table uses "detection mode status identifier" as the primary key for the row index and "linkage execution delay compensation amount" in the corresponding mode as the column attribute to construct a one-to-one mapping relationship between "mode" and "delay". The global delay configuration table is updated in real time by the delay mean archiving update step in step S1. Whenever the system completes a sliding mean calculation for the corresponding visual detection mode in the background, the calculation result is immediately overwritten into the corresponding cell in the table, ensuring that the values ​​stored in the table always represent the latest delay characteristics of the current physical channel. This table provides a standard retrieval interface for step S2. By comparing the currently active detection mode status identifier, a target value with high confidence is retrieved. The extracted detection mode status identifier is used as the hierarchical retrieval identifier; primary key comparison and matching are performed in the application layer global latency configuration table to retrieve the corresponding target execution latency compensation amount.

[0092] It needs to be explicitly stated and confirmed here that the high-confidence value generated in step S1 above, which involves logically isolating and archiving the absolute time difference and round-trip response time, and then performing independent moving average calculation, is the linkage execution delay compensation amount defined in this method. This parameter is represented as a dynamic parameter set containing multiple mode benchmarks. In this step, the linkage execution delay compensation amount value is matched and extracted from the global delay configuration table corresponding to this dynamic parameter set using the detection mode status identifier corresponding to the current streaming frame as a unique index, and is defined as the target execution delay compensation amount.

[0093] Extract the current global control coordination benchmark (denoted as) In this embodiment, the preferred range is configured to be between 30ms and 80ms. The current global control coordination benchmark is specifically statically equivalent to the pre-calibrated and fixed extreme value of the basic network transmission delay of the current system, representing the pure physical relative time consumption parameter under ideal no-load conditions. The target execution delay compensation amount is directly used as the advance compensation factor and subjected to a pre-algebraic summation operation with the current global control coordination benchmark. Its corresponding superposition mechanism is expressed as follows: ; to generate a dynamic advance trigger compensation duration with predictive attributes that characterizes the pure relative time series lead correction amount. This smoothing step size is represented by a pure interval value perfectly aligned with the physical execution timeline, serving as the sole relative compensation benchmark for generating the final absolute timestamp. Substituting this pure interval value into the equation... Scale alignment is performed to generate dimensionless timing constraint parameters: This relative compensation amount is combined with the real-time master control timestamp only when the control signal is issued in step S3, and the output is the absolute activation time of the pneumatic valve in the controlled pneumatic actuator.

[0094] In response to the detection mode status being identified as full-scale deep visual detection mode, the pre-loaded defect semantic assessment model is retrieved. Based on the spatial coordinates and scale anchors attached to the basic defect risk identifier, the corresponding local high-dimensional image tensor feature matrix is ​​extracted from the current surface state perception data stream through the image cropping engine. This high-dimensional image tensor feature matrix is ​​used as a standard multi-dimensional feature source and input into the multi-layer feature cascade perception network deployed within the model for forward inference operations. Its semantic distribution features in the preset dimension space are extracted and then mapped by the terminal classifier to generate quantified target hazard weights.

[0095] In the system architecture, the image cropping engine is represented by a spatial semantic mapping logic unit. This unit is responsible for parsing the metadata embedded in the basic defect risk identifier, specifically the spatial coordinates of the physical center point of the defect identified in the preceding step S1 and the scale anchor points of the coverage area. The "cropping" action represents using the above coordinates and anchor points as dynamic window parameters to perform local pixel block addressing and extraction on the surface state perception data stream in the buffer, thereby generating a size-normalized local high-dimensional image tensor feature matrix. The forward inference operation is macroscopically represented as a deterministic mapping function from features to semantics. This operation, through cascaded computation of multi-layer perception nodes, abstracts the input image pixel tensor layer by layer into probability distribution features representing the severity of the defect, and finally, the classifier outputs a discretized target hazard weight that matches the current physical attributes of the defect.

[0096] In response to the detection mode status indicating a shallow dimensionality reduction visual detection mode, the high-dimensional image tensor feature matrix extraction is actively suspended. Through application-layer static mapping bypass, the shallow contour risk markers of the extracted local anomaly edge information are directly mapped to the highest hazard level. The target hazard weight is then applied. The static assignment value is its logical upper limit extreme value of 5. At this time, the parameter in this state is defined as the extreme value target hazard weight.

[0097] In this embodiment, the multi-layer feature cascaded sensing network is instantiated sequentially in the virtual computing node as the following logical processing layers:

[0098] The input dimension mapping layer is configured to receive and normalize a high-dimensional image tensor feature matrix, mapping it to a matrix with a fixed dimension. The initial business data stream carrier; the spatial feature dimensionality reduction layer preferably includes at least three sets of cascaded two-dimensional convolution operators and max pooling operators, which are used to extract the spatial gradient and texture connectivity identifiers of defect features at different scales based on the local receptive field, and compress the high-dimensional features into a one-dimensional deep business feature sequence.

[0099] The fully connected semantic fusion layer is configured to receive a one-dimensional deep business feature sequence, perform cross-dimensional weighted inner product operations through non-linear activation nodes, and output a discrete feature vector of dimension K, thereby extracting its semantic distribution features in a preset dimensional space. The control refers to the scale parameter of the multidimensional feature matrix tensor, which represents the spatial resolution width, height, and number of channels of the input feature source. In this embodiment, the preferred range is a pixel constant, with a preferred value of 224×224×3. K represents a one-dimensional scalar representing the total number of discrete business priority state categories in the final output. Its value is 5, corresponding to the 5 level intervals of the target hazard weight. The quantified target hazard weight is generated by mapping through the terminal classifier. The defect semantic evaluation model adopts a deep convolutional neural network (CNN) architecture, which is trained under supervision based on an industrial defect benchmark dataset before system deployment. This dataset collects and labels various fatal breakage hazards and shallow surface scratches images that have been cleaned and screened, covering multiple types of real production extreme lighting deviation samples. The training process uses a cross-entropy loss function to guide the weight gradient descent update until its loss value converges to the preset application tolerance threshold. In this embodiment, to meet the high confidence interception requirements of the rolling line for minor defects, the application tolerance threshold is preferably limited to a value range of 0.01 to 0.05. In this embodiment, it is specifically calibrated to 0.02. In this mapping relationship, the target hazard weight serves as a scalar for the control response priority that determines the global control bus resource scheduling tilt and action triggering authority. It is a finite discrete integer sequence {1,2,3,4,5}, with the extreme value 5 representing an inescapable fatal flaw, ensuring that the generated scalar has objective industrial discriminative power.

[0100] Furthermore, this embodiment's supervised training process not only relies on statically cleaned samples but also introduces adversarial generation under constrained physical conditions and dynamic perturbation data augmentation mechanisms. During the data loading and forward propagation phases of network training, the system bypasses and calls the physical degradation simulation engine. According to a preset trigger probability, it randomly injects an anisotropic motion blur kernel simulating real high-frequency rolling flutter and a Poisson dynamic illumination noise simulating flickering interference from heterogeneous light sources in a factory into the high-dimensional image tensor of the training batch. Based on this multi-dimensional perturbation, in addition to the cross-entropy loss function, a structural consistency constraint loss is further superimposed, forcing the model to still extract semantic feature boundaries that are highly homologous to the clean, undamaged samples even when facing severe degradation of the input image caused by the extreme fluctuations of the underlying physical equipment. This perturbation-resistant training specification, mounted on the model's underlying weight topology space, endows it with absolute anti-interference discriminative power, preventing the misjudgment of dynamic blur as a fatal flaw when facing spatiotemporal decoupling and transient high-frequency oscillation environments.

[0101] The system performs scale alignment and dimensionality reduction. The application layer transfer control unit reads the system's limit boundary configuration library and dynamically calculates and obtains the first control parameter normalization coefficient matching the target hazard weight, and the second control parameter normalization coefficient matching the dynamic pre-triggered compensation duration, through a defined calibration mechanism. The first control parameter normalization coefficient is generated by calculating the ratio of the unit extreme value constant to the maximum lethal hazard state identifier value supported by the model; the second control parameter normalization coefficient is generated by calculating the ratio of the unit extreme value constant to the maximum physical delay limit constant allowed by the system architecture. The scale transformation mechanism corresponding to the first and second control parameter normalization coefficients is as follows:

[0102]

[0103]

[0104] The two parameters are applied to perform a multiplicative mapping operation on the target hazard weight and the dynamic pre-trigger compensation duration, respectively, to generate a dimensionless priority evaluation parameter and a timing constraint parameter.

[0105] In this embodiment, the preferred ranges for the normalization coefficients of the first and second control parameters are as follows: consecutive decimals within an interval; This refers to the maximum critical hazard state extreme value that the defect semantic evaluation model can output; its value is 5. The meaning is the maximum physical delay limit constant that the system can tolerate for cross-node communication and actuators; in this embodiment, the preferred range is 150ms to 300ms, and the preferred value is 200ms.

[0106] By accessing the system's global configuration bus, the pre-configured process safety tolerance boundary limit values ​​within the application layer flow control unit are extracted. The determination of these limit values ​​specifically includes the minimum priority release threshold and the maximum delay expiration threshold. The configuration of the minimum priority release threshold is based on statistical analysis of historical production line yield data. Its setting logic is as follows: extract the physical limit defect tolerance level identifier from the system quality specification that does not affect the downstream finished product pass rate, and convert it proportionally into a dimensionless scalar falling within the (0,1] interval and in the same logical evaluation domain as the aforementioned priority evaluation parameters, using this as a benchmark.

[0107] The physical baseline configuration for the maximum delay expiration threshold is obtained by dividing the maximum physical spatial span from the end of the physical production line conveyor belt to the actuator by the minimum conservative flow velocity. During system initialization, the pre-configured second control parameter normalization coefficient is further extracted, and the physical baseline configuration is scaled proportionally to a unified evaluation domain through algebraic multiplication mapping to generate a dimensionless final scalar with the same dimension as the timing constraint parameter, which serves as the maximum delay expiration threshold for cross-validation.

[0108] The safety tolerance limits of the two aforementioned processes are constrained within the range (0,1). The continuous range of the lowest priority release threshold is limited to [0.2, 0.8]. In the rolling control condition of this embodiment, the preferred critical value is set to 0.4. Since the original discrete target hazard weight extreme value is 5, after introducing the normalization coefficient (1 / 5=0.2), setting the threshold to 0.4 means that the system will automatically release micron-level surface oxide spots with a weight of 1 or 2 (corresponding to values ​​of 0.2 or 0.4), while initiating unconditional physical interception for any substantial cracks equal to or higher than level 3 (corresponding to values ​​≥0.6), thereby achieving the optimal energy efficiency balance between yield and physical action frequency.

[0109] The maximum delay expiration threshold is continuously limited to a range of [0.75, 1.0]. In this embodiment, the preferred safety threshold is set to 0.95. This value indicates that when the predicted relative compensation time reaches 95% of the absolute physical response limit tolerance, it is considered to be at high risk of delay.

[0110] Perform two-dimensional logical AND cross-validation on the priority evaluation parameters and timing constraint parameters:

[0111] Branch 1 is normal conduction: If and only if the priority evaluation parameter is greater than the lowest priority release threshold and the timing constraint parameter is less than the maximum delay expiration threshold, it is determined to be legal in both time and space. The two are merged and encapsulated into structured data streams and mapped to generate a comprehensive defect judgment status.

[0112] Branch 2 is fatal circuit breaking: If the priority evaluation parameter is greater than the lowest priority release threshold (indicating the existence of a high-risk defect), but the timing constraint parameter exceeds the maximum delay expiration threshold (indicating that the physical rejection action has expired), the current control logic skips the encapsulation and linkage rejection steps and directly throws a system-level global emergency stop control signal to the upper-level bus, forcibly blocking the operation of the global service transmission mechanism to ensure that the fatal defect has no escape.

[0113] Branch 3 is for security blocking: If the priority evaluation parameter does not reach the minimum priority release threshold and any dimension does not meet the above conditions, then silent discard processing is performed to block the data flow encapsulation.

[0114] Specific implementation details for step S3: The linkage rejection control unit acts as the physical execution node, internally equipped with a controlled pneumatic actuator. When this control unit receives the synchronization control signal and parses out the start and end time windows of the rejection action, which are presented as an absolute time-domain closed interval, its physical execution mechanism follows the state machine transition logic constrained by the endpoints of the closed interval. Specifically, the controller of the linkage rejection control unit extracts the minimum and maximum timestamps of the absolute time-domain closed interval. As the system master clock advances to the minimum timestamp, the controller instantaneously outputs an electrical trigger level, activating the controlled pneumatic actuator. Since the minimum timestamp has already been pre-deducted by the exhaust hysteresis constant and superimposed with the dynamic pre-trigger compensation time when it is generated at the application layer, the physical response time for the pneumatic circuit to establish air pressure and overcome mechanical damping is offset on the time axis. The pneumatic mechanical components reach their maximum physical performance when the target material physical segment with extreme target hazard weight enters the cross-station absolute transmission distance of the controlled pneumatic actuator at a real-time transmission speed, and performs fixed-point physical rejection or segmented cutting action on the continuously transmitted strip.

[0115] In this embodiment, the linkage rejection control unit, in an actual restricted variable speed rolling production line, is mechanically integrated with a flying shear cutting mechanism based on the thickness specifications and defect type of the target material. The controlled pneumatic actuator in this method specifically refers to the pneumatic component within the cutting mechanism responsible for controlling the engagement and disengagement of the blade holder, its downward pressure, or the diversion of waste material. Electrical pulses within an absolute time-domain closed interval are used to trigger the pneumatic circuit; through the inherent exhaust hysteresis and mechanical damping of the pre-flush pneumatic component, it is ensured that the controlled cutting blade holder can enter the effective working stroke within the extreme time window when it is relatively stationary with the aluminum strip flow rate, thereby achieving precise physical cutting and rejection of the defective section.

[0116] When the clock precisely advances to its maximum timestamp, the controller immediately reverses the trigger level, causing the controlled pneumatic actuator to rapidly depressurize and mechanically reset. The hardware response parameters of the mechanical equipment associated with the downstream linkage rejection control unit are obtained. During system initialization, the hardware configuration specifications of the controlled pneumatic actuator are parsed; the factory test constant, indicating the lower limit of the maximum load command response concurrency, is extracted and loaded into the global static variable area of ​​the system application layer configuration library. By calling a preset one-dimensional static physical dimension mapping rule, the factory test constant is converted into a globally static read-only actuator response failure limit parameter in the same logical calculation domain as the extreme difference in operating speed.

[0117] In this embodiment, the continuous absolute value range of the actuator response failure limit parameter is defined as [1.2, 1.8], in m / s. 2 For the physical environment of the rolling and pressing linkage in this system, the preferred absolute critical value is precisely set to 1.5 m / s. 2 .

[0118] The threshold for intercepting extreme fluctuations configured in this embodiment is set to 0.8 m / s. 2 This is merely the software-layer defense limit that triggers dimensionality reduction in visual computing power; however, when the extreme difference in operating speed continues to surge and exceeds 1.5m / s... 2 When the actuator response reaches its failure limit, it indicates that the controlled pneumatic actuator is approaching the lethal critical point of nonlinear deformation and pressure build-up failure. This prevents the indiscriminate issuance of invalid high-frequency commands to the controlled pneumatic actuator at the edge of extreme collapse. This parameter serves as an absolute blocking benchmark for determining whether the current control logic has triggered the control safety circuit breaker isolation defense, avoiding the issuance of invalid timing commands exceeding the controlled pneumatic actuator's execution capability under extreme acceleration conditions.

[0119] In response to the determination that the extreme difference in operating speed has not exceeded the set failure limit of the aforementioned actuator response, a synchronization control signal is generated based on the comprehensive defect determination state generated in step S2 to trigger the downstream linkage rejection control unit.

[0120] In response to the generation of the synchronization control signal, the action of the start and end time window of the rejection action issued by the locking linkage rejection control unit is executed. The current transfer speed of the target material and the physical distance of the rejection actuator are obtained, and algebraic division is performed to determine the extreme time range in which the rejection action must take effect.

[0121] Based on the longitudinal physical length of the defect, the inherent exhaust hysteresis constant, and the set physical response time compensation, the effective time window interval of the physical action is calculated: before constructing this absolute time domain closed interval, the real-time pulse data uploaded by the underlying speed measurement encoding node is called through the application layer data bus and converted into a real-time transmission speed parameter characterizing the current physical motion state of the aluminum plate; in this embodiment, the communication between the application layer data bus and the linkage rejection control unit preferably adopts a deterministic communication protocol based on real-time industrial Ethernet (EtherCAT or PROFINET-IRT), so as to ensure at the physical link layer that the queue-jumping control beacon can penetrate the network noise with sub-millisecond deterministic jitter to reach the actuator.

[0122] Based on the basic defect risk identifier extracted in step S1, when deriving the longitudinal physical length of the defect, a data unpacking operation is performed on the basic defect risk identifier to directly extract the upper and lower edge pixel coordinates of its target bounding rectangle in the direction parallel to the main conveying axis (the Y-axis direction in the two-dimensional image coordinate system). Then, by calculating the absolute difference between the two, it is used as the absolute pixel span along the main conveying axis in the two-dimensional image space. The pre-fixed pixel equivalent conversion ratio coefficient is read from the system environment configuration. It should be noted that, in order to adapt to the high-frequency mechanical vibration conditions and the identification requirements of micron-level defects (scratches, roll marks) in the continuous rolling production line of aluminum plates, and at the same time take into account the real-time computing power load boundary of the main control hardware in processing high-dimensional image matrices, the effective value range of the pixel equivalent conversion ratio coefficient is limited to 0.05mm / pixel to 0.8mm / pixel.

[0123] Furthermore, in the full-volume deep vision inspection mode, the preferred range of the pixel equivalent conversion ratio coefficient is configured to be 0.1 mm / pixel to 0.3 mm / pixel, and the optimal constant value pre-fixed in the system is 0.15 mm / pixel.

[0124] During the system initialization phase, the optimal constant value is derived from the focal length of the optical lens of the front-end visual perception node, the physical size of the photosensitive component target, and the actual installation distance, through a standard physical calibration board and offline calibration. In the subsequent production and operation cycle, it is stored in read-only memory as an immutable system-level environmental constant.

[0125] By performing a one-way algebraic scalar multiplication, multiplying the pixel span by the pixel equivalent conversion factor, the absolute longitudinal physical length of the defect in physical space (denoted as ) is derived losslessly. The application laminar flow control unit performs a bidirectional temporal extension mapping based on physical spatiotemporal relationships on the triggering time points of the fundamental theory, according to the following core interval equations:

[0126] ;in, The final calculation generates the start and end time windows of the elimination action. To control the electrical level transition, the minimum timestamp is triggered at the start. To maintain the maximum termination timestamp for the reset action of the control mechanism. This is the trigger timestamp based on the baseline theory. This timestamp is obtained by acquiring the real-time absolute timestamp of the currently flowing target material captured by the detection node. The absolute transmission distance across workstations of the controlled pneumatic actuator (as denoted by...) Divide by the real-time transmission speed parameter The time is then converted to a time-dimension scale to obtain a purely theoretical transmission time that is consistent with the system's master clock. The real-time absolute timestamp is added to this purely theoretical transmission time, and the exhaust hysteresis constant is algebraically subtracted from it according to the timing alignment rules. Compensation duration for dynamic pre-triggered events generated earlier This ultimately leads to the derived baseline theoretical trigger timestamp. .

[0127] It is important to clarify here that the exhaust hysteresis constant... The inherent mechanical dynamic dead zone property of the controlled pneumatic actuator is specifically characterized by the inherent time delay required from the physical moment when the controller outputs the electrical trigger level until the pneumatic solenoid valve completes the commutation, the main air chamber establishes an effective working air pressure, and the piston overcomes the static friction force to generate the initial phase displacement.

[0128] The linkage elimination control unit executes the controlled pneumatic actuator, and the absolute transmission distance across workstations is... It is a spatial geometric static constant. In the physical coordinate system of the calendering production line, this distance is defined as: the absolute physical length that any microscopic particle on the surface of the target material must travel along the actual movement trajectory of the conveyor belt from the "visual detection reference point" to the "physical rejection reference point". The "visual detection reference point" refers to the perpendicular intersection of the central optical axis of the front-end vision camera and the conveying plane of the target material; the "physical rejection reference point" refers to the blade closure line or physical contact center line when the downstream controlled pneumatic actuator performs mechanical action. By extracting this solidified absolute conveying distance across stations, which can be directly measured by conventional physical measuring tools, and dividing it by the real-time conveying speed parameter, the purely theoretical flight time required for the target material to cross this spatial span can be solved.

[0129] A preset physical response time compensation (preferred range 5ms~15ms) is used to compensate for microscopic execution errors caused by air pressure fluctuations. This physical response time compensation is obtained based on the microfluidic anti-disturbance verification action during the system initialization phase. Specifically, a high-frequency sampling oscilloscope node is scheduled to perform offline clustering extraction of physical time delay deviations caused by extreme air pressure fluctuations under continuous operation of this type of pneumatic actuator. After cleaning and removing outlier data, the maximum absolute deviation value of this deviation from the cluster distribution boundary is extracted, calibrated, and solidified as the physical response time compensation. This margin serves as a temporal safety redundancy scalar in the closed-loop control strategy to ensure physical command coverage; the preferred value is 10ms.

[0130] This embodiment expands a single timestamp into an absolute time-domain closed interval, constructing and locking the start and end time window of the removal action. This is a one-dimensional timestamp array composed of minimum and maximum values. By recombining the longitudinal physical length of the defect detected by the front-end vision with the real-time transmission speed parameter, the risk of physical spatial misalignment is eliminated. This is then transformed into a control parameter constraining the instruction activation time of the lower-level hardware actuator, ensuring that the physically linked actions fall within the time-domain window where the spatial positions completely overlap.

[0131] Synchronization of flow status is triggered synchronously within the global scheduling unit. Based on a preset multi-task collaborative allocation strategy, each business execution entity is structurally encapsulated and transformed into a standard cross-module interactive control data flow. The specific state synchronization action is based on the internal application layer state control logic, which transforms it into a global change of the business flow progress identifier; the application layer control unit reads the current concurrent task sequence dynamically constructed by the collaborative requests initiated by each business sub-node within the system in the global scheduling unit, extracts the scheduling identifier representing the processing order of each related collaborative task; in response to the generated synchronization control signaling, it constructs a queue-jumping control beacon for the current material flow batch;

[0132] When the comprehensive defect judgment state generated by the application layer control logic parsing meets the extreme interception conditions, a queue-jumping control beacon with the highest interruption authority (its control load status bit is directly assigned to 0xFF) is immediately generated in the application layer global state control node for the material flow batch. Based on the generated 0xFF queue-jumping control beacon, the application layer flow control unit initiates an asynchronous interrupt request signaling with a preemptive communication identifier to the global scheduling unit through the internal system bus. After receiving the request, the global scheduling unit locks the flow execution pointer of the current concurrent queue. In the locked state, the application layer control node directly overwrites the scheduling identifier corresponding to the current material flow batch based on the queue-jumping control beacon (replacing the original 0x00 normal logic sorting position with the 0xFF rapid preemption identifier) ​​to advance the logic execution sorting of the current material flow batch. After the replacement is confirmed, the global scheduling unit releases the bus suspension state and broadcasts a priority reload completion synchronization response packet to the application layer and downstream linkage elimination control units to trigger the immediate queue reordering of the underlying execution stack. The asynchronous interrupt request signaling with a strong communication lock identifier signifies a high-priority logic control bit that triggers a strong lock on layout-level resources; its state transition value is a binary trigger state. The priority reload completion synchronous response signifies a verification receipt signaling that releases cross-module communication lock and globally confirms the completion of the route topology update; its state transition value is a confirmation release status identifier. The queue-jumping control beacon and scheduling identifier are essentially hexadecimal discrete status words that determine the execution sequence of the system-level concurrent task queue. Based on assigning equal high-priority interception confirmation status (primary key) to each concurrent task, the timestamp base corresponding to the dynamic pre-triggered compensation duration is embedded as a secondary sorting index (secondary key) in the scheduling identifier. This ensures that even when multiple highest-priority tasks are concurrent, the queue can still follow the absolute physical time / space sequence to execute the pre-ordering of micro-instructions, guaranteeing the orderly continuation of continuous culling actions.

[0133] By forcibly replacing the aforementioned scheduling and sorting identifiers, the control scheduling status of the current material flow batch is mapped to a high-priority interception confirmation status. At the end of this normal branch, a synchronization control signal is specifically sent to the downstream linkage rejection control unit. This synchronization control signal is used to instruct the underlying controller to perform physical action compensation, thereby eliminating the coordination delay between the front-end visual perception node and the linkage rejection control unit on the application layer flow control unit side. Based on this, the effective time of the issued synchronization control signal can be dynamically synchronized with the real-time transmission location of the target flow material.

[0134] In response to the determination that the extreme difference in operating speed exceeds the set failure limit of the actuator response, a blocking signal is generated as a global control command; based on the blocking signal, the dynamic generation process of the dynamic pre-triggered compensation duration is intercepted, and the pre-configured system preset static compensation constant is forcibly rolled back to replace the dynamic parameter.

[0135] When the value of the extreme difference in the operating condition speed overflows and exceeds the response failure limit of the actuator, the application layer flow control unit immediately stops the dynamic step size superposition calculation data stream and instead calls the system's preset static compensation constant. The baseline configuration logic of this constant is as follows: extract the linkage execution delay compensation amount records in the system's historical operating cycle, remove the extreme and occasional data at both ends, calculate the arithmetic mean, and use it as the constant backup parameter under the blocking branch.

[0136] Based on the blocking signaling, the subsequent detection mode status identifier is forcibly downgraded and locked to a shallow dimensionality reduction visual detection mode. This disaster recovery mechanism sacrifices the detection accuracy of micron-level minor defects to forcibly preserve the deterministic interception of core quality hazards at the band breakage level in cross-node control. During the downgrade period, the application layer state control logic continuously monitors the extreme value difference of the operating speed of subsequent flow frames. If, within M consecutive monitoring periods (M being a safety threshold constant set between 3 and 10), the extreme value difference of the current operating speed falls back to the normal fluctuation safety range, the system actively triggers a state self-healing branch: canceling the blocking command for the feature traversal path and smoothly rolling back the extraction strategy of subsequent target flow materials to a full-scale deep visual detection mode, thereby achieving a complete logical closed loop for cross-domain business collaborative management in complex industrial environments.

[0137] The method in this embodiment is based on the mapping of dynamic input conditions to achieve adaptive adjustment of the core control parameters:

[0138] The extreme difference in operating speed serves as the benchmark for separating full-scale deep analysis and shallow dimensionality reduction interception. When it exceeds the limit, high-concurrency computation is forcibly reduced in dimensionality to eliminate instruction delivery misalignment under extreme fluctuations from the root cause of control.

[0139] The linkage execution delay compensation is used as the core benchmark for deriving timing constraint parameters. By dynamically extracting historical averages and performing pre-compensation, it absorbs the random delay variance of network nodes and physical links in advance.

[0140] The target hazard weight serves as the priority criterion for controlling bus resources, ensuring that the highest-level hazard (extreme value 5) has unconditional interception authority in cross-module collaborative pipelines.

[0141] Based on the aforementioned mechanism, the internal processing logic of this method adaptively updates according to the operating conditions: when the extracted extreme value difference does not reach the set limit fluctuation interception threshold, the full-scale deep visual detection mode is maintained, and the configured first control parameter normalization coefficient is aligned with the extracted target hazard weight execution scale to derive the priority evaluation parameter; simultaneously, the timing constraint parameter is derived based on the second control parameter normalization coefficient and the dynamic pre-triggered compensation duration. When the operating condition input changes drastically, the response calculated by this method based on the above model is shown in the table below.

[0142] Table 1: Numerical Derivation Examples Based on the Dynamic Rolling Acceleration Extreme Perturbation of This Embodiment

[0143]

[0144] The numerical projection trends shown in Table 1 are specifically calibrated under the dynamic fluctuations of the spindle absolute acceleration variation parameter and the objective constant of the data transmission hysteresis of the linkage elimination control unit under dynamic frequency conversion conditions. When the extreme difference of the operating speed jumps to 0.95, the front-end computational load expands exponentially, resulting in a sharp increase in the corresponding extracted dynamic pre-trigger compensation time. Scale alignment is performed according to the pre-configured normalization coefficient of the second control parameter. If dimensionality reduction interception is not performed, the system will generate a failure timing constraint parameter with a value as high as 1.50. Since this value directly exceeds the process safety tolerance boundary (exceeding the maximum delay expiration threshold, preferably 0.95), it will be directly judged as expired and discarded by the application layer flow control logic. Conversely, this method, upon detecting the out-of-bounds error, instantly and proactively suspends the high-dimensional image tensor feature matrix truncation action and forces feature dimensionality reduction through application layer static mapping bypass. Based on the data in Table 1, it can be seen that after adopting this method, the generated time-series constraint parameters are significantly optimized and converged from the baseline value of 1.50 in the existing technology without dimensionality reduction to the legal output value of 0.60 of this method, which theoretically improves the compliance performance of time-series constraints by 60% (calculation logic: by extracting the absolute value of the difference between 1.50 and 0.60 and dividing it by the baseline value of 1.50).

[0145] Figure 1In this diagram: Controlled pneumatic components represent controlled pneumatic actuators; VSP represents the front-end vision sensing node, a vision hardware device deployed above the rolling mill line, responsible for capturing surface defects on continuous aluminum sheets and outputting surface state perception data streams. RMB represents the rolling mill / heavy-duty conveyor belt, representing the linkage rejection control unit's execution transmission platform carrying continuous aluminum strips, where its spindle motor generates a physical working environment of drastic high-frequency acceleration and deceleration. ALC represents the application layer flow control unit, representing the control center algorithm carrier, responsible for eliminating cross-node coordination delays between the front-end vision sensing node and the linkage rejection control unit. PRU represents the linkage rejection control unit, representing the downstream physical execution end, containing controlled pneumatic actuators. It performs point-to-point physical rejection actions on continuous strips by parsing the endpoint advance activation of the absolute time domain closed interval to completely offset the exhaust hysteresis constant through the cutting mechanism. VIS represents the surface state perception data stream, representing the static visual feature data base collected by the front-end vision sensing node. RCDS represents the real-time production line control data stream, representing the dynamic pulse sequence collected from the underlying rolling mill, characterizing the strip's transmission position and physical transmission rhythm. ESD represents the extreme difference of operating conditions speed, which is the core calculation parameter characterizing the dynamic acceleration and deceleration limit boundary conditions of the main drive equipment, serving as the criterion for triggering timing correction and mode degradation. BDR represents the basic defect risk identifier, which is the static visual feature tensor extracted from the surface state perception data stream, characterizing the initial state of surface defects on the aluminum plate. VDM represents the detection mode state identifier, which is the working mode identifier dynamically determined by comparing the extreme difference of operating conditions speed with the limit fluctuation interception threshold, resulting in system feature dimensionality reduction or full-element traversal.

[0146] CDD represents the linkage execution delay compensation amount, which is a time base constant associated with the current visual inspection mode and used to characterize network communication lag and physical flow errors. FDSS represents the dynamic pre-trigger compensation duration, which is a pure interval physical action lead time benchmark generated after summing the timing benchmarks by calling the linkage execution delay compensation amount. CDDS represents the comprehensive defect judgment status, which is the comprehensive defect judgment status generated after associating and mapping the dynamic pre-trigger compensation duration with the basic defect risk identifier.

[0147] like Figure 1 As shown, the method of this invention is executed by the application layer flow control unit. During execution, it receives surface state perception data streams of the target flowing material from the front-end visual perception node and real-time production line control data streams. Based on the real-time production line control data stream, it extracts the extreme difference in operating speed, characterizing the dynamic acceleration and deceleration limit boundary conditions of the main drive equipment, and extracts the basic defect risk identifier from the surface state perception data stream. The extreme difference in operating speed is compared with a set limit fluctuation interception threshold to dynamically determine the detection mode state identifier corresponding to the surface state perception data stream, and obtains the associated linkage execution delay compensation amount.

[0148] Using the extreme difference in operating speed and the amount of compensation for linkage execution delay as input, the linkage execution delay compensation amount matching the detection mode status identifier is called to perform timing benchmark summation processing, generating a dynamic pre-trigger compensation duration characterizing the physical action advance benchmark; then, the dynamic pre-trigger compensation duration is correlated and mapped with the basic defect risk identifier to generate a comprehensive defect judgment status. The extreme difference in operating speed is verified with the set actuator response failure limit, and based on the comprehensive defect judgment status and verification results, a synchronization control signal is sent to the downstream linkage rejection control unit. This signal is converted into a rigid electrical pulse that constrains the working stroke of the underlying controlled pneumatic actuator. By extracting the endpoint of the absolute time domain closed interval for advance activation to completely offset the mechanical exhaust lag, within the physical execution window wrapped by time and space, a fixed-point physical rejection action is completed on the continuously transmitted aluminum strip, thereby eliminating the coordination delay between the front-end visual perception node and the linkage rejection control unit on the application layer flow control unit side, and on this basis, enabling the effective time of the issued synchronization control signal to be dynamically synchronized with the real-time transmission position of the target flow material.

[0149] Figure 3 The horizontal axis represents the extreme difference in operating speed, and the vertical axis represents the timing constraint parameters. Different line types represent the traditional extrapolation curve without dimensionality reduction and the reduced convergence extrapolation curve of this method, respectively. Figure 3 Numerical calculations demonstrate that the threshold comparison rule for extreme fluctuations proposed in this invention can eliminate the divergence of timing parameters caused by high load. Within the stable test range where the difference in extreme speeds under operating conditions is less than the threshold of 0.8 for extreme fluctuations, the two calculated curves coincide, exhibiting a steady, gradual increase. In contrast, within the extreme extreme range of high loads exceeding the threshold of 0.8 for extreme fluctuations, the traditionally derived curve shows a divergence trend exceeding the normalization coefficient of the second control parameter. The degradation convergence curve of this method, due to the active suspension of feature extraction actions and extreme value weighting, exhibits a step-like algebraic decrease at the threshold point and reconverges within the safe and legal domain. This confirms that introducing pre-judgment of extreme operating conditions and feature-based dimensionality reduction and diversion control can qualitatively resolve the synchronization problem of command actions and target physical positions missing the target.

[0150] Figure 4While maintaining the full-feature traversal comparison path, the system receives the surface state perception data stream from the front end (corresponding to the image data of the defect area of ​​the target flowing material in the figure). For this input source, the system executes a parallel heterogeneous feature extraction branch: it extracts the boundary coordinate dataset representing the two-dimensional spatial occupancy of defects based on the connected component labeling algorithm (denoted as the defect two-dimensional boundary coordinate dataset in the figure); secondly, it calls the gray-level co-occurrence matrix algorithm to extract texture connectivity distribution parameters; and thirdly, it calls the Sobel edge operator to extract depth gray-level gradient features. Through the system's built-in feature cascade layer, the three sets of heterogeneous feature matrices extracted above are uniformly flattened and serialized and spliced ​​according to a preset data protocol, ultimately generating a one-dimensional / multi-dimensional visual feature set of constant length. For example... Figure 4 The multidimensional visual feature set is used to achieve lossless fusion of multidimensional visual features, which serves as the high-dimensional data base for subsequent aggregation to generate basic defect risk identifiers.

[0151] The computational logic involved in this application can be constructed using algorithms such as regression analysis in machine learning, establishing a mathematical model by analyzing the inherent trends and interrelationships of the collected parameters. This process can be implemented using specialized computational tools (such as Python's Scikit-learn library or the R language environment). Throughout all calculations, to eliminate the influence of different physical dimensions and ensure that data is compared and analyzed on the same scale, the input parameters in each formula are dimensionless. The dimensionless techniques used include, but are not limited to, max-min normalization or Z-score standardization.

[0152] To decouple the core algorithm from specific application strategies and ensure the configurability and ease of debugging of the technical solution, all configurable operating parameters in the specific implementation path of this invention are read through a standardized "configuration interface". The data source of this configuration interface is a "data storage module" (e.g., a non-transitory computer-readable storage medium, such as a configuration file, database entry, or cloud configuration service), which is configured to store configuration data in key-value pair format.

[0153] It should be emphasized that the foregoing embodiments are merely illustrative of preferred implementations of the present invention and are not intended to limit the scope of protection of the present invention. This application also provides a computer-readable storage medium having computer program instructions stored thereon.

Claims

1. A linkage execution control method with adaptive time delay compensation, executed by an application layer flow control unit, characterized in that, The specific steps include: Step S1: Receive the surface state perception data stream of the target material and the real-time production line control data stream from the front-end vision perception node via the industrial fieldbus; extract the extreme difference of the working speed that characterizes the dynamic acceleration and deceleration limit boundary conditions of the main drive equipment based on the real-time production line control data stream. Extract the basic defect risk identifier from the surface condition sensing data stream; compare the extreme difference of the working condition speed with the set limit fluctuation interception threshold to dynamically determine the detection mode status identifier corresponding to the surface condition sensing data stream; and obtain the linkage execution delay compensation amount associated with the detection mode status identifier. Step S2: Using the extreme difference of the working condition speed and the linkage execution delay compensation amount as input, and the detection mode status identifier as the mapping index, call the linkage execution delay compensation amount that matches the detection mode status identifier to perform timing benchmark summation processing, and generate a dynamic advance trigger compensation duration characterizing the physical action advance amount benchmark; and associate the dynamic advance trigger compensation duration with the basic defect risk identifier to generate a comprehensive defect judgment status; Step S3: Verify the extreme difference of the working condition speed with the set failure limit of the actuator response, and generate a synchronous control signal or global control command to trigger the linkage elimination control unit based on the comprehensive defect judgment status and the verification result. Specifically, the synchronous control signal is sent to the downstream linkage rejection control unit, or the global control command is sent to the upper-level bus, in order to eliminate the coordination delay between the front-end visual perception node and the linkage rejection control unit on the application layer flow control unit side. On this basis, the effective time of the issued synchronous control signal can be dynamically synchronized with the real-time transmission position of the target flow material.

2. The linkage execution control method with adaptive time delay compensation according to claim 1, characterized in that: The steps of extracting the extreme value difference of the working speed representing the dynamic acceleration and deceleration limit boundary conditions of the main drive equipment based on the real-time production line control data stream, and extracting the basic defect risk identifier and dynamically determining the detection mode state identifier corresponding to the surface state perception data stream, specifically include: The real-time production line control data stream is analyzed, and the absolute value difference between the macroscopic speed peak parameter and the macroscopic speed trough parameter within the set monitoring period is calculated to generate the extreme value difference of the operating condition speed. In response to the extreme difference of the working condition speed exceeding the set limit fluctuation interception threshold, the full feature traversal comparison path is intercepted at the application layer, and the shallow contour risk identifier is extracted as the basic defect risk identifier. The detection mode status identifier of the surface state perception data stream is updated to the shallow dimension reduction visual detection mode simultaneously. In response to the fact that the extreme difference of the working condition speed does not exceed the set limit fluctuation interception threshold, the full feature traversal comparison path is maintained to extract the standardized multi-dimensional feature cluster vector as the basic defect risk identifier, and is simultaneously updated to the full-volume deep visual detection mode.

3. The linkage execution control method with adaptive time delay compensation according to claim 2, characterized in that: The step of obtaining the linkage execution delay compensation amount associated with the detection mode status identifier specifically includes: Read the historical control execution log, extract the pre-judgment time marker and the action confirmation time marker, and calculate and generate the absolute time difference; If the comprehensive defect judgment state is not generated or the downstream action confirmation feedback is not received within the set control silence time window, a virtual detection synchronization signal with a terminal physical action shielding identifier is generated and sent to the linkage elimination control unit. Extract the round-trip response time of the virtual probe synchronization signal transmitted via the control bus; According to the detection mode status identifier corresponding to the extracted absolute time difference or round-trip response time, the absolute time difference and the round-trip response time are isolated and archived together. Calculate the average value of the isolated archived data, and generate and dynamically refresh the linkage execution delay compensation amount corresponding to the status identifier of each detection mode.

4. The linkage execution control method with adaptive time delay compensation according to claim 3, characterized in that: The steps for performing timing baseline summation and generating a comprehensive defect determination status specifically include: The extracted detection mode status identifier is used as the hierarchical retrieval identifier. The corresponding linkage execution delay compensation amount is matched and called in the global delay configuration table, and it is used as the target execution delay compensation amount. The target execution delay compensation amount is used as an advance compensation factor and is added to the current global control coordination benchmark in advance to generate the dynamic advance trigger compensation duration. In response to the detection mode status being identified as full-scale deep visual detection mode, based on a preset defect semantic evaluation model, the image feature matrix is ​​extracted according to the basic defect risk identifier and forward reasoning is performed to output the matching target hazard weight. In response to the detection mode status being identified as a shallow dimensionality reduction visual detection mode, the basic defect risk identifier is directly converted into an extreme target hazard weight through application layer static mapping bypass. Based on the pre-configured control parameter normalization coefficient, a scale alignment mapping operation is performed on the target hazard weight and the dynamic pre-trigger compensation duration to generate dimensionless priority evaluation parameters and time-series constraint parameters. Based on the process safety tolerance boundary, a cross-threshold comparison is performed on the priority evaluation parameters and time-series constraint parameters to map and generate the comprehensive defect judgment status.

5. The linkage execution control method with adaptive time delay compensation according to claim 4, characterized in that: The steps of sending the synchronization control signaling to the downstream linkage rejection control unit or sending the global control command to the upper-level bus specifically include: In response to the fact that the extreme difference of the working condition speed does not exceed the set failure limit of the actuator response, for the current material flow batch corresponding to the target flow material, the start and end time window of the rejection action used to constrain the linkage rejection control unit is locked, and the state synchronization is triggered in the global scheduling unit to map the control scheduling state of the current material flow batch to the high priority interception confirmation state, and the synchronization control signal is generated and issued based on the above state. In response to the extreme difference in operating speed exceeding the set failure limit of the actuator response, a blocking signal is generated as the global control command; based on the blocking signal, the dynamic generation process of the dynamic pre-trigger compensation duration is intercepted, the pre-configured system preset static compensation constant is forcibly rolled back to replace the dynamic pre-trigger compensation duration, and the subsequent detection mode status identifier is forcibly downgraded and locked to the shallow dimensionality reduction visual detection mode.

6. The linkage execution control method with adaptive time delay compensation according to claim 5, characterized in that: The step of triggering state synchronization within the global scheduling unit to map the control scheduling state of the current material flow batch to a high-priority interception confirmation state specifically includes: Read the current concurrent task sequence in the global scheduling unit and extract the scheduling identifier that represents the processing order of each associated collaborative task; In response to the generated synchronization control signaling, a queue-jumping control beacon is constructed for the current material flow batch; Based on the queue-jumping control beacon, the scheduling identifier corresponding to the current material flow batch is overwritten to prioritize the logical execution of the current material flow batch, thereby mapping and generating the high-priority interception confirmation state.

7. The linkage execution control method with adaptive time delay compensation according to claim 6, characterized in that: The linkage rejection control unit is equipped with a controlled pneumatic actuator; the step of locking the start and end time window of the rejection action of the linkage rejection control unit specifically includes: Read the real-time transmission speed parameters of the target material and the pre-configured absolute transmission distance across workstations; and read the pre-configured physical response time compensation. Based on the aforementioned basic defect risk identifier, the corresponding vertical physical length of the defect is parsed and derived. Based on the real-time transmission speed parameter, the absolute transmission distance across workstations, the longitudinal physical length of the defect, and the exhaust hysteresis constant extracted in advance from the controlled pneumatic actuator, the reference theoretical trigger timestamp is calculated. Based on the physical response time compensation, a bidirectional time-domain extension mapping is performed on the reference theoretical trigger timestamp to construct a removal action start and end time window that is presented as an absolute time-domain closed interval.

Images