A method and system for identifying defects in weld seams on pressure steel pipe surface
By improving the YOLOv5 algorithm model and wall climbing robot system, weld defects on the surface of pressure steel pipes of hydropower stations are identified, and the problems of low efficiency and low accuracy of traditional manual detection are solved, efficient and accurate weld defect detection is achieved, and the safety of hydropower stations is improved. sex.
Patent Information
- Application Number
- CN202410637575.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-22
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-05-22
AI Technical Summary
Traditional manual detection methods are difficult to efficiently and accurately identify the weld defects on the surface of pressure steel pipes of hydropower stations, resulting in low detection efficiency and long construction period, and poses major safety hazards.
The improved YOLOv5 algorithm model is used to combine the wall-climbing robot, and the weld image of the steel pipe surface is obtained through the wall-climbing robot and used the pre-constructed YOLOv5 algorithm model to identify it, and the defect category and location are output, overcoming the cost and accuracy imbalance of manual detection.
It improves the efficiency and accuracy of the detection of weld defects on the surface of steel pipes, reduces the inspection cost, enhances safety, avoids the cumbersome construction of scaffolding, and provides a reliable safety maintenance solution.
Smart Images

Figure CN118396974B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a method and a system for identifying defects on a weld seam on a pressure steel pipe surface, belonging to the technical field of surface defect detection. Background Art
[0002] In order to improve the utilization rate of water energy in hydropower stations, the capacity of hydropower units in hydropower stations has been continuously increased. However, as the main component of the generator set, the pressure-bearing area of the penstock is much larger than that of the turbine unit, so it is more susceptible to long-term erosion of water flow and welding defects. The fragility of the penstock weld means that once a burst occurs, it will pose a huge threat to the safe operation of the hydropower station and cause serious losses of life and property of the country and the people. Therefore, it is particularly important to regularly inspect and maintain the inner surface of the penstock and promptly repair weld defects and other surface quality problems.
[0003] Traditional manual inspection methods have a series of defects: Since the penstock of a hydropower station is round and has a large vertical height, it is difficult and costly to build scaffolding; manual visual inspection is subjective and uncertain, resulting in low efficiency and long construction period for weld defect detection. These problems seriously hinder the timely discovery and treatment of weld defects, resulting in major safety hazards for penstocks. Summary of the invention
[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide a method and system for identifying defects in welds on the surface of a pressure steel pipe, which can effectively improve the efficiency and detection accuracy of identifying defects in welds on the surface of the steel pipe.
[0005] To achieve the above object, the present invention is implemented by adopting the following technical solutions:
[0006] In a first aspect, the present invention provides a method for identifying defects on a surface weld of a pressure steel pipe, comprising:
[0007] Acquire the image of the weld on the surface of the steel pipe to be identified;
[0008] Input the image to be identified into a pre-built second improved YOLOv5 algorithm model to obtain the weld type and weld position on the steel pipe surface;
[0009] The construction process of the second improved YOLOv5 algorithm model includes:
[0010] Obtain training images of welds on the surface of steel pipes;
[0011] Constructing a first improved YOLOv5 algorithm model, training the first improved YOLOv5 algorithm model through the training image, and obtaining model parameters of the first improved YOLOv5 algorithm model;
[0012] A second improved YOLOv5 algorithm model is constructed, and model parameters of the first improved YOLOv5 algorithm model are loaded into the second improved YOLOv5 algorithm model to obtain a pre-constructed second improved YOLOv5 algorithm model.
[0013] Furthermore, the first improved YOLOv5 algorithm model and the second improved YOLOv5 algorithm model both include a trunk, a neck and a head;
[0014] The backbone is used to extract features from weld image data on the surface of a steel pipe, and includes a basic unit, an InceptionNeXt unit, and a fast spatial pyramid pooling unit;
[0015] The neck is used to perform feature fusion on the extracted features, and includes a simplified convolution unit, a re-parameterized visual geometry group structure unit and an upsampling unit;
[0016] The head includes three detection heads, which are used to output defect categories and defect locations.
[0017] Furthermore, the InceptionNeXt unit includes four branches, wherein the first branch includes a deep convolution layer with a convolution kernel size of 3×3, the second branch includes a deep convolution layer with a convolution kernel size of 1×11, the third branch includes a deep convolution layer with a convolution kernel size of 11×1, and the fourth branch is an identity mapping. The four branches are all connected to the multi-layer perception unit, so that the output results of the four branches are spliced and normalized and then input into the multi-layer perception unit, and the input features are added to the output features of the multi-layer perception unit as the output.
[0018] Furthermore, the reparameterized visual geometry group structure unit includes a plurality of residual structures connected in sequence, and the residual structure in the first improved YOLOv5 algorithm model includes three branches, wherein the first branch includes a convolution layer with a convolution kernel size of 3×3 and a batch normalization layer, the second branch includes a convolution layer with a convolution kernel size of 1×1 and a batch normalization layer, and the third branch includes a batch normalization layer, which is valid only when the input channel is equal to the output channel, and the output results of the three branches are output after being activated by the ReLU function;
[0019] The residual structure in the second improved YOLOv5 algorithm model includes a convolution layer with a convolution kernel size of 3×3.
[0020] Furthermore, the first improved pre-training method of the YOLOv5 algorithm model includes:
[0021] Acquire a steel pipe surface weld image dataset, wherein the image dataset includes defect-free images and images of different defect categories;
[0022] Label the image data set with defect category names and defect locations to obtain a labeled image data set;
[0023] Divide the labeled image dataset into training set and test set;
[0024] The training set data is used as input, and the identified defect type and defect location are used as output to train the first improved YOLOv5 algorithm model;
[0025] Using the test set data as input, the first improved YOLOv5 algorithm model is tested to obtain test results, and it is determined whether the test results meet the convergence conditions. If the convergence conditions are met, the pre-trained first improved YOLOv5 algorithm model is obtained. If the convergence conditions are not met, the training steps are repeated until the test results meet the convergence conditions to obtain the pre-trained first improved YOLOv5 algorithm model.
[0026] Furthermore, the defect categories include cracks, slag inclusions, incomplete penetration, and pores.
[0027] In a second aspect, the present invention provides a pressure steel pipe surface weld defect recognition system, the system is used to implement any of the pressure steel pipe surface weld defect recognition methods described above, and comprises a wall climbing robot and a steel pipe surface weld defect positioning and recognition module connected to each other by electrical signals;
[0028] The wall-climbing robot is used to obtain the image to be identified of the weld on the surface of the steel pipe, and transmit the image to be identified of the weld on the surface of the steel pipe to the steel pipe weld defect positioning and identification module;
[0029] The steel pipe surface weld defect positioning and identification module is used to process the image to be identified based on the pre-built second improved YOLOv5 algorithm model to obtain the weld category and weld position on the steel pipe surface;
[0030] It also includes a model parameter acquisition module, which is electrically connected to the steel pipe surface weld defect positioning and identification module. The model parameter acquisition module is used to train the first improved YOLOv5 algorithm model, obtain the model parameters of the first improved YOLOv5 algorithm model, and load the model parameters of the first improved YOLOv5 algorithm model into the second improved YOLOv5 algorithm model.
[0031] Optionally, it also includes a data transmission module, which is connected to the steel pipe surface weld defect positioning and identification module and the mobile terminal electrical signal, and is used to transmit the defect category and defect location to the mobile terminal;
[0032] The mobile terminal is also connected to the wall-climbing robot by electrical signals, and can control the movements of the wall-climbing robot.
[0033] Optionally, the wall-climbing robot can perform wall-climbing movement along the surface of the steel pipe to capture images of the welds on the surface of the steel pipe, and obtain images to be identified of the welds on the surface of the steel pipe. The wall-climbing robot includes a camera mechanism, and the camera mechanism is electrically connected to a module for locating and identifying weld defects on the surface of the steel pipe. The camera mechanism can photograph the welds on the surface of the steel pipe to obtain images of the welds on the surface of the steel pipe, and transmit the images of the welds on the surface of the steel pipe to the module for locating and identifying weld defects on the surface of the steel pipe.
[0034] Optionally, the wall-climbing robot includes a base plate, a plurality of track wheels are provided on both sides of the base plate, and a CCD camera, a controller, a permanent magnetic suction cup, a battery pack and a switch button are provided on the lower surface of the base plate, the permanent magnetic suction cup is used to attach the wall-climbing robot to the surface of the steel pipe, a top plate is provided above the base plate, a panoramic camera and a lighting mechanism are provided on the top plate, and the controller is electrically connected to the track wheels, CCD camera, panoramic camera, mobile terminal and steel pipe weld defect positioning and identification module.
[0035] Compared with the prior art, the present invention has the following beneficial effects:
[0036] The present invention combines a wall-climbing robot with a steel pipe surface weld defect positioning and recognition module. The wall-climbing robot takes a steel pipe surface weld image, and the steel pipe surface weld defect positioning and recognition module recognizes the steel pipe surface weld image and outputs the defect category and defect position, thereby overcoming the limitations of manual detection cost and uneven detection accuracy of steel pipe surface welds, and can effectively improve the detection efficiency and detection accuracy of steel pipe surface weld defects.
[0037] The wall-climbing robot provided by the present invention has the ability to walk flexibly on complex and vertical structures, avoiding the tediousness of scaffolding construction; the improved YOLOv5 algorithm model ensures the accuracy and real-time performance of weld micro-defect detection, greatly improves the efficiency and safety of weld defect detection, and provides a reliable solution for the safe operation and maintenance of hydropower stations;
[0038] The steel pipe surface weld defect positioning and identification module of the present invention adopts the second improved YOLOv5 algorithm model, and the first improved YOLOv5 algorithm model is used during pre-training. The difference between the first improved YOLOv5 algorithm model and the second improved YOLOv5 algorithm model is only the branch structure of the re-parameterized visual geometry group structure unit. Each 3×3 convolution layer in the re-parameterized visual geometry group structure unit of the first improved YOLOv5 algorithm model is added with an identity mapping branch and a residual branch with a convolution kernel of 1×1, which can effectively prevent gradient disappearance and network degradation, improve the training accuracy of the model, and accelerate the convergence of the network model. The re-parameterized visual geometry group structure unit of the second improved YOLOv5 algorithm model combines the identity mapping branch and the residual branch with a convolution kernel of 1×1 into a convolution with a convolution kernel of 3×3 through structural re-parameterization, which not only reduces the amount of reasoning calculations and improves the reasoning speed of the model, but also reduces the consumption of memory resources, making the deployment of the model more convenient. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 A schematic diagram of a flow chart of a method for identifying weld defects on a pressure steel pipe surface in an embodiment of the present invention;
[0040] Figure 2 A schematic diagram of the structure of an improved YOLOv5 algorithm model in an embodiment of the present invention;
[0041] Figure 3 A schematic diagram of the structure of an InceptionNeXt unit in an embodiment of the present invention;
[0042] Figure 4 1 is a schematic diagram of the structure of a re-parameterized visual geometry group structure unit in an embodiment of the present invention, wherein a is a schematic diagram of the structure of a re-parameterized visual geometry group structure unit in a first improved YOLOv5 algorithm model, and b is a schematic diagram of the structure of a re-parameterized visual geometry group structure unit in a second improved YOLOv5 algorithm model;
[0043] Figure 5 A schematic cross-sectional view of a moving end of a wall-climbing robot in an embodiment of the present invention;
[0044] Figure 6 A schematic diagram of the structure of a wall-climbing robot in one embodiment of the present invention;
[0045] Figure 7 A schematic diagram of a top view of a wall-climbing robot in an embodiment of the present invention;
[0046] In the figure: 1 bottom plate, 2 top plate, 3 track wheels, 4 permanent magnetic suction cup, 5 CCD camera, 6 panoramic camera, 7 lighting device, 8 controller, 9 battery pack, 10 switch button. DETAILED DESCRIPTION
[0047] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.
[0048] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In the description of the present invention, unless otherwise specified, "plurality" means two or more.
[0049] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood by specific circumstances. Example
[0050] The pressure steel pipe surface weld defect recognition system provided by the embodiment of the present invention includes a wall-climbing robot, a steel pipe surface weld defect positioning and recognition module, and a data transmission module which are connected in sequence by electrical signals.
[0051] Combination Figure 6 and Figure 7 In this embodiment, the wall-climbing robot adopts a wheel-type robot, and the specific structure includes a base plate 1, and a plurality of track wheels 3 are arranged on both sides of the base plate 1. The wall-climbing robot can be driven to move by powering the track wheels 3. A CCD camera 5, a controller 8, a permanent magnetic suction cup 4, a battery pack 9 and a switch button 10 are arranged on the lower surface of the base plate 1. The number of CCD cameras 5 is 2, which are respectively arranged on the head and tail sides of the base plate 1. The main optical axis of the lens of the CCD camera 5 is perpendicular to the surface of the steel pipe. The number of permanent magnetic suction cups 4 is set to 3 groups, which do not contact the surface of the pressure steel pipe and do not generate friction, but can provide magnetic force to attach the wall-climbing robot to the surface of the pressure steel pipe. The battery pack 9 provides power for the CCD camera 5, the panoramic camera 6, the track wheels 3, etc.
[0052] A top plate 2 is provided above the bottom plate 1, and a panoramic camera 6 and a lighting device 7 are provided on the top plate 2. The panoramic camera 6 can not only photograph the weld, but also realize a 360° all-round real-time view of the surrounding conditions of the wall-climbing robot to avoid obstacles. The lighting device 7 is used in conjunction with the panoramic camera to capture the weld state more clearly.
[0053] The controller 8 is connected to the track wheels 3, CCD camera 5, panoramic camera 6, and mobile terminal via electrical signals. In some embodiments, the mobile terminal controls the track wheels via the controller 8, thereby controlling the moving direction of the wall-climbing robot. The mobile terminal can also control the movements of the panoramic camera 6 and CCD camera 5 via the controller 8.
[0054] The wall-climbing robot performs wall-climbing motion along the surface of the steel pipe to collect images of the weld seams on the steel pipe surface, and transmits the image data of the weld seams on the steel pipe surface to the steel pipe weld defect positioning and recognition module through the controller. The steel pipe weld defect positioning and recognition module identifies the weld seam image on the steel pipe surface to obtain the defect category and defect location, and transmits the defect category and defect location to the data transmission module, which is connected to the mobile terminal electrical signal and is configured to transmit the defect category and defect location to the mobile terminal.
[0055] The steel pipe weld defect location and recognition module is based on the pre-trained second improved YOLOv5 algorithm model. In this embodiment, it also includes a model parameter acquisition module, which is electrically connected to the steel pipe weld defect location and recognition module, and is configured to acquire the model parameters of the pre-trained first improved YOLOv5 algorithm model and load the model parameters into the steel pipe weld defect location and recognition module.
[0056] The first improved YOLOv5 algorithm model and the second improved YOLOv5 algorithm model are both composed of a trunk, a neck and a head. The trunk is used to extract features from image data, which includes a basic unit, an InceptionNeXt unit and a fast spatial pyramid pooling (SPPF) unit; the neck is used to fuse the extracted features, which includes a simplified convolution unit, a reparameterized visual geometry group structure unit and an upsampling unit; the head includes three detection heads, which are used to output defect categories and defect locations.
[0057] In this embodiment, the CSP-n module of the backbone of the traditional YOLOv5 algorithm model is replaced by an InceptionNeXt unit, the CSP-n module of the neck is replaced by a reparameterized visual geometry group (RepVGG) structure unit, and the CBS module of the neck is replaced by a simplified convolution (SimConv) unit, that is, the SiLU activation function in the basic unit is changed to the ReLU activation function.
[0058] The InceptionNeXt unit includes four branches, of which the first branch includes a deep convolution layer with a convolution kernel size of 3×3, the second branch includes a deep convolution layer with a convolution kernel size of 1×11, the third branch includes a deep convolution layer with a convolution kernel size of 11×1, and the fourth branch is an identity mapping. All four branches are connected to the multi-layer perception unit, so that the output results of the four branches are spliced and normalized before being input into the multi-layer perception unit, and the input features are added to the output features of the multi-layer perception unit as the final output.
[0059] The processing steps of the InceptionNeXt unit are:
[0060] The input to the InceptionNeXt module X Decomposed into four branches along the channel dimension:
[0061]
[0062] in, X is the input vector, They represent the decomposition vectors with channel dimensions of 1~g, g~2g, 2g~3g, and 3g~C, respectively. g is the number of channels of the convolution branch, and g=r g C, r g is the ratio, C is the total number of channels, The corresponding decomposition vectors are The vector name of .
[0063] The grouped inputs are fed into four parallel branches, each performing a depthwise convolution operation:
[0064]
[0065] in, The kernel size is k i × k j , the input channel is g i , the output channel is g j The depth of convolution. k s The default is a square kernel with a kernel size of 3. k b The default is a banded kernel with a kernel size of 11. They are The output vector of the depthwise convolution operation.
[0066] Concatenate the output of each branch:
[0067]
[0068] Among them, Concat represents the vector concatenation operation. is the vector obtained after concatenation.
[0069] Concatenate vectors Perform normalization:
[0070]
[0071] Among them, Norm represents the normalization operation, yes The normalized vector of .
[0072] The features obtained The input is sent to the multi-layer perception unit, which consists of two fully connected layers with an activation function in between. The two fully connected layers can be implemented by 1×1 convolution and shortcut connection is also used. This process is expressed as:
[0073]
[0074] in, The kernel size is k × k , Input Channel C i and output channels C o The convolution of r is the expansion ratio, σ Represents the activation function.
[0075] The re-parameterized visual geometry group (RepVGG) structural unit is used to improve the neck of the YOLOv5 algorithm model. The re-parameterized visual geometry group structural unit consists of multiple branch structures connected in sequence. The basic idea is to introduce a special branch structure to assist training during network training. Specifically:
[0076] The branch structure in the first improved YOLOv5 algorithm model includes three branches, wherein the first branch includes a convolution layer with a convolution kernel size of 3×3 and a batch normalization layer, the second branch includes a convolution layer with a convolution kernel size of 1×1 and a batch normalization layer, and the third branch includes 1 batch normalization layer, which is only valid when the input channel is equal to the output channel, and the output results of the three branches are then passed through the ReLU activation function as the final output. The three-branch structure can effectively prevent gradient disappearance and network degradation, improve the training accuracy of the model, and accelerate the convergence of the network model.
[0077] The branch structure in the second improved YOLOv5 algorithm model includes a convolution layer with a convolution kernel size of 3 × 3. The use of structural reparameterization technology to merge branch structures not only reduces the amount of inference operations and improves the inference speed of the model, but also reduces the consumption of memory resources and makes it easier to deploy the model. Example
[0078] This embodiment also provides a method for identifying defects on the surface weld of a pressure steel pipe, which is implemented by the system for identifying defects on the surface weld of a pressure steel pipe described in Example 1, and includes the following steps:
[0079] The wall-climbing robot is controlled by a mobile terminal to move forward, and the weld on the surface of the steel pipe is photographed to obtain an image of the weld on the surface of the steel pipe.
[0080] Constructing a first improved YOLOv5 algorithm model and pre-training it to obtain a pre-trained first improved YOLOv5 algorithm model, specifically including:
[0081] a. Obtain a dataset of weld images on the surface of a steel pipe, the dataset including defect-free images and images of different defect categories, the defect categories including cracks, slag inclusions, incomplete penetration, pores, etc. In this embodiment, the image dataset comes from the JPEGWD weld public dataset;
[0082] b. Label the image data set with defect category names and defect locations to obtain a labeled image data set;
[0083] c. Divide the labeled image dataset into a training set and a test set;
[0084] d. Using the training set data as input and the identified defect type and defect location as output, the first improved YOLOv5 algorithm model is trained;
[0085] e. Use the test set data as input to test the first improved YOLOv5 algorithm model, obtain the test results, and determine whether the test results meet the convergence conditions. If the convergence conditions are met, the pre-trained first improved YOLOv5 algorithm model is obtained. If the convergence conditions are not met, repeat c~e until the test results meet the convergence conditions to obtain the pre-trained first improved YOLOv5 algorithm model.
[0086] The model parameters in the pre-trained first improved YOLOv5 algorithm model are obtained, and loaded into the second improved YOLOv5 algorithm model to obtain the pre-trained second improved YOLOv5 algorithm model.
[0087] Based on the pre-trained second improved YOLOv5 algorithm model, the weld image on the steel pipe surface is processed to obtain the weld category and weld position on the steel pipe surface.
[0088] 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 method for identifying defects in weld seams on a pressure steel pipe surface, characterized in that: include: Acquire the image of the weld on the surface of the steel pipe to be identified; Input the image to be identified into a pre-built second improved YOLOv5 algorithm model to obtain the weld type and weld position on the steel pipe surface; The pre-construction process of the second improved YOLOv5 algorithm model includes: Obtain training images of welds on the surface of steel pipes; Constructing a first improved YOLOv5 algorithm model, training the first improved YOLOv5 algorithm model through the training image, and obtaining model parameters of the first improved YOLOv5 algorithm model; Constructing a second improved YOLOv5 algorithm model, and loading the model parameters of the first improved YOLOv5 algorithm model into the second improved YOLOv5 algorithm model by reparameterization to obtain a pre-constructed second improved YOLOv5 algorithm model; The first improved YOLOv5 algorithm model and the second improved YOLOv5 algorithm model both include a trunk, a neck and a head; The backbone is used to extract features from weld image data on the surface of a steel pipe, and includes a CBS module, an InceptionNeXt unit, and a fast spatial pyramid pooling unit; The neck is used to perform feature fusion on the extracted features, and includes a simplified convolution unit, a re-parameterized visual geometry group structure unit and an upsampling unit; The head includes three detection heads, which are used to output defect categories and defect locations; The reparameterized visual geometry group structure unit includes a plurality of residual structures connected in sequence, and the residual structure in the first improved YOLOv5 algorithm model includes three branches, wherein the first branch includes a convolution layer with a convolution kernel size of 3×3 and a batch normalization layer, the second branch includes a convolution layer with a convolution kernel size of 1×1 and a batch normalization layer, and the third branch includes a batch normalization layer, which is only valid when the input channel is equal to the output channel, and the output results of the three branches are output after being activated by the ReLU function; The residual structure in the second improved YOLOv5 algorithm model includes a convolution layer with a convolution kernel size of 3×3; The construction method of the first improved YOLOv5 algorithm model and the second improved YOLOv5 algorithm model includes: The backbone CSP-n module of the traditional YOLOv5 algorithm model is replaced with the InceptionNeXt unit, the neck CSP-n module is replaced with the reparameterized visual geometry group structure unit, and the neck CBS module is replaced with the simplified convolution unit, that is, the SiLU activation function in the CBS module is changed to the ReLU activation function.
2. The method for identifying defects in weld seams on a pressure steel pipe according to claim 1, characterized in that: The InceptionNeXt unit includes four branches, wherein the first branch includes a deep convolution layer with a convolution kernel size of 3×3, the second branch includes a deep convolution layer with a convolution kernel size of 1×11, the third branch includes a deep convolution layer with a convolution kernel size of 11×1, and the fourth branch is an identity mapping. The four branches are all connected to the multi-layer perception unit, so that the output results of the four branches are spliced and normalized and then input into the multi-layer perception unit, and the input features are added to the output features of the multi-layer perception unit as the output.
3. The method for identifying defects in weld seams on a pressure steel pipe according to claim 1, characterized in that: The first improved pre-training method of the YOLOv5 algorithm model includes: Acquire a steel pipe surface weld image dataset, wherein the image dataset includes defect-free images and images of different defect categories; Label the image data set with defect category names and defect locations to obtain a labeled image data set; Divide the labeled image dataset into training set and test set; The training set data is used as input, and the identified defect type and defect location are used as output to train the first improved YOLOv5 algorithm model; Using the test set data as input, the first improved YOLOv5 algorithm model is tested to obtain test results, and it is determined whether the test results meet the convergence conditions. If the convergence conditions are met, the pre-trained first improved YOLOv5 algorithm model is obtained. If the convergence conditions are not met, the training steps are repeated until the test results meet the convergence conditions to obtain the pre-trained first improved YOLOv5 algorithm model.
4. The method for identifying defects in weld seams on a pressure steel pipe according to claim 1, characterized in that: The defect categories include cracks, slag inclusions, incomplete penetration, and pores.
5. A pressure steel pipe surface weld defect recognition system, characterized in that: The system is used to implement the pressure steel pipe surface weld defect identification method according to any one of claims 1 to 4, and comprises a wall-climbing robot and a steel pipe surface weld defect positioning and identification module connected to each other by electrical signals; The wall-climbing robot is used to obtain the image to be identified of the weld on the surface of the steel pipe, and transmit the image to be identified of the weld on the surface of the steel pipe to the weld defect positioning and identification module on the surface of the steel pipe; The steel pipe surface weld defect positioning and identification module is used to process the image to be identified based on the pre-built second improved YOLOv5 algorithm model to obtain the weld category and weld position on the steel pipe surface; It also includes a model parameter acquisition module, which is electrically connected to the steel pipe surface weld defect positioning and identification module. The model parameter acquisition module is used to train the first improved YOLOv5 algorithm model, obtain the model parameters of the first improved YOLOv5 algorithm model, and load the model parameters of the first improved YOLOv5 algorithm model into the second improved YOLOv5 algorithm model.
6. The pressure steel pipe surface weld defect recognition system according to claim 5 is characterized in that: It also includes a data transmission module, which is connected to the steel pipe surface weld defect positioning and identification module and the mobile terminal electrical signal, and is used to transmit the defect category and defect location to the mobile terminal; The mobile terminal is also connected to the wall-climbing robot by electrical signals, and the mobile terminal can control the movements of the wall-climbing robot.
7. The pressure steel pipe surface weld defect recognition system according to claim 5 is characterized in that: The wall-climbing robot can perform wall-climbing motion along the surface of the steel pipe to capture images of the welds on the surface of the steel pipe, and obtain images to be identified of the welds on the surface of the steel pipe. The wall-climbing robot includes a camera mechanism, which is electrically connected to a module for locating and identifying weld defects on the surface of the steel pipe by signals. The camera mechanism can photograph the welds on the surface of the steel pipe to obtain images of the welds on the surface of the steel pipe, and transmit the images of the welds on the surface of the steel pipe to the module for locating and identifying weld defects on the surface of the steel pipe.
8. The pressure steel pipe surface weld defect recognition system according to claim 7 is characterized in that: The wall-climbing robot includes a base plate, a plurality of track wheels are arranged on both sides of the base plate, and a CCD camera, a controller, a permanent magnetic suction cup, a battery pack and a switch button are arranged on the lower surface of the base plate, the permanent magnetic suction cup is used to attach the wall-climbing robot to the surface of the steel pipe, a top plate is arranged above the base plate, a panoramic camera and a lighting mechanism are arranged on the top plate, and the controller is connected to the track wheels, CCD camera, panoramic camera, mobile terminal and steel pipe weld defect positioning and identification module by electrical signals.
Citation Information
Cited By
Safety assessment method for initial defect of steel pipe welded joint
CN120539270A
