Color-coated steel plate weather resistance evaluation method based on multi-source data fusion

Abstract

The invention discloses a color-coated steel plate weather resistance evaluation method based on multi-source data fusion, and relates to the technical field of color-coated steel plate weather resistance evaluation. The method comprises the following steps: acquiring color-coated steel plate data based on Internet of Things sensing equipment and preprocessing the color-coated steel plate data; in an experimental environment, performing subitem weather resistance evaluation based on the preprocessed data of the color-coated steel plate, and determining a weather resistance evaluation grade of each subitem; if the weather resistance evaluation grades of all the sub-items are qualified, comprehensive weather resistance evaluation is conducted on the color-coated steel plate based on different simulated use environments, the risk grade of the color-coated steel plate in the simulated use environment after the weather resistance evaluation of the sub-items is determined, and multi-dimensional data are obtained through Internet of Things sensing equipment and fused and analyzed; practical accurate evaluation and comprehensive risk judgment of the appearance, the corrosion resistance and the mechanical property of the coating are achieved, and the problems that in an existing evaluation method, due to a single data source, the evaluation result is one-sided, and the weather resistance of the color-coated steel plate in the complex environment cannot be comprehensively reflected are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of weathering performance evaluation of color-coated steel plates, and in particular to a weathering performance evaluation method of color-coated steel plates based on multi-source data fusion. Background Art

[0002] Color-coated steel sheets are produced on a continuous production line using cold-rolled steel strip or galvanized steel strip (electrogalvanized or hot-dip galvanized). After surface pretreatment (degreasing and chemical treatment), they are coated with one or more layers of liquid coating using a roller coating method, followed by baking and cooling. Color-coated steel sheets must meet certain requirements for corrosion resistance and a novel appearance, making weather resistance assessment an integral part of the production process.

[0003] Chinese patent application publication number CN116759019A discloses a method for using machine learning to assist in establishing a weathering steel weathering resistance assessment model. The method includes: 1) establishing a machine learning model of weathering steel, its alloy composition, and service environment factors; 2) measuring the alloy composition of the steel to be tested, and detecting the environmental influencing factors of the service environment. The alloy composition and the environmental factors of the service environment are input into the model to obtain the corrosion rate of the weathering steel in this service environment, and its weathering resistance is evaluated based on the corrosion rate.

[0004] However, the data used in the existing evaluation methods are relatively single, resulting in one-sided evaluation results that are difficult to fully reflect the weather resistance of color-coated steel plates in complex environments. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a method for evaluating the weathering resistance of color-coated steel plates based on multi-source data fusion, which solves the problem that the evaluation results of the existing evaluation methods are one-sided due to a single data source and cannot fully reflect the weathering resistance of color-coated steel plates in complex environments.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for evaluating the weather resistance of color-coated steel plates based on multi-source data fusion, comprising the following steps: obtaining color-coated steel plate data based on an Internet of Things sensing device and preprocessing it; in an experimental environment, performing sub-item weather resistance evaluation based on the preprocessed color-coated steel plate data, and determining the weather resistance evaluation level of each sub-item, wherein the preprocessed color-coated steel plate data includes surface image data, electrochemical and corrosion morphology data, and macro- and micromechanical data, the sub-item weather resistance evaluation includes coating appearance performance evaluation, corrosion resistance performance evaluation, and mechanical performance evaluation, and the sub-item weather resistance evaluation level includes coating appearance performance evaluation level, corrosion resistance performance evaluation level, and mechanical performance evaluation level; if the sub-item weather resistance evaluation levels are all qualified, then performing a comprehensive weather resistance performance evaluation on the color-coated steel plate based on different simulated use environments, and determining the risk level of the color-coated steel plate in the simulated use environment after the sub-item weather resistance evaluation.

[0007] Furthermore, the coating appearance performance evaluation process is as follows: obtain the trained VGG16 convolutional neural network model and support vector machine model stored in the database; input the surface image data into the VGG16 convolutional neural network model to obtain the corresponding defect category prediction probability vector, including the color change defect prediction probability, the powdering defect prediction probability and the peeling defect prediction probability; perform feature processing on the surface image data to obtain color features, texture features and shape features, and concatenate the color features, texture features and shape features to form a multidimensional feature vector; input the multidimensional feature vector into the support vector machine model to obtain the defect classification result vector, including the color change defect classification result value, the powdering defect classification result value and the peeling defect classification result value. The method comprises the following steps: obtaining the result value and the peeling defect classification result value of the surface image data; obtaining the surface defect index, including the color difference uniformization value, the powdering area ratio and the peeling area ratio; processing the defect category prediction probability vector, the defect classification result vector and the surface defect index based on the coating appearance performance evaluation model to obtain the coating appearance performance evaluation coefficient; determining the level of the coating appearance performance evaluation coefficient based on the scoring rules stored in the database to obtain the coating appearance performance evaluation level, wherein the coating appearance performance evaluation level includes excellent coating appearance, good coating appearance, medium coating appearance and poor coating appearance. If the coating appearance performance evaluation level is excellent coating appearance or good coating appearance, the coating appearance performance evaluation level is recorded as qualified.

[0008] Furthermore, the training process of the VGG16 convolutional neural network model is as follows: retain the convolution layer of the classic VGG16 convolutional neural network, replace the fully connected layer, and set the output nodes of the fully connected layer to three, corresponding to color change defects, powdering defects and peeling defects respectively; annotate the surface images of color-coated steel plates taken regularly under different environments to obtain historical color-coated steel plate surface image defect data, where the annotations include color change area defect annotations, powdering defect area annotations and peeling defect area annotations; divide the historical color-coated steel plate surface image defect data into training set, validation set and test set; use the training set as input, adopt the cross entropy loss function, use the Adam optimizer, and set the learning rate to 1e-4 for training; perform verification and testing based on the verification set and test set, and output the VGG16 convolutional neural network model after the verification and testing are completed.

[0009] Furthermore, the support vector machine model training process is as follows: feature extraction is performed on historical color-coated steel plate surface image defect data to obtain color training features, texture training features and shape training features, and the color training features, texture training features and shape training features are concatenated to form a multidimensional training feature vector; principal component analysis is used to reduce the dimensionality of the multidimensional training feature vector to obtain a multidimensional training feature vector after dimensionality reduction; the multidimensional training feature vector after dimensionality reduction in the training set is used as the input of the support vector machine, and the defect classification result vector is predicted as the output. The kernel function and penalty parameter C of the SVM are optimized through grid search and cross-validation; verification and testing are performed based on the verification set and the test set, and the support vector machine model is output after verification and testing are completed.

[0010] Furthermore, the corrosion resistance evaluation process is as follows: a polarization curve test is performed on the color-coated steel plate to obtain a polarization curve, and the corrosion current density is obtained from the polarization curve; a corrosion weight loss test is performed on the color-coated steel plate to obtain the weight loss rate of the color-coated steel plate after corrosion; a color-coated steel plate sample after the corrosion weight loss test is obtained, and an electron microscope scanning observation is performed to obtain a microscopic corrosion morphology image, and a corrosion morphology score is determined based on the microscopic corrosion morphology image; the corrosion current density, the weight loss rate of the color-coated steel plate after corrosion and the corrosion morphology score are recorded as electrochemical and corrosion morphology data; the corrosion resistance evaluation grade is determined based on the electrochemical and corrosion morphology data, wherein the corrosion resistance evaluation grades include excellent corrosion resistance, good corrosion resistance, neutral corrosion resistance and poor corrosion resistance. If the corrosion resistance evaluation grade is excellent corrosion resistance or good corrosion resistance, the corrosion resistance evaluation grade is recorded as qualified.

[0011] Furthermore, the corrosion morphology score is determined based on the microscopic corrosion morphology image, including the following steps: performing feature extraction on the microscopic corrosion morphology image to obtain corrosion pit features, including corrosion pit shape, maximum cross-sectional area of corrosion pits, number of corrosion pits and corrosion pit distribution; performing energy spectrum analysis on the color-coated steel plate sample to determine corrosion products; concatenating the corrosion pit features and corrosion products into score comparison data, performing cosine similarity analysis on the score comparison data and each score pointing data stored in the database to obtain each cosine similarity comparison value; determining the score pointing data corresponding to the maximum cosine similarity comparison value, and determining the corrosion resistance performance evaluation grade corresponding to the score pointing data from the database.

[0012] Furthermore, the mechanical property evaluation process is as follows: a macroscopic mechanical test is performed on the color-coated steel plate to obtain a macroscopic mechanical property data set; a microscopic mechanical test is performed on the color-coated steel plate to obtain a microscopic mechanical property data set; the macroscopic mechanical property data set and the microscopic mechanical property data set are normalized to obtain macroscopic and microscopic mechanical data; the macroscopic and microscopic mechanical data are input into the trained mechanical vector machine model to obtain a mechanical property evaluation grade, where the mechanical property evaluation grade includes excellent mechanical property, good mechanical property, medium mechanical property and poor mechanical property. If the mechanical property evaluation grade is excellent mechanical property or good mechanical property, the mechanical property evaluation grade is recorded as qualified.

[0013] Furthermore, determining the risk level of the color-coated steel plate in the simulated use environment after the sub-item weathering evaluation includes the following steps: obtaining simulated weathering data of the color-coated steel plate in each simulated use environment after the sub-item weathering evaluation, including simulated surface image data, simulated corrosion morphology data and simulated environment matching data; obtaining simulated weathering parameter data stored in a database, including parameterized simulated surface image data, parameterized simulated corrosion morphology data and parameterized simulated environment matching data; performing cosine similarity analysis on the simulated weathering data and parameterized simulated surface image data in each simulated use environment after standardization to obtain the simulated environment weathering cosine similarity in each simulated use environment; determining the model diagnosis trigger threshold based on the performance of the color-coated steel plate in the experimental environment; If the cosine similarity of the simulated environmental weathering under any simulated use environment is less than the model diagnosis trigger threshold, the simulated weathering data under each simulated use environment is diagnosed based on the decision tree model to determine the abnormal category probability vector of the color-coated steel plate under each simulated use environment; the abnormal category probability vectors of the coated steel plate under each simulated use environment are spliced, and the cosine similarity is calculated with the various pointing abnormal category probability vectors stored in the database to obtain the cosine similarity comparison value; the pointing abnormal category probability vector corresponding to the maximum cosine similarity comparison value is determined, and the pointing abnormal category probability vector is determined to correspond to the risk level stored in the database in advance.

[0014] Furthermore, based on the performance of the color-coated steel plates in the experimental environment, a model diagnosis trigger threshold is determined, including the following steps: obtaining qualified performance data of the color-coated steel plates in the experimental environment, including the number of excellent performance and the number of good performance; performing weighted summation processing on the number of excellent performance and the number of good performance to obtain a qualified performance coefficient; and determining the model diagnosis trigger threshold based on a qualified performance coefficient-model diagnosis trigger threshold mapping set stored in a database.

[0015] Furthermore, the simulated weathering data under each simulated use environment is diagnosed based on the decision tree model to determine the abnormal category probability vector of the color-coated steel plate, including the following steps: determining the parameter weight data under each simulated use environment based on the simulated use environment-parameter weight mapping set stored in the database, including simulated surface image weight data, simulated corrosion morphology weight data and simulated environment matching weight data; performing weighted processing on the simulated weathering data under each simulated use environment based on the parameter weight data to obtain a fused simulated weathering vector under each simulated use environment; and inputting the fused simulated weathering vector under each simulated use environment into the decision tree model respectively to obtain the abnormal category probability vector of the color-coated steel plate under each simulated use environment.

[0016] The present invention has the following beneficial effects: This method for evaluating the weathering resistance of color-coated steel plates based on multi-source data fusion obtains multi-dimensional data through IoT sensing equipment and integrates and analyzes it, achieving precise sub-item evaluation and comprehensive risk judgment of the coating appearance, corrosion resistance, and mechanical properties, improving the comprehensiveness, accuracy, and reliability of the evaluation results, and solving the problem in existing evaluation methods that the evaluation results are one-sided and cannot fully reflect the weathering resistance performance of color-coated steel plates in complex environments due to a single data source.

[0017] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of the method for evaluating the weathering performance of color-coated steel plates based on multi-source data fusion according to the present invention; Figure 2 This is a flow chart for determining the risk level of color-coated steel plates in a simulated use environment after item-by-item weathering evaluation in the present invention. DETAILED DESCRIPTION

[0019] See also Figure 1 The embodiment of the present invention provides a technical solution: a method for evaluating the weathering performance of color-coated steel plates based on multi-source data fusion, comprising the following steps: obtaining color-coated steel plate data based on Internet of Things sensing equipment and preprocessing; preprocessing mainly includes data cleaning, i.e., removing image noise, eliminating numerical outliers and erroneous data; data conversion, quantifying non-numeric data and unifying the dimensions of numerical data; data missing processing, filling missing values through mean filling, interpolation or model prediction, and finally forming a clean and standardized data set that can be used for analysis.

[0020] In an experimental environment (for example, the experimental environment is centered on a climate simulation cabin with set temperature (-20℃-80℃), relative humidity (10%-95%), light intensity (0-2000W / m²), and salt spray concentration (50-500g / m³) to simulate typical weathering scenarios of high temperature and humidity, strong ultraviolet rays, and salt spray corrosion), a sub-item weathering evaluation is performed based on the pre-treated color-coated steel plate data to determine the weathering evaluation grade of each sub-item.

[0021] The pre-treated color-coated steel plate data includes surface image data, electrochemical and corrosion morphology data, and macro- and micro-mechanical data. The sub-item weathering evaluation includes coating appearance performance evaluation, corrosion resistance evaluation, and mechanical performance evaluation. The sub-item weathering evaluation grades include coating appearance performance evaluation grade, corrosion resistance evaluation grade, and mechanical performance evaluation grade. The coating appearance performance evaluation process is as follows: obtain the trained VGG16 convolutional neural network model and support vector machine model stored in the database; input the surface image data (extract color distribution and texture structure, etc. from the surface image data) into the VGG16 convolutional neural network model to obtain the corresponding defect category prediction probability vector, including the color change defect prediction probability, the chalking defect prediction probability and the peeling defect prediction probability; for example, the color change defect prediction probability, the chalking defect prediction probability and the peeling defect prediction probability are 0.85, 0.62, 0.25 and 0.16 respectively.

[0022] The surface image data is feature processed to obtain color features, texture features, and shape features (such as RGB mean, gray-level co-occurrence matrix texture features, and defect contour perimeter-to-area ratio). The color features, texture features, and shape features are concatenated to form a multidimensional feature vector. The multidimensional feature vector is input into the support vector machine model to obtain a defect classification result vector, including color change defect classification result values, powdering defect classification result values, and peeling defect classification result values (for example, a color change defect classification result value of 1 indicates the presence of the defect). The structured feature vector is highly interpretable, making it easier for technicians to understand the cause of the defect.

[0023] The surface image data is used to obtain indicators to obtain surface defect indicators, including color difference uniformity value, powdering area ratio and peeling area ratio.

[0024] Based on the coating appearance performance evaluation model, the defect category prediction probability vector, the defect classification result vector and the surface defect index are processed to obtain the coating appearance performance evaluation coefficient. It should be noted that the coating appearance performance evaluation coefficient can be obtained by: Perform weighted summation on each probability in the defect category prediction probability vector to obtain the defect probability coefficient; Perform weighted summation on each result value in the defect classification result vector to obtain the result value coefficient (it should be noted that the result value is 1 or 0); The defect index coefficient is obtained by weighted summing the color difference homogenization value, the powdering area ratio, and the peeling area ratio. Finally, the defect probability coefficient, result value coefficient and defect index coefficient are weighted and summed to obtain the coating appearance performance evaluation coefficient.

[0025] The coating appearance performance evaluation coefficient is graded based on the scoring rules stored in the database to obtain the coating appearance performance evaluation grade, where the coating appearance performance evaluation grade includes excellent coating appearance, good coating appearance, medium coating appearance and poor coating appearance. If the coating appearance performance evaluation grade is excellent coating appearance or good coating appearance, the coating appearance performance evaluation grade is recorded as qualified.

[0026] For example, the scoring rule is 0-0.2 corresponds to excellent coating appearance, 0.2-0.5 corresponds to good coating appearance, 0.5-0.7 corresponds to fair coating appearance, and 0.7-1 corresponds to poor coating appearance. If the coating appearance performance evaluation coefficient is 0.21, the grade is determined to be fair coating appearance.

[0027] Through the three-layer architecture of deep learning automatic feature extraction + traditional machine learning classification + quantitative indicator analysis, the automation, precision and standardization of coating appearance performance evaluation are achieved.

[0028] The training process of the VGG16 convolutional neural network model is as follows: retain the convolution layer of the classic VGG16 convolutional neural network (to extract basic image features), replace the fully connected layer, and set the output nodes of the fully connected layer to three, corresponding to color change defects, powdering defects, and peeling defects respectively; annotate the surface images of color-coated steel plates taken regularly under different environments (natural environment and artificial accelerated aging environment) to obtain historical color-coated steel plate surface image defect data, where the annotations include color change area defect annotation, powdering defect area annotation, and peeling defect area annotation; divide the historical color-coated steel plate surface image defect data into training set, validation set, and test set according to the ratio of 7:2:1; use the training set as input, adopt the cross entropy loss function, use the Adam optimizer, and set the learning rate to 1e-4 for training; perform verification and testing based on the verification set and test set, and output the VGG16 convolutional neural network model after the verification and testing are completed.

[0029] Intelligent detection of coating appearance defects provides a high-precision, low-latency solution, especially suitable for application scenarios with strict requirements on weather resistance, significantly reducing manual inspection costs and improving quality control.

[0030] The support vector machine model training process is as follows: feature extraction is performed on historical color-coated steel plate surface image defect data to obtain color training features, texture training features, and shape training features, which are then concatenated to form a multidimensional training feature vector; principal component analysis is used to reduce the dimensionality of the multidimensional training feature vector to obtain a reduced-dimensional multidimensional training feature vector, for example, retaining the principal components with a cumulative variance contribution rate ≥ 95%; reducing the feature space dimension can reduce computational complexity, increase model training speed, and improve the accuracy of subsequent SVM classification.

[0031] The multidimensional training feature vector after dimensionality reduction in the training set is used as the input of the support vector machine, and the predicted defect classification result vector is used as the output. The kernel function (such as the radial basis kernel function RBF) and the penalty parameter C of the SVM are optimized through grid search and cross-validation. K-fold cross-validation (such as k=5) is performed on each set of parameters. Verification and testing are performed based on the validation set and test set. After verification and testing are completed, the support vector machine model is output.

[0032] Through the standardized process of data dimensionality reduction-parameter optimization-model verification, the efficiency and accuracy of the support vector machine model in classifying coating defects have been improved, and the key issues of high-dimensional feature processing and model generalization have been solved. This has made the analysis of coating appearance performance in the sub-item weathering performance evaluation more reliable, laying the foundation for subsequent comprehensive weathering performance evaluation.

[0033] The corrosion resistance evaluation process is as follows: a polarization curve test is performed on the color-coated steel plate to obtain a polarization curve, and the corrosion current density is obtained from the polarization curve; a corrosion weight loss test is performed on the color-coated steel plate to obtain the weight loss rate of the color-coated steel plate after corrosion, where the weight loss rate = (mass before corrosion - mass after corrosion) / mass before corrosion × 100%; a color-coated steel plate sample after the corrosion weight loss test is obtained, and an electron microscope (SEM) scanning observation is performed to obtain a microscopic corrosion morphology image, and the corrosion morphology score is determined based on the microscopic corrosion morphology image; the corrosion current density, the weight loss rate of the color-coated steel plate after corrosion, and the corrosion morphology score are recorded as electrochemical and corrosion morphology data.

[0034] Corrosion current density is a key parameter for quantifying corrosion rate. It directly reflects the electrochemical corrosion activity of materials under specific conditions and provides a kinetic basis for corrosion resistance evaluation.

[0035] The corrosion resistance evaluation grade is determined based on the electrochemical and corrosion morphology data, where the corrosion resistance evaluation grades include excellent corrosion resistance, good corrosion resistance, neutral corrosion resistance and poor corrosion resistance. If the corrosion resistance evaluation grade is excellent corrosion resistance or good corrosion resistance, the corrosion resistance evaluation grade is recorded as qualified.

[0036] For example, each corrosion resistance assessment level corresponds to a set of baseline electrochemical and corrosion morphology data. Similarity analysis is performed between the acquired electrochemical and corrosion morphology data and the baseline electrochemical and corrosion morphology data. The closest baseline electrochemical and corrosion morphology data is identified, and the corrosion resistance assessment level is then determined. By integrating multi-dimensional corrosion data (electrochemical parameters, corrosion weight loss, and micromorphology), the one-sidedness of single-metric assessments is resolved, enabling a comprehensive analysis from corrosion rate to corrosion mechanism.

[0037] Determining a corrosion morphology score based on a microscopic corrosion morphology image includes the following steps: performing feature extraction (such as edge detection and threshold segmentation) on the microscopic corrosion morphology image to obtain corrosion pit features, including the shape of the corrosion pit (circular / irregular), the maximum cross-sectional area of the corrosion pit, the number of corrosion pits, and the distribution of the corrosion pits (such as uniform distribution / clustered distribution); performing energy spectrum analysis on the color-coated steel plate sample to determine the corrosion products (such as FeOOH, Fe3O4, etc.); concatenating the corrosion pit features and the corrosion products into score comparison data (such as [shape code, area, number, Cl content, S content]), performing cosine similarity analysis on the score comparison data and each score pointing data stored in a database to obtain each cosine similarity comparison value; determining the score pointing data corresponding to the maximum cosine similarity comparison value, and determining the corrosion resistance performance evaluation grade corresponding to the score pointing data from the database.

[0038] Through a three-tiered analysis framework comprised of electrochemical testing, macroscopic weight loss, and microscopic morphology, corrosion resistance assessment progresses from "rate quantification" to "mechanistic analysis," addressing the one-sidedness of traditional methods and enhancing the scientific nature of the assessment results. A standardized process combining image algorithms, energy spectrum analysis, and similarity matching transforms the complex characteristics of microscopic corrosion morphology into quantifiable and comparable data indicators, addressing the subjectivity and inefficiency of traditional manual assessments.

[0039] The mechanical properties evaluation process is as follows: macroscopic mechanical tests are conducted on color-coated steel sheets to obtain a macroscopic mechanical properties dataset (including tensile strength, yield strength, and elongation); microscopic mechanical tests are conducted on the color-coated steel sheets to obtain a microscopic mechanical properties dataset (including hardness and elastic modulus); the macroscopic and microscopic mechanical properties datasets are normalized to obtain macroscopic and microscopic mechanical data; the macroscopic and microscopic mechanical data are input into a trained mechanical vector machine model to obtain a mechanical properties evaluation grade, which includes excellent mechanical properties, good mechanical properties, moderate mechanical properties, and poor mechanical properties. If the mechanical properties evaluation grade is excellent or good, the mechanical properties evaluation grade is considered qualified. This technical approach of macroscopic and microscopic data fusion, normalization, and intelligent model evaluation addresses the one-sidedness and inefficiency of traditional mechanical properties evaluation.

[0040] If all weathering assessment sub-items meet the required rating, a comprehensive weathering performance assessment is conducted on the pre-coated steel sheet based on various simulated operating environments to determine the risk level of the pre-coated steel sheet in these simulated operating environments. These simulated operating environments include, but are not limited to, salt spray corrosion, strong UV radiation, high temperature and humidity, and high temperature oxidation, covering typical scenarios in actual applications. This generates assessment data that closely reflects real-world operating conditions, avoiding the limitations of a single laboratory environment.

[0041] like Figure 2 As shown, after the sub-item weathering evaluation, the simulated weathering data of the color-coated steel plate in each simulated use environment are obtained, including simulated surface image data, simulated corrosion morphology data and simulated environment matching data (different simulated use environments correspond to different simulated environment matching data, which are stored in the database. For example, the simulated weathering data corresponding to the salt spray corrosion environment are salt spray deposition, blister density and FeCl3 content); the simulated weathering parameter data stored in the database are obtained, including parameterized simulated surface image data, parameterized simulated corrosion morphology data and parameterized simulated environment matching data; the simulated weathering data and parameterized simulated surface image data under each simulated use environment are standardized and then cosine similarity analysis is performed to obtain the simulated environment weathering cosine similarity under each simulated use environment; based on the performance of the color-coated steel plate in the experimental environment, the model diagnosis trigger threshold is determined, such as the model diagnosis trigger threshold is set to 0.7.

[0042] If the cosine similarity of the simulated environmental weathering under any simulated use environment is less than the model diagnosis trigger threshold, the simulated weathering data under each simulated use environment is diagnosed based on the decision tree model to determine the abnormal category probability vector of the color-coated steel plate under each simulated use environment; the abnormal category probability vectors of the coated steel plate under each simulated use environment are spliced, and the cosine similarity is calculated with the various pointing abnormal category probability vectors stored in the database to obtain the cosine similarity comparison value; the pointing abnormal category probability vector corresponding to the maximum cosine similarity comparison value is determined, and the pointing abnormal category probability vector is determined to correspond to the risk level stored in the database in advance.

[0043] Through the full-process design of scenario simulation-data comparison-intelligent diagnosis, the problem of insufficient predictive ability of traditional evaluation methods in practical applications is solved.

[0044] Based on the performance of the color-coated steel plates in an experimental environment, a model diagnosis trigger threshold is determined, which includes the following steps: obtaining qualified performance data of the color-coated steel plates in the experimental environment, including the number of excellent performance and the number of good performance (performing multiple sub-item weathering evaluations, and counting the sum of the number of coatings with excellent appearance, excellent corrosion resistance, and excellent mechanical properties, which is recorded as the number of excellent performance, and counting the sum of the number of coatings with good appearance, good corrosion resistance, and excellent mechanical properties, which is recorded as the number of good performance); performing weighted summation processing on the number of excellent performance and the number of good performance to obtain a qualified performance coefficient; determining the model diagnosis trigger threshold based on the qualified performance coefficient-model diagnosis trigger threshold mapping set stored in the database, such as the model diagnosis trigger threshold is 0.6 when the qualified performance coefficient is ≥8, and the model diagnosis trigger threshold is 0.9 when the qualified performance coefficient is 6-8. The threshold setting process of data statistics-weighted calculation-dynamic mapping solves the problem that traditional fixed thresholds cannot adapt to material property fluctuations.

[0045] The simulated weathering data under each simulated use environment are diagnosed based on a decision tree model to determine the probability vector of abnormal categories of color-coated steel plates, including the following steps: based on the simulated use environment-parameter weight mapping set stored in the database, the parameter weight data under each simulated use environment are determined, including simulated surface image weight data, simulated corrosion morphology weight data and simulated environment matching weight data.

[0046] Analyze historical data and determine the impact weight of each parameter for different simulated usage environments (such as ocean salt spray, industrial acid rain, and strong ultraviolet rays on the plateau). For example: Marine salt spray environment: corrosion morphology weight (0.5) > surface image weight (0.3) > environment matching weight (0.2) (because salt spray corrosion is the dominant factor); Industrial acid rain environment: Environmental matching weight (0.4) > Corrosion morphology weight (0.3) > Surface image weight (0.3) (priority should be given to the reaction between acid components and coatings).

[0047] The weight data is stored in a database to form a "simulated usage environment-parameter weight mapping set".

[0048] After weighting the simulated weathering data for each simulated usage environment based on parameter weight data, a fused simulated weathering vector for each simulated usage environment was obtained. This fused simulated weathering vector for each simulated usage environment was then input into a decision tree model to obtain a probability vector for abnormal classification of the color-coated steel plate for each simulated usage environment. This technical approach of context-sensitive weight allocation, data fusion, and customized diagnosis addresses the lack of scenario adaptability of traditional fixed-weight assessments.

[0049] An electronic device comprises: a processor; and a memory, wherein computer program instructions are stored in the memory, and when the computer program instructions are executed by the processor, the processor executes the above-mentioned method for evaluating the weathering resistance of color-coated steel plates based on multi-source data fusion.

[0050] A computer-readable storage medium is used to store a program, which, when executed by a processor, implements the above-mentioned method for evaluating the weathering performance of color-coated steel plates based on multi-source data fusion.

[0051] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0052] The present invention is described with reference to flowcharts and / or block diagrams of systems, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0053] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0054] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0055] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0056] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for evaluating the weathering performance of color-coated steel plates based on multi-source data fusion, characterized in that: The following steps are involved: Acquire color-coated steel plate data and perform pre-processing based on IoT sensing equipment; Under the experimental environment, a sub-item weathering evaluation is performed based on the pre-treated color-coated steel plate data to determine the weathering evaluation grade of each sub-item. The pre-treated color-coated steel plate data includes surface image data, electrochemical and corrosion morphology data, and macro- and micro-mechanical data. The sub-item weathering evaluation includes coating appearance performance evaluation, corrosion resistance evaluation, and mechanical performance evaluation. The sub-item weathering evaluation grades include coating appearance performance evaluation grade, corrosion resistance evaluation grade, and mechanical performance evaluation grade. If the weather resistance assessment levels of each sub-item are qualified, a comprehensive weather resistance performance evaluation of the color-coated steel plate will be conducted based on different simulated usage environments to determine the risk level of the color-coated steel plate in the simulated usage environment after the sub-item weather resistance evaluation.

2. The method for evaluating the weathering performance of color-coated steel sheets based on multi-source data fusion according to claim 1, characterized in that: The coating appearance performance evaluation process is as follows: Get the trained VGG16 convolutional neural network model and support vector machine model stored in the database; The surface image data is input into the VGG16 convolutional neural network model to obtain the corresponding defect category prediction probability vector, including the color change defect prediction probability, the chalking defect prediction probability, and the peeling defect prediction probability; Perform feature processing on the surface image data to obtain color features, texture features and shape features, and then concatenate the color features, texture features and shape features to form a multi-dimensional feature vector; Inputting the multidimensional feature vector into the support vector machine model to obtain a defect classification result vector, including a color change defect classification result value, a powdering defect classification result value, and a peeling defect classification result value; Obtain indicators from surface image data to obtain surface defect indicators, including color difference uniformity value, powdering area ratio and peeling area ratio; Based on the coating appearance performance evaluation model, the defect category prediction probability vector, defect classification result vector and surface defect index are processed to obtain the coating appearance performance evaluation coefficient; The coating appearance performance evaluation coefficient is graded based on the scoring rules stored in the database to obtain the coating appearance performance evaluation grade, where the coating appearance performance evaluation grade includes excellent coating appearance, good coating appearance, medium coating appearance and poor coating appearance. If the coating appearance performance evaluation grade is excellent coating appearance or good coating appearance, the coating appearance performance evaluation grade is recorded as qualified.

3. The method for evaluating the weathering performance of color-coated steel sheets based on multi-source data fusion according to claim 2, characterized in that: The VGG16 convolutional neural network model training process is as follows: The convolutional layer of the classic VGG16 convolutional neural network is retained and replaced with the fully connected layer. The output nodes of the fully connected layer are set to three, corresponding to color change defects, chalking defects, and peeling defects respectively; The surface images of color-coated steel plates taken regularly under different environments are annotated to obtain historical surface defect data of color-coated steel plates, including the annotation of color change defect areas, chalking defect areas, and peeling defect areas. The historical color-coated steel plate surface image defect data is divided into training set, validation set and test set; The training set is used as input, the cross entropy loss function is adopted, the Adam optimizer is used, and the learning rate is set to 1e-4 for training; Verification and testing are performed based on the validation set and test set, and the VGG16 convolutional neural network model is output after verification and testing are completed.

4. The method for evaluating weathering resistance of color-coated steel sheets based on multi-source data fusion according to claim 2, characterized in that: The support vector machine model training process is as follows: Feature extraction is performed on the historical color-coated steel plate surface image defect data to obtain color training features, texture training features, and shape training features. The color training features, texture training features, and shape training features are connected in series to form a multi-dimensional training feature vector. Use principal component analysis to reduce the dimensionality of the multidimensional training feature vector to obtain the multidimensional training feature vector after dimensionality reduction; The multi-dimensional training feature vector after dimensionality reduction in the training set is used as the input of the support vector machine, and the predicted defect classification result vector is used as the output. The kernel function and penalty parameter C of the SVM are optimized through grid search and cross-validation. Verification and testing are performed based on the validation set and test set, and the support vector machine model is output after verification and testing are completed.

5. The method for evaluating weathering resistance of color-coated steel sheets based on multi-source data fusion according to claim 1, characterized in that: The corrosion resistance evaluation process is as follows: Polarization curve test is performed on the color-coated steel plate to obtain the polarization curve, and the corrosion current density is obtained from the polarization curve; Carry out corrosion weight loss test on color-coated steel plate to obtain weight loss rate of color-coated steel plate after corrosion; Obtain color-coated steel plate samples after the corrosion weight loss test, perform electron microscope scanning observation, obtain microscopic corrosion morphology images, and determine the corrosion morphology score based on the microscopic corrosion morphology images; The corrosion current density, weight loss rate of color-coated steel plate after corrosion and corrosion morphology score were recorded as electrochemical and corrosion morphology data; The corrosion resistance evaluation grade is determined based on the electrochemical and corrosion morphology data, where the corrosion resistance evaluation grades include excellent corrosion resistance, good corrosion resistance, neutral corrosion resistance and poor corrosion resistance. If the corrosion resistance evaluation grade is excellent corrosion resistance or good corrosion resistance, the corrosion resistance evaluation grade is recorded as qualified.

6. The method for evaluating weathering resistance of color-coated steel sheets based on multi-source data fusion according to claim 5, characterized in that: Determining the corrosion morphology score based on the microscopic corrosion morphology image includes the following steps: Extract features from microscopic corrosion morphology images to obtain corrosion pit characteristics, including corrosion pit shape, maximum cross-sectional area of corrosion pits, number of corrosion pits, and corrosion pit distribution; Energy spectrum analysis was performed on color-coated steel samples to identify corrosion products; The corrosion pit features and corrosion products are concatenated into score comparison data, and the score comparison data is subjected to cosine similarity analysis with each score pointing data stored in the database to obtain each cosine similarity comparison value; The scoring data corresponding to the maximum cosine similarity comparison value is determined, and the corrosion resistance evaluation grade corresponding to the scoring data is determined from the database.

7. The method for evaluating weathering resistance of color-coated steel sheets based on multi-source data fusion according to claim 1, characterized in that: The mechanical properties evaluation process is as follows: Macro-mechanical tests were conducted on color-coated steel sheets to obtain a macro-mechanical properties dataset; Micromechanical tests were conducted on color-coated steel sheets to obtain a micromechanical properties data set; Normalizing the macro-mechanical properties data set and the micro-mechanical properties data set to obtain macro- and micro-mechanical data; The macro- and micro-mechanical data are input into the trained mechanical vector machine model to obtain the mechanical property evaluation grade, where the mechanical property evaluation grade includes excellent mechanical performance, good mechanical performance, medium mechanical performance and poor mechanical performance. If the mechanical property evaluation grade is excellent mechanical performance or good mechanical performance, the mechanical property evaluation grade is recorded as qualified.

8. The method for evaluating weathering resistance of color-coated steel sheets based on multi-source data fusion according to claim 1, characterized in that: Determining the risk level of the color-coated steel sheet in the simulated use environment after the sub-item weathering assessment includes the following steps: Obtain simulated weathering data of color-coated steel sheets in various simulated usage environments after item-by-item weathering evaluation, including simulated surface image data, simulated corrosion morphology data, and simulated environment matching data; Acquire simulated weathering parameter data stored in the database, including parameter simulated surface image data, parameter simulated corrosion morphology data, and parameter simulated environment matching data; After standardizing the simulated weathering data and the number of reference simulated surface images under each simulated use environment, cosine similarity analysis is performed to obtain the simulated environmental weathering cosine similarity under each simulated use environment; Based on the performance of color-coated steel plates in the experimental environment, the model diagnosis trigger threshold is determined; If the simulated environment weathering cosine similarity under any simulated use environment is less than the model diagnosis trigger threshold, the simulated weathering data under each simulated use environment is diagnosed based on the decision tree model to determine the abnormal category probability vector of the color-coated steel plate under each simulated use environment; The abnormal category probability vectors of the coated steel plates under each simulated use environment are spliced together, and the cosine similarity is calculated with the probability vectors of each pointing abnormal category stored in the database to obtain the cosine similarity comparison value; The probability vector pointing to the abnormal category corresponding to the maximum cosine similarity comparison value is determined, and the probability vector pointing to the abnormal category is determined to correspond to the risk level pre-stored in the database.

9. The method for evaluating weathering resistance of color-coated steel sheets based on multi-source data fusion according to claim 8, characterized in that: Based on the performance of the color-coated steel plate in the experimental environment, the model diagnosis trigger threshold is determined, including the following steps: Obtain qualified performance data of color-coated steel sheets in experimental environments, including the number of sheets with excellent performance and the number of sheets with good performance; Perform weighted summation on the number of excellent performance and the number of good performance to obtain the qualified performance coefficient; The model diagnosis trigger threshold is determined based on a qualified performance coefficient-model diagnosis trigger threshold mapping set stored in a database.

10. The method for evaluating weathering performance of color-coated steel sheets based on multi-source data fusion according to claim 8, characterized in that: The simulated weathering data under each simulated use environment are diagnosed based on the decision tree model to determine the abnormal category probability vector of the color-coated steel plate, including the following steps: Determining parameter weight data under each simulated use environment based on the simulated use environment-parameter weight mapping set stored in the database, including simulated surface image weight data, simulated corrosion morphology weight data, and simulated environment matching weight data; After weighted processing of the simulated weathering data under each simulated use environment based on the parameter weight data, a fusion simulated weathering vector under each simulated use environment is obtained; The fused simulated weathering vectors under each simulated usage environment are input into the decision tree model respectively to obtain the abnormal category probability vectors of the color-coated steel plate under each simulated usage environment.

Images