Step transition bridging inhibition method based on artificial intelligence

Abstract

The invention provides a step transition bridging suppression method based on artificial intelligence, and the method comprises the steps: collecting the temperature of a pressing point, the pressing force, the tension, the remaining distance in front of a step and the feeding speed in a unified manner in a control time period in front of the step, and constructing a working condition time sequence; through a sequential network containing a material softening index and a stress distribution index, an intervention scoring curve is generated under the constraint of an approach restriction signal, and an intervention time point is determined by adopting a fixed window threshold and a first exceeding rule; in combination with the mechanism and heating response delay, the starting and ending moments of roller micro-lateral deviation and short-time preheating are calculated by using a unified time mapping formula, a linkage execution instruction is formed after sampling grid alignment and boundary check, and the two actions are enabled to take effect synchronously at the step boundary; step transition bridging can be restrained, and the forming quality and the process stability of the step area are improved.

Description

Technical Field

[0001] This invention relates to the field of thermoplastic composite material layup and molding process control technology, and in particular to an artificial intelligence-based method for suppressing step transition bridging. Background Technology

[0002] Predictive control of layup deformation in thermoplastic composites is crucial for achieving high-quality component forming. During layup, the material undergoes complex thermo-mechanical coupling deformation under heating, pressing, and traction, particularly in stepped areas where thickness or cross-section changes abruptly, easily leading to defects such as ply arching, bridging, and localized insufficient compaction. Existing processes typically rely on adjusting parameters such as pressing point temperature, pressing force, tension, and feed rate to minimize defect risks in stepped transition areas while ensuring production cycle time, and to provide a stable initial state for subsequent molding or multi-pass layup.

[0003] To suppress bridging of thermoplastic composites in stepped areas, existing technologies typically rely on process experience or offline process simulation to pre-plan control curves for the temperature, pressing force, and tension at the pressing point, and implement constant or segmented control on the laying equipment. One approach measures the remaining distance before the step and the feed speed; when the remaining distance falls below a set threshold, it triggers lateral roller offset and local heating according to a preset time lead. Another approach introduces a simple time series model or multivariate logic judgment, providing intervention control signals before the step based on the combined changes in temperature, pressure, and tension. A PLC or motion controller then generates roller offset and preheating commands according to a fixed mapping relationship, achieving auxiliary control of the step transition process. Some processes also incorporate molding performance prediction results to limit the allowable temperature and stress range in the stepped area and adjust the aforementioned control curves accordingly.

[0004] However, existing technologies in this field mostly rely on preset process windows or simple rules, making it difficult to provide real-time intervention times that match specific working conditions when considering multi-source timing information such as temperature, pressing force, tension, remaining distance, and feed rate. The trigger times for roller lateral deflection and localized heating are often set independently, without systematically incorporating the differences between mechanism response delays and heating response delays. This makes it difficult to strictly align the actual effective times of these two actions at the step boundary, potentially leaving bridging risks under complex step transition conditions. Furthermore, the correspondence between the control logic and the control time period before the step is not clear enough, hindering precise coordination on the sampling time scale.

[0005] Therefore, a method for suppressing step transition bridging in thermoplastic composite materials that can overcome the shortcomings of the prior art is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an artificial intelligence-based step transition bridging suppression method. The core technical problem to be solved is: when thermoplastic composite materials are laid through abrupt cross-sectional areas such as steps, how to determine and implement the intervention control timing and start and end time of roller micro-lateral deviation and short-term preheating simultaneously at the step boundary within the control period before the step, based on multi-source working condition timing information and combined with mechanism response delay and heating response delay.

[0007] An artificial intelligence-based step transition bridging suppression method according to an embodiment of the present invention includes:

[0008] S1. Obtain the temperature of the pressing point, pressing force, tension, remaining distance before the step and feed speed, calculate the control time period before the step and align the data according to time to form a working condition time series;

[0009] S2. Based on the working condition time series and the step-before control time period, construct the intervention time determination network, generate the approach limit signal according to the step-before remaining distance data in the working condition time series, limit the time range of calculation and output, and prohibit out-of-bounds output within the step-before control time period. Send the combined data of temperature, pressure force and tension into the time series calculation unit, calculate the intervention score under the limitation of the approach limit signal and generate the intervention score curve.

[0010] S3. Set the intervention judgment threshold within the pre-step control period based on the intervention scoring curve, and determine the intervention time point according to the rule of the first time the threshold is exceeded.

[0011] S4. Obtain equipment response delay data and establish a time mapping for the linkage control module based on the intervention time point. Under the constraints of mechanism response delay and heating response delay, calculate the start and end times of roller micro-lateral deviation and short-term preheating. Move the start and end times forward and align them according to the time mapping to form a linkage time table.

[0012] S5. According to the linkage schedule, perform a consistency check during the control period before the step to form a linkage execution command for roller micro-lateral deviation and short-term preheating.

[0013] S6. Based on the linkage execution command, a step transition control command is generated, and the drive roller micro-lateral deviation and short-term preheating are executed before reaching the step.

[0014] Optionally, S1 is as follows:

[0015] Set the pressing point temperature, pressing force, tension, remaining distance before the step and feed speed as the unified sampling objects, establish a sampling trigger synchronized with the feed speed, calibrate the acquisition time mark of each channel and unify the measurement units;

[0016] The time to reach the step boundary is calculated based on the remaining distance in front of the step and the feed rate. The start and end times of the control time period in front of the step are determined, and a time index sequence consistent with the sampling trigger is generated.

[0017] During the control period before the step, the temperature of the pressing point, the pressing force, the tension, the remaining distance before the step and the feed speed are sampled at equal intervals. Data from two hundred time points are collected and time indexes are marked to ensure that each time point contains the above five-dimensional inputs.

[0018] The five-dimensional inputs collected during the pre-step control period are aligned and sorted according to the time index. Data outside the pre-step control period are removed, and the data are combined in chronological order to form a working condition time series containing two hundred time points. The start and end indices of the pre-step control period are marked to limit the time range.

[0019] Optionally, S2 is as follows:

[0020] Using the working condition time series and the control period before the step as input, an intervention time determination network is constructed to calculate the working condition derived data, including material softening index and stress distribution index. The five-dimensional input and two-dimensional derived data are aligned by time to form a seven-dimensional working condition feature vector.

[0021] At the approach limit layer, an approach limit signal is generated based on the control time period before the step and the remaining distance before the step. Time points within the control time period before the step are marked as valid, and time points outside the control time period before the step are marked as invalid, ensuring that the approach limit signal corresponds one-to-one with the working condition time series.

[0022] The seven-dimensional working condition feature vector is fed into the feature fusion layer. One hundred and twenty-eight one-dimensional convolutional neurons are used to extract short-term changes, and thirty-two linear projection neurons are used to map to the input channels of the time-series computing unit to form a fusion sequence consistent with the time index.

[0023] Four parallel state channels are set in the timing calculation unit. Each state channel contains sixty-four linear state neurons and thirty-two gated neurons. The proximity limit signal is input to the gated neurons. When invalid marking occurs, the state of the linear state neurons is frozen and the channel output is blocked. When valid marking occurs, state updates and output are allowed.

[0024] The outputs of the four state channels are concatenated over time and fed into thirty-two linear fusion neurons to obtain a temporal fusion vector. The combined changes of the pressure point temperature, pressure force and tension are then continuously expressed in the temporal fusion vector for score generation.

[0025] In the scoring generation layer, linear output neurons are used to generate intervention scores over time. Near-limit signals are simultaneously input into the gating neurons of the scoring generation layer to restrict the scoring generation. The intervention score is only output at the time point with valid labeling and not output when invalid labeling occurs.

[0026] The output of the scoring generation layer is arranged by time index and hard-truncated at the boundary of the control time period before the step to prevent out-of-bounds output, resulting in the intervention scoring curve. The intervention scoring curve only contains continuous one-dimensional output within the control time period before the step, which is used for threshold determination.

[0027] Optionally, the material softening index in the derived data of the calculation conditions is calculated using the following derived formula:

[0028] ;

[0029] in, For the dimensionless value of the material softening index, For time index Temperature at the crimping point The reference temperature is calibrated based on the temperature baseline within the stable operating range. The temperature scale constant is used for temperature normalization and is calibrated by combining the temperature sensor's range and the temperature standard deviation within the stable operating range. For time index The remaining distance in front of the steps, For time index The feed rate at that location, The characteristic time constant for heating is determined by the response half-life of a short-time preheating step test. It is a natural exponential function.

[0030] Optionally, S3 specifically refers to:

[0031] Read the interventional scoring curve and align it with the time index. Inherit the valid markers of the approach limit signal into the interventional scoring curve, limit the analysis range to the pre-step control time period, and remove the score values ​​corresponding to invalid markers.

[0032] At the beginning of the pre-step control period, the baseline values ​​and fluctuation amplitudes of the intervention score calculations for the first twenty time points are selected. An intervention judgment threshold is generated based on a preset threshold ratio, and the intervention judgment threshold is fixed and does not change with time during the pre-step control period, so that the intervention judgment threshold and the amplitude scale of the intervention score curve are consistent and a single threshold is maintained.

[0033] The intervention score is detected from front to back according to the time index. The time point when the intervention score is first strictly greater than the intervention judgment threshold is found and the time point is determined as a candidate time point. When there is no time point that exceeds the intervention judgment threshold, the end time of the control period before the step is determined as a candidate time point.

[0034] The candidate time points are checked against the boundaries to confirm that they are within the time index range of the control period before the step, and the corresponding actual time is determined as the intervention time point.

[0035] Optionally, S4 specifically refers to:

[0036] Using the intervention time point and equipment response delay data as input, the mechanism response delay and heating response delay are decomposed, a time mapping is established, and the constraint that the effective time is consistent with the intervention time point is written into the time mapping. The start and end boundary constraints of the control time period before the step are added to the time mapping.

[0037] The start and end times of the roller micro-biasing are calculated based on the mechanism response delay and the preset duration of roller micro-biasing. The start time is then shifted forward by the mechanism response delay according to the time mapping, so that the effective time of the roller micro-biasing at the step boundary is consistent with the intervention time.

[0038] The start and end times of short-term preheating are calculated based on the heating response delay and the preset short-term preheating duration. The start time is then shifted forward by the heating response delay according to the time mapping, so that the effective time of short-term preheating at the step boundary is consistent with the intervention time.

[0039] Align the start and end times of the roller micro-lateral deviation and short-term preheating, round the two start and end times to the sampling interval on the time index, set the alignment error threshold as one sampling interval, and if the difference between the two effective times is greater than the alignment error threshold, adjust the start time synchronously until the difference is not greater than the alignment error threshold.

[0040] Boundary checks are performed on the aligned start and end times to restrict the start and end times to within the control time period before the step. If there is a boundary violation, the boundary constraints in the time mapping are used to resolve the start and end times to ensure that the two effects are effective at the step boundary and do not exceed the start and end boundaries of the control time period before the step.

[0041] The start and end times of the roller micro-lateral deviation after alignment and boundary verification, and the start and end times of the short-term preheating are combined in time index order to form a linkage timetable. The effective time corresponding to the step boundary is marked in the linkage timetable for use in calling.

[0042] Optionally, the steps of establishing a time mapping and calculating the start and end times of the roller micro-biasing based on the mechanism response delay and the preset roller micro-biasing duration, and calculating the start and end times of the short-term preheating based on the heating response delay and the preset short-term preheating duration, are performed using a unified time mapping formula, wherein the time mapping formula is specifically as follows:

[0043] ;

[0044] in, For action The starting time, For action At the end of the day, For action identifiers, This marks the start time of the control period before the steps. This is the end time of the control period before the steps. For action Duration The sampling interval is... This is the integer rounding function, which rounds the integer to the nearest integer. The effective time corresponding to the step boundary. For action Response delay, For the operation of taking the larger of two numbers, This is a function to take the smaller of two numbers.

[0045] Optional, S5 specifically includes:

[0046] Based on the linkage schedule, extract the start and end times of roller micro-lateral deviation, the start and end times of short-term preheating, and the effective time corresponding to the step boundary during the control period before the step for consistency check.

[0047] Boundary and sequence checks are performed on the extracted times. Each start and end time is restricted to the control time period before the step. Each end time must be strictly greater than the corresponding start time. The time index must be monotonically increased according to the sampling interval. The difference between the two effective times is compared with a sampling interval. If the difference is not greater than a sampling interval, it is considered to pass.

[0048] The verified linkage timetable is mapped with instructions. The start and end times of roller micro-lateral deviation are mapped to roller micro-lateral deviation start instruction and roller micro-lateral deviation end instruction, respectively. The start and end times of short-term preheating are mapped to short-term preheating start instruction and short-term preheating end instruction, respectively. The effective time mark corresponding to the step boundary is added to the trigger time of the two start instructions.

[0049] The four instructions are sorted by time index and merged into a linkage execution instruction. The linkage execution instruction is triggered within the control time period before the step, and the effective time of the two start instructions is consistent with the effective time corresponding to the step boundary.

[0050] Optional, S6 specifically includes:

[0051] Read the linkage execution instructions, parse the roller micro-lateral deviation start instruction, roller micro-lateral deviation end instruction, short-time preheating start instruction, and short-time preheating end instruction according to the time index, and inherit the effective time markers corresponding to the control time period boundary before the step and the step boundary;

[0052] The four instructions are converted into trigger entries for step transition control instructions, and corresponding trigger times are generated. The effective times of the two start instructions are required to be consistent with the effective times corresponding to the step boundary. All trigger times are required to be within the step pre-control time period and are rounded according to the sampling interval to align the time index.

[0053] The trigger entries are arranged sequentially, with the start instruction placed before the corresponding end instruction according to the time index. When the two instructions have the same time, they are triggered simultaneously, and the effective time is kept consistent. No new time points are inserted between the trigger times.

[0054] The step transition control command is triggered and executed according to the time index. Two start commands are triggered sequentially before reaching the step, and two end commands are triggered at the corresponding end time. During the execution process, the boundary constraints of the control time period before the step are followed, so that the roller micro-lateral deviation and short-term preheating take effect at the step boundary at the same time.

[0055] The beneficial effects of this invention are:

[0056] 1. This proposal presents an improved intervention time determination method. Based on the Mamba-2 state-space sequence model, it introduces working condition-derived features of material softening and stress distribution indices. It also incorporates proximity-limited gating and hard-truncation mechanisms within the network, updating and scoring the five-dimensional time-series data of the pressing point temperature, pressing force, tension, remaining distance before the step, and feed rate only during the control period before the step. Unlike existing intervention determination methods that rely on preset process windows or simple time-series models, this proposal combines one-dimensional convolution with multi-channel linear state units to continuously model the inter-temporal changes in thermo-mechanical conditions. It uses a fixed window threshold and a first-overlap rule to determine the intervention time point on a single time index. This ensures that the intervention determination is strictly constrained by the effective range before the step while dynamically matching the specific working conditions, thereby improving the consistency between the intervention time and the actual step transition state without increasing online complexity.

[0057] 2. This proposal presents a novel time mapping method for the linkage between roller micro-lateral offset and short-term preheating. It simultaneously incorporates the mechanism response delay and heating response delay into a unified time mapping formula, using the effective time corresponding to the step boundary as the constraint target. The start and end times of both types of control actions are calculated through an integrated approach of advancing the start time, rounding the sampling grid, and trimming the boundary. Unlike existing technologies that separately set trigger advance amounts or use fixed mapping relationships without systematically handling the differences in the two types of response delays, this proposal further quantifies and fine-tunes the error of the two effective times on the sampling time axis. It iteratively corrects asynchronicity cases exceeding one sampling interval and resolves the out-of-bounds start and end times by combining the start and end boundaries of the control time period before the step. This ensures that roller micro-lateral offset and short-term preheating achieve synchronous effectiveness within the sampling scale at the step boundary, avoiding misalignment or local absence of the two types of actions at key sections, thus better suppressing step transition bridging.

[0058] 3. This proposal presents a holistic control method for suppressing step transition bridging in thermoplastic composite material layup. Through a series of techniques—"unified sampling and time alignment—Mamba-2 intervention scoring—threshold determination of intervention time point—response delay time mapping—linked execution instructions—step transition control instructions"—multi-source operating condition timing information and equipment response characteristics are integrated under the same time indexing system. Unlike existing technologies that only provide control curves or simple triggering logic, this proposal completes the generation of intervention scoring curves, determination of intervention time points, alignment of the start and end times of the two types of controls, and encoding of executable instructions on the same sampling benchmark. This ensures a clear and verifiable mapping relationship between the start and end actions of roller micro-lateral deviation and short-term preheating, the control time period before the step, and the effective time of the step boundary. Therefore, when laying through areas of abrupt cross-sectional change such as steps, it provides a clear, well-defined, and adaptable time-series control framework for simultaneously implementing two types of controls, which helps reduce the risk of step transition bridging and improve process stability. Attached Figure Description

[0059] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0060] Figure 1 This is a flowchart of an artificial intelligence-based step transition bridging suppression method proposed in this invention;

[0061] Figure 2 This is a flowchart of the working condition time series acquisition and organization process for an artificial intelligence-based step transition bridging suppression method proposed in this invention.

[0062] Figure 3This is a flowchart of the intervention time determination network for an artificial intelligence-based step transition bridging suppression method proposed in this invention.

[0063] Figure 4 This is a flowchart illustrating the intervention time point determination of an artificial intelligence-based step transition bridging suppression method proposed in this invention.

[0064] Figure 5 This is a flowchart of the time mapping and start / end time calculation of the linkage control module of the artificial intelligence-based step transition bridging suppression method proposed in this invention.

[0065] Figure 6 This is a schematic diagram illustrating the linkage control between the laying head and the step area in an artificial intelligence-based step transition bridging suppression method proposed in this invention.

[0066] Figure 7 This is a schematic diagram of the time-determining network structure of an artificial intelligence-based step transition bridging suppression method proposed in this invention. Detailed Implementation

[0067] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0068] refer to Figures 1 to 7 A step transition bridging suppression method based on artificial intelligence, characterized by comprising:

[0069] S1. Obtain the temperature of the pressing point, pressing force, tension, remaining distance before the step and feed speed, calculate the control time period before the step and align the data according to time to form a working condition time series;

[0070] S2. Based on the working condition time series and the step-before control time period, construct the intervention time determination network, generate the approach limit signal according to the step-before remaining distance data in the working condition time series, limit the time range of calculation and output, and prohibit out-of-bounds output within the step-before control time period. Send the combined data of temperature, pressure force and tension into the time series calculation unit, calculate the intervention score under the limitation of the approach limit signal and generate the intervention score curve.

[0071] S3. Set the intervention judgment threshold within the pre-step control period based on the intervention scoring curve, and determine the intervention time point according to the rule of the first time the threshold is exceeded.

[0072] S4. Obtain equipment response delay data and establish a time mapping for the linkage control module based on the intervention time point. Under the constraints of mechanism response delay and heating response delay, calculate the start and end times of roller micro-lateral deviation and short-term preheating. Move the start and end times forward and align them according to the time mapping to form a linkage time table.

[0073] S5. According to the linkage schedule, perform a consistency check during the control period before the step to form a linkage execution command for roller micro-lateral deviation and short-term preheating.

[0074] S6. Based on the linkage execution command, a step transition control command is generated, and the drive roller micro-lateral deviation and short-term preheating are executed before reaching the step.

[0075] In this embodiment, step S1 specifically includes:

[0076] To acquire and organize five-dimensional working condition data within the pre-step control period, a consistent input and time range marker are provided for the timing input layer and proximity constraint layer of the intervention time determination network. The pressure point temperature is denoted as... The pressing force is recorded as Let tension be denoted as The remaining distance in front of the steps is denoted as The feed rate is denoted as The total number of sampling points is recorded as The sampling interval is denoted as Record the time index as The start and end times of the control period before the steps are recorded as follows: and The time to reach the boundary of the step is recorded as The start and end indices of the control time period before the steps are denoted as... and The operating condition time series is recorded as The distance threshold for determining the step boundary is denoted as... The fixed time delay difference of the channel is denoted as ;

[0077] First, , , , , Set as a uniform sampling object, to As a sampling trigger reference, a sampling trigger synchronized with the feed rate is established so that any trigger moment corresponds to the material's forward position. The time stamps of the five acquisition channels are calibrated, and the calibration pulse is used as a unified time reference. The fixed time delay difference is measured for each channel. Furthermore, a translation calibration is performed on the time markers to align the sampling records corresponding to the same trigger moment on the time axis, and the units of measurement are standardized. Use a unified unit of temperature. and Adopt a unified unit of force. Use a uniform unit of length. A unified unit of speed is adopted to ensure that the dimensions of the five-dimensional input are consistent under the same time index;

[0078] Secondly, based on The time to reach the step boundary is determined by the synchronous value triggered by sampling. ,Will Set as the distance threshold for determining the step boundary. Pre-calibrated based on the equipment's ranging resolution and step boundary positioning accuracy, the equipment is then inspected point by point along the time index from front to back. When it first appears no greater than Determine the trigger time This moment will be taken as the end of the pre-step control period. ,Will Set to two hundred time points, Backtrack along the time axis by sampling interval until coverage is achieved. At each trigger moment, the control time period before the step is obtained. ,exist Internally generate a time index sequence consistent with the sampling trigger, so that... from Monotonically increasing to The trigger time interval corresponding to adjacent indices is ;

[0079] Secondly, sampling is performed at equal intervals during the control period before the step, and the settings are as follows: For two hundred time points, make Covering two hundred discrete moments, indexing each time point. collection , , , and , will pass The calibrated timestamps are appended to the sampling records of this index to ensure that all channels are on the same track. Strict alignment is required, and integrity checks are performed on the sampled records. Each time index must contain the aforementioned five-dimensional inputs, with no missing or duplicate samples, ensuring accuracy regarding... The five-dimensional data frames are continuous and traceable;

[0080] Subsequently, the five-dimensional inputs collected during the control period before the steps were aligned and sorted according to the time index, and those not in the time index were sorted. The sampled records were directly discarded, and cross-segment splicing was not performed. The samples were combined chronologically to obtain a working condition time series containing two hundred time points. This working condition time series is denoted as... ,in In the index The place contains , , , and Synchronous value retrieval and synchronous annotation and This is used to limit the valid range of subsequent timing calculations and prevent out-of-bounds output. It also checks the consistency between the time index and the trigger time, ensuring that any... Corresponding to a unique trigger time and connected with Maintain consistency;

[0081] Finally, the operating condition time series and start and end index , Provided to the intervention time determination network, wherein... Entering the timing input layer, In time index The following sequence and Together, they act on the proximity limit layer to generate a proximity limit signal, which is used to limit the time range of state updates and scoring outputs within the time-series calculation unit. Through the above-mentioned unified sampling, boundary judgment, and sequence organization, the working condition time series is synchronized with the feed rate, consistent with the pre-step control time period, and matched with two hundred time points. This provides a stable and alignable five-dimensional input basis for forming an intervention scoring curve related to step transition bridging suppression in the time-series calculation unit, and provides a clear time reference and effective index range for threshold determination and intervention time point determination after generating the intervention scoring curve.

[0082] In this embodiment, step S2 specifically includes:

[0083] This step outlines the construction and operation process of the intervention time determination network. The inputs are the operating condition time series and the pre-step control period, and the output is the intervention scoring curve. The time index is denoted as... The start and end indices are denoted as follows: and The five-dimensional input vector of the operating condition time series is denoted as... ,in Synchronous values ​​of contact point temperature, pressing force, tension, remaining distance before step, and feed rate are included, indexed by time. The temperature at the crimping point is denoted as index the time The pressure at the point is denoted as index the time The tension at the point is denoted as index the time The remaining distance in front of the steps is denoted as index the time The feed rate at point is denoted as The material softening index is included in the index. The scalar at the location is denoted as The force distribution index is placed in the index. The scalar at the location is denoted as Let the seven-dimensional working condition feature vector be denoted as The proximity limit signal will be at the index. The mark at the location is denoted as The output of the feature fusion layer is denoted as The fused output of the timing computation unit is denoted as The output of the rating generation layer is denoted as The number of one-dimensional convolutional neurons is denoted as The number of linear projection neurons is denoted as The number of parallel state channels is denoted as The number of linear state neurons in each channel is denoted as... The number of gated neurons in each channel is denoted as The number of linearly fused neurons is denoted as ;

[0084] This application proposes a derived formula based on Calculation of operating condition derived data and ,right A normalized characterization based on temperature and remaining time to the step boundary is adopted to couple temperature and travel conditions under the same dimension, and to ensure that this characterization can be directly calculated from the collected data and calibration constants. The source is clearly defined, and it is constructed based on the underlying numerical values ​​of compressive force and tension. It is generated using a combination of ratios, differences, and short-term fluctuations within a window, and is indexed at the same time. Alignment to form a scalar for network input:

[0085] ;

[0086] in, For the dimensionless value of the material softening index, For time index Temperature at the crimping point The reference temperature is calibrated based on the temperature baseline within the stable operating range. The temperature scale constant is used for temperature normalization and is calibrated by combining the temperature sensor's range and the temperature standard deviation within the stable operating range. For time index The remaining distance in front of the steps, For time index The feed rate at that location, The characteristic time constant for heating is determined by the response half-life of a short-time preheating step test. It is a natural exponential function;

[0087] Secondly, for In time index The calculation is performed using a fixed-length window, and the window length is denoted as . The sampling interval and the mechanism response time are used for calibration. Covering several sampling intervals, the ratio sequence and difference sequence of compressive force and tension are extracted within this window. The mean of the ratio and the mean of the difference are calculated, and the joint fluctuation amplitude of the two within the window is calculated. The weighting constant is denoted as... , and The values ​​were obtained from the force distribution calibration test. The three quantities were weighted, summed, and linearly normalized. The result was used as... ,make It produces a discernible response during the redistribution of compressive force and tension, and maintains consistency with... The amount of data collected at the lowest level directly corresponds to the data collected at that level. and Perform time index alignment, By concatenating the two at the same index, a seven-dimensional feature vector of the working condition is obtained. ;

[0088] Subsequently, near the constraint layer, the distance is generated based on the control time period before the step and the remaining distance before the step. ,Will to Time indices within the specified range are marked as valid, while time indices outside the range are marked as invalid. and Perform a consistency check, requiring exist The interval maintains a valid label, in or The interval is kept invalid, so that It corresponds one-to-one with the working condition time series at the index-by-index level, and forms boundary constraints for subsequent state calculations and scoring outputs;

[0089] Will Feed into the feature fusion layer and set For 128 one-dimensional convolutional neurons, Short-term change extraction is performed, encoding the coordinated changes in temperature, compressive force, and tension within a small window as local features. With thirty-two linear projection neurons, the channels after convolution are mapped to the input channel dimension of the temporal computation unit, resulting in a fusion sequence consistent with the time index. This is used to drive subsequent state calculations;

[0090] An improved implementation of the Mamba-2 architecture is adopted in the timing computation unit, and the following settings are made: There are four parallel state channels, each channel is set with For sixty-four linear state neurons and For thirty-two gated neurons, Linear state neurons fed into each channel in parallel perform state propagation across time. The gated neurons fed into each channel are time-domain restricted, in When an invalid label is displayed, the internal state of the linear neuron is frozen and the channel output is masked. This means the state value of the previous valid index remains unchanged, and no channel output is generated for the current index. To allow state updates and channel output when valid tags are set, the four channels are indexed. The output is concatenated with the input at the location. Generate temporal fusion vectors for thirty-two linear fusion neurons. This causes the combined changes of heat, force, and velocity to occur in Maintain continuous expression in, and Generate usable timing representations within the range;

[0091] Next, The data is fed into the scoring generation layer, which consists of linear output neurons and gating neurons. The linear output neurons generate a single-channel output according to the time index. Gated neurons receive And restrict the scoring output, so that Only A numerical value is generated when a valid tag is created. No output is generated when the flag is invalid. Sort by time index, and in and A hard truncation is performed at the boundary, clearing the output at the out-of-bounds index to zero, resulting in an intervention scoring curve that contains only the continuous one-dimensional output during the pre-step control time period. ;

[0092] Then, and The corresponding relationships need to be checked, requiring any If If it is an invalid tag, then There is no valid score, which is used to verify whether the proximity limit signal is effective in constraining state updates and output generation within the network, thereby ensuring that the intervention score curve falls strictly within the pre-step control time period and avoiding the generation of output within the invalid time range.

[0093] Finally, the intervention scoring curve will be used. This will serve as the sole basis for subsequent threshold determination and intervention timing, and will be maintained. and , , The correspondence remains unchanged, through the above... For input, with To restrict, with As an intermediate representation of the serial link, it completes the continuous modeling of the combined changes of the pressure point temperature, pressing force and tension in the parallel state channel of the time-series computing unit, and forms a one-dimensional intervention scoring curve directly related to the step transition bridging suppression scenario in the scoring generation layer, providing a single time series output consistent with the physical process for subsequent threshold determination and time mapping.

[0094] In this embodiment, step S3 specifically includes:

[0095] This step focuses on threshold determination and intervention time point identification for the interventional scoring curve. The inputs are the interventional scoring curve and the proximity limit signal; the output is the intervention time point. The time index is denoted as... The start and end indices of the control time period before the step are denoted as follows: and The intervention score curve will be in the index The value at point is denoted as The proximity limit signal will be at the index. The mark at the location is denoted as The sampling interval is denoted as The threshold calculation window length is denoted as The baseline value used for threshold calculation is denoted as The fluctuation range used for threshold calculation is denoted as The preset threshold multiplier is denoted as The intervention threshold is denoted as The candidate time index is denoted as The actual time corresponding to the intervention point is recorded as ;

[0096] Align the interventional scoring curve with time index. and Arrange them in a one-to-one correspondence, Valid mark inheritance to ,exist Remove the score value at the index of the invalid marker. and Within the index range, no rating entries are retained; only [the following are retained] For consecutive scoring items, perform monotonicity checks on the index sequence, requiring... Increment by sampling interval, and in and Hard truncation is performed at the boundary, so that the intervention score curve forms a detectable single time series within the pre-step control period.

[0097] Secondly, a threshold calculation window is set at the beginning of the control time period before the step, and... Set up twenty time points, and take to Within range The arithmetic mean of the samples is used as the threshold for calculation. The difference between the maximum and minimum values ​​of the sample is used as... ,Will Set to a preset constant and provided by offline calibration to keep the threshold ratio stable under different operating conditions. Set as and fix exist The range does not change over time, making and Maintain a consistent amplitude scale and use only a single threshold;

[0098] Then, perform the first exceedance rule check based on the time index, and Traverse from front to back, when encountering... and Simultaneously, a comparison is performed at the index where both the score exists and the label is valid. Strictly greater than The index is recorded at that time. And immediately stop traversing when traversing to If no index satisfies the strict greater than condition, then Set as This ensures that the candidate time index always exists during the control period before the step. Perform uniqueness checks, requiring that only the single index that meets the conditions for the first time be retained, without duplicate records, and without inserting new time points during the detection process;

[0099] Then, boundary checks and time mapping are performed on the candidate time indices. and , Compare and confirm. satisfy ,Will Mapped to actual time The method of using the start time of the control period before the step as a reference and accumulating the data indexed sequentially according to the sampling interval is adopted to obtain the result. This ensures that the time mapping is consistent with the sampling trigger and with the time reference of the operating condition time series. Perform mark verification, requiring Corresponding index To ensure effective marking and consistency between the intervention time point and the boundary constraints of the pre-step control period;

[0100] Finally, Used as an intervention point for subsequent steps and maintained and , , The correspondence remains unchanged. Through the linear process of reading and alignment, setting fixed window thresholds, first exceeding rule detection and boundary verification, the intervention scoring curve forms a determinable time point under the constraint of the approach limit signal. This ensures that the intervention time point comes from a single threshold judgment within the control period before the step and is consistent with the output of the intervention time judgment network, supporting the time mapping and start and end time calculation of the subsequent linkage control module.

[0101] In this embodiment, step S4 specifically includes:

[0102] This specific implementation focuses on the time mapping and start / end time calculation of the linkage control module. The inputs are the intervention time point and equipment response delay data, and the output is the linkage time schedule. The intervention time point is denoted as... Record the device response delay data as The delay in the agency's response will be recorded as The heating response delay is denoted as The start and end times of the control period before the steps are recorded as follows: and The effective time corresponding to the step boundary is recorded as The sampling interval is denoted as Record the time index as The duration of the slight lateral deviation of the roller is recorded as The duration of short-term preheating is recorded as The start and end times of the slight lateral deflection of the roller are denoted as follows: and The start and end times of the short-term preheating are denoted as follows: and The alignment error threshold is denoted as Record the linkage schedule as ;

[0103] First, and As input, Perform channel decomposition to obtain and The calibration was performed by recording the channel response time constant during the step-triggered test within the control period before the step. Set the effective time corresponding to the step boundary, with To unify the target time, a time-reverse calculation is carried out so that the two types of effects take effect at the same target time.

[0104] Secondly, a time mapping is established and constraints are written. The time mapping includes effective time constraints and boundary constraints. The effective time constraints require that both types of actions... For the boundary constraints to take effect, all start and end times must be within the specified range. Using sampling interval As a discretization benchmark, all triggering moments can be mapped to a time index. The corresponding sampling time should be used to avoid generating misaligned moments.

[0105] A unified time mapping formula is used to calculate the start and end times of the two types of actions, and the action identifier is recorded as follows. ,in This indicates a slight lateral deviation of the roller. This indicates a short preheating period, the corresponding duration of which is... The response delay is :

[0106] ;

[0107] in, For action The starting time, For action At the end of the day, For action identifiers, This is the start time of the control period before the steps. This is the end time of the control period before the steps. For action Duration The sampling interval is... This is the integer rounding function, which rounds the integer to the nearest integer. The effective time corresponding to the step boundary. For action Response delay, For the operation of taking the larger of two numbers, This is the operation of taking the smaller of two numbers;

[0108] Based on the above time mapping, we obtain , and , This mapping simultaneously performs forward shifting, sampling rounding, and boundary clipping within a single formula, ensuring that it does not exceed the limits. Under the premise of Each response delay is used to backtrack to the executable trigger time, and the two end times are kept consistent with their respective durations;

[0109] The alignment process is then executed, calculating the two types of effective errors affecting the sampling alignment. and ,Will Set a sampling interval, when any error is greater than When choosing to take effect later than One side adjusts its starting time forward by one When it takes effect earlier than When the error on one side is larger, adjust its starting time backward by one. After each adjustment, the error is rechecked until the error on both sides is no greater than [value missing]. And keep the rounding time unchanged up to the sampling time;

[0110] Next, perform boundary verification. , , and and A comparison is performed; if any boundary is exceeded, then... Without changing the constraints, shift the corresponding start time in the same direction along the effective equivalent constraints so that the new start and end times fall within the constraints. And press again Round down until the boundary check is passed;

[0111] Then, a linkage schedule was constructed. ,Will , , and Index by Time Write in ascending order In relation to The corresponding time index is marked with the effective time marker of the step boundary. A sequential check is performed, requiring that the start time of each type be strictly earlier than the corresponding end time, and that the effective activation times of the two types of effects point to the same point. ;

[0112] Finally, As the sole time reference for the linkage control module, executable trigger entries are generated within the control period before the step. The linkage control module does not perform amplitude calculations or trajectory replanning; instead, it employs a sequential process of time mapping, sampling alignment, and boundary verification to ensure that the roller's micro-lateral deviation and short-term preheating occur within the specified time frame. Effective simultaneously, and The system provides a directly executable start and end time sequence, providing a complete, alignable, and constraint-compliant time base for the formation of subsequent linkage execution instructions and step transition control instructions.

[0113] In this embodiment, step S5 specifically includes:

[0114] This step focuses on generating the linkage execution command. The inputs are the linkage timetable and the pre-step control time period, and the output is the linkage execution command. The linkage timetable is denoted as... Record the time index as The sampling interval is denoted as The start and end times of the control period before the steps are recorded as follows: and The start and end times of the slight lateral deflection of the roller are denoted as follows: and The start and end times of the short-term preheating are denoted as follows: and The effective time corresponding to the step boundary is recorded as The delay in the agency's response will be recorded as The heating response delay is denoted as The start and end commands for the roller micro-deviation are denoted as follows: and The short-term preheating start and end commands are denoted as follows: and The set of instructions for coordinated execution is denoted as ;

[0115] First, from Extract time sequence entries within the control time period before the step, and read them according to the time index. , , , Simultaneously read Extracting entries and A consistency check is performed, requiring each entry to have a unique time index, which monotonically increases with the sampling interval. The entry originates from... The same time reference and consistent with the control period before the step;

[0116] Secondly, boundary checks and sequence checks are performed on the extracted time points. , , , respectively with Perform a comparison to restrict all start and end times from falling into the range. When an out-of-bounds error occurs, the order of entries should be maintained, and the sequence should be backtracked or moved forward according to the most recent sampling time until the boundary constraints are met. The sequence relationship should be checked. Strictly greater than ,Require Strictly greater than The effective times of the two types are calculated one by one, so that the effective time of the roller micro-bias is... and Adding them together, we get the effective time of the short preheating from... and The sum is obtained by taking the absolute value of the difference between the two effective times and multiplying it by a sampling interval. Perform a comparison; when the difference is no greater than If the result is deemed successful, it is recorded as unsuccessful and the process enters the initial fine-tuning procedure until it is successful.

[0117] Subsequently, instructions will be mapped through the verification and linkage schedule, in order to and Mapped to respectively and ,by and Mapped to respectively and In the trigger records of the two types of start instructions, an effective time marker corresponding to the step boundary is added, and the marker content is as follows: This is used for effectiveness verification during subsequent steps, and the instruction mapping remains consistent with... Consistent time index, maintaining consistency with A consistent sampling triggering benchmark is maintained, without altering the order of time entries;

[0118] Next, the four instructions are sorted by time index and merged into a set of instructions for simultaneous execution. During the sorting process, index deduplication and gap filling are performed. It is required that no time index repeatedly records different types of end instructions, that the time index matches the sampling interval, and that no non-sampling moments are inserted. The trigger range is limited to And check the effective time of the two types of start instructions in the set. Consistency, meaning the start command for slight roller misalignment is triggered after... achieve The short-term preheating start command is triggered after... achieve If a discrepancy is found during the verification, the process returns to the previous step to fine-tune the sampling grid at the starting time and regenerate. ;

[0119] Then, to Perform integrity checks and consistency records, requiring Earlier Furthermore, the index difference between the two covers the duration of the slight lateral deviation of the roller, requiring... Earlier Furthermore, the index difference between the two covers the duration of the short-term warm-up, requiring the addition of two types of start instructions. Consistent marking requires all instructions to be in the same place. The data should be arranged in a monotonically increasing order, and the set of records should have a uniform format and include three fields: instruction type, trigger time, and effective flag.

[0120] Finally, output As a coordinated execution command, it is triggered by time index within the control period before the step. It directly provides the control system with executable trigger entries, without introducing amplitude calculation or trajectory replanning, maintaining consistency with the time base of the linkage schedule, and ensuring that the effective time of the two start commands corresponds to the step boundary. It is consistent and provides an aligned and verifiable set of time-series instructions for the formation of step transition control instructions.

[0121] In this embodiment, step S6 specifically includes:

[0122] This step focuses on the construction and triggering of step transition control instructions. The inputs are a set of linked execution instructions and a pre-step control time period; the output is the step transition control instructions. The set of linked execution instructions is denoted as... Record the time index as The sampling interval is denoted as The start and end times of the control period before the steps are recorded as follows: and The effective time corresponding to the step boundary is recorded as The start and end commands for the roller micro-deviation are denoted as follows: and The short-term preheating start and end commands are denoted as follows: and The delay in the agency's response will be recorded as The heating response delay is denoted as The effective time of the roller micro-bias is recorded as The effective time of short-term preheating is recorded as The set of step transition control commands is denoted as ;

[0123] First, read and according to Four instructions were parsed to obtain , , and The trigger index and trigger time on the sampling grid will and As a boundary constraint, it is inherited into this step, As the effective time marker, it is inherited as the first entry of the four instructions. Perform consistency checks on the records, requiring all triggering indexes to be unique, and the indexes are based on... Incremental, with all records using the same time base as the control period before the step;

[0124] Secondly, the four instructions are converted into trigger entries for step transition control instructions, so as to and Generate start and end trigger entries for the roller micro-biasing respectively, so as to... and Generate start and end trigger entries for short-term preheating respectively, and apply them to the four trigger times. Round the data, write the rounded trigger index into the trigger entry, and calculate... After the start trigger entry for the roller micro-bias is triggered, Calculate the arrival time. The start of the short-term warm-up trigger entry is triggered after... The arrival time is checked against the records of the two types of entries that began to trigger. and All with Consistency, and maintained in records As an effective flag field;

[0125] Then, the trigger entries are sequentially arranged, with the start trigger entry placed before the corresponding end trigger entry. It is required that the start trigger index of the same type be strictly less than the end trigger index. When two types of trigger entries appear at the same time index, they are triggered simultaneously, and this is maintained between the two start trigger entries. Consistent sequential arrangement ensures that no new time points are inserted between trigger times, the rounding result of the sampling grid is not changed, and the index sequence is maintained according to the specified order. Monotonically increasing;

[0126] Next, boundary checks are performed on the sequentially arranged trigger entries, comparing all trigger times with... A comparison is performed; if any out-of-bounds entries are found, they are either moved back or forward to within the boundary based on the most recently sampled index, and then re-verified. and and The consistency is maintained; when consistency is broken, the triggering index is synchronized and fine-tuned within the sampling grid until the effective times of the two types are consistent. Maintain consistency; do not change the type or order of items during the verification process.

[0127] Subsequently, a set of step transition control instructions is generated. The four types of trigger entries are sorted as follows: Ascending order write And append to the two types of start-trigger entries. As an effective flag field, in The internal execution integrity check requires that the start and end trigger entries for roller micro-bias cover the actual duration range, the start and end trigger entries for short-term preheating cover the actual duration range, and the set record format be uniform and include four fields: instruction type, trigger index, trigger time, and effective flag.

[0128] Finally, execution is triggered by time index. At each trigger index within the control time period before reaching the step, the triggers are activated sequentially. and And trigger at their respective end trigger indexes and Continuously follow during the execution process Boundary constraints require and Arrive at the edge of the steps simultaneously To ensure that the roller's slight lateral deviation and short-term preheating are within the range of time required... Simultaneously take effect, and complete the alignment triggering and closing execution of the step transition control command within the step pre-control time period.

[0129] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An artificial intelligence-based step transition bridging suppression method, characterized by, The method comprises the following steps: S1, acquiring the temperature of the crimping point, the pressing force, the tension, the remaining distance before the step, and the feeding speed, calculating the control time period before the step and aligning the data by time to form a working condition time sequence; S2, constructing an intervention time judgment network based on the working condition time sequence and the control time period before the step, generating an approach limiting signal according to the remaining distance before the step in the working condition time sequence, limiting the time range of calculation and output and prohibiting the output from exceeding the limit within the control time period before the step, inputting the combined data of the temperature, the pressing force and the tension into a time sequence calculation unit, calculating an intervention score under the limitation of the approach limiting signal and generating an intervention score curve; S3, setting an intervention judgment threshold within the control time period before the step according to the intervention score curve, determining the time point as the intervention time point according to the rule of first exceeding the threshold; S4, acquiring the device response delay data, and establishing a time mapping of the linkage control module according to the intervention time point, calculating the start and end time of the roller micro-lateral deviation and short-time preheating under the constraints of the mechanism response delay and the heating response delay, moving the start and end time forward according to the time mapping and aligning them to form a linkage time table; S5, performing consistency check within the control time period before the step according to the linkage time table to form linkage execution instructions of the roller micro-lateral deviation and short-time preheating; S6, forming a step transition control instruction according to the linkage execution instruction to drive the roller micro-lateral deviation and short-time preheating to execute before reaching the step.

2. The step transition bridging suppression method based on artificial intelligence according to claim 1, characterized in that, S1 specifically comprises: Setting the temperature of the crimping point, the pressing force, the tension, the remaining distance before the step, and the feeding speed as unified sampling objects, establishing a sampling trigger synchronized with the feeding speed, calibrating the collection time markers of each channel and unifying the units of measurement; Calculating the time of reaching the step boundary according to the remaining distance before the step and the feeding speed, determining the start time and end time of the control time period before the step, and generating a time index sequence consistent with the sampling trigger; Acquiring the temperature of the crimping point, the pressing force, the tension, the remaining distance before the step, and the feeding speed at equal intervals within the control time period before the step, collecting data of two hundred time points and labeling time indexes, ensuring that each time point contains the above five-dimensional input; Aligning and sorting the five-dimensional input collected within the control time period before the step according to the time index, eliminating data not within the control time period before the step, combining to form a working condition time sequence containing two hundred time points in chronological order, and labeling the start and end indexes of the control time period before the step for time range limitation. 3.The step transition bridging suppression method based on artificial intelligence according to claim 1, wherein, S2 specifically comprises: Taking the working condition time sequence and the control time period before the step as inputs, constructing an intervention time judgment network, calculating working condition derived data including material softening indicators and stress distribution indicators, aligning the five-dimensional input and two-dimensional derived data by time and forming a seven-dimensional working condition feature vector; Generating an approach limiting signal according to the control time period before the step and the remaining distance before the step in the approach limiting layer, marking the time points within the control time period before the step as valid and the time points outside the control time period before the step as invalid, ensuring that the approach limiting signal and the working condition time sequence correspond one-to-one; The seven-dimensional working condition characteristic vector is sent to a characteristic fusion layer, one-dimensional convolutional neurons are used to extract short-time changes, and linear projection neurons are used to map to input channels of a time sequence calculation unit, so as to form a fusion sequence consistent with a time index; Four parallel state channels are arranged in the time sequence calculation unit, each state channel contains sixty-four linear state neurons and thirty-two gate neurons, a proximity limiting signal is input into the gate neurons, the state of the linear state neurons is frozen and the channel output is shielded when the invalid flag is invalid, and the state is updated and output when the valid flag is valid; The outputs of the four state channels are spliced by time and sent to thirty-two linear fusion neurons to obtain a time sequence fusion vector, and the combined changes of the pressure point temperature, the rolling force and the tension are continuously expressed in the time sequence fusion vector for score generation; Linear output neurons are used in the score generation layer to generate an intervention score by time, a proximity limiting signal is input into the gate neurons of the score generation layer at the same time, the score generation is limited, and the intervention score is output only at the time point of the valid flag, and no output is generated when the invalid flag is invalid; The output of the score generation layer is arranged according to the time index and hard cut is performed at the boundary of the pre-step control time period to prevent the generation of out-of-boundary output, so as to obtain an intervention score curve, and the intervention score curve only contains continuous one-dimensional output in the pre-step control time period for threshold determination.

4. The step transition bridging suppression method based on artificial intelligence according to claim 3, characterized in that, The material softening index in the calculation working condition derived data is calculated by the following derivation formula: ； wherein, is a dimensionless value of a material softening indicator, is a time index is a crimping point temperature at the time index, is a reference temperature and is calibrated from a temperature baseline within the stable operating interval, is a temperature scaling constant for temperature normalization and is calibrated from a combination of the temperature sensor range and the temperature standard deviation of the stable operating interval, is a time index is a remaining distance to the step at the time index, is a time index is a feed speed at the time index, is a heating characteristic time constant and is calibrated from the response half-life of a short pre-heat step test, is a natural exponential function.

5. The step transition bridging suppression method based on artificial intelligence according to claim 1, wherein, S3 is specifically: The intervention score curve is read and aligned with the time index, the valid flag of the proximity limiting signal is inherited to the intervention score curve, the analysis range is limited to the pre-step control time period, and the score values corresponding to the invalid flags are removed; The intervention score at the starting point of the pre-step control time period is selected to calculate a reference value and a fluctuation amplitude, an intervention determination threshold is generated according to a preset threshold multiple, and the intervention determination threshold is fixed and does not change with time in the pre-step control time period, so that the intervention determination threshold is consistent with the amplitude scale of the intervention score curve and remains a single threshold; The intervention score is detected from the front to the back according to the time index, the time point at which the intervention score is first strictly greater than the intervention determination threshold is found, and the time point is determined as a candidate time point, and when there is no time point exceeding the intervention determination threshold, the end time of the pre-step control time period is determined as the candidate time point; Boundary checking is performed on the candidate time point to confirm that it is located within the time index range of the pre-step control time period, and the actual time corresponding to the candidate time point is determined as an intervention time point.

6. The step transition bridging suppression method based on artificial intelligence according to claim 1, wherein, S4 is specifically: The intervention time point and the equipment response delay data are taken as inputs to obtain the mechanism response delay and the heating response delay through decomposition, a time mapping is established, a constraint that the effective time is consistent with the intervention time point is written into the time mapping, and the start and end boundary constraints of the pre-step control time period are added to the time mapping; The start time and the end time of the roller micro-lateral deviation are calculated according to the mechanism response delay and the preset roller micro-lateral deviation duration, and the start time is moved forward by the mechanism response delay according to the time mapping, so that the effective time of the roller micro-lateral deviation at the step boundary is consistent with the intervention time point. The start and end times of short-term preheating are calculated based on the heating response delay and the preset short-term preheating duration. The start time is then shifted forward by the heating response delay according to the time mapping, so that the effective time of short-term preheating at the step boundary is consistent with the intervention time. Align the start and end times of the roller micro-lateral deviation and short-term preheating, round the two start and end times to the sampling interval on the time index, set the alignment error threshold as one sampling interval, and if the difference between the two effective times is greater than the alignment error threshold, adjust the start time synchronously until the difference is not greater than the alignment error threshold. Boundary checks are performed on the aligned start and end times to restrict the start and end times to within the control time period before the step. If there is a boundary violation, the boundary constraints in the time mapping are used to resolve the start and end times to ensure that the two effects are effective at the step boundary and do not exceed the start and end boundaries of the control time period before the step. The start and end times of the roller micro-lateral deviation after alignment and boundary verification, and the start and end times of the short-term preheating are combined in time index order to form a linkage timetable. The effective time corresponding to the step boundary is marked in the linkage timetable for use in calling.

7. The step transition bridging suppression method based on artificial intelligence according to claim 6, characterized in that, The steps of establishing a time mapping and calculating the start and end times of the roller micro-biasing based on the mechanism response delay and the preset roller micro-biasing duration, and calculating the start and end times of the short-term preheating based on the heating response delay and the preset short-term preheating duration, are performed using a unified time mapping formula, wherein the time mapping formula is specifically as follows: ； wherein is a start time of the action , is an end time of the action , is an action identifier, is a start time of a pre-step control time period, is an end time of a pre-step control time period, is a duration of the action , is a sampling interval, is a rounding function to round to the nearest integer, is an effective time corresponding to a step boundary, is a response delay of the action , is a max operation of two numbers, is a min operation of two numbers.

8. The step transition bridging suppression method based on artificial intelligence according to claim 1, wherein, S5 specifically refers to: Based on the linkage schedule, extract the start and end times of roller micro-lateral deviation, the start and end times of short-term preheating, and the effective time corresponding to the step boundary during the control period before the step for consistency check. Boundary and sequence checks are performed on the extracted times. Each start and end time is restricted to the control time period before the step. Each end time must be strictly greater than the corresponding start time. The time index must be monotonically increased according to the sampling interval. The difference between the two effective times is compared with a sampling interval. If the difference is not greater than a sampling interval, it is considered to pass. The verified linkage timetable is mapped with instructions. The start and end times of roller micro-lateral deviation are mapped to roller micro-lateral deviation start instruction and roller micro-lateral deviation end instruction, respectively. The start and end times of short-term preheating are mapped to short-term preheating start instruction and short-term preheating end instruction, respectively. The effective time mark corresponding to the step boundary is added to the trigger time of the two start instructions. The four instructions are sorted by time index and merged into a linkage execution instruction. The linkage execution instruction is triggered within the control time period before the step, and the effective time of the two start instructions is consistent with the effective time corresponding to the step boundary.

9. The step transition bridging suppression method based on artificial intelligence according to claim 1, wherein, S6 specifically refers to: Read the linkage execution instructions, parse the roller micro-lateral deviation start instruction, roller micro-lateral deviation end instruction, short-time preheating start instruction, and short-time preheating end instruction according to the time index, and inherit the effective time markers corresponding to the control time period boundary before the step and the step boundary; The four instructions are converted into trigger entries of the step transition control instructions, and corresponding trigger time points are generated. The effective time points of the two start instructions are required to be consistent with the effective time points of the step boundaries. All trigger time points are required to be located in the pre-step control time period, and are rounded according to the sampling interval to align the time index. The trigger entries are sequentially arranged, the start instructions are arranged before the corresponding end instructions according to the time index, the two instructions are triggered at the same time when the time points of the two instructions are the same, and the effective time points are kept consistent, and no new time point is inserted between the trigger time points. The step transition control instructions are triggered and executed according to the time index. The two start instructions are triggered in sequence before the step, and the two end instructions are triggered at the corresponding end time points. In the execution process, the boundary constraint of the pre-step control time period is followed, so that the roll micro-side deviation and the short-time preheating are simultaneously effective at the step boundary.

Images