Thermoplastic laying bandwidth stabilization method based on nerve guidance symbol regression and process three-parameter selector

Abstract

The invention provides a thermoplastic laying bandwidth stabilization method based on nerve guidance symbol regression and a process three-parameter selector, and the method comprises the steps: synchronously collecting power, speed, pressure, strip temperature, effective bandwidth and material identification, and calculating a compaction condition; constructing a process influence item set by using input alignment, segmented gating and a symbolic regression structure, explicitly calculating an edge outflow risk value, and identifying safety adjustment intervals and maximum adjustable amplitudes of power, speed and pressure; constructing a single-parameter adjustment path in a process three-parameter selector, performing path scoring on the risk reduction amount, the compaction influence amount and the response time delay, and generating a first adjustment parameter instruction; and through single-step adjustment and adjustment response data acquisition, according to risk change, an effective bandwidth difference value, response time delay and a compaction condition, carrying out bandwidth restability judgment, and outputting a sequence control instruction. The edge outflow risk of the strip can be reduced, the compaction quality is guaranteed, and the stability and the forming quality in the thermoplastic laying process are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic placement process control of thermoplastic composites, and particularly to a method for stabilizing the bandwidth of thermoplastic placement based on neural-guided symbolic regression and process triple parameter selector. Background Art

[0002] Due to characteristics such as remelting and short forming cycle, thermoplastic composites have received extensive attention in the fields of automatic placement and compression molding. To ensure the dimensional accuracy and interface quality of components, the industry has carried out a large number of studies on the prediction and regulation of placement deformation of plastic composites and the prediction of the compression molding performance of thermoplastic composites. By establishing material constitutive models, placement path geometric models, and temperature and pressure field models during the forming process, the placement deformation, residual stress, and compression molding performance are analyzed and optimized. However, most of these studies focus on offline simulation and process window design, and relatively less attention is paid to the online bandwidth stability and the regulation of edge outflow risk during the automatic placement process.

[0003] In the existing thermoplastic placement process control, the common practice is to determine process windows such as power, speed, and pressure based on experience or experiments, and perform manual or simple closed-loop adjustments on the basis of monitoring the strip temperature and forming quality; there are also those that use sensors to collect power, speed, pressure, temperature, and placement quality data, and construct regression or neural network models to predict the placement bandwidth, temperature field, or forming defects, which are used to guide parameter setting; some solutions attempt to combine online measurement results to perform amplitude limiting adjustment on a single parameter or perform linkage adjustment based on preset rules to reduce the probability of defects. These methods have improved the stability of the thermoplastic placement process to a certain extent, but mostly based on fixed process windows or black-box prediction models, lacking a systematic control strategy that is closely coupled with the temperature domain characteristics of the material and takes into account both bandwidth and compaction quality.

[0004] In the prior art, the bandwidth control and process parameter adjustment usually do not explicitly introduce the segmented temperature domain constraints of the material from vitrification to melting, making it difficult to accurately reflect the influence of temperature changes on the placement bandwidth and edge outflow behavior; the parameter recommendations based on experience or black-box models mostly stay at given recommended values or single corrections, lacking quantitative evaluation and closed-loop verification of the bandwidth changes and edge outflow risks during the linkage adjustment of power, speed, and pressure within the safe range; at the same time, there is generally a lack of interpretable risk indicators, clear safe adjustment intervals, and descriptions of the maximum adjustable amplitude, making it difficult to automatically generate sequential control instructions and single-step adjustment paths under the premise of ensuring compaction conditions, and it is difficult to meet the online bandwidth stabilization requirements under complex working conditions.

[0005] Therefore, a method for stabilizing the bandwidth of thermoplastic placement that can solve the above deficiencies 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 a method for stabilizing the bandwidth of thermoplastic layup based on neural-guided symbolic regression and a three-parameter process selector. The core technical problem to be solved is: in the automatic layup process of thermoplastic composite materials, how to construct a closed-loop control method with an interpretable edge outflow risk assessment and a safety adjustment mechanism for power, velocity, and pressure under the combined effect of segmented constraints from the glass transition to the melting temperature range and compaction conditions, so as to achieve online stabilization and sequential adaptive adjustment of the effective layup bandwidth.

[0007] A method for stabilizing thermoplastic layup bandwidth based on neural-guided symbolic regression and a three-parameter process selector according to an embodiment of the present invention includes:

[0008] S1. Acquire power, speed, pressure, strip temperature, effective bandwidth and material identification, generate working condition data and calculate compaction conditions;

[0009] S2. Based on the operating condition data and material identification, construct a set of process influence items and calculate the edge outflow risk value in neural-guided symbolic regression. The neural network provides the weight and monotonic direction of the influence items. Under the segmented constraint from the glass transition to the melting temperature range, identify the safe adjustment range and the maximum adjustable range of power, speed, and pressure, and obtain the edge outflow risk value, the safe adjustment range, and the maximum adjustable range.

[0010] S3. Under compaction conditions, based on the edge outflow risk value, the safety adjustment range, and the maximum adjustable range, a power adjustment path, a speed adjustment path, and a pressure adjustment path are formed. The risk reduction, impact on compaction, and response delay of each path within the allowable range are measured, and a first adjustment parameter instruction including the parameter to be adjusted, the adjustment direction, and the adjustment range is determined.

[0011] S4. Perform single-step adjustment according to the initial adjustment parameter instruction and in accordance with the safe adjustment range and the maximum adjustable range, and generate adjustment response data;

[0012] S5. Based on the adjusted response data, edge outflow risk value and compaction conditions, calculate the bandwidth stabilization judgment result. When the set threshold and compaction conditions are met, the bandwidth stabilization judgment result is formed.

[0013] S6. Based on the initial adjustment parameter command and the bandwidth stabilization judgment result, generate the sequence control command.

[0014] Optionally, S1 is as follows:

[0015] Power, speed, pressure, strip temperature, effective bandwidth, and material identification data are collected at the same time point to form a numerical sequence.

[0016] The original signal set is arranged in the order of power, speed, pressure, strip temperature and effective bandwidth to generate working condition data containing five values. The material identifier is converted into a numerical vector of dimension 10 and aligned with the working condition data in the input layer.

[0017] Based on the power, speed, pressure and strip temperature in the working condition data, the effective bandwidth is determined by setting a threshold, and the compaction conditions are calculated as the upper and lower boundaries of the power, speed and pressure ranges and the upper and lower boundaries of the strip temperature in the glass transition to melting temperature range.

[0018] The working condition data and material identifiers are used to calculate the set of process influence items in neural-guided symbolic regression. The values ​​and trends of strip temperature are extracted and aligned with the effective bandwidth to ensure that the compaction conditions are consistent with the physical constraints for identifying edge outflow risk values.

[0019] Optionally, S2 is as follows:

[0020] The operating condition data and material identifiers are aligned in the input alignment layer to form a 15-dimensional input vector, which is then mapped to the vector positions in the order of power, speed, pressure, strip temperature and effective bandwidth, and used as the input for the influence term generation layer.

[0021] The set of process influence items is calculated on the 15-dimensional input vector at the influence item generation layer. It is generated item by item according to the combination of power-related items, speed-related items, pressure-related items, temperature-related items, bandwidth-related items and material identification-related items, and is aligned with the operating condition data and material identification.

[0022] Based on the material identification and strip temperature, segmented gating signals from the glass transition to the melting temperature range are calculated in the segmented gating layer to obtain gating coefficients for the low-temperature and high-temperature segments, which are then used to segmentally enable and suppress the set of process influence items.

[0023] The output of the influence term generation layer and the gating coefficient of the segmented gating layer are input into the symbol coefficient layer to generate the influence term weights and monotonic directions. The monotonic directions correspond to power, velocity and pressure, respectively, and are used to limit the direction of parameter changes.

[0024] The process impact items set and their weights are explicitly expressed in the symbolic regression head assembly, constrained by the gating coefficient of the segmented gating layer, to calculate the edge outflow risk value and keep it aligned with the operating condition data;

[0025] Based on the monotonic direction and the gating coefficient of the segmented gating layer, the calculation head performs a unidirectional scan of power, speed and pressure in the safe adjustment range, and outputs the upper and lower boundaries of the three segments to form the safe adjustment range;

[0026] Based on the changing trends of strip temperature and effective bandwidth, and combined with the length of the safe adjustment range, the maximum adjustable range of head generation power, speed and pressure is calculated within the maximum adjustable range to obtain the edge outflow risk value, the safe adjustment range and the maximum adjustable range.

[0027] Optionally, the step of explicitly expressing and calculating the marginal outflow risk value using the set of process influence items and their weights in a symbolic regression head assembly includes calculating the marginal outflow risk value using a risk formula:

[0028] ;

[0029] in, This represents the risk value for edge outflow, a single value aligned with the operating condition. For the index of affected items, take numbers one to twenty-four. The weight of the influencing item A number, and Alignment The first of the set of process influence items A number, by generate, The low-temperature gating coefficient is in the first... The values ​​of each influencing item The high-temperature section gate coefficient is in the th... The values ​​of each influencing item This is the segmentation factor, with a value between zero and one, determined by the strip temperature. Relative to the glass transition temperature threshold With melting temperature threshold Generated using linear interpolation, and truncated when exceeding a threshold range to maintain a range of zero to one. This represents the glass transition temperature threshold, which is the value corresponding to the material identifier. Here, represents the melting temperature threshold, and represents the value corresponding to the material identifier. All of these variables are obtained directly in numerical form within the model and are consistent with... and Keep them aligned.

[0030] Optionally, S3 specifically refers to:

[0031] Input the compaction conditions, edge outflow risk value, safe adjustment range, maximum adjustable range and operating condition data into the process three-parameter selector to determine the path starting point as the power, speed and pressure in the operating condition data;

[0032] Within the safe adjustment range, with the maximum adjustable range as the step size, power adjustment path, speed adjustment path and pressure adjustment path are constructed sequentially in the monotonic direction. Ten candidate points are selected for each path to form path data.

[0033] For each path, the risk reduction is calculated based on the operating status data and the edge outflow risk value. The difference between the edge outflow risk values ​​at the beginning and end of the path is taken as the risk reduction.

[0034] For each path, the impact on compaction is calculated based on the compaction conditions and the working condition data at the end of the path. A set threshold is used to determine the offset and generate the impact value.

[0035] The time it takes for a parameter adjustment to be transmitted to the effective bandwidth is taken as the response delay. The response delay is calculated for each path based on the relationship between the operating status data and the change in the effective bandwidth.

[0036] The risk reduction, the impact on compaction, and the response delay are aligned to form a path calculation vector with three dimensions. This vector is then integrated with the upper and lower boundaries of the safety adjustment range and the maximum adjustable range to form a path constraint vector with nine dimensions. The path score is then calculated.

[0037] The path with the highest score is selected based on the path score value. The parameters to be adjusted, the adjustment direction, and the adjustment range are output to form the initial adjustment parameter instruction. Constraint checks are performed based on the safe adjustment range and the maximum adjustable range. Instructions that do not meet the constraints are truncated in range and corrected in direction to obtain compliant initial adjustment parameter instructions.

[0038] Optionally, in the step of aligning the risk reduction, the impact on compaction, and the response delay into a path calculation vector with three dimensions, and merging it with the upper and lower boundaries of the safety adjustment range and the maximum adjustable range to form a path constraint vector with nine dimensions, the path score value is calculated using a path score formula:

[0039] ;

[0040] in, The path rating is a numerical value, which is related to the path index. Aligned single numerical values, The constraint coefficient takes a value between zero and one. The risk reduction amount is weighted, and the value is dimensionless. The weight of the influence on compaction is a dimensionless value. For response delay weighting, the unit is the reciprocal of time. The risk reduction is the difference between the marginal outflow risk values ​​at the start and end points of the path. The influence of this on compaction is denoted as , and the three offsets are denoted as . The dimensionless value obtained by the largest of the ratios is... To address latency, in order to ensure that the effective bandwidth first exceeds The difference between the detection time and the parameter adjustment time, for The calculation is performed according to the following process: The parameter belonging to the current path is denoted as... When the endpoint of the path crosses the boundary, Set to zero. When the boundary is not exceeded, calculate the remaining distance from the endpoint to the nearest boundary and divide it by the corresponding interval length to obtain the interval remaining ratio, denoted as . Calculate the complement of the ratio of the cumulative step size of the endpoint relative to the starting point to the maximum adjustable range of this parameter, and obtain the remaining step size ratio, denoted as . ,Will and Both are truncated to the range of zero to one, and the smaller value is taken as... .

[0041] Optionally, S4 specifically refers to:

[0042] Align the initial adjustment parameter command with the safe adjustment range and the maximum adjustable range. Determine compliance based on the upper and lower boundaries and the upper limit of the step size. If non-compliant, truncate according to the upper and lower boundaries and correct according to the monotonic direction.

[0043] Under compaction conditions, a single-step adjustment is adopted, in which the parameter to be adjusted is changed once according to the corrected adjustment range, while keeping the two unselected parameters unchanged.

[0044] Immediately after adjustment, power, speed, pressure, strip temperature and effective bandwidth are collected to form adjusted operating condition data, and the data is aligned with the adjusted operating condition data to establish a correspondence between the before and after.

[0045] The difference in effective bandwidth and response latency are calculated based on the correspondence between the two values. The adjusted adjustment range, the difference in effective bandwidth, and the response latency are combined to generate adjusted response data, which is then aligned with the edge outflow risk value.

[0046] Optional, S5 specifically includes:

[0047] Align the adjusted response data, edge outflow risk value, and compaction conditions, organize them into inputs for judgment according to the time sequence of the adjusted working condition status data, and keep them aligned with the working condition status data.

[0048] The corrected adjustment range in the adjustment response data is verified based on the safe adjustment range and the maximum adjustable range, and calculation is only performed when the upper and lower boundaries are not exceeded and the step size is not exceeded.

[0049] Based on the difference in effective bandwidth and response latency in the adjusted response data, the risk reduction is calculated in conjunction with the change in edge outflow risk value. The difference in effective bandwidth is compared with the set threshold, and the response latency is also compared with the set threshold.

[0050] Based on the compaction conditions, verify whether the power, speed and pressure in the adjusted working condition data are within the upper and lower boundaries of the compaction conditions. If they are not within the range, it is determined that the system will not stabilize.

[0051] A bandwidth stabilization judgment result is generated when the risk reduction amount meets the set threshold, the difference in effective bandwidth meets the set threshold, and the compaction condition is met; no bandwidth stabilization judgment result is generated if any of the conditions are not met.

[0052] Optional, S6 specifically includes:

[0053] Align the initial adjustment parameter command with the bandwidth stabilization judgment result, and verify compliance based on the upper and lower boundaries of the safe adjustment range and the upper limit of the maximum adjustable step size to obtain an executable instruction set.

[0054] When the bandwidth stabilization judgment result meets the set threshold and compaction conditions, the sequence control command is set to stop adjustment, and the power, speed and pressure are kept constant.

[0055] When the bandwidth stabilization judgment result has not been formed, the sequence control command is set to follow the monotonic direction of the first adjustment parameter command, and advance to the next candidate point within the safe adjustment range with a step size not exceeding the maximum adjustable range;

[0056] The activation temperature range of the sequential control command is segmented and limited according to the gating coefficient of the segmented gating layer, so that the sequential control is aligned with the segmented constraint of the glass transition to melting temperature range.

[0057] The beneficial effects of this invention are:

[0058] 1. This proposal presents an improved edge outflow risk assessment method based on neural-guided symbolic regression. Through an input alignment layer, an influence item generation layer, and a segmented gating layer, power, velocity, pressure, strip temperature, effective bandwidth, and material identifiers are uniformly mapped to a set of process influence items. The symbolic coefficient layer generates the influence item weights and their monotonic directions relative to power / velocity / pressure. Then, the symbolic regression head explicitly calculates the edge outflow risk value, safe adjustment range, and maximum adjustable range under segmented constraints from the glass transition to the melting temperature range. Compared to existing methods relying on empirical windows or black-box regression models, this proposal introduces segmented material temperature range gating and interpretable symbolic representation into the risk model, tightly coupling compaction conditions, temperature ranges, and edge outflow risk. This enables quantitative identification of the safe adjustment boundaries and maximum adjustment step size for the three parameters of power, velocity, and pressure under the premise of satisfying compaction conditions and temperature range constraints, thus providing a physically consistent and traceable decision-making basis for subsequent bandwidth stabilization control.

[0059] 2. This proposal presents a novel method for selecting and scoring the three process parameters and their adjustment paths. Utilizing a three-parameter selector, it constructs single-parameter adjustment paths for power, speed, and pressure within identified safe adjustment ranges and maximum adjustable amplitudes. Using edge outflow risk as the core indicator, and combining its impact on compaction and bandwidth response delay, it forms path calculation and constraint vectors. A path scoring formula is used to weight and constrain the risk reduction, compaction deviation, and response characteristics of different paths, automatically selecting the parameters to be adjusted, the adjustment direction, and the adjustment amplitude, thus generating the initial adjustment parameter command. Compared to existing schemes that adjust based solely on single-parameter limits or fixed rules, this proposal comprehensively considers bandwidth risk reduction, compaction safety, and dynamic response characteristics at the path level. This makes the trade-off process between the three parameters quantifiable and comparable, prioritizing the adjustment path that is more conducive to bandwidth stabilization without compromising compaction conditions, thereby improving the targeting and stability of online bandwidth control under multi-parameter coupled conditions.

[0060] 3. This proposal presents a bandwidth stabilization method for automated layup of thermoplastic composites. It integrates the aforementioned risk modeling and three-parameter selection mechanism into a single-step adjustment and sequential control chain. Through initial parameter adjustment commands, single-step adjustment execution, adjustment response data acquisition, bandwidth stabilization judgment, and sequential control command generation, a closed-loop control path is constructed, from risk assessment to parameter correction and stabilization determination. Under the constraint of segmented gating from the glass transition to the melting temperature range, the activation temperature range and adjustment range of control commands are limited, enabling step-by-step management of power, speed, and pressure, as well as stopping conditions. Unlike traditional methods that only provide recommended parameters or one-time correction suggestions, this proposal aligns and manages operating condition data, risk values, adjustment response data, and sequential control commands along the timeline. This ensures that bandwidth stabilization judgment is based on underlying numerical changes and compaction condition verification after single-step adjustment, thereby achieving online stabilization control of the effective bandwidth in actual layup scenarios, balancing edge outflow risk suppression with compaction quality maintenance. Attached Figure Description

[0061] 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:

[0062] Figure 1 The flowchart shows a method for stabilizing thermoplastic layup bandwidth based on neural-guided symbolic regression and a process three-parameter selector proposed in this invention.

[0063] Figure 2 This invention presents a flowchart of a process for obtaining working condition data and calculating compaction conditions based on neural-guided symbolic regression and a process three-parameter selector.

[0064] Figure 3 This is a flowchart of the calculation of the neural-guided symbolic regression, safety adjustment range, and maximum adjustable range for a thermoplastic layup bandwidth stabilization method based on neural-guided symbolic regression and a process three-parameter selector proposed in this invention.

[0065] Figure 4 This invention presents a flowchart of the process three-parameter selector and initial parameter adjustment command forming a thermoplastic layup bandwidth stabilization method based on neural-guided symbolic regression and process three-parameter selector.

[0066] Figure 5 This is a flowchart of the single-step adjustment and adjustment response data generation process of a thermoplastic layup bandwidth stabilization method based on neural-guided symbolic regression and process three-parameter selector proposed in this invention.

[0067] Figure 6 This is a flowchart of the bandwidth stabilization judgment process for a thermoplastic layup bandwidth stabilization method based on neural-guided symbolic regression and a process three-parameter selector proposed in this invention.

[0068] Figure 7 This is a schematic diagram of a neural-symbolic hybrid multi-head network structure for a thermoplastic layup bandwidth stabilization method based on neural-guided symbolic regression and a process three-parameter selector proposed in this invention.

[0069] Figure 8 This is a top-view schematic diagram illustrating the process from instability to stabilization of thermoplastic layup bandwidth, which is a method for stabilizing thermoplastic layup bandwidth based on neural-guided symbolic regression and a three-parameter process selector proposed in this invention. Detailed Implementation

[0070] 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.

[0071] refer to Figures 1 to 8 A method for stabilizing thermoplastic layup bandwidth based on neural-guided symbolic regression and a three-parameter process selector, characterized by comprising:

[0072] S1. Acquire power, speed, pressure, strip temperature, effective bandwidth and material identification, generate working condition data and calculate compaction conditions;

[0073] S2. Based on the operating condition data and material identification, construct a set of process influence items and calculate the edge outflow risk value in neural-guided symbolic regression. The neural network provides the weight and monotonic direction of the influence items. Under the segmented constraint from the glass transition to the melting temperature range, identify the safe adjustment range and the maximum adjustable range of power, speed, and pressure, and obtain the edge outflow risk value, the safe adjustment range, and the maximum adjustable range.

[0074] S3. Under compaction conditions, based on the edge outflow risk value, the safety adjustment range, and the maximum adjustable range, a power adjustment path, a speed adjustment path, and a pressure adjustment path are formed. The risk reduction, impact on compaction, and response delay of each path within the allowable range are measured, and a first adjustment parameter instruction including the parameter to be adjusted, the adjustment direction, and the adjustment range is determined.

[0075] S4. Perform single-step adjustment according to the initial adjustment parameter instruction and in accordance with the safe adjustment range and the maximum adjustable range, and generate adjustment response data;

[0076] S5. Based on the adjusted response data, edge outflow risk value and compaction conditions, calculate the bandwidth stabilization judgment result. When the set threshold and compaction conditions are met, the bandwidth stabilization judgment result is formed.

[0077] S6. Based on the initial adjustment parameter command and the bandwidth stabilization judgment result, generate the sequence control command.

[0078] In this embodiment, step S1 specifically includes:

[0079] In the data collection at the same time point, the power is recorded as... , denoted as Record the pressure as The strip temperature is recorded as The effective bandwidth is denoted as The material identification vector is denoted as ,in For a numerical vector of dimension 10, the sampling time is denoted as... At the laying station , , , , and Synchronously acquire data to form a time-ordered numerical sequence for input, and record the operating condition data as follows: Let the input vector be denoted as After acquisition, the original signal set is arranged in a fixed order of power, speed, pressure, strip temperature, and effective bandwidth, generating a dataset containing five values. and will Keep it as a ten-dimensional numerical vector, and Position mapping is performed at the input alignment layer to form a fifteen-dimensional structure. ,make Position order and and Alignment provides aligned input for the computation of the influence term generation layer;

[0080] In the calculation of compaction conditions, the compaction conditions are denoted as... The threshold for determining effective bandwidth is denoted as... The glass transition temperature threshold is denoted as The melting temperature threshold is denoted as Record the equipment parameter library as , Based on the material identification and equipment model index, the upper and lower boundaries of the power range are denoted as follows: and The upper and lower boundaries of the velocity range are denoted as and The upper and lower boundaries of the pressure range are denoted as and The aforementioned boundaries are provided in numerical form and correspond to the workstations. The effective bandwidth determination threshold library is denoted as... , Output by material identification index The material temperature range library is denoted as , Output by material identification index and ,in accordance with In , , and The value is adopted. right Make a judgment and compare the judgment result with... In , , , , , Alignment, defining the upper and lower boundaries of the power, speed, and pressure ranges respectively, based on Output and right Limit the temperature range so that Located within the glass transition to melting temperature range, the upper and lower boundaries of the above range are summarized as follows: ,in This includes the upper and lower boundaries of the power range, the upper and lower boundaries of the speed range, the upper and lower boundaries of the pressure range, and the upper and lower boundaries of the strip temperature within the glass transition to melting temperature range, and is related to... The order of their positions must remain consistent;

[0081] To align with the calculation of the process influence term set in neural-guided symbolic regression, and The input influence term generation layer generates combined terms of power-related terms, velocity-related terms, pressure-related terms, temperature-related terms, bandwidth-related terms, and material identification-related terms. The segmented gating vector of the segmented gating layer is denoted as... ,Depend on and Generate in segmented gating layer This is used to segment and activate the set of process influence items within the glass transition to melting temperature range, with the sampling period denoted as... The most recent sampling time is recorded as ,in and The interval is the sampling period. , in order to and The temperature difference of the collected strip is denoted as , in order to and The effective bandwidth difference of the acquisition is denoted as and with and Alignment, forming the numerical values ​​and trends of strip temperature and effective bandwidth, will , , and and After alignment, it serves as an auxiliary input to the sign coefficient layer and the sign regression head, enabling the identification of marginal outflow risk values. The upper and lower boundaries of the range remain consistent under segmented constraints, ensuring that the identification of the safe adjustment range and the maximum adjustable range is based on physically consistent aligned data. Through the above implementation, data collected at the same time point are... , , , , and Forming alignment and and with , , Clearly define the compaction conditions At the same time , and By introducing the segmented temperature range constraint and the dynamic correlation between strip temperature and effective bandwidth into the input of the influence term generation layer and the symbolic regression head, the identification of compaction conditions and edge outflow risk values ​​is kept aligned and consistent within the model structure.

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

[0083] Record the operating condition data as , which includes power ,speed ,pressure strip temperature With effective bandwidth The five values ​​are used to denote the material identification vector as follows: ,in Let the output vector of the input alignment layer be a numerical vector of dimension 10. In the input alignment layer and Perform position mapping to form a fifteen-dimensional representation. , and according to , , , , The fixed order and The ten-dimensional positional mapping is maintained, ensuring that the reading order of each layer is consistent with... and Align, As input to the impact item generation layer, the calculation process for the process impact item set is initiated;

[0084] Let the set of process influence items be denoted as In the generation layer of influence terms, the fifteen-dimensional Expanding item by item, generating items based on combinations of power-related items, speed-related items, pressure-related items, temperature-related items, bandwidth-related items, and material identification-related items, so that... The dimension is twenty-four, and each item is compared with... Establish a one-to-one correspondence between the positions and maintain Alignment attributes with operating condition data and material identifiers ensure that gating and coefficient assembly are enabled and suppressed in a unified location. The gating vector of the segmented gating layer is denoted as... The low-temperature gating coefficient is denoted as The high-temperature section gating coefficient is denoted as In the segmented gating layer according to and Calculate the segmented gating signal for the glass transition to melting temperature range, so that... Simultaneously includes and And based on the temperature range Each item is enabled and suppressed in stages;

[0085] The influence term weight of the symbol coefficient layer is denoted as... ,in Let the weight vector be of dimension 24, and let the monotonic direction be denoted as... ,in Let the direction vectors be three-dimensional, corresponding to the directions of power, velocity, and pressure, respectively. In the sign coefficient layer... Output and The gating coefficients are used as input to generate the symbolic representation of the assembly. With the use of safety adjustment and keep Align to The twenty-four items, maintain Align to , , The three parameters in the symbolic regression head and Explicit assembly is performed, using a fixed combination of addition and multiplication operations, and is subject to... The piecewise gating coefficient constraint is used to calculate the edge outflow risk value, and the risk value is output as... Its calculation is based on the risk formula:

[0086] ;

[0087] in, This represents the risk value for edge outflow, a single value aligned with the operating condition. For the index of affected items, take numbers one to twenty-four. The weight of the influencing item A number, and Alignment The first of the set of process influence items A number, by generate, The low-temperature gating coefficient is in the first... The values ​​of each influencing item The high-temperature section gate coefficient is in the th... The values ​​of each influencing item This is the segmentation factor, with a value between zero and one, determined by the strip temperature. Relative to the glass transition temperature threshold With melting temperature threshold Generated using linear interpolation, and truncated when exceeding a threshold range to maintain a range of zero to one. This represents the glass transition temperature threshold, which is the value corresponding to the material identifier. Here, represents the melting temperature threshold, and represents the value corresponding to the material identifier. All of these variables are obtained directly in numerical form within the model and are consistent with... and Keep aligned;

[0088] Let the boundary vector of the safety adjustment interval be denoted as ,in There are six boundary values, each corresponding to... , , The upper and lower boundaries are calculated within the safe adjustment range based on the monotonic direction. Piecewise gated vectors Perform a unidirectional scan, with the current In , , Starting from, according to The specified direction determines the upper and lower boundaries of each parameter within the gated temperature range, and the output... and Align the positions in the correct order, and denote the maximum adjustable amplitude vector as... ,in There are three values, each corresponding to... , , The maximum adjustable range is denoted as the strip temperature change trend. The trend of change in effective bandwidth is denoted as Based on sampling time Compared with the most recent sampling time The fixed sampling period is denoted as Acquired at adjacent sampling times and and and Alignment, the effective bandwidth change threshold is denoted as The temperature change threshold is denoted as The calculation of the maximum adjustable range is based on... and The numerical values ​​and trends, combined with the safety adjustment range Length of To generate, so that in Exceed Or Exceed Under these circumstances, the amplitude of the corresponding parameter converges. and If it is within the threshold The remaining length is limited to the upper limit, forming a temperature range segment that is consistent with the monotonic direction. ;

[0089] In the above implementation process, the input alignment layer is completed. and The fifteen-dimensional mapping is formed The generation layer of the influence item forms a 24-dimensional structure. The segmented gating layer generates a number of components. and of Symbol coefficient layer generation and Aligned Aligned with parameters The symbolic regression head outputs the above formula and... Alignment risk value The safety adjustment interval calculation head outputs six boundary values. The maximum adjustable range calculation head is in , and The output consists of three values ​​under the constraints. ,Will , and Maintaining alignment between the operating condition data and material identifiers within the model is crucial for forming power, speed, and pressure adjustment paths. This alignment is also used to generate initial adjustment parameter commands under compaction conditions, ensuring that edge outflow risk values ​​and parameter adjustment rules are linked and consistent within the same model. Through these implementations, the operating condition data and material identifiers are aligned at the input alignment layer. The output is generated by the coupling between the influence term generation layer and the segmented gating layer. and The temperature range segmentation is consistent with that formed by the symbol coefficient layer and the symbol regression head. And combine the calculation head of the safe adjustment range with the calculation head of the maximum adjustable range. , , and generate Ultimately, the risk value of marginal outflow was obtained. Safety adjustment range With maximum adjustable range It provides aligned and executable inputs for the process three-parameter selector.

[0090] In this embodiment, step S3 specifically includes:

[0091] Record the operating condition data as ,in Includes power ,speed ,pressure strip temperature With effective bandwidth The five values ​​are used to denot the compaction conditions as follows: ,in Give the upper and lower boundaries of the power, speed, and pressure ranges, as well as the upper and lower boundaries of the strip temperature within the glass transition to melting temperature range. Denote the edge outflow risk value as... For a single value aligned with the current operating condition, the safety adjustment range is denoted as . ,in There are six boundary values, corresponding to the upper and lower boundaries of the power, respectively. , upper and lower boundaries of velocity , Upper and lower pressure boundaries , The maximum adjustable range is denoted as ,in There are three values, each corresponding to the upper limit of the power step size. Upper limit of speed step size Upper limit of pressure step size Let the monotonic direction be denoted as ,in The three directional markers correspond to the directions of power, speed, and pressure, respectively, and are used in the process three-parameter selector. , , , and Input is the command to adjust the initial parameters;

[0092] In determining the path start point, the power value in the current operating condition data is recorded as... The speed value is denoted as The pressure value is recorded as ,by , , As the starting point for the power adjustment path, speed adjustment path, and pressure adjustment path, the path index is denoted as... ,in Use numbers from 0 to 9 to mark ten candidate points for each path, and then select them within the selector. Verify the upper and lower boundaries to ensure the starting point is within the safe adjustment range, and in accordance with... The direction markings should remain consistent;

[0093] In path construction, the maximum adjustable range is used. As a step size, within the safe adjustment range Inner monotonic direction Proceeding sequentially, the power adjustment path is denoted as... In each advancement, at no more than Candidate points are generated along the power direction using a step size, and the step size is truncated when the limit is exceeded to maintain the position. to Within this range, a total of ten candidate points are identified, and the speed adjustment path is recorded as follows: The pressure adjustment path is denoted as , according to The same rules apply to their respective intervals. to , to Ten candidate points are generated internally. The candidate point sequences of the three paths are used as path data and aligned to the path index. ;

[0094] In the calculation of risk reduction, the edge outflow risk value at the starting point of each path is denoted as... The edge outflow risk value at the end of the path is denoted as The amount of risk reduction is recorded as ,right and Find the difference and path index Align the endpoint positions and calculate the corresponding values ​​on each of the three paths. ;

[0095] In calculating the impact of compaction, the working condition data at the end of the path and the compaction conditions are compared. Alignment is performed by calculating the offset between the endpoint values ​​of power, speed, and pressure and the upper and lower boundaries of their corresponding ranges. The power offset is denoted as... When the endpoint is less than the lower boundary of the range, the result is obtained by subtracting the endpoint from the lower boundary of the range. When the endpoint is greater than the upper boundary of the range, the result is obtained by subtracting the upper boundary of the range from the endpoint. When within range Set to zero, and set the velocity offset. With pressure offset Obtained in the same way, the set threshold is denoted as... The amount of impact on compaction is denoted as The three offsets are respectively compared with The one with the largest ratio is taken as ,exist Paths exceeding one are marked as requiring penalty. A path not exceeding one time is marked as a compliant path and aligned with the path endpoint;

[0096] In response delay calculation, the parameter adjustment time is denoted as... The effective bandwidth change detection time is denoted as The effective bandwidth change threshold is denoted as After a single parameter adjustment of the path endpoint, the change in effective bandwidth is monitored, and the bandwidth is first exceeded. Time Record ,by Receive response delay And align with the end point of the path;

[0097] In the fusion of path vectors and constraints, the path calculation vector is denoted as... ,in Includes risk reduction The impact on compaction With response latency The three values ​​are used to denote the path constraint vector as follows: ,in There are nine values, including The six upper and lower boundaries and The three upper limits of the step size are used in the scoring module to generate a path score value by weighting the risk priority, the impact on compaction, and the response time delay as penalties. A constraint coefficient is introduced to suppress paths that are close to the boundary or have insufficient step size. The path score value is denoted as... Its calculation uses the path scoring formula:

[0098] ;

[0099] in, The path rating is a numerical value, which is related to the path index. Aligned single numerical values, The constraint coefficient takes a value between zero and one. The risk reduction amount is weighted, and the value is dimensionless. The weight of the influence on compaction is a dimensionless value. For response delay weighting, the unit is the reciprocal of time. The risk reduction is the difference between the marginal outflow risk values ​​at the start and end points of the path. The influence of this on compaction is denoted as , and the three offsets are denoted as . The dimensionless value obtained by the largest of the ratios is... To address latency, in order to ensure that the effective bandwidth first exceeds The difference between the detection time and the parameter adjustment time, for The calculation is performed according to the following process: The parameter belonging to the current path is denoted as... When the endpoint of the path crosses the boundary, Set to zero. When the boundary is not exceeded, calculate the remaining distance from the endpoint to the nearest boundary and divide it by the corresponding interval length to obtain the interval remaining ratio, denoted as . Calculate the complement of the ratio of the cumulative step size of the endpoint relative to the starting point to the maximum adjustable range of this parameter, and obtain the remaining step size ratio, denoted as . ,Will and Both are truncated to the range of zero to one, and the smaller value is taken as... ;

[0100] In the path selection and instruction generation process, the scores of the three paths are compared, and the path with the highest score is selected. The parameters to be adjusted, the adjustment direction, and the adjustment range are then output to form the initial adjustment parameter instruction, denoted as... During constraint checks, the safe adjustment range is used. Upper and lower boundaries and maximum adjustable range upper limit of step size Verification is performed, and the amplitude is truncated when it exceeds the upper and lower boundaries, in the monotonic direction. If there is inconsistency, directional correction is performed to obtain compliant initial adjustment parameter instructions. Operating condition data Alignment, entering the single-step adjustment process and used to generate adjustment response data;

[0101] In the above implementation process, the process three-parameter selector consists of a path construction module, a calculation module, a scoring module, and a constraint checking module. The path construction module adopts... , and The data for three single-parameter paths is generated, and the calculation module uses... , Calculate path-by-path using path data as input. , and Output path calculation vector The scoring module is based on the above formula. and Generate rating values ​​and maintain them with the path index. Alignment and constraint check module based on and right Verification and correction are performed to ensure that the initial adjustment parameter command and the safe adjustment range are consistent with the maximum adjustable range. The linearly progressive calculation link ensures that the parameter to be adjusted, the adjustment direction and the adjustment range are generated on the carrier of the single parameter path, and are consistent with the segmented constraints and monotonic direction in the scoring and constraint checks.

[0102] In this embodiment, step S4 specifically includes:

[0103] Record the operating condition data as ,in Includes power ,speed ,pressure strip temperature With effective bandwidth The five values ​​are used to denot the compaction conditions as follows: ,in Give the upper and lower boundaries of the power, speed, and pressure ranges, as well as the upper and lower boundaries of the strip temperature within the glass transition to melting temperature range. The safe adjustment range is denoted as... ,in There are six boundary values, corresponding to the upper and lower boundaries of the power, respectively. , upper and lower boundaries of velocity , Upper and lower pressure boundaries , The maximum adjustable range is denoted as ,in There are three upper limit values ​​for the step size, each corresponding to the upper limit of the power step size. Upper limit of speed step size Upper limit of pressure step size Let the monotonic direction be denoted as ,in The three directional markers correspond to the directions of power, velocity, and pressure, respectively. The gating vector of the segmented gating layer is denoted as... ,in The activation and suppression are segmented and marked within the glass transition to melting temperature range, and the initial parameter adjustment command is recorded as follows: This includes the parameter to be adjusted, the adjustment direction, and the adjustment range. The parameter adjustment time is recorded as... For the time stamp of single-step adjustment, the effective bandwidth change threshold is denoted as... This is used to trigger the detection of response latency;

[0104] In compliance assessment and correction, and and Alignment, read the parameter marker to be adjusted, and record the adjustment range of the parameter as... ,in Only , , Take the value from the middle, for With the corresponding step size upper limit The values ​​are compared with those in the range. If the value does not exceed the upper limit of the step size, it remains unchanged; if it exceeds the upper limit, it is truncated. The target value after execution is compared with the corresponding range. Verify the upper and lower boundaries, and if the boundary is crossed, proceed along the monotonic direction. Reverse correction and truncation by the nearest boundary, the gated vector corresponding to the current temperature range. When not enabled, the correction result will be marked as non-compliant and no adjustment will be performed. The adjusted amount will be recorded as... Record the compliance mark as Set it to the enabled flag when compliant, and set it to the disabled flag when non-compliant;

[0105] In single-step adjustment execution, To enable marking and meet compaction conditions At that time, the parameter to be adjusted is adjusted step by step. according to Perform one change while keeping the other two parameters unchanged, if the compaction conditions are not met or To prevent changes from being implemented and to ensure alignment between compaction and temperature range segmentation constraints when the marker is disabled, the parameter adjustment time is recorded at the end of the execution. This serves as the starting point for response latency;

[0106] In the adjusted data acquisition, power is acquired immediately after the execution is completed. ,speed ,pressure strip temperature With effective bandwidth The adjusted operating condition data is recorded as follows: , and according to , , , , The order and The alignment establishes a correspondence between the before and after alignment, maintaining consistent position mapping during the alignment process. The effective bandwidth before adjustment is denoted as... The adjusted effective bandwidth is denoted as Used for difference calculation and time delay detection, maintaining the strip temperature and material identification and the gating vector of the segmented gating layer within the same sampling period. Alignment is performed to ensure sampling consistency between the low-temperature and high-temperature ranges;

[0107] In response quantity generation, based on the correspondence between the preceding and following data... relatively The difference is taken as the difference in effective bandwidth and is denoted as... and the parameter to be adjusted is marked Alignment, record the corresponding adjustment range after correction. In response latency calculation, changes in effective bandwidth are monitored, and the threshold for effective bandwidth change is first exceeded. The detection time is recorded with a time tag, which is then recorded as... ,by relatively The time difference is taken as the response delay, and is denoted as . And align it with the time stamp of the adjusted operating condition data, and record the adjustment response data as... ,in Including the revised adjustment range The difference in effective bandwidth With response latency The three values, along with a flag indicating the parameter to be adjusted. With the sampling time stamp, within the same sampling period The risk value of marginal outflow is denoted as The updated value alignment is used for bandwidth stabilization judgment and the formation of sequential control instructions. Through the above implementation, the initial adjustment parameter instructions, the safety adjustment range and the maximum adjustable range are executed in compliance with segmented constraints. The difference between the adjusted operating status data and the effective bandwidth and the response delay are formed on the same time axis to form adjustment response data and are aligned with the edge outflow risk value, supporting the linear progressive link of judgment and control.

[0108] In this embodiment, step S5 specifically includes:

[0109] Record the operating condition data as ,in Includes power ,speed ,pressure strip temperature With effective bandwidth The adjusted operating condition data will be recorded as the value. ,and Align the location and time, and record the compaction conditions as follows: ,in Give the upper and lower boundaries of the power, speed, and pressure ranges, as well as the upper and lower boundaries of the strip temperature within the glass transition to melting temperature range. Denote the edge outflow risk value as... For a single numerical sequence aligned with the operating conditions, the safety adjustment range is denoted as... ,in There are six boundary values, corresponding to the upper and lower boundaries of the power, respectively. , upper and lower boundaries of velocity , Upper and lower pressure boundaries , The maximum adjustable range is denoted as ,in There are three upper limit values ​​for the step size, each corresponding to the upper limit of the power step size. Upper limit of speed step size Upper limit of pressure step size The adjusted response data will be recorded as ,in Including the revised adjustment range The difference in effective bandwidth With response latency The value, along with a marker for the parameter to be adjusted. ,in exist , , The value is taken from the middle, and the risk threshold is recorded as . The bandwidth difference threshold is denoted as The response delay threshold is denoted as The three are the set thresholds based on the material identification index;

[0110] In the input organization used for judgment, , and according to Align and sort the time tags to maintain consistency with... The positional order is consistent, and the effective bandwidth before adjustment is recorded as... The adjusted effective bandwidth is recorded as The risk value before adjustment is recorded as The adjusted risk value is recorded as Establish a correspondence between the samples within the same sampling period for verification and comparison;

[0111] During compliance verification, read and The corresponding interval boundary is denoted as and The corresponding upper limit of step size is denoted as The target value after execution and and The scope is verified; if it is outside the scope, it is not included in the calculation. and The comparison is performed, and samples exceeding the step size limit are not included in the calculation. Compliant samples are retained for risk and bandwidth assessment.

[0112] In the comparison of risk, latency, and bandwidth, the decrease in risk is denoted as... ,by and The difference is obtained ,by and Comparison, in When the bandwidth is less than the threshold, it is marked as insufficient. and Comparison, in If the response is greater than a threshold, it is marked as slow. and Comparison, in If the values ​​are less than the threshold, it is marked as insufficient risk reduction. The results of the above three comparisons are kept consistent with the time label and used for the final judgment.

[0113] During the compaction condition verification, read The power, speed, and pressure endpoint values ​​are respectively compared with... Verify the upper and lower boundaries of the corresponding range, and record the verification mark as follows. When any parameter is outside the range, Set as not satisfied, and set as not satisfied when all are within the range. To meet the requirements, maintain consistency with the verification process. Position and time are aligned;

[0114] In the bandwidth stabilization assessment, the bandwidth stabilization assessment result is recorded as... Simultaneously satisfying the verification of risk reduction amount, effective bandwidth difference, and response latency, and To meet the requirements Set as stable, and if any condition is not met, Set as non-stabilizing, and and The time-stamped output is used for sequential control and parameter updates. When the system fails to stabilize, the current sample is retained, and the selection and adjustment process for the next sampling period begins. When the system stabilizes, the current parameters are frozen, and steady-state monitoring begins. Through the above implementation, , , , , and Alignment and verification are performed in a linearly progressive link, so that the bandwidth stabilization judgment is performed at the lowest level value after a single-step adjustment, ensuring that the judgment and security constraints are consistent with the compaction conditions.

[0115] In this embodiment, step S6 specifically includes:

[0116] Record the operating condition data as ,in Includes power ,speed ,pressure strip temperature With effective bandwidth The value of the compaction condition is denoted as . ,in Give the upper and lower boundaries of the power, speed, and pressure ranges, and denote the safe adjustment range as . This includes the upper and lower power boundaries. , upper and lower boundaries of velocity , Upper and lower pressure boundaries , The maximum adjustable range is denoted as This includes the upper limit of the power step size. Speed ​​step size limit Pressure step size limit Let the monotonic direction be denoted as This includes directional markers for power, velocity, and pressure, which are either positive or negative. The gating coefficient of the segmented gating layer is denoted as... This includes the low-temperature coefficient. Coefficient of high temperature section The gating activation threshold is recorded as , which is the judgment threshold of the segmented gating layer output, and the initial adjustment parameter command is recorded as This includes the parameter to be adjusted, the adjustment direction, and the adjustment range. The bandwidth stabilization judgment result is recorded as... This serves as a flag to determine whether the system has stabilized or not; the sequence control command is recorded as... For instructions that can be issued by the execution layer, the set of executable instructions is denoted as... For the set of instructions that have passed verification, the time stamp is recorded as... Used to identify the current control cycle, the current parameter value is recorded as , where is the value of the parameter to be adjusted in the current control cycle;

[0117] In compliance verification, and Alignment, read the parameter markers to be adjusted ,in exist , , Take the value from the middle and record the requested adjustment range as . The corresponding upper limit of step size is denoted as The corresponding monotonic direction is marked as The corresponding interval boundary is denoted as , ,when Exceed Cut off at time Calculate the remaining boundary distance along the monotonic direction, denoted as . The way to obtain it is: when When it is positive, with and The difference as ,when When it is negative, with and The difference as When the difference is negative, Set to zero, and adjust the requested range accordingly. If the comparison exceeds the limit, the requested adjustment range will be truncated. The truncated amplitude is denoted as As the execution scope, construct executable instruction elements and load them. , and ,join in and time tags Alignment;

[0118] In the stop condition settings, when To stabilize and When satisfied, Set to stop adjustment, without changing power, speed, or pressure, and maintain [the desired state]. Unchanged, will and Align the output and freeze it. Unexecuted elements in the process are entered into steady-state monitoring;

[0119] In setting the conditions for advancement, when To prevent a return to stability, Set along Proceeding in a monotonous direction, from Read the current and , not exceeding The step size is to Proceed to the next candidate point within the range, the advancement method is: when When it is positive, set the target value to be in Add on the basis ,when When it is negative, set the target value to be at Reduce on the basis Recalculate the latest version before proceeding. When the advance reaches the boundary, the execution range will be truncated to... The next candidate point is set at the boundary position. After advancement, the two unselected parameters remain unchanged, and the advanced target value is compared with... Alignment updates are set as the starting point for the next control cycle; failure to advance is determined by... If the value is zero or the gate is disabled, no adjustments will be issued if the process fails; instead, the process will be returned to compliance verification to wait for the next cycle.

[0120] In the segmented limitation settings, read the strip temperature. Gating coefficient corresponding to material identification Activated when in low temperature range Activated when in high temperature range Record the gating enable flag as When the enabled gating coefficient is less than When Set to disabled when not less than When Set to enabled when When disabled, Setting the execution range to zero does not trigger execution; it only retains the execution value. The elements are used for re-verification in the next temperature range, when To enable, issue commands according to the above-mentioned stop or advance conditions. At the end of each control cycle, and Archive the data for use in aligning with the adjusted operating condition data;

[0121] Through the above implementation, , , , , and Alignment within the same control link, stopping adjustment when stabilization and compaction conditions are met, advancing in a monotonic direction with an upper limit step size when stabilization fails, and enabling or suppressing sequential control under segmented constraints from glass transition to melting temperature, along with an executable instruction set. With sequence control instructions In time tags It remains traceable and consistent with operating condition data.

[0122] 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. A method for stabilizing the bandwidth of thermoplastic layup based on neural-guided symbolic regression and a three-parameter process selector, characterized in that, include: S1. Acquire power, speed, pressure, strip temperature, effective bandwidth and material identification, generate working condition data and calculate compaction conditions; S2. Based on the operating condition data and material identification, construct a set of process influence items and calculate the edge outflow risk value in neural-guided symbolic regression. The neural network provides the weight and monotonic direction of the influence items. Under the segmented constraint from the glass transition to the melting temperature range, identify the safe adjustment range and the maximum adjustable range of power, speed, and pressure, and obtain the edge outflow risk value, the safe adjustment range, and the maximum adjustable range. S3. Under compaction conditions, based on the edge outflow risk value, the safety adjustment range, and the maximum adjustable range, a power adjustment path, a speed adjustment path, and a pressure adjustment path are formed. The risk reduction, impact on compaction, and response delay of each path within the allowable range are measured, and a first adjustment parameter instruction including the parameter to be adjusted, the adjustment direction, and the adjustment range is determined. S4. Perform single-step adjustment according to the initial adjustment parameter instruction and in accordance with the safe adjustment range and the maximum adjustable range, and generate adjustment response data; S5. Based on the adjusted response data, edge outflow risk value and compaction conditions, calculate the bandwidth stabilization judgment result. When the set threshold and compaction conditions are met, the bandwidth stabilization judgment result is formed. S6. Based on the initial adjustment parameter command and the bandwidth stabilization judgment result, generate the sequence control command.

2. The method for stabilizing thermoplastic layup bandwidth based on neural-guided symbolic regression and a three-parameter process selector according to claim 1, characterized in that, S1 specifically refers to: Power, speed, pressure, strip temperature, effective bandwidth, and material identification data are collected at the same time point to form a numerical sequence. The original signal set is arranged in the order of power, speed, pressure, strip temperature and effective bandwidth to generate working condition data containing five values. The material identifier is converted into a numerical vector of dimension 10 and aligned with the working condition data in the input layer. Based on the power, speed, pressure and strip temperature in the working condition data, the effective bandwidth is determined by setting a threshold, and the compaction conditions are calculated as the upper and lower boundaries of the power, speed and pressure ranges and the upper and lower boundaries of the strip temperature in the glass transition to melting temperature range. The working condition data and material identifiers are used to calculate the set of process influence items in neural-guided symbolic regression. The values ​​and trends of strip temperature are extracted and aligned with the effective bandwidth to ensure that the compaction conditions are consistent with the physical constraints for identifying edge outflow risk values.

3. The method for stabilizing thermoplastic layup bandwidth based on neural-guided symbolic regression and a three-parameter process selector according to claim 1, characterized in that, S2 specifically refers to: The operating condition data and material identifiers are aligned in the input alignment layer to form a 15-dimensional input vector, which is then mapped to the vector positions in the order of power, speed, pressure, strip temperature and effective bandwidth, and used as the input for the influence term generation layer. The set of process influence items is calculated on the 15-dimensional input vector at the influence item generation layer. It is generated item by item according to the combination of power-related items, speed-related items, pressure-related items, temperature-related items, bandwidth-related items and material identification-related items, and is aligned with the operating condition data and material identification. Based on the material identification and strip temperature, segmented gating signals from the glass transition to the melting temperature range are calculated in the segmented gating layer to obtain gating coefficients for the low-temperature and high-temperature segments, which are then used to segmentally enable and suppress the set of process influence items. The output of the influence term generation layer and the gating coefficient of the segmented gating layer are input into the symbol coefficient layer to generate the influence term weights and monotonic directions. The monotonic directions correspond to power, velocity and pressure, respectively, and are used to limit the direction of parameter changes. The process impact items set and their weights are explicitly expressed in the symbolic regression head assembly, constrained by the gating coefficient of the segmented gating layer, to calculate the edge outflow risk value and keep it aligned with the operating condition data; Based on the monotonic direction and the gating coefficient of the segmented gating layer, the calculation head performs a unidirectional scan of power, speed and pressure in the safe adjustment range, and outputs the upper and lower boundaries of the three segments to form the safe adjustment range; Based on the changing trends of strip temperature and effective bandwidth, and combined with the length of the safe adjustment range, the maximum adjustable range of head generation power, speed and pressure is calculated within the maximum adjustable range to obtain the edge outflow risk value, the safe adjustment range and the maximum adjustable range.

4. The method for stabilizing thermoplastic layup bandwidth based on neural-guided symbolic regression and a three-parameter process selector according to claim 3, characterized in that, The step of explicitly expressing and calculating the edge outflow risk value using the set of process influence items and their weights in the symbolic regression head assembly includes calculating the edge outflow risk value using a risk formula: ； in, This represents the risk value for edge outflow, a single value aligned with the operating condition. For the index of affected items, take numbers one to twenty-four. The weight of the influencing item A number, and Alignment The first of the set of process influence items A number, by generate, The low-temperature gating coefficient is in the first... The values ​​of each influencing item The high-temperature section gate coefficient is in the th... The values ​​of each influencing item This is the segmentation factor, with a value between zero and one, determined by the strip temperature. Relative to the glass transition temperature threshold With melting temperature threshold Generated using linear interpolation, and truncated when exceeding a threshold range to maintain a range of zero to one. This represents the glass transition temperature threshold, which is the value corresponding to the material identifier. Here, represents the melting temperature threshold, and represents the value corresponding to the material identifier. All of these variables are obtained directly in numerical form within the model and are consistent with... and Keep them aligned.

5. The method for stabilizing thermoplastic layup bandwidth based on neural-guided symbolic regression and a three-parameter process selector according to claim 1, characterized in that, S3 specifically refers to: Input the compaction conditions, edge outflow risk value, safe adjustment range, maximum adjustable range and operating condition data into the process three-parameter selector to determine the path starting point as the power, speed and pressure in the operating condition data; Within the safe adjustment range, with the maximum adjustable range as the step size, power adjustment path, speed adjustment path and pressure adjustment path are constructed sequentially in the monotonic direction. Ten candidate points are selected for each path to form path data. For each path, the risk reduction is calculated based on the operating status data and the edge outflow risk value. The difference between the edge outflow risk values ​​at the beginning and end of the path is taken as the risk reduction. For each path, the impact on compaction is calculated based on the compaction conditions and the working condition data at the end of the path. A set threshold is used to determine the offset and generate the impact value. The time it takes for a parameter adjustment to be transmitted to the effective bandwidth is taken as the response delay. The response delay is calculated for each path based on the relationship between the operating status data and the change in the effective bandwidth. The risk reduction, the impact on compaction, and the response delay are aligned to form a path calculation vector with three dimensions. This vector is then integrated with the upper and lower boundaries of the safety adjustment range and the maximum adjustable range to form a path constraint vector with nine dimensions. The path score is then calculated. The path with the highest score is selected based on the path score value. The parameters to be adjusted, the adjustment direction, and the adjustment range are output to form the initial adjustment parameter instruction. Constraint checks are performed based on the safe adjustment range and the maximum adjustable range. Instructions that do not meet the constraints are truncated in range and corrected in direction to obtain compliant initial adjustment parameter instructions.

6. The method for stabilizing thermoplastic layup bandwidth based on neural-guided symbolic regression and a three-parameter process selector according to claim 5, characterized in that, In the step of aligning the risk reduction, the impact on compaction, and the response delay into a path calculation vector with three dimensions, and merging it with the upper and lower boundaries of the safety adjustment range and the maximum adjustable range to form a path constraint vector with nine dimensions, the path score value is calculated using the path score formula: ； in, The path rating is a numerical value, which is related to the path index. Aligned single numerical values, The constraint coefficient takes a value between zero and one. The risk reduction amount is weighted, and the value is dimensionless. The weight of the influence on compaction is a dimensionless value. For response delay weighting, the unit is the reciprocal of time. The risk reduction is the difference between the marginal outflow risk values ​​at the start and end points of the path. The influence of this on compaction is denoted as , and the three offsets are denoted as . The dimensionless value obtained by the largest of the ratios is... To address latency, in order to ensure that the effective bandwidth first exceeds The difference between the detection time and the parameter adjustment time, for The calculation is performed according to the following process: The parameter belonging to the current path is denoted as... When the endpoint of the path crosses the boundary, Set to zero. When the boundary is not exceeded, calculate the remaining distance from the endpoint to the nearest boundary and divide it by the corresponding interval length to obtain the interval remaining ratio, denoted as . Calculate the complement of the ratio of the cumulative step size of the endpoint relative to the starting point to the maximum adjustable range of this parameter, and obtain the remaining step size ratio, denoted as . ,Will and Both are truncated to the range of zero to one, and the smaller value is taken as... .

7. The method for stabilizing thermoplastic layup bandwidth based on neural-guided symbolic regression and a three-parameter process selector according to claim 1, characterized in that, S4 specifically refers to: Align the initial adjustment parameter command with the safe adjustment range and the maximum adjustable range. Determine compliance based on the upper and lower boundaries and the upper limit of the step size. If non-compliant, truncate according to the upper and lower boundaries and correct according to the monotonic direction. Under compaction conditions, a single-step adjustment is adopted, in which the parameter to be adjusted is changed once according to the corrected adjustment range, while keeping the two unselected parameters unchanged. Immediately after adjustment, power, speed, pressure, strip temperature and effective bandwidth are collected to form adjusted operating condition data, and the data is aligned with the adjusted operating condition data to establish a correspondence between the before and after. The difference in effective bandwidth and response latency are calculated based on the correspondence between the two values. The adjusted adjustment range, the difference in effective bandwidth, and the response latency are combined to generate adjusted response data, which is then aligned with the edge outflow risk value.

8. The method for stabilizing thermoplastic layup bandwidth based on neural-guided symbolic regression and a three-parameter process selector according to claim 1, characterized in that, S5 specifically refers to: Align the adjusted response data, edge outflow risk value, and compaction conditions, organize them into inputs for judgment according to the time sequence of the adjusted working condition status data, and keep them aligned with the working condition status data. The corrected adjustment range in the adjustment response data is verified based on the safe adjustment range and the maximum adjustable range, and calculation is only performed when the upper and lower boundaries are not exceeded and the step size is not exceeded. Based on the difference in effective bandwidth and response latency in the adjusted response data, the risk reduction is calculated in conjunction with the change in edge outflow risk value. The difference in effective bandwidth is compared with the set threshold, and the response latency is also compared with the set threshold. Based on the compaction conditions, verify whether the power, speed and pressure in the adjusted working condition data are within the upper and lower boundaries of the compaction conditions. If they are not within the range, it is determined that the system will not stabilize. A bandwidth stabilization judgment result is generated when the risk reduction amount meets the set threshold, the difference in effective bandwidth meets the set threshold, and the compaction condition is met; no bandwidth stabilization judgment result is generated if any of the conditions are not met.

9. The method for stabilizing thermoplastic layup bandwidth based on neural-guided symbolic regression and a three-parameter process selector according to claim 1, characterized in that, S6 specifically refers to: Align the initial adjustment parameter command with the bandwidth stabilization judgment result, and verify compliance based on the upper and lower boundaries of the safe adjustment range and the upper limit of the maximum adjustable step size to obtain an executable instruction set. When the bandwidth stabilization judgment result meets the set threshold and compaction conditions, the sequence control command is set to stop adjustment, and the power, speed and pressure are kept constant. When the bandwidth stabilization judgment result has not been formed, the sequence control command is set to follow the monotonic direction of the first adjustment parameter command, and advance to the next candidate point within the safe adjustment range with a step size not exceeding the maximum adjustable range; The activation temperature range of the sequential control command is segmented and limited according to the gating coefficient of the segmented gating layer, so that the sequential control is aligned with the segmented constraint of the glass transition to melting temperature range.

Images