A waterproof material laser cutting path optimization method based on deep learning

Abstract

The present application relates to the technical field of laser cutting, and particularly relates to a waterproof material laser cutting path optimization method based on deep learning, which comprises the following steps: obtaining basic size data, thickness change rate distribution data and preloaded stress distribution data of a target test piece, identifying key / non-key path segments through discretization grid units; cutting in the non-key path segments by using basic cutting parameters, monitoring edge size data in real time and dynamically adjusting parameters to form verification cutting parameters; screening first confidence path segments based on stable cutting quality, extracting verification parameters to construct confidence data set; determining key cutting parameters through parameter fluctuation analysis and correlation calculation to form confidence data set; cutting key path segments by using the parameters; and finally establishing a deep learning model based on historical confidence data set to output basic cutting parameters of a new test piece. The method realizes adaptive optimization of cutting parameters and path, and significantly improves the cutting precision of waterproof material test pieces.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of laser cutting, in particular to a waterproof material laser cutting path optimization method based on deep learning. BACKGROUND

[0002] The waterproof material test piece preparation equipment is mainly used for waterproof material test piece preparation, which can ensure the standardization of the test piece during the preparation of the waterproof material test piece. At present, the mature waterproof material test piece preparation mainly adopts the template or cutter method to process the test piece, which has many defects: the cutting equipment lacks effective control over the material deformation, tool wear and other conditions during cutting, resulting in large size deviation of the cut test piece, which cannot meet the precision millimeter level processing requirement of the detection standard; burr problem exists in the formed test piece, which affects the physical and chemical properties of the waterproof material and greatly affects the reliability of the detection result; the waterproof material detection parameters are complex, and the test piece shape and size requirements are different, the existing cutting equipment is limited to simple straight line cutting, and it is difficult to process complex test piece; the test piece preparation time cycle is long, which causes the detection cycle to be prolonged and the detection equipment utilization rate to be reduced, increases the equipment investment and labor cost, and reduces the production benefit. At the same time, the existing automatic sample cutting equipment can only realize basic cutting control, lacks intelligent optimization and self-learning ability, and cannot optimize the cutting path and parameters according to the material characteristics, cutting historical data and other conditions, the operator still needs to manually adjust a large number of parameters, the operation is complicated and easy to make mistakes, at the same time, the software is difficult to compatible with new detection standards and material types, and the upgrade is difficult, which limits the long-term use value of the equipment.

[0003] The laser cutting method has the following advantages: completely solves the influence of template wear or deformation, reduces the cost of template replacement; laser cutting uses computer drawing, and has test piece preparation one-time forming, which can greatly improve the work efficiency, and the running process is safe and stable, accurate and reliable, which can effectively ensure the accuracy of the test piece preparation and detection data. Therefore, the laser cutting path optimization method of the waterproof material with self-learning function has important use value for the preparation of the waterproof material test piece. SUMMARY

[0004] Therefore, the present application provides a waterproof material laser cutting path optimization method based on deep learning, which can overcome the problem of poor cutting precision caused by the inability to adjust the cutting laser parameters in real time according to the cutting precision in the prior art.

[0005] To achieve the above purpose, the present application provides a waterproof material laser cutting path optimization method based on deep learning, which comprises:

[0006] Step S1: Obtain the basic size data, thickness change rate distribution data, and preloaded stress distribution data of the target specimen of the waterproof material, and determine the critical path segment and non-critical path segment of the cutting path of the target specimen.

[0007] Step S2: Determine basic cutting parameters based on the basic size data of the target specimen, cut the non-critical path segment of the target specimen using the basic cutting parameters, and monitor the cutting edge size data of the target specimen in real time.

[0008] Step S3: Based on the comparison between the obtained cutting edge size data of the target specimen and the sample preparation preset data, the basic cutting parameters are adjusted to obtain the verification cutting parameters, and the laser cutting of the non-critical path segment is completed using the verification cutting parameters.

[0009] Step S4: Determine the first confidence path segment based on the cutting edge size data of the non-critical path segment, and extract the corresponding verification cutting parameters to form an accepted dataset;

[0010] Step S5: Based on the fluctuation of the verification cutting parameters in the accepted dataset, determine the correlation of the verification cutting parameters, obtain the key cutting parameters, and form a confidence dataset;

[0011] Step S6: Laser cut the critical path segment using the key cutting parameters in the confidence dataset to obtain the target specimen finished product;

[0012] Step S7: Establish a deep learning optimization model based on several confidence datasets, and output the basic cutting parameters for subsequent specimens.

[0013] Further, step S1 includes:

[0014] Step S11: Obtain the basic dimensional data of the target specimen, including the cutting path and thickness reference value;

[0015] Step S12: Obtain the thickness change rate distribution data of the sample containing the target specimen;

[0016] Step S13: Based on the testing conditions of the target specimen and the basic size data, perform finite element analysis to generate the preloaded stress distribution data;

[0017] Step S14: Based on the thickness change rate distribution data and the preloaded stress distribution data, determine the critical path segments and non-critical path segments of the cutting path.

[0018] Further, step S14 includes:

[0019] Step S141: Discretize the cutting path into several grid cells, wherein the grid resolution of each grid cell is set according to the thickness reference value.

[0020] Step S142: Perform binary classification on each mesh element. If the judgment conditions of thickness change rate threshold or stress threshold are met, it is marked as a critical mesh element; otherwise, it is marked as a non-critical mesh element.

[0021] Step S143: Based on the testing conditions of the target specimen, forcibly mark the mesh elements corresponding to the stress-bearing area as critical mesh elements;

[0022] Step S144: Remove isolated single-point critical mesh cells;

[0023] Step S145: Merge adjacent grid cells of the same type to form continuous critical path segments and non-critical path segments.

[0024] Further, step S145 includes:

[0025] Step S1451: Perform morphological closing operation on the key mesh elements to fill the gaps;

[0026] Step S1452: Perform connected component analysis on key grid cells and merge discrete segments with spacing less than a preset merging threshold.

[0027] Step S1453: Output the topologically continuous closed path segments.

[0028] Furthermore, in step S144, the isolated single-point critical mesh cell is a single critical mesh cell in which a predetermined number of consecutive mesh cells in the cutting path direction are all of non-critical type.

[0029] Furthermore, step S1 also includes:

[0030] Step S15: When the length of the longest non-critical path segment in the non-critical path segments is less than the minimum characterization length, a compensation cutting segment is generated in the longest non-critical path segment of the target specimen.

[0031] Step S16: The compensated cutting segment is incorporated into the non-critical path segment to form an extended non-critical path segment, so that the length of the extended non-critical path segment meets the minimum representation length requirement.

[0032] Further, step S3 includes:

[0033] Step S31: Compare the real-time monitored cutting edge size data with the sample preparation preset data one by one, and calculate the corresponding size deviation values.

[0034] Step S32: When a predetermined number of consecutive dimensional deviation values ​​exceed the allowable tolerance range, the basic cutting parameters are adjusted, wherein the basic cutting parameters are adjusted based on the direction and magnitude of the dimensional deviation values, and the basic cutting parameters include laser power and cutting speed.

[0035] Step S33: Complete the cutting of the current non-critical path segment using the adjusted verification cutting parameters, and continuously acquire the cutting edge size data of the target specimen and the corresponding relationship of the verification cutting parameters during the cutting process.

[0036] Further, step S4 includes;

[0037] Step S41: Filter out continuous segments in the non-critical path segments where the cutting edge size data is continuously within the preset tolerance zone, and mark them as the first confidence path segments;

[0038] Step S42: Extract the set of verification cutting parameters in the first confidence path segment, denoted as the acceptance cutting parameters, and extract the cutting edge size data corresponding to the acceptance cutting parameters, denoted as the acceptance size data;

[0039] Step S43: Store the acceptance cutting parameters and acceptance size data in spatial order to form an acceptance dataset, and record the mapping relationship between the acceptance cutting parameters and acceptance size data.

[0040] Further, step S5 includes:

[0041] Step S51: Divide each of the first confidence path segments according to the change of the acceptance cutting parameters to form a number of second confidence path segments;

[0042] Step S52: Calculate the correlation coefficient between the acceptance cutting parameter and the corresponding acceptance size data in each second confidence path segment to determine the degree of correlation between each acceptance cutting parameter and the corresponding acceptance size data;

[0043] Step S53: Remove the confidence cut parameters whose correlation coefficients are less than a preset correlation coefficient threshold to obtain the key cut parameters and form the confidence dataset.

[0044] Further, step S7 includes:

[0045] Step S71: Collect the confidence datasets and corresponding cutting quality rating datasets of multiple sets of target specimen finished products;

[0046] Step S72: Construct a supervised learning model with the cutting parameters as input and the cutting quality rating as output;

[0047] Step S73: When the basic size data of the new specimen is input, the optimized initial values ​​of the cutting parameters are output.

[0048] Compared with the prior art, the beneficial effects of the present invention are that, in view of the functional characteristics differences of different areas of waterproof test specimens, the present invention adopts a graded processing mechanism of critical / non-critical path segments to ensure ultra-high precision cutting of critical areas (such as stress areas and sealing surfaces) while using non-critical segments to quickly complete parameter self-learning, which greatly reduces the overall debugging time and establishes a closed-loop system of dynamic parameter adjustment, confidence data extraction and deep learning optimization, so that the cutting process has the ability to continuously evolve and significantly improves the adaptability and stability under complex working conditions.

[0049] Furthermore, this invention uses a dual-threshold determination mechanism of thickness change rate and stress distribution to accurately identify critical path segments affecting waterproof performance. Based on finite element analysis and pre-loaded stress data, combined with forced marking of highly sensitive areas under testing conditions, it ensures that special cutting strategies are used in areas of abrupt changes in material thickness and stress concentration areas, thus avoiding defects such as weld beads and cracks from the source. At the same time, it generates continuous cutting paths through morphological closing operations and connected domain analysis to eliminate parameter oscillations caused by data noise, ensuring the energy stability of cutting in critical areas, and enabling the specimen to demonstrate true and reliable performance in subsequent tests.

[0050] Furthermore, this invention achieves intelligent improvement in cutting efficiency by dynamically adjusting parameters and extracting confidence data in non-critical path segments. Through a real-time closed-loop "cutting-monitoring-feedback" system, power and speed parameters are automatically optimized based on the direction and magnitude of dimensional deviations, and a confidence dataset is constructed based on continuous stable segments. By compensating for cutting segment expansion and filtering parameter correlations, the invention ensures that non-critical segment data has sufficient statistical significance, avoiding ineffective debugging and providing highly reliable parameters for critical segments. This transforms traditional trial-and-error parameter optimization into a data-driven, efficient process, significantly shortening the new product development cycle.

[0051] Furthermore, this invention endows the cutting system with continuous evolution capabilities by constructing a deep learning model based on historical confidence datasets. Using optimized cutting parameters and quality ratings as training samples, it establishes an intelligent mapping relationship from basic dimensions to initial parameters, solving the cold start problem under new materials and new working conditions. The model automatically outputs initial parameters adapted to specific thicknesses and material properties, significantly reducing the need for manual intervention. At the same time, through periodic data iteration updates, the system is made robust to cope with material batch fluctuations and environmental changes, maintaining a high-precision cutting level in the long term. Attached Figure Description

[0052] Figure 1 This is a flowchart of the deep learning-based laser cutting path optimization method for waterproof materials according to the present invention;

[0053] Figure 2 This is a flowchart of step S1 of the deep learning-based method for optimizing the laser cutting path of waterproof materials according to the present invention.

[0054] Figure 3 This is a flowchart of step S14 of the deep learning-based method for optimizing the laser cutting path of waterproof materials according to the present invention.

[0055] Figure 4 This is a schematic diagram of step S1 of the deep learning-based laser cutting path optimization method for waterproof materials according to the present invention.

[0056] In the diagram: 11 - Isolated single-point critical mesh element; 21 - Critical mesh element; 22 - Critical path segment; 31 - Non-critical mesh element; 32 - Non-critical path segment; 41 - Compensation cut segment; 42 - Deleted segment; a1 - First front mesh element; a2 - Second front mesh element; a3 - Third front mesh element; b1 - First rear mesh element; b2 - Second rear mesh element; b3 - Third rear mesh element. Detailed Implementation

[0057] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0058] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0059] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0060] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0061] Please see Figure 1The diagram shows a flowchart of the laser cutting path optimization method for waterproof materials based on deep learning, as described in this invention. This embodiment of the invention provides a laser cutting path optimization method for waterproof materials based on deep learning, comprising:

[0062] Step S1: Obtain the basic size data, thickness change rate distribution data, and preloaded stress distribution data of the target specimen of the waterproof material, and determine the critical path segment and non-critical path segment of the cutting path of the target specimen.

[0063] Please see Figures 2 to 4 The diagrams shown are, respectively, flowcharts of step S1 of the deep learning-based laser cutting path optimization method for waterproof materials according to the present invention, flowcharts of step S14 of the deep learning-based laser cutting path optimization method for waterproof materials according to the present invention, and a schematic diagram of step S1 of the deep learning-based laser cutting path optimization method for waterproof materials according to the present invention. Specifically, step S1 includes:

[0064] Step S11: Obtain the basic dimensional data of the target specimen, including the cutting path and thickness reference value;

[0065] During implementation, the reference thickness value of the specimen, in millimeters (mm), denoted as T, is obtained from the CAD drawing of the target specimen; as well as the preset cutting path of the specimen;

[0066] Step S12: Obtain the thickness change rate distribution data of the sample containing the target specimen;

[0067] In practice, a laser thickness gauge is used to scan the entire surface of the sample containing the target specimen, and the thickness change rate distribution data of the sample containing the target specimen is calculated and denoted as R. ij Preferably, the detection step size of the above thickness change rate distribution data is 0.1 mm to 1 mm;

[0068] Step S13: Based on the testing conditions of the target specimen and the basic size data, perform finite element analysis to generate the preloaded stress distribution data;

[0069] In practice, the stress distribution of the specimen under the corresponding testing conditions is simulated through finite element analysis, and an equivalent stress cloud diagram is output to obtain the preloaded stress distribution data, in megapascals (MPa), denoted as σ. ij .

[0070] In this embodiment, the finite element analysis specifically involves:

[0071] A three-dimensional solid model is established based on the basic dimensional data of the target specimen, with the model boundary strictly following the specimen contour. Constitutive model parameters of the waterproof material are input, including elastic modulus, Poisson's ratio, and yield strength. Load conditions are set according to the test condition type and the GB / T328 standard, and load application areas are preset in the model. For example, for tensile testing, clamping surface areas are generated at both ends of the specimen to simulate clamp constraints; for hydrostatic testing, a uniformly distributed normal pressure load is applied to the specimen surface. Static structural analysis is performed, and an equivalent stress cloud map is output to obtain preloaded stress distribution data.

[0072] Those skilled in the art will understand that the testing conditions include tests that require load application, such as the peel test of the lap joint, tensile performance test, tear performance test, and impermeability test of waterproof membranes as specified in GB / T328 "Test Methods for Waterproof Membranes in Buildings". These will not be elaborated further.

[0073] Those skilled in the art will understand that finite element analysis is an existing and mature technology, and its specific operational details will not be elaborated here.

[0074] It is understandable that areas of abrupt thickness change in waterproof materials are prone to forming weld beads due to heat accumulation during laser cutting, while laser cutting defects in high-stress areas may become crack sources, interfering with the accuracy of subsequent testing.

[0075] Step S14: Based on the thickness change rate distribution data and the preloaded stress distribution data, determine the critical path segment 22 and non-critical path segment 32 of the cutting path.

[0076] Specifically, step S14 includes:

[0077] Step S141: Discretize the cutting path into several grid cells, wherein the grid resolution of each grid cell is set according to the thickness reference value.

[0078] In practice, the formula for calculating the grid resolution is as follows:

[0079] δ = 2 - 0.1 × T, and δ ≥ 1 mm.

[0080] Where δ is the grid resolution, which is the side length of a single grid cell, in millimeters (mm).

[0081] Understandably, the grid resolution needs to match the laser processing precision (slightly larger than the line width of the cutting laser, which is usually no more than 1mm) to avoid path deviations caused by discrete errors. At the same time, the thinner the waterproof material, the more sensitive it is to the processing parameters of laser cutting, and the grid resolution should be increased.

[0082] Step S142: Perform binary classification on each mesh cell. If the judgment conditions of thickness change rate threshold or stress threshold are met, it is marked as a critical mesh cell 21; otherwise, it is marked as a non-critical mesh cell 31.

[0083] In implementation, the thickness change rate distribution data R ij and preloaded stress distribution data σ ij Map the corresponding spatial coordinates to the grid cells, and judge each grid cell. When the thickness change rate distribution data R is satisfied... ij Greater than or equal to the thickness change rate threshold R0, or the preloaded stress distribution data σ ij If any stress threshold σ0 is greater than or equal to any of the stress thresholds, the mesh element is marked as a critical mesh element 21; otherwise, it is marked as a non-critical mesh element 31. The thickness change rate threshold R0 is preferably 0.05. The stress threshold σ0 is determined based on the yield strength of the waterproof material. Preferably, σ0 is 0.3 times the yield strength of the waterproof material.

[0084] Understandably, the thickness change rate threshold R0 is set based on the critical value for melt formation calibrated from historical experimental data: when the thickness change rate is ≥5%, uneven heat conduction leads to a significant increase in the risk of molten material accumulation. The stress threshold σ0 corresponds to the crack initiation stress threshold; exceeding this value will significantly increase the probability of cutting defect propagation. The dual-condition logic ensures that any risk factor is captured.

[0085] Step S143: Based on the testing conditions of the target specimen, the mesh elements corresponding to the stress-bearing area are forcibly marked as key mesh elements 21;

[0086] Understandably, the stress-bearing area is the core sensitive area for sample failure. For example, in a water pressure test, the sealing surface is subjected to 0.5 MPa water pressure. Micron-level cutting defects can cause leakage and affect the experimental results. Therefore, the priority of forcibly marking the stress-bearing area as the key mesh unit 21 is higher than that of binary classification, to ensure that these areas are always cut with high precision and to avoid the special working conditions being missed by the general threshold.

[0087] Step S144: Remove isolated single-point key mesh cells 11;

[0088] Specifically, in step S144, the isolated single-point critical mesh cell 11 is a single critical mesh cell 21 in which a predetermined number of consecutive mesh cells in the cutting path direction are all of non-critical type.

[0089] In implementation, the cutting path is discretized into a point column based on the grid cells, specifically as follows:

[0090] P k k = 1, 2, ..., N;

[0091] Traversing point sequence P k All points, if P k The corresponding grid cell is a critical grid cell, but the first n grid cells and the last n grid cells in the cutting path direction are all non-critical grid cells. Then, the critical grid cell is determined to be the isolated single-point critical grid cell, and the grid cell is remarked as a non-critical grid cell. Here, n≥10, preferably, n is 12.

[0092] In this embodiment, for the isolated single-point critical mesh cell 11, in the cutting path direction, there are a first front mesh cell a1, a second front mesh cell a2, a third front mesh cell a3, a first rear mesh cell b1, a second rear mesh cell b2, and a third rear mesh cell b3, all of which are non-critical mesh cells 31. Therefore, the mesh cell is determined to be an isolated single-point critical mesh cell 11, and the mesh cell is re-marked as a non-critical mesh cell 31.

[0093] It is understandable that isolated single-point critical meshes are usually data noise rather than real risk areas, and they will interfere with the merging of critical path segment 22. If isolated single-point critical meshes are not removed, it will lead to frequent power switching and heat accumulation during cutting, which is not conducive to the path optimization of laser cutting.

[0094] Step S145: Merge adjacent grid cells of the same type to form continuous critical path segments 22 and non-critical path segments 32.

[0095] Specifically, step S145 includes:

[0096] Step S1451: Perform morphological closing operation on the key mesh element 21 to fill the gaps;

[0097] In practice, the morphological closing operation is as follows: first, the critical mesh element 21 is expanded in the cutting path direction (preferably by 3 to 5 mesh elements of the expanded boundary), and then the cutting path direction is eroded (the number of mesh elements of the contracted boundary and the expanded boundary are the same).

[0098] Understandably, this operation is similar to "bridge construction." During the laser cutting of waterproof materials, due to the accuracy limitations of the thickness measurement equipment, false gaps may appear between the critical mesh units 21. These gaps formed by non-critical mesh units 31 will result in a large number of discontinuous critical path segments 22 and non-critical path segments 32, making subsequent steps difficult to implement. First, an expansion operation is performed to extend the boundary of the critical mesh unit 21 outward by a certain number of mesh units, so that the false gaps are permanently closed. Then, an erosion operation is performed to restore the original range of the critical mesh unit 21, ensuring the continuity of the critical path segment 22 and non-critical path segment 32.

[0099] Step S1452: Perform connected component analysis on the key grid cell 21 and merge discrete segments with a spacing smaller than a preset merging threshold.

[0100] In implementation, all key grid cells 21 are identified and spacing is detected: if the spacing between two key grid cells 21 is less than or equal to a preset merging threshold, they are merged into a single continuous segment, wherein the preset merging threshold is preferably the length of two grid cells.

[0101] It is understandable that by processing the small gaps that still exist after morphological closing operations through connected component analysis (similar to hole filling in image processing), the frequent fluctuations in laser energy caused by parameter changes in the cutting laser due to the interruption of the critical path segment 22 can be avoided.

[0102] Step S1453: Output the topologically continuous closed path segments.

[0103] In implementation, the processed mesh is reorganized into continuous path segments according to type, and a closure check is performed to ensure that all path segments are connected end to end. The output is a vector path that can be executed by the laser cutting system, including critical path segments 22 and non-critical path segments 32.

[0104] Specifically, step S1 further includes:

[0105] Step S16: When the length of the longest non-critical path segment 32 in the non-critical path segment 32 is less than the minimum characterization length, a compensation cutting segment 41 is generated in the longest non-critical path segment of the target specimen.

[0106] In implementation, the lengths of all non-critical path segments 32 are calculated, and the longest segment is identified, with its length denoted as L. max If L max If the length is less than L0, a compensation cutting segment 41 and a deletion segment 42 are generated on the outside of the path segment. Here, L0 is the minimum characterization length in millimeters (mm). Preferably, L0 is 80 mm.

[0107] The calculation formula for the compensation cutting segment 41 is as follows:

[0108] L b =L0-L max +L d ,

[0109] Among them, L b To compensate for the length of the cut segment 41, the unit is millimeters (mm); L d This represents the length of segment 42 to be deleted, in millimeters (mm).

[0110] Step S17: The compensated cutting segment 41 is incorporated into the non-critical path segment 32 to form an extended non-critical path segment 32, so that the length of the extended non-critical path segment 32 meets the minimum representation length requirement.

[0111] In implementation, the compensation cutting segment 41 is connected to the original non-critical path segment 32, and the deleted segment 42 is deleted at the same time; the merged whole path is uniformly marked as the extended non-critical path segment 32.

[0112] In practice, after the laser cutting of the target specimen is completed, the deleted non-critical path segment 32 should be repeated to remove the extra part added by the extended non-critical path segment 32, which will not be described in detail here.

[0113] Understandably, the minimum characterization length is a critical value for the stability of cutting parameters based on historical experimental data. Below this length, the laser power / speed cannot reach a steady state, the data fluctuates greatly, and the data sample obtained is small. It is difficult to characterize the quality data of laser cutting in non-critical path segments 32. Therefore, it needs to be extended to ensure that the cutting quality data has statistical significance.

[0114] Step S2: Determine basic cutting parameters based on the basic size data of the target specimen, cut the non-critical path segment 32 of the target specimen using the basic cutting parameters, and monitor the cutting edge size data of the target specimen in real time.

[0115] During implementation, the basic cutting parameters are determined by referring to a table (see Table 1 below) based on the specimen thickness T and material type. These basic cutting parameters include the basic cutting speed and the basic laser power. The basic cutting speed is denoted as V0, with units of meters per minute (m / min); the basic laser power is denoted as P. 0， The unit is watts (W); non-critical path segment 32 is cut with P0 and V0, and the cutting edge size is monitored in real time using a laser displacement sensor. A measurement point is collected every 0.1 mm; the cutting edge size data includes the kerf width and the edge perpendicularity. The kerf width is denoted as W, and the unit is millimeters (mm); the edge perpendicularity is denoted as θ, and the unit is degrees (°).

[0116] Table 1 Basic Cutting Parameter Lookup Table

[0117]

[0118] It is understandable that the above basic cutting parameter lookup table is a parameter reference table summarized based on the characteristics of polymer waterproof membrane and industrial laser cutting practice. The data is based on industry experience, literature research and historical experimental calibration. Those skilled in the art can make adaptive adjustments according to the actual situation, which will not be elaborated here.

[0119] Step S3: Based on the comparison between the obtained cutting edge size data of the target specimen and the sample preparation preset data, the basic cutting parameters are adjusted to obtain the verification cutting parameters, and the laser cutting of the non-critical path segment 32 is completed using the verification cutting parameters.

[0120] Specifically, step S3 includes:

[0121] Step S31: Compare the real-time monitored cutting edge size data with the sample preparation preset data one by one, and calculate the corresponding size deviation values.

[0122] In practice, the sample preparation preset data for the kerf width W is W0, in millimeters (mm), preferably W0 is 0.8mm; the sample preparation preset data for the edge perpendicularity θ is θ0, in degrees (°), preferably θ0 is 90°.

[0123] Dimensional deviations include the kerf width deviation, denoted as △W, in millimeters (mm); and the edge perpendicularity deviation, denoted as △θ, in degrees (°).

[0124] ΔW=W-W0,

[0125] Δθ=θ-θ0.

[0126] Step S32: When a predetermined number of consecutive dimensional deviation values ​​exceed the allowable tolerance range, the basic cutting parameters are adjusted, wherein the basic cutting parameters are adjusted based on the direction and magnitude of the dimensional deviation values, and the basic cutting parameters include laser power and cutting speed.

[0127] In this embodiment, the allowable tolerance range for the kerf width deviation is △W≤±0.1mm; the allowable tolerance range for the edge perpendicularity deviation is △θ≤±2°.

[0128] In this embodiment, the adjustment rules for the basic cutting parameters are as follows:

[0129] When five consecutive data points satisfy ΔW > 0 and ΔW is greater than the allowable tolerance range, adjust the laser power, P. 新 =0.95×P0;

[0130] When five consecutive data points satisfy ΔW < 0 and ΔW is greater than the allowable tolerance range, adjust the laser power, P. 新 =1.05 × P0;

[0131] When five consecutive data points satisfy the condition that Δθ is greater than the allowable tolerance range, adjust the cutting speed, V. 新 =0.9×V0; where, P 新 For the adjusted laser power, V 新P0 represents the adjusted cutting speed, and V0 represents the laser power before adjustment.

[0132] Understandably, the core principle of dynamic parameter adjustment is based on the causal relationship between cutting quality defects and the laser energy input. When real-time monitoring detects deviations in the cutting edge dimensions, the system automatically analyzes the defect type and its physical causes, and adjusts the laser parameters accordingly: When the kerf is too wide (positive deviation), this is usually caused by excessive melting of the material due to excessive laser power. In principle, excessive energy input will cause the molten pool to expand, and the liquid material to flow to both sides, thus widening the kerf. At this time, the system reduces the laser power proportionally, reducing the molten pool size by reducing heat input. When the kerf is too narrow (negative deviation), this is usually caused by insufficient power leading to incomplete melting. When the energy is insufficient to completely vaporize the material, the melt cannot be fully discharged, forming a narrow kerf. At this time, the system increases the power accordingly to enhance the material vaporization ability and ensure that the kerf reaches the preset width. When the edge is tilted (perpendicularity deviation), the main cause is the mismatch of cutting speed. Too fast a speed will cause the laser beam to have insufficient residence time inside the material, forming asymmetrical heat conduction and producing a tilted cut. At this time, the system will reduce the cutting speed and extend the laser action time to make the heat distribution more uniform. Meanwhile, the continuous multi-point triggering mechanism (default 5 consecutive points out of tolerance) is used to distinguish between real defects and transient interference. The deviation of a single data point may be caused by local impurities in the material or detection noise, while the deviation of multiple consecutive points indicates that there is a systematic parameter imbalance. The continuous multi-point triggering mechanism ensures the timeliness of adjustment and avoids excessive parameter oscillation.

[0133] Step S33: Complete the cutting of the current non-critical path segment 32 using the adjusted verification cutting parameters, and continuously acquire the cutting edge size data of the target specimen and the corresponding relationship of the verification cutting parameters during the cutting process.

[0134] With new parameter P 新 and V 新 Continue laser cutting to obtain verified cutting parameters and establish a mapping table of cutting parameters and cutting quality (dimensional deviation value), that is, the correspondence between the cutting edge size data and the corresponding verified cutting parameters.

[0135] Step S4: Determine the first confidence path segment based on the cutting edge size data of the non-critical path segment 32, and extract the corresponding verification cutting parameters to form an acceptance dataset;

[0136] Specifically, step S4 includes:

[0137] Step S41: Filter out continuous segments in non-critical path segments 32 whose cutting edge size data are continuously within the preset tolerance zone, and mark them as the first confidence path segments;

[0138] In implementation, continuous segments in non-critical path segments 32 whose cutting edge dimension data are continuously within the preset tolerance zone are selected. When the dimension data of 50 consecutive measurement points (corresponding to a length of 5mm) are all within the tolerance zone, the continuous segment is marked as the first confidence path segment.

[0139] In this embodiment, the preset tolerance zone is △W≤±0.08mm for the kerf width deviation value and △θ≤±1° for the edge perpendicularity deviation value.

[0140] Understandably, only a sufficiently long (50 consecutive measurement points) continuous and stable cutting section can demonstrate that the cutting parameters used in that section are reliable and suitable for the material properties. By setting a minimum length requirement (5mm), the accidental acceptance of qualified points is avoided, ensuring that the parameters have statistical significance. For the preset tolerance zone, the allowable tolerance range that is stricter than the tolerance requirements of the target specimen size should be selected so that the cutting accuracy of the edge size data of the verification cutting parameters can meet the requirements and characterize the cutting parameters as the optimal energy input combination for the corresponding target specimen.

[0141] Step S42: Extract the set of verification cutting parameters in the first confidence path segment, denoted as the acceptance cutting parameters, and extract the cutting edge size data corresponding to the acceptance cutting parameters, denoted as the acceptance size data;

[0142] In implementation, for each first confidence path segment, the verification cutting parameters corresponding to that segment are extracted and recorded as the accepted cutting parameters, including laser power and cutting speed values, and recorded as the laser power-cutting speed combination; the cutting edge size data corresponding to the accepted cutting parameters are extracted and recorded as the accepted size data.

[0143] It is understandable that the cutting accuracy of the edge size data cut by the verification cutting parameters used in the first confidence path segment can meet the requirements, which can characterize that the cutting parameters are suitable for the specific material properties and are the optimal energy input combination for the corresponding target specimen.

[0144] Step S43: Store the acceptance cutting parameters and acceptance size data in spatial order to form an acceptance dataset, and record the mapping relationship between the acceptance cutting parameters and acceptance size data.

[0145] In implementation, the parameter set is stored in spatial order to form the acceptance dataset. Specifically, the data is arranged in the order of the cutting path. Each data unit contains: start coordinates, end coordinates, corresponding acceptance cutting parameters and corresponding acceptance size data. At the same time, the topological relationship is simplified. Adjacent first confidence path segments are automatically connected, and non-adjacent first confidence path segments retain the position mark.

[0146] Step S5: Based on the fluctuation of the verification cutting parameters in the accepted dataset, determine the correlation of the verification cutting parameters, obtain the key cutting parameters, and form a confidence dataset;

[0147] Specifically, step S5 includes:

[0148] Step S51: Divide each of the first confidence path segments according to the change of the acceptance cutting parameters to form a number of second confidence path segments;

[0149] In implementation, the first confidence path segment is finely segmented, specifically by scanning the laser power value and cutting speed value at 0.1mm intervals along the cutting path. When the laser power changes by ≥20% or the cutting speed changes by ≥30%, it is marked as a segmentation point. At the segmentation point, the original first confidence path segment is divided into a new sub-segment, namely the second confidence path segment.

[0150] Understandably, although the first confidence path segment is of acceptable quality overall, there may be minor parameter adjustments within it. By segmenting according to the points where parameters change, it is ensured that the cutting parameters within each second confidence path segment are not completely consistent, thus providing usable data samples for subsequent correlation analysis.

[0151] Step S52: Calculate the correlation coefficient between the acceptance cutting parameter and the corresponding acceptance size data in each second confidence path segment to determine the degree of correlation between each acceptance cutting parameter and the corresponding acceptance size data;

[0152] In practice, the average cutting speed and the average laser power in each second confidence path segment are calculated; the average kerf width deviation ΔW and the average perpendicularity deviation Δθ in each second confidence path segment are calculated; then the correlation between ΔW and laser power and the correlation between Δθ and cutting speed for each second confidence path are calculated.

[0153] In this embodiment, the correlation coefficient is specifically calculated using the Pearson correlation coefficient to obtain the correlation coefficient r1 between laser power and ΔW and the correlation coefficient r2 between cutting speed and Δθ.

[0154] It is understandable that the correlation degree represents the specific mechanism by which cutting parameters affect cutting quality. Specifically, when r1 > 0, power changes lead to a wider kerf, and when r1 < 0, power changes lead to a narrower kerf. When r2 > 0, increased speed leads to improved perpendicularity, and when r2 < 0, increased speed leads to increased tilting.

[0155] Step S53: Remove the confidence cut parameters whose correlation coefficients are less than a preset correlation coefficient threshold to obtain the key cut parameters and form the confidence dataset.

[0156] In implementation, the preset correlation coefficient threshold is 0.7. For the correlation between laser power and ΔW, r1, the cutting parameters corresponding to |r1|>0.7 are retained. For the correlation between cutting speed and Δθ, r2, the cutting parameters corresponding to r2>0.7 are retained. Other cutting parameters are removed. The cutting parameters with the highest correlation among the retained cutting parameters are compared and calculated to obtain the key cutting parameters, and then the confidence dataset is obtained.

[0157] Understandably, the correlation coefficient threshold of 0.7 is set based on the principle of statistical significance. Eliminating parameters with low correlation can avoid overfitting and ensure that the recommended parameters have universality.

[0158] Step S6: Laser cut the critical path segment 22 using the key cutting parameters in the confidence dataset to obtain the target specimen finished product;

[0159] Step S7: Establish a deep learning optimization model based on several confidence datasets, and output the basic cutting parameters for subsequent specimens.

[0160] Specifically, step S7 includes:

[0161] Step S71: Collect the confidence datasets and corresponding cut quality rating datasets of multiple sets of target specimen finished products;

[0162] During implementation, a complete confidence dataset containing at least 50 sets of historical cutting specimens is collected. Each dataset includes: optimized cutting parameters (laser power / cutting speed combination) and corresponding quality rating (pass / fail).

[0163] The specific criteria for grading cut quality are as follows:

[0164] Acceptable: Cut width ±0.02mm and perpendicularity ±0.3°;

[0165] Unacceptable: Any indicator exceeds the tolerance (cut width or perpendicularity exceeds the tolerance);

[0166] The data is stored as a structured database, with each record containing metadata such as material type and thickness.

[0167] Step S72: Construct a supervised learning model with the cutting parameters as input and the cutting quality rating as output;

[0168] In implementation, the model uses a standard multilayer perceptron (MLP) neural network. The input layer encodes the basic size data and material type; the output layer provides the recommended initial laser power and recommended initial cutting speed; the training process involves training with 70% historical data, tuning parameters on a 15% validation set, and evaluating with a 15% test set.

[0169] Step S73: When the basic size data of the new specimen is input, the optimized initial values ​​of the cutting parameters are output.

[0170] In this embodiment, when the basic size data of a new specimen is input, the initial value of the output optimized cutting parameters is used as the basic cutting parameters to cut the non-critical path segment 32, so as to continuously optimize the cutting path and the cutting parameters of the corresponding cutting path.

[0171] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for optimizing the laser cutting path of waterproof materials based on deep learning, characterized in that, include: Step S1: Obtain the basic size data, thickness change rate distribution data, and preloaded stress distribution data of the target specimen of the waterproof material, and determine the critical path segment and non-critical path segment of the cutting path of the target specimen. Step S2: Determine basic cutting parameters based on the basic size data of the target specimen, cut the non-critical path segment of the target specimen using the basic cutting parameters, and monitor the cutting edge size data of the target specimen in real time. Step S3: Based on the comparison between the obtained cutting edge size data of the target specimen and the sample preparation preset data, the basic cutting parameters are adjusted to obtain the verification cutting parameters, and the laser cutting of the non-critical path segment is completed using the verification cutting parameters. Step S4: Determine the first confidence path segment based on the cutting edge size data of the non-critical path segment, and extract the corresponding verification cutting parameters to form an accepted dataset; Step S5: Based on the fluctuation of the verification cutting parameters in the accepted dataset, determine the correlation of the verification cutting parameters, obtain the key cutting parameters, and form a confidence dataset; Step S6: Laser cut the critical path segment using the key cutting parameters in the confidence dataset to obtain the target specimen finished product; Step S7: Establish a deep learning optimization model based on several confidence datasets, and output the basic cutting parameters for subsequent specimens.

2. The method for optimizing the laser cutting path of waterproof materials based on deep learning according to claim 1, characterized in that, Step S1 includes: Step S11: Obtain the basic dimensional data of the target specimen, including the cutting path and thickness reference value; Step S12: Obtain the thickness change rate distribution data of the sample containing the target specimen; Step S13: Based on the testing conditions of the target specimen and the basic size data, perform finite element analysis to generate the preloaded stress distribution data; Step S14: Based on the thickness change rate distribution data and the preloaded stress distribution data, determine the critical path segments and non-critical path segments of the cutting path.

3. The method for optimizing the laser cutting path of waterproof materials based on deep learning according to claim 2, characterized in that, Step S14 includes: Step S141: Discretize the cutting path into several grid cells, wherein the grid resolution of each grid cell is set according to the thickness reference value. Step S142: Perform binary classification on each mesh element. If the judgment conditions of thickness change rate threshold or stress threshold are met, it is marked as a critical mesh element; otherwise, it is marked as a non-critical mesh element. Step S143: Based on the testing conditions of the target specimen, forcibly mark the mesh elements corresponding to the stress-bearing area as critical mesh elements; Step S144: Remove isolated single-point critical mesh cells; Step S145: Merge adjacent grid cells of the same type to form continuous critical path segments and non-critical path segments.

4. The method for optimizing the laser cutting path of waterproof materials based on deep learning according to claim 3, characterized in that, Step S145 includes: Step S1451: Perform morphological closing operation on the key mesh elements to fill the gaps; Step S1452: Perform connected component analysis on key grid cells and merge discrete segments with spacing less than a preset merging threshold. Step S1453: Output the topologically continuous closed path segments.

5. The method for optimizing the laser cutting path of waterproof materials based on deep learning according to claim 3, characterized in that, In step S144, the isolated single-point critical mesh cell is a single critical mesh cell in which a predetermined number of consecutive mesh cells in the cutting path direction are all of non-critical type.

6. The method for optimizing the laser cutting path of waterproof materials based on deep learning according to claim 2 or 4, characterized in that, Step S1 further includes: Step S15: When the length of the longest non-critical path segment in the non-critical path segments is less than the minimum characterization length, a compensation cutting segment is generated in the longest non-critical path segment of the target specimen. Step S16: The compensated cutting segment is incorporated into the non-critical path segment to form an extended non-critical path segment, so that the length of the extended non-critical path segment meets the minimum representation length requirement.

7. The method for optimizing the laser cutting path of waterproof materials based on deep learning according to claim 4 or 5, characterized in that, Step S3 includes: Step S31: Compare the real-time monitored cutting edge size data with the sample preparation preset data one by one, and calculate the corresponding size deviation values. Step S32: When a predetermined number of consecutive dimensional deviation values ​​exceed the allowable tolerance range, the basic cutting parameters are adjusted, wherein the basic cutting parameters are adjusted based on the direction and magnitude of the dimensional deviation values, and the basic cutting parameters include laser power and cutting speed. Step S33: Complete the cutting of the current non-critical path segment using the adjusted verification cutting parameters, and continuously acquire the cutting edge size data of the target specimen and the corresponding relationship of the verification cutting parameters during the cutting process.

8. The method for optimizing the laser cutting path of waterproof materials based on deep learning according to claim 6, characterized in that, Step S4 includes: Step S41: Filter out continuous segments in the non-critical path segments where the cutting edge size data is continuously within the preset tolerance zone, and mark them as the first confidence path segments; Step S42: Extract the set of verification cutting parameters in the first confidence path segment, denoted as the acceptance cutting parameters, and extract the cutting edge size data corresponding to the acceptance cutting parameters, denoted as the acceptance size data; Step S43: Store the acceptance cutting parameters and acceptance size data in spatial order to form an acceptance dataset, and record the mapping relationship between the acceptance cutting parameters and acceptance size data.

9. The method for optimizing the laser cutting path of waterproof materials based on deep learning according to claim 7, characterized in that, Step S5 includes: Step S51: Divide each of the first confidence path segments according to the change of the acceptance cutting parameters to form a number of second confidence path segments; Step S52: Calculate the correlation coefficient between the acceptance cutting parameter and the corresponding acceptance size data in each second confidence path segment to determine the degree of correlation between each acceptance cutting parameter and the corresponding acceptance size data; Step S53: Remove the confidence cut parameters whose correlation coefficients are less than a preset correlation coefficient threshold to obtain the key cut parameters and form the confidence dataset.

10. The method for optimizing the laser cutting path of waterproof materials based on deep learning according to claim 9, characterized in that, Step S7 includes: Step S71: Collect the confidence datasets and corresponding cutting quality rating datasets of multiple sets of target specimen finished products; Step S72: Construct a supervised learning model with the cutting parameters as input and the cutting quality rating as output; Step S73: When the basic size data of the new specimen is input, the optimized initial values ​​of the cutting parameters are output.

Images