Automatic fracture identification method for drop weight test based on artificial intelligence technology

Abstract

The invention discloses a drop weight test fracture automatic identification method based on an artificial intelligence technology, and the method comprises the steps: carrying out data collection, employing a shooting system, preparing 1000 pictures as a training set, and enabling the pictures to have as many data types as possible, thereby guaranteeing the identification accuracy of a model; preprocessing the training set, performing labeling processing on the sample set by using labeling software, and labeling target areas of a fragile area, a toughness area and a semi-fragile area in the image; calculating the area of a shearing region, and training and establishing a model; the invention relates to the technical field of steel detection, in particular to a drop weight test fracture automatic identification method based on the artificial intelligence technology, the artificial intelligence fracture identification technology and means are introduced, and a quadrature method is used. The shape of the brittle region can be accurately recognized, the area of the brittle region can be automatically calculated, recognition and calculation are completed within 3-5 seconds, the defect that time is consumed in a quadrature method is overcome, and a comparison method amp is adopted; and the detection result of the measurement method is not accurate enough.

Description

Technical Field

[0001] The invention relates to the technical field of steel material detection, and in particular to an automatic identification method for a drop weight test fracture based on artificial intelligence technology. Background Art

[0002] 9.2.4 Calculation of Shear Area Percentage in GBT8363 "Drop-Weight Tear Test Method for Steel" recommends the use of three calculation methods, namely, the quadrature method, the comparison method, and the measurement method, for the treatment of brittle fracture zones on the fracture separation surface of controlled-rolled pipeline steel and the determination of abnormal fractures. It is clearly stated that the quadrature method shall be used for the final determination in case of disputes or arbitration.

[0003] Defects: Appendix A of SY / T6476 "Drop-weight tear test method for pipeline steel pipes" has clearly stated that the three methods have certain deficiencies in the measurement process and measurement results.

[0004] Quadrature method

[0005] The method of measuring the shear area using a planimeter has relatively accurate measurement results, but the detection process is very time-consuming due to reasons such as the measuring equipment and calculation method.

[0006] Comparative Method

[0007] Using the comparison method is fast but not accurate enough.

[0008] Measurement method

[0009] The calculation formula of the approximate method used is only applicable to the fracture situation of 45%-100% content or dissociative fracture extending to one wall thickness on the hammer side of the test test.

[0010] Based on the above deficiencies, this case is proposed to solve the above problems. Summary of the invention

[0011] In view of the deficiencies in the prior art, the present invention provides an automatic identification method for a drop weight test fracture based on artificial intelligence technology, which solves the problems of the prior art that the measurement is not accurate enough or the application is limited.

[0012] To achieve the above objectives, the present invention is implemented by the following technical solutions: a method for automatically identifying a fracture surface in a drop weight test based on artificial intelligence technology, comprising the following steps:

[0013] Carry out data collection, use the shooting system, and prepare 1,000 pictures as a training set. These pictures have as many data types as possible to ensure the recognition accuracy of the model;

[0014] Preprocessing of the training set: using annotation tools to annotate the sample set and annotate the target areas of the image, such as brittle area, tough area, and semi-brittle area;

[0015] Calculate the area of ​​the clipping region, train and establish the model;

[0016] Model verification: input the prepared test set into the model for verification. When the accuracy meets the requirement, the model is established. If the accuracy does not meet the requirement, re-preprocess the data or increase the number of training sets, and repeat the above steps for re-training.

[0017] For relatively clear, well-defined areas, the test accuracy can reach 90-97%.

[0018] Preferably, model training and establishment include the following steps:

[0019] According to the task objectives, a deep learning segmentation model is selected as the model used for training;

[0020] The backbone network used in the model is VGG, which consists of 5 convolutional layers, 3 fully connected layers, and 1 softmax output layer. Maxpool is used to separate the layers, and the activation units of all hidden layers use ReLU functions.

[0021] The size of the model input image is 512*512;

[0022] The training set prepared above is used for data training through the model. When training, the model first performs forward calculation to obtain the loss, then performs reverse calculation based on the loss, and calculates the gradient of the parameters layer by layer based on the chain rule. Finally, the parameter value can be updated based on the gradient of the parameter and the learning rate.

[0023] After multiple rounds of iterations, all parameter values ​​in the model were finally determined, and ultimately the most ideal model data was obtained.

[0024] Preferably, in the drop weight tear test, the fracture zone on the fracture surface of the thrust fracture that is shiny and crystalline is the brittle zone, the fracture zone on the fracture surface of the thrust fracture that is dark gray and fibrous is the tough zone, and the area on the fracture surface of the thrust fracture where the brittle zone and the tough zone are twinned or alternately generated is the semi-brittle zone.

[0025] Preferably, the shear area percentage is measured as SA, and the value is expressed in %, calculated according to the formula:

[0026]

[0027] Where:

[0028] CD is the area of ​​the brittle zone in the fracture, in square millimeters (mm2);

[0029] H is the area of ​​the semi-brittle zone in the fracture, in square millimeters (mm2);

[0030] L is the area of ​​secondary cracks in the fracture, in square millimeters (mm2).

[0031] Preferably, the shooting system consists of a light source, a lens, and a CCD camera;

[0032] The overall brightness of the light source is 200,000-400,000 LUX, and multiple sets of oblique low-angle oblique lighting methods are used to ensure that the mirror reflection light in the brittle area and the diffuse reflection light in the toughness area in the picture are clearly presented, while the shadow area in the picture cannot affect regional recognition.

[0033] Beneficial Effects

[0034] The present invention provides a method for automatically identifying the fracture of a drop hammer test based on artificial intelligence technology. It has the following beneficial effects: the method for automatically identifying the fracture of a drop hammer test based on artificial intelligence technology uses the quadrature method by introducing artificial intelligence fracture recognition technology and means. It can accurately identify the morphology of the brittle area and automatically calculate the area of ​​the brittle area. The recognition and calculation time are completed within 3 to 5 seconds, which solves the disadvantages of the quadrature method being "time-consuming" and the disadvantages of the comparison method and measurement method being inaccurate in the detection results.

[0035] This method is based on the U-Net deep learning model. It requires the collection of a certain amount of fracture morphology photos. Specialized image annotation software is used to annotate the brittle areas in the image at the pixel level. The fracture images are then trained using the model. The trained model can be used for automatic identification of brittle areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of the detection results of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] Through the personnel in this field, all the electrical components in this case are connected to their corresponding power supplies through wires, and a suitable controller should be selected according to the actual situation to meet the control requirements. The specific connection and control sequence should refer to the following working principle, and the electrical connection between the electrical components is completed in the order of working in sequence. The detailed connection means are well-known technologies in this field. The following mainly introduces the working principle and process, and does not explain the electrical control.

[0039] See also Figure 1The present invention provides a technical solution: a method for automatically identifying a drop weight test fracture surface based on artificial intelligence technology, comprising the following steps:

[0040] Carry out data collection, use the shooting system, and prepare 1,000 pictures as a training set. These pictures have as many data types as possible to ensure the recognition accuracy of the model;

[0041] Preprocessing of the training set: using annotation tools to annotate the sample set and annotate the target areas of the image, such as brittle area, tough area, and semi-brittle area;

[0042] Furthermore, the annotation tool is Labelme, an online annotation tool created by the MIT Computer Science and Artificial Intelligence Laboratory. It provides a dataset of digital images with annotations. Labelme supports six different annotation types: polygon, rectangle, circle, line, point, and line strip. One of the limitations is that the file can only be saved and exported in JSON format.

[0043] The main functions of LabelMe are as follows: point labeling, line labeling, rectangular box labeling, and polygonal box labeling.

[0044] Calculate the area of ​​toughness zone, train and establish the model;

[0045] Model verification: input the prepared test set into the model for verification. When the accuracy meets the requirement, the model is established. If the accuracy does not meet the requirement, re-preprocess the data or increase the number of training sets, and repeat the above steps for re-training.

[0046] For relatively clear, well-defined areas, the test accuracy can reach 90-97%.

[0047] Model training and building includes the following steps:

[0048] According to the task objectives, a deep learning segmentation model is selected as the model used for training;

[0049] The backbone network used in the model is VGG, which consists of 5 convolutional layers, 3 fully connected layers, and 1 softmax output layer. Maxpool is used to separate the layers, and the activation units of all hidden layers use ReLU functions.

[0050] The size of the model input image is 512*512;

[0051] The training set prepared above is used for data training through the model. When training, the model first performs forward calculation to obtain the loss, then performs reverse calculation based on the loss, and calculates the gradient of the parameters layer by layer based on the chain rule. Finally, the parameter value can be updated based on the gradient of the parameter and the learning rate.

[0052] After multiple rounds of iterations, all parameter values ​​in the model were finally determined, and ultimately the most ideal model data was obtained.

[0053] In the drop weight tear test, the fracture zone on the fracture surface that appears shiny and crystalline is the brittle zone, the fracture zone on the fracture surface that appears dark gray and fibrous is the ductile zone, and the area on the fracture surface where the brittle zone and the ductile zone are twinned or alternately generated is the semi-brittle zone.

[0054] The shear area percentage is measured in SA meter, and the value is expressed in %, calculated according to the formula:

[0055]

[0056] Where:

[0057] CD is the area of ​​the brittle zone in the fracture, in square millimeters (mm2);

[0058] H is the area of ​​the semi-brittle zone in the fracture, in square millimeters (mm2);

[0059] L is the area of ​​secondary cracks in the fracture, in square millimeters (mm2).

[0060] The shooting system is composed of a light source, a lens, and a CCD camera;

[0061] The overall brightness of the light source is 200,000-400,000 LUX, and multiple sets of oblique low-angle oblique lighting methods are used to ensure that the mirror reflection light in the brittle area and the diffuse reflection light in the toughness area in the picture are clearly presented, while the shadow area in the picture cannot affect regional recognition.

[0062] 1) It solves the problem that the traditional planimetry method is accurate but time-consuming. The traditional planimetry method is to use a planimeter to manually circle the edge of each area on a fracture photo or optical projection image with a ruler, measure the area of ​​each brittle fracture area one by one, and then calculate. AI can automatically calculate the result within 3-5 seconds after taking a photo.

[0063] 2) Fully automated testing operations, high efficiency and savings

[0064] With the help of the automatic operation coordination of the robot arm / hand and the automatic fracture recognition and calculation technology of artificial intelligence, it can be realized at any time, inspection at any time, and data sending at any time. It avoids the waiting problems of sample delivery time, freezing time, hammer drop time, inspection time, and data collection time in the process of division of labor and cooperation, greatly shortens the inspection cycle, and improves production efficiency.

[0065] 3) No one is involved in the whole operation, which is light and safe

[0066] With the help of the automatic operation coordination of the robotic arm / hand, the entire operation can be done without human intervention, freeing the operator's hands, avoiding operational hazards such as scratches and frostbite caused by low temperatures, and improving harsh working environments such as heavy objects, noise, and dust.

[0067] 4) Test results can be stored for a long time

[0068] With traditional testing methods, samples will rust and become unrecognizable after being left for a few hours due to condensation of water mist after the hammer falls. The AI ​​testing method not only retains the test results, but also supports the retention, retrieval, and multi-person re-inspection and analysis of sample result images, laying the foundation for digital archiving of results and in-depth data analysis.

[0069] 5) The test results are stable and reliable

[0070] With the help of artificial intelligence fracture recognition and calculation technology, the human errors caused by different shifts and different inspectors are avoided, and the repeatability and reproducibility of the testing process are greatly improved, providing an effective grip and assistance for production quality control.

[0071] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions. The sentence "includes an element defined by ... does not exclude the existence of other identical elements in the process, method, article or device including the element".

[0072] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An automatic identification method for drop weight test fracture based on artificial intelligence technology, It is characterized in that The following steps are involved: Collect data using the camera system and prepare 1,000 images as a training set. These images should contain as many data types as possible to ensure the recognition accuracy of the model. Preprocessing of the training set: using annotation tools to annotate the sample set and annotate the target areas of the image, such as the brittle area, semi-brittle area, and tough area; Calculate the area of ​​the clipping region, train and establish the model; Model verification: input the prepared test set into the model for verification. When the accuracy meets the requirement, the model is established. If the accuracy does not meet the requirement, re-preprocess the data or increase the number of training sets, and repeat the above steps for re-training. For relatively clear, well-defined areas, the test accuracy can reach 90-97%.

2. According to claim 1, a method for automatically identifying fracture surfaces in a drop weight test based on artificial intelligence technology, Features ,Model training and building include the following steps: According to the task objectives, a deep learning segmentation model is selected as the model used for training; The backbone network used in the model is VGG, which consists of 5 convolutional layers, 3 fully connected layers, and 1 softmax output layer. Maxpool is used to separate the layers, and the activation units of all hidden layers use ReLU functions. The size of the model input image is 512*512; The training set prepared above is used for data training through the model. When training, the model first performs forward calculation to obtain the loss, then performs reverse calculation based on the loss, and calculates the gradient of the parameters layer by layer based on the chain rule. Finally, the parameter value can be updated based on the gradient of the parameter and the learning rate. This process is repeated multiple times to determine all parameter values ​​in the model and ultimately obtain the most ideal model data.

3. According to the method of automatic identification of drop weight test fracture surface based on artificial intelligence technology in claim 1, Features In the drop-weight tearing test, the fracture zone on the fracture surface that is shiny and crystalline is the brittle zone, the fracture zone on the fracture surface that is dark gray and fibrous is the ductile zone, and the area on the fracture surface where the brittle zone and the ductile zone are twinned or alternately generated is the semi-brittle zone.

4. According to claim 1, a method for automatically identifying fracture surfaces in a drop weight test based on artificial intelligence technology, Features , the shear area percentage is measured by SA meter, the value is expressed in %, and is calculated according to the formula: Where: CD is the area of ​​the brittle zone in the fracture, in square millimeters (mm2); H is the area of ​​the semi-brittle zone in the fracture, in square millimeters (mm2); L is the area of ​​secondary cracks in the fracture, in square millimeters (mm2).

5. The method for automatic identification of fracture surfaces in a drop weight test based on artificial intelligence technology according to claim 1, Features ,The shooting system is composed of a light source, a lens and a CCD camera; The overall brightness of the light source is 200,000-400,000 LUX, and multiple sets of oblique low-angle oblique lighting methods are used to ensure that the mirror reflection light in the brittle area and the diffuse reflection light in the toughness area in the picture are clearly presented, while the shadow area in the picture cannot affect regional recognition.

6. The method for automatically identifying fracture surfaces in a drop weight test based on artificial intelligence technology according to claim 1, Features ,The input of the previously prepared test set into the model for ,validation includes the following steps: Convert the model from Pytorch to ONNX format; Package the model, integrate one or more models together by pre- and post-processing the model, and then add descriptive files to achieve complete verification functions; In addition to the efficient implementation of some common pre- and post-processing and the pre- and post-processing logic during alignment training, the model also needs to have sufficient scalability to cope with different scenarios and facilitate the expansion of new functions by the business line.

Images