Structural ceramic bending strength prediction method and system and storage medium

Abstract

The invention discloses a structural ceramic bending strength prediction method and system and a storage medium, and belongs to the technical field of material performance evaluation.The method comprises the steps that microscopic images of potential fractures of to-be-detected ceramic are processed through image processing and machine vision, and a to-be-detected hole defect feature set is obtained; inputting the to-be-detected defect feature set into a pre-trained bending strength prediction model to obtain the bending strength of the to-be-detected ceramic, the acquisition method of the bending strength prediction model comprises the following steps: processing a historical microscopic image through image processing and machine vision to obtain a hole defect feature set of each sample; obtaining a key feature set according to the hole defect feature set; respectively training a plurality of pre-constructed machine learning algorithm models according to the key feature set and the bending strength value of each sample to obtain a plurality of prediction models; and optimizing the plurality of prediction models to obtain a bending strength prediction model, thereby realizing accurate extraction of micro defects on the surface of the ceramic and accurate prediction of the strength of the ceramic.

Description

Technical Field

[0001] The present invention belongs to the technical field of material performance evaluation, and in particular relates to a method, system and storage medium for predicting the bending strength of structural ceramics. Background Art

[0002] Silicon nitride ceramics have been widely used in high-tech fields such as aerospace, mechanical equipment, and medical devices due to their excellent properties such as high strength, high hardness, high temperature resistance, and corrosion resistance. Considering the intrinsic brittleness and strong defect sensitivity of ceramics, under the action of tensile stress, defects are prone to cause crack initiation and rapid expansion due to stress concentration, leading to catastrophic damage. Therefore, silicon nitride ceramics are currently mainly used in compressive stress environments, such as bearing balls, cutting tools, etc. In order to broaden its application scenarios, it is necessary to deeply understand the harm of microscopic defects represented by holes to material strength in non-compressive stress environments.

[0003] Although there has been great progress in the field of quantitative prediction of micropore-strength, it has remained at the level of describing the harmfulness of simple-shaped holes (elliptical, approximately rectangular, etc.), and the strength attenuation under the action of inherently complex micropores is still difficult to effectively evaluate. According to the different types of descriptors in the feature space, it can be divided into: considering elemental composition, additives, raw material characteristics and sintering process; considering microstructures dominated by grain size and grain boundaries. The existing models are good at dealing with the mechanical properties of ceramics with different components, but cannot effectively explain the randomness of ceramic strength when the material system is the same but the microstructure has slight differences due to differences in sampling positions. Considering the disadvantages of high-precision CT detection equipment, such as high cost, long time, and general reproducibility of detection results, it is urgent to find an efficient, low-cost, and easily reproducible defect detection and extraction method to accurately predict the strength of ceramics based on the extracted micropore defects to meet the dual needs of industrial production for quality control and efficiency improvement. Summary of the invention

[0004] The present invention provides a method, system and storage medium for predicting the flexural strength of structural ceramics. By combining image processing and machine vision, the hole defect features in the microscopic image of the structural ceramics are extracted, and then a pre-trained flexural strength prediction model with the best performance is used to predict the extracted hole defect features, thereby obtaining the flexural strength evaluation result of the structural ceramics under the action of complex micropores.

[0005] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0006] In a first aspect, the present invention provides a method for predicting the flexural strength of structural ceramics, comprising:

[0007] Obtaining a microscopic image of the potential fracture surface of the structural ceramic to be tested;

[0008] Processing the microscopic image of the potential fracture through image processing and machine vision to obtain a feature set of the hole defect to be tested;

[0009] Inputting the hole defect feature set to be tested into a pre-trained bending strength prediction model to obtain the bending strength of the structural ceramic to be tested;

[0010] Wherein, the method for obtaining the bending strength prediction model includes:

[0011] Obtain the historical microscopic images of the potential fractures on the surface of each sample in the structural ceramic bending strength test and the corresponding bending strength values ​​of each sample;

[0012] Processing the historical microscopic images by image processing and machine vision to obtain a hole defect feature set of each sample;

[0013] Obtaining a key feature set corresponding to each sample according to the hole defect feature set of each sample;

[0014] According to the key feature sets and flexural strength values ​​corresponding to each sample, multiple pre-built machine learning algorithm models are trained separately to obtain multiple prediction models;

[0015] The multiple prediction models are optimized to obtain a bending strength prediction model.

[0016] Optionally, obtaining a microscopic image of a potential fracture on the surface of the structural ceramic to be tested includes: collecting a microscopic image of a surface area of ​​the structural ceramic by a scanning electron microscope, wherein the collection range of the microscopic image is an area extending outward by 30 to 40% of the span from the geometric center of the structural ceramic to be tested as the center point.

[0017] Optionally, the historical microscopic images are processed by image processing and machine vision to obtain a hole defect feature set of each sample, including:

[0018] Perform image preprocessing on the historical microscopic images of each sample to obtain a valid image set of each sample;

[0019] Performing image segmentation on the effective image set of each sample to obtain a target area image set of each sample, wherein the target area image contains hole defects on the surface of the sample;

[0020] The feature extraction of the target area image set of each sample is performed through machine vision to obtain a geometric feature set of hole defects on the surface of each sample.

[0021] Optionally, obtaining a key feature set corresponding to each sample according to the hole defect feature set of each sample includes:

[0022] Standardizing the geometric feature set of the hole defects on the surface of each sample to obtain a standard feature set of each sample;

[0023] The standard feature set of each sample is screened for correlation to obtain a key feature set of each sample.

[0024] Optionally, the geometric feature set of the hole defects on the surface of each sample is standardized to obtain a standard feature set of each sample, including:

[0025] The characteristic values ​​in the geometric characteristic set of each sample are averaged and statistically analyzed to obtain the average characteristic set of each sample;

[0026] The average feature set of each sample is normalized to obtain a standard feature set.

[0027] Optionally, the correlation screening of the standard feature set of each sample to obtain the key feature set of each sample includes:

[0028] The Spearman correlation calculation was performed on the standard feature set of each specimen to obtain the correlation between each standard feature and the flexural strength value;

[0029] Weakly correlated features in the standard feature set of each sample are removed, and Pearson correlation calculation is performed on the standard feature set after the weakly correlated features are removed to obtain the correlation between the standard features of each sample;

[0030] The standard features with strong correlation between features are classified into the same group of features, and the standard features with strong correlation with the bending strength value are selected from the same group of features of each specimen to represent the group of features, so as to obtain the key feature set corresponding to each specimen.

[0031] Optionally, the pre-built multiple machine learning algorithm models are trained separately according to the key feature sets and flexural strength values ​​corresponding to each sample to obtain multiple prediction models, including:

[0032] Acquire a sample set according to the key feature set corresponding to each sample and the bending strength value corresponding to each sample;

[0033] Dividing the sample set into a training set, a validation set and a test set;

[0034] The training set is respectively input into a plurality of different pre-built machine learning algorithm models, the machine learning algorithm models are trained, and during the training process, the hyperparameters of the machine learning algorithm models are adjusted using the validation set to obtain a plurality of trained prediction models, wherein the key feature set in the training set is the input feature of the model training, and the bending strength value is the target feature of the model training.

[0035] Optionally, optimizing the multiple prediction models to obtain the bending strength prediction model includes: inputting the test set into the multiple trained prediction models respectively, evaluating the results output by each prediction model through the determination coefficient R² and the root mean square error RMSE, and obtaining the prediction model with the highest score as the bending strength prediction model.

[0036] In a second aspect, the present invention provides a structural ceramic bending strength prediction system, comprising:

[0037] Fracture image acquisition module: used to obtain the microscopic image of the potential fracture on the surface of the structural ceramic to be tested;

[0038] Defect feature acquisition module: processes the microscopic image of the potential fracture through image processing and machine vision to obtain a feature set of the hole defect to be tested;

[0039] Bending strength prediction module: used for inputting the hole defect feature set to be tested into a pre-trained bending strength prediction model to obtain the bending strength of the structural ceramic to be tested;

[0040] Wherein, the method for obtaining the bending strength prediction model includes:

[0041] Obtain the historical microscopic images of the potential fractures on the surface of each sample in the structural ceramic bending strength test and the corresponding bending strength values ​​of each sample;

[0042] Processing the historical microscopic images by image processing and machine vision to obtain a hole defect feature set of each sample;

[0043] Obtaining a key feature set corresponding to each sample according to the hole defect feature set of each sample;

[0044] According to the key feature sets and flexural strength values ​​corresponding to each sample, multiple pre-built machine learning algorithm models are trained separately to obtain multiple prediction models;

[0045] The multiple prediction models are optimized to obtain a bending strength prediction model.

[0046] In a third aspect, the present invention provides a computer storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for predicting the bending strength of structural ceramics as described in any step of the first aspect is implemented.

[0047] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: by combining image processing algorithms with machine vision, accurate extraction of tiny defects on the ceramic surface in historical samples is achieved, and then feature screening is performed through the extracted defect information to obtain key features that play an important role in the prediction of bending strength, and then a high-precision and generalized machine learning bending strength prediction model is constructed through the screened key features, which enables effective prediction of the bending strength of structural ceramics based on random microscopic defects on the surface of structural ceramics. The problem of difficulty in accurately evaluating the ceramic strength caused by the randomness of the ceramic strength when the structural ceramic material system is the same but there are slight differences in the microstructure is solved by efficient means, meeting the dual needs of industrial production for quality control and efficiency improvement. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Shown is a flow chart of a method for predicting the bending strength of structural ceramics in one embodiment of the present invention;

[0049] Figure 2 Shown is a flow chart of extracting surface defects of structural ceramics by machine vision in one embodiment of the present invention;

[0050] Figure 3 FIG. 3 is a schematic diagram of a visualization process of extracting defect features from a microscopic image of a potential fracture in an embodiment of the present invention, wherein 3(a) represents a microscopic image of a potential fracture scanned by a microscope, 3(b) is an image after preprocessing 3(a), 3(c) is an image after image segmentation of 3(b), and 3(d) is a minimum circumscribed circle area A and convex hull area A obtained by feature extraction of 3(c). c 3(e) is a magnified image of 3(a), 3(f) is a magnified image of 3(b), 3(g) is a magnified image of 3(c), 3(h) is a microscopic image of another potential fracture scanned by a microscope, 3(i) is an image after preprocessing of 3(h), 3(j) is an image after image segmentation of 3(i), 3(k) is a schematic diagram of feature extraction of 3(j) to obtain the major axis a and minor axis b of the circumscribed minimum ellipse and the angle θ, 3(l) is a magnified image of 3(h), 3(m) is a magnified image of 3(i), and 3(n) is a magnified image of 3(j);

[0051] Figure 4 Shown is a schematic diagram of the comparison and verification results of the predicted value and measured value of the AdaBoost algorithm prediction model in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0053] Example 1

[0054] like Figure 1 As shown, this embodiment provides a method for predicting the bending strength of structural ceramics, comprising:

[0055] Obtaining a microscopic image of the potential fracture surface of the structural ceramic to be tested;

[0056] Processing the microscopic image of the potential fracture through image processing and machine vision to obtain a feature set of the hole defect to be tested;

[0057] Inputting the hole defect feature set to be tested into a pre-trained bending strength prediction model to obtain the bending strength of the structural ceramic to be tested;

[0058] Wherein, the method for obtaining the bending strength prediction model includes:

[0059] Obtain the historical microscopic images of the potential fractures on the surface of each sample in the structural ceramic bending strength test and the corresponding bending strength values ​​of each sample;

[0060] Processing the historical microscopic images by image processing and machine vision to obtain a hole defect feature set of each sample;

[0061] Obtaining a key feature set corresponding to each sample according to the hole defect feature set of each sample;

[0062] According to the key feature sets and flexural strength values ​​corresponding to each sample, multiple pre-built machine learning algorithm models are trained separately to obtain multiple prediction models;

[0063] The multiple prediction models are optimized to obtain a bending strength prediction model.

[0064] Through microscopic image processing combined with machine vision technology, the characteristic data of the holes on the surface of each sample are extracted, and then the extracted characteristic data are processed and optimized and correlation screened to obtain the standardized key feature space of each sample. Combined with the actual flexural strength of each sample, the models constructed by multiple machine learning algorithms are trained to obtain a flexural strength prediction model with high precision and strong generalization ability, which realizes the subsequent guidance of accurate and efficient flexural strength prediction of the structural ceramics to be tested.

[0065] Example 2

[0066] Based on Example 1, this example also makes the following design.

[0067] The method for predicting the flexural strength of structural ceramics is divided into the following steps:

[0068] Step S1: using a scanning electron microscope to collect images of potential fracture areas on the surface of silicon nitride ceramics.

[0069] In this embodiment, the silicon nitride ceramic is in the shape of a bar, with a size of 6×5×48 mm³, the image collection area is 6×48 mm² surface, the photographing area is 6×10 mm², and the center line of the material extends 5 mm on both sides in the span direction. Figure 2 As shown, a serpentine-like photography method was used, the microscope magnification was 100x, 60 images were taken for each sample, and the size of each image was 1280 pixels × 960 pixels.

[0070] Step S2: Through machine vision and image processing technology, the geometric information of the surface hole defects is extracted as the initial feature set.

[0071] Among them, the process of extracting defects through machine vision refers to Figure 2 ,It is divided into four parts: image acquisition, image preprocessing, image segmentation and feature extraction.

[0072] Image acquisition is the image collection using a scanning electron microscope in step S1. The potential fracture area images obtained by scanning are shown in Figures 3(a) and 3(h). For easy viewing, Figures 3(a) and 3(h) are enlarged in proportion to obtain Figures 3(e) and 3(l) for display.

[0073] Image preprocessing: Median filtering and top-hat transformation are combined to reduce problems such as uneven illumination and noise in the image during microscopic photography. The median filter uses a 5×5 matrix window to calculate the median of each pixel, and the top-hat transformation kernel size is set to 50×50. The preprocessed image is shown in the figure below. Figure 3 As shown in Figures 3(b) and 3(i), for easier viewing, Figures 3(b) and 3(i) are enlarged in equal proportion to obtain Figures 3(f) and 3(m) for display.

[0074] Image segmentation: The image is segmented using the region growing algorithm. The grayscale histogram analysis based on multiple images is used to determine the grayscale value distribution difference between the background and the target. The seed pixels are manually selected between 80 and 120 to ensure the representativeness of the selected area. The difference between the maximum grayscale pixel value and the seed pixel grayscale value is used as the similarity criterion for iterative region expansion. The region area threshold is set to 100. If it is less than this value, it is regarded as a noise point and is not extracted. The image obtained after image segmentation is as follows Figure 3 As shown in 3(c) and 3(j), for easy viewing, 3(c) and 3(j) are enlarged in proportion to obtain 3(g) and 3(n) for display.

[0075] Feature extraction: With the help of machine vision, the size, shape, position and orientation information of the hole are obtained from the segmented image, and the minimum circumscribed circle diameter D can be extracted. c , perimeter P, area A, convex hull area A c (the number of pixels of the minimum convex enclosed package containing any concave shape), the major axis a and minor axis b of the circumscribed minimum ellipse, the angle θ (the angle between the major axis a of the hole and the y-axis direction of the sample, between 0° and 90°), the absolute coordinates of the center of mass (X, Y) and other geometric information. Defect features extracted using machine vision are as follows: Figure 3 As shown in 3(d) and 3(k), the minimum circumscribed circle area A and the convex hull area A are extracted in 3(d). c , the major axis a, minor axis b and angle θ of the circumscribed minimum ellipse are extracted in 3(k).

[0076] Based on the geometric information extracted by machine vision, the characteristic information such as convexity ratio CR, major-minor axis ratio AR, roundness C, shape coefficient S, eccentricity e, and relative coordinates of center of mass (x, y) are calculated and the obtained hole geometric feature data set is used as the initial feature data set. The calculation formulas of geometric features are as follows:

[0077] ,

[0078] ,

[0079] ,

[0080] ,

[0081] ,

[0082] ,

[0083] ,

[0084] Where W is the specimen width and L is the bending test span.

[0085] Step S3: Optimize the initial feature set to obtain a key feature set.

[0086] Feature average processing: select the top 30 hole data by diameter for average statistics and calculate the average diameter D c-avg , area and A sum , average shape factor S avg , average major-minor ratio AR avg , average convexity ratio CR avg , average roundness C avg , average angle θ avg , average eccentricity emax , average absolute coordinate (X avg ,Y avg ), average relative coordinate (x avg ,y avg ) and other eigenvalues, and construct the calculated data set into a new feature space.

[0087] Feature normalization processing: The newly constructed feature space data is processed by Z-score normalization. The processed data conforms to the standard normal distribution, that is, the mean is 0 and the standard deviation is 1, so as to obtain a standard feature data set; the Z-score normalization calculation formula is as follows:

[0088] ,

[0089] Among them, x is the original data value of the feature, z j is the normalized value, and are the mean and standard deviation of each feature respectively.

[0090] Feature screening: Perform correlation analysis and screening on multiple features to obtain the final input features.

[0091] The Spearman correlation calculation was performed on the standard feature data set. The absolute value of the Spearman correlation coefficient was less than 0.05, which was considered to be weakly correlated with the bending strength value, and the features with weak correlation were directly removed.

[0092] For the data set after the Spearman correlation calculation and screening, the Pearson correlation calculation was used, and the absolute value of the Pearson correlation coefficient greater than 0.8 was considered a strong correlation. Features with strong correlation were classified into the same group. For each group, the features with a high degree of correlation with the bending strength value were screened out according to the Spearman correlation coefficient analysis to represent the features of the group. The representative features of each group were combined into a key feature set as the input features.

[0093] In this embodiment, the average convexity ratio CR is screened out. avg , average y relative coordinate y avg , average diameter D c-avg , maximum hole diameter and average angle θ avg as input features.

[0094] Step S4: Construct and train multiple prediction models based on multiple machine learning algorithms.

[0095] The bending strength value is taken as the target feature, and the input features of each sample and its corresponding target features are combined to form a sample set.

[0096] In this embodiment, in the sample set, nine samples with the fracture source located on the lower surface of the sample are selected as the training set and the test set according to the position of the fracture source in the fracture surface of the sample, and the machine learning algorithm model is trained and adjusted by the training set and the test set respectively to obtain multiple prediction models.

[0097] In this embodiment, machine learning algorithm models are constructed according to each machine learning algorithm, and the machine learning algorithm models are adaptive boosting model AdaBoost, extreme gradient boosting model XGBoost, support vector machine model SVR, decision tree model DT, K nearest neighbor classification KNN, extreme random number model ETR and random forest model RF.

[0098] During the training and parameter adjustment process, network search is used to tune the model parameters of the adaptive boosting model. Network search is to call GridSearch CV in the Scikit-learn library in Python. The parameter values ​​after model tuning are shown in Table 1 below:

[0099] Table 1: Optimal values ​​of hyperparameters for the Adaptive Boosting model (AdaBoost)

[0100] parameter Search Space Optimal parameter values n_estimators [30, 50, 100, 150, 200] 150 learning_rate [0.01, 0.05, 0.1, 0.5] 0.1 max_depth [2, 3, 4, 5] 2

[0101] Step S5: Evaluate the performance of each prediction model to obtain the bending strength prediction model with the best performance.

[0102] In this embodiment, in the sample set, according to the location of the fracture source in the fracture surface of the sample, three samples with the fracture source located at one sample surface and two sample edges are used as the validation set to verify the accuracy and generalization of the trained model. In this embodiment, the performance of each prediction model is comprehensively evaluated based on the determination coefficient R², the root mean square error RMSE and the average relative error Validation error. The calculation formulas of the determination coefficient R² and the root mean square error RMSE are as follows:

[0103] ,

[0104] ,

[0105] Among them, n is the number of samples in the validation set, is the predicted value of the bending strength of the i-th validation set sample, is the true value of the bending strength of the i-th validation set sample, is the average bending strength of the validation set samples.

[0106] The performance of each prediction model is evaluated using the validation set data. The performance graphs of different prediction models are shown in Table 2 below:

[0107] Table 2: Performance of prediction models constructed by different algorithms

[0108] Algorithm Model AdaBoost XGBoost SVR DT KNN ETR RF R² 0.96 0.99 0.99 0.95 0.93 0.98 0.95 RMSE 17.7 4.26 0.14 18.1 22.2 10.4 18.4 Validation error 0.030 0.112 0.078 0.113 0.110 0.055 0.056

[0109] As can be seen from Table 2, the prediction error of the prediction model based on the AdaBoost algorithm is very low. In contrast, although the R² values ​​of the XGBoost, SVR and ETR machine learning algorithm models are all very high, the verification data all have large errors, and even dangerous situations where the predicted values ​​are significantly higher than the measured values. Therefore, the prediction model based on the AdaBoost algorithm has the best comprehensive performance.

[0110] from Figure 4 It can be seen that the prediction model built based on the AdaBoost algorithm can effectively predict the destructive strength of the samples with fracture sources located on the surface or near the ridges in the verification set, with an error of no more than 5%.

[0111] In this embodiment, the prediction model constructed based on the AdaBoost algorithm is used as the bending strength prediction model for the subsequent prediction process.

[0112] Step S6: According to the contents of steps S1-S3, a key feature set of surface defects of the silicon nitride ceramic to be tested is obtained as input features, and the input features are input into a bending strength prediction model based on AdaBoost to obtain a predicted bending strength value of the silicon nitride ceramic to be tested.

[0113] Example 3

[0114] This embodiment provides a structural ceramic bending strength prediction system, including:

[0115] Fracture image acquisition module: used to obtain the microscopic image of the potential fracture on the surface of the structural ceramic to be tested;

[0116] Defect feature acquisition module: processes the microscopic image of the potential fracture through image processing and machine vision to obtain a feature set of the hole defect to be tested;

[0117] Bending strength prediction module: used for inputting the hole defect feature set to be tested into a pre-trained bending strength prediction model to obtain the bending strength of the structural ceramic to be tested;

[0118] Wherein, the method for obtaining the bending strength prediction model includes:

[0119] Obtain the historical microscopic images of the potential fractures on the surface of each sample in the structural ceramic bending strength test and the corresponding bending strength values ​​of each sample;

[0120] Processing the historical microscopic images by image processing and machine vision to obtain a hole defect feature set of each sample;

[0121] Obtaining a key feature set corresponding to each sample according to the hole defect feature set of each sample;

[0122] According to the key feature sets and flexural strength values ​​corresponding to each sample, multiple pre-built machine learning algorithm models are trained separately to obtain multiple prediction models;

[0123] The multiple prediction models are optimized to obtain a bending strength prediction model.

[0124] Example 4

[0125] This embodiment provides a computer storage medium on which a computer program is stored. When the computer program is executed by a processor, the method for predicting the bending strength of structural ceramics as described in any step of Embodiment 2 is implemented.

[0126] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0127] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, 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 generate 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 flowchart and / or block diagram. 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.

[0128] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0129] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0130] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A method for predicting the flexural strength of structural ceramics, characterized in that: include: Obtaining a microscopic image of the potential fracture surface of the structural ceramic to be tested; Processing the microscopic image of the potential fracture through image processing and machine vision to obtain a feature set of the hole defect to be tested; Inputting the hole defect feature set to be tested into a pre-trained bending strength prediction model to obtain the bending strength of the structural ceramic to be tested; Wherein, the method for obtaining the bending strength prediction model includes: Obtain the historical microscopic images of the potential fractures on the surface of each sample in the structural ceramic bending strength test and the corresponding bending strength values ​​of each sample; Processing the historical microscopic images by image processing and machine vision to obtain a hole defect feature set of each sample; Obtaining a key feature set corresponding to each sample according to the hole defect feature set of each sample; According to the key feature sets and flexural strength values ​​corresponding to each sample, multiple pre-built machine learning algorithm models are trained separately to obtain multiple prediction models; The multiple prediction models are optimized to obtain a bending strength prediction model.

2. The method for predicting the flexural strength of structural ceramics according to claim 1, characterized in that: The method of obtaining a microscopic image of a potential fracture on the surface of the structural ceramic to be tested comprises: collecting a microscopic image of the surface area of ​​the structural ceramic by a scanning electron microscope, wherein the collection range of the microscopic image is an area extending outward by 30-40% of the span from the geometric center of the structural ceramic to be tested as the center point.

3. The method for predicting the flexural strength of structural ceramics according to claim 1, characterized in that: The historical microscopic images are processed by image processing and machine vision to obtain a hole defect feature set of each sample, including: Perform image preprocessing on the historical microscopic images of each sample to obtain a valid image set of each sample; Performing image segmentation on the effective image set of each sample to obtain a target area image set of each sample, wherein the target area image contains hole defects on the surface of the sample; The feature extraction of the target area image set of each sample is performed through machine vision to obtain a geometric feature set of hole defects on the surface of each sample.

4. The method for predicting the flexural strength of structural ceramics according to claim 3, characterized in that: The key feature set corresponding to each sample is obtained according to the hole defect feature set of each sample, including: Standardizing the geometric feature set of the hole defects on the surface of each sample to obtain a standard feature set of each sample; The standard feature set of each sample is screened for correlation to obtain a key feature set of each sample.

5. The method for predicting the flexural strength of structural ceramics according to claim 4, characterized in that: The geometric feature set of the hole defects on the surface of each sample is standardized to obtain a standard feature set of each sample, including: The characteristic values ​​in the geometric characteristic set of each sample are averaged and statistically analyzed to obtain the average characteristic set of each sample; The average feature set of each sample is normalized to obtain a standard feature set.

6. The method for predicting the flexural strength of structural ceramics according to claim 4, characterized in that: The correlation screening of the standard feature set of each sample is performed to obtain the key feature set of each sample, including: The Spearman correlation calculation was performed on the standard feature set of each specimen to obtain the correlation between each standard feature and the flexural strength value; Weakly correlated features in the standard feature set of each sample are removed, and Pearson correlation calculation is performed on the standard feature set after the weakly correlated features are removed to obtain the correlation between the standard features of each sample; The standard features with strong correlation between features are classified into the same group of features, and the standard features with strong correlation with the bending strength value are selected from the same group of features of each specimen to represent the group of features, so as to obtain the key feature set corresponding to each specimen.

7. The method for predicting the flexural strength of structural ceramics according to claim 1, characterized in that: The multiple pre-built machine learning algorithm models are trained separately according to the key feature sets and flexural strength values ​​corresponding to each sample to obtain multiple prediction models, including: Acquire a sample set according to the key feature set corresponding to each sample and the bending strength value corresponding to each sample; Dividing the sample set into a training set, a validation set and a test set; The training set is respectively input into a plurality of different pre-built machine learning algorithm models, the machine learning algorithm models are trained, and during the training process, the hyperparameters of the machine learning algorithm models are adjusted using the validation set to obtain a plurality of trained prediction models, wherein the key feature set in the training set is the input feature of the model training, and the bending strength value is the target feature of the model training.

8. The method for predicting the flexural strength of structural ceramics according to claim 7, characterized in that: The optimizing the multiple prediction models to obtain the bending strength prediction model includes: inputting the test set into the multiple trained prediction models respectively, evaluating the results output by each prediction model through the determination coefficient R² and the root mean square error RMSE, and obtaining the prediction model with the highest score as the bending strength prediction model.

9. A structural ceramic flexural strength prediction system, characterized in that: include: Fracture image acquisition module: used to obtain the microscopic image of the potential fracture on the surface of the structural ceramic to be tested; Defect feature acquisition module: processes the microscopic image of the potential fracture through image processing and machine vision to obtain a feature set of the hole defect to be tested; Bending strength prediction module: used for inputting the hole defect feature set to be tested into a pre-trained bending strength prediction model to obtain the bending strength of the structural ceramic to be tested; Wherein, the method for obtaining the bending strength prediction model includes: Obtain the historical microscopic images of the potential fractures on the surface of each sample in the structural ceramic bending strength test and the corresponding bending strength values ​​of each sample; Processing the historical microscopic images by image processing and machine vision to obtain a hole defect feature set of each sample; Obtaining a key feature set corresponding to each sample according to the hole defect feature set of each sample; According to the key feature sets and flexural strength values ​​corresponding to each sample, multiple pre-built machine learning algorithm models are trained separately to obtain multiple prediction models; The multiple prediction models are optimized to obtain a bending strength prediction model.

10. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting the flexural strength of structural ceramics according to any one of claims 1 to 8 is implemented.

Images