Sodium-cooled fast reactor sodium pipe weld crack infrared nondestructive testing method

Abstract

This invention belongs to the technical field of non-destructive testing in nuclear power, specifically disclosing an infrared non-destructive testing method for weld cracks in sodium-cooled fast reactor pipelines. When an anomaly is detected by thermocouples in the secondary loop pipelines of a sodium-cooled fast reactor, the reactor needs to be shut down. At this time, the liquid sodium in the pipeline will slowly cool down. Due to structural differences, a surface temperature difference will form between defective and non-defective areas during the cooling process. This invention utilizes an infrared thermal imager equipped with a telephoto lens and a laser rangefinder to photograph the weld seam of the cooled pipeline after the insulation layer has been removed. Then, a trained image segmentation network is used to segment and identify the image. Through curve fitting, the pixel length of the segmented area is extracted, and combined with laser ranging information, the true size of the weld crack is determined. This invention overcomes the limitations of traditional methods for non-destructive testing of high-temperature pipelines through non-contact infrared non-destructive testing technology.

Description

Technical Field

[0001] This invention belongs to the technical field of non-destructive testing in nuclear power, specifically relating to an infrared non-destructive testing method for weld cracks in sodium-related pipelines in sodium-cooled fast reactors. Background Technology

[0002] Sodium-cooled fast reactors typically use welding to connect sodium-related pipelines. Due to factors such as high temperatures and corrosion, cracks inevitably form at the welds. If these cracks are not detected early, they can develop into pipeline ruptures, leading to the leakage of liquid sodium. Because sodium is chemically reactive, it reacts upon contact with air, creating a dangerous hazard and affecting equipment operation. Therefore, regular non-destructive testing (NDT) is necessary for the welds of sodium-related high-temperature pipelines. Common NDT methods used in the nuclear industry include penetrant testing, visual inspection, ultrasonic testing, eddy current testing, and radiographic testing. However, the high temperatures involved in sodium-related pipelines prevent conventional NDT methods from reaching the target area, leading to testing failures.

[0003] Therefore, it is necessary to develop a non-destructive testing method for sodium-cooled fast reactors that can be used in nuclear power plant environments to solve the problem of detecting weld cracks in high-temperature pipelines. Summary of the Invention

[0004] The purpose of this invention is to provide an infrared non-destructive testing method for weld cracks in sodium-cooled fast reactor pipelines. This method utilizes the temperature difference of the object itself to identify defects. Through non-contact infrared non-destructive testing, non-destructive testing is carried out without getting close to the target object. In high-temperature target scenarios, the infrared thermal imager is equipped with a dedicated infrared telephoto lens to achieve non-destructive testing at medium and long distances, thus solving the problems mentioned in the background art.

[0005] Technical solution to achieve the purpose of this invention:

[0006] An infrared non-destructive testing method for weld cracks in sodium-cooled fast reactor pipelines, the method comprising:

[0007] Step 1: Produce weld-cracked pipes of different sizes, connect the pipes to a high-temperature circulation pipeline, and conduct high-temperature simulation experiments;

[0008] Step 2: Using the pipe with weld cracks as the inspection target, an infrared thermal imager equipped with a telephoto lens is used to take pictures of the inspection target from a distance to obtain infrared image sequence data; the laser rangefinder on the infrared thermal imager is used to record the distance between the thermal imager and the inspection target to obtain the length of the laser rangefinder.

[0009] Step 3: Collect infrared image sequence data of different lengths, depths, and temperatures, and preprocess the infrared image sequence data to obtain single-frame image data;

[0010] Step 4: Use a recurrent adversarial generative network to augment the weld crack image data to obtain an augmented weld crack image dataset.

[0011] Step 5: Use an image segmentation network to segment and identify the expanded dataset of images with weld cracks, obtain the mask information of the cracks, and then use image post-processing methods to obtain the true length of the cracks.

[0012] Furthermore, in step 1, the temperature of the high-temperature circulation pipeline is set below 300°C to simulate the temperature of a real sodium-related pipeline after it is shut down, and the pipeline is simulated to have no shielding or insulation layer on the outside.

[0013] Furthermore, in step 1, the high-temperature simulation experiment method is as follows: after heating the medium in the pipeline to 280℃~320℃, it is naturally cooled. During the cooling process, the medium temperature should not be lower than 100℃.

[0014] Furthermore, in step 2, the infrared image sequence data includes infrared image sequence data of weld cracks and infrared image sequence data of cracks without welds.

[0015] Furthermore, in step 3, the infrared image sequence data is preprocessed using the following formula to obtain single-frame image data:

[0016] T i (x,y)={T i (x,y,t) max -T i (x,y,t) min} 2

[0017]

[0018] In the formula, T i (x,y,t) max and T i (x,y,t) min These represent the highest and lowest temperatures of pixel (x, y) in the sequence; T i (x,y) represents the temperature value of a single pixel (x,y) after processing, T min and T max For {T i The lowest and highest temperatures in (x,y)}, and pi(x,y) is the gray value of a single pixel (x,y).

[0019] Furthermore, step 4 specifically includes:

[0020] Step 4.1: Divide the single-frame image data obtained in Step 3 into two groups of data and save them: images with weld cracks and images without weld cracks.

[0021] Step 4.2: Use the two sets of data as the two inputs of the recurrent adversarial generative network to train it, and obtain the trained recurrent adversarial generative network model.

[0022] Step 4.3: Use the image without weld cracks as input to the trained recurrent adversarial generative network model to generate images with weld cracks. Select the images with high crack generation quality and combine them with the real weld crack images collected in Step 3 to form the expanded dataset of images with weld cracks.

[0023] Furthermore, step 5 specifically includes:

[0024] Step 5.1: Use an open-source annotation tool to annotate all the weld crack image data in the expanded weld crack image dataset to obtain the annotation file;

[0025] Step 5.2: Feed the labeled file and the image with weld cracks into the image segmentation network for training. When the AP50 of the image segmentation network is greater than 0.8, end the training; otherwise, adjust the network parameters and retrain.

[0026] Step 5.3: The image segmentation network outputs crack detection boxes and segmented mask information. Using the image skeletonization method, the mask information is converted into crack information with a single pixel width.

[0027] Step 5.4: Calculate the number of non-zero pixel values ​​after mask skeletonization within the crack detection box to obtain the crack skeleton length;

[0028] Step 5.5: Calculate the actual length of the crack based on the length of the crack skeleton.

[0029] Furthermore, in step 5.5, the formula for calculating the true length of the crack is:

[0030]

[0031] In the formula, L is the actual length of the crack, l is the skeleton length of the crack, f is the focal length of the infrared thermal imager, and D is the length of the laser ranging in step 2.

[0032] The beneficial technical effects of this invention are as follows:

[0033] This invention utilizes an infrared thermal imager equipped with a telephoto lens to inspect welds in high-temperature sodium-related pipelines of sodium-cooled fast reactors. It leverages the varying cooling rates in different areas during the natural cooling process after reactor shutdown, resulting in temperature differences, to perform flaw detection. Therefore, it features non-contact and rapid detection, solving the problem of mainstream non-destructive testing methods in the nuclear industry failing in high-temperature pipeline scenarios. Furthermore, this invention addresses the issue of scarce real crack data from nuclear power plants, which leads to failures in automatic target identification, through pipeline simulation experiments and data generation via a cyclic adversarial generative network. Attached Figure Description

[0034] Figure 1 The present invention provides a flowchart of an infrared non-destructive testing method for weld cracks in sodium-cooled fast reactor pipelines.

[0035] Figure 2 This invention provides a method for converting an infrared image sequence into a single-frame image in an infrared nondestructive testing method for weld cracks in sodium-related pipelines in a sodium-cooled fast reactor.

[0036] Figure 3 This is a schematic diagram of the cyclic adversarial generative network data augmentation method in the infrared non-destructive testing method for weld cracks in sodium-cooled fast reactor pipelines provided by the present invention. Detailed Implementation

[0037] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0038] In sodium-cooled fast reactors, the secondary loop piping needs to be shut down when an anomaly is detected by the thermocouple. During this process, the liquid sodium in the piping slowly cools down. Due to structural differences, a surface temperature difference forms between the defective and non-defective areas. This invention utilizes an infrared thermal imager equipped with a telephoto lens and a laser rangefinder to photograph the weld seam of the cooling piping after the insulation layer has been removed. A trained image segmentation network is then used to segment and identify the image. Through curve fitting, the pixel length of the segmented region is extracted. Combined with laser ranging information, the true size of the weld crack is determined.

[0039] like Figure 1 As shown, this invention provides an infrared non-destructive testing method for weld cracks in sodium-contaminated pipelines in sodium-cooled fast reactors, specifically including the following steps:

[0040] Step 1: Artificial cracks of varying shapes and sizes are created on the welds of stainless steel pipes of different diameters using electrical discharge machining (EDM). These pipes are then connected to a high-temperature liquid circulation loop. The temperature of the high-temperature circulation loop is set below 300℃ to simulate the temperature of a real sodium-containing pipeline after shutdown. No shielding or insulation layer is installed on the outside of the simulated pipe. The circulation loop uses media with viscosities and specific heat capacities similar to liquid sodium, such as high-temperature silicone oil (dimethyl silicone oil, etc.) and low-viscosity PAO synthetic oil. The simulated liquid medium is heated to 280℃–320℃ and then allowed to cool naturally, ensuring that the liquid temperature does not fall below 100℃ during cooling (liquid sodium solidifies below 100℃). If the temperature falls below 100℃, the pipe is reheated to 280℃–320℃, and natural cooling data is collected again.

[0041] Step 2: Using a simulated stainless steel weld cracked pipe as the inspection target, an infrared thermal imager equipped with a telephoto lens is used to photograph the inspection target (simulated stainless steel weld cracked pipe). The objects photographed include areas with weld cracks and areas without weld cracks, obtaining infrared image sequence data of weld cracks and infrared image sequence data of no weld cracks; the number of frames acquired in the image sequence is set to N (e.g., a value of 100). If the actual number of frames n acquired during the process of the pipe cooling from 280℃ to 320℃ to 100℃ is less than N, then the actual number of frames n acquired is used as the final number of frames; if the actual number of frames n acquired is greater than N, then the first N frames acquired are used.

[0042] Simultaneously, the laser rangefinder on the infrared thermal imager is used to record the distance between the thermal imager and the target being detected, thus obtaining the length of the laser rangefinder.

[0043] Step 3: Collect infrared temperature image sequences of weld cracks at different lengths, depths, and temperatures, as well as infrared temperature image sequences of weld cracks without welds. Preprocess each infrared temperature image sequence using the following formula to convert the infrared temperature image sequence data into single-frame image data:

[0044] T i (x,y)={T i (x,y,t) max -T i (x,y,t) min} 2

[0045]

[0046] In the formula, T i (x,y,t) max and T i (x,y,t) min These represent the highest and lowest temperatures of pixel (x, y) in the sequence; T i(x,y) represents the temperature value of a single pixel (x,y) after processing, T min and T max For {T i The lowest and highest temperatures are defined in the region (x,y), and pi(x,y) is the grayscale value of a single pixel at (x,y). The specific processing steps are as follows: Figure 2 As shown in the figure. This processing method can be used to enhance the image of the cracked area.

[0047] Step 4: Due to the limited number of simulated crack images, a CycleGAN (Recurrent Generative Adversarial Network) is used to augment the crack images. The processing method is as follows: Figure 3 As shown, the specific processing flow is as follows:

[0048] Step 4.1: Divide the single-frame image data obtained in Step 3 into infrared images with weld cracks and infrared images without weld cracks. Input all infrared images with weld cracks as the dataset X. A Infrared images of weld-free cracks are used as the input dataset X. B The remaining infrared images of the weldless cracks are used as dataset B.

[0049] Step 4.2, X A and X B The two inputs are used as the two inputs to train the CycleGAN network model. When the CycleGAN network model reaches training equilibrium, training is stopped until the CycleGAN network model is fully trained.

[0050] Step 4.3: The weld-free crack-with-weld-crack generator G obtained from the CycleGAN network model... BA As a model used in the application phase, the infrared image dataset B of weld-free cracks from step 4.1 is used as G. BA The input is used to obtain an infrared image dataset A containing weld cracks. At the same time, images with high crack generation quality are selected, and images with poor crack generation effect are removed from the image dataset A.

[0051] Step 4.4: Combine the image dataset A obtained in Step 4.3 with X from Step 4.1. A Combined together, they form an expanded dataset of infrared images of weld cracks, which serves as the dataset for the segmentation network.

[0052] Step 5: Use the expanded infrared image dataset with weld cracks obtained in Step 4.4 as input to train the YOLO v11-seg image segmentation network, and obtain the final crack length information through image post-processing methods. The specific process is as follows:

[0053] Step 5.1: Use the open-source annotation tool labelme to annotate the expanded infrared image dataset with weld cracks. The annotation includes fitting closed polygon contours of the cracks and drawing rectangular boxes around the cracks, generating an annotated JSON file.

[0054] Step 5.2: Feed the labeled JSON file and the infrared image with weld crack together into the YOLO v11-seg image segmentation network for training (one weld crack image corresponds to one labeled file. When training the image segmentation network, the weld crack image and its corresponding labeled file need to be fed into the image segmentation network for training). When the network's AP50 (AP represents the area under the precision and recall curves, and 50 represents the cross-union ratio between the YOLO v11-seg predicted mask and the real mask) is greater than 0.8, end the training; otherwise, adjust the network parameters and retrain until the network's AP50 is greater than 0.8 and end the training.

[0055] Step 5.3: The YOLO v11-seg image segmentation network outputs the rectangular detection box of the crack and the mask information of the crack. For the binary mask in each detection box, the skeleton is extracted using the skeletonize() function of OpenCV to obtain the crack information with a single pixel width.

[0056] Step 5.4: Cracks are usually random spline curves. Therefore, for each crack detection box, count the number of non-zero pixel values ​​after mask skeletonization within the detection box. This number is the skeleton length (pixel length) l of the crack.

[0057] Step 5.5: Using the following formula, convert the crack skeleton length l into the actual crack length L.

[0058]

[0059] In the formula, f is the focal length of the infrared thermal imager, and D is the length of the laser ranging in step 2.

[0060] In the actual detection of weld cracks in sodium-cooled fast reactor pipelines, the external insulation layer of the sodium-containing pipelines needs to be removed first. Then, the pipeline weld data is collected using an infrared thermal imager with a long focal length in step 2. A single-frame image is generated by the single-frame image data acquisition method in step 3 and used as input. The final detection result is obtained by sequentially going through steps 5.3-5.5 in step 5 (segmentation and recognition using an image segmentation network and image post-processing).

[0061] The present invention has been described in detail above with reference to the accompanying drawings and embodiments. However, the present invention is not limited to the above embodiments, and various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention. All contents not described in detail in the present invention can be derived from existing technologies.

Claims

1. A non-destructive infrared detection method for weld cracks in sodium-contaminated pipelines in sodium-cooled fast reactors, characterized in that: The method includes: Step 1: Produce weld-cracked pipes of different sizes, connect the pipes to a high-temperature circulation pipeline, and conduct high-temperature simulation experiments; Step 2: Using the pipe with weld cracks as the inspection target, an infrared thermal imager equipped with a telephoto lens is used to take pictures of the inspection target from a distance to obtain infrared image sequence data; the laser rangefinder on the infrared thermal imager is used to record the distance between the thermal imager and the inspection target to obtain the length of the laser rangefinder. Step 3: Collect infrared image sequence data of different lengths, depths, and temperatures, and preprocess the infrared image sequence data to obtain single-frame image data; Step 4: Use a recurrent adversarial generative network to augment the weld crack image data to obtain an augmented weld crack image dataset. Step 5: Use an image segmentation network to segment and identify the expanded dataset of images with weld cracks, obtain the mask information of the cracks, and then use image post-processing methods to obtain the true length of the cracks.

2. The method as described in claim 1, characterized in that: In step 1, the temperature of the high-temperature circulation pipeline is set below 300°C to simulate the temperature of a real sodium-related pipeline after it is shut down, and the pipeline is simulated to have no shielding or insulation layer on the outside.

3. The method as described in claim 1, characterized in that: In step 1, the high-temperature simulation experiment method is as follows: after heating the medium in the pipeline to 280℃~320℃, it is naturally cooled. During the cooling process, the medium temperature should not be lower than 100℃.

4. The method as described in claim 1, characterized in that: In step 2, the infrared image sequence data includes infrared image sequence data of weld cracks and infrared image sequence data of cracks without welds.

5. The method as described in claim 1, characterized in that: In step 3, the infrared image sequence data is preprocessed using the following formula to obtain single-frame image data: T i (x,y)={T i (x,y,t) max -T i (x,y,t) min } 2 In the formula, T i (x,y,t) max and T i (x,y,t) min These represent the highest and lowest temperatures of pixel (x, y) in the sequence; T i (x,y) represents the temperature value of a single pixel (x,y) after processing, T min and T max For {T i The lowest and highest temperatures in (x,y)}, and pi(x,y) is the gray value of a single pixel (x,y).

6. The method as described in claim 1, characterized in that, Step 4 specifically involves: Step 4.1: Divide the single-frame image data obtained in Step 3 into two groups of data and save them: images with weld cracks and images without weld cracks. Step 4.2: Use the two sets of data as the two inputs to train the recurrent adversarial generative network to obtain the trained recurrent adversarial generative network model. Step 4.3: Use the image without weld cracks as input to the trained recurrent adversarial generative network model to generate images with weld cracks. Select the images with high crack generation quality and combine them with the real weld crack images collected in Step 3 to form the expanded dataset of images with weld cracks.

7. The method as described in claim 1, characterized in that, Step 5 specifically involves: Step 5.1: Use an open-source annotation tool to annotate all the weld crack image data in the expanded weld crack image dataset to obtain the annotation file; Step 5.2: Feed the labeled file and the image with weld cracks into the image segmentation network for training. When the AP50 of the image segmentation network is greater than 0.8, end the training; otherwise, adjust the network parameters and retrain. Step 5.3: The image segmentation network outputs crack detection boxes and segmented mask information. Using the image skeletonization method, the mask information is converted into crack information with a single pixel width. Step 5.4: Calculate the number of non-zero pixel values ​​after mask skeletonization within the crack detection box to obtain the crack skeleton length; Step 5.5: Calculate the actual length of the crack based on the length of the crack skeleton.

8. The method as described in claim 1, characterized in that, In step 5.5, the formula for calculating the true length of the crack is: In the formula, L is the actual length of the crack, l is the skeleton length of the crack, f is the focal length of the infrared thermal imager, and D is the length of the laser ranging in step 2.

Images