Piston detection method

Abstract

The invention relates to a piston detection method, which comprises the step of detecting defects of a piston and specifically comprises the following steps of: 1, constructing a CT image segmentation model, and training the constructed CT image segmentation model to obtain a trained CT image segmentation model; 2, S pistons are placed on a fixing tool to be detected, N pistons can be placed on each layer of the fixing tool, both S and N are positive integers, CT tomography is conducted on the top, the middle and the bottom of an inner cooling oil duct of the piston placed on each layer of the fixing tool, multiple CT images are obtained, two-channel images are obtained from all the CT images, and the two-channel images are used for detecting the pistons; and inputting the obtained dual-channel image into the trained CT image segmentation model to obtain a predicted defect. The method has the advantage that the detection accuracy and the detection efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the field of detection technology, and in particular to a piston detection method. Background Art

[0002] The piston is the core component of the engine. Under high temperature and high pressure operating conditions, its quality directly affects the performance of the engine. The piston inner cooling oil channel is generally formed by the water-soluble salt core method. By spraying the oil into the oil channel, it absorbs heat in the oil channel and then flows out, which can effectively reduce the temperature and thermal deformation of important parts such as the piston head and ring groove, but it also weakens the strength of the piston head structure accordingly.

[0003] Once the piston inner cooling oil channel has a large axial or radial deviation, the corresponding deviation side will be easily punctured during operation, thus causing the dangerous situation of piston failure. In addition, casting residues (aluminum inclusions, residual iron, residual salt) in the inner cooling oil channel may cause the oil channel to be blocked, and the piston cannot dissipate heat in time; casting defects (looseness, holes, inclusions, cracks) on the piston base will reduce the mechanical properties of the piston, which also poses a major safety hazard.

[0004] In industrial applications, the use of industrial CT detection technology can effectively detect the internal quality of the piston. When performing industrial CT detection, the piston is first digitally rayed, and the axial offset of the piston's internal cooling oil channel is measured based on the digital ray projection. Then, tomography is performed on the top, middle, and bottom of the piston's internal cooling oil channel to determine whether there are defects. Finally, the radial offset of the piston's internal cooling oil channel is measured based on the tomography image of the middle. During the detection process, multiple staff members are required to collaborate to measure and record the corresponding data and evaluate the internal quality of the piston. However, due to the large production batch of pistons, the efficiency of manual evaluation is low, the labor cost required is high, and the evaluation results are limited by the experience of the inspectors, etc., which is prone to measurement errors, false detections, missed detections, etc.; in addition, traditional image processing technology is limited by interference such as noise and structural artifacts, and is also prone to errors. For this reason, further improvements to the existing technology are needed. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a piston detection method which can improve detection accuracy and detection efficiency in view of the above-mentioned prior art.

[0006] The technical solution adopted by the present invention to solve the above technical problem is: a piston detection method, characterized in that it includes detecting defects of the piston, specifically the following steps:

[0007] Step 1: construct a CT image segmentation model, and train the constructed CT image segmentation model to obtain a trained CT image segmentation model;

[0008] The specific process of training the CT image segmentation model is as follows:

[0009] Step 1-1, collecting a certain number of CT images of pistons, and identifying the piston contour in each CT image;

[0010] Step 1-2: Perform the following cutting operations on each CT image:

[0011] The center of the piston circle identified in the current CT image is taken as the center of the piston image corresponding to the current CT image, and the piston diameter is taken as the length of the piston image corresponding to the current CT image, and a square piston image corresponding to the current CT image is obtained by cutting;

[0012] Step 1-3, enhancing the local contrast of the piston image obtained in step 1-2 to obtain a piston grayscale image;

[0013] Step 1-4, construct a distance feature matrix for each piston image, assign the pixel value of any point in the matrix to the Euclidean distance from the point to the center of the piston image, and evenly map the distance feature matrix to a grayscale space of 0 to 255 to obtain a distance feature grayscale image;

[0014] Step 1-5, merging the piston grayscale image and the distance feature grayscale image into a dual-channel image;

[0015] Step 1-6: input the dual-channel image into the constructed CT image segmentation model to obtain the prediction result, and annotate the piston image obtained in step 1-2 to obtain the piston defect mask image; calculate the loss function using the prediction result and the piston defect mask image, and then use the loss function to update the parameters of the constructed CT image segmentation model to obtain the CT image segmentation model after training;

[0016] Step 2: Place S pistons on a fixed tool for inspection. Each layer of the fixed tool can hold N pistons, where S and N are both positive integers. Perform a CT scan on the top, middle, and bottom of the inner cooling oil passage of the pistons placed on each layer of the fixed tool to obtain multiple CT images. Obtain dual-channel images from all CT images in the same manner as steps 1-1 to 1-5, and input the acquired dual-channel images into the trained CT image segmentation model to obtain predicted defects.

[0017] Preferably, in step 1-1, Hough circle transform is performed on all CT images to identify the piston contour in each CT image.

[0018] Furthermore, it also includes calculating the axial offset value of the inner cooling oil channel of the piston, specifically:

[0019] A digital radiographic image segmentation model is constructed, a certain number of digital radiographic images of pistons are collected, and the oil channel contours and the piston inner cavity contours in the digital radiographic images of the pistons are annotated to form a data set; the constructed digital radiographic image segmentation model is trained using the data set to obtain a trained digital radiographic image segmentation model;

[0020] Perform digital ray scanning on the pin hole-facing direction of multiple pistons to be tested to obtain a digital ray image of the piston, and input the digital ray image of the piston into the trained digital ray image segmentation model to obtain the oil channel contour and the piston inner cavity contour in the digital ray image of the piston;

[0021] The oil channel contour in the piston digital radiographic image is detected by the connected domain detection method to obtain 2S connected domains. The center coordinates of each connected domain are calculated. The 2S connected domains are numbered 1 to 2S according to the center coordinates. The relationship between the center coordinates is:

[0022] when hour:

[0023] x 1 <x 2 <… <x 2S

[0024] Among them, x 1 is the central horizontal coordinate of the first connected domain, x 2 is the central horizontal coordinate of the second connected domain, x 2S is the central horizontal coordinate of the 2Sth connected domain;

[0025] when hour:

[0026]

[0027] Wherein, j is a positive integer greater than 1, is a universal quantifier; k∈{0, 1, 2…j-1}, (x 1 ,y 1 ) are the center coordinates of the first connected domain;

[0028] Calculate the axial offset value of the inner cooling oil channel of each piston 1~S:

[0029] d 轴向偏移 (1)=|y max (1)-y max (2)|

[0030] d 轴向偏移 (2)=|y max (3)-y max (4)|

[0031] …

[0032] d 轴向偏移 (S)=|y max (2S-1)-y max (2S)|

[0033] Among them, d 轴向偏移 (1) is the axial offset value of the inner cooling oil channel of the first piston, y max (1) represents the maximum value of the ordinate in the first connected domain, y max (2) represents the maximum value of the ordinate in the second connected domain, d 轴向偏移 (2) is the axial offset value of the inner cooling oil channel of the second piston, y max (3) represents the maximum value of the ordinate in the third connected domain, y max (4) represents the maximum value of the ordinate in the fourth connected domain, d 轴向偏移 (S) is the axial offset value of the inner cooling oil channel of the Sth piston, y max (2S-1) represents the maximum value of the ordinate in the 2S-1th connected domain, y max (2S) represents the maximum value of the ordinate in the 2Sth connected domain.

[0034] Furthermore, the method further includes verifying defects near the central area of ​​the piston cavity, and the specific verification steps are as follows:

[0035] Step a: According to the piston inner cavity contour in the obtained piston digital radiographic image, obtain the inner cavity part of the piston at (x 1 ,x 2 ), and according to the piston CT tomography height H and tomography slice thickness d, the piston inner cavity is obtained Coordinates at altitude

[0036] Step b: and In a rectangle formed by four points,

[0037] The grayscale cumulative value of each column:

[0038]

[0039] Among them, ρ 1 , 2 They represent the gray values ​​of the cavity part and the piston base in the CT image respectively;

[0040] Step c: Calculate the grayscale average value of each column:

[0041] Step d: With the piston axis area as the horizontal coordinate and the piston center as the coordinate center, a coordinate system is constructed on the piston CT cutting image to extract the gray value change curve S′(x) of the defect area in the y=0 direction, with a range of x 1 ' <x<x 2 ′, where the identified defect range (x 1 ′,x 2 ′)∈(x 1 ,x 2 );

[0042] Step e: Calculate S′(x) and The similarity SSIM is calculated as follows:

[0043]

[0044] in, express In (x 1 ′,x 2 ′), the average value within S′(x) It shows that S′(x) in (x 1 ′,x 2 ′), express and S′(x) in (x 1 ′,x 2 ′), C 1 and C 2 Represents a preset constant, express In (x 1 ′,x 2 ′) within the square of the standard deviation, σ S′(x) 2 It means that S′(x) is 1 ′,x 2 ′) is the square of the standard deviation within;

[0045] Step f: Determine whether SSIM is less than the preset constant C 3 If yes, it is considered that there is a defect near the center area of ​​the piston cavity; if no, it is determined that there is no defect in the center area of ​​the piston cavity.

[0046] Furthermore, it also includes calculating the radial offset value of the inner cooling oil channel of the piston, specifically:

[0047] Cutting the piston image in the manner of step 1-2, and extracting the CT tomographic image of the middle portion of the inner cooling oil passage of the piston image;

[0048] Perform circular fitting on the outer wall of the piston in the CT tomographic image of the middle part of the inner cooling oil channel and calculate the coordinates of the center of the fitted circle

[0049] Calculate the coordinates of the center of the fitted circle The Euclidean distance between the piston center in step 1-2 is the radial offset value of the inner cooling oil channel of the piston.

[0050] Compared with the prior art, the advantages of the present invention are: by constructing a CT image segmentation model, training the constructed CT image segmentation model, and predicting piston defects through the trained CT image segmentation model, on the one hand, the method uses image processing technology to realize piston detection, without the need for multiple staff members to collaborate in measuring and recording corresponding data, and to evaluate the internal quality of the piston, and the detection results are not limited by the experience of the detection personnel and other aspects. In addition, by merging the piston grayscale image and the distance feature grayscale image into a dual-channel image, the image can be effectively prevented from being disturbed by noise, structural artifacts, etc., thereby improving the detection accuracy and detection efficiency; on the other hand, by placing multiple pistons on a fixed tooling for detection, multiple pistons can be detected at one time, further improving the detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is an image of a piston in an embodiment of the present invention;

[0052] Figure 2 for Figure 1 The piston defect mask image corresponding to the piston image;

[0053] Figure 3 is a distance feature grayscale image in an embodiment of the present invention;

[0054] Figure 4 is a digital radiographic image of a piston in an embodiment of the present invention;

[0055] Figure 5 Based on Figure 4 The obtained oil channel profile;

[0056] Figure 6 It is a CT tomographic image of the middle part of the internal cooling oil channel in the embodiment of the present invention. DETAILED DESCRIPTION

[0057] The present invention is further described in detail below with reference to the accompanying drawings.

[0058] The piston detection method in this embodiment includes detecting defects of the piston, calculating the axial offset value of the inner cooling oil channel of the piston, verifying defects near the central area of ​​the inner cavity of the piston, and calculating the radial offset value of the inner cooling oil channel of the piston.

[0059] In this embodiment, the specific steps of detecting defects of the piston are:

[0060] Step 1: construct a CT image segmentation model, and train the constructed CT image segmentation model to obtain a trained CT image segmentation model;

[0061] The CT image segmentation model in this embodiment can adopt a deep learning network commonly used in the prior art, which will not be described in detail here; for example: U-Net network;

[0062] The specific process of training the CT image segmentation model is as follows:

[0063] Step 1-1, collecting a certain number of CT images of pistons, and identifying the piston contour in each CT image; in this embodiment, performing Hough circle transform on all CT images to identify the piston contour in each CT image;

[0064] Step 1-2: Perform the following cutting operations on each CT image:

[0065] The center of the piston circle identified in the current CT image is taken as the center of the piston image corresponding to the current CT image, and the piston diameter is taken as the length of the piston image corresponding to the current CT image, and a square piston image corresponding to the current CT image is obtained by cutting;

[0066] Step 1-3, enhancing the local contrast of the piston image obtained in step 1-2 to obtain a piston grayscale image;

[0067] In this embodiment, contrast-limited adaptive histogram equalization is used to enhance the local contrast of the piston image;

[0068] Step 1-4: construct a distance feature matrix for each piston image, assign the pixel value of any point in the matrix to the Euclidean distance from the point to the center of the piston image, and evenly map the distance feature matrix to the grayscale space of 0 to 255 to obtain a distance feature grayscale map; the distance feature grayscale map is shown in Figure 3 As shown;

[0069] Step 1-5, merging the piston grayscale image and the distance feature grayscale image into a dual-channel image;

[0070] Step 1-6: Input the dual-channel image into the constructed CT image segmentation model to obtain the prediction result, and annotate the piston image obtained in step 1-2 to obtain the piston defect mask image; use the prediction result and the piston defect mask image to calculate the loss function, and then use the loss function to update the parameters of the constructed CT image segmentation model to obtain the CT image segmentation model after training; Figure 1 As shown in the figure, this is one of the piston images. The piston defect mask image corresponding to this piston image is as follows: Figure 2 As shown;

[0071] Step 2, place S pistons on a fixed tool for inspection, each layer of the fixed tool can hold N pistons, S and N are both positive integers, perform a CT scan on the top, middle and bottom of the inner cooling oil passage of the pistons placed on each layer of the fixed tool, obtain multiple CT images, and obtain dual-channel images from all CT images in the same manner as steps 1-1 to 1-5, and input the obtained dual-channel images into the trained CT image segmentation model to obtain predicted defects; in this embodiment, one layer of the fixed tool can hold 3 pistons, the fixed tool has a total of 4 layers, and 12 pistons can be placed at a time;

[0072] The axial offset value of the inner cooling oil channel of the piston is calculated as follows:

[0073] A digital radiographic image segmentation model is constructed, a certain number of digital radiographic images of pistons are collected, and the oil channel contours and the piston inner cavity contours in the digital radiographic images of the pistons are annotated to form a data set; the constructed digital radiographic image segmentation model is trained using the data set to obtain a trained digital radiographic image segmentation model; the digital radiographic image segmentation model in this embodiment also adopts a deep learning network commonly used in the prior art, which will not be described in detail here;

[0074] Digital ray scanning is performed on the pin hole-facing direction of multiple pistons to be tested to obtain a digital ray image of the piston, and the digital ray image of the piston is input into the trained digital ray image segmentation model to obtain the oil channel contour and the piston inner cavity contour in the digital ray image of the piston; Figure 4 As shown, it is a piston digital radiographic image, which includes digital radiographic images of 12 pistons. Figure 5 Based on Figure 4 The obtained oil channel profile;

[0075] The oil channel contour in the piston digital radiographic image is detected by the connected domain detection method to obtain 2S connected domains. The center coordinates of each connected domain are calculated. The 2S connected domains are numbered 1 to 2S according to the center coordinates. The relationship between the center coordinates is:

[0076] when hour:

[0077] x 1 <x 2 <… <x 2S

[0078] Among them, x 1 is the central horizontal coordinate of the first connected domain, x 2 is the central horizontal coordinate of the second connected domain, x 2S is the central horizontal coordinate of the 2Sth connected domain;

[0079] when hour:

[0080]

[0081] Wherein, j is a positive integer greater than 1, is a universal quantifier; k∈{0, 1, 2…j-1}, (x 1 ,y 1 ) are the center coordinates of the first connected domain;

[0082] Calculate the axial offset value of the inner cooling oil channel of each piston 1~S:

[0083] d 轴向偏移 (1)=|y max (1)-y max (2)|

[0084] d 轴向偏移 (2)=|y max (3)-y max (4)|

[0085] …

[0086] d 轴向偏移 (S)=|y max (2S-1)-y max (2S)|

[0087] Among them, d 轴向偏移 (1) is the axial offset value of the inner cooling oil channel of the first piston, y max (1) represents the maximum value of the ordinate in the first connected domain, y max (2) represents the maximum value of the ordinate in the second connected domain, d 轴向偏移 (2) is the axial offset value of the inner cooling oil channel of the second piston, y max (3) represents the maximum value of the ordinate in the third connected domain, y max (4) represents the maximum value of the ordinate in the fourth connected domain, d 轴向偏移 (S) is the axial offset value of the inner cooling oil channel of the Sth piston, y max (2S-1) represents the maximum value of the ordinate in the 2S-1th connected domain, y max (2S) represents the maximum value of the ordinate in the 2Sth connected domain.

[0088] The specific steps for verifying defects near the center area of ​​the piston cavity are:

[0089] Step a: According to the piston inner cavity contour in the obtained piston digital radiographic image, obtain the inner cavity part of the piston at (x 1 ,x 2), and according to the piston CT tomography height H and tomography slice thickness d, the piston inner cavity is obtained Coordinates at altitude

[0090] Step b: and In a rectangle formed by four points,

[0091] The grayscale cumulative value of each column:

[0092]

[0093] Among them, ρ 1 , 2 They represent the gray values ​​of the cavity part and the piston base in the CT image respectively;

[0094] Step c: Calculate the grayscale average value of each column:

[0095] Step d: With the piston axis area as the horizontal coordinate and the piston center as the coordinate center, a coordinate system is constructed on the piston CT cutting image to extract the gray value change curve S′(x) of the defect area in the y=0 direction, with a range of x 1 ' <x<x 2 ′, where the identified defect range (x 1 ′,x 2 ′)∈(x 1 ,x 2 );

[0096] Step e: Calculate S′(x) and The similarity SSIM is calculated as follows:

[0097]

[0098] in, express In (x 1 ′,x 2 ′), the average value within S′(x) It shows that S′(x) in (x 1 ′,x 2 ′), express and S′(x) in (x 1 ′,x 2 ′), C 1 and C 2 Represents a preset constant, express In (x 1 ′,x 2′) within the square of the standard deviation, σ S′(x) 2 It means that S′(x) is 1 ′,x 2 ′) is the square of the standard deviation within;

[0099] Step f: Determine whether SSIM is less than the preset constant C 3 If yes, it is considered that there is a defect near the center of the piston cavity; if no, it is considered that there is no defect in the center of the piston cavity. 3 The specific value of is determined based on experience.

[0100] The radial offset value of the inner cooling oil channel of the piston is calculated as follows:

[0101] Cutting the piston image in the manner of step 1-2, and extracting the CT tomographic image of the middle portion of the inner cooling oil passage of the piston image;

[0102] Perform circular fitting on the outer wall of the piston in the CT tomographic image of the middle part of the inner cooling oil channel and calculate the coordinates of the center of the fitted circle like Figure 6 As shown;

[0103] Calculate the coordinates of the center of the fitted circle The Euclidean distance between the piston center in step 1-2 is the radial offset value of the inner cooling oil channel of the piston.

[0104] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A piston detection method, characterized in that Including the detection of piston defects, the specific steps are: Step 1: construct a CT image segmentation model, and train the constructed CT image segmentation model to obtain a trained CT image segmentation model; The specific process of training the CT image segmentation model is as follows: Step 1-1, collecting a certain number of CT images of pistons, and identifying the piston contour in each CT image; Step 1-2: Perform the following cutting operations on each CT image: The center of the piston circle identified in the current CT image is taken as the center of the piston image corresponding to the current CT image, and the piston diameter is taken as the length of the piston image corresponding to the current CT image, and a square piston image corresponding to the current CT image is obtained by cutting; Step 1-3, enhancing the local contrast of the piston image obtained in step 1-2 to obtain a piston grayscale image; Step 1-4, construct a distance feature matrix for each piston image, assign the pixel value of any point in the matrix to the Euclidean distance from the point to the center of the piston image, and evenly map the distance feature matrix to a grayscale space of 0 to 255 to obtain a distance feature grayscale image; Step 1-5, merging the piston grayscale image and the distance feature grayscale image into a dual-channel image; Step 1-6: input the dual-channel image into the constructed CT image segmentation model to obtain the prediction result, and annotate the piston image obtained in step 1-2 to obtain the piston defect mask image; calculate the loss function using the prediction result and the piston defect mask image, and then use the loss function to update the parameters of the constructed CT image segmentation model to obtain the CT image segmentation model after training; Step 2: Place S pistons on a fixed tool for inspection. Each layer of the fixed tool can hold N pistons, where S and N are both positive integers. Perform a CT scan on the top, middle, and bottom of the inner cooling oil passage of the pistons placed on each layer of the fixed tool to obtain multiple CT images. Obtain dual-channel images from all CT images in the same manner as steps 1-1 to 1-5, and input the acquired dual-channel images into the trained CT image segmentation model to obtain predicted defects.

2. The piston detection method according to claim 1, characterized in that: In step 1-1, Hough circle transform is performed on all CT images to identify the piston contour in each CT image.

3. The piston detection method according to claim 1, characterized in that: It also includes the calculation of the axial offset value of the inner cooling oil channel of the piston, specifically: A digital radiographic image segmentation model is constructed, a certain number of digital radiographic images of pistons are collected, and the oil channel contours and the piston inner cavity contours in the digital radiographic images of the pistons are annotated to form a data set; the constructed digital radiographic image segmentation model is trained using the data set to obtain a trained digital radiographic image segmentation model; Perform digital ray scanning on the pin hole-facing direction of multiple pistons to be tested to obtain a digital ray image of the piston, and input the digital ray image of the piston into the trained digital ray image segmentation model to obtain the oil channel contour and the piston inner cavity contour in the digital ray image of the piston; Using the connected component detection method for the oil passage contour in the piston digital radiography image, 2S connected components are obtained; calculate the central coordinates of each connected component. For the 2S connected components, number them from 1 to 2S according to the central coordinates. The relationship between the central coordinates is as follows: when hour: x1 <x2<…<x 2S Among them, x1 is the central horizontal coordinate of the first connected domain, x2 is the central horizontal coordinate of the second connected domain, and x 2S is the central horizontal coordinate of the 2Sth connected domain; when hour: Wherein, j is a positive integer greater than 1, is a universal quantifier; k∈{0, 1, 2…j-1}, (x1, y1) is the center coordinate of the first connected domain; Calculate the axial offset value of the internal cooling oil passage of each piston from 1 to S: d 轴向偏移 (1)=|y max (1)-y max (2)| d 轴向偏移 (2)=|y max (3)-y max (4)| …… d 轴向偏移 (S)=|and max (2S-1)-and max (2S)| Among them, d 轴向偏移 (1) is the axial offset value of the inner cooling oil channel of the first piston, y max (1) represents the maximum value of the ordinate in the first connected domain, y max (2) represents the maximum value of the ordinate in the second connected domain, d 轴向偏移 (2) is the axial offset value of the inner cooling oil channel of the second piston, y max (3) represents the maximum value of the ordinate in the third connected domain, y max (4) represents the maximum value of the ordinate in the fourth connected domain, d 轴向偏移 (S) is the axial offset value of the inner cooling oil channel of the Sth piston, t max (2S-1) represents the maximum value of the ordinate in the 2S-1th connected domain, y max (2S) represents the maximum value of the ordinate in the 2Sth connected domain.

4. The piston detection method according to claim 3, characterized in that: It also includes verifying the defects near the central region of the piston inner cavity. The specific verification steps are as follows: Step a: According to the piston inner cavity contour in the obtained piston digital radiographic image, obtain the distribution cumulative function P(x) of the piston inner cavity part between (x1, x2), and obtain the piston inner cavity according to the piston CT tomography height H and tomography slice thickness d. Coordinates at altitude Step b: and In a rectangle formed by four points, The gray level cumulative value of each column: Among them, ρ1 and ρ2 respectively represent the gray level values of the inner cavity part and the piston matrix in the CT image; Step c: Calculate the grayscale average value of each column: Step d: Taking the axis region of the piston as the abscissa and the piston center as the coordinate center, construct a coordinate system on the piston CT cut image, and extract the gray level value change curve S′(x) in the y = 0 direction of the defect region, with the range x1′ < x < x2′, where the identified defect range (x1′, x2′) ∈ (x1, x2); Step e: Calculate S′(x) and The similarity SSIM is calculated as follows: in, express The average value in (x1′,x2′), μ S′(x) shows the average value of S′(x) in (x1′,x2′), express and the covariance of S′(x) in (x1′,x2′), C1 and C2 represent preset constants, express The square of the standard deviation in (x1′,x2′), σ S′(x) 2 represents the square of the standard deviation of S′(x) in (x1′,x2′); Step f: Judge whether the SSIM is less than the preset constant C3. If so, it is considered that there are defects near the central region of the piston inner cavity; if not, it is determined that there are no defects in the central region of the piston inner cavity.

5. The piston detection method according to any one of claims 1 to 4, characterized in that: It also includes calculating the radial offset value of the internal cooling oil passage of the piston. Specifically: Cut the piston image in the manner described in steps 1 - 2, and extract the CT tomographic image of the middle part of the internal cooling oil passage of the piston image; Perform circular fitting on the outer wall of the piston in the CT tomographic image of the middle part of the inner cooling oil channel and calculate the coordinates of the center of the fitted circle Calculate the coordinates of the center of the fitted circle The Euclidean distance between the piston center in step 1-2 is the radial offset value of the inner cooling oil channel of the piston.

Images