Surface roughness nondestructive testing method based on machine learning

Abstract

The invention relates to a surface roughness nondestructive testing method based on machine learning. The method comprises the following steps: measuring the surface roughness of a sample, acquiring surface roughness data, acquiring an SEM image and a metallographic image of the sample, preprocessing the image, extracting a texture feature value in the image, and forming a data set with the surface roughness data; establishing a linear mapping relation model between the surface roughness and the texture feature value, training the linear mapping relation model by adopting the training set, and verifying the accuracy of the linear mapping relation model by adopting the test set; and simulating a to-be-measured object by using the adjusted linear mapping relation model to obtain the surface roughness of the to-be-measured object. Compared with the prior art, the method has the advantages that the linear mapping relation model between the object image feature value and the roughness is constructed by using the machine learning technology, the current situation of discreteness or lack of a mapping model in a traditional detection method is changed, non-contact and rapid measurement of the surface roughness of the object is realized, and the detection cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of material surface performance detection, and in particular to a surface roughness nondestructive detection method based on machine learning. Background Art

[0002] Proton exchange membrane fuel cell (PEMFC) is a battery pack formed by multiple single fuel cells connected in series or in parallel, and a single fuel cell is mainly composed of four parts: proton exchange membrane, diffusion layer, catalyst layer and bipolar plate. The cost of proton exchange membrane fuel cells is too high to achieve commercial application, and the bipolar plate accounts for about 50% of the total cost of the battery. To reduce the cost of proton exchange membrane fuel cells, it is necessary to develop cheaper, thinner, and bipolar plates with good mechanical properties, high surface conductivity, low gas permeability and high corrosion resistance, as well as lower cost. Among them, surface roughness analysis plays an extremely important role in the surface quality and conductivity of BPP.

[0003] At present, stylus profilometers are widely used in surface roughness detection, but since the stylus needs to directly contact the surface to be measured, it is inevitable that scratches will be caused on the workpiece. In addition, this method uses linear sampling, which cannot fully reflect the contour characteristics of the entire surface, and the measurement efficiency is low. Non-contact roughness detection methods mainly include optical detection and electronic detection, but their popularity in the industrial field is limited due to their small measurement area, high equipment cost and low efficiency. In recent years, roughness detection methods based on machine vision have gradually attracted attention, but the detection process lacks an effective mapping model between roughness and characteristic indicators, which has a certain impact on the accuracy and efficiency of detection.

[0004] Therefore, it is urgent to study a surface roughness detection method that can improve detection accuracy and efficiency. Summary of the invention

[0005] The purpose of the present invention is to provide a surface roughness nondestructive testing method based on machine learning in order to overcome the defects of the above-mentioned prior art. It integrates machine learning, image preprocessing, and image roughness detection, which is convenient and fast; it uses machine learning technology to construct an effective linear mapping relationship model between the object image feature value and the roughness, changing the current situation of discreteness or lack of mapping model in traditional detection methods.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] The present invention provides a surface roughness nondestructive testing method based on machine learning, comprising the following steps:

[0008] S1: Measure the surface roughness of the sample and obtain the surface roughness data. At the same time, obtain the SEM image and metallographic image of the sample, extract the texture feature value in the image after preprocessing, and form a data set with the surface roughness data;

[0009] S2: Establish a linear mapping relationship model between surface roughness and texture feature values, and divide the data set obtained in S1 into a training set and a test set, use the training set to train the linear mapping relationship model, use the test set to verify the accuracy of the linear mapping relationship model, and obtain the adjusted linear mapping relationship model;

[0010] S3: Input the texture feature value of the object to be measured into the linear mapping relationship model adjusted in S2 to obtain the surface roughness of the object to be measured.

[0011] Furthermore, in S1, a probe-type roughness meter is used to measure the roughness of the sample surface.

[0012] Furthermore, in S1, the preprocessing process includes grayscale conversion, median filtering and grayscale enhancement of the image.

[0013] Furthermore, in S1, a transmission electron microscope is used to obtain a SEM image of the sample.

[0014] Furthermore, in S1, a metallographic microscope is used to obtain a metallographic image of the sample.

[0015] Furthermore, in S2, the texture feature values ​​include entropy, contrast, correlation, energy, and homogeneity. The entropy is used to represent the degree of non-uniformity or complexity of the texture in the image, the contrast is used to reflect the clarity of the image and the depth of the texture grooves, the correlation reflects the similarity of the matrix elements in the row or column direction, the energy is used to reflect the uniformity of the image grayscale distribution and the coarseness of the texture, and the homogeneity is used to reflect the amount of local changes in the measured image texture.

[0016] Furthermore, in S2, a linear mapping relationship model is established based on the KNN algorithm, and the K value is set to 5.

[0017] Furthermore, MATLAB is used to calculate the entropy value in the preprocessed image.

[0018] Furthermore, MATLAB is used to read the image, calculate the gray-level co-occurrence matrix and extract the texture eigenvalues.

[0019] Furthermore, in S3, MATLAB is used to extract texture feature values ​​from the metallographic image and SEM image of the object to be tested.

[0020] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0021] The present invention integrates machine learning, image preprocessing and image roughness detection, which is convenient and fast. It uses machine learning technology to construct an effective linear mapping relationship model between the object image feature value and the roughness, which changes the current situation of discreteness or lack of mapping model in traditional detection methods, realizes non-contact and rapid measurement of the surface roughness of the object, and reduces the cost of detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flow chart of a surface roughness nondestructive testing method based on machine learning;

[0023] Figure 2 These are metallographic images of the platinum-plated titanium plate in Example 1 before and after grayscale processing. DETAILED DESCRIPTION

[0024] The specific implementation methods of the present invention are described in detail below through examples. These examples are implemented on the premise of the scheme described in the present invention, and provide detailed implementation methods and specific operating processes, but the protection scope of the present invention is not limited to the following examples.

[0025] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments. Any features such as component models, material names, connection structures, preparation methods, materials, structures or composition ratios not clearly described in this technical solution are deemed to be common technical features disclosed in the prior art.

[0026] Example 1

[0027] This embodiment provides a surface roughness nondestructive testing method based on machine learning, such as Figure 1 As shown, the following steps are included:

[0028] S1: Use a probe roughness meter to measure the surface roughness of the sample and obtain surface roughness data. At the same time, obtain the SEM image and metallographic image of the sample, such as Figure 2 As shown, MATLAB is used to perform grayscale conversion, median filtering, and grayscale enhancement on the image, and then the texture feature values ​​in the image are extracted to form a data set with the surface roughness data;

[0029] Texture feature values ​​include entropy, contrast, correlation, energy, and homogeneity. The entropy is used to indicate the degree of non-uniformity or complexity of the texture in the image. The contrast is used to reflect the clarity of the image and the depth of the texture grooves. The correlation reflects the similarity of the matrix elements in the row or column direction. The energy is used to reflect the uniformity of the image grayscale distribution and the coarseness of the texture. The homogeneity is used to reflect the amount of local changes in the measured image texture.

[0030] S2: Based on the KNN algorithm, a linear mapping relationship model between surface roughness and texture feature values ​​is established, and the K value is set to 5. The data set obtained in S1 is divided into a training set and a test set. The training set is used to train the linear mapping relationship model, and the test set is used to verify the accuracy of the linear mapping relationship model to obtain the adjusted linear mapping relationship model.

[0031] The KNN algorithm, also known as the K nearest neighbor algorithm, is an instance-based learning algorithm. The basic idea is: first, a sample data set containing known labels is required, called a training sample set. When a new unlabeled data is input, the algorithm will compare the data with the data in the training sample set, find the K samples closest to it (i.e., the K nearest neighbors), and then determine the category of the new data based on the category information of these K nearest neighbors through the majority voting principle.

[0032] By measuring the distance between different data points for classification or regression prediction, the KNN algorithm has the advantages of being intuitive, easy to understand and simple to implement.

[0033] S3: A scanning electron microscope is used to obtain a SEM image of the platinum-coated titanium plate to be tested, and a metallographic microscope is used to obtain a metallographic image of the platinum-coated titanium plate to be tested. The image is then preprocessed through MATLAB to obtain the entropy value, contrast, correlation, energy, and homogeneity of the platinum-coated titanium plate to be tested, and the entropy value, contrast, correlation, energy, and homogeneity of the platinum-coated titanium plate to be tested are input into the adjusted linear mapping relationship model in S2 to obtain the surface roughness of the object to be tested. The results are shown in Table 1.

[0034] Table 1 Prediction results of surface roughness of platinum-coated titanium plate based on KNN algorithm

[0035]

[0036] The above description of the embodiments is to facilitate the understanding and use of the invention by those skilled in the art. It is obvious that those skilled in the art can easily make various modifications to these embodiments and apply the general principles described herein to other embodiments without creative work. Therefore, the present invention is not limited to the above embodiments, and improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the present invention should be within the scope of protection of the present invention.

Claims

1. A nondestructive testing method for surface roughness based on machine learning, characterized in that: The following steps are involved: S1: Measure the surface roughness of the sample and obtain the surface roughness data. At the same time, obtain the SEM image and metallographic image of the sample, extract the texture feature value in the image after preprocessing, and form a data set with the surface roughness data; S2: Establish a linear mapping relationship model between surface roughness and texture feature values, and divide the data set obtained in S1 into a training set and a test set. Use the training set to train the linear mapping relationship model, and use the test set to verify the accuracy of the linear mapping relationship model to obtain the adjusted linear mapping relationship model. S3: Input the texture feature value of the object to be measured into the linear mapping relationship model adjusted in S2 to obtain the surface roughness of the object to be measured.

2. The surface roughness nondestructive testing method based on machine learning according to claim 1 is characterized in that: In S1, a probe-type roughness meter is used to measure the roughness of the sample surface.

3. The surface roughness nondestructive testing method based on machine learning according to claim 1 is characterized in that: In S1, the preprocessing process includes grayscale conversion, median filtering and grayscale enhancement of the image.

4. The surface roughness nondestructive testing method based on machine learning according to claim 1, characterized in that: In S1, a transmission electron microscope was used to obtain a SEM image of the sample.

5. The surface roughness nondestructive testing method based on machine learning according to claim 1, characterized in that: In S1, a metallographic microscope is used to obtain a metallographic image of the sample.

6. The surface roughness nondestructive testing method based on machine learning according to claim 1, characterized in that: In S2, the texture feature values ​​include entropy, contrast, correlation, energy, and homogeneity. The entropy is used to represent the degree of non-uniformity or complexity of the texture in the image. The contrast is used to reflect the clarity of the image and the depth of the texture grooves. The correlation reflects the similarity of the matrix elements in the row or column direction. The energy is used to reflect the uniformity of the image grayscale distribution and the coarseness of the texture. The homogeneity is used to reflect the amount of local changes in the measured image texture.

7. The surface roughness nondestructive testing method based on machine learning according to claim 1, characterized in that: In S2, a linear mapping relationship model is established based on the KNN algorithm, and the K value is set to 5.

8. The method for nondestructive testing of surface roughness based on machine learning according to claim 6, characterized in that: MATLAB was used to calculate the entropy value in the preprocessed image.

9. The method for nondestructive testing of surface roughness based on machine learning according to claim 6, characterized in that: MATLAB is used to read the image, calculate the gray-level co-occurrence matrix and extract the texture eigenvalues.

10. The method for nondestructive testing of surface roughness based on machine learning according to claim 1, characterized in that: In S3, MATLAB is used to extract texture feature values ​​from the metallographic image and SEM image of the object to be tested.

Images