阅读说明：本技术 一种基于四元数奇异值熵指标的磨削表面粗糙度检测方法 (Grinding surface roughness detection method based on quaternion singular value entropy index ) 是由 易怀安 赵欣佳 舒爱华 于 2019-11-11 设计创作，主要内容包括：本发明提供一种基于四元数奇异值熵指标的磨削表面粗糙度检测方法,属于粗糙度检测技术领域。包括以下步骤：利用纯四元数的三个虚部分别代替彩色图像的三个颜色分量；对纯四元数矩阵进行奇异值分解；截取前30个奇异值,计算其奇异值熵指标；利用支持向量机对奇异值熵指标和粗糙度值进行训练拟合,从而预测粗糙度。本发明的方法将颜色信息作为一个整体处理,并给出了合理的数据结构表达,可直接获取彩色光源照射在待检测物表面的虚像对应的图片来检测其粗糙度,该图像是彩色图像,非灰度图像,因而不存在图像降质的过程,使得图像处理的准确性提高,从而使得粗糙度检测结果的准确性更高。(The invention provides a grinding surface roughness detection method based on quaternion singular value entropy indexes, and belongs to the technical field of roughness detection. The method comprises the following steps: three imaginary parts of pure four-element numbers are used for replacing three color components of the color image respectively; performing singular value decomposition on the pure four-element number matrix; intercepting the first 30 singular values, and calculating the singular value entropy indexes of the singular values; and training and fitting the singular value entropy index and the roughness value by using a support vector machine so as to predict the roughness. The method of the invention treats the color information as a whole, gives reasonable data structure expression, and can directly obtain the picture corresponding to the virtual image of the color light source irradiating on the surface of the object to be detected to detect the roughness of the object, wherein the picture is a color picture and a non-gray level picture, so that the process of image degradation does not exist, the accuracy of image treatment is improved, and the accuracy of the roughness detection result is higher.)

1. A grinding surface roughness detection method based on quaternion singular value entropy indexes is characterized by comprising the following steps:

(1) the method comprises the following steps of taking a color light source as a reference object, enabling the reference object to form a color virtual image on the surfaces of a group of standard test blocks with different roughness, and representing three color components at any point (x, y) of the color virtual image by using a pure quaternion Q (x, y), wherein the formula is as follows:

Q(x，y)＝R(x，y)i+G(x，y)j+B(x，y)k

wherein Q (x, y) is a color image pure four-element number matrix; r (x, y), G (x, y) and B (x, y) are color vectors based on a matrix form, as real numbers; i. j, k represent 3 imaginary parts of the quaternion respectively;

(2) performing singular value decomposition on the pure quaternion matrix in the step (1);

(3) intercepting the first 30 singular values in the step (2), and calculating the singular value entropy indexes of the singular values;

(4) and training and fitting the singular value entropy index and the roughness value by using a support vector machine, so as to obtain a prediction model for predicting roughness.

2. The grinding surface roughness detection method based on quaternion singular value entropy index as claimed in claim 1, characterized in that: the step (2) specifically comprises the following steps: singular value decomposition is carried out on the pure quaternion matrix in the step (1) to obtain a singular value matrix sigma_r＝diag(δ₁，δ₂…δ_r) (ii) a Normalizing each singular value, i.e.

In the formula (I), the compound is shown in the specification,

let k equal to 1, construct a calculation formula for calculating singular value entropy QE

Wherein r is the total number of the selected singular values.

3. The grinding surface roughness detection method based on quaternion singular value entropy index as claimed in claim 1, characterized in that: the device for acquiring the color virtual image in the step (1) comprises a detection platform, a customized color light source, an image acquisition unit and an intelligent terminal;

the detection platform is used for bearing an object to be detected;

the customized color light source comprises a lamp holder, a lamp bead and a square lampshade, and the square lampshade is arranged on the surface of the lamp holder; the image acquisition unit is the camera, customization colored light source and image acquisition unit pass through the support setting in detection platform's top, and the light and the image acquisition unit of customization colored light source transmission all face the last object's that awaits measuring surface of detection platform, the contained angle of the plane center normal of square lampshade and the normal of the object upper surface that awaits measuring is theta₁The included angle between the optical axis of the image acquisition unit and the normal of the upper surface of the object to be measured is theta₂Of the plane central normal of the square lampshade and the image-capturing unitThe optical axis is located on both sides of the normal of the upper surface of the object to be measured, and the angle theta₁、θ₂Are all less than 90 DEG, and theta₁＝θ₂；

The image acquisition element is connected with the intelligent terminal and used for acquiring an image corresponding to a virtual image formed by the customized color light source on the surface of the object to be measured and sending the image to the intelligent terminal;

the intelligent terminal comprises a storage module and a roughness calculation module; the storage module is used for storing the corresponding relation between the singular value entropy index of the virtual image of the customized color light source and the object surface roughness, the roughness calculation module is used for calculating the singular value entropy index of the image according to the virtual image acquired by the image acquisition unit, and then calculating the surface roughness of the object to be measured according to the corresponding relation stored in the storage module.

4. The grinding surface roughness detection method based on quaternion singular value entropy index as claimed in claim 3, characterized in that: the customized color light source comprises 2 or 4 lamp beads, and each lamp bead is independently covered by the same square lampshade; half of the red light beads are red light beads, and the other half of the green light beads are green light beads; when the number of the lamp beads is 2, arranging one row; when the number of the lamp beads is 4, two rows of 2 lamp beads are arranged, and the colors of the adjacent lamp beads are different.

5. The grinding surface roughness detection method based on quaternion singular value entropy index as claimed in claim 4, characterized in that: the device also comprises a multi-channel digital display light source controller, wherein each channel of the light source controller controls one of the lamp beads, and the light source controller can control the brightness of the lamp beads.

[ technical field ] A method for producing a semiconductor device

The invention relates to the technical field of workpiece surface roughness detection, in particular to a grinding surface roughness detection method based on quaternion singular value entropy indexes.

[ background of the invention ]

The machine vision detection method of surface roughness generally analyzes a gray image, performs objective quantization aiming at image gray value information, and then applies an image texture analysis technology (the image texture analysis technology is generally summarized into three categories of frequency spectrum, structure and statistics) to obtain the relation between the surface roughness and a machine vision index. Because the gray level image is a degraded image, the sensitivity of the image characteristics to the surface roughness is reduced to a certain extent, and the gray level image cannot be subjectively judged visually; moreover, the application of more gray level co-occurrence matrix methods needs to be combined with a microscopic device, so that the test view field is small, the operation is inconvenient, and the working efficiency is low. In addition, the frequency spectrum characteristics of Fourier transform have better robustness to the texture characteristics with periodic regularity, but for the grinding workpiece with larger randomness of the texture characteristics, the sensitivity is not strong when the surface roughness is judged by the frequency spectrum characteristics, the prediction accuracy of the artificial neural network is greatly influenced by the factors such as the number of training samples, training parameters, network structures and the like, and the prediction accuracy cannot be ensured in the case of small samples. Literature indicates that color is a sensitive delineation factor that often simplifies the extraction and recognition of objects from a scene, and few quantitatively predict surface roughness by studying color image quality. Because the grinding sample does not have the characteristic of clear color, the obvious color difference information is difficult to obtain by directly shooting the sample. The ancient people used copper as a mirror, observed the appearance by utilizing a virtual image very early, whether the appearance is clear or not is related to the smoothness of the surface of the copper mirror, and most metal surfaces are easy to generate virtual images. In the present invention, the customized light source can be regarded as a reference object, and the surface roughness of the sample can be judged according to the definition change of the virtual image on the surface of the sample. Meanwhile, the objective index of definition is combined with subjective evaluation of a Human Visual System (HVS), so that the surface roughness of the workpiece can be detected quickly and accurately.

According to the reflection law of geometric optics, the light source points generate diffuse reflection diffusion on the surface of a grinding sample, and the energy diffusion degree is larger when the surface roughness value is larger, so that the grinding surface roughness can be evaluated according to the color image energy distribution index. The singular value entropy is an index which can filter noise and well reflect the reflection energy difference of different surface roughness, so that the singular value entropy index can be used for representing the surface roughness. At present, no relevant report for representing the surface roughness by using singular value entropy indexes exists.

[ summary of the invention ]

The invention aims to: aiming at the existing problems, the grinding surface roughness detection method based on the quaternion singular value entropy index is provided, the method and the device can objectively, simply and accurately express the surface roughness of an object, and a new method is provided for the detection of the grinding roughness.

In order to achieve the purpose, the technical scheme adopted by the invention is as follows:

a grinding surface roughness detection method based on quaternion singular value entropy indexes comprises the following steps:

(1) the method comprises the following steps of taking a color light source as a reference object, enabling the reference object to form a color virtual image on the surfaces of a group of standard test blocks with different roughness, and representing three color components at any point (x, y) of the color virtual image by using a pure quaternion Q (x, y), wherein the formula is as follows:

Q(x,y)＝R(x,y)i+G(x,y)j+B(x,y)k

wherein Q (x, y) is a color image pure four-element number matrix; r (x, y), G (x, y) and B (x, y) are based on a matrix form of color vectors, being real numbers; i. j, k represent 3 imaginary parts of the quaternion respectively;

(2) performing singular value decomposition on the pure quaternion matrix in the step (1);

(3) intercepting the first 30 singular values in the step (2), and calculating the singular value entropy indexes of the singular values;

(4) and training and fitting the singular value entropy index and the roughness value by using a support vector machine, so as to obtain a prediction model for predicting roughness.

In the present invention, further, the step (2) specifically includes: singular value decomposition is carried out on the pure quaternion matrix in the step (1) to obtain a singular value matrix sigma_r＝diag(δ₁，δ₂…δ_r) (ii) a Normalizing each singular value, i.e.

Formula (II)

Then there are

The information entropy normalization condition is met; calculation formula according to information entropy

And (5) constructing a calculation formula for calculating the singular value entropy QE by taking k as 1:

wherein r is the total number of the selected singular values.

Further, the device for acquiring the color virtual image in the step (1) comprises a detection platform, a customized color light source, an image acquisition unit and an intelligent terminal;

the detection platform is used for bearing an object to be detected;

the customized color light source comprises a lamp holder, a lamp bead and a square lampshade, and the square lampshade is arranged on the surface of the lamp holder; the image acquisition unit is a camera, the customized color light source and the image acquisition unit are arranged above the detection platform through a support, the light emitted by the customized color light source and the image acquisition unit all face the surface of an object to be detected on the detection platform, and the included angle between the plane center normal of the square lamp shade and the normal of the upper surface of the object to be detected is theta₁The included angle between the optical axis of the image acquisition unit and the normal of the upper surface of the object to be measured is theta₂The plane center normal of the square lampshade and the optical axis of the image acquisition unit are positioned on two sides of the normal of the upper surface of the object to be detected, and the theta is₁、θ₂Are all less than 90 DEG, and theta₁＝θ₂；

The image acquisition element is connected with the intelligent terminal and used for acquiring an image corresponding to a virtual image formed by the customized color light source on the surface of the object to be detected and sending the image to the intelligent terminal;

the intelligent terminal comprises a storage module and a roughness calculation module; the storage module is used for storing the corresponding relation between the singular value entropy index of the virtual image of the customized color light source and the object surface roughness, and the roughness calculation module is used for calculating the singular value entropy index of the image according to the virtual image acquired by the image acquisition unit and then calculating the surface roughness of the object to be measured according to the corresponding relation stored in the storage module.

Preferably, the customized color light source comprises 2 or 4 lamp beads, and each lamp bead is independently covered by the same square lampshade; half of the red light beads are red light beads, and the other half of the green light beads are green light beads; when the number of the lamp beads is 2, arranging one row; when the number of the lamp beads is 4, two rows of 2 lamp beads are arranged, and the colors of the adjacent lamp beads are different.

Preferably, the device also comprises a multi-channel digital display light source controller, each channel of the light source controller controls one lamp bead, and the light source controller can control the brightness of the lamp beads

In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:

1. the method of the invention treats the color information as a whole, gives reasonable data structure expression (quaternion modeling), can directly obtain the picture corresponding to the virtual image of the color light source irradiating on the surface of the object to be detected to detect the roughness, and the picture is a color picture and a non-gray level picture, so that the process of image degradation does not exist, the accuracy of image treatment is improved, and the accuracy of the roughness detection result is higher.

2. The roughness detection device for acquiring the virtual image, which is adopted by the invention, takes the common light source as a reference object, and can reduce the primary light reflection process compared with the mode of matching the color block with the LED light source, so that the light path design is simplified, and a simpler and more convenient detection method is provided.

[ description of the drawings ]

2. The roughness detection device for acquiring the virtual image, which is adopted by the invention, takes the common light source as a reference object, and can reduce the primary light reflection process compared with the mode of matching the color block with the LED light source, so that the light path design is simplified, and a simpler and more convenient detection method is provided.

FIG. 2 is a point light source and corresponding virtual image;

FIG. 3 is a flowchart of a method for detecting roughness of a grinding surface based on quaternion singular value entropy indexes;

fig. 4 is a schematic structural diagram of a customized light source roughness detecting device used in the present invention.

The system comprises a support 1, a customized color light source 2, a camera 3, an intelligent terminal 4, an object to be detected 5, a detection platform 6, a virtual image 7 and a light source controller 8.

FIG. 5 is a graph showing the relationship between singular value entropy and roughness in a specific experimental example.

[ detailed description ] embodiments

The invention aims to explore a new RGB color space-based index capable of objectively, simply and accurately expressing surface roughness. In order to better illustrate the principle and application of the present invention, the mechanism of surface roughness evaluation to which the present invention is applied and the mathematical basic theory to which the present invention is applied will be described.

1. Mechanism of surface roughness evaluation

The principle of imaging a rough surface is shown in fig. 1. When the light source irradiates on the surface of the object A, the object A can be regarded as a point light source. Assuming that only two light beams are emitted from the point a, for the grinding sample reflecting surface, the light emitted from the point a is reflected by the rough surface and then is diffused all around, and the same point will generate a plurality of virtual images with reduced brightness, namely, the situation of one object and multiple images, such as a1 and a2, as shown in the left image in fig. 1. Meanwhile, when the included angle (less than or equal to 90 ℃) between the light source direction and the optical axis of the camera is increased, the divergence range S of the imaging point of the target A on the imaging surface is in direct proportion to the roughness height value, as shown in the right diagram in FIG. 1. Therefore, the virtual image formed on the rough surface can represent a change in the roughness height, specifically, a change in the roughness Ra.

Obviously, the light emitted from point a is more than two beams far away, and for a rough plane, the virtual image point formed by point a on its surface will also diverge differently in area due to the roughness of the surface. From the aspect of energy conservation, it is theoretically assumed that the larger the roughness is, the more virtual image points are formed by the points a, the larger the divergence plane is, and the lower the brightness of each point is, and this is also the theoretical basis of the gray scale image detection surface roughness in practice.

Assuming that there are A, B red and green points on the target, the divergent virtual images formed on the sample surfaces with different roughness are ideal circles, as shown in fig. 2. Fig. 2(a) is virtual images a1 and B1 of a and B light source points formed on a surface having a roughness value of R1 (R1R 2), and C1 is an aliasing region of a1 and B1. Fig. 2(B) shows virtual images a2 and B2 formed by the light source points a and B on the surface with roughness value R2, C2 is an aliasing region of a2 and B2, and it can be seen from fig. 2 that the aliasing region becomes larger in area as the roughness increases.

Meanwhile, according to the reflection law of geometric optics, the aliasing phenomenon shown in fig. 2 is just a diffusion phenomenon that light source points generate diffuse reflection on the surface of a grinding sample, namely, the energy diffusion degree is larger when the surface roughness value is larger, so that the grinding surface roughness can be evaluated according to the color image energy distribution index. The singular value entropy is an index which can filter noise and well reflect the reflection energy difference of different surface roughness degrees, so that the singular value entropy index can be used for representing the surface roughness.

2. Quaternion matrix and singular value decomposition

A quaternion contains four components, a real component and three imaginary components, of the form:

q＝a+bi+cj+dk (1)

in the formula, a, b, c and d are real numbers, a represents a real part, and i, j and k respectively represent 3 imaginary parts of quaternions. Let the real part of the quaternion be 0, then q is called the pure quaternion. For a color image of a single pixel point, a relatively fitting corresponding relation exists between the self mathematical structure of a pure quaternion and three color channels of the color image, so that the two are connected, and the R, G and B values at any point (x, y) of the color image are represented by the pure quaternion, and the formula is as follows:

Q(x，y)＝R(x，y)i+G(x，y)j+B(x，y)k (2)

where R, G and B are color vectors based on a matrix form, and are therefore also referred to as a color image pure quaternion matrix. By modeling the color image in a pure four-element structure, a mathematical reasonable data structure expression of color information is established, and the expression space of the color information is wider.

The matrix operation of real numbers and complex numbers in the general sense cannot be directly applied to the quaternion matrix, and the correlation operation of the quaternion matrix can be converted into a complex expression matrix for calculation. And performing singular value decomposition on the complex representation matrix of the quaternion matrix. For quaternion matrix A ∈ Q^m×nA may be represented as follows:

A＝A₁+A₂j (3)

wherein A1 and A2 are m × n complex matrices, Q^m×nFor a four-element number, a complex representation matrix A of A is defined_αThe following were used:

in the formula

Is A₁The conjugate matrix of (a) is determined,

is the conjugate matrix of a. According to the relation between the quaternion matrix and the complex expression matrix given by the formula (4), for A ∈ Q^m×ⁿThe singular value decomposition can be performed by decomposing a complex representation matrix, which can be expressed as follows:

in the formula_r＝diag(δ₁，δ₂…δ_r) Is a diagonal matrix having r non-zero values, constituting a singular value vector of A, and satisfying δ₁≥δ₂≥…δ_rU and V are quaternion unitary matrices,is the conjugate matrix of V.

3. Singular value entropy evaluation index

Singular Value Decomposition (SVD) is performed on the image quaternion matrix, the distribution of the singular value sizes is mapped with the energy distribution of the image, and the singular value matrix represents the case of the distribution of the singular value sizes. In addition, a singular value is an inherent characteristic of a matrix, and is an index for measuring the stability of the matrix, and when a certain element in the matrix changes, the singular value also changes correspondingly. Therefore, in order to quantitatively characterize the change of the image energy distribution on different roughness surfaces, an information entropy theory is introduced to describe the change.

The shannon (c.e. shannon) rate firstly populates the entropy into the information theory, applies it to the quantitative analysis of information, proposes that the entropy is used to measure the information of the object and the information transmission amount in the process of the change of the object, and the calculation formula of the information entropy is:

according to the shannon information theory, the larger the entropy is, the larger the information content is, and the larger the entropy is in the image quality evaluation, the clearer the image is. As can be seen from the surface roughness evaluation mechanism, when the roughness is smaller, the surface diffusion degree is smaller, the reflected light concentration is strong, and the surface image is clearer. Therefore, the image energy distribution is more concentrated when the image is more clear, and conversely, the energy distribution is more dispersed. The singular value matrix can represent the image energy distribution, and the singular value entropy can describe the condition of the energy distribution. When the singular value entropy is larger, the image is clearer, the energy distribution is more concentrated, and vice versa, the energy distribution is more dispersed. The quaternion matrix is subjected to singular value decomposition by using the formula (5) to obtain a singular value matrix sigma_r＝diag(δ₁，δ₂…δ_r) Normalizing each singular value, i.e.

In the formula (I), the compound is shown in the specification,

and (3) constructing a calculation formula for calculating the singular value entropy QE by taking k as 1 in the formula (6):

the present invention will be described in further detail with reference to specific embodiments.

The invention discloses a grinding surface roughness detection method based on quaternion singular value entropy indexes, the flow of the method is shown in figure 3, and the method comprises the following steps:

(1) the method comprises the following steps of taking a color light source as a reference object, enabling the reference object to form a color virtual image on the surfaces of a group of standard test blocks with different roughness, and representing three color components at any point (x, y) of the color virtual image by using a pure quaternion Q (x, y), wherein the formula is as follows:

Q(x，y)＝R(x,y)i+G(x,y)j+B(x,y)k

wherein Q (x, y) is a pure four-element number matrix of the color virtual image; r (x, y), G (x, y) and B (x, y) are color vectors based on a matrix form, as real numbers; i. j, k represent 3 imaginary parts of the quaternion respectively;

(2) performing singular value decomposition on the pure quaternion matrix in the step (1);

matrix sigma of singular values can be obtained_r＝diag(δ₁，δ₂…δ_r) (ii) a Normalizing each singular value, i.e.

In the formula

Then there are

The information entropy normalization condition is met; calculation formula according to information entropy

And (5) constructing a calculation formula for calculating the singular value entropy QE by taking k as 1:

wherein r is the total number of the selected singular values.

(3) Intercepting the first 30 singular values in the step (2), and calculating the singular value entropy indexes of the singular values;

(4) and training and fitting the singular value entropy index and the roughness value by using a support vector machine, so as to obtain a prediction model for predicting roughness.

In the present invention, the instrument used for implementing the roughness test is shown in fig. 4, and the structure is described as follows:

the system comprises a detection platform 6, a customized color light source 2, an image acquisition unit 3 and an intelligent terminal 4;

the detection platform 6 is used for bearing an object 5 to be detected, and preferably adopts a precise optical test bench; (ii) a

The customized color light source 2 comprises a lamp holder, a lamp bead and a square lampshade, and the square lampshade is arranged on the surface of the lamp holder; the image acquisition unit 3 is a camera, preferably a CCD camera. The customized color light source 2 and the image acquisition unit 3 are both arranged above the detection platform 6 through the support 1, the light emitted by the customized color light source 1 and the lens of the image acquisition unit all face the surface of an object to be detected 5 on the detection platform, and the included angle between the plane center normal of the square lampshade and the normal of the upper surface of the object to be detected is theta₁The included angle between the optical axis of the image acquisition unit and the normal of the upper surface of the object to be measured is theta₂The plane center normal of the square lampshade and the optical axis of the image acquisition unit are positionedOn both sides of the normal line of the upper surface of the object to be measured, theta₁、θ₂Are all less than 90 DEG, and theta₁＝θ₂；

The image acquisition element 3 is connected with the intelligent terminal 4, and the image acquisition element 3 is used for acquiring an image corresponding to a virtual image formed by the customized color light source on the surface of the object 5 to be measured and sending the image to the intelligent terminal 4 through the wireless communication module;

the intelligent terminal 4 comprises a storage module and a roughness calculation module, and preferably adopts a computer; the intelligent terminal 4 is connected with the image acquisition element 3 in a communication mode. The device comprises a storage module, a roughness calculation module and an image acquisition unit 3, wherein the storage module is used for storing the corresponding relation between the singular value entropy of a virtual image of a customized color light source and the surface roughness of an object; the corresponding relation between the singular value entropy index of the virtual image and the object surface roughness is obtained by calculation through the method.

In the embodiment, the customized color light source comprises 2 or 4 lamp beads, and each lamp bead is independently covered by the same square lampshade; half of the red light beads are red light beads, and the other half of the green light beads are green light beads; when the number of the lamp beads is 2, arranging one row; when the number of the lamp beads is 4, two rows of 2 lamp beads are arranged, and the colors of the adjacent lamp beads are different. In the embodiment, the lamp shade is a plastic lamp shade with the side length of 40mm, and the distance between the lamp shades is set to be 2 mm. The voltage of a customized red and green light source is 24V, the power of a red light is 0.9W, the power of a green light is 1.8W, the brightness interval is 85-120, and a lampshade is made of transparent plastic.

In other embodiments, the device further comprises a multi-channel digital display light source controller, each channel of the light source controller controls one of the lamp beads, and the light source controller can control the brightness of the lamp beads.

When the color block is used as a reference object, the virtual image of the color block on the surface of the workpiece is obtained, and the imaging principle is that light irradiates on the color block, is reflected to the surface of the workpiece through the color block and is reflected to a camera for imaging through the surface of the workpiece. The light emitted from the light source can reach the camera for imaging only by at least secondary reflection, which undoubtedly increases the difficulty of light path design for engineering practice, and if the color block is directly designed into the light source, the reflection process of the primary light can be reduced, and the feasibility for realizing online detection of singular value entropy indexes is improved.

The detection case is as follows:

the method is adopted to detect the grinding roughness, 11 standard grinding parts with different roughness are selected, the standard roughness value (Ra) of the standard grinding parts is shown in the table 1, the device is adopted to obtain the color virtual image of the color light source, the quaternion matrix and the singular value decomposition of the device are utilized to calculate and obtain the corresponding singular value entropy (QE) of the color virtual image respectively shown in the table 1.

Table 1: support vector machine training sample SE value

NO	1	2	3	4	5	6	7	8	9	10	11
												Ra	0.055	0.148	0.177	0.213	0.269	0.357	0.765	1.034	1.134	1.534	2.341
QE	0.4351	0.4309	0.4296	0.4276	0.4256	0.4148	0.3830	0.3681	0.3519	0.3294	0.3242

According to the corresponding relation of the table 1, a relation model between the singular value entropy (QE) index and the roughness value (Ra) is obtained as shown in fig. 5, and the singular value entropy (QE) indexes are all reduced along with the increase of the roughness value (Ra), so that the prediction feasibility of the singular value entropy (QE) index is shown. Another 10 standard grinding parts are taken, and respective roughness predicted values are measured by the method of the invention and are compared with the measured values of the contact pins, which are shown in table 2.

Table 2: roughness prediction and contact measurement

NO	1	2	3	4	5	6	7	8	9	10	11
												Measured value of stylus	0.118	0.154	0.200	0.216	0.348	0.563	0.940	1.103	1.271	1.930	2.430
Prediction value	0.120	0.154	0.182	0.206	0.349	0.590	0.976	1.014	1.373	1.663	2.058
												Error rate (%)	1.7％	0	9％	4.6％	0.3％	4.7％	3.8％	8.1％	8％	13.8％	15.3％

As can be seen from the results in Table 2, the method of the present invention has high test accuracy of roughness, especially for grinding parts with roughness less than 1, and can be used for testing grinding roughness.

The above description is intended to describe in detail the preferred embodiments of the present invention, but the embodiments are not intended to limit the scope of the invention, and all equivalent changes and modifications made within the technical spirit of the present invention should fall within the scope of the present invention.