Parameter adaptive optimization control system and method for heat-conducting film cutting process

Abstract

The invention relates to the technical field of industrial control, and particularly discloses a parameter adaptive optimization control system and method for a heat-conducting film cutting process. According to the system, a first workpiece confirmation process configuration module, a trial cutting parameter characterization confirmation sub-module, a trial cutting evaluation vector calibration sub-module, a parameter characterization control sub-module and a cutting process parameter optimization module are arranged; the first piece confirmation process configuration module reads an initial parameter set of the current heat-conducting film, inputs the initial parameter set into the process knowledge model to output a historical prior sensitivity group and historical first piece confirmation parameter representation of a similar cutting process, and configures a first piece confirmation process of the current cutting process according to the historical prior sensitivity group and the historical first piece confirmation parameter representation; therefore, the blindness of trial cutting of the first workpiece is reduced, and the startup parameter adjustment time is shortened; and the cutting process parameter optimization module executes a cutting task after the first workpiece is confirmed successfully, forms a parameter updating quantity and updates cutting process parameter representation in a closed loop manner, so that online self-adaptive optimization is realized, chain defects caused by abnormal fluctuation are reduced, and the process consistency and yield are improved.

Description

Technical Field

[0001] This invention relates to the field of industrial control technology, specifically to a parameter adaptive optimization control system and method for the cutting process of surface-guided thermal film. Background Technology

[0002] Thermal conductive film cutting refers to the process of processing thermally conductive interface materials, which are composed of thermally conductive fillers and polymer matrix and usually have single-sided or double-sided release films, into specified shapes and sizes according to drawings. Common forms include full cut (cutting the film and its release film as a whole to form an independent pad) and half cut / kiss cut (cutting only through the adhesive layer or upper structure while retaining the bottom release film for waste removal and mounting). This process must ensure dimensional accuracy and contour consistency while controlling burrs, stringing, contamination and deformation of the cut edges to meet the requirements of subsequent mounting, heat dissipation contact and assembly reliability.

[0003] The current parameter optimization and process control of thermal conductive film cutting mainly rely on the empirical parameter window method, which is usually implemented by relying on the closed loop of general equipment in the existing cutting production line: Before production, the operator presets the execution settings of the cutting station in the human-machine interface (HMI) or production line formula management system based on the incoming material information such as material thickness, release film type and die specifications (for example, the pressing pressure or equivalent cutting depth is set by the cylinder / servo pressing mechanism of the die-cutting station, the linear speed is set by the cutter roller / spindle servo, the belt tension and feed stability are set by the unwinding / traction roller servo and tension controller, and the registration compensation is set by the CCD vision alignment system and registration servo). Then, through the first piece or small batch trial cut, the cutting penetration degree, bottom film integrity (half-cut scenario), waste discharge stability, dimensional deviation and burr pulling are judged by online vision inspection device, encoder / displacement sensor and waste detection sensor of peeling / rewinding section. If problems such as incomplete cutting, damage to the base film, or abnormal waste discharge occur, the operator mainly relies on experience to manually adjust the die-cutting station pressing settings (or cutter roller gap / cutting depth), spindle speed, unwinding / traction tension settings, and alignment compensation on the HMI, and verifies them through repeated trial cuts until they are usable. During mass production, the process stability is maintained by relying on periodic inspections, spot checks, and manual parameter readjustment during abnormal shutdowns. Overall, the basic servo / tension / alignment control capabilities of the equipment are still the means of execution, but parameter update decisions are highly dependent on manual experience and discrete spot checks to deal with parameter drift caused by material drift and die wear.

[0004] Based on the above technical solutions, it was found that existing thermal conductive film cutting methods often employ "process card solidification" in the first-piece trial cut confirmation stage. This involves first providing a set of set values ​​based on material type and experience, then conducting a small-scale trial cut to observe whether the film cuts open, whether the cut edge appearance is obviously abnormal, whether the underlying layer is visibly damaged, and whether waste material can be peeled off. Once qualified, this set of set values ​​is used as the standard for the shift and the film is released directly. Because thermal conductive films contain adhesive and a release layer, they are sensitive to changes in load and cycle time. Cutting needs to simultaneously consider the cutting effect, underlying layer protection, clean cut edges, and smooth peeling. The allowable adjustment margin is already small, and the first piece "appearing qualified" often only falls at the critical point. Once the material condition or equipment / conveying condition slightly shifts, this critical point will deviate, leading to micro-damage to the underlying layer or insufficient cutting in certain areas. These risks are often difficult to quantify and identify in the first-piece stage. Subsequently, if any abnormality occurs during mass production, the on-site staff tend to make single-point corrections based on the first piece benchmark, fixing the problem wherever it occurs. However, cutting quality is determined by the coupling of multiple factors, and single-point corrections can easily trigger chain changes, making it difficult to converge parameters through repeated trial and error. This ultimately leads to intermittent edge defects, unstable peeling, and dimensional fluctuations, which are concentrated in downstream processes such as film removal and mounting. This forces the production line to maintain stability by slowing down and tightening spot checks, resulting in simultaneous damage to yield and production capacity. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a parameter adaptive optimization control system and method for the cutting process of surface-guided thermally conductive films, which can effectively solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of the present invention provides a parameter adaptive optimization control system for a thermal conductive film cutting process, comprising: a first-piece confirmation process configuration module, used to read the initial parameter set of the current thermal conductive film to form a description vector of the current cutting process as input to a process knowledge model, output historical prior sensitivity groups of similar cutting processes and historical first-piece confirmation parameter representations, and execute a control unit to configure the first-piece confirmation process of the current cutting process; the first-piece confirmation process includes a trial cutting parameter representation confirmation submodule, a trial cutting evaluation vector calibration submodule, and a parameter representation control submodule; and a cutting process parameter optimization module, used to execute the cutting task of the current thermal conductive film based on the first-piece confirmation result, generate a cutting evaluation vector of the cutting process based on a set inspection cycle, and perform adaptive optimization control on the cutting process parameter representation.

[0007] The second aspect of the present invention provides a parameter adaptive optimization control method for a surface-mount thermal conductive film cutting process, comprising: S1. Reading the initial parameter set of the current thermal conductive film to form a description vector of the current cutting process as input to a process knowledge model, outputting the prior sensitivity group of similar cutting processes and the historical first-piece confirmation parameter representation, and executing the control unit to configure the first-piece confirmation process of the current cutting process; the first-piece confirmation process includes a trial cutting parameter representation confirmation submodule, a trial cutting evaluation vector calibration submodule, and a parameter representation control submodule; S2. Executing the cutting task of the current thermal conductive film based on the first-piece confirmation result, generating a cutting evaluation vector of the cutting process based on a set inspection cycle, and performing adaptive optimization control on the parameter representation of the cutting process.

[0008] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0009] (1) This invention provides a parameter adaptive optimization control system and method for the cutting process of thermal conductive film. The first piece confirmation process configuration module reads the initial parameter set of the current thermal conductive film (including incoming material information and key tags of process requirements) and constructs the description vector of the current cutting process. It inputs the vector into the process knowledge model to output the historical prior sensitivity group and historical first piece confirmation parameter representation of similar cutting processes. Based on this, the first piece confirmation process of the current cutting process is configured, so that the start-up parameters have a "reusable starting point" and the trial cutting strategy can automatically converge with the working conditions, thereby reducing the first piece trial... The module reduces the blindness of cutting and shortens the startup parameter adjustment time. After the first piece is confirmed to be passed, the cutting process parameter optimization module executes the cutting task and collects the cutting quality response under the set inspection cycle to generate a cutting evaluation vector (accuracy / stability / loss). Based on this, the risk level and threshold proximity of key risk items are calculated. Combined with the sensitivity of the current calibration side, the priority adjustment of the calibration side parameter group, adjustment direction and adjustment range are determined, forming the parameter update volume and updating the cutting process parameter characterization in a closed loop. This achieves online adaptive optimization, reduces the chain defects caused by abnormal fluctuations and improves process consistency and yield.

[0010] (2) By outputting the historical global sensitivity and the sensitivity of each calibration side under similar working conditions, this solution can identify in advance "which parameter groups are more sensitive and have smaller adjustable margins" in the first piece stage, thereby automatically tightening the trial cutting step size, reasonably configuring the number of trial cuttings and priorities, avoiding excessive adjustment under sensitive working conditions that leads to over-limit or repeated callbacks; at the same time, it allows for faster convergence under low-sensitivity working conditions, reducing unnecessary trial cuttings and material consumption, and achieving a balance between start-up efficiency and risk control.

[0011] (3) This solution integrates “working condition description vector - first piece confirmation parameter characterization - calibration side sensitivity - evaluation vector - parameter update amount” into a reusable data closed loop: the same set of evaluation vectors is used for risk assessment of the current first piece confirmation, as well as for decision-making on the direction and step size of adjacent trial cuts, and is continuously reused through the inspection cycle during the stable production stage to drive online optimization; at the same time, the current working condition data and sensitivity results can be directly used as the preliminary start-up basis for the next similar working condition after being stored in the database, so that the historical calibration data can continue to play a value in different batches and different time periods, reducing the repeated trial and error of “starting from zero every time the machine is started”.

[0012] (4) Compared with the existing method of "fixed parameter table, small number of first-piece trial cuts and release, and inspection callback", this solution upgrades the first-piece confirmation from a single-point decision of "visual qualification is solidified" to a closed-loop confirmation based on "previous similar working conditions, evaluation vector quantification, and risk level constraints". In the mass production stage, the parameters are continuously adaptively corrected through the inspection cycle. Therefore, it can identify and suppress implicit deviations and coupling instability earlier, reduce the ebb and flow and repeated trial and error caused by single-point callback, reduce downstream defects and rework caused by edge defects, peeling instability or size drift, and ultimately achieve improved yield, reduced intervention and more stable output of production capacity. Attached Figure Description

[0013] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0014] Figure 1 This is a schematic diagram of the system module connections of the present invention.

[0015] Figure 2 This is a schematic diagram of the method steps of the present invention.

[0016] Figure 3 This is a flowchart of the first-article confirmation configuration process for the present invention.

[0017] Figure 4 This is a flowchart illustrating the optimization of cutting process parameters according to the present invention.

[0018] Figure 5 This is a schematic diagram of the reasoning chain in the process knowledge model. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0020] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0021] It needs to be explained that, in practice, the thermally conductive film is placed on the bearing surface of the cutting station. The control system uses a moving seat as the conveying / feeding execution entity, driving the moving seat to position and feed along two directions corresponding to the first and second moving plates, so that the area to be cut sequentially reaches the action position of the cutting blade. After each positioning, the system first uses a clamping and fixing structure to flatten the film and suppress slippage, and then drives the cutting blade to cut the target contour line during the downward stroke. After cutting, the cutting blade is lifted and reset, and enters the next point for repeated positioning and cutting until the entire piece (or section) of the cutting trajectory is completed. The cut material then enters the peeling station and forms a folding peeling angle through the peeling plate, so that the waste layer that has been cut and separated is pried off at the edge of the peeling plate and pulled and collected along the waste path, while the finished product layer continues to be output along the main path. This realizes the integrated execution process of cutting, conveying, and peeling, with the moving seat completing the positioning and feeding, the cutting blade completing the cutting and shaping, and the peeling plate completing the separation of waste and finished product.

[0022] Example 1: Refer to Figure 1 As shown, the first aspect of the present invention provides a parameter adaptive optimization control system for a surface-guided thermal film cutting process, comprising: a first-piece confirmation process configuration module, a cutting process parameter optimization module, and a database. The first-piece confirmation process configuration module includes a trial-cutting parameter characterization confirmation submodule, a trial-cutting evaluation vector calibration submodule, and a parameter characterization control submodule. The database is used to store preset values ​​for various parameters.

[0023] The first piece confirmation process configuration module is connected to the cutting process parameter optimization module. The first piece confirmation process configuration module includes a trial cutting parameter characterization confirmation submodule, a trial cutting evaluation vector calibration submodule, and a parameter characterization control submodule. The trial cutting parameter characterization confirmation submodule is connected to the trial cutting evaluation vector calibration submodule, and the trial cutting evaluation vector calibration submodule is connected to the parameter characterization control submodule.

[0024] The first-piece confirmation process configuration module reads the initial parameter set of the current thermal conductive film to form the description vector of the current cutting process, which serves as input to the process knowledge model. It outputs historical prior sensitivity groups of similar cutting processes and historical first-piece confirmation parameter representations, configuring the first-piece confirmation process for the current cutting process. The specific process is based on... Figure 3 As shown, Figure 3 This is a flowchart of the first-article confirmation configuration process for the present invention.

[0025] In the diagram, the process begins by reading the initial parameter set of the thermal conductive film. Material information and process requirement tags are encoded into a condition description vector and input into the process knowledge model to obtain historical first-piece confirmation parameter representations and prior sensitivity groups for similar conditions. These parameters are used to determine the starting parameters and adaptive number of trial cuts for the first test cut. Subsequently, an evaluation vector (accuracy / stability / loss) is generated by collecting quality responses from the trial cuts. Based on this, a risk level and threshold proximity are constructed: if the risk is unacceptable, the process enters the "stability priority" branch, suppressing out-of-bounds risks by tightening the step size, backtracking, or adding more trial cuts; if the risk is acceptable, the sensitivity of each calibration side is further calculated, a priority parameter group is selected, and the adjustment direction is determined. The adjustment magnitude is then determined by combining the threshold proximity, generating the parameter update amount for adjacent trial cuts to configure the next trial cut. This closed-loop cycle continues until convergence and release conditions are met, ultimately solidifying the parameter center point and allowable fluctuation range. The current condition data and calibration / global sensitivity are stored in a database, providing a reusable prior start-up basis for subsequent similar conditions.

[0026] In this embodiment, the above-mentioned process knowledge model can be compared with the case-based reasoning model (CBR). It adopts the existing CBR mechanism of "case database construction - similarity retrieval - result reuse - result correction - write-back and sedimentation" to realize the feasible reasoning process of "inputting the working condition description vector and outputting the prior sensitivity group and historical first article confirmation parameter representation".

[0027] Specific examples Figure 5 As shown, Figure 5 This is a schematic diagram of the reasoning chain of the process knowledge model. In the input layer, incoming material information, key tag information and working condition constraints are read respectively. After feature fusion, a current process description vector x with a unified dimension is generated. In the core layer, the knowledge model first writes the historical cutting records into the historical case library in a structured way through case library construction. Each historical batch forms a case, which includes the historical working condition description vector xi, the corresponding first piece confirmation parameter representation pi and the historical prior sensitivity group si, and the case accumulation module completes the entry and update of the library.

[0028] Subsequently, under the constraints of the similarity rule base, the knowledge model calls the similarity retrieval module to calculate and sort the similarity scores between the current x and each historical xi, and selects the top T cases with the highest similarity as the set of similar working conditions. In the result reuse module, the si and pi of the set of similar working conditions are weighted and aggregated according to the similarity scores to obtain the historical prior sensitivity group s and the historical first piece confirmation parameter representation p of the current working condition, which are used as the output of "historical prior sensitivity group / first piece confirmation parameter" in the figure.

[0029] On one hand, s is used to guide the matching and correction of the number of trial cuts and the adjustment step size strategy in the first piece confirmation configuration; on the other hand, p serves as the initial parameter starting point for the first piece confirmation configuration and is issued to subsequent cutting processes. In the execution feedback layer, after the optimization system completes the first piece confirmation configuration, it sends it back to the correction and optimization module of the core layer; the correction and optimization module performs local correction on p under the constraints of the correction rule base to obtain p', and outputs the optimized process parameters for the next round of first piece confirmation or adjacent trial cut configuration.

[0030] Finally, the case accumulation module writes the current working condition description vector x, the corrected first piece confirmation parameter representation p', and the sensitivity group s' obtained from the statistics of the current first piece and subsequent processes back to the historical case library to form new cases or update the original cases, thereby realizing the reuse reasoning and continuous accumulation of process knowledge for thermal conductive film cutting process.

[0031] The initial parameter set of the current thermal conductive film mentioned above includes the incoming material information and process requirements of the current thermal conductive film, wherein the process requirements include the key label information of the current cutting process.

[0032] In this embodiment, the aforementioned incoming material information includes at least the material type / grade, batch number, width and roll length, nominal thickness and thickness tolerance of the thermal conductive film, structural layer information (including the number of adhesive layers and release layers and their interlayer configuration), surface coating type and release treatment type, material hardness or equivalent stiffness grade, viscosity grade or adhesion grade, release force grade and its test conditions, filler type and ratio grade (or thermal conductivity grade), packaging and storage conditions, and warehousing time.

[0033] Key label information includes the outline / size specifications of the target finished product, the complexity level of the cut shape (whether it contains fine structures such as small holes, sharp corners, and narrow bridges), the permissible edge defect level, the permissible level of impact on the underlying layer, waste stripping and collection method requirements, target cycle time / capacity requirements, target yield threshold, inspection cycle and sampling rules, and environmental requirement labels (including temperature range, humidity range, and cleanliness level requirements), thus forming an initial parameter set that can be used to characterize the current cutting conditions and serve as input for the process knowledge model.

[0034] The incoming material information of the thermal conductive film and the key label information of the current cutting process form the description vector of the current cutting process.

[0035] The aforementioned historical prior sensitivity group for similar cutting processes includes the historical global sensitivity of similar cutting processes and the historical sensitivity of each calibration side of similar cutting processes.

[0036] In this embodiment, sensitivity is used to characterize the "response strength" of the thermal conductive film cutting process to small parameter deviations. That is, when the parameters of the cutting process undergo minor changes within allowable limits, the degree and speed at which the cutting quality and process stability indicators improve or deteriorate accordingly. Historical global sensitivity is used to characterize the overall sensitivity of similar processes to parameter disturbances. By performing regression analysis on multivariate data in historical cutting records, it identifies which process parameters are most sensitive to quality and stability, thus providing a basis for parameter adjustment. Historical sensitivity on each calibration side is used to characterize the sensitivity of different parameter groups (cutting side, conveying side, and peeling side) to quality and loss responses. These sensitivities are calculated using techniques such as sensitivity analysis and response surface methodology (RSM). Combining machine learning algorithms (such as random forests and XGBoost) further improves prediction accuracy. The optimization system can provide a basis for configuring the number of trial cuts, prioritizing disturbances, and adaptive optimization control based on the sensitivity of each side, ensuring that accuracy is improved while controlling risks and losses during optimization, thereby increasing production efficiency and reducing defect rates.

[0037] The aforementioned historical first-piece confirmation parameters represent the corresponding historical calibration data sets, including the cutting blade parameter set on the cutting side, the moving seat parameter set on the conveying side, and the peeling plate parameter set on the peeling side. The historical calibration data sets are used as the first trial cut confirmation parameters for the current thermal conductive film cutting process.

[0038] In this embodiment, the historical calibration data set is used as the first trial cut confirmation parameters for the current thermal conductive film cutting process, serving as the starting point for startup parameters. The cutting blade parameter set on the cutting side may include set values ​​characterizing the cutting blade's execution strength (such as the cutting blade's pressing strength, equivalent stroke setting, and equivalent setting of the blade contact state), set values ​​characterizing the cutting cycle time and synchronization relationship (such as cutting frequency and phase / trigger timing setting), and set values ​​characterizing the relative alignment and pressure guidance state between the blade and the material (such as blade alignment error, material tension compensation, and pressing stability parameters). These set values ​​are mainly dynamically controlled by the cutting blade execution unit (servo motor and blade module), pressure controller, and vision / displacement sensor.

[0039] The parameter set of the conveyor-side moving seat may include setting values ​​for characterizing the matching of feeding and traction cycles (such as feeding speed reference, cycle synchronization coefficient, and positioning compensation setting), setting values ​​for characterizing centering and correction control (such as correction gain, correction response threshold, and path guidance compensation setting), and setting values ​​for characterizing conveying stability suppression (such as jitter suppression coefficient, slip compensation setting, and start / stop slope setting). These setting values ​​are mainly composed of the moving seat execution unit (servo drive and guide rail module), correction controller, and tension / displacement sensor to ensure accurate positioning and stable transmission of materials during the conveying process.

[0040] The peeling plate parameter set on the peeling side can include settings for characterizing the peeling path shape and turning method (such as equivalent settings for peeling angle / detour path and guide roller path configuration settings), settings for characterizing the peeling and collection coordination cycle (such as peeling speed ratio and collection / rewinding synchronization coefficient settings), and settings for characterizing peeling stability adjustment (such as waste breakage prevention threshold, re-adhesion suppression setting, and abnormal triggering backlash amplitude setting). These settings are mainly composed of the peeling plate execution unit (servo drive and traction roller module) and the waste rewinding unit, ensuring smooth peeling and recycling of waste while avoiding re-adhesion or waste breakage.

[0041] S101. Based on the historical calibration data set, the first trial cut is confirmed, and the parameters of each trial cut are characterized and confirmed under the adaptive number of trial cuts.

[0042] The specific process for limiting the number of adaptive trial cuts mentioned above is as follows:

[0043] Extract the historical global sensitivity matching of similar cutting processes to obtain the number of adaptive trial cuts for the current cutting process, calculate the average of the number of trial cuts set for the predefined process, and obtain the adaptive number of trial cuts for the current cutting process.

[0044] It should be explained that the above average calculation specifically involves rounding to the nearest integer to ensure that the number of trial cuts is an integer, which meets the requirements of actual operation. This operation ensures that the number of trial cuts can flexibly adapt to different working conditions, while avoiding material waste or operational complexity caused by excessive precision, and at the same time ensuring production efficiency and stability.

[0045] The number of adaptive trial cuts satisfies the minimum information content constraint, which includes a lower limit for the minimum number of trial cuts and an upper limit for the maximum number of trial cuts.

[0046] In this embodiment, the adaptive trial cutting number limitation also includes a dynamic judgment mechanism for configuring the trial cutting process to achieve adaptive convergence control of the trial cutting rounds: the parameter optimization system pre-sets an early stop condition. When the evaluation vector corresponding to several consecutive trial cuts meets the stability criterion, specifically, in several adjacent trial cuts, when the change amplitude of the evaluation vector is within the preset fluctuation threshold, it indicates that the cutting effect under the current start-up parameters has entered a repeatable stable state. It is determined that the current start-up parameters have reached a stable execution confirmation state, thereby terminating the remaining trial cutting rounds in advance to reduce start-up losses. At the same time, the parameter optimization system pre-sets a forced addition condition. When the fluctuation amplitude of the evaluation vector exceeds the preset range and the quality trend continues to deteriorate during the trial cutting process (the evaluation vector continuously deteriorates in multiple adjacent trial cutting cycles and the amount of deterioration reaches the preset upper limit), it is determined that the existing trial cutting information is insufficient to support reliable confirmation, triggering additional trial cutting rounds, thereby balancing start-up efficiency and material loss while ensuring cutting stability.

[0047] S102. The cutting process is calibrated in groups to obtain the calibration side. During the trial cutting cycle of the current trial cutting confirmation process, the quality response of the calibration side in the current trial cutting is collected, and the evaluation vector of the current trial cutting is generated.

[0048] The aforementioned calibration sides include the cutting side, the conveying side, and the peeling side.

[0049] The evaluation vector for the current test cut includes the accuracy evaluation index, the stability evaluation index, and the loss evaluation index for the current test cut.

[0050] The accuracy evaluation index is obtained by combining the accuracy evaluation parameters in the cutting blade parameter group on the cutting side, the accuracy evaluation parameters in the moving seat parameter group on the conveying side, and the accuracy evaluation parameters in the peeling plate parameter group on the peeling side. The stability evaluation index and the loss evaluation index are also obtained by combining the corresponding evaluation parameters in the cutting blade parameter group on the cutting side, the moving seat parameter group on the conveying side, and the peeling plate parameter group on the peeling side.

[0051] Specifically, the stability evaluation index for each trial cut is obtained by combining the stability evaluation parameters from the cutting blade parameter group on the cutting side, the moving seat parameter group on the conveying side, and the peeling plate parameter group on the peeling side. The wear evaluation index for each trial cut is obtained by combining the wear evaluation parameters from the cutting blade parameter group on the cutting side, the moving seat parameter group on the conveying side, and the peeling plate parameter group on the peeling side.

[0052] In this embodiment, the parameter optimization system constructs an evaluation vector for the current trial cut. The evaluation vector includes at least an accuracy evaluation index, a stability evaluation index, and a loss evaluation index, so as to achieve a unified quantitative characterization of the effects of the three parameter groups: the cutting blade on the cutting side, the moving seat on the conveying side, and the peeling plate on the peeling side.

[0053] This enables the optimization system to perform comparable evaluations of the effects of minor adjustments to different parameter groups within the same evaluation framework, and provides an interpretable optimization basis for adaptive control in adjacent trial cutting cycles.

[0054] The calculation process for accuracy evaluation, stability evaluation, and loss evaluation indices is based on the influence of each process parameter on the final quality, stability, and loss. These indices are obtained by quantifying and normalizing the relevant parameters. The normalized parameters are then weighted and combined to ensure consistency in weight and dimension among all parameters. Specifically, the optimization system first standardizes or normalizes each parameter in the cutting process to ensure that dimensional differences between different parameters do not cause deviations in the final calculation results. For example, for the tool depth and cutting speed on the cutting side, the optimization system converts them into dimensionless values ​​through normalization, typically using p...nor =(pp min ) / (p max -p min Normalize p, where p nor is the normalized parameter value; p is the original parameter value, i.e., the actual measured value of a certain cutting process parameter, such as tool depth, cutting speed, etc. min and p max These are the minimum and maximum values ​​of the parameter in historical data, respectively, which maps the parameter to the interval [0, 1]. The normalized parameter values ​​can be directly used to calculate evaluation indicators, avoiding bias caused by different parameter magnitudes.

[0055] The accuracy evaluation index reflects the consistency of the contour dimensions, the cleanliness of the cut edge, and the risk of incomplete cutting in certain areas during the trial cut. The calculation method is a weighted combination of various relevant parameter groups:

[0056] A k =[1 / (1+w1*|p depth -p depth '|+w2*|p speed -p depth '|)]+w3*p al A k p is the accuracy evaluation index for the k-th trial cut. depth p represents the cutting depth of the cutting blade, indicating the depth of cut during the cutting process. The blade position is monitored in real-time by a sensor, or the cutting depth is measured using a displacement sensor. depth ' represents the cutting tool reference depth; p speed The cutting speed of the cutting blade represents the speed at which the blade moves during the cutting process. It is obtained by measuring the blade's movement speed or by setting a value in the equipment control unit. depth ' represents the cutting reference speed. p al This is a parameter for evaluating the alignment of the cutting blade, representing the centering deviation between the cutting blade and the workpiece. The degree of centering deviation between the cutting head and the workpiece is detected by a vision system or displacement sensor. w1, w2, and w3 are preset weights for various relevant parameters in the database.

[0057] In calculating the aforementioned accuracy evaluation indicators, tool depth, cutting speed, and alignment assessment parameters are closely correlated. Tool depth directly affects the depth of cut during the cutting process; too shallow a depth may result in incomplete cutting, while too deep a depth may cause damage to the underlying material or edge burrs. Cutting speed determines the rate at which the tool cuts the material; if the speed is too high, it may lead to uneven cutting and irregular edges; if the speed is too slow, it may affect production efficiency and increase tool wear. The alignment assessment parameter ensures the accurate relative position of the tool and the workpiece; if the alignment is inaccurate, even if the tool depth and speed are appropriate, it will be difficult to achieve the expected cutting effect. These parameters work together to affect the consistency of contour dimensions, the cleanliness of the cut edge, and the risk of insufficient cutting in certain areas, thus requiring a high degree of consistency and coordination among the process parameters.

[0058] The stability evaluation index characterizes the repeatability and risk fluctuation level of the current trial cut under continuous production conditions, and is calculated as a weighted combination of various relevant parameter groups:

[0059] S k =[1 / (1+w4*p flu +w5*p vi )]+w6*p acc ;where S k p is the stability evaluation index for the k-th cut. flu This represents the load fluctuation of the moving seat during the cutting process. Load changes are measured using a force sensor or a current / power fluctuation detection device. vi The vibration of the moving seat indicates the vibration level of the moving seat during the cutting process. Vibration of the equipment is monitored using vibration sensors or accelerometers. acc The feed displacement deviation of the moving head represents the difference between the actual feed position / displacement and the target feed position / displacement. This deviation is measured using a displacement sensor or conveyor belt encoder. w4, w5, and w6 are preset weights for various parameters in the database.

[0060] As mentioned above, there is a clear correlation between load fluctuation, vibration, and feed displacement deviation when calculating stability evaluation indicators. Load fluctuation characterizes the changes in load borne by the equipment during the cutting process, which directly affects the operational stability of the equipment. Large load fluctuations may cause equipment vibration and instability in accuracy. Vibration reflects the instability of the equipment during operation; excessive vibration not only affects cutting accuracy but may also accelerate equipment wear. Feed displacement deviation ensures the stable movement of the workpiece during the cutting process. If the feed displacement deviation is large, it may lead to uneven cutting, affecting workpiece quality and production rhythm. These three factors are closely related and jointly determine the quality of the stability evaluation indicators. Therefore, controlling the stability of these parameters is crucial to ensuring the repeatability of the entire cutting process.

[0061] The loss evaluation index quantifies the comprehensive cost of a trial cut in terms of material consumption, time commitment, and capacity loss. The calculation method is a weighted combination of various relevant parameter groups:

[0062] C k =w7*p rate +w8*p dow +w9*p waste C k p is the loss evaluation index for the k-th trial cut. rate The scrap rate represents the proportion of waste material caused by poor cutting. It is recorded either through a production monitoring unit or through manual inspection and counting. dow This refers to the downtime of the stripping board, indicating the duration of downtime caused by stripping board malfunction or intervention. The downtime is obtained from the stripping board's operating records (e.g., PLC system logs). waste The unit length of good product scrap is calculated as the ratio between the cumulative length of scrap reeling (obtained by the encoder) and the number of good products produced in this trial cutting cycle (obtained by the counter). w7, w8, and w9 are preset weights for various relevant parameters in the database.

[0063] The above calculations of loss evaluation indicators also reveal a significant correlation between scrap rate, downtime, and material waste. Scrap rate indicates the proportion of defective products due to poor cutting, edge defects, etc., directly impacting material utilization. Downtime represents production interruptions caused by equipment malfunctions or debugging; excessive downtime leads to decreased production efficiency, indirectly increasing losses. Material waste primarily results from increased residual waste due to insufficient cutting or improper adjustments; excessive material waste not only affects costs but may also prolong production cycles. These three factors are interconnected; a high scrap rate is often accompanied by longer downtime and higher material waste, and vice versa. Therefore, controlling the relationship between these parameters can effectively reduce overall losses and improve production efficiency.

[0064] S103. Based on the evaluation vector of the current trial cut, the execution control unit performs adaptive control on the characterization of adjacent trial cut parameters in adjacent trial cut cycles.

[0065] The above adaptive control of adjacent trial cutting parameters is as follows:

[0066] Based on the accuracy evaluation index of the current trial cut, the risk level and threshold proximity of key risk items are constructed. The priority queue of adjacent trial cut control is determined by the risk level. Under the constraints of the above priority queue, the calibration side parameter group to be adjusted first is selected, the parameter adjustment direction is determined, and the sensitivity of each calibration side in the current trial cut is calculated. The adjustment range is determined in conjunction with the above threshold proximity.

[0067] In this embodiment, after completing the trial cut, the parameter optimization system first quantifies the key risk items based on the accuracy evaluation index of the trial cut. The key risk items are the set of risks that have the greatest impact on the yield of thermal conductive film cutting and the availability of downstream processes after cutting. They include at least the risk of insufficient cutting (manifested as local uncut or remnants), the risk of the bottom layer being affected (manifested as indentation, micro-damage or reduced integrity of the bottom layer), the risk of edge defects (manifested as burrs, stringing, gaps or contour damage), and the risk of peeling and pulling (manifested as pulling, re-attaching or material carrying tendency on the edge of the finished product during the waste peeling process).

[0068] To provide a calculable assessment of the aforementioned risks, the parameter optimization system monitors tool depth, cutting speed, and alignment status in real time, extracting evaluation parameters related to each risk item to assess the risks of the cutting process. Specifically, the risk of insufficient cutting is assessed by evaluating the impact of changes in tool depth and cutting speed on uncut features and residual probability; the risk of impact on the underlying material is assessed by evaluating the impact of tool depth on the indentation, micro-damage, and integrity of the underlying material, evaluating the abnormal features and integrity score of the underlying material; the risk of edge defects is assessed by evaluating the impact of cutting speed and alignment status on the burrs, notches, and roughness scores of the edges; and the risk of peeling and pulling is assessed by evaluating the peeling anomaly rate and pulling features through the combination of alignment status and cutting speed. The optimization system converts these evaluation parameters into risk scores using preset mapping rules, converts them into risk levels based on preset risk grading thresholds, and calculates the proportional relationship between the risk score and its threshold to obtain the threshold proximity, which characterizes how close the risk is to the boundary of exceeding the limit.

[0069] Furthermore, a priority queue for adjacent trial cut control is constructed based on risk level. The priority queue is constructed according to the risk level, threshold proximity, and process requirements of each key risk item. Specifically, risk level is the primary ranking dimension. The level is first determined based on the risk score of the control target for each trial cut cycle. High risk level usually refers to the risk score exceeding the preset high risk threshold. The corresponding process has problems such as insufficient cutting, damage to the underlying material, and out-of-tolerance critical dimensions, which seriously affect process stability or cause significant quality loss and require priority intervention. Medium risk level is the risk score within the preset medium risk threshold range. The overall process risk is controllable, but there are problems such as local precision deviation and minor edge defects, which need to be addressed after stability is ensured. Low risk level is the risk score below the preset low risk threshold. The process is running stably and only production efficiency related optimization needs to be focused on.

[0070] If multiple key risk items belong to the same risk level, the optimization system will further sort them by threshold proximity. The greater the proximity of the key risk items, the closer the actual detection value is to the process critical threshold and the closer the risk is to being out of control, requiring priority handling. At the same time, the system will make final fine adjustments based on the process requirements of the specific production scenario (e.g., the cutting process of core functional areas can be given higher priority at the same level).

[0071] After sorting, the optimization system determines the priority adjustment targets and calibration parameter groups for the adjacent trial cutting cycles under priority queue constraints. Then, it executes control according to the logic of "stability first, precision then optimization". First, it prioritizes risk suppression and stability control for high-risk processes. By adjusting key parameters such as tool depth, cutting speed, and conveyor belt positioning accuracy, it solves stability problems such as insufficient cutting, bottom layer damage, and cutting path deviation, ensuring process stability and avoiding the expansion of losses. Second, after stability is guaranteed, it turns to precision improvement control. For precision problems such as edge burrs, contour consistency deviation, and uncut local areas, it improves cutting precision by fine-tuning parameters such as alignment accuracy, tool angle, and segmented cutting speed. Finally, under the premise that risks are controlled and precision meets the standards (dimensional tolerances, edge quality, and other indicators meet preset standards), it performs loss optimization or cycle time improvement control. By optimizing the cutting path layout to reduce material waste, reasonably improving equipment operating cycle time and loading / unloading efficiency, and shortening the single-process cycle, it improves production efficiency and reduces production costs.

[0072] Subsequently, the parameter adjustment direction is determined by combining the micro-amplitude perturbation records of the parameter group in the historical calibration data with the change results of the evaluation vector in the current trial cut. That is, the effect of each candidate perturbation direction of the parameter group is judged. The candidate perturbation direction is a number of small-step adjustment directions pre-enumerated within the allowable adjustment range of the selected priority calibration side parameter group. Its form can be: "adjusting up / down" for a single parameter, or a combination direction of "synchronously adjusting up / synchronously adjusting down / one increasing and one decreasing" for a group of parameters. Therefore, the candidate perturbation direction is not the final positive adjustment direction, but a set of candidates for screening. Usually, multiple candidate directions exist at the same time.

[0073] The effect of each candidate perturbation direction is evaluated separately. Let j be the candidate direction number. Based on historical samples of similar disturbances, estimate its "local response" to the evaluation vector to obtain the predicted change of the accuracy evaluation index in that direction. And the predicted changes in the proximity of the risk scores / risk levels of each key risk item to the threshold. , Then, the prediction results are compared with the preset threshold to determine whether the direction meets the requirement of "no change in risk and effective improvement in accuracy". That is, the perturbation direction that can reduce or keep at least one key risk level, or reduce the proximity of the threshold (i.e., the risk is far away from the boundary), and at the same time make the accuracy evaluation index reach the preset minimum effective improvement amount is selected as the positive adjustment direction of the adjacent trial cutting cycle.

[0074] When multiple feasible positive adjustment directions exist, the optimization system further selects the best direction based on priority queue constraints. The system determines the positive adjustment direction for adjacent trial cutting cycles according to preset quantification judgment conditions. Specifically, the optimization system calculates candidate directions respectively. The corresponding change in the proximity of the key risk item thresholds and the change in accuracy evaluation indicators Where k represents the k-th cut, >0 indicates that the risk is far from the boundary. ≤0 indicates that the risk has not improved or has worsened. This indicates that after the k-th trial cut, if the next cut follows the candidate perturbation direction... The predicted / estimated value of the threshold proximity corresponding to the m-th key risk item when a slight adjustment is made to the parameters. This means that after the k-th trial cut, if the next press... During adjustment, the predicted / estimated values ​​of the accuracy evaluation index are obtained.

[0075] The optimization system first performs hard constraint gating: for any critical risk item m, if ≥ Or if the risk level rises, then Deemed infeasible and removed, among which The proximity of the warning threshold for the m-th key risk item; and if <ΔA min Then it is judged as an invalid improvement and removed, ΔA min This represents the minimum effective improvement. For the remaining feasible directions, the system calculates a ranking score: Where m* represents the highest priority risk item in the priority queue, the first item The first term represents the change in threshold proximity of the highest priority risk item; α, β, and γ are preset weights in the database; the second penalty term constrains the deterioration trend of other key risk items besides the highest priority risk item m*: when the threshold proximity of any other risk item increases instead of decreasing under the candidate perturbation direction, the system calculates the penalty amount according to its deterioration magnitude; the third penalty term constrains insufficient accuracy improvement: when the improvement in accuracy evaluation indicators brought about by the candidate perturbation direction does not reach the preset target improvement amount, the system calculates the penalty amount according to the insufficient magnitude. This indicates the pre-set target improvement amount.

[0076] Optimize the system to minimize the score. As the final positive adjustment direction; if there are ties, then further selection is made to make If the smallest direction is still parallel, then choose the direction with the smaller step size to reduce the risk of overshoot.

[0077] If a candidate perturbation direction can achieve the preset minimum effective improvement in the accuracy evaluation index, but causes any key risk level to rise or the threshold proximity to enter the warning range, then the perturbation direction is determined to have reached the risk boundary and is suppressed. Here, "reaching the risk boundary" covers two situations: first, the candidate direction itself does not meet the risk constraints and must be eliminated; second, when all candidate directions cannot simultaneously meet the constraints of "effective improvement in accuracy and no change in risk," the optimization system no longer forcibly pursues accuracy improvement, but switches to a stability-priority direction or remains unchanged. That is, it only selects conservative adjustments that can reduce threshold proximity / lower risk level, or triggers a control trial / additional trial to supplement local response data before updating the adjustment direction for the next cycle.

[0078] Furthermore, based on the quality response of the current trial cut and the disturbance amplitude of each calibration side parameter group, the sensitivity of each calibration side in the current trial cut is calculated. This sensitivity is used to characterize the influence intensity of small changes in each calibration side parameter group on the accuracy evaluation index and risk score. The sensitivity and threshold proximity are combined to determine the adjustment range, thereby outputting the parameter update amount of adjacent trial cut cycles and realizing adaptive control of the parameter characterization of adjacent trial cuts.

[0079] The above calculation of the sensitivity of each calibration side in the current trial cut is specifically implemented in this embodiment by using the control trial cut (or the previous benchmark trial cut) as a reference. Let the parameter set of the calibration side that is disturbed in the k-th trial cut be g∈{cutting side, transport side, stripping side}, and the parameter representation vector of this parameter set be p. g，k The parameter representation vector corresponding to the test cut is p. g，0 Then this small disturbance can be expressed as: Δp g，k =p g，k -p g，0 The normalized magnitude of this perturbation (used to eliminate differences in the dimensions of different parameters) is defined as follows: , where m g The number of parameters in parameter group g. Let j be the disturbance amount of the j-th parameter. ε is the scaling factor (e.g., standard deviation or allowable fluctuation range) for this parameter in a historical stable sample, and ε is a very small positive number to prevent the denominator from being zero. This represents the norm after normalization.

[0080] Furthermore, let the evaluation vector for this trial cut be: E k = k S k C k >, where A k S is the accuracy evaluation index. k C is a stability evaluation index. k This is the loss evaluation index; the corresponding control test cut evaluation vector is E0=<A0,S0,C0> To unify the dimensions, normalized changes are defined as follows: A range S range C range This refers to the fluctuation range or allowable range of available historical samples.

[0081] The sensitivity of the calibration side of the current trial cut for parameter set g can be defined by weighted finite difference as follows: , where w A w S w C Predefined weighting coefficients in the database are used to reflect the importance of "accuracy / stability / loss" under the current control objectives.

[0082] The determined adjustment direction and adjustment range are applied to the adjacent trial cutting parameter characterization to generate updated parameter characterization configurations for adjacent trial cutting cycles.

[0083] Specifically, the optimization system first represents the adjacent trial cutting parameters as a parameter vector to be updated, P(k) = [P1(k), ..., P2(k)]. d [(k)], where each dimension corresponds to a controllable parameter within the selected calibration side parameter group; the adjustment direction is represented as a direction vector d(k), whose components take {−1, 0, +1} to respectively characterize the parameter’s downward adjustment, maintenance, or upward adjustment in the next cycle; the adjustment magnitude is represented as a step size coefficient η(k) and combined with the reference step size s=[s1, …, s d ​The parameter update amount ΔP(k) = η(k) * (d(k) ⊙s) is obtained, where ⊙ represents element-wise multiplication. Based on this, the system generates the updated parameter representation P(k+1) = P(k) + ΔP(k), and performs amplitude limiting and constraint projection processing on P(k+1) to make it satisfy the allowable range and risk constraint conditions of each parameter; when any parameter exceeds the allowable range after the update, it is truncated to the boundary value and the convergence step size is reduced accordingly. Finally, the system sends P(k+1) to the execution units on the cutting, conveying, and stripping sides, enabling adjacent trial cuts to run under the new parameter configuration. After each adjacent trial cut, a new quality response is collected to generate an evaluation vector, which is used to verify whether the adjustment has improved the accuracy evaluation index, moved the key risk items away from the threshold, and controlled the loss evaluation index. If the verification result meets the preset target and risk constraints, the updated parameter representation P(k+1) is solidified as the candidate benchmark for the next stage. Otherwise, a rollback is triggered to correct the parameter representation. Rollback means restoring the parameter representation to the benchmark P(k) that passed the verification in the previous cycle or halving the update amount according to the preset rollback coefficient (such as P(k)+λΔP(k), 0<λ<1) to reduce overshoot, thereby achieving closed-loop convergence of the parameter representation in consecutive adjacent trial cut cycles, thereby improving cutting consistency and reducing anomalies and losses.

[0084] The above-mentioned joint determination of the adjustment range is as follows:

[0085] Based on the threshold proximity of key risk items, an adaptive step size factor is obtained through the preset matching rules in the database. This factor is then multiplied by the baseline adjustment step size to obtain the adaptive adjustment step size.

[0086] The sensitivity of each calibration side in the current trial cut is compared with the predefined reference sensitivity. If the sensitivity of a calibration side in the current trial cut is greater than the reference sensitivity, the adjustment step size is set to the first adaptive adjustment step size; otherwise, it is set to the second adaptive adjustment step size. The first adaptive adjustment step size limits the maximum adjustment amount in a single cycle.

[0087] In this embodiment, the optimization system employs a joint mechanism of "threshold proximity and sensitivity" to determine the adjustment range of adjacent trial cutting cycles. First, an adaptive step size factor is obtained based on the threshold proximity matching of key risk items, so that the step size automatically shrinks when the risk is close to the threshold and moderately expands when the risk is far from the threshold, thereby avoiding excessive parameter tuning near the risk boundary. On this basis, the sensitivity of each calibration side in the current trial cutting is further compared with a predefined benchmark sensitivity to determine the "response strength" of the current working condition to parameter fine-tuning. When the sensitivity of a calibration side is greater than the benchmark sensitivity, it indicates that a small change in the parameter on that side will cause a large fluctuation in the evaluation vector. Accordingly, the optimization system uses a first adaptive adjustment step size to limit the maximum adjustment amount in a single cycle, making the parameter tuning more conservative to improve stability. When the sensitivity of the calibration side is not greater than the benchmark sensitivity, it indicates that the change in the parameter on that side has a relatively slow impact on the evaluation vector. Accordingly, the optimization system uses a second adaptive adjustment step size to accelerate the convergence efficiency. It should be noted that the first adaptive adjustment step size is smaller than the second adaptive adjustment step size. Its function is to automatically adopt a more conservative small step update when the optimization system judges that it is "more easily affected by fine-tuning" or "closer to the risk boundary", while adopting a larger step size to improve optimization efficiency when the risk margin is sufficient and the response is relatively smooth.

[0088] The first piece confirmation configuration is completed by iteratively calibrating the evaluation vector and characterizing the parameters according to the adaptive trial cutting number iterations.

[0089] The cutting process parameter optimization module is used to execute the cutting task of the current thermal conductive film based on the first piece confirmation result. It generates a cutting evaluation vector for the cutting process based on the set inspection cycle and performs adaptive optimization control on the cutting process parameter representation. The specific process is based on, for example... Figure 4 As shown, Figure 4 This is a flowchart illustrating the optimization of cutting process parameters according to the present invention.

[0090] In the diagram, this process targets the stable production stage. It periodically collects cutting quality response data based on a preset inspection cycle and generates a cycle-specific evaluation vector (accuracy / stability / loss). This vector is then compared with the previous cycle's evaluation vector to obtain the change in indicators. Simultaneously, the risk level and its proximity to the threshold are calculated to determine if abnormal fluctuations or risk approaches are present. When risk approaches or abnormal fluctuations are detected, the process enters the "stability priority" branch, tightening the parameter update step size and triggering a control trial cut if necessary to supplement attributable data before issuing the update. When the risk is under control and fluctuations are normal, parameter updates are generated based on the change and sensitivity and executed. After the update, the execution effect is verified in a closed loop. If there is no improvement or the risk is out of control, the process returns to the stability priority branch for conservative correction. If the effect improves and the risk is under control, the cycle-specific evaluation vector and sensitivity / global sensitivity are stored in the database and iterate continuously in the next inspection cycle. This achieves long-term adaptive optimization and stable operation of the cutting process parameters without relying on frequent manual intervention.

[0091] The above-mentioned adaptive optimization control is performed on the parameters of the cutting process. The specific control process is as follows:

[0092] Within each inspection cycle, the cutting quality response data corresponding to that cycle is collected and a cutting evaluation vector is generated. The cutting evaluation vector includes accuracy evaluation index, stability evaluation index, and loss evaluation index.

[0093] In this embodiment, the optimization system performs periodic adaptive optimization control on the thermal conductive film cutting process by setting an inspection cycle. The inspection cycle is the time / length trigger interval for a centralized evaluation and parameter optimization of the cutting process. It can be limited to any one or a combination of a fixed duration, a fixed output quantity, or a fixed material running length according to preset rules. When a material batch state change is detected, the environmental state change exceeds a preset threshold, or the equipment enters a new lifespan stage, the inspection cycle is automatically shortened to increase the evaluation frequency. Within each inspection cycle, the optimization system acquires cutting quality response data from the online detection units and process acquisition units on the cutting side, conveying side, and peeling side. The cutting quality response data is statistically summarized and combined to obtain a cutting evaluation vector. The calculation methods for the accuracy evaluation index, stability evaluation index, and loss evaluation index are as described above.

[0094] The above-mentioned cutting evaluation vector is compared with the cutting evaluation vector of the previous inspection cycle. The changes in each evaluation index and the proximity of the threshold of the key risk items are extracted. The calibration side parameter group that should be adjusted first in this inspection cycle is determined. The adjustment direction and adjustment range under the current working conditions are determined. The parameter update amount representing the cutting process parameters is obtained and adaptive optimization control is completed.

[0095] After completing the evaluation of the current inspection cycle, the optimization system compares the current cutting evaluation vector with the cutting evaluation vector of the previous inspection cycle, and calculates the change in each evaluation indicator. The change refers to the relative rate of change of the same evaluation indicator between adjacent inspection cycles, used to characterize the degree of improvement or deterioration of accuracy, stability, and loss in the cycle dimension. The optimization system further constructs preset key risk items based on evaluation parameters directly related to risk assessment in the accuracy evaluation indicators, and calculates the risk score, risk level, and threshold proximity of each key risk item relative to a preset threshold to characterize the degree to which the risk approaches the boundary. Based on the change and threshold proximity, the system determines the control priority within the current inspection cycle and selects the calibration-side parameter group to be adjusted first. Then, it determines the adjustment direction by combining the local response patterns of this parameter group in historical calibration data, and jointly determines the adjustment magnitude based on the step size adaptive factor obtained by threshold proximity matching and the calibration-side sensitivity calculated in the current inspection cycle. This generates the parameter update quantity characterizing the cutting process parameters and issues it for execution.

[0096] After performing parameter updates, the optimization system enters the next inspection cycle and repeats the above process, so that the parameter representation of the cutting process gradually converges to a stable parameter center point and allowable fluctuation range that meets risk constraints and takes into account accuracy, stability and loss within the continuous inspection cycle.

[0097] It should be explained that, in this embodiment, after completing the adaptive parameter optimization control of the current cutting process, the parameter optimization system writes the current working condition data, the sensitivity of each calibration side, and the global sensitivity into the database as a priori sensitivity source when starting subsequent similar working conditions. The current working condition data includes at least the working condition description vector, the parameter representation confirmed by the first trial cut, the evaluation vector of each trial cut, and the corresponding parameter perturbation record. The global sensitivity is used to characterize the "comprehensive influence intensity of small overall parameter deviations on the cutting evaluation results" under the current working condition. It is assigned a global aggregation weight based on the risk contribution or adjustment priority of each calibration side parameter group in the current working condition. The normalized sensitivity of each calibration side is then weighted and summed according to the aggregation weight to obtain the global sensitivity. This allows the global sensitivity to simultaneously reflect the overall "adjustable margin" under multi-calibration side coupling conditions and is used as a priori criterion for configuring the number of adaptive trial cuts and tightening / loosening the adjustment step size during the next similar working condition retrieval.

[0098] Example 2:

[0099] Based on the threshold proximity matching step size adaptive factor and the selection of the first / second adaptive adjustment step size in combination with the calibration side sensitivity, under the condition that other conditions remain unchanged in Example 1, an adaptive learning rate step size update scheme can also be used as an equivalent alternative: The optimization system uses the "ratio between the change in evaluation vector and the update of parameters" in adjacent trial cutting cycles as equivalent gradient information, and statistically analyzes the magnitude and dispersion of the equivalent gradient in several consecutive trial cutting cycles to determine the update step size of the next cycle. Specifically, when the magnitude of the equivalent gradient exceeds the preset magnitude threshold, or the difference (or variance) between the maximum and minimum values ​​of the equivalent gradient in the sliding window exceeds the preset fluctuation threshold, or when the threshold proximity of any key risk item in the evaluation vector enters the warning interval, the system determines that updating the same parameters will cause greater evaluation fluctuations, and then lowers the step size adaptive factor to the preset lower limit, thereby reducing the probability of going out of bounds due to over-update.

[0100] When the absolute value of the equivalent gradient is lower than the preset amplitude threshold and its fluctuation is lower than the preset fluctuation threshold and shows a decreasing trend, as shown by the gradual decrease of the sliding window variance or the gradual decrease of the maximum and minimum difference, the upper limit of the parameter update amount in the next period is relaxed proportionally, so that the parameter update can reach the target range in fewer periods. The relaxation trigger condition can be clearly defined as follows: when the equivalent gradient amplitude is lower than the preset amplitude threshold and the fluctuation is lower than the preset fluctuation threshold, the optimization system will adjust the step size adaptive factor to the preset upper limit or gradually increase it according to the preset increment rate, thereby improving the parameter convergence efficiency without sacrificing risk constraints.

[0101] Reference Figure 2 As shown, the second aspect of the present invention provides a parameter adaptive optimization control method for a surface-mount thermal conductive film cutting process, comprising: S1. reading the initial parameter set of the current thermal conductive film to form a description vector of the current cutting process as input to the process knowledge model, outputting the prior sensitivity group of similar cutting processes and the historical first piece confirmation parameter representation, and executing the control unit to configure the first piece confirmation process of the current cutting process.

[0102] The first-piece confirmation process includes a trial-cut parameter characterization confirmation submodule, a trial-cut evaluation vector calibration submodule, and a parameter characterization control submodule.

[0103] S2. Based on the first piece confirmation result, execute the current thermal conductive film cutting task, generate the cutting evaluation vector of the cutting process based on the set inspection cycle, and perform adaptive optimization control on the parameter characterization of the cutting process.

[0104] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined by the present invention, they should all fall within the protection scope of the present invention.

Claims

1. A parameter adaptive optimization control system for the cutting process of surface-guided thermally conductive film, characterized in that, include: The first piece confirmation process configuration module is used to read the initial parameter set of the current thermal conductive film to form the description vector of the current cutting process as the input of the process knowledge model, output the historical prior sensitivity group of similar cutting processes and the historical first piece confirmation parameter representation, and execute the control unit to configure the first piece confirmation process of the current cutting process. The first-piece confirmation process includes a trial cutting parameter characterization confirmation submodule, a trial cutting evaluation vector calibration submodule, and a parameter characterization control submodule. The cutting process parameter optimization module is used to execute the cutting task of the current thermal conductive film based on the first piece confirmation result, generate the cutting evaluation vector of the cutting process based on the set inspection cycle, and perform adaptive optimization control on the cutting process parameter characterization.

2. The parameter adaptive optimization control system for the surface-guided thermal film cutting process according to claim 1, characterized in that, Includes the following steps: The initial parameter set of the current thermal conductive film includes the incoming material information and process requirements of the current thermal conductive film, and the process requirements include the key label information of the current cutting process; The incoming material information of the thermal conductive film and the key label information of the current cutting process form the description vector of the current cutting process.

3. The parameter adaptive optimization control system for the surface-guided thermal film cutting process according to claim 1, characterized in that, Includes the following steps: The historical prior sensitivity group of similar cutting processes includes the historical global sensitivity of similar cutting processes and the historical sensitivity of each calibration side of similar cutting processes; The historical first-piece confirmation parameters represent the corresponding historical calibration data set, including the cutting blade parameter set on the cutting side, the moving seat parameter set on the conveying side, and the peeling plate parameter set on the peeling side. The historical calibration data set is used as the first trial cut confirmation parameters for the current thermal conductive film cutting process.

4. The parameter adaptive optimization control system for the surface-guided thermal film cutting process according to claim 1, characterized in that, The configuration process for confirming the first piece in the current cutting process is as follows: The trial cutting parameter characterization and confirmation submodule is used to perform the first trial cutting confirmation based on the historical calibration data set, and to perform the characterization and confirmation of each trial cutting parameter under the adaptive number of trial cuttings. The trial cut evaluation vector calibration submodule is used to perform group calibration of the cutting process to obtain the calibration side. During the current trial cut cycle, the quality response of the calibration side in the current trial cut is collected, and the evaluation vector of the current trial cut is generated. The parameter characterization control submodule is used to adaptively control the parameter characterization of adjacent trial cuts in adjacent trial cut cycles by executing the control unit based on the evaluation vector of the current trial cut; The first piece confirmation configuration is completed by iteratively calibrating the evaluation vector and characterizing the parameters according to the adaptive trial cutting number iterations.

5. The parameter adaptive optimization control system for the surface-guided thermally conductive film cutting process according to claim 4, characterized in that, Includes the following steps: The specific process for limiting the number of adaptive trial cuts is as follows: Extract the historical global sensitivity matching of similar cutting processes to obtain the number of adaptive trial cuts for the current cutting process, and calculate the average with the number of trial cuts set for the predefined process to obtain the adaptive number of trial cuts for the current cutting process; The adaptive number of trial cuts satisfies the minimum information content constraint, which includes a lower limit for the minimum number of trial cuts and an upper limit for the maximum number of trial cuts.

6. The parameter adaptive optimization control system for the surface-guided thermal film cutting process according to claim 4, characterized in that, Includes the following steps: The calibration side includes the cutting side, the conveying side, and the peeling side; The evaluation vector for the current test cut includes the accuracy evaluation index, the stability evaluation index, and the loss evaluation index for the current test cut. The accuracy evaluation index is obtained by combining the accuracy evaluation parameters in the cutting blade parameter group on the cutting side, the accuracy evaluation parameters in the moving seat parameter group on the conveying side, and the accuracy evaluation parameters in the peeling plate parameter group on the peeling side. The stability evaluation index and the loss evaluation index are also obtained by combining the corresponding evaluation parameters in the cutting blade parameter group on the cutting side, the moving seat parameter group on the conveying side, and the peeling plate parameter group on the peeling side.

7. The parameter adaptive optimization control system for the surface-guided thermal film cutting process according to claim 6, characterized in that, Includes the following steps: The adaptive control of the adjacent trial cutting parameter characterization process is as follows: Based on the accuracy evaluation index of the current trial cut, the risk level and threshold proximity of key risk items are constructed. The priority queue of adjacent trial cut control is determined by the risk level. Under the constraints of the priority queue, the calibration side parameter group to be adjusted first is selected, the parameter adjustment direction is determined, and the sensitivity of each calibration side in the current trial cut is calculated. The adjustment range is determined in conjunction with the threshold proximity. The determined adjustment direction and adjustment range are applied to the adjacent trial cutting parameter characterization to generate updated parameter characterization configurations for adjacent trial cutting cycles.

8. The parameter adaptive optimization control system for the surface-guided thermal film cutting process according to claim 7, characterized in that, Includes the following steps: The joint determination of the adjustment range is specifically determined as follows: The adaptive step size factor is obtained by matching the threshold proximity of key risk items, and multiplied by the baseline adjustment step size to obtain the adaptive adjustment step size. The sensitivity of each calibration side in the current trial cut is compared with the predefined reference sensitivity. If the sensitivity of a calibration side in the current trial cut is greater than the reference sensitivity, the adjustment step size is set to the first adaptive adjustment step size; otherwise, it is set to the second adaptive adjustment step size. The first adaptive adjustment step size limits the maximum adjustment amount in a single cycle.

9. The parameter adaptive optimization control system for the surface-guided thermal film cutting process according to claim 1, characterized in that, Includes the following steps: The adaptive optimization control of the cutting process parameters is performed as follows: Within each inspection cycle, the cutting quality response data corresponding to that cycle is collected and a cutting evaluation vector is generated. The cutting evaluation vector includes accuracy evaluation index, stability evaluation index and loss evaluation index. The cutting evaluation vector is compared with the cutting evaluation vector of the previous inspection cycle. The changes in each evaluation index and the proximity of the threshold of the key risk items are extracted. The calibration side parameter group that should be adjusted first in this inspection cycle is determined. The adjustment direction and adjustment range under the current working conditions are determined. The parameter update amount representing the cutting process parameters is obtained and adaptive optimization control is completed.

10. A parameter adaptive optimization control method for the cutting process of surface-guided thermally conductive film, characterized in that: include: S1. Read the initial parameter set of the current thermal conductive film to form the description vector of the current cutting process as the input of the process knowledge model, output the prior sensitivity group of similar cutting processes and the historical first piece confirmation parameter representation, and execute the control unit to configure the first piece confirmation process of the current cutting process. The first-piece confirmation process includes a trial cutting parameter characterization confirmation submodule, a trial cutting evaluation vector calibration submodule, and a parameter characterization control submodule. S2. Based on the first piece confirmation result, execute the current thermal conductive film cutting task, generate the cutting evaluation vector of the cutting process based on the set inspection cycle, and perform adaptive optimization control on the parameter characterization of the cutting process.

Images