Method, apparatus, and medium for evaluating composite repair consumables

Abstract

The application discloses a kind of composite material repair consumable evaluation method, device, equipment and medium, it is related to composite material structure maintenance technical field.By introducing a series of controllable disturbance factors around the benchmark process parameters to simulate actual fluctuations, the performance of repair consumables under non-ideal process conditions is tested, and the safety boundary of its qualified product output is determined to provide a more comprehensive and practical evaluation system.By introducing the process window volume as an objective and comparable robustness quantitative index, and with the help of Gaussian process regression model to efficiently predict the process space boundary, the high cost and low efficiency problem of traditional exhaustive test method is avoided.The process robustness that determines the stability of repair consumables in actual maintenance environment is converted into quantifiable, comparable equivalent curing process window and its measurement value, which realizes the scientific selection of repair consumables and significantly reduces the quality risk in actual maintenance operation.

Description

Technical Field

[0001] This invention relates to the field of composite material structure repair technology, and in particular to an evaluation method, apparatus, equipment and medium for composite material repair consumables. Background Technology

[0002] Currently, the robustness evaluation of composite material repair consumables is typically based on single or limited performance tests using ideal process parameters recommended by the consumable supplier. However, the actual repair environment for composite materials is generally subject to various process fluctuation factors, such as deviations in equipment control precision, differences in operator execution, and variations in ambient temperature and humidity. These factors can inevitably cause deviations in actual process parameters from their set values. Traditional evaluation methods fail to incorporate such fluctuations into the evaluation system. Therefore, they cannot objectively reflect whether, under real production conditions, when process parameters deviate from recommended values ​​within a reasonable range, the consumable can still consistently and stably output repair results that meet quality requirements. This leads to the potential risk of repair failure due to process fluctuations in the practical application of composite material repair. Summary of the Invention

[0003] In view of this, the present invention provides an evaluation method, apparatus, electronic device and medium for composite material repair consumables, in order to solve the technical problem that traditional evaluation methods ignore process fluctuations, cannot truly evaluate the process robustness of consumables, and thus lead to the risk of failure in actual repairs.

[0004] Firstly, a method for evaluating composite material repair consumables is provided, the method comprising: Obtain the process parameter space of multiple preset disturbance factors for repair consumables used for composite material repair, wherein the process parameter space includes the process parameter baseline value and fluctuation range of each preset disturbance factor. Based on the process parameter space, multiple combinations of test parameters are generated, and through these combinations, disturbance repair tests are conducted on the repair consumables to generate multiple measured data for multiple preset quality indicators. Based on multiple combinations of experimental parameters and multiple measured data, a multi-output Gaussian process regression model is trained to obtain a quality index prediction model. Based on the quality index prediction model and the preset constraints of each preset quality index, a process window boundary search is performed in the process parameter space to determine the feasible process window for repair consumables and the number of sample points in the process parameter space that satisfy all preset constraints. Based on the process parameter space, feasible process window, and number of sample points, a robustness index for repair consumables is generated. Evaluation results for repair consumables are generated based on the robustness index.

[0005] Secondly, an evaluation device for composite material repair consumables is provided, the device comprising: The acquisition module is used to acquire the process parameter space of multiple preset disturbance factors for repair consumables used for composite material repair. The process parameter space includes the baseline values ​​of the process parameters of each preset disturbance factor and their fluctuation range. The first generation module is used to generate multiple combinations of test parameters based on the process parameter space, and to conduct disturbance repair tests on repair consumables through multiple combinations of test parameters, generating multiple measured data of multiple preset quality indicators; The model training module is used to train a multi-output Gaussian process regression model based on multiple combinations of experimental parameters and multiple measured data to obtain a quality index prediction model. The determination module is used to perform process window boundary search within the process parameter space based on the quality index prediction model and the preset constraints of various preset quality indices, to determine the feasible process window for repair consumables and the number of sample points in the process parameter space that satisfy all preset constraints. The second generation module is used to generate the robustness index of repair consumables based on the process parameter space, feasible process window and number of sample points. The third generation module is used to generate evaluation results for repair consumables based on the robustness index.

[0006] Thirdly, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described evaluation method for composite material repair consumables.

[0007] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the above-described evaluation method for composite material repair consumables.

[0008] The aforementioned evaluation method, apparatus, electronic equipment, and storage medium for composite material repair consumables utilize a series of controllable disturbance factors introduced around benchmark process parameters to simulate actual fluctuations. This systematically tests the performance of repair consumables under non-ideal process conditions, thereby determining the safety boundary for producing qualified products. This expands consumable performance evaluation from traditional single-point verification to a multi-dimensional parameter space, providing a more comprehensive and practical evaluation system. By introducing the process window volume as an objective and comparable robustness quantification indicator and efficiently predicting the process space boundary using a Gaussian process regression model, the high cost and low efficiency of traditional exhaustive testing methods are avoided. The process robustness, which determines the stability of repair consumables in actual maintenance environments, is transformed into a quantifiable and comparable equivalent curing process window and its metric, enabling scientific screening of repair consumables and significantly reducing quality risks in actual maintenance operations. Attached Figure Description

[0009] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a flowchart illustrating the evaluation method for composite material repair consumables in one embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of an evaluation device for composite material repair consumables in one embodiment of the present invention. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in the present invention are only for illustrative and descriptive purposes and are not intended to limit the scope of protection of the present invention.

[0011] Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this invention illustrate operations implemented according to some embodiments of the invention. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or performed simultaneously. Moreover, those skilled in the art, guided by the content of this invention, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0012] Furthermore, the embodiments described herein are merely some, not all, of the embodiments of the invention. The components of the embodiments of the invention described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0013] It should be noted that the term "comprising" will be used in the embodiments of the present invention to indicate the presence of a feature subsequently declared, but does not exclude the addition of other features. It should also be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0014] Composite materials have been widely used in the secondary and primary load-bearing structures of various aircraft, and their proportion in the weight of aircraft structures continues to increase. However, defects or damage are unavoidable in the production, use, and maintenance of composite material structures, which makes the repair of composite material components increasingly important.

[0015] In recent years, with the vigorous promotion of self-reliance and control over the large aircraft industry chain, the localization of composite material repair consumables has become a key development direction, and its goal is moving from "usable" to "easy to use." However, relevant units have not yet established a dedicated standard system for domestically produced aviation composite material repair consumables. With the C919 and other domestically produced large aircraft entering the mass production stage and the increasing uncertainty of the international supply chain, increasing the localization rate of maintenance consumables has become an inevitable trend.

[0016] Currently, the evaluation method for maintenance consumables typically involves conducting one or a few performance tests under the "ideal" process parameters recommended by the material supplier, and then deeming it "qualified." However, this method ignores the process fluctuations that are common in actual repair workshop environments. For example, deviations in equipment control precision, differences in operator execution, and variations in ambient temperature and humidity can all cause unavoidable fluctuations in actual process parameters around the set values. Existing evaluation systems do not include such fluctuations in their testing scope, thus failing to answer a key engineering question: Under real production conditions, when process parameters deviate from the recommended values ​​within a certain range, can the consumable still consistently and stably output repair results that meet quality requirements? This inadequacy in evaluation leads to a lack of objective understanding of the "process robustness" of consumables. If domestically produced consumables have a narrow process window, they are highly susceptible to repair failures due to common fluctuations in practice, such as excessive porosity and interlayer debonding, leading to high rework costs and even safety hazards, ultimately hindering the smooth progress of domestic production. Therefore, there is an urgent need to establish a method that can systematically evaluate the stability of composite material repair consumables under actual fluctuating process conditions, so as to scientifically identify and screen domestically produced consumables with excellent robustness, ensure repair quality, control costs, and support the implementation of the supply chain self-sufficiency strategy.

[0017] Based on the above problems, this application provides an evaluation method for composite material repair consumables, which transforms the process robustness that determines the stability of domestic consumables in actual maintenance environments into a quantifiable and comparable equivalent curing process window and its metric value, thereby realizing the scientific screening of domestic consumables and significantly reducing the quality risks and supply chain uncertainties in actual maintenance operations.

[0018] The following is a detailed description of this case, in conjunction with the relevant accompanying drawings in the instruction manual.

[0019] Please see Figure 1 This description and embodiment provide an evaluation method for composite material repair consumables, specifically including the following steps: S10: Obtain the process parameter space for multiple preset disturbance factors of repair consumables used for composite material repair; The process parameter space includes the baseline values ​​of process parameters for each preset disturbance factor and their fluctuation range.

[0020] It is understood that the executing entity of this invention can be an evaluation device for composite material repair consumables, or it can be a terminal or a server; no specific limitation is made here. This embodiment of the invention will be described using a server as an example.

[0021] In this step, the repair consumables used for composite material repair refer to specialized material combinations specifically designed for repairing damage to composite material structures such as aircraft, such as a domestically produced prepreg and adhesive film system. Before carrying out composite material repair, it is necessary to evaluate the performance of the selected repair consumables under fluctuating process conditions to determine whether they can reliably replace imported consumables in actual, fluctuating repair environments. To this end, it is necessary to identify several pre-defined disturbance factors affecting the quality of composite material repair, and for each pre-defined disturbance factor, based on the process manual, equipment capabilities, and operating procedures, determine the baseline values ​​and fluctuation ranges of the process parameters that can be studied. The baseline values ​​and fluctuation ranges of all disturbance factors together define the complete process parameter space.

[0022] Optionally, the repair consumables can be domestically produced consumables used to replace imported ones. Further, a widely accepted or validated curing process curve is selected as a benchmark, clearly defining the key process parameters and their changes over time. Based on practical experience in the repair workshop, equipment accuracy data, and operating procedures, the key disturbance factors that have the greatest impact on repair quality and are most prone to fluctuation in practice are identified, and their reasonable fluctuation ranges are determined, including but not limited to: heating rate, holding temperature, holding time, and cooling rate. Among these, the heating rate affects resin viscosity changes, flow wetting, and internal stress; the holding temperature directly determines the resin curing reaction rate and its final conversion rate; the holding time ensures the curing reaction is fully completed; and the cooling rate affects the residual stress state and interlayer bonding quality of the cured component.

[0023] S20: Based on the process parameter space, generate multiple combinations of test parameters, and conduct disturbance repair tests on repair consumables through multiple combinations of test parameters to generate multiple measured data of multiple preset quality indicators.

[0024] In this step, based on the process parameter space defined by the baseline values ​​and fluctuation ranges of the process parameters for various disturbance factors, an experimental design method is used to generate test schemes containing multiple combinations of parameters for different disturbance factors. Subsequently, in a real or simulated maintenance environment, the repair curing process corresponding to each combination of test parameters is sequentially executed on the repair consumables according to the test scheme to complete the disturbance repair test. For the repair specimens produced in each test, non-destructive and destructive testing is performed according to relevant standards, and the measured data of all preset quality indicators used to evaluate the process robustness of the repair consumables are measured and recorded.

[0025] Optionally, based on the fundamental goal of ensuring the safe and reliable restoration of the designed function of the composite material repair structure, and strictly in accordance with the core requirements of aviation maintenance industry standards and engineering practices, several preset quality indicators are set, including at least one of the following: average porosity (%), maximum porosity in sensitive areas (%), percentage of area exceeding porosity (%), mean absolute thickness deviation (mm), maximum positive thickness deviation (mm), maximum negative thickness deviation (mm), standard deviation of thickness deviation (mm), edge bond strength (MPa), bond strength at steps (MPa), delamination defect density (number / cm²), low-resin / high-resin defect density (number / cm²), and wrinkle defect density (number / cm²). Through the setting of these multiple quality indicators, a multi-dimensional comprehensive evaluation system is constructed, aiming to systematically and comprehensively determine whether the repair results meet the stringent requirements for airworthiness safety and reliable service from aspects such as structural strength, internal quality, dimensional accuracy, and process consistency.

[0026] In one embodiment of this application, a specific disturbance repair test scheme is provided. In S20, multiple test parameter combinations are generated based on the process parameter space, and disturbance repair tests are conducted on repair consumables using these multiple test parameter combinations to generate multiple measured data for multiple preset quality indicators. Specifically, this includes the following steps S21-S22: S21: Through orthogonal experimental design, based on orthogonal arrays and process parameter space, multiple combinations of experimental parameters are determined, wherein each combination of experimental parameters includes process parameters from different preset disturbance factors; Each combination of test parameters includes process parameters derived from different preset disturbance factors.

[0027] In this step, for each disturbance factor, several levels are reasonably set within the process parameter space, such as high temperature, medium temperature, and low temperature. An orthogonal experimental design is employed, selecting a suitable orthogonal array and substituting each preset disturbance factor and its level into the table. The mathematical properties of the orthogonal array ensure that the generated experimental schemes possess balanced dispersion and uniform comparability. Finally, a matrix containing multiple combinations of experimental parameters is output, where each combination consists of specific level parameters from different preset disturbance factors.

[0028] For example, with four preset perturbation factors and three levels for each factor, a full experiment would require 81 trials, resulting in high costs. By employing orthogonal experimental design, the nine most representative experimental combinations can be scientifically selected from all 81 possible combinations based on an orthogonal array, thereby effectively inferring the main effects of each preset perturbation factor. Specifically, the factors and levels for the low-temperature, medium-temperature, and high-temperature experiments are set as follows: In the low-temperature group experiment: Factor A: heating rate (Level 1: 1℃ / min; Level 2: 3℃ / min; Level 3: 5℃ / min); Factor B: cooling rate (Level 1: 1℃ / min; Level 2: 3℃ / min; Level 3: 5℃ / min); Factor C: holding temperature (Level 1: 55℃; Level 2: 60℃; Level 3: 65℃); Factor D: holding time (Level 1: 90min; Level 2: 120min; Level 3: 150min).

[0029] In the intermediate temperature group experiment: Factor A: heating rate (Level 1: 1℃ / min; Level 2: 3℃ / min; Level 3: 5℃ / min); Factor B: cooling rate (Level 1: 1℃ / min; Level 2: 3℃ / min; Level 3: 5℃ / min); Factor C: holding temperature (Level 1: 115℃; Level 2: 120℃; Level 3: 125℃); Factor D: holding time (Level 1: 120min; Level 2: 150min; Level 3: 180min).

[0030] In the high-temperature group experiment: Factor A: heating rate (Level 1: 1℃ / min; Level 2: 3℃ / min; Level 3: 5℃ / min); Factor B: cooling rate (Level 1: 1℃ / min; Level 2: 3℃ / min; Level 3: 5℃ / min); Factor C: holding temperature (Level 1: 170℃; Level 2: 175℃; Level 3: 180℃); Factor D: holding time (Level 1: 60min; Level 2: 90min; Level 3: 120min).

[0031] Table 1 shows the specific details of the three sets of experiments. Taking experiment number 1 as an example, it represents the parameter combination of factor A at level 1 (e.g., 1℃ / min), factor B at level 1 (e.g., 1℃ / min), factor C at level 1 (55°C in the low-temperature group, 115°C in the medium-temperature group, and 170°C in the high-temperature group), and factor D at level 1 (90 min in the low-temperature group, 120 min in the medium-temperature group, and 60 min in the high-temperature group). The other experiment numbers follow the same pattern. Thus, by using the same efficient and balanced experimental design scheme, three different typical process temperature ranges can be simultaneously covered in only 9 experiments, significantly saving costs while systematically studying the influence of the four factors within each temperature range.

[0032] Table 1

[0033] S22: Based on multiple combinations of test parameters, perform disturbance repair tests on repair consumables to obtain measured data of various preset quality indicators under each combination of test parameters.

[0034] In this step, based on the generated combinations of multiple test parameters, in a controlled real or high-fidelity simulated maintenance environment, the repair consumables to be evaluated are used to sequentially perform the repair process operations corresponding to each set of parameters on standardized composite material test pieces, thereby completing a series of systematic disturbance repair tests. After each test, the prepared repair test pieces are subjected to comprehensive non-destructive testing (including ultrasonic C-scanning, X-ray inspection, etc.) and destructive testing (including mechanical property testing, metallographic analysis, etc.) in strict accordance with relevant aerospace maintenance and testing standards (such as ASTM, SACMA, etc.). All preset quality index measured data used for comprehensively evaluating repair quality and process robustness are measured and accurately recorded. This yields a complete and clear initial dataset mapping process parameters to quality indices.

[0035] By employing the orthogonal experimental design method to design disturbance repair tests, we can obtain data reflecting the systematic relationship between process parameters and quality indicators with the fewest number of tests. This significantly reduces the initial cost and time expenditure of the evaluation, while ensuring the representativeness and scientific nature of the data. This provides a crucial high-quality data foundation for the subsequent establishment of high-precision prediction models and the realization of reliable quantitative evaluation of process robustness.

[0036] S30: Based on multiple combinations of experimental parameters and multiple measured data, train a multi-output Gaussian process regression model to obtain a quality index prediction model.

[0037] In this step, all combinations of experimental parameters are used as model input, and the measured data of the multi-dimensional preset quality indicators corresponding to each combination of experimental parameters are used as model output. An input-output dataset is constructed, and a multi-output Gaussian process regression model is trained. This allows the model to learn and accurately characterize the nonlinear mapping relationship between complex process parameters and multiple related quality indicators. After training, a quality indicator prediction model is obtained that can predict the values ​​of various preset quality indicators and their uncertainties for any unknown combination of process parameters.

[0038] In one embodiment of this application, a specific model training scheme is provided. In S30, a multi-output Gaussian process regression model is trained based on multiple combinations of experimental parameters and multiple measured data to obtain a quality index prediction model. This specifically includes the following steps S31-S34: S31: Construct a multi-output Gaussian process regression model using the squared exponential kernel function and the linear model kernel.

[0039] In this step, due to the complex characteristics of the experimental data, such as multiple inputs, multiple outputs, nonlinearity, and small sample size, this application chooses to construct a multi-output Gaussian process regression (MOGPR) model that combines a squared exponential kernel and a linear model kernel, and supplements it with targeted data preprocessing to achieve accurate learning and joint prediction of the complex nonlinear relationships between multiple process parameters and multiple quality indicators.

[0040] Specifically, the core performance of the Gaussian process regression model lies in its covariance function, i.e., the kernel function. This application adopts a composite kernel function structure: First, to effectively measure the similarity between different process parameters, ensure the smoothness of the prediction results, and improve the robustness to variance, a squared exponential kernel function is selected for each input dimension, with the basic form as follows:

[0041] in, For input x and The covariance function values ​​between them; d is the signal variance; d is the dimension index of the preset disturbance factor; D is the total dimension of the preset disturbance factor; l d x is the length scale of the d-th dimension; d The standardized value of the input index x; Input metrics Standardized values; / The normalized squared distance on the d-th dimension; This is a noise term, only if x = The time is not zero; For noise variance; For the Kronecker delta function, when x = The value is 1 if it is true, and 0 otherwise.

[0042] Optionally, hyperparameters The initial value is set to the variance of the output index; the hyperparameter l d The initial value is set to the standard deviation of the input index; hyperparameters The initial value is set to 0.1 × the variance of the output index.

[0043] Secondly, to model the intrinsic correlation among the outputs of M preset quality indicators, a linear model kernel is adopted. This kernel introduces a set of latent functions and a weight matrix W to couple and model multiple related outputs. Its multi-output covariance matrix can be expressed as:

[0044] in, The multi-output covariance matrix describes the joint correlation of multiple preset quality indicators under different combinations of experimental parameters; R represents the number of potential tasks, corresponding to the implicit independent process influencing factors; A k Let be a positive semi-definite matrix, representing the covariance structure among the k-th output components, used to control the coupling strength between different quality indices, where , It is the initialization weight matrix The k-th column vector, , This represents the weight of the j-th potential task to the i-th output metric, with its initial value randomly generated and automatically optimized subsequently; The Kronecker product is used to extend the effect of the kernel function to a multi-output space; Describe the impact of process parameters on a potential factor.

[0045] Traditional single-output Gaussian process regression models cannot effectively capture the inherent physical relationships between different output indicators, while linear model kernels, through the matrix A in their covariance structure... k This demonstrates the modeling of such relationships. When different output indicators are influenced by the same process parameter or potential process mechanism, the linear model kernel can achieve joint learning of these indicators, thereby effectively improving the overall prediction accuracy of the model. In addition, the linear model kernel significantly enhances the model's ability to jointly model and predict multiple quality indicators of composite materials by introducing a small number of potential factors (usually satisfying R ≤ number of output indicators) to explain the coupling effect between complex quality indicators.

[0046] S32: Standardize the process parameters in multiple combinations of test parameters.

[0047] In this step, the experimental parameter combinations X are standardized. The process parameters for each preset perturbation factor are standardized to eliminate numerical deviations caused by differences in dimensions and value ranges, ensuring the stability of the model optimization process and making the length-scale hyperparameter l... d The initialization is comparable (usually initialized to 1).

[0048] For example, four preset disturbance factor input data are established: heating rate (°C / min), cooling rate (°C / min), holding temperature (°C), and holding time (min).

[0049]

[0050] Standardize the input data:

[0051] in, For the first The standardized values ​​of the input data; For the first One set of original input data; The mean of the input dataset X is calculated using the following formula: ,in, This represents the total number of data points. The standard deviation of the input dataset X is calculated using the following formula: .

[0052] S33: Perform change processing on the measured data of each preset quality indicator.

[0053] In this step, the measured data of each preset quality indicator are specifically transformed to improve their statistical characteristics and make them more consistent with the modeling assumptions of a Gaussian process. Specifically, for porosity-related indicators (such as average porosity and maximum porosity): [The following is a more detailed explanation of the transformation process.] Transformation (where) (As a minimal constant), to compress the right-skewed distribution of the data and enhance the model's ability to resolve low porosity regions; for defect density indices (such as delamination, low / high resin content, and wrinkle defect density): [The text abruptly ends here, likely due to an incomplete sentence or missing information.] Transformation makes discrete counting data continuous and stabilizes its variance; for indices such as thickness deviation and bond strength, their original numerical scale is usually maintained to preserve the intuitiveness of the physical meaning.

[0054] For example, 12 safety and quality index output values ​​are established: average porosity (%), maximum porosity of sensitive area (%), percentage of area with porosity exceeding the standard (%), mean absolute thickness deviation (mm), maximum positive thickness deviation (mm), maximum negative thickness deviation (mm), standard deviation of thickness deviation (mm), edge bonding strength (MPa), bonding strength at step (MPa), density of delamination defects (units / cm2), density of low-adhesion / high-adhesion defects (units / cm2), and density of wrinkle defects (units / cm2).

[0055]

[0056] Transform the output data: porosity indices y1, y2, and y3 are used... Changes (of which) Take 10 -5 This makes the data distribution closer to a normal distribution and improves the model's sensitivity to changes in porosity; the defect density index y 10 y 11 y 12 ,use The transformation makes the discrete counting data continuous, improving the Gaussian process's ability to model rare events. The thickness deviation indices y4, y5, y6, and y7 and the bond strength indices y8 and y9 retain their original values ​​to maintain the intuitiveness of their physical meaning, avoid introducing additional complexity due to the transformation, and facilitate the interpretation of subsequent results.

[0057] S34: Using the standardized combination of experimental parameters as input indicators and the modified measured data as output indicators, the multi-output Gaussian process regression model is trained and hyperparameters are optimized based on the log-marginal likelihood function to obtain the quality indicator prediction model.

[0058] In this step, the standardized experimental parameters are used as input metrics, and the transformed measured data are used as output metrics to form a training set. The model is trained and all hyperparameters are optimized simultaneously by maximizing the log-marginal likelihood function, including the length scale of each kernel function, signal variance, noise variance, and the weight matrix in the linear model kernel, so that the model best fits the actual observed data.

[0059] The log-marginal likelihood function is:

[0060] in, Let Y be the log-marginal likelihood of the observed data Y given input X; Y is the observed output vector, with dimension n×l; X is the input data matrix, with dimension n×d; K is the covariance matrix generated by the kernel function; I is the identity matrix, with dimension n×n. The covariance matrix is ​​noisy; It is a quadratic form used to measure the goodness of fit between data and the model; C is a constant, C= .

[0061] In the log-marginal likelihood function, the first term... This indicates the goodness of fit between the data and the model, measuring the degree of match between the predicted results and the actual observations; if the data fits well with the covariance structure defined by the kernel function, this value is smaller (a smaller negative value). The second term... This represents a penalty for model complexity, used to suppress overfitting; the larger the determinant of the covariance matrix, the higher the model complexity, and the more negative the value of this term, thus playing a regularization role.

[0062] Optionally, the optimization process can employ numerical optimization algorithms such as the conjugate gradient method or the quasi-Newton method. If the optimized l d If the value is very small, it indicates that the input data has a significant impact on the output metric, and the input needs to be strictly controlled; if... If the value is large, the measurement accuracy needs to be improved; if If the value is very small, the output index fluctuation range is small, indicating high process stability.

[0063] Furthermore, after training, a quality indicator prediction model is obtained. This model can take any new, untested standardized process parameters as input and output the predicted mean vector and complete covariance matrix of all preset quality indicators, thus providing both the predicted values ​​and the quantification results of their uncertainties.

[0064] Using the above methods, a multi-output Gaussian process regression model combining a squared exponential kernel and a linear model kernel was constructed. With targeted data preprocessing, this model achieved accurate learning and joint prediction of complex nonlinear relationships between multiple process parameters and quality indicators. This model not only outputs point prediction results but also provides valuable information on prediction uncertainties, thus laying a solid theoretical and computational foundation for subsequent active learning strategies and quantitative evaluation of process robustness.

[0065] S40: Based on the quality index prediction model and the preset constraints of various preset quality indices, perform a process window boundary search within the process parameter space to determine the feasible process window for repairing consumables and the number of sample points in the process parameter space that satisfy all preset constraints.

[0066] In this step, predefined constraints must be met for each preset quality indicator. Combined with the trained quality indicator prediction model, a process window boundary search is performed within the entire process parameter space. This search is achieved iteratively by executing the following steps: predicting the quality indicators and uncertainties of a large number of sample points in the parameter space using the current model; calculating the degree of violation of quality constraints at each sample point; intelligently selecting the next experimental point with the most information based on the uncertainty and the degree of constraint violation using a data acquisition function; conducting a real experiment at this point to obtain new data; and updating the model using the new data. This process is repeated until the process window boundary converges, ultimately determining the set of process parameters that satisfy all quality indicator constraints, thus obtaining the feasible process window. This window visually represents the safe operating area where the repair consumable can produce qualified repair parts. Subsequently, the number of sample points in the process parameter space that satisfy all preset constraints is obtained. This value represents the number of process parameter points determined to be within the safe operating area under the Monte Carlo sampling framework.

[0067] In one embodiment of this application, a specific feasible process window determination scheme is provided. In S40, based on the quality index prediction model and the preset constraints of various preset quality indices, a process window boundary search is performed in the process parameter space to determine the feasible process window for repair consumables and the number of sample points in the process parameter space that satisfy all preset constraints. Specifically, this includes the following steps S41-S46: S41: Based on the process parameter space, the first sample dataset is generated using the Latin hypercube sampling method; The first sample dataset includes multiple sample points, each of which consists of process parameters with different preset disturbance factors.

[0068] In this step, based on the process parameter space, the Latin hypercube sampling method is used to generate a large number of sample points (e.g., N=10,000) uniformly and without repetition in this multidimensional space. Each sample point corresponds to a combination of process parameters with different preset perturbation factors. All sample points together constitute the first sample dataset, which is used to evaluate and select the next experimental point with the most information.

[0069] S42: Input each sample point of the first sample dataset into the quality index prediction model to obtain the prediction data of each preset quality index under each sample point; The predicted data includes the predicted mean and predicted variance of each preset quality indicator.

[0070] In this step, the first sample dataset is input into the quality indicator prediction model. Within the framework of Gaussian process regression, the model outputs two key pieces of information for each sample point: the predicted mean and the predicted variance of each preset quality indicator. The predicted mean refers to the best estimate of the preset quality indicator most likely to be measured at each sample point, while the predicted variance characterizes the model's uncertainty about that predicted value, i.e., the model's confidence in the prediction.

[0071] S43: Calculate the acquisition function value for each sample point based on the predicted data and preset constraints.

[0072] In this step, based on the predicted data, the degree to which each sample point violates the pre-defined quality index constraints is calculated, and further combined with its prediction variance, the acquisition function value of each sample point is calculated. The acquisition function value is a comprehensive scoring index used in the active learning (Bayesian optimization) framework to quantify the information value of candidate sample points. This value identifies sample points that minimize model cognitive ambiguity (exploring unknown regions) and are most likely located near the process qualification boundary (utilizing existing knowledge). This guides each real experiment to target the parameter positions that contribute most to accurately characterizing the process window boundary, achieving efficient convergence to the optimal solution with minimal experimental cost.

[0073] In one embodiment of this application, a specific scheme for calculating the acquisition function value is provided. In S43, that is, based on the prediction data and preset constraints, the acquisition function value of each sample point is calculated, which specifically includes the following steps S431-S435: S431: Based on the difference between the predicted mean of each preset quality indicator and the value of the preset constraint, the violation amount of each preset quality indicator at each sample point is obtained.

[0074] S432: Use the maximum value of multiple violations as the degree of constraint violation for each sample point.

[0075] For steps S431-S432, the predicted mean of each preset quality indicator is substituted into the constraint violation degree calculation formula along with the value of its corresponding preset constraint condition to calculate the constraint violation degree of each preset quality indicator.

[0076] The formula for calculating the degree of constraint violation is:

[0077] in, The degree of constraint violation for sample point x; Let J be the violation amount of the j-th quality indicator at sample point x, where j = 1, 2, ..., J; The formula for calculating the number of quality indicator violations is:

[0078] in, Let x be the predicted mean of the j-th quality index at sample point x; Let x be the acceptable limit value for the j-th quality indicator at sample point x.

[0079] Specifically, for any sample point, the predicted mean of each preset quality indicator and the corresponding pass / fail limit value are substituted into the formula for calculating the violation amount of the quality indicator to assess whether the point exceeds the standard in each indicator. Here, the pass / fail limit value is the quantitative value of the constraint condition. For example, if the preset constraint condition for the average porosity is not higher than 2, then 2 is the pass / fail upper limit value for the average porosity.

[0080] Furthermore, the default value of the j-th quality index for sample point x is calculated as follows: ≤0 indicates the predicted mean of that point. Not exceeding the upper limit That is, the sample point x is qualified to predict the index; if If the value is greater than 0, it indicates that the predicted mean for that point is... Exceeded the limit That is, sample point x is predicted to be unqualified on this indicator, and The value is the excess. For example, if the predicted average porosity of sample point x is μ1(x) = 2.5, and the upper limit of acceptable value is a1 = 2, then... =2.5-2=0.5>0, indicating that the predicted average porosity of this point is unqualified and exceeds the standard by 0.5.

[0081] Then, all violations are substituted into the constraint violation degree calculation formula, and the maximum value is taken. If all If all values ​​are less than or equal to 0, then the maximum value is 0. =0 indicates that the prediction for this point is entirely accurate and it is within the estimated safe zone; if any g j (x)>0, then If the value equals the maximum positive violation, it means that at least one indicator at that point is predicted to be non-compliant, and The value reflects the severity of the violation for the most serious indicator; the higher the value, the more serious the violation.

[0082] S433: Based on the prediction variance of each preset quality indicator, determine the prediction standard deviation of each preset quality indicator.

[0083] S434: Obtain the target standard deviation that maximizes the predicted standard deviation.

[0084] For steps S433-S434, for each sample point x in the first sample dataset, the prediction standard deviation of each indicator is obtained by taking the square root of the prediction variance output by the quality indicator prediction model. This standard deviation intuitively represents the degree of confidence of the model in predicting the j-th quality indicator. The larger the value, the higher the cognitive uncertainty of the model at that point.

[0085] The formula for calculating standard deviation is:

[0086] in, Let be the predicted standard deviation of the j-th quality indicator; The predicted standard deviation and predicted variance of the j-th quality indicator.

[0087] Furthermore, in order to comprehensively reflect the overall uncertainty of sample point x across all quality indicators with a scalar, the maximum value among the predicted standard deviations of each indicator at that point is selected as its target standard deviation σ(x). This strategy ensures that as long as the predicted uncertainty of any quality indicator is high, the point is considered to be highly uncertain overall, thereby effectively guiding the algorithm to explore such information-scarce regions.

[0088] Alternatively, the target standard deviation can be determined solely based on the standard deviation of pre-selected key quality indicators (such as porosity and bond strength).

[0089] S435: Calculate the acquisition function value for each sample point based on the degree of constraint violation and the target standard deviation.

[0090] In this step, for any sample point x in the first sample dataset, the degree of constraint violation reflecting the severity of the violation and the target standard deviation representing the overall uncertainty of the model prediction are substituted into the formula for calculating the acquisition function value of the sample point to obtain the acquisition function value of the sample point.

[0091] The formula for calculating the collected function value is:

[0092] in, The value of the function collected at sample point x; As a balancing parameter, α∈[0,1], it is usually taken as 0.5 to balance the contributions of the two. The target standard deviation for sample point x; To constrain the degree of violation.

[0093] By combining the quantitative prediction uncertainty (target standard deviation) with the degree of constraint violation, an information-driven acquisition function was constructed to intelligently identify and prioritize the exploration of parameter points that are most valuable for reducing global uncertainty and clarifying process boundaries. This enabled efficient and accurate searching of process window boundaries, significantly improving the cost-effectiveness and scientific rigor of the evaluation process.

[0094] S44: Obtain the target sample point with the largest acquisition function value, and conduct a disturbance repair test on the repair consumables based on the target sample point to obtain the target measured data of each preset quality index under the target sample point.

[0095] In this step, the target sample point with the largest acquisition function value is selected. In a real or simulated maintenance environment, a complete disturbance repair test is performed using the repair consumables to be evaluated, strictly following the process parameters corresponding to the target sample point. After the test, the manufactured specimen is comprehensively inspected to obtain the target measured data of all preset quality indicators under this set of process parameters.

[0096] S45: Based on the target sample points, target measured data, multiple experimental parameter combinations and multiple measured data, an amplified training set is constructed, and the model parameters of the quality index prediction model are iteratively updated based on the amplified training set until the termination condition is met (when the change in the rate of change of the process window volume is less than the preset threshold or the number of iterations is equal to the maximum number of iterations), and the updated quality index prediction model is output.

[0097] In this step, the newly obtained target sample points and their corresponding measured data are added to the existing combination of multiple experimental parameters and measured datasets to form an expanded training set. Using this expanded training set, all model parameters of the quality index prediction model are retrained and optimized, thereby improving the model's predictive ability and accuracy in the process window boundary region. Subsequently, it is checked whether the current iteration meets the preset termination condition; if it does, the iteration process terminates, and the updated quality index prediction model is output; otherwise, the process returns to the step of calculating the prediction and acquisition functions based on the candidate sample set, and the next iteration begins.

[0098] Optionally, the termination conditions include: after each round of active learning iteration, determining a temporary feasible process window (temporary target sample dataset) based on the current model, obtaining the number of sample points in the temporary target sample dataset, calculating that the rate of change of the number of sample points in the feasible window between the current round and the previous round is less than a preset threshold (e.g., less than 5%), or the number of iterations has reached the preset maximum allowed number of experiments.

[0099] S46: Based on the updated quality index prediction model, preset constraints, and process parameter space, determine the feasible process window for repair consumables and the number of sample points in the process parameter space that satisfy all preset constraints.

[0100] In this step, a global evaluation is performed across the entire process parameter space based on the final updated, high-confidence quality indicator prediction model and all preset quality indicator constraints. The set of all process parameter points that satisfy all constraints is defined as the feasible process window for the repair consumable; this window represents the safe operating area precisely delineated through intelligent search. Subsequently, the number of sample points in the process parameter space that satisfy all preset constraints is obtained. This value represents the number of process parameter points determined to be within the safe operating area under the Monte Carlo sampling framework.

[0101] By combining the active learning framework with the multi-output Gaussian process model in the above manner, experimental resources can be actively guided to the process parameter points with the most information. This allows for rapid and accurate convergence to the true boundary of the process window with far fewer experiments than grid search or random search, significantly improving evaluation efficiency and reducing R&D costs.

[0102] In one embodiment of this application, a specific feasible process window search scheme is provided. In S46, based on the updated quality index prediction model, preset constraints, and process parameter space, the feasible process window for repair consumables and the number of sample points in the process parameter space that satisfy all preset constraints are determined. This specifically includes the following steps S461-S464: S461: Generate a second sample dataset based on the process parameter space using the Monte Carlo method.

[0103] S462: Input each sample point of the second sample dataset into the updated quality index prediction model to generate the predicted mean of each quality index for each sample point.

[0104] S463: Based on the predicted mean and preset constraints, determine the target sample dataset that satisfies all preset constraints in the process parameter space, and use it as a feasible process window for repair consumables.

[0105] For steps S461-S463, after obtaining a high-confidence updated quality indicator prediction model through active learning loops, Monte Carlo sampling is performed on the entire process parameter space to generate a massive set of sample points covering this space (typically 10,000 to 100,000), i.e., the second sample dataset. All sample points from the second sample dataset are input into the updated quality indicator prediction model, and the model outputs the predicted mean of each preset quality indicator for each sample point. Subsequently, for each sample point, its predicted mean is compared one by one with the preset constraints of each preset quality indicator to determine whether all predicted means meet the constraints. All sample points are traversed, and all sample points that simultaneously meet all preset constraints are selected to form the target sample dataset, which is the feasible process window for repair consumables. This window visually represents all safe operating areas in the process parameter space that can stably produce qualified repair parts.

[0106] The above method enables automated and objective identification of safe operating areas within complex, high-dimensional process parameter spaces. Compared to the traditional approach that relies on engineers' experience and manually explores process ranges through sporadic experiments, this method directly outputs clear, quantifiable, and visualized feasible process windows. This provides a solid and reliable data basis for subsequent consumable performance comparisons, significantly improving the scientific rigor, systematic approach, and efficiency of process development.

[0107] S464: Get the number of sample points contained in the target sample dataset.

[0108] In this step, the number of sample points contained in the target sample dataset is obtained. This number represents the number of process parameter points that are determined to be within the safe operating area under the Monte Carlo sampling framework.

[0109] S50: Generates the robustness index of repair consumables based on the process parameter space, feasible process window, and number of sample points.

[0110] In this step, after finding the feasible process window for repair consumables, the volume of the feasible process window is calculated using the process parameter space and the number of sample points that meet all preset constraints. Then, the robustness index of the repair consumables is calculated. This index is a relative evaluation index used to directly compare the performance of the consumables to be evaluated with the industry-recognized benchmark.

[0111] In one embodiment of this application, a specific robustness index calculation scheme is provided. In S50, the robustness index of repair consumables is generated based on the process parameter space, feasible process window, and number of sample points. This specifically includes the following steps S51-S55: S51: Obtain the spatial volume of the process parameter space.

[0112] In this step, the process parameter space is a multidimensional hypercube spanned by the fluctuation ranges of all preset disturbance factors, and its volume is obtained by calculating the product of the fluctuation range lengths of each factor.

[0113] The formula for calculating the volume of space is:

[0114] Among them, V X D represents the spatial volume; D represents the number of preset disturbance factors. The maximum value of the d-th disturbance factor; It represents the minimum value of the d-th disturbance factor.

[0115] S52: Get the total number of sample points contained in the second sample dataset.

[0116] In this step, after generating a second sample dataset covering the entire process parameter space through Monte Carlo sampling, the total number of sample points in the dataset is obtained.

[0117] S53: Calculate the target volume of the feasible process window based on the spatial volume, the number of sample points, and the total number of sample points.

[0118] In this step, the target volume of the feasible process window is obtained by multiplying the volume of the entire process parameter space by the probability that a random sample point falls within the window. The larger the target volume, the wider the allowable range of process parameter fluctuations for the repair consumable, and the stronger its production robustness.

[0119] The formula for calculating the feasible process window volume is:

[0120] Where PWV is the target volume; N in N represents the number of sample points. total This represents the total number of sample points.

[0121] S54: Obtain the reference volume of the baseline feasible process window.

[0122] In this step, in order to make a horizontal comparison, a recognized import benchmark repair consumable's process window volume under the same process parameter space and the same evaluation criteria is obtained as a reference volume.

[0123] S55: Generate a robustness index for repair consumables based on the ratio of the target volume to the reference volume.

[0124] In this step, the robustness index of the repair consumables is obtained by calculating the ratio of the target volume to the reference volume.

[0125] The robustness index is calculated using the following formula:

[0126] Wherein, RI is the robustness index; For the target volume; For reference volume.

[0127] S60: Based on the robustness index, generate evaluation results for repair consumables.

[0128] In this step, based on the calculated robustness index and according to preset evaluation criteria, evaluation results are automatically generated. Specifically, if the robustness index is greater than 1, it indicates that the process window volume of the current consumable to be evaluated is larger than the imported benchmark, and its process robustness is better. The evaluation result can be generated as "Process robustness is better than the benchmark, and replacement is recommended"; if the robustness index is equal to 1, it indicates that the process robustness of the two is comparable, and the evaluation result can be output as "Process robustness is comparable to the benchmark, and replacement requirements are met"; if the robustness index is less than 1, it indicates that the process robustness of the current consumable to be evaluated is worse than the benchmark, and the evaluation result can be generated as "Process robustness needs to be improved, and direct replacement is not recommended at present".

[0129] By introducing the robustness index—a standardized and relative evaluation indicator—the complex process of comparing process performance is transformed into an intuitive numerical result and clear decision-making recommendations. This completely solves the technical problem of traditional evaluations relying on subjective experience and being difficult to quantify, providing an objective, impartial, and auditable scientific basis for the selection of repair consumables.

[0130] In practical applications, a structured evaluation report is automatically generated based on all acquired key data and the final evaluation results. The report includes: the process window volume of the consumable to be evaluated and the baseline consumable; the calculated robustness index; the multiple preset disturbance factors used; the preset quality indicators and their corresponding constraints; and the clearly defined final evaluation results.

[0131] As can be seen, the above scheme, by introducing a series of controllable disturbance factors around the benchmark process parameters to simulate actual fluctuations, systematically tests the performance of repair consumables under non-ideal process conditions, thereby determining the safety boundary for producing qualified products. This expands the performance evaluation of consumables from traditional single-point verification to a multi-dimensional parameter space, providing a more comprehensive and practical evaluation system. By introducing the process window volume as an objective and comparable robustness quantification index, and using a Gaussian process regression model to efficiently predict the process space boundary, the high cost and low efficiency of traditional exhaustive testing methods are avoided. The process robustness, which determines the stability of repair consumables in the actual maintenance environment, is transformed into a quantifiable and comparable equivalent fixed process window and its metric, enabling the scientific screening of repair consumables and significantly reducing quality risks in actual maintenance operations.

[0132] In one embodiment, an evaluation device for composite material repair consumables is provided, which corresponds one-to-one with the evaluation method for composite material repair consumables in the above embodiments. For example... Figure 2 As shown, the evaluation device 100 for composite material repair consumables includes: an acquisition module 101, a first generation module 102, a model training module 103, a determination module 104, a second generation module 105, and a third generation module 106. Detailed descriptions of each functional module are as follows: The acquisition module 101 is used to acquire the process parameter space of multiple preset disturbance factors for repair consumables used for composite material repair, wherein the process parameter space includes the process parameter reference value and fluctuation range of each preset disturbance factor. The first generation module 102 is used to generate multiple test parameter combinations based on the process parameter space, and to conduct disturbance repair tests on repair consumables through multiple test parameter combinations, thereby generating multiple measured data of multiple preset quality indicators. The model training module 103 is used to train a multi-output Gaussian process regression model based on multiple combinations of experimental parameters and multiple measured data to obtain a quality index prediction model. The determination module 104 is used to perform a process window boundary search in the process parameter space based on the quality index prediction model and the preset constraints of various preset quality indices, and to determine the feasible process window for repair consumables and the number of sample points in the process parameter space that satisfy all preset constraints. The second generation module 105 is used to generate a robustness index for repair consumables based on the process parameter space, feasible process window and number of sample points. The third generation module 106 is used to generate evaluation results for repair consumables based on the robustness index.

[0133] In one embodiment, the first generation module 102 is specifically used for: Through orthogonal experimental design, based on orthogonal arrays and process parameter space, multiple combinations of experimental parameters are determined, where each combination of experimental parameters includes process parameters from different preset disturbance factors; Based on multiple combinations of test parameters, disturbance repair tests are performed on repair consumables to obtain measured data of various preset quality indicators under each combination of test parameters.

[0134] In one embodiment, the model training module 103 is specifically used for: A multi-output Gaussian process regression model is constructed using the squared exponential kernel function and the linear model kernel. Standardize the process parameters in multiple combinations of experimental parameters; The measured data of each preset quality indicator are processed to account for changes; The standardized experimental parameter combinations are used as input indicators, and the measured data after variation processing are used as output indicators. The multi-output Gaussian process regression model is trained and hyperparameters are optimized based on the log-marginal likelihood function to obtain the quality indicator prediction model.

[0135] In one embodiment, the determining module 104 is specifically used for: Based on the process parameter space, a first sample dataset is generated using the Latin hypercube sampling method. The first sample dataset includes multiple sample points, each of which consists of process parameters with different preset perturbation factors. Input each sample point of the first sample dataset into the quality index prediction model to obtain the prediction data of each preset quality index under each sample point. The prediction data includes the prediction mean and prediction variance of each preset quality index. Based on the predicted data and preset constraints, the acquisition function value of each sample point is calculated. Obtain the target sample point with the largest acquisition function value, and conduct a disturbance repair test on the repair consumables based on the target sample point to obtain the target measured data of each preset quality index under the target sample point; An augmented training set is constructed based on target sample points, target measured data, multiple combinations of experimental parameters, and multiple measured data. The model parameters of the quality index prediction model are iteratively updated based on the augmented training set until the termination condition is met, and the updated quality index prediction model is output. Based on the updated quality index prediction model, preset constraints, and process parameter space, the feasible process window for repair consumables and the number of sample points in the process parameter space that satisfy all preset constraints are determined.

[0136] In one embodiment, the determining module 104 is further configured to: Based on the difference between the predicted mean of each preset quality indicator and the value of the preset constraint, the violation amount of each preset quality indicator at each sample point is obtained. The maximum value of multiple violations is used as the degree of constraint violation for each sample point; Based on the predicted variance of each preset quality indicator, the predicted standard deviation of each preset quality indicator is determined. Find the target standard deviation that maximizes the predicted standard deviation; The acquisition function value for each sample point is calculated based on the degree of constraint violation and the target standard deviation.

[0137] In one embodiment, the determining module 104 is further configured to: Based on the process parameter space, a second sample dataset is generated using the Monte Carlo method. Input each sample point of the second sample dataset into the updated quality index prediction model to generate the predicted mean of each quality index for each sample point. Based on the predicted mean and preset constraints, a target sample dataset that satisfies all preset constraints is determined in the process parameter space as a feasible process window for repair consumables. Get the number of sample points contained in the target sample dataset.

[0138] In one embodiment, the third generation module 106 is specifically used for: Obtain the spatial volume of the process parameter space; Obtain the total number of sample points contained in the second sample dataset; Calculate the target volume of the feasible process window based on the spatial volume, the number of sample points, and the total number of sample points. Obtain the reference volume of the baseline feasible process window; The robustness index of repair consumables is generated based on the ratio of the target volume to the reference volume.

[0139] This invention provides an evaluation device 100 for composite material repair consumables. By introducing a series of controllable disturbance factors around benchmark process parameters to simulate actual fluctuations, the system tests the performance of repair consumables under non-ideal process conditions, thereby determining the safety boundary for producing qualified products. This expands consumable performance evaluation from traditional single-point verification to a multi-dimensional parameter space, providing a more comprehensive and practical evaluation system. By introducing the process window volume as an objective and comparable robustness quantification index, and using a Gaussian process regression model to efficiently predict the process space boundary, the high cost and low efficiency of traditional exhaustive testing methods are avoided. The process robustness, which determines the stability of repair consumables in actual maintenance environments, is transformed into a quantifiable and comparable equivalent curing process window and its metric, enabling scientific screening of repair consumables and significantly reducing quality risks in actual maintenance operations.

[0140] Specific limitations regarding the evaluation device for composite material repair consumables can be found in the limitations of the evaluation method for composite material repair consumables mentioned above, and will not be repeated here. Each module in the aforementioned evaluation device for composite material repair consumables can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the electronic device in hardware form, or stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of each module.

[0141] In one embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the aforementioned evaluation method for composite material repair consumables.

[0142] In one embodiment, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the above-described evaluation method for composite material repair consumables.

[0143] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or electronic device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0144] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0145] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0146] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for evaluating composite material repair consumables, characterized in that, include: A process parameter space for multiple preset disturbance factors of repair consumables used for composite material repair is obtained, wherein the process parameter space includes the process parameter reference value and fluctuation range of each preset disturbance factor. Based on the process parameter space, multiple test parameter combinations are generated, and the repair consumables are subjected to disturbance repair tests through the multiple test parameter combinations to generate multiple measured data of multiple preset quality indicators. Based on the combination of multiple experimental parameters and the multiple measured data, a multi-output Gaussian process regression model is trained to obtain a quality index prediction model. Based on the quality index prediction model and the preset constraints of each preset quality index, a process window boundary search is performed in the process parameter space to determine the feasible process window of the repair consumables and the number of sample points in the process parameter space that satisfy all preset constraints. Based on the process parameter space, the feasible process window, and the number of sample points, the robustness index of the repair consumables is generated. Based on the robustness index, the evaluation results of the repair consumables are generated; The step of performing a process window boundary search within the process parameter space based on the quality index prediction model and the preset constraints of each preset quality index, to determine the feasible process window for the repair consumables and the number of sample points in the process parameter space that satisfy all preset constraints, specifically includes: Based on the process parameter space, a first sample dataset is generated using the Latin hypercube sampling method. The first sample dataset includes multiple sample points, each of which consists of process parameters with different preset perturbation factors. Each sample point of the first sample dataset is input into the quality index prediction model to obtain the prediction data of each preset quality index under each sample point, wherein the prediction data includes the prediction mean and prediction variance of each preset quality index. Based on the predicted data and the preset constraints, calculate the acquisition function value for each sample point; Obtain the target sample point with the largest acquisition function value, and conduct a disturbance repair test on the repair consumables based on the target sample point to obtain the target measured data of each preset quality index under the target sample point; An augmented training set is constructed based on target sample points, target measured data, multiple combinations of experimental parameters, and multiple measured data. The model parameters of the quality index prediction model are iteratively updated based on the augmented training set until the termination condition is met, and the updated quality index prediction model is output. Based on the updated quality index prediction model, preset constraints, and the process parameter space, the feasible process window for the repair consumables and the number of sample points in the process parameter space that satisfy all preset constraints are determined. The step of determining the feasible process window and the number of sample points satisfying all preset constraints in the process parameter space based on the updated quality index prediction model, preset constraints, and the process parameter space specifically includes: Based on the process parameter space, a second sample dataset is generated using the Monte Carlo method. Input each sample point of the second sample dataset into the updated quality index prediction model to generate the predicted mean of each quality index for each sample point. Based on the predicted mean and preset constraints, a target sample dataset that satisfies all preset constraints is determined in the process parameter space as a feasible process window for repair consumables. Obtain the number of sample points contained in the target sample dataset.

2. The evaluation method for composite material repair consumables according to claim 1, characterized in that, The step of conducting disturbance tests on composite material repair consumables based on multiple sets of process parameters to generate multiple measured data of multiple preset quality indicators specifically includes: Through orthogonal experimental design, based on orthogonal arrays and process parameter space, the multiple combinations of experimental parameters are determined, wherein each combination of experimental parameters includes process parameters from different preset disturbance factors; Based on the multiple test parameter combinations, a disturbance repair test is performed on the repair consumables to obtain measured data of various preset quality indicators under each test parameter combination.

3. The evaluation method for composite material repair consumables according to claim 1, characterized in that, The step of training a multi-output Gaussian process regression model based on the multiple combinations of experimental parameters and the multiple measured data to obtain a quality index prediction model specifically includes: The multi-output Gaussian process regression model is constructed using the squared exponential kernel function and the linear model kernel. Standardize the process parameters in multiple combinations of experimental parameters; The measured data of each preset quality indicator are processed to account for changes; The standardized combination of experimental parameters is used as the input index, and the measured data after variation processing is used as the output index. The multi-output Gaussian process regression model is trained and its hyperparameters are optimized based on the log-marginal likelihood function to obtain the quality index prediction model.

4. The evaluation method for composite material repair consumables according to claim 1, characterized in that, The step of calculating the acquisition function value of each sample point based on the predicted data and the preset constraints specifically includes: Based on the difference between the predicted mean of each preset quality indicator and the value of the preset constraint, the violation amount of each preset quality indicator at each sample point is obtained. The maximum value of multiple violations is used as the degree of constraint violation for each sample point; Based on the predicted variance of each preset quality indicator, the predicted standard deviation of each preset quality indicator is determined. Find the target standard deviation that maximizes the predicted standard deviation; The acquisition function value for each sample point is calculated based on the degree of constraint violation and the target standard deviation.

5. The evaluation method for composite material repair consumables according to claim 1, characterized in that, The step of generating the robustness index of the repair consumables based on the process parameter space, the feasible process window, and the number of sample points specifically includes: Obtain the spatial volume of the process parameter space; Obtain the total number of sample points contained in the second sample dataset; Based on the spatial volume, the number of sample points, and the total number of sample points, calculate the target volume of the feasible process window; Obtain the reference volume of the baseline feasible process window; The robustness index of the repair consumable is generated based on the ratio of the target volume to the reference volume.

6. An evaluation device for composite material repair consumables, characterized in that, include: The acquisition module is used to acquire the process parameter space of multiple preset disturbance factors for repair consumables used for composite material repair, wherein the process parameter space includes the process parameter reference value and fluctuation range of each preset disturbance factor. The first generation module is used to generate multiple test parameter combinations based on the process parameter space, and to conduct disturbance repair tests on the repair consumables through the multiple test parameter combinations to generate multiple measured data of multiple preset quality indicators. The model training module is used to train a multi-output Gaussian process regression model based on the multiple combinations of experimental parameters and the multiple measured data to obtain a quality index prediction model. The determination module is used to perform a process window boundary search within the process parameter space based on the quality index prediction model and the preset constraints of each preset quality index, to determine the feasible process window of the repair consumables and the number of sample points in the process parameter space that satisfy all preset constraints. The second generation module is used to generate the robustness index of the repair consumables based on the process parameter space, the feasible process window, and the number of sample points. The third generation module is used to generate the evaluation results of the repair consumables based on the robustness index; Specifically, the determining module is used for: Based on the process parameter space, a first sample dataset is generated using the Latin hypercube sampling method. The first sample dataset includes multiple sample points, each of which consists of process parameters with different preset perturbation factors. Each sample point of the first sample dataset is input into the quality index prediction model to obtain the prediction data of each preset quality index under each sample point, wherein the prediction data includes the prediction mean and prediction variance of each preset quality index. Based on the predicted data and the preset constraints, calculate the acquisition function value for each sample point; Obtain the target sample point with the largest acquisition function value, and conduct a disturbance repair test on the repair consumables based on the target sample point to obtain the target measured data of each preset quality index under the target sample point; An augmented training set is constructed based on target sample points, target measured data, multiple combinations of experimental parameters, and multiple measured data. The model parameters of the quality index prediction model are iteratively updated based on the augmented training set until the termination condition is met, and the updated quality index prediction model is output. Based on the updated quality index prediction model, preset constraints, and the process parameter space, the feasible process window for the repair consumables and the number of sample points in the process parameter space that satisfy all preset constraints are determined. Specifically, the determining module is used for: Based on the process parameter space, a second sample dataset is generated using the Monte Carlo method. Input each sample point of the second sample dataset into the updated quality index prediction model to generate the predicted mean of each quality index for each sample point. Based on the predicted mean and preset constraints, a target sample dataset that satisfies all preset constraints is determined in the process parameter space as a feasible process window for repair consumables. Obtain the number of sample points contained in the target sample dataset.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the evaluation method for composite material repair consumables as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the evaluation method for composite material repair consumables as described in any one of claims 1 to 5.

Images