WO2021185035A1 - Method for detecting abnormal wear spot image on basis of appearance feature - Google Patents

Abstract

Provided in the present invention is a method for detecting an abnormal wear spot image on the basis of an appearance feature. The method comprises: on the basis of segmenting a wear mark region and measuring a direction angle, representing row-direction and column-direction appearance features of the wear mark region by means of row-direction and column-direction lengths; on the basis of interval filtering, respectively calculating gradient change values of a row length and a column length; and performing binarization on the gradient change values by using an adaptive threshold value, and determining whether a wear spot image is abnormal on the basis of the monotonicity of the gradient. In the present invention, the abnormal state of a wear spot image is determined on the basis of an appearance feature. The invention is more universal, avoids errors of subjective determination, achieves higher accuracy, and provides a theoretical basis for the availability of test data.

Description

A Detection Method of Abnormal Wear Spot Images Based on Shape Features Technical field

The invention relates to the extended application of a four-ball friction tester to determine the lubricating performance of lubricating oil, in particular to a method for detecting abnormal wear scar images based on shape characteristics.

Background technique

The lubricating oil with good lubricity can protect the machine and prolong the working life. It is often measured by the four-ball wear tester. According to China's petrochemical industry standards (GB-T 12583-1998 and HT 0762-2005), the determination method of lubricant anti-wear performance is: clamp three steel balls with a diameter of 12.7 mm in an oil box, and test After oil immersion, place a steel ball with a diameter on the top of the three balls. After the oil test temperature reaches 75℃±2℃, apply 147N or 392N force, the top ball rotates at a certain speed for 60min, and then take out the three steel balls at the bottom , Measure the wear scar diameters of three steel balls under a microscope with a measurement accuracy of 0.01mm. A total of six sets of measurement data are obtained from the three steel balls. The arithmetic mean of the spot diameter is evaluated. In an ideal state, the wear scar image of the four-ball friction test is elliptical, but when the experimenter's experience is insufficient or the operation is not standardized during the operation, the wear scar image presents an abnormal shape and cannot be used to determine the friction coefficient of the lubricating oil. At present, testers mainly judge the validity of test data based on experience, and subjective judgment errors are unavoidable. Based on this, we propose a detection method for abnormal wear scar images based on shape features.

Summary of the invention

The purpose of the present invention is to provide a method for detecting abnormal wear scar images based on appearance characteristics, so as to solve the defect of error in the abnormal detection of the existing wear scar images.

In order to achieve the above objective, the technical solution adopted by the present invention is:

The present invention provides a method for detecting abnormal wear spot images based on appearance characteristics, including the following steps:

Step 1. Preprocess the collected wear scar image to obtain a wear scar area map;

Step 2: Extract the wear scar area in the wear scar area map;

Step 3. Calculate the row and column diameters of the wear scar area;

Step 4. Determine whether the wear scar image is abnormal according to the row and column long diameters of the wear scar level map obtained in step 3. If the wear scar image is abnormal, the algorithm ends; otherwise, go to step 5;

Step 5: Perform interval filtering on the row length and column length of the wear spot area to obtain the interval row length and the interval column length;

Step 6. Calculate the gradient change value according to the interval row length and interval column length obtained in step 5;

Step 7. Determine the adaptive segmentation threshold;

Step 8. Binarize the gradient change value obtained in step 6 according to the adaptive segmentation threshold obtained in step 7, to obtain gradient binary data;

Step 10: Determine whether the shape of the wear scar area is normal according to the gradient binary data obtained in step 9 and the row major diameter and column major diameter of the wear scar area in step 3.

Preferably, in step 2, the wear scar area is extracted in the wear scar area map, and the specific method is:

S1, determine the wear scar direction angle of the wear scar area map obtained in step 1;

S2: Rotate the wear scar area map clockwise around the center of the image to obtain a wear scar level map;

S3, extract the wear scar area in the wear scar level map, where i ₀ and i _{1 are used to} represent the first and last rows of the wear scar area in the wear scar level graph; j ₀ and j _{1 are used to} represent the wear scar in the wear scar level graph The first and last columns of the area:

i ₀ <i and f(i ₀ ,j)=f(i,j)=1,

i ₁ >i and f(i ₁ ,j)=f(i,j)=1,

j ₀ <j and f(i,j ₀ )=f(i,j)=1,

j ₁ >j and f(i,j ₁ )=f(i,j)=1,

Preferably, in step 3, the line length diameter of the wear scar area is calculated according to the following formula:

Among them, _rh is the length of the row; D _h (i) is the length of the i-th row,

i ₀ is the first row of the wear scar area; i ₁ is the last row of the wear scar area _{; j 0 is the first row of the wear scar area; j 1} is the last row of the wear scar area;

Calculate the column long diameter by the following formula:

Among them, r _l is the length of the column; D _l (j) is the length of the j-th column,

Preferably, in step 4, judging whether the wear scar image is abnormal according to the row length diameter and column length diameter of the wear scar level map obtained in step 3, the specific method is that if the row length diameter and the column length diameter satisfy the following formula, it is considered Abnormal wear spot image:

Among them, β is the gap threshold.

Preferably, in step 5, interval filtering is performed on the row length and column length of the wear spot area in combination with the interval length set by the following formula:

Among them, s and t are the indexes of row interval and column interval respectively; L _h (s) represents the row length of the sth interval; L _l (t) is the column length of the t-th interval; q is the set interval length; [] Represents the round-down operator.

Preferably, in step 6, the gradient change value is calculated by the interval row length and interval column length obtained in step 5 of the following structure:

Among them, I _h (s) is the row gradient in the s interval, and I _l (t) is the column gradient in the t interval.

Preferably, in step 7, the adaptive segmentation threshold is determined, and the specific method is:

Will gather

The elements in are arranged in ascending order, and the order of

The element corresponding to the bit is used as the row segmentation threshold _Th ;

Will gather

After the elements are arranged in ascending order, the order of

The corresponding element value is used as the column segmentation threshold T ₁ ;

in,

It is a rounding operation.

Preferably, in step 8, the gradient change obtained in step 6 is binarized according to the adaptive segmentation threshold obtained in step 7 to obtain gradient binary data, and the specific method is calculated using the following formula:

Among them, B _h (s) is the row gradient binary data; B _l (t) is the column gradient binary data.

Preferably, in step 10, judging whether the shape of the wear scar area is normal according to the gradient binary data obtained in step 9 and the row length diameter and column length diameter of the wear scar area in step 3, the specific method is:

Determine whether the row-direction shape and the column-direction shape are both abnormal. When the row-direction shape and the column-direction shape are both normal, the wear spot image is determined to be normal; otherwise, the wear spot image is abnormal.

Preferably, the integers m and n are arbitrarily selected, and when the integers satisfy the following formula at the same time, it is considered that the shape of the row direction is abnormal:

Among them, α is the length threshold;

If the integers k and p are selected arbitrarily, when the integers satisfy the following formula at the same time, it is considered that the shape of the row direction is abnormal:

Compared with the prior art, the beneficial effects of the present invention are:

The present invention provides a method for detecting abnormal wear scar images based on shape features. Based on the segmentation of the wear scar area and detection of the direction angle, the row and column lengths are used to characterize the row and column shape characteristics of the wear scar area. ; On the basis of interval filtering, the gradient change values of row length and column length are calculated separately; adaptive threshold is used to binarize the gradient change value, and based on the monotonicity of the gradient, it is determined whether the wear spot image is abnormal. Compared with the current method that only uses the diameter difference in two directions to determine, the method of the present invention is based on the abnormal state of the wear spot image based on the shape feature, and has more universal applicability, avoids the error of subjective judgment, and has higher accuracy. Provide a theoretical basis for the availability of test data.

Description of the drawings

Figure 1 is a wear spot image F;

Figure 2 is a diagram f of the wear scar area;

Figure 3 is a graph g of the level of wear scars.

Detailed ways

The present invention will be further described in detail below in conjunction with the accompanying drawings.

The present invention provides a method for detecting abnormal wear spot images based on appearance characteristics, including the following steps:

Step 1: Acquire the wear scar image of the test steel ball through a scanning electron microscope. Specifically: after the four-ball friction test, the three bottom steel balls used in the test were taken out and placed in the scanning electron microscope, and parameters such as the illumination and magnification of the scanning electron microscope were adjusted to clearly collect the wear spots. Image, the collected wear spot image is denoted by F. At the same time, the pixel size of the obtained wear spot image F is M×N, for example, the size of the wear spot image is 768×1024, that is, M=768, N=1024. At the same time, use (i, j) to represent the coordinates of any pixel of the wear spot image F, then i and j represent the row and column of the pixel respectively, and i and j are integers, satisfying: 1≤i≤M And 1≤j≤N holds.

In this embodiment, the wear spot image F is shown in FIG. 1.

Step 2: Automatic segmentation of the wear scar area. Using the inventor mentioned in the patent "a method for measuring wear scar diameter based on steel ball wear scar gradient" (ZL201310752721.3), the wear scar area is automatically segmented, and the divided wear scar area is obtained to obtain a wear scar area map. It is denoted by f, the pixel with f(i,j)=1 represents the wear scar, and the pixel with f(i,j)=0 represents the non-wear scar.

In this embodiment, the wear scar area map f is shown in FIG. 2.

Step 3: Detect the direction angle of the wear scar. Using the inventor mentioned in the patent "a four-ball friction test wear scar image wear scar direction angle automatic measurement method" (ZL201710018314.8) to detect the direction angle of the wear scar, the detected direction angle of the wear scar is represented by w.

In this embodiment, the direction angle of the wear scar is w=35°.

Step 4: Rotate the wear scar to the direction of the line axis according to the direction angle of the wear scar. Rotate the wear scar area map f clockwise around the center of the image by w degrees to obtain the wear scar horizontal map, which is represented by the symbol g. At this time, the wear scar direction is consistent with the row axis direction (horizontal direction). After the rotation transformation, the size of the image is changed from M×N to M _w ×N _w , and i ₀ and i _{1 are used to} represent the first and last rows of the wear scar area in the wear scar level map, which satisfies the calculation formula (1-2) ; Use j ₀ and j _{1 to} denote the first and last columns of the wear scar area in the wear scar level map, which satisfies the calculation formula (3-4); the subsequent operations are only performed on the wear scar area, which can reduce the calculation amount of the method.

i ₀ <i and f(i ₀ ,j)=f(i,j)=1,

i ₁ >i and f(i ₁ ,j)=f(i,j)=1,

j ₀ <j and f(i,j ₀ )=f(i,j)=1,

j ₁ >j and f(i,j ₁ )=f(i,j)=1,

In this embodiment, the wear scar level map g is shown in FIG. 3, and the image size is: M _w =1217, N _w =1281, i ₀ =206, i ₁ =934, j ₀ =230, j ₁ =1013.

Step 5: Calculate the line length. When the wear marks are in the horizontal direction, the length of each line of wear marks is used to characterize the line length. In an ideal state, the wear spot area is elliptical, and the row length increases with the increase of the row number to present an inverted U shape, that is, the row length curve first rises and then drops, and the largest row length is the row length diameter. The row length and row length diameter are shown in formula (5-6).

Among them, D _h (i) represents the length of the i-th row; _rh is the length of the row.

In this embodiment, the row length diameter is _rh =784.

Step 6: Calculate the column length. Similarly, the column-wise length calculation method is the same as the column-wise length. The length of each column represents the column-wise length. The column-wise length of the wear scar area increases with the column number in an inverted U shape. The length of the jth column is D _l (j) and the column The long diameter r _{l is} shown in formula (7-8):

In this embodiment, the column long diameter is r _l =728.

Step 7: Detection of abnormal images based on the long diameter. If there is a large difference between the row length diameter and the column length diameter, and the equation (9) is satisfied, the wear spot image is considered to be abnormal, and then go to step 13; otherwise, go to step 8:

Among them, β is the gap threshold, and the value is 0.167 based on experience.

In this embodiment,

Step 8: Perform interval filtering on the row length and column length of the wear spot area. The wear scar image is composed of wear scars, and the edges of the detected wear scar area are often not smooth. The row length and column length are filtered according to the set interval length to make the small jagged edges smoother. After interval filtering, the interval row length and interval column length of the wear spot area are obtained, and the calculation formula (10-11) is as follows:

Among them, s and t are the indexes of the row interval and column interval respectively, which are integers; L _h (s) represents the row length of the s-th interval; L _l (t) is the column length of the t-th interval; q is the set interval length , Usually between 3-20; [] represents the round-down operator.

In this embodiment, the interval length q is 10, s=0,1,...,72 and t=0,1,...,78.

Step 9: Calculate the gradient change value. The gradient change of the data can reflect the monotonicity of the data curve. When the gradient value is positive, the curve increases monotonously, and vice versa, it can be judged whether the shape of the wear scar area exhibits first increase and then decrease characteristics. The gradient of the interval row length and column length after interval filtering is shown in formula (12-13):

Among them, I _h (s) is the row gradient in the s interval, and I _l (t) is the column gradient in the t interval.

Step 10: Determine the adaptive segmentation threshold. The method uses an adaptive threshold to perform binary segmentation on the gradient change value. The specific operation is:

The elements in are arranged in ascending order, and the order is in the first

The element of the bit is selected as the row segmentation threshold, denoted as _Th ; in the same way, the column segmentation threshold T ₁ can be determined: set

After the elements are arranged in ascending order, the order of

The corresponding element value is the column segmentation threshold T ₁ .

in,

It is a rounding operation.

In this embodiment, T _h = 14.4 and T ₁ = 5.7.

Step 11: Gradient binarization. Binarize the gradient data based on the adaptive threshold to obtain gradient binary data. The data with a value of 1 corresponds to a monotonic increasing curve (that is, the rising section of the curve), and the rest are monotonic decreasing curves. If the monotonic increasing curve segment is not monotonously decreasing The curve separation means that the data change conforms to the change characteristics of first increase and then decrease. Use B _h (s) and B _l (t) to represent the binary data of row and column gradients respectively, and the calculation formula is as shown in (14-15):

In this embodiment, the row and column gradient binary data are shown in Table 1.

Table 1 Row and column gradient binary data B _h (s) and B _l (t)

s	1-31	32-37	38-47	48-52	53-54	55-56	57-58	59-65	66
B h (s)	1	0	1	0	1	0	1	0	1
t	1-50	51-52	53-54	55	56	57-78	79-80	81-102	103
B l (t)	1	0	1	0	1	0	1	0	1

Step 12: Detection of abnormal wear scar images. Use the gradient binary data and the row length diameter and column length diameter of the wear scar area in step 3 to judge whether the shape of the wear scar area is normal: when there is any integer m, n satisfies formula (16-18), the curve The ascending segment of is separated by the descending segment, and the curve shows ascending, descending, ascending and descending, which is inconsistent with the standard current curve, indicating that the line shape is abnormal, otherwise the line shape is normal; similarly, if there is any integer k, p satisfies the formula at the same time (19-21), indicating that the column-wise shape is abnormal; when the row or column direction is abnormal, it is an abnormal wear spot image, and only when the row and column directions are both normal, it is regarded as a normal wear spot image. Normal image data can be used for subsequent analysis, while abnormal image data cannot be used for subsequent analysis and need to be re-tested.

Among them, α is the length threshold, and its value is 2-10.

In this embodiment, the value of α is 4; there is m=32, 33; n=38, 39, 40, 41, 42, 43 satisfies formula (16-18), indicating that the shape of the row is abnormal; there is no k and l Satisfies the formula (19-21), indicating that the column-direction shape is normal, so it is an abnormal wear spot image and needs to be tested again.

Step 13: The method ends.

Claims (10)

1. A method for detecting abnormal wear scar images based on shape features is characterized in that it comprises the following steps:

   Step 1. Preprocess the collected wear scar image to obtain a wear scar area map;

   Step 2: Extract the wear scar area in the wear scar area map;

   Step 3. Calculate the row and column diameters of the wear scar area;

   Step 4. Determine whether the wear scar image is abnormal according to the row and column long diameters of the wear scar level map obtained in step 3. If the wear scar image is abnormal, the algorithm ends; otherwise, go to step 5;

   Step 5: Perform interval filtering on the row length and column length of the wear spot area to obtain the interval row length and the interval column length;

   Step 6. Calculate the gradient change value according to the interval row length and interval column length obtained in step 5;

   Step 7. Determine the adaptive segmentation threshold;

   Step 8. Binarize the gradient change value obtained in step 6 according to the adaptive segmentation threshold obtained in step 7, to obtain gradient binary data;

   Step 10: Determine whether the shape of the wear scar area is normal according to the gradient binary data obtained in step 9 and the row major diameter and column major diameter of the wear scar area in step 3.
2. The method for detecting abnormal wear scar images based on appearance features according to claim 1, characterized in that, in step 2, the wear scar area is extracted in the wear scar area map, and the specific method is:

   S1, determine the wear scar direction angle of the wear scar area map obtained in step 1;

   S2: Rotate the wear scar area map clockwise around the center of the image to obtain a wear scar level map;

   S3, extract the wear scar area in the wear scar level map, where i ₀ and i _{1 are used to} represent the first and last rows of the wear scar area in the wear scar level graph; j ₀ and j _{1 are used to} represent the wear scar in the wear scar level graph The first and last columns of the area:

   i ₀ <i and

   

   i ₁ >i and

   

   j ₀ <j and

   

   j ₁ >j and

   
3. The method for detecting abnormal wear scar images based on shape features according to claim 1, characterized in that, in step 3, the line length diameter of the wear scar area is calculated according to the following formula:

   

   Among them, _rh is the length of the row; D _h (i) is the length of the i-th row,

   

   i ₀ is the first row of the wear scar area; i ₁ is the last row of the wear scar area _{; j 0 is the first row of the wear scar area; j 1} is the last row of the wear scar area;

   Calculate the column long diameter by the following formula:

   

   Among them, r _l is the length of the column; D _l (j) is the length of the j-th column,

   
4. The method for detecting abnormal wear scar images based on appearance characteristics according to claim 1, characterized in that, in step 4, the wear scar is judged according to the row length diameter and the column length diameter of the wear scar level map obtained in step 3. Whether the image is abnormal, the specific method is, if the row length diameter and column length diameter meet the following formula, the wear spot image is considered abnormal:

   

   Among them, β is the gap threshold.
5. The method for detecting abnormal wear scar images based on appearance features according to claim 1, characterized in that, in step 5, the row length and column length of the wear scar area are divided by the following formula combined with the set interval length: Filtering:

   

   

   Among them, s and t are the indexes of the row interval and column interval respectively; L _h (s) represents the row length of the s-th interval; L _l (t) is the column length of the t-th interval; q is the set interval length; [] Represents the round-down operator.
6. The method for detecting abnormal wear scar images based on appearance characteristics according to claim 1, wherein in step 6, the gradient change value is calculated by the interval row length and interval column length obtained in step 5 of the following structure:

   

   

   Among them, I _h (s) is the row gradient in the s interval, and I _l (t) is the column gradient in the t interval.
7. The method for detecting abnormal wear spot images based on shape features according to claim 6, characterized in that, in step 7, the adaptive segmentation threshold is determined, and the specific method is:

   Will gather

   

   The elements in are arranged in ascending order, and the order of

   

   The element corresponding to the bit is used as the row segmentation threshold _Th ;

   Will gather

   

   After the elements are arranged in ascending order, the order of

   

   The corresponding element value is used as the column segmentation threshold T ₁ ;

   in,

   

   It is a rounding operation.
8. The method for detecting abnormal wear scar images based on shape features according to claim 1, characterized in that, in step 8, the gradient change obtained in step 6 is binary-valued according to the adaptive segmentation threshold obtained in step 7 To obtain the gradient binary data, the specific method uses the following formula to calculate:

   

   

   Among them, B _h (s) is the row gradient binary data; B _l (t) is the column gradient binary data.
9. The method for detecting abnormal wear scar images based on shape features according to claim 1, characterized in that, in step 10, according to the gradient binary data obtained in step 9 and the line length diameter of the wear scar area in step 3 Determine whether the shape of the wear scar area is normal by using the long diameter and the column diameter. The specific method is:

   Determine whether the row-direction shape and the column-direction shape are both abnormal. When the row-direction shape and the column-direction shape are both normal, the wear spot image is determined to be normal; otherwise, the wear spot image is abnormal.
10. The method for detecting abnormal wear scar images based on appearance features according to claim 9, characterized in that integers m and n are arbitrarily selected, and when the integers satisfy the following formula at the same time, it is considered that the shape of the row direction is abnormal:

    

    

    

    Among them, α is the length threshold;

    If the integers k and p are selected arbitrarily, when the integers satisfy the following formula at the same time, it is considered that the shape of the row direction is abnormal:

    

    

    

Images