A method and device for detecting thermal cross-penetration and expansion tubes based on AI image recognition technology

Abstract

The present invention discloses a method and device for detecting heat exchange tubes based on AI image recognition technology, which belongs to the field of production quality inspection of heat exchange devices. An image acquisition device is provided at one end of the heat exchanger tube plate, and a first light source with adjustable brightness, color and position is provided at the other end, and a second light source is added to assist in detecting leaking expansion tubes. The collected image data is input into the AI ​​heat exchange tube detection model. The model calculates the category similarity coefficient based on the data of each frame of the streaming media image. For heat exchange tubes whose coefficient exceeds the set threshold and whose abnormal category detection frequency exceeds the set category attribution coefficient threshold, the display device marks and reminds them to re-inspect. Six types of material training models are provided, and the first light source uses a surface light source to improve detection efficiency. The detection device includes a light source device, a camera device, a display device and a control center. The light source device and the camera device can be moved by a plane moving device for accurate detection. This technology improves the accuracy and efficiency of heat exchange tube detection.

Description

Technical Field

[0001] The present invention relates to the field of production quality inspection of heat exchange devices, and in particular to a method and device for inspecting heat exchange tubes based on AI image recognition technology. Background Art

[0002] The basic structure of a heat exchanger consists of a matrix of heat transfer tubes (also known as heat exchange tubes) within a shell. These tubes are typically made of high-thermal-conductivity metals such as copper, stainless steel, and titanium. The channels within the tubes are called the tube side, while the channel between the tubes and the shell is called the shell side. The tube side and shell side are isolated from each other. The heat exchanger operates by allowing the hot and cold media to flow through the tube side and shell side, respectively, to exchange heat through the tubes. Hereinafter, heat exchangers will be referred to as "heat exchangers."

[0003] In the heat exchanger, the matrix-arranged heat transfer tubes do not pass through the shell one by one. Instead, a tube sheet is provided inside the shell. The tube sheet has preset mounting holes that are compatible with the heat transfer tubes. The end of each heat transfer tube is inserted into the mounting hole, and then the end of the heat transfer tube is expanded through the tube expansion process to achieve a sealed connection with the tube sheet.

[0004] During the heat exchange assembly process, after inserting and expanding the heat transfer tubes, all tube holes on the tube sheet must be inspected to prevent problems such as missing insertions, missing expansions, and insufficient expansion. Due to the large number and density of tubes on the tube sheet, traditional manual inspections can easily lead to visual fatigue and high workload, making it very easy to miss inspections. Even with counters and helium inspections, it is still difficult to avoid missing inspections. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the present invention provides a method and device for detecting thermal cross-penetration and expansion tubes based on AI image recognition technology.

[0006] The specific technical solution of the present invention to solve the above technical problems is:

[0007] A heat exchange tube expansion and penetration detection method based on AI image recognition technology includes an image acquisition device disposed at one end of a heat exchanger tube sheet to be inspected, and a first light source disposed at the other end of the heat exchanger. The first light source is used to illuminate the interior of the heat exchange tube. The image acquisition device captures image data of one end of the tube sheet to be inspected and inputs the image data into a trained AI heat exchange tube expansion and penetration detection model. The AI ​​heat exchange tube expansion and penetration detection model determines the expansion quality of each heat exchange tube in the image data. Heat exchange tubes with potentially abnormal expansion quality are prominently marked on a display device to remind staff to re-inspect.

[0008] Compared to existing technologies, the present invention has the following beneficial effects: The AI ​​heat exchange tube detection model determines whether the similarity between each heat exchange tube in the image data and different categories exceeds a set similarity threshold. If so, it is labeled as the corresponding category. For the same tube opening, for any category, if the frequency of detection exceeding the set similarity threshold exceeds the set category affiliation threshold, the tube opening is marked as belonging to the corresponding category and prominently annotated on the display device. This solution uses AI image automatic detection technology to train the AI ​​heat exchange tube detection model based on manually annotated large amounts of normal and abnormal tube materials. This enables the AI ​​heat exchange tube detection model to identify abnormal tubes, freeing up staff, assisting manual inspection, and reducing the requirements for operators. Furthermore, the AI's stable and high-speed characteristics ensure the quality of the inspection process and improve operation speed.

[0009] Furthermore, the AI ​​heat exchange tube detection model is based on streaming media image data, and performs comprehensive calculations on the heat exchange tubes at the same position, considering the data of each frame in the streaming media image, to determine the final category.

[0010] The beneficial effect of adopting the above further solution is that, as the camera equipment and the light source move, for the same heat exchange tube, multiple photos of different lighting angles can be obtained in the streaming media. Based on the comprehensive judgment of photos from multiple angles, the expansion of the tube can be judged more accurately, preventing errors in judgment when the light is too weak or too strong.

[0011] Furthermore, the brightness and color of the first light source are adjustable to form a sharp contrast effect with the color of the heat exchange tube itself.

[0012] The beneficial effect of adopting the above further solution is that the brightness and color of the first light source are adjusted in order to obtain an image with a sharp contrast effect at the observation end, thereby improving the success rate of AI large model recognition.

[0013] Furthermore, the color of the first light source is a complementary color to the color of the heat exchange tube itself.

[0014] The beneficial effect of adopting the above further solution is that complementary colors can achieve a good contrast effect.

[0015] Furthermore, the position of the image acquisition device or the position of the first light source is adjustable.

[0016] The beneficial effect of adopting the above further solution is that by adjusting the position of the image acquisition device or the position of the first light source, photos under different lighting angles can be obtained, so as to obtain image data that is more suitable for AI large model analysis and improve judgment accuracy.

[0017] Furthermore, a second light source is provided on the image acquisition device side, and the brightness of the second light source is lower than that of the first light source. The AI ​​heat exchange tube expansion detection model assists in detecting the leakage of the expansion tube based on whether there is a gap between the heat exchange tube and the tube sheet.

[0018] The beneficial effect of adopting the above further solution is that another characteristic of a leaky expansion tube is the presence of a wide gap between the heat exchange tube and the tube sheet (this gap is occupied and disappears after the tube is expanded). However, this characteristic is not as obvious as the characteristic inside the tube, so it can be used as an auxiliary judgment and secondary verification condition.

[0019] Furthermore, when providing training materials for the AI ​​thermal cross-penetration and expansion tube detection model, six types of materials are included:

[0020] The first category is material with a tube that has been penetrated but not expanded. This is done by manually determining whether there are arc segments or circular outlines in the inner cavity of the tube end, and whether there is a clear gap between the tube opening and the tube hole. If the tube end image has a similarity greater than X (X is the abnormal similarity threshold, a percentage typically set between [50% and 60%]) with the first category, it is marked as suspected non-expanded tube. This means that the probability of this tube hole having X indicates that the tube has been penetrated but not expanded.

[0021] The second category is leaked pipe material, which is manually judged to have no bright tube end image at the tube sheet end. If the tube end image has a similarity greater than X (the empirical value of X is between [50% and 60%]) with the second category material, it is marked as a suspected leaked pipe;

[0022] The third category is insufficiently expanded material, which is manually judged to have poor roundness of the tube mouth or a clear gap between the tube mouth and the tube hole. If the similarity between the tube end image and the third category material is greater than X (the empirical value of X is between [50% and 60%]), it is marked as suspected insufficiently expanded.

[0023] The fourth category is normal footage of pipes. The similarity between the pipe end image and the normal image is determined. If the similarity is greater than Y (Y is the normal similarity threshold, with an empirical value between [60% and 70%], which is higher than X), and there are no cases in the first, second, or third categories, the pipe end is marked as a normal pipe end. Otherwise, no label is made and the pipe end is treated as the sixth category of footage.

[0024] The fifth category is materials from other material pipes, which provide generalized learning training for the model, enabling the model to recognize the above situations for different material pipes;

[0025] The sixth category is the undetected situation where the pipe opening is detected but the status cannot be identified. Generally, the outer pipe opening of the first light source cannot be detected. The position of this type of pipe opening can be used as a position reference during automatic detection to plan the movement path of the light source and image acquisition device during the automatic detection process.

[0026] Furthermore, the first light source is a surface light source.

[0027] The beneficial effect of adopting the above further solution is that the light from the surface light source is soft, uniform and parallel, and the lighting area is large, so a large number of heat exchange tubes can be inspected at one time, thereby improving the efficiency of detection.

[0028] The present invention also relates to a device for implementing the aforementioned heat exchange tube penetration and expansion detection method based on AI image recognition technology, comprising a light source device, a camera device, a display device and a control center, wherein the light source device is used to provide illumination, the camera device is used to collect images, the heat exchange tube plate to be detected is arranged between the light source device and the camera device, and the images collected by the camera device are transmitted to the control center. The control center has a built-in trained AI heat exchange tube penetration and expansion detection model, which is used to analyze and determine whether there are any missing tubes, missing expanded tubes or insufficient expansion at the ends of the heat exchange tubes in the image, and mark the positions with higher probabilities of missing tubes, missing expanded tubes or insufficient expansion, and display them to the staff via the display device.

[0029] Furthermore, the light source device and the camera device are respectively arranged on independent plane moving devices, and the plane moving device includes a base, a vertical motion device fixed on the base, a horizontal motion device driven to rise and fall by the vertical motion device, and a bearing part driven to move horizontally by the horizontal motion device, and the bearing part is used to carry the light source device or the camera device; the vertical motion device and the horizontal motion device are both electrically connected to the control center, and the control center is also used to control the plane moving device to carry the light source or the camera device to move within the plane.

[0030] The present invention has the following beneficial effects: the control center can independently control the movement of the camera device, independently control the movement of the light source device, or simultaneously control the movement of both. If a heat exchange tube is not detected based on the AI ​​heat exchange tube penetration and expansion detection model, the control center can control the light source of the light source device and the camera of the camera device to reach the target position for special inspection based on the coordinates of the target heat exchange tube. If a missing tube is detected, the control center controls the camera and light source to stop operation, issues a warning message on the display device, and waits for staff to review and confirm before continuing. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 Flow chart of the assembly process of the heat exchanger part;

[0032] Figure 2 Schematic diagram of the thermal cross-penetration and expansion tube detection method based on AI image recognition technology of the present invention;

[0033] Figure 3This is a structural schematic diagram of the first movement control part and the second movement control part in the thermal cross-penetration and expansion tube detection device based on AI image recognition technology of the present invention.

[0034] Figure 4 Schematic diagram of the original photo and the annotated photo in the thermal cross-penetration and expansion tube detection method based on AI image recognition technology of the present invention;

[0035] Figure 5 Schematic diagram of the original photo and annotated photo when the heat exchange tube detection method based on AI image recognition technology of the present invention is used to detect copper heat exchange tubes;

[0036] Figure 6 This is a schematic diagram showing the situation where the pipe end is not fully expanded;

[0037] Figure 7 Schematic diagram of the situation of detecting leaks at the pipe end.

[0038] In the accompanying drawings, the names of the components represented by the reference numerals are listed as follows:

[0039] 100. First light source; 200. Image acquisition device; 300. Display device; 400. Heat exchanger; 500. Vertical motion servo motor; 501. Horizontal motion servo motor. DETAILED DESCRIPTION

[0040] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0041] like Figure 1 The figure shows the heat exchange tube assembly process. After the initial threading and expansion, AI image inspection equipment (embedded with a trained AI heat exchange tube threading and expansion detection model) automatically detects any missing threading, missing expansion, or inadequate expansion. The AI ​​heat exchange tube threading and expansion detection model then provides prominent warnings for any tube openings that may be abnormal. A manual review is then conducted to check the locations of potentially problematic tube holes. If all threading and expansion are normal, the assembly process proceeds to the next step. Otherwise, the problematic tube holes are addressed and the process is repeated. This process inspects all tube holes to ensure they are functioning properly, preventing issues such as missing threading, missing expansion, and inadequate expansion. The AI ​​heat exchange tube threading and expansion detection model automatically identifies potentially problematic tube hole locations. The operator then uses the location warnings to recheck and, if any problems remain, repair them.

[0042] like Figure 2FIG2 is a schematic diagram of a method for detecting heat exchange tubes based on AI image recognition technology. An image acquisition device 200 is provided at one end of a tube sheet to be inspected in a heat exchanger 400, and a first light source 100 is provided at the other end of the heat exchanger 400. The first light source 100 is used to illuminate the interior of the heat exchange tubes. The image acquisition device 200 captures image data of one end of the tube sheet to be inspected and inputs the image data into a trained AI heat exchange tube detection model.

[0043] When training the AI ​​thermal cross-section expansion tube detection model, the training materials provided to the AI ​​thermal cross-section expansion tube detection model include six types of materials:

[0044] The first category is material that has been threaded but not expanded. This involves checking whether the heat exchange tube ends have curved lines or circular outlines within their inner cavities, and whether there are visible gaps between the tube ends and the tube holes in the tube sheet. If the tube end image has a similarity of greater than 50% with material from the first category, it is marked as a suspected non-expanded tube.

[0045] The second type is the leaked tube material. When the tube is leaked, there is no bright tube end image at the tube plate end, such as Figure 7 As shown in FIG, if the similarity between the pipe end image and the second type of material is greater than 50%, it is marked as a suspected leaked pipe.

[0046] The third type is the material that is not fully expanded. When the expansion is not sufficient, the roundness of the tube mouth is poor or there is still an obvious gap between the tube mouth and the tube hole, such as Figure 6 As shown in FIG, if the similarity between the tube end image and the third type of material is greater than 50%, it is marked as NG, which means that the tube expansion is suspected to be insufficient.

[0047] The fourth category is normal footage with pipes inserted through it. The similarity between the pipe end image and the normal image is determined. If the similarity is greater than 60% and there are no cases in the first, second, or third categories, the image is marked as OK; otherwise, it is marked as NG.

[0048] The fifth category is materials from other material tubes, which provide generalized learning training for the model, enabling the model to recognize the above situations for different material tubes.

[0049] The sixth category is the situation where the presence of a pipe opening is detected but the state cannot be identified. This is mostly because the position of the first light source 100 causes the image acquisition device 200 to be unable to detect the pipe opening of the first light source 100. The position of such pipe openings can be used to plan the movement path of the first light source 100 and the image acquisition device 200 during the fully automatic detection process to ensure that all pipe openings are detected.

[0050] The aforementioned expansion marks and gaps inside the tube are features that can be observed relatively intuitively by the human eye and are the result of judgments compiled when manually annotating training materials. However, during the AI ​​model training process, the features that determine whether the material is OK or NG are automatically extracted by the AI ​​model and are not limited to the features described above. More features are automatically extracted by the AI ​​model.

[0051] The first light source 100 includes a light source portion and a first movement control portion. The first movement control portion can control the first light source 100 to move freely in a vertical plane. The first light source 100 can be configured with different brightness and colors depending on the type of heat exchange tube. The image acquisition device 200 includes a camera and a second movement control portion. The second movement control portion can carry the camera and move freely in a vertical plane.

[0052] The first movement control part and the second movement control part are identical in hardware, such as Figure 3 As shown, both include a vertical motion servo motor 500 and a horizontal motion servo motor 501. The vertical motion servo motor 500 drives the horizontal motion servo motor 501 to move in the vertical direction, and the horizontal motion servo motor 501 drives the carried light source part or camera equipment to move in the horizontal direction. The two cooperate to realize the free movement of the light source part or camera equipment in the vertical plane. The first movement control part and the second movement control part are connected to each other for communication, so as to respectively carry the light source part and the camera equipment to realize synchronous matching movement of the position, ensuring that the camera equipment can collect good image data like a camera. The automatic detection path is determined by two parts of logic. The first part is a preset S-shaped return main path covering the entire tube plate surface. The second part is a secondary path for supplementing the local position fluctuation near the main path based on the position of the local uninspected pipe opening detected according to the sixth type of material. The two path algorithms work together to ensure that every pipe opening is detected.

[0053] The AI ​​heat exchange tube detection model provides the similarity between each heat exchange tube in the image data and various situations, namely, the category similarity. The category similarity takes a value of 0-1, where 0 represents that the situation does not belong to this category at all, and 1 represents that the situation belongs to this category 100%. It is determined whether the category similarity exceeds the set similarity threshold. If it does, the corresponding pipe opening is marked with the corresponding category. If the frequency of the same pipe opening being detected and classified as a certain category exceeds the set category attribution threshold, a prominent category identifier is added to the corresponding pipe opening position via the display device 300. The abnormal similarity threshold for abnormal conditions is set between X = [50% and 60%]. In this example, it is set to 50%. This means that for a single photo, if the AI ​​HTPE detection model determines that a tube with a similarity of more than 50% indicates leaking, leaking, or insufficiently expanded, it will be identified as abnormal. The normal similarity threshold for normal features is set between Y = [60% and 70%]. In this example, it is set to 60%. This means that for a single photo, if the AI ​​HTPE detection model determines that a tube with a similarity of more than 60% indicates normal, it will be identified as normal. During actual detection, the AI ​​HTPE detection model recognizes multiple features simultaneously, with the abnormal classification taking precedence over the normal classification. That is, the detection priorities are the same, but the abnormal classification takes precedence when determining the final classification. For example, for the same photo, if any abnormal feature has a similarity of more than 50% while the normal feature has a similarity of more than 60%, the image will be identified as abnormal.

[0054] The AI ​​heat exchange tube expansion detection model uses streaming media image data, capturing 30 or more images per second for inspection. For heat exchange tubes at the same location, comprehensive calculations are performed based on the data from each frame of the streaming media image. As the camera and light source move, multiple images of the same heat exchange tube can be captured from the streaming media at different lighting angles. Based on a comprehensive assessment of these images, if n = {4, 5, 6} of the same tube opening in 100 consecutive images are identified as belonging to a specific abnormality category (n is a configurable parameter), the system will pause inspection and mark the corresponding tube opening to prompt manual re-inspection. In this example, n is 4. If, in 100 consecutive images, two are judged as insufficiently expanded, three are judged as not fully expanded, and four are judged as leaking, the system will classify this tube opening as "suspected leaking," suspend inspection, and mark the opening to remind personnel to review it. If fewer than three out of 100 images are identified as abnormal, it is assumed to be an occasional misidentification.

[0055] The AI ​​thermal cross-penetration and expansion tube detection model's detection accuracy for the category to which the tube orifice belongs, i.e., the detection accuracy of the classification attribution, is derived from the accumulation of two probabilities:

[0056] The first part is the probability of automatic recognition by the AI ​​model, that is, the probability that a single nozzle is in a certain situation. This is determined by the similarity threshold. For example, in this example, the similarity threshold for determining whether a nozzle in the image has a certain abnormal feature is X, and the similarity threshold for determining whether it is a normal feature is Y. This controls the accuracy of the control of a certain category of situations.

[0057] The second part is the artificial experience probability, which controls the sensitivity of stopping detection and issuing an alarm. It is determined by the category attribution threshold. That is, if a certain nozzle is judged as a certain abnormal category n times in 100 images, the detection will be suspended and an alarm will be issued. In summary, the final abnormal category attribution coefficient is ,in ,in x The AI ​​thermal cross-penetration and expansion tube detection model determines the similarity (percentage) between the tube opening and the specific abnormal category. X For the corresponding similarity threshold, the final alarm coefficient for a certain situation is p=n / 100; taking the leakage expansion situation as an example, set X =60%, n=5, then p=0.05, that is, when g(x)>0.05, the logic judgment is that there is leakage or expansion at the pipe mouth; in actual detection, if leakage or expansion is detected n=6 times, that is, the probability of leakage or expansion is 0.06, it is considered to be leakage or expansion, and the detection is suspended. The abnormal pipe mouth is significantly prompted. Since the recognition of abnormal features and normal features is carried out simultaneously, the system will use the situation with the first 6 abnormalities detected as the abnormal category to which the pipe mouth finally belongs.

[0058] Under the aforementioned settings, the threshold Y for judging whether a nozzle is normal must be greater than X. Y is the corresponding normal similarity threshold. For example, Y = 70%. That is, for a single photo, a nozzle is judged to be normal only if the similarity with a normal nozzle exceeds 70%.

[0059] like Figure 4 The following is a comparison chart with and without annotations. Figure 4 On the right side, the heat exchange pipe framed in red may have abnormalities, and the pipe opening framed in green is normal.

[0060] like Figure 5 As shown in FIG, a schematic diagram of the detection of a copper heat exchange pipe is shown. In this figure, due to the problem of color contrast, the recognition accuracy of the original model decreases, so the color of the first light source 100 is preferably selected as the complementary color of the heat exchange pipe itself. The complementary color can obtain better contrast image data at the pipe mouth, and then increase the collection of some samples of such situations to enhance the generalization ability of the model.

[0061] A second light source is also provided on the image acquisition device 200. The brightness of the second light source is lower than that of the first light source 100. Natural light is generally sufficient. When the natural light is too strong, a certain degree of masking is required to ensure that the image acquisition device 200 can capture image data with clear features. The provision of the second light source enables the AI ​​heat exchange tube detection model to assist in detecting leaking expansion tubes based on whether there is a gap between the heat exchange tube and the tube sheet. In addition, for black hole tube openings that cannot be illuminated by the first light source 100, the second light source can also enable the AI ​​heat exchange tube detection model to detect unrecognized tube openings, assisting in planning the movement path during automatic detection.

[0062] The control principle of the entire device is as follows: the camera equipment continuously captures images of the tube sheet end and sends them to the AI ​​thermal cross-expansion tube detection model. After identification and processing by the AI ​​thermal cross-expansion tube detection model, if a pipe end with a potential problem is found, the vertical motion device and the horizontal motion device will be synchronously stopped, and the location of the possible problematic pipe end will be marked and displayed on the display device 300, prompting the operator to perform further inspection and repair work. After manual confirmation, movement and detection can continue.

[0063] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for detecting thermal cross-extension tubes based on AI image recognition technology, characterized in that: An image acquisition device is installed at one end of the heat exchanger's tube sheet to be inspected, and a first light source is installed at the other end of the heat exchanger to illuminate the interior of the heat exchange tubes. The image acquisition device captures image data of one end of the tube sheet to be inspected and inputs the image data into a trained AI heat exchange tube expansion detection model. The AI ​​heat exchange tube expansion detection model determines the expansion quality of each heat exchange tube in the image data. Heat exchange tubes with potentially abnormal expansion quality are prominently marked on a display device to remind staff to re-inspect. When providing training materials for the AI ​​thermal cross-entanglement and expansion detection model, at least the first type of materials should be included. The first type of materials are materials of pipes that have been penetrated but not expanded. The model determines whether there are arc segments or circular contours in the inner cavity of the pipe end, and whether there is a clear gap between the pipe opening and the pipe hole. If the similarity between the pipe end image and the first type of materials exceeds the abnormal similarity threshold X, the image is marked as NG.

2. The thermal cross-entry expansion tube detection method based on AI image recognition technology according to claim 1 is characterized in that: The AI ​​heat exchange tube detection model is based on streaming media image data. It performs comprehensive calculations on the heat exchange tubes at the same location, considering the data of each frame in the streaming media image, and determines the final category of each pipe opening.

3. The thermal cross-entry expansion tube detection method based on AI image recognition technology according to claim 2 is characterized in that: The brightness and color of the first light source are adjustable to form a sharp contrast effect with the color of the heat exchange tube itself.

4. The thermal cross-penetration and expansion tube detection method based on AI image recognition technology according to claim 3 is characterized in that: The color of the first light source is a complementary color to the color of the heat exchange tube.

5. The method for detecting thermal cross-entry expansion tubes based on AI image recognition technology according to any one of claims 1 to 4, characterized in that: The position of the image acquisition device and / or the position of the first light source are adjustable.

6. The method for detecting thermal cross-entry expansion tubes based on AI image recognition technology according to any one of claims 1 to 4, characterized in that: A second light source is further provided on the image acquisition device side. The brightness of the second light source is lower than that of the first light source. The AI ​​heat exchange tube expansion detection model assists in detecting leaking expansion tubes based on whether there is a gap between the heat exchange tube and the tube sheet.

7. The method for detecting thermal cross-entry expansion tubes based on AI image recognition technology according to any one of claims 1 to 4, characterized in that: When providing training materials for the AI ​​thermal cross-penetration and expansion tube detection model, the following five types of materials are also included: The second category is the leaked tube material. When the tube is leaked, there is no bright tube end image at the tube plate end. If the similarity between the tube end image and the second category material is greater than the abnormal similarity threshold X, it is marked as NG. The third category is materials with insufficient tube expansion. When the tube is insufficiently expanded, the roundness of the tube mouth is poor or there is still a clear gap between the tube mouth and the tube hole. If the similarity between the tube end image and the third category material is greater than the abnormal similarity threshold X, it is marked as NG. The fourth category is normal footage with pipes inserted through it. The similarity between the pipe end image and the normal image is determined. If the similarity is greater than the normal similarity threshold Y, and Y is higher than X, and there is no situation in the first, second, or third categories, then it is marked as OK. The fifth category is materials from other material pipes, which provide generalized learning training for the model, enabling the model to recognize the above situations for different material pipes; The sixth category is the undetected situation where the pipe opening is detected but the status cannot be identified. Most of the time, the outer pipe opening of the first light source cannot be detected. It is used to provide path guidance supplement during automatic detection.

8. The method for detecting thermal cross-entry expansion tubes based on AI image recognition technology according to any one of claims 1 to 3, characterized in that: The first light source is a surface light source.

9. A device for implementing the thermal cross-penetration and expansion tube detection method based on AI image recognition technology according to any one of claims 1 to 8, characterized in that: It includes a light source device, a camera device, a display device and a control center. The light source device is used to provide lighting, the camera device is used to collect images, and the heat exchange tube plate to be detected is arranged between the light source device and the camera device. The images collected by the camera device are transmitted to the control center. The control center has a built-in trained AI heat exchange tube penetration and expansion detection model, which is used to analyze and determine whether there are any missing tubes, missing expanded tubes or insufficient expansion at the ends of the heat exchange tubes in the image, and mark the positions with higher probabilities of missing tubes, missing expanded tubes or insufficient expansion, and display them to the staff through the display device.

10. The device according to claim 9, characterized in that The light source device and the camera device are respectively arranged on independent plane moving devices, and the plane moving device includes a base, a vertical motion device fixed on the base, a horizontal motion device driven to rise and fall by the vertical motion device, and a bearing part driven to move horizontally by the horizontal motion device, and the bearing part is used to carry the light source device or the camera device; the vertical motion device and the horizontal motion device are both electrically connected to the control center, and the control center is also used to control the plane moving device to carry the light source device or the camera device to move in the plane.

Images