Through-wall pipeline hole defect detection method and system based on convolutional neural network

Abstract

The invention relates to the field of nuclear power, in particular to a through-wall pipeline hole defect detection method and system based on a convolutional neural network. The method comprises the following steps: establishing a training sample set; constructing a convolutional neural network; training the convolutional neural network by using the training sample set to obtain a detection model with a function of identifying through-wall pipeline hole defects; and identifying a to-be-detected electromagnetic ultrasonic image by using the detection model, and outputting a detection result. The system comprises a training sample set acquisition module, a convolutional neural network construction module, a training module and a detection module. According to the invention, the hole defect of the through-wall pipeline can be preliminarily detected, and the method has the advantage of rapid detection.

Description

Technical Field

[0001] The present invention relates to the field of nuclear power, and in particular to a method and system for detecting through-wall pipe hole defects based on a convolutional neural network. Background Art

[0002] Through-the-wall pipes are widely used in power plants, chemical plants, water supply, gas supply and other fields. As the service time increases, the pipelines are subject to the combined effects of environmental factors and have the risk of corrosion, thinning and even perforation. Timely and effective monitoring and detection of pipeline corrosion conditions are important for ensuring their operational safety. However, due to the special location, complex structure and inaccessible detection equipment of through-the-wall pipes, the corrosion of the through-the-wall pipe sections cannot be effectively monitored and detected for a long time, which increases the risk of system operation.

[0003] Therefore, how to detect hole defects in wall-penetrating pipes has become an urgent problem to be solved. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for detecting hole defects of through-wall pipes based on convolutional neural networks, which can not only perform preliminary detection of hole defects of through-wall pipes, but also has the advantage of fast detection.

[0005] The present invention provides a method for detecting through-wall pipe hole defects based on a convolutional neural network, comprising the following steps:

[0006] Step 1: Establish a training sample set, wherein the training sample set includes an electromagnetic ultrasonic image set with hole defects and an electromagnetic ultrasonic image set without hole defects;

[0007] Step 2: Build a convolutional neural network;

[0008] Step 3: using the training sample set to train the convolutional neural network to obtain a detection model with the function of identifying through-wall pipe hole defects;

[0009] Step 4: Use the detection model to identify the electromagnetic ultrasonic image to be tested and output the detection result.

[0010] In a specific embodiment of the present invention, in the step 1,

[0011] Putting a first mark on all electromagnetic ultrasonic images contained in the electromagnetic ultrasonic image set with hole defects; the first mark includes the hole defect, the peak width, peak height and peak position of the electromagnetic ultrasonic image corresponding to the hole defect;

[0012] A second mark is put on all electromagnetic ultrasonic images included in the electromagnetic ultrasonic image set of non-porous defects, wherein the second mark includes non-porous defects.

[0013] In a specific embodiment of the present invention, the step three further includes:

[0014] The detection model is updated.

[0015] In a specific embodiment of the present invention, updating the detection model specifically includes:

[0016] Comparing the detection result with the result obtained by manually identifying the electromagnetic ultrasonic image to be tested;

[0017] If the comparison results are similar or identical, the electromagnetic ultrasonic image to be tested is used as a training sample to update the detection model.

[0018] In a specific embodiment of the present invention, if the electromagnetic ultrasonic image to be tested is an electromagnetic ultrasonic image with a hole defect;

[0019] Similar comparison results include: one or two of the peak width, peak height and peak position of the electromagnetic ultrasonic wave corresponding to the hole defect are the same;

[0020] The same comparison results include: the peak width, peak height and peak position of the electromagnetic ultrasonic wave corresponding to the hole defect are the same;

[0021] Before updating the detection model, the method further includes: adding the manually identified peak width, peak height and peak position, and the hole defect, as a mark, and marking the mark on the electromagnetic ultrasonic image to be tested;

[0022] If the electromagnetic ultrasonic image to be tested is an electromagnetic ultrasonic image without hole defects;

[0023] The comparison results are the same including: no hole defects;

[0024] Before updating the detection model, the method further includes: taking the non-porous defect as a mark and marking the mark on the electromagnetic ultrasonic image to be tested.

[0025] In a specific embodiment of the present invention, the step 1 further includes pre-processing the collected electromagnetic ultrasonic images to form a corresponding image set.

[0026] In a specific embodiment of the present invention, in the step 2, the constructed convolutional neural network includes 26 layers, including a data layer, a convolution layer, a pooling layer, a fully connected layer, and a softmax layer, wherein the data layer is used to provide input data, the convolution layer is used to train model parameters, the pooling layer is used for downsampling to avoid overfitting, and the fully connected layer and the softmax layer work together to output scores for judgment.

[0027] In a specific embodiment of the present invention, in step three, iterative training is performed by gradient descent method and back propagation algorithm to obtain a detection model with the function of identifying through-wall pipe hole defects;

[0028] The detection model is optimized using at least one of dense links, residual networks, and batch normalization.

[0029] The present invention also provides a convolutional neural network-based through-wall pipe hole defect detection device, comprising:

[0030] A training sample set collection module is used to collect training samples;

[0031] Convolutional neural network building blocks;

[0032] A training module, using the training sample set to train the convolutional neural network to obtain a detection model with the function of identifying through-wall pipe hole defects;

[0033] The detection module uses the detection model to identify the electromagnetic ultrasonic image to be tested obtained by the electromagnetic ultrasonic detection device and outputs the detection result.

[0034] In a specific embodiment of the present invention, the training sample set acquisition module includes an electromagnetic ultrasonic image set of defects with holes and an electromagnetic ultrasonic image set of defects without holes.

[0035] Compared with the prior art, the method and system for detecting hole defects in through-wall pipes based on convolutional neural networks of the present invention establish a convolutional neural network and use a training sample set for training to obtain a detection model, thereby realizing the recognition of the electromagnetic ultrasonic image to be tested, and finally confirming whether the through-wall pipe has a hole defect. The present application can not only perform preliminary detection of hole defects in through-wall pipes, but also has the advantage of fast detection. As a preliminary detection, the present application is very friendly to technicians who are still inexperienced and can give certain preliminary inspection results. At the same time, the preliminary detection results given by the present application and the results of the technicians' direct analysis of the electromagnetic ultrasonic images to be tested can be mutually verified, thereby improving the accuracy of the judgment of hole defects in through-wall pipes. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A flow chart of a method for detecting through-wall pipe hole defects based on a convolutional neural network provided in an embodiment of the present application;

[0037] Figure 2 Block diagram of a through-wall pipe hole defect detection system based on a convolutional neural network provided in an embodiment of the present application. DETAILED DESCRIPTION

[0038] In order to further understand the present invention, the embodiments of the present invention are described below in conjunction with examples, but it should be understood that these descriptions are only for further illustrating the features and advantages of the present invention, rather than limiting the present invention.

[0039] See also Figure 1 As shown, the embodiment of the present application provides a method for detecting through-wall pipe hole defects based on a convolutional neural network, which includes the following steps:

[0040] 101: Establish a training sample set, wherein the training sample set includes an electromagnetic ultrasonic image set of defects with holes and an electromagnetic ultrasonic image set of defects without holes, wherein all electromagnetic ultrasonic images included in the electromagnetic ultrasonic image set of defects with holes and the electromagnetic ultrasonic image set of defects without holes are obtained by using an electromagnetic ultrasonic detection device.

[0041] It can be understood that all the electromagnetic ultrasonic images obtained by the electromagnetic ultrasonic detection device are historical results of actual measurements of through-wall pipes using the electromagnetic ultrasonic detection device in actual factory buildings and other scenarios, and these results are consistent with the actual conditions of the corresponding through-wall pipes. The advantage of using such historical results is that it can ensure the accuracy of the subsequent trained detection model.

[0042] The detection result of the electromagnetic ultrasonic detection device is an electromagnetic ultrasonic wave waveform diagram, in which the horizontal axis represents the distance, and the position of the electromagnetic ultrasonic detection device in the wall-penetrating pipe is taken as the origin.

[0043] As an example, the electromagnetic ultrasonic detection device is an electromagnetic ultrasonic multifunctional guided wave detector. Of course, other devices can also be used to ensure that the device used for the sample set and the device used for the subsequent images to be tested are the same model of equipment.

[0044] After the electromagnetic ultrasonic image is exported from the electromagnetic ultrasonic detection device, it can be preprocessed, for example, the edge of the image can be cropped to obtain a cropped image that meets the size requirements for training the convolutional neural network; when cropping, the center of the image should be retained as much as possible, and the text information on the edge should be removed to avoid edge effects.

[0045] 102: Build a convolutional neural network.

[0046] The convolutional neural network is constructed by the caffenet network model. The convolutional neural network constructed in this embodiment includes 26 layers, including a data layer, a convolution layer, a pooling layer, a fully connected layer, and a softmax layer, wherein the data layer is used to provide input data, the convolution layer is used to train model parameters, the pooling layer is used for downsampling to avoid overfitting, and the fully connected layer and the softmax layer work together to output scores for judgment.

[0047] 103: Using the training sample set to train the convolutional neural network to obtain a detection model capable of identifying through-wall pipe hole defects.

[0048] For example, through iterative training using the gradient descent method and the back-propagation algorithm, a detection model with the function of identifying defects in through-wall pipe holes is obtained.

[0049] The specific steps of the gradient descent method are: calculate the gradient of the loss function with respect to each weight, start from any point, move a distance in the opposite direction of the gradient at that point, and continue to move a distance in the opposite direction of the gradient at the new position, thus continuously updating the weights of the network.

[0050] The specific steps of the back-propagation algorithm are as follows: when the weights of each layer of the convolutional neural network are iteratively updated using the gradient descent method, the gradient is propagated forward from the last layer of the network in sequence according to the chain rule.

[0051] In order to optimize the recognition rate and recognition speed, this embodiment also uses at least one of dense links, residual networks (ResNets) and batch normalization (BatchNormalization) to optimize the detection model.

[0052] 104: Using the detection model, recognize the electromagnetic ultrasonic image to be detected obtained by the electromagnetic ultrasonic detection device, and output the detection result.

[0053] In step 104, in order to facilitate the review and use and to facilitate the comparison and confirmation by the technical personnel, the electromagnetic ultrasonic image to be tested is further integrated into the test results.

[0054] The detection method provided in the present application establishes a convolutional neural network and uses a training sample set for training to obtain a detection model, thereby realizing the recognition of the electromagnetic ultrasonic image to be tested and ultimately confirming whether there is a hole defect in the wall-penetrating pipe.

[0055] The present application can not only perform preliminary detection of hole defects of wall-penetrating pipes, but also has the advantage of rapid detection.

[0056] This application is a preliminary test and is very friendly to inexperienced technicians, and can provide certain preliminary test results.

[0057] At the same time, the preliminary test results given in this application can verify each other with the results of the technicians' direct analysis of the electromagnetic ultrasonic images to be tested, thereby improving the accuracy of the judgment of the defects of the through-wall pipe holes.

[0058] In order to be able to train the constructed convolutional neural network, in the present application, a first mark is marked on all electromagnetic ultrasonic images contained in the electromagnetic ultrasonic image set with hole defects; wherein the first mark includes the hole defect and the peak width, peak height and peak position of the electromagnetic ultrasonic image corresponding to the hole defect.

[0059] A second mark is placed on all electromagnetic ultrasonic images included in the non-porous defect electromagnetic ultrasonic image set; wherein the second mark includes a non-porous defect.

[0060] In this application, the electromagnetic ultrasonic image set with hole defects and the electromagnetic ultrasonic image set without hole defects are marked differently. Because the images are divided into only two categories, either with hole defects or without hole defects. Therefore, the electromagnetic ultrasonic image set without hole defects only needs to be marked with "without hole defects".

[0061] As for the electromagnetic ultrasonic image set with hole defects, in addition to the hole defects, there are also differences in the location and size of the hole defects.

[0062] Therefore, in the electromagnetic ultrasonic image collection of hole defects, the labels include not only "hole defect" but also "peak width and peak height of electromagnetic ultrasonic wave" which reflect the size of hole defect.

[0063] Furthermore, in order to enhance the self-learning capability of the detection model, the method provided in the present application also includes a step of updating the detection model.

[0064] Wherein, updating the detection model specifically includes:

[0065] The detection result is compared with the result obtained by manually identifying the electromagnetic ultrasonic image to be tested.

[0066] If the comparison results are similar or identical, the electromagnetic ultrasonic image to be tested is used as a training sample to update the detection model.

[0067] Since the electromagnetic ultrasonic images to be tested are very good samples themselves, local materials are used as training samples to update the detection model. By dynamically updating the detection model parameters, the self-learning ability of the detection model is enhanced.

[0068] It should be noted that if both the test result and the manual recognition result indicate that the electromagnetic ultrasonic image to be tested does have a hole defect, that is, the electromagnetic ultrasonic image to be tested is an electromagnetic ultrasonic image with a hole defect;

[0069] At this time, the similar comparison results include: one or two of the peak width, peak height and peak position of the electromagnetic ultrasonic wave corresponding to the hole defect are the same; the same comparison results include: the peak width, peak height and peak position of the electromagnetic ultrasonic wave corresponding to the hole defect are the same;

[0070] At this time, before updating the detection model, it also includes: adding the manually identified peak width, peak height and peak position, and the hole defect as a mark, and marking the mark on the electromagnetic ultrasonic image to be tested.

[0071] The manual recognition result is used as a mark here, which actually defaults to manual recognition being more accurate. It should be noted that the manual recognition here is performed by experienced technicians to ensure that the results are accurate and reliable.

[0072] If both the test result and the manual recognition result indicate that the electromagnetic ultrasonic image to be tested is indeed a non-porous defect, that is, the electromagnetic ultrasonic image to be tested is an electromagnetic ultrasonic image without a pore defect;

[0073] At this time, the comparison results are the same including: no hole defects;

[0074] At this time, before updating the detection model, the method further includes: taking the non-porous defect as a mark and marking the mark on the electromagnetic ultrasonic image to be tested.

[0075] See also Figure 2 As shown, the embodiment of the present application also provides a through-wall pipe hole defect detection system based on a convolutional neural network, including a training sample set acquisition module, a convolutional neural network construction module, a training module and a detection module, wherein:

[0076] A training sample set acquisition module, used for acquiring training samples; the training sample set includes an electromagnetic ultrasonic image set of defects with holes and an electromagnetic ultrasonic image set of defects without holes, wherein all electromagnetic ultrasonic images contained in the electromagnetic ultrasonic image set of defects with holes and the electromagnetic ultrasonic image set of defects without holes are obtained by using an electromagnetic ultrasonic detection device;

[0077] Convolutional neural network building blocks;

[0078] A training module, using the training sample set to train the convolutional neural network to obtain a detection model with the function of identifying through-wall pipe hole defects;

[0079] The detection module uses the detection model to identify the electromagnetic ultrasonic image to be tested obtained by the electromagnetic ultrasonic detection device and outputs the detection result.

[0080] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0081] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0082] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0083] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0084] The above embodiments are only used to help understand the method and core idea of ​​the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

[0085] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting through-wall pipe hole defects based on convolutional neural network, characterized in that: The following steps are involved: Step 1: Establish a training sample set, wherein the training sample set includes an electromagnetic ultrasonic image set with hole defects and an electromagnetic ultrasonic image set without hole defects; Step 2: Build a convolutional neural network; Step 3: using the training sample set to train the convolutional neural network to obtain a detection model with the function of identifying through-wall pipe hole defects; Step 4: Use the detection model to identify the electromagnetic ultrasonic image to be tested and output the detection result.

2. The method for detecting defects of through-wall pipe holes based on convolutional neural networks according to claim 1 is characterized in that: In the step 1, Putting a first mark on all electromagnetic ultrasonic images contained in the electromagnetic ultrasonic image set with hole defects; the first mark includes the hole defect, the peak width, peak height and peak position of the electromagnetic ultrasonic image corresponding to the hole defect; A second mark is put on all electromagnetic ultrasonic images included in the electromagnetic ultrasonic image set of non-porous defects, wherein the second mark includes non-porous defects.

3. The method for detecting defects of through-wall pipe holes based on convolutional neural network according to claim 2 is characterized in that: The step three also includes: The detection model is updated.

4. The method for detecting defects of through-wall pipe holes based on convolutional neural networks according to claim 3 is characterized in that: Updating the detection model specifically includes: Comparing the detection result with the result obtained by manually identifying the electromagnetic ultrasonic image to be tested; If the comparison results are similar or identical, the electromagnetic ultrasonic image to be tested is used as a training sample to update the detection model.

5. The method for detecting defects of through-wall pipe holes based on convolutional neural network according to claim 4 is characterized in that: If the electromagnetic ultrasonic image to be tested is an electromagnetic ultrasonic image with a hole defect; Similar comparison results include: one or two of the peak width, peak height and peak position of the electromagnetic ultrasonic wave corresponding to the hole defect are the same; The same comparison results include: the peak width, peak height and peak position of the electromagnetic ultrasonic wave corresponding to the hole defect are the same; Before updating the detection model, the method further includes: adding the manually identified peak width, peak height and peak position, and the hole defect, as a mark, and marking the mark on the electromagnetic ultrasonic image to be tested; If the electromagnetic ultrasonic image to be tested is an electromagnetic ultrasonic image without hole defects; The comparison results are the same including: no hole defects; Before updating the detection model, the method further includes: taking the non-porous defect as a mark and marking the mark on the electromagnetic ultrasonic image to be tested.

6. The method for detecting wall-penetrating pipe hole defects based on convolutional neural network according to claim 1, characterized in that: The step 1 also includes preprocessing the collected electromagnetic ultrasonic images to form a corresponding image set.

7. The method for detecting wall-penetrating pipe hole defects based on convolutional neural network according to claim 1 is characterized in that: In the step 2, the constructed convolutional neural network includes 26 layers, including a data layer, a convolution layer, a pooling layer, a fully connected layer, and a softmax layer, wherein the data layer is used to provide input data, the convolution layer is used to train model parameters, the pooling layer is used for downsampling to avoid overfitting, and the fully connected layer and the softmax layer work together to output scores for judgment.

8. The method for detecting defects of through-wall pipe holes based on convolutional neural networks according to claim 1 is characterized in that: In the step 3, iterative training is performed by gradient descent method and back propagation algorithm to obtain a detection model with the function of identifying through-wall pipe hole defects; The detection model is optimized using at least one of dense links, residual networks, and batch normalization.

9. A device for detecting defects in through-wall pipe holes based on convolutional neural networks, characterized in that: include: A training sample set collection module is used to collect training samples; Convolutional neural network building blocks; A training module, using the training sample set to train the convolutional neural network to obtain a detection model with the function of identifying through-wall pipe hole defects; The detection module uses the detection model to identify the electromagnetic ultrasonic image to be tested obtained by the electromagnetic ultrasonic detection device and outputs the detection result.

10. The device for detecting defects of through-wall pipe holes based on convolutional neural network according to claim 9, characterized in that: The training sample set acquisition module includes an electromagnetic ultrasonic image set of defects with holes and an electromagnetic ultrasonic image set of defects without holes.

Images