A mold air tightness detection parameter optimization method

Abstract

The application provides a mold air tightness detection parameter optimization method, and belongs to the technical field of parameter optimization. The method comprises the following steps: simulating the detection of a physical model of a target mold, and simulating the optimization and actual use of the physical model; dividing and counting the running time ratio of each working condition according to the high-temperature forming condition, the normal-temperature storage condition and the high-pressure test condition of the actual use log of the target mold, simultaneously, synchronously collecting the multi-modal data of each working condition and performing redundancy processing; based on the running time ratio of each working condition and the multi-modal data after redundancy, a multi-working-condition collaborative optimization objective function is constructed, and the multi-working-condition collaborative optimization function is subjected to objective optimization; the detection threshold of the multi-working-condition collaborative optimization objective function is reversely updated according to the optimization result, the rationality of the detection parameter setting and the accuracy of the detection result are ensured, the reliability of the mold air tightness detection is improved, and the cross-working-condition self-adaptive optimization of the mold air tightness detection parameter is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of parameter optimization, in particular to a mold air tightness detection parameter optimization method. BACKGROUND

[0002] Mold air tightness detection is a core link for guaranteeing product quality in high-end fields such as aerospace and automobile manufacturing (for example, leakage of an aero-engine case mold will cause thrust loss, and leakage of an automobile cylinder body mold will cause cooling liquid leakage). However, the traditional air tightness detection adopts fixed parameters (such as uniform pressure 10 MPa and pressure maintaining time 60 s), without distinguishing the differences in working conditions such as high-temperature forming (above 150℃, plastic melt injection causes thermal deformation and creep), normal-temperature storage (long-term accumulation of micro-leakage), and high-pressure testing (extreme pressure above 50 MPa causes plastic damage): in the high-temperature forming working condition, thermal expansion of the mold changes the sealing gap, and fixed pressure is easy to misjudge thermal deformation leakage as real leakage; in the high-pressure testing working condition, permanent gap caused by plastic deformation of the mold is detected as continuous leakage by static parameter detection, increasing the misjudgment rate (the misjudgment rate of a certain aero-mold detection reaches 12%). And although the existing method collects pressure, temperature, and ultrasonic multi-modal data, there is a problem of data redundancy, that is, multi-modal data is not subjected to working condition-specific feature screening (for example, temperature data of the normal-temperature storage working condition is greatly affected by environmental interference, but is processed equally with pressure data), which increases the proportion of invalid features and slows down the optimization efficiency, resulting in unreasonable detection parameter setting, inaccurate detection results, and low efficiency, thereby affecting the detection of mold air tightness.

[0003] Therefore, the present application provides a mold air tightness detection parameter optimization method. SUMMARY

[0004] The present application provides a mold air tightness detection parameter optimization method to solve the above technical problems.

[0005] The present application provides a mold air tightness detection parameter optimization method, comprising:

[0006] Step 1: Simulate detection on a physical model of a target mold, and simulate optimization and actual use of the physical model;

[0007] Step 2: Divide the actual use log of the target mold according to high-temperature forming working condition, normal-temperature storage working condition, and high-pressure testing working condition, and statistically analyze the running time ratio of each working condition, and simultaneously collect multi-modal data of each working condition;

[0008] Step 3: Based on the running time ratio of each working condition and the multi-modal data, a multi-working condition cooperative optimization objective function is constructed, and the multi-working condition cooperative optimization function is subjected to objective optimization;

[0009] Step 4: According to the optimization result, the detection threshold of the multi-working condition collaborative optimization objective function is updated reversely.

[0010] Preferably, the physical model of the target mold is simulated and detected, and the physical model is simulated and optimized, comprising:

[0011] According to the attribute characteristics of the target mold, an attribute vector is constructed, wherein the attribute characteristics are related to the material type, cavity complexity, dimensional tolerance, and surface roughness of the mold;

[0012] The similarity of the attribute vector and each scheme is calculated from the attribute-scheme database and combined with the cosine similarity algorithm, a candidate air tightness scheme set is screened, and the physical model of the target mold is simulated and detected in turn according to each candidate air tightness scheme in the candidate air tightness scheme set, and an initial simulation matrix is constructed;

[0013] An ideal detection vector of the historical optimal data of each candidate air tightness scheme is obtained, and an ideal detection matrix is constructed;

[0014] Based on the initial simulation matrix and the ideal detection matrix, and combined with the historical precision decay rate of each candidate air tightness scheme, the air tightness coefficient of the target mold under the corresponding candidate air tightness scheme is determined;

[0015] The optimization direction of the physical model is determined according to the difference detection matrix of the initial simulation matrix and the ideal detection matrix, and the optimization parameters of the physical model are determined combined with the air tightness coefficient and the historical precision decay rate, wherein the feature dimensions are related to the pressure decay feature, the ultrasonic echo feature, and the helium leakage feature;

[0016] The physical model is simulated and optimized according to the optimization parameters.

[0017] Preferably, the air tightness coefficient of the target mold under the corresponding candidate air tightness scheme is determined, comprising:

[0018] The historical detection precision sequence of the kth candidate air tightness scheme is extracted The corresponding historical precision decay rate is calculated , wherein represents the historical detection precision of the kth candidate air tightness scheme under the ith detection; m represents the historical detection times of the kth candidate air tightness scheme; represents the minimum value of all ; represents the maximum value of all ;

[0019] The air tightness coefficient of the kth candidate air tightness scheme is calculated combined with the Euclidean distance difference of the initial simulation matrix and the ideal detection matrix and the precision decay rate ;

[0020]

[0021] wherein, , 2 respectively represent weights, and ; represents the normalized row vector of the kth candidate airtightness scheme based on the initial simulation matrix; represents the normalized row vector of the kth candidate airtightness scheme based on the ideal detection matrix.

[0022] Preferably, the optimization parameter of the physical model is determined, comprising:

[0023] According to the column vector of each feature dimension in the difference detection matrix, the optimization direction of the physical model is determined;

[0024] According to the detection side direction of each candidate airtightness scheme, the first scheme matching the optimization direction is screened from all candidate airtightness schemes, and the airtightness coefficient of the first scheme is sorted from small to large;

[0025] According to the sorting result, starting from the minimum airtightness coefficient, the parameter optimization strategy corresponding to the minimum airtightness coefficient is coded into a knowledge vector and combined with the corresponding precision decay rate to determine the optimization parameter of the physical model.

[0026] Preferably, the multi-modal data of each working condition is synchronously collected and redundantly processed, comprising:

[0027] All initial features are extracted from the multi-modal data of each working condition to form an original feature set of the corresponding working condition wherein, respectively represent the 1st initial feature, the 2nd initial feature, and the n1th initial feature based on the multi-modal data;

[0028] From the original feature set, a relevant feature subset H of the airtightness detection target is screened wherein, respectively represent the 1st screening feature, the 2nd screening feature, and the n2th screening feature, and the information gain rate of each screening feature is calculated wherein, is the entropy of the original data set; represents the conditional entropy of the j2th screening feature in H after division, and j2 takes the value of 1, 2, 3,..., n2;

[0029] If the information gain rate of the j2th screening feature is less than the corresponding experimental calibration threshold, the j2th screening feature is removed from H ​The corresponding screening feature is given a first rejection label;

[0030] An equipment working log of the acquisition equipment corresponding to each screening feature is obtained, a start communication field of the acquisition equipment and a first field attached to the screening feature are extracted;

[0031] The state influence relationship between the start communication field and the first field is analyzed according to the communication type of the acquisition equipment, and preset thermal data corresponding to the working condition are obtained, and the mapping dependency relationship between each screening feature and the preset thermal data is determined respectively, wherein the preset thermal data is a direction of focus under the corresponding working condition;

[0032] Based on the state influence relationship and the mapping dependency relationship, it is determined whether to give the screening feature a second rejection label;

[0033] When the corresponding screening feature is given a first rejection label and a second rejection label, the corresponding screening feature is redundantly rejected.

[0034] Preferably, based on the running time ratio of each working condition and the multi-modal data, a multi-working condition collaborative optimization objective function is constructed, including:

[0035] A multi-physical field coupling model is constructed , wherein the input layer of the single-working condition coupling model is constructed based on the fusion of the core features of the corresponding working condition according to the coupling mechanism, and the output layer is constructed by the predicted value of the air tightness index under the corresponding working condition, wherein, , , , , respectively represent the leakage amount, the detection time, and the damage value; is a detection parameter vector, and represents the ith detection parameter, and n01 represents the number of detection parameters; is a model structure parameter vector, and represents the ith model structure parameter, and n02 represents the number of model structure parameters; is a model output, and ;

[0036] The contribution of the core feature of the multi-modal data after redundancy of each working condition to the corresponding working condition is quantified, and the cross-working condition contribution of the multi-modal feature is integrated, and the importance weight of the working condition is combined to construct a multi-working condition collaborative optimization objective function: , wherein, , , respectively represent the weights of the leakage amount, the detection time, and the damage value; represents the importance weight of the jth working condition determined by the running time ratio; , , respectively represent the actual leakage amount, actual detection time, and actual damage value under the j3th working condition; , , respectively represent the ideal leakage amount, ideal detection time, and ideal damage value under the j3th working condition;

[0037] finding the optimal parameter combination of the multi-working condition collaborative optimization objective function .

[0038] Preferably, the pressure decay feature is related to the detection pressure, air pressure holding time, and pressure rise rate;

[0039] The ultrasonic echo feature is related to the ultrasonic frequency, coupling agent thickness, and focusing depth;

[0040] The helium leakage feature is related to the helium concentration, vacuum pumping time, and mass spectrum scanning frequency.

[0041] Preferably, according to the optimization result, the working condition importance ranking verification and the reverse update of the detection threshold of the multi-working condition collaborative optimization objective function, comprising:

[0042] Receiving the optimization result of the multi-working condition collaborative optimization objective function, extracting the optimal detection parameter combination under each working condition and the corresponding objective function value;

[0043] Input the actual detection data under the optimization parameters into the DBSCAN clustering algorithm to identify abnormal data clusters, and calculate the sensitivity of each objective function threshold to the detection result;

[0044] If the detection error of the corresponding working condition exceeds the corresponding threshold for ne consecutive times, trigger the threshold fine-tuning based on Kalman filtering;

[0045] At the same time, after every nn detection period, the weight matrix of the multi-working condition collaborative optimization objective function is corrected, and a threshold correlation matrix is established. When there is working condition threshold update, the threshold of the associated working condition is adjusted synchronously through the threshold correlation matrix.

[0046] Compared with the prior art, the beneficial effects of the present application are as follows:

[0047] Through the whole process design of physical model iteration → working condition data driving → honor processing → multi-objective optimization → threshold closed-loop update, the proportion of invalid features is reduced, the rationality of the detection parameter setting and the accuracy of the detection result are guaranteed, the reliability of the mold air tightness detection is improved, and the cross-working condition adaptive optimization of the mold air tightness detection parameters is realized.

[0048] Additional features and advantages of the present application will be set forth in the description that follows, and in part will be apparent from the description, or can be learned by practice of the application. The objectives and other advantages of the present application will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

[0049] The technical solutions of the present application are described in further detail below with reference to the accompanying drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0050] The accompanying drawings are included to provide a further understanding of the present application, and are incorporated in and constitute a part of the specification, illustrate embodiments of the present application and explain the present application, but do not limit the present application. In the drawings:

[0051] Figure 1 The flow chart of the mold air tightness detection parameter optimization method in the embodiment of the present application. DETAILED DESCRIPTION

[0052] The preferred embodiments of the present application are described below with reference to the accompanying drawings, and it should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and do not limit the present application.

[0053] The present application provides a mold air tightness detection parameter optimization method, as shown in Figure 1 , comprising:

[0054] Step 1: Simulate the detection of the physical model of the target mold, and simulate the optimization and actual use of the physical model;

[0055] Step 2: Divide the actual use log of the target mold according to high-temperature forming conditions, normal-temperature storage conditions, and high-pressure test conditions, and count the running time ratio of each condition, while synchronously collecting multi-modal data of each condition;

[0056] Step 3: Based on the running time ratio of each condition and the multi-modal data, a multi-condition collaborative optimization objective function is constructed, and the multi-condition collaborative optimization function is optimized;

[0057] Step 4: According to the optimization result, the detection threshold of the multi-condition collaborative optimization objective function is updated in reverse.

[0058] In this embodiment, the target mold refers to an industrial mold (such as an aero-engine casing mold or an automobile cylinder body mold) to be detected for air tightness. Taking an aluminum alloy automobile cylinder body mold as an example, the complexity of its cavity is 6 levels (industry classification 1-10, the larger the value, the more complex), the size tolerance is ±0.05mm, and the surface roughness Ra=1.6μm.

[0059] The physical model is built by ANSYS TwinBuilder to simulate the gas leakage, thermal deformation and structural damage of the mold under pressure and temperature loads, integrating the gas flow equation (Navier-Stokes), heat conduction equation (Fourier's law) and structural mechanics equation (elastic-plastic constitutive).

[0060] The simulation test is preset with initial parameters (test pressure 10 MPa, holding time 60 s), and the pressure decay curve, leakage distribution and mold equivalent plastic strain are simulated. For example, if the standard deviation of pressure fluctuation reaches 0.5 kPa (exceeding the standard 0.3 kPa), the simulation test is determined to be failed.

[0061] The simulation optimization is to adjust the holding time (e.g. shortened to 45 s) and optimize the grid accuracy of the sealing surface (from 1 mm to 0.5 mm) to re-simulate until the pressure fluctuation is less than or equal to 0.3 kPa, .

[0062] The actual use is to load the optimized parameters into the air tightness detection equipment (such as helium mass spectrometry leak detection system) to perform detection on the real mold, record the actual leakage, detection time, mold surface temperature rise and other data, and form a simulation-actual measurement feedback loop.

[0063] In this embodiment, the actual use log is the historical data recorded by the mold management system, including the timestamp, temperature, pressure, and leakage result of each detection. Taking the monthly use log of a certain mold as an example: high temperature forming condition (150℃, 30MPa) accounts for 60% (300 hours), normal temperature storage condition (25℃, normal pressure) accounts for 25% (125 hours), and high pressure test condition (25℃, 50MPa) accounts for 15% (75 hours).

[0064] Condition division:

[0065] High temperature forming condition: production condition of mold in service (temperature≥80℃, pressure 5-50MPa), which needs to simulate the thermal-mechanical coupling leakage when plastic melt is injected;

[0066] Normal temperature storage condition: storage condition of mold when idle (10℃≤temperature≤40℃, normal pressure), which needs to detect the micro-leakage during long-term storage;

[0067] High pressure test condition: extreme condition of factory inspection (pressure≥10MPa, normal temperature), which verifies the limit sealing capacity.

[0068] Multi-modal data: synchronously collect three types of data:

[0069] Pressure data: adopt differential pressure sensor with precision 0.01 kPa to record pressure fluctuation curve;

[0070] Temperature data: Infrared thermal imager (temperature resolution 0.1℃) scans the mold surface to generate temperature field distribution;

[0071] Ultrasonic data: 5MHz phased array ultrasonic transducer detects internal gaps in the mold, extracts echo amplitude, frequency offset characteristics.

[0072] In this embodiment, the runtime ratio is used as the working condition importance weight, and the high-temperature forming weight is the highest, which prioritizes the detection accuracy.

[0073] The multi-modal data is the data collected in step 2, and the core features such as pressure fluctuation standard deviation, temperature gradient, and ultrasonic echo amplitude are extracted, and the principal component analysis (PCA) is used for dimensionality reduction, and the features with a contribution degree of ≥80% are retained (such as pressure standard deviation and ultrasonic amplitude).

[0074] The multi-condition cooperative optimization objective function is realized based on the integration of leakage error, detection time, damage, and working condition weight.

[0075] The improved firefly algorithm is used for target optimization, and the detection parameters are divided into air pressure population (pressure holding time) and ultrasonic population (frequency, coupling agent thickness) for independent evolution, and the knowledge distillation mechanism (optimal population transfer strategy) is introduced, and the optimal parameters are output after 50 iterations.

[0076] In this embodiment, the optimization result is, for example, after optimization of the high-pressure test working condition, the optimal parameters are helium concentration 5%, vacuum time 30s, and the actual detection error is stable at 2.5% (lower than the original threshold of 5%).

[0077] The detection threshold is the original high-pressure working condition leakage error threshold of 5%, based on the optimization result and K-means++ clustering analysis (abnormal data ratio <5%), the threshold is adjusted to 3%; at the same time, through the threshold correlation matrix (high-pressure and high-temperature working condition pressure correlation degree 0.7), the pressure fluctuation threshold of high-temperature working condition is tightened from ±0.5kPa to ±0.3kPa.

[0078] The reverse update is to dynamically adjust the threshold using Kalman filter (updated every 10 detections), and to optimize the objective function weight (such as increasing the high-pressure working condition damage weight from 0.2 to 0.3) using reinforcement learning (DQN), to ensure that the threshold dynamically adapts to the actual performance of the mold.

[0079] The beneficial effects of the above technical solutions are: through the whole process design of physical model iteration → working condition data driving → honor processing → multi-objective optimization → threshold closed-loop update, the proportion of invalid features is reduced, the rationality of detection parameter setting and the accuracy of detection result are guaranteed, the reliability of mold air tightness detection is improved, and the cross-condition self-adaptive optimization of mold air tightness detection parameters is realized.

[0080] The application provides a mold air tightness detection parameter optimization method, which comprises the following steps of: performing simulation detection on a physical model of a target mold, and performing simulation optimization on the physical model, wherein the simulation detection comprises the following steps of: obtaining attribute characteristics of the target mold, and constructing an attribute vector according to the attribute characteristics, wherein the attribute characteristics are related to a material type, cavity complexity, size tolerance and surface roughness of the mold; obtaining a scheme database according to the attribute vector, and calculating a similarity between the attribute vector and each scheme in the scheme database to obtain a candidate air tightness scheme set; performing simulation detection on the physical model of the target mold according to each candidate air tightness scheme in the candidate air tightness scheme set to construct an initial simulation matrix; obtaining an ideal detection vector of historical optimal data of each candidate air tightness scheme to construct an ideal detection matrix; determining an air tightness coefficient of the target mold under a corresponding candidate air tightness scheme based on the initial simulation matrix and the ideal detection matrix and in combination with a historical precision decay rate of each candidate air tightness scheme; determining an optimization direction of the physical model according to a difference detection matrix of the initial simulation matrix and the ideal detection matrix, and determining an optimization parameter of the physical model in combination with the air tightness coefficient and the historical precision decay rate, wherein the feature dimension is related to a pressure decay feature, an ultrasonic echo feature and a helium leakage feature; and performing simulation optimization on the physical model according to the optimization parameter.

[0081] According to the attribute characteristics of the target mold and constructing an attribute vector, wherein the attribute characteristics are related to the material type, the cavity complexity, the size tolerance and the surface roughness of the mold;

[0082] The similarity between the attribute vector and each scheme is calculated from the attribute-scheme database and in combination with a cosine similarity algorithm, a candidate air tightness scheme set is screened, and the physical model of the target mold is sequentially simulated and detected according to each candidate air tightness scheme in the candidate air tightness scheme set to construct an initial simulation matrix;

[0083] The ideal detection vector of the historical optimal data of each candidate air tightness scheme is obtained to construct an ideal detection matrix;

[0084] Based on the initial simulation matrix and the ideal detection matrix, and in combination with the historical precision decay rate of each candidate air tightness scheme, the air tightness coefficient of the target mold under the corresponding candidate air tightness scheme is determined;

[0085] The optimization direction of the physical model is determined according to the difference detection matrix of the initial simulation matrix and the ideal detection matrix, and the optimization parameter of the physical model is determined in combination with the air tightness coefficient and the historical precision decay rate, wherein the feature dimension is related to the pressure decay feature, the ultrasonic echo feature and the helium leakage feature;

[0086] The physical model is simulated and optimized according to the optimization parameter.

[0087] Preferably, the pressure decay feature is related to the detection pressure, the air pressure retention time and the pressure rising rate;

[0088] The ultrasonic echo feature is related to the ultrasonic frequency, the coupling agent thickness and the focusing depth;

[0089] The helium leakage feature is related to the helium concentration, the vacuum pumping time and the mass spectrum scanning frequency.

[0090] In this embodiment, the attribute characteristics are the core physical attributes of the focusing mold, including a material type (such as titanium alloy commonly used in aviation molds, coded as 1; aluminum alloy commonly used in automobile molds, coded as 2), cavity complexity (industry classification 1-10 levels, such as the cavity complexity of the engine cylinder body mold is 8 levels because it contains a multi-depth cavity thin-wall structure), size tolerance (such as the tolerance of a certain precision mold is ±0.03 mm, and the absolute value 0.03 is directly taken), and surface roughness (measured by the arithmetic average deviation Ra, such as the optical mold Ra=0.8 μm).

[0091] Attribute vector construction is to quantify the above attributes into a four-dimensional vector, such as titanium alloy, cavity complexity 8, tolerance 0.03, Ra 0.8, corresponding vector [1, 8, 0.03, 0.8], through standardization (such as tolerance divided by 0.1, roughness divided by 1.6) to unify the dimension, which is convenient for subsequent similarity calculation.

[0092] In this embodiment, the attribute-scheme database pre-stores the applicable attribute intervals of various air tightness schemes, for example:

[0093] Air pressure detection scheme: adapt to high material stiffness (such as steel, code 1), cavity complexity ≤6, tolerance ≥±0.1mm, Ra≥3.2μm;

[0094] Helium mass spectrometry detection scheme: adapt to high precision (tolerance ≤±0.05mm), smooth surface (Ra≤1.6μm), cavity complexity ≤8;

[0095] Ultrasonic detection scheme: adapt to non-metallic mold (such as composite material, code 3), cavity complexity ≥5 (need to penetrate thick wall).

[0096] Cosine similarity calculation: for the target vector (such as [1, 8, 0.03, 0.8]) and the scheme vector (such as helium mass spectrometry scheme vector [1, 7, 0.04, 1.2]) in the database, calculate the similarity according to the cosine similarity formula, select the scheme ≥0.8 (such as helium mass spectrometry and ultrasonic detection, exclude air pressure detection because the roughness does not meet the requirements), form a candidate air tightness scheme set.

[0097] In this embodiment, the physical model simulation detection is to use ANSYSTwinBuilder to build a mold multi-physical field model, and simulate the detection for each candidate scheme:

[0098] The helium mass spectrometry scheme is to input helium concentration 5%, vacuum pumping time 30s, simulate the output pressure decay curve (standard deviation 0.2kPa), helium leakage amount ( ), detection time (2.5 minutes);

[0099] The ultrasonic scheme is to input frequency 5MHz, coupling agent thickness 0.5mm, simulate the output echo amplitude (80dB), frequency offset (200Hz), focusing depth error (0.1mm).

[0100] The initial simulation matrix is to arrange the multi-features of each scheme into a matrix, with rows representing schemes and columns representing features.

[0101] The ideal detection matrix is to extract the historical optimal data of the candidate schemes to construct an ideal feature vector, which is combined into an ideal detection matrix.

[0102] Historical precision decay rate refers to the rate at which the detection precision of a certain detection scheme (including the equipment used, the operation process, etc.) gradually decreases with the use time, the number of detections, or the change of the environment in long-term application, which is used to quantify the stability decay trend.

[0103] The air tightness coefficient is calculated by quantifying the difference between the initial and ideal matrices (such as a helium mass spectrometry scheme difference of 0.3) combined with the weight, and the smaller the coefficient, the better the scheme.

[0104] The difference detection matrix is the initial minus the ideal matrix, and the column vector (such as a pressure decay standard deviation difference of 0.1 kPa) indicates the optimization direction (the pressure fluctuation needs to be reduced, and the pressure holding time needs to be optimized);

[0105] The scheme screening and strategy migration, such as screening the helium mass spectrometry scheme with the smallest air tightness coefficient, extracting its parameter optimization strategy (such as vacuumizing time from 30 s to 25 s, encoded as a knowledge vector), combining the decay rate (which needs to improve the precision stability), and determining the final optimization parameters (helium concentration 5.5%, vacuumizing time 25 s, and ultrasonic frequency adjusted to 6 MHz for compensation).

[0106] In this embodiment, the pressure decay characteristics are related to the detection pressure (high pressure accelerates leakage, such as from 10 MPa to 12 MPa, and the pressure decay is more significant), the pressure holding time (from 60 s to 45 s, shortening the invalid pressure holding), and the pressure increasing rate (from 0.5 MPa / s to 0.8 MPa / s, quickly reaching the pressure to reduce thermal disturbance);

[0107] The ultrasonic echo characteristics are related to the ultrasonic frequency (5 MHz to 6 MHz, improving the resolution but reducing the penetration, suitable for thin-walled molds), the coupling agent thickness (0.5 mm to 0.3 mm, reducing signal attenuation), and the focusing depth (5 mm to 4.5 mm, accurately positioning the leakage point);

[0108] The helium leakage characteristics are related to the helium concentration (5% to 5.5%, improving the detection sensitivity), the vacuumizing time (30 s to 25 s, balancing the efficiency and vacuum degree), and the mass spectrometry scanning frequency (100 Hz to 120 Hz, speeding up data acquisition).

[0109] The beneficial effects of the above technical scheme are: through the closed-loop design of mold attribute precise matching scheme, multi-modal simulation quantitative difference, historical data driven coefficient optimization, and feature correlation directional parameter adjustment, the scheme-level adaptation and precise optimization of the air tightness detection parameters are realized, the detection performance and mold life are effectively balanced, and the fine quality control of high-end molds is supported.

[0110] The present application provides a mold air tightness detection parameter optimization method, which determines the air tightness coefficient of the target mold under the corresponding candidate air tightness scheme, comprising:

[0111] extract the historical detection accuracy sequence of the kth candidate air-tightness scheme , calculate the corresponding historical accuracy decay rate , wherein represents the historical detection accuracy of the kth candidate air-tightness scheme under the ith detection; m represents the historical detection times of the kth candidate air-tightness scheme; represents the minimum value in all ; represents the maximum value in all ;

[0112] In combination with the Euclidean distance difference of the initial simulation matrix and the ideal detection matrix and the accuracy decay rate, calculate the air-tightness coefficient of the kth candidate air-tightness scheme ;

[0113]

[0114] , wherein , 2 respectively represent weights, and ; represents the normalized row vector of the kth candidate air-tightness scheme based on the initial simulation matrix; represents the normalized row vector of the kth candidate air-tightness scheme based on the ideal detection matrix.

[0115] In this embodiment, focuses on the static deviation of the current simulation vs. the ideal state; focuses on the dynamic trend of the accuracy degradation of the historical data, assuming that the mold is newly designed / just put into production, the historical detection times are few, and the statistical reliability of the historical accuracy decay rate is low, the ideal detection matrix is based on the theoretical optimal model to construct the reference of the current simulation result, at this time, = 0.7, = 0.3.

[0116] Assuming that the mold is in long-term service, the historical detection times are many, and the historical accuracy decay rate can stably reflect the decay law of the precision interval, the deviation of the initial simulation matrix from the ideal state is more significant due to the aging and wear of the mold, and the influence of historical degradation, at this time, 1 = 0.3, = 0.7, it should be noted that only is required.

[0117] In this embodiment, the kth candidate air-tightness scheme refers to the detection method (such as air pressure detection, helium mass spectrometry detection, ultrasonic detection) selected from the attribute-scheme database, assuming that k = 2 corresponds to the helium mass spectrometry detection scheme (used for high-precision leakage detection).

[0118] historical detection accuracy sequence , extract the precision data of the scheme for m=100 times of historical detection (such as the first time 98%, the 50th time 95%, the 100th time 92%), form a sequence {0.98,..., 0.95,.., 0.92}, quantify the long-term performance change of the scheme, and measure the degree of precision attenuation by the ratio of the maximum precision (optimal performance when the new device) to the minimum precision (worst performance after aging) in the sequence.

[0119] In this embodiment, the initial simulation matrix row vector For example, for a target mold (such as an aero-engine case mold), a helium mass spectrometry scheme is used for simulation detection (input helium concentration 5%, vacuum extraction time 30s), and features such as pressure decay standard deviation, helium leakage, and detection time are extracted to form a row vector after normalization (such as =[0.6, 0.5, 0.4], and the smaller the value represents the closer to the ideal state).

[0120] The ideal detection matrix row vector is to extract the historical optimal detection data of the helium mass spectrometry scheme (such as a batch of new case detection, pressure decay standard deviation 0.1kPa, leakage , time 2 minutes), and form a row vector after normalization (such as =[0.2, 0.3, 0.2], which represents the theoretical optimal performance of the scheme).

[0121] In this embodiment, the larger the air tightness coefficient, the worse the current performance and long-term stability of the scheme, which needs to be optimized or replaced first.

[0122] The beneficial effects of the above technical scheme are: by quantifying the aging trend of the scheme through the historical precision attenuation, the dual-dimensional evaluation of the simulation-ideal matrix difference to characterize the current performance, and the air tightness coefficient to realize the unified quantification of the real-time ability and long-term stability of the detection scheme: it avoids the misselection of new schemes with high precision but easy to age, and prevents the misuse of old schemes with stability but poor current performance, providing a more scientific priority basis for subsequent parameter optimization.

[0123] The present application provides a mold air tightness detection parameter optimization method, which determines the optimization parameters of the physical model, comprising:

[0124] According to the column vector of each feature dimension in the difference detection matrix, the optimization direction of the physical model is determined;

[0125] According to the detection emphasis direction of each candidate air tightness scheme, a first scheme matching the optimization direction is selected from all candidate air tightness schemes, and the air tightness coefficients of the first scheme are sorted from small to large;

[0126] Based on the ranking results, starting from the minimum airtightness coefficient, the parameter optimization strategy corresponding to the minimum airtightness coefficient is encoded into a knowledge vector and combined with the corresponding accuracy decay rate to determine the optimization parameters of the physical model.

[0127] In this embodiment, the elements in the difference detection matrix represent the deviation between the current performance and the ideal state. For example, in a helium mass spectrometry detection scheme for a certain automobile cylinder block mold, the initial simulated pressure decay standard deviation is 0.5 kPa (0.6 after normalization), the ideal value is 0.2 kPa (0.2 after normalization), and the deviation of this feature in the difference matrix is ​​0.4.

[0128] The column vectors of the feature dimension are each column in the matrix corresponding to a detection feature (such as pressure attenuation, ultrasonic echo amplitude, and detection time). The variance contribution of the column vectors is calculated by principal component analysis (PCA). If the variance of the pressure attenuation column accounts for 80% (far higher than other columns), the optimization direction is focused to reduce the standard deviation of pressure attenuation and improve the accuracy of pressure detection.

[0129] In this embodiment, the detection focus is the core advantage of each candidate scheme. For example, helium mass spectrometry focuses on high-precision leakage quantification (suitable for high-pressure testing conditions, with a detection limit of up to 100%). Air pressure testing focuses on low-cost batch testing (suitable for room temperature storage conditions, with a single testing cost of <50 yuan), while ultrasonic testing focuses on rapid defect location (suitable for high-temperature molding conditions, with a testing speed of >10 times / minute).

[0130] The first selection and ranking of schemes is as follows: if the optimization direction is to improve pressure accuracy, select schemes that focus on pressure detection (helium mass spectrometry, gas pressure detection); sort them in ascending order of their airtightness coefficients (e.g., helium mass spectrometry Ck=0.8, gas pressure Ck=1.2), and prioritize the scheme with the smallest coefficient (better current performance and long-term stability).

[0131] In this embodiment, the knowledge vector encodes the parameter optimization strategy of the optimal solution (such as helium mass spectrometry) into a vector. For example, helium concentration 5% → 5.5%, vacuuming time 30s → 25s, mass spectrometry scanning frequency 100Hz → 120Hz are converted into parameter change vectors [+0.5%,−5s,+20Hz], quantifying the transferability of the strategy.

[0132] Combined with accuracy attenuation rate compensation: If the historical accuracy attenuation rate of the helium mass spectrometer leads to a decrease in sensitivity due to equipment aging, the parameters are adjusted through a linear compensation model: the helium concentration is increased by 0.5% (compensating for sensitivity), the vacuuming time is shortened by 5s (compensating for efficiency), and the scanning frequency is increased by 20Hz (compensating for data acquisition speed). The final optimized parameters are: helium concentration 5.5%, vacuuming time 25s, and scanning frequency 120Hz. The parameters are then input into the physical model for iterative optimization.

[0133] The beneficial effects of the above technical solutions are: through the three-layer logic of difference matrix positioning core deviation, scheme focusing matching screening, knowledge vector + attenuation compensation parameter adjustment, the precise directional optimization of detection parameters is realized, blind adjustment of all parameters is avoided, and the influence of equipment aging is offset through historical attenuation compensation.

[0134] The application provides a mold air tightness detection parameter optimization method, which synchronously collects multi-modal data of each working condition and performs redundancy processing, and comprises the following steps:

[0135] All initial features are extracted from the multi-modal data of each working condition to form an original feature set corresponding to the working condition , wherein, respectively represent the 1st initial feature, the 2nd initial feature and the n1th initial feature based on the multi-modal data;

[0136] A feature subset H related to the air tightness detection target is screened from the original feature set , wherein, respectively represent the 1st screening feature, the 2nd screening feature and the n2th screening feature, and the information gain rate of each screening feature is calculated , wherein, is the entropy of the original data set; represents the conditional entropy of the j2th screening feature in H , and the value of j2 is 1, 2, 3,..., n2;

[0137] If the information gain rate is less than the corresponding experimental calibration threshold, a first elimination label is assigned to the screening feature corresponding to the information gain rate .

[0138] The device working log of the acquisition device corresponding to each screening feature is obtained, and the start communication field of the acquisition device and the first field attached to the screening feature are extracted;

[0139] According to the state influence relationship of the start communication field and the first field, the preset hot data of the corresponding working condition is obtained, and the mapping dependency relationship between each screening feature and the preset hot data is determined, wherein the preset hot data is the direction of attention under the corresponding working condition;

[0140] Based on the state influence relationship and the mapping dependency relationship, it is determined whether to assign a second elimination label to the screening feature;

[0141] When the corresponding screening feature is assigned the first elimination label and the second elimination label, the corresponding screening feature is redundantly eliminated.

[0142] ​In this embodiment, the multi-modal data is collected synchronously for a certain working condition of the mold (e.g., a high-temperature forming working condition with a temperature of 150°C and a pressure of 30 MPa), including three types of data, i.e., pressure data (accuracy: 0.01 kPa), infrared temperature data (resolution: 0.1°C), and ultrasonic echo data (frequency: 5 MHz), covering multi-dimensional information such as pressure fluctuation, temperature distribution, and structural defects.

[0143] Initial features: basic features are extracted for each type of data.

[0144] Pressure data: standard deviation of pressure fluctuation and peak difference are calculated.

[0145] Temperature data: temperature field gradient and average temperature difference are calculated.

[0146] Ultrasonic data: echo amplitude and frequency shift are calculated.

[0147] In this embodiment, the target of the air tightness detection is to determine whether the leakage amount of the mold is less than or equal to 1×10-6 Pa·m3 / s (aviation standard).

[0148] The relevant feature subset H is selected by the mutual information algorithm to be strongly associated with the leakage, such as the standard deviation of pressure fluctuation (σp), which increases dramatically when there is a leakage, the temperature field gradient (G), which is large when there is a leakage point with fast heat dissipation, and the ultrasonic echo amplitude (A), which attenuates at the defect. Thus, H ={hu1,hu2,hu3}, and n2=3.

[0149] The entropy H0 of the original data set is a measure of the uncertainty of the leakage judgment (e.g., an entropy value of 0.8 represents a 50% probability of missed detection), and the conditional entropy Huj2 is a measure of the uncertainty of the missed detection probability after dividing the data by the feature (temperature gradient) (e.g., reduced to 0.4 with an entropy value of 0.6).

[0150] For example: =0.60.8−0.6≈0.33. If the experimental calibration threshold is 0.35, then <0.35, hu2 is assigned the first rejection label (insufficient information contribution).

[0151] ​​​In this embodiment, the device operation log is the log record of different types of devices, the start time (2025-07-06-10:00:00), the communication state (wireless connection), the feature acquisition timestamp (10:00:02), the extracted start communication field ("Wireless_Init_10:00") and the first field ("Pressure_Std_0.5kPa", such as the feature value of hu1).

[0152] The state influence relationship is analyzed when wireless communication, for example, the start delay (such as 2 seconds) causes the feature acquisition to lag, and the standard deviation of the pressure fluctuation of hu1 is smoothed (the actual value is 0.5kPa, and the acquisition value is 0.4kPa), which weakens the sensitivity to leakage, and it is determined that the start communication field and the first field are negatively correlated.

[0153] In this embodiment, for example, the high-temperature forming condition focuses on the leakage caused by thermal deformation, at this time, the preset thermal data is the mold surface temperature field distribution (resolution 0.1℃), and the degree of thermal deformation is quantified (for example, if the temperature difference in a certain area is greater than 5℃, it is determined that there is a risk of thermal deformation).

[0154] The mapping dependency relationship is analyzed, for example, the correlation between hu2 (temperature gradient) and thermal deformation: the temperature gradient describes the temperature change rate, and the thermal deformation is determined by the temperature difference, and the physical correlation is weak (the Pearson correlation coefficient is 0.3<0.6), and it is determined that hu2 and the preset thermal data are insufficiently dependent on each other.

[0155] In this embodiment, the second elimination label determination is combined with the start communication to cause the corresponding feature distortion (state influence) and the weak correlation with the thermal deformation (mapping dependency), and the second elimination label is assigned to the corresponding feature.

[0156] Redundancy elimination: when the corresponding feature meets the first label (IR is insufficient) + the second label (correlation distortion) at the same time, the feature subset Hu is eliminated, and finally the core feature is reserved, for example, {hu1, hu3}.

[0157] The beneficial effects of the above technical solutions are: through the four-layer screening of multi-modal feature full extraction→information gain rate screening correlation→device log analysis distortion→thermal data verification dependency, the dual-dimension determination of feature redundancy is realized, that is, the weakly correlated features (such as temperature gradient) with low information contribution are eliminated, and the pseudo-correlated features (such as invalid features caused by communication lag) disturbed by devices or with weak physical correlation are excluded, thereby indirectly improving the accuracy of the air tightness detection.

[0158] The present application provides a mold air tightness detection parameter optimization method, based on the running time ratio of each working condition and multi-modal data, a multi-working condition cooperative optimization objective function is constructed, including:

[0159] Constructing a multi-physical field coupling model wherein, the input layer of the single-condition coupling model is constructed based on the fusion of the core features of the corresponding condition according to the coupling mechanism, and the output layer is constructed by the predicted value of the air tightness index under the corresponding condition, wherein, , , , respectively represent the leakage amount, the detection time, and the damage value; is a detection parameter vector, and represents the ith detection parameter, and n01 represents the number of detection parameters; is a model structure parameter vector, and represents the ith2 model structure parameter, and n02 represents the number of model structure parameters; is a model output, and ;

[0160] The core features of the multi-modal data after quantifying the redundancy of each condition are contributed to the corresponding condition, and the cross-condition contribution of the multi-modal features is integrated, and the importance weight of the condition is combined to construct a multi-condition collaborative optimization objective function: wherein, , , respectively represent the weight of the leakage amount, the detection time, and the damage value; represents the importance weight of the determined j3 condition; , , respectively represent the actual leakage amount, the actual detection time, and the actual damage value under the j3 condition; , , respectively represent the ideal leakage amount, the ideal detection time, and the ideal damage value under the j3 condition;

[0161] find the optimal parameter combination of the multi-condition collaborative optimization objective function .

[0162] In this embodiment, the multi-physical field coupling is for the mold high-temperature forming condition (150℃, 30MPa), which integrates the fluid field (gas leakage flow, Navier-Stokes equation), temperature field (heat conduction, Fourier's law), and structure field (mold deformation, elastic-plastic constitutive relation), and simulates the coupling effect of temperature rise→mold thermal expansion→seal gap change→leakage change.

[0163] The input layer (core feature fusion) adopts the redundant core features such as pressure fluctuation standard deviation (fu1, such as 0.5kPa) and ultrasonic echo amplitude (fu3, such as 80dB) to construct the input vector X1 through physical equation coupling.

[0164] Output layer (airtightness index): leakage amount xl: mass flow of simulated gas leakage, detection time tx: total time consumption of simulated sensor acquisition and signal processing (e.g. 6 minutes), damage value sx: calculated equivalent plastic strain of the mold (e.g. 0.12%, reflecting creep damage), which constitute an output vector.

[0165] In this embodiment, the detection parameter vector includes, for example: pressure P = 10 MPa, holding time T = 60 s, etc., model structure parameters such as elastic modulus E = 70 GPa, thermal expansion coefficient ℃, initial sealing gap d0 = 0.1 mm, etc.

[0166] In this embodiment, the core features after redundancy are, for example, pressure fluctuation (fu1) and ultrasonic echo (fu3) under high-temperature working conditions, which are retained after removing noise (such as temperature gradient distortion due to equipment communication) through the previous steps, and their contribution degrees are quantified by linear regression + variance decomposition:

[0167] Pressure fluctuation explains 80% of the leakage amount change (R2 = 0.8), with high contribution degree;

[0168] Ultrasonic echo explains 75% of the defect positioning error (R2 = 0.75), with high contribution degree.

[0169] Cross-condition contribution integration: the core features of high-pressure test conditions are helium concentration (fu5) and vacuum pumping time (fu6), and cross-condition feature association (such as the Pearson correlation coefficient between pressure and helium concentration 0.65) is established by association rule mining to realize multi-condition feature collaboration.

[0170] In this embodiment, the importance weight of working conditions = 0.6 for high-temperature forming (60%), = 0.25 for normal temperature storage (25%), = 0.15 for high-pressure testing (15%), = 0.15.

[0171] Target weight , , : priority is given to leakage detection accuracy ( = 0.6), followed by efficiency ( = 0.3), and finally mold damage ( = 0.1), which meets .

[0172] In this embodiment, the optimization algorithm is an improved firefly algorithm, which globally searches for detection parameters (such as pressure P, holding time T, helium concentration CHe, etc.) to minimize minD(X), and outputs the optimal parameter combination (P=12MPa, T=45s, CHe=5.5%).

[0173] The beneficial effects of the above technical solutions are: through the three-layer logic of multi-physical field coupling modeling (describing cross-field action) → core feature contribution quantization (focusing on key information) → multi-working condition weight optimization (adapting to service scenarios), the full-working condition precision cooperation of the air tightness detection parameter is realized, which not only breaks through the local optimization of a single working condition (such as excessive pursuit of precision leading to waste of efficiency in high-pressure test), but also restores the real coupling effect through the physical model (such as the influence of thermal expansion on leakage), so that the detection precision is improved.

[0174] The present application provides a mold air tightness detection parameter optimization method, which comprises:

[0175] Receiving the optimization result of the multi-working condition cooperative optimization target function, extracting the optimal detection parameter combination under each working condition and the corresponding target function value;

[0176] Inputting the actual detection data under the optimization parameters into the DBSCAN clustering algorithm to identify abnormal data clusters, and calculating the sensitivity of each target function threshold to the detection result;

[0177] If the detection error of the corresponding working condition exceeds the corresponding threshold for consecutive ne times, trigger the threshold fine-tuning based on Kalman filtering;

[0178] At the same time, the weight matrix of the multi-working condition cooperative optimization target function is corrected after every nn detection period, and a threshold correlation matrix is established, and when there is working condition threshold update, the threshold of the associated working condition is adjusted synchronously through the threshold correlation matrix.

[0179] In this embodiment, the target function value is a multi-objective weighted sum under the corresponding parameters (such as F1=0.85 in high-temperature working condition, representing the comprehensive performance of leakage, efficiency and damage, and the smaller the value is, the better it is).

[0180] The actual detection data under the optimization parameters are measured according to the optimal parameters, and 100 groups of data are collected (such as measuring every 2 minutes in high-temperature working condition, recording the leakage amount, detection time and mold strain).

[0181] DBSCAN clustering: set the neighborhood radius to 0.5 (leakage amount standard deviation unit) and the minimum point number minPts=5, and identify the abnormal data cluster (such as 10 consecutive points of leakage amount, which is determined as "continuous leakage" anomaly).

[0182] In this embodiment, the ratio of the threshold change rate (such as ±1%) and the detection error change rate (such as from 3% to 5% or 1%) is calculated when the leakage threshold is 3%, and the sensitivity (the greater the value, the more significant the threshold on the result) is obtained.

[0183] In this embodiment, the threshold is fine-tuned if the high-temperature working condition has a leakage error >3% for 5 consecutive times.

[0184] Kalman filter model:

[0185] State equation: (A=0.99, simulate threshold slow change characteristics);

[0186] Observation equation: zk=Hxk+vk (H=1, observation value is actual detection error);

[0187] Iterative update: the threshold is updated by combining the predicted threshold and the observation error, for example, to achieve accurate fine-tuning of the threshold from 3% to 2.8%.

[0188] In this embodiment, every nn detection period: set nn=50, after every 50 detections, analyze the target function weight deviation (such as damage weight=0.1, the actual damage exceeds the standard working condition by 30%), and correct the weight by gradient descent method (such as ε3 from 0.1 to 0.2, to improve the damage attention).

[0189] Threshold correlation matrix: construct a 3x3 matrix (rows and columns correspond to working conditions), and quantify the threshold correlation degree between working conditions (such as the pressure threshold correlation degree C13 between high temperature and high pressure is 0.7). If the high-temperature pressure threshold is from ±0.5kPa to ±0.3kPa, the high-pressure pressure threshold is adjusted to ±0.35kPa (0.5−0.7×0.2) at the same time, to realize cross-working condition threshold coordination.

[0190] The beneficial effects of the above technical solutions are: through the closed loop of landing of optimization results → abnormal data identification → dynamic fine-tuning of threshold → cross-working condition coordinated update, the full life cycle adaptation of detection threshold is realized, which not only captures abnormal data caused by device aging (such as sensor drift) in time through DBSCAN, but also dynamically balances the threshold between accuracy guarantee and working condition adaptation (such as the high-pressure test threshold is optimized at the same time when the high-temperature is adjusted due to the correlation effect, to avoid local optimization of isolated parameter tuning.

[0191] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. A method for optimizing mold airtightness testing parameters, characterized in that, include: Step 1: Simulate and test the physical model of the target mold, and optimize and apply the physical model in practice. Step 2: Divide the actual usage log of the target mold into high temperature molding condition, normal temperature storage condition and high pressure test condition, and calculate the running time ratio of each condition. At the same time, collect multimodal data of each condition and perform redundancy processing. Step 3: Based on the running time ratio of each working condition and the redundant multimodal data, construct a multi-working-condition collaborative optimization objective function, and perform objective optimization on the multi-working-condition collaborative optimization function; Step 4: Update the detection threshold of the multi-condition collaborative optimization objective function in reverse based on the optimization results; The process includes simulating and testing the physical model of the target mold, and then simulating and optimizing the physical model, including: Based on the attribute characteristics of the target mold, an attribute vector is constructed, where the attribute characteristics are related to the mold's material type, cavity complexity, dimensional tolerance, and surface roughness. The similarity between the attribute vector and each scheme is calculated from the attribute-scheme database and combined with the cosine similarity algorithm. A set of candidate airtightness schemes is obtained by filtering. The physical model of the target mold is simulated and tested in turn according to each candidate airtightness scheme in the set of candidate airtightness schemes to construct an initial simulation matrix. Obtain the ideal detection vector of the historical best data for each candidate airtightness scheme and construct the ideal detection matrix; Based on the initial simulation matrix and the ideal detection matrix, and combined with the historical accuracy decay rate of each candidate airtightness scheme, the airtightness coefficient of the target mold under the corresponding candidate airtightness scheme is determined. The optimization direction of the physical model is determined based on the difference detection matrix of the initial simulation matrix and the ideal detection matrix, and the optimization parameters of the physical model are determined by combining the airtightness coefficient and the historical accuracy decay rate. The optimization direction of the physical model is determined based on the column vectors of each feature dimension in the difference detection matrix. The feature dimensions are related to pressure attenuation features, ultrasonic echo features, and helium leakage features; The physical model is simulated and optimized according to the optimization parameters.

2. The method for optimizing mold airtightness testing parameters according to claim 1, characterized in that, Determining the airtightness coefficient of the target mold under the corresponding candidate airtightness scheme includes: Extract the historical detection accuracy sequence of the k-th candidate airtightness scheme Calculate the corresponding historical accuracy attenuation rate ,in, represents the historical detection accuracy of the k-th candidate airtightness scheme under the i-th test; m represents the historical test count of the k-th candidate airtightness scheme; Indicates all The minimum value in; Indicates all The maximum value in; By combining the Euclidean distance difference between the initial simulation matrix and the ideal detection matrix, and the accuracy attenuation rate, the airtightness coefficient of the k-th candidate airtightness scheme is calculated. ; ； in, , 2 represents the weights, and ; This represents the normalized row vector of the k-th candidate airtightness scheme based on the initial simulation matrix; Let represent the normalized row vector of the k-th candidate airtightness scheme based on the ideal detection matrix.

3. The method for optimizing mold airtightness testing parameters according to claim 2, characterized in that, Determining the optimization parameters of the physical model includes: Based on the detection focus of each candidate airtightness scheme, a first scheme that matches the optimization direction is selected from all candidate airtightness schemes, and the airtightness coefficients of the first schemes are sorted from small to large. Based on the ranking results, starting from the minimum airtightness coefficient, the parameter optimization strategy corresponding to the minimum airtightness coefficient is encoded into a knowledge vector and combined with the corresponding accuracy decay rate to determine the optimization parameters of the physical model.

4. The method for optimizing mold airtightness testing parameters according to claim 1, characterized in that, Synchronously collect multimodal data for various operating conditions and perform redundancy processing, including: All initial features are extracted from the multimodal data of each working condition to form the original feature set for that working condition. ,in, These represent the first, second, and n1th initial features based on multimodal data, respectively. Select a subset H of features relevant to the airtightness detection target from the original feature set. ,in, Let n represent the first, second, and n2th filtering features, respectively, and calculate the information gain ratio for each filtering feature. ,in, The entropy of the original dataset; H represents The conditional entropy after the j2-th filtered feature is partitioned, and the value of j2 is 1, 2, 3, ..., n2; like Less than the corresponding experimental calibration threshold, towards The corresponding filtering feature is assigned the first rejection label; Obtain the device operation log of the acquisition device corresponding to each filtering feature, and extract the startup communication field of the acquisition device and the first field attached to the filtering feature; Based on the communication type of the acquisition device, analyze the state influence relationship between the start communication field and the first field. At the same time, obtain the preset hot data of the corresponding working condition, and determine the mapping dependency relationship between each filtering feature and the preset hot data. The preset hot data is the focus of attention under the corresponding working condition. Based on the state influence relationship and mapping dependency relationship, determine whether to assign a second rejection label to the screening feature; When the corresponding filtering feature is assigned a first rejection label and a second rejection label, the corresponding filtering feature will be redundantly rejected.

5. The method for optimizing mold airtightness testing parameters according to claim 4, characterized in that, Based on the running time ratio of each operating condition and the redundant multimodal data, a multi-operating condition collaborative optimization objective function is constructed, including: Constructing a multiphysics coupling model In this model, the input layer is constructed by fusing the core features of the corresponding working condition based on the coupling mechanism, and the output layer is constructed from the predicted values ​​of the airtightness index under the corresponding working condition. , , , These represent the leakage amount, detection time, and damage value, respectively. For the detection parameter vector, and This represents the i1th detection parameter, and n01 represents the number of detection parameters; Let be the model structure parameter vector, and This represents the i2th model structure parameter, and n02 represents the number of model structure parameters; For model output, and ; The contribution of the core features of the redundant multimodal data for each working condition is quantified in the corresponding working condition, and the cross-working condition contributions of the multimodal features are integrated. Combined with the importance weights of the working conditions, a multi-working-condition collaborative optimization objective function is constructed: ,in, , , These represent the weights of leakage amount, detection time, and damage value, respectively. This indicates the importance weight of the runtime compared to the determined j3rd operating condition; , , These represent the actual leakage, actual detection time, and actual damage value under the j3rd operating condition, respectively. , , These represent the ideal leakage rate, ideal detection time, and ideal damage value under the j3rd operating condition, respectively. Find the optimal parameter combination of the multi-condition collaborative optimization objective function. .

6. The method for optimizing mold airtightness testing parameters according to claim 1, characterized in that, The pressure decay characteristics are related to the detection pressure, the pressure holding time, and the pressurization rate. The ultrasonic echo characteristics are related to the ultrasonic frequency, the coupling agent thickness, and the focusing depth. The characteristics of helium leakage are related to helium concentration, vacuuming time, and mass spectrometry scanning frequency.

7. The method for optimizing mold airtightness testing parameters according to claim 1, characterized in that, Based on the optimization results, the importance of each working condition is graded and verified, and the detection threshold of the objective function for multi-working-condition collaborative optimization is updated in reverse, including: Receive the optimization results of the multi-condition collaborative optimization objective function, and extract the optimal combination of detection parameters and the corresponding objective function value under each condition; The actual detection data under the optimization parameters are input into the DBSCAN clustering algorithm to identify abnormal data clusters, and the sensitivity of each objective function threshold to the detection results is calculated. If the detection error exceeds the corresponding threshold for n consecutive times under the corresponding working condition, threshold fine-tuning based on Kalman filtering is triggered. Meanwhile, after every n detection cycles, the weight matrix of the multi-condition collaborative optimization objective function is corrected, and a threshold correlation matrix is ​​established. When there is a condition threshold update, the threshold of the associated condition is adjusted synchronously through the threshold correlation matrix.

Images