Defect Detection and Identification System and Method for Hot Melt Butt Joints of High-Density Polyethylene Pipes

Abstract

The present invention discloses a defect detection and identification system and method for a hot melt butt joint of a high-density polyethylene pipeline, including: a probe sends an ultrasonic detection signal to the hot melt butt joint of the high-density polyethylene pipe body and receives an ultrasonic reflection signal, and an ultrasonic phased array synthesizes and amplifies the ultrasonic signal to generate an A-scan signal; performs image reconstruction on the A-scan signal to generate a B-scan image; controls the probe to scan along the surface of the high-density polyethylene pipe body to generate a C-scan image; uses short-time Fourier transform to extract features from the A-scan information; uses the histogram of oriented gradients image processing algorithm to extract features from the B-scan and C-scan images; uses a convolutional neural network to perform deep feature extraction on the signal features and image features, and then a classifier performs classification and output. The present invention uses the histogram of oriented gradients and a convolutional neural network for image feature recognition, and uses a softmax classifier for classification, greatly improving the detection accuracy and precision.

Description

Technical Field

[0001] The present invention relates to the technical field of pipeline damage detection and identification, and particularly relates to a defect detection and identification system and method for the hot melt butt joint of a high-density polyethylene pipeline based on ultrasonic phased array. Background Art

[0002] Compared with metal pipelines, high-density polyethylene pipelines have the advantages of high strength, corrosion resistance, long service life, and convenient installation, and are widely used in various aspects such as urban gas supply, water supply, and agricultural irrigation. Hot melt butt joint is the most commonly used connection method in high-density polyethylene pipelines, which has low cost and strong applicability; however, its process parameters are numerous and the operation and construction are complex, and various types of defects are likely to occur during the butt joint process. Its external defects can be judged by the naked eye, and its internal defects need to rely on non-destructive testing technology. Ultrasonic technology has the advantages of fast speed, simplicity, and no need for safety protection, and is the preferred method for non-destructive testing of the hot melt butt joints of high-density polyethylene pipelines.

[0003] The detection of the hot melt butt joints of high-density polyethylene pipelines has main difficulties such as complex weld structure and defect location; currently in ultrasonic testing, it mainly relies on the detection personnel to visually identify the defect information in the image and judge the product grade based on work experience. The above method is too dependent on the experience of on-site personnel, and the subjective judgment of personnel is also easily affected by factors such as personal experience, on-site environment, thinking mode, and technical level. Summary of the Invention

[0004] In view of the above problems existing in the prior art, the present invention provides a defect detection and identification system and method for the hot melt butt joint of a high-density polyethylene pipeline based on ultrasonic phased array, which uses the Histogram of Oriented Gradients (HOG) and Convolutional Neural Network (CNN) for image feature recognition, and uses a softmax classifier for classification, greatly improving the detection accuracy and precision.

[0005] The present invention discloses a defect detection and identification system for the hot melt butt joint of a high-density polyethylene pipeline, comprising: an ultrasonic phased array, a probe, and a scanning frame, wherein the probe is respectively connected to the ultrasonic phased array and the scanning frame;

[0006] The probe acts on the hot melt butt joint of the high-density polyethylene pipe body and is used for:

[0007] Sending an ultrasonic detection signal to the hot melt butt joint of the high-density polyethylene pipe body;

[0008] Receiving the ultrasonic reflection signal after passing through the high-density polyethylene pipe body;

[0009] The ultrasonic phased array is used for:

[0010] Add the time gain corresponding to the ultrasonic signal for signal synthesis and amplification to generate the A-scan signal;

[0011] Perform image reconstruction on the A-scan signal to generate the B-scan image;

[0012] Control the probe to scan along the surface of the high-density polyethylene pipe body through the scanning frame. After the A-scan signal and the B-scan image are transformed and calculated, a C-scan image is generated;

[0013] Use the short-time Fourier transform signal analysis method to extract features from the A-scan information;

[0014] Use the histogram of oriented gradients image processing algorithm to extract features from the B-scan and C-scan images;

[0015] Use a convolutional neural network to perform deep feature extraction on the signal features and image features, and then classify and output by a classifier.

[0016] As a further improvement of the present invention, the scanning frequency of the ultrasonic phased array is 2.5 - 10 MHz, and the probe is connected to the ultrasonic phased array through a 64-pin shielded wire.

[0017] As a further improvement of the present invention, a coupling agent is provided on the contact surface between the probe and the high-density polyethylene pipe body.

[0018] As a further improvement of the present invention, the short-time Fourier transform is defined as:

[0019]

[0020] In the formula, t is the time domain and ω is the frequency domain.

[0021] As a further improvement of the present invention, the gradient calculation formula is defined as:

[0022]

[0023] The magnitude of the gradient is:

[0024]

[0025] The direction of the gradient is:

[0026]

[0027] As a further improvement of the present invention, during the convolution process, the feature map of the previous layer is convolved by a learnable convolution kernel to obtain the output feature map; each output feature map contains the convolution of multiple input feature maps, and its formula is:

[0028]

[0029]

[0030] In the formula, represents the j-th feature map in the l-th layer, is called the activation function, is the convolution kernel matrix, is the bias of the feature map after convolution.

[0031] As a further improvement of the present invention, it is classified and output by a softmax classifier, which maps the input real number to the interval [0, 1] through the sigmoid function, as follows:

[0032]

[0033] Its loss function is as follows:

[0034]

[0035] The softmax function is as follows:

[0036]

[0037] As a further improvement of the present invention, it further includes:

[0038] Select the defective area map and the non-defective area from the classification output results for segmentation, label them, and store the labeled pictures in the training library;

[0039] According to the corresponding relationship between its characteristic information and the defect, to determine whether there is a defect inside the butt joint and the type, position and size of the defect.

[0040] The present invention also discloses a method for defect detection and recognition of a hot melt butt joint of a high-density polyethylene pipe based on the above system, including:

[0041] Sending an ultrasonic detection signal to the hot melt butt joint of the high-density polyethylene pipe body;

[0042] Receiving the ultrasonic reflection signal after passing through the high-density polyethylene pipe body;

[0043] Adding a time gain corresponding to the ultrasonic signal for signal synthesis and amplification to generate an A-scan signal;

[0044] Performing image reconstruction on the A-scan signal to generate a B-scan image;

[0045] Controlling the probe to scan along the surface of the high-density polyethylene pipe body, and after the A-scan signal and the B-scan image are transformed and calculated, generating a C-scan image;

[0046] Using the short-time Fourier transform signal analysis method to extract features from the A-scan information;

[0047] Use the histogram of oriented gradients (HOG) image processing algorithm to extract features from B-scan and C-scan images;

[0048] Use a convolutional neural network to perform deep feature extraction on the signal features and image features, and then classify and output by a classifier.

[0049] As a further improvement of the present invention, it further includes:

[0050] Select the defective area map and the non-defective area from the classification output results for segmentation, annotate them, and store the annotated pictures in the training library;

[0051] According to the corresponding relationship between its feature information and the defect, determine whether there are defects inside the butt joint, and the type, position and size of the defects.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] 1. Through the movement of the probe, the present invention can achieve full-range scanning and detection of the hot melt butt joint of high-density polyethylene pipes, and can record detection data and detection maps through ultrasonic phased arrays. The detection efficiency is high, the operation is simple, and the results are reliable;

[0054] 2. The present invention uses the short-time Fourier transform signal analysis method to extract features from the A-scan information, uses the histogram of oriented gradients (HOG) image processing algorithm to extract features from the B-scan image and C-scan image, uses a convolutional neural network to perform deep feature extraction on the signal features and image features, and then classify and output by a classifier; as the number of detections increases, the generalization ability of the model can be gradually increased, and the recognition accuracy can be improved;

[0055] 3. The present invention is applied to the defect detection of the hot melt butt joint of high-density polyethylene pipes, can effectively detect various defects of the hot melt butt joint of high-density polyethylene pipes, greatly reduces the requirements for professional technicians, avoids the interference of human factors, makes the defect recognition results more reliable and the evaluation results more objective, and can achieve accurate positioning, qualitative and quantitative of the defects of the hot melt butt joint. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic structural diagram of a defect detection and recognition system for the hot melt butt joint of high-density polyethylene pipes based on ultrasonic phased arrays disclosed in an embodiment of the present invention;

[0057] Figure 2 It is a schematic flow diagram of a defect detection and recognition method for the hot melt butt joint of high-density polyethylene pipes based on ultrasonic phased arrays disclosed in an embodiment of the present invention.

[0058] In the figure:

[0059] 1. Ultrasonic phased array; 2. 64-pin shielded cable; 3. Probe; 4. High-density polyethylene pipe body; 5. Hot melt butt joint; 6. Couplant; 7. Defect detection module; 8. Defect identification module. Specific embodiments

[0060] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0061] The present invention will be further described in detail below with reference to the accompanying drawings:

[0062] As Figure 1 、 2 shown, the present invention provides a defect detection and identification system for a hot melt butt joint of a high-density polyethylene pipeline, including: an ultrasonic phased array 1, a 64-pin shielded cable 2, a probe 3, a high-density polyethylene pipe body 4, a hot melt butt joint 5, a couplant 6 and a scanning frame (not shown in the figure). The ultrasonic phased array 1 includes a defect detection module 7 and a defect identification module 8; wherein,

[0063] The probe 3 is placed on the scanning frame and connected to the ultrasonic phased array 1 through a 64-pin shielded cable 2. The probe 3 acts on the hot melt butt joint 5 of the high-density polyethylene pipe body 4. Preferably, the probe 3 is a wedge-shaped probe. Preferably, a couplant 6 is provided on the contact surface between the probe 3 and the high-density polyethylene pipe body 4; the ultrasonic phased array 1 is started, and its scanning parameters are set. The scanning frequency is 2.5 - 10 MHz, and the specific frequency setting is selected according to the wall thickness.

[0064] As Figure 2 shown, based on the above defect detection and identification system, the defect detection and identification method for the hot melt butt joint of the high-density polyethylene pipeline of the present invention includes:

[0065] S1. The ultrasonic phased array 1 emits ultrasonic signals to the measured hot melt butt joint 5 of the high-density polyethylene pipe body 4 through the wedge-shaped probe 3;

[0066] S2. The ultrasonic signals pass through the couplant 6 and enter the inside of the high-density polyethylene pipe body 4. When the ultrasonic signals encounter defects, the change in acoustic impedance causes them to be reflected at the defects, and the reflected ultrasonic signals are received by the same probe 3;

[0067] S3. The ultrasonic phased array 1 adds corresponding time gains to the ultrasonic signals for signal synthesis and amplification to generate its A-scan signal;

[0068] S4. The A-scan signal is reconstructed inside the ultrasonic phased array 1 to generate its B-scan image;

[0069] S5. The scanning frame controls the probe to scan along the surface of the pipeline. After the A-scan signal and the B-scan image are transformed and calculated, its C-scan image is generated;

[0070] S6. Record the ultrasonic phased array detection data, and use the short-time Fourier transform signal analysis method to extract the features of the A-scan information; among them,

[0071] The short-time Fourier transform is defined as:

[0072]

[0073] In the formula, t is the time domain and ω is the frequency domain;

[0074] S7. Use the histogram of oriented gradients image processing algorithm to extract the features of the B-scan and C-scan images;

[0075] Specifically:

[0076] Record the ultrasonic phased array detection map, intercept the B-scan image and the C-scan image in it, and save them in RGB format; when there are edges in the image, the gray values of some adjacent pixels change greatly, that is, there must be relatively large gradient values, so the gradient of the image can be calculated to determine the edge of the image. The gradient calculation formula is defined as:

[0077]

[0078] The magnitude of the gradient is:

[0079]

[0080] The direction of the gradient is:

[0081]

[0082] According to the above gradient, calculate the gradient histogram of the picture. After normalization, calculate the feature vectors of the B-scan image and the C-scan image.

[0083] S8. Use the convolutional neural network to perform deep feature extraction on the signal features and image features, and then classify and output by the classifier;

[0084] Specifically:

[0085] The surface features of the image extracted by the histogram of gradients are imported into the convolutional neural network for deep feature extraction. During the convolution process, the feature map of the previous layer is convolved by a learnable convolutional kernel to obtain the output feature map. Each output feature map contains the convolution of multiple input feature maps, and its formula is as follows:

[0086]

[0087]

[0088] In the formula, represents the j-th feature map in the l-th layer, is called the activation function, is the convolutional kernel matrix, is the bias of the feature map after convolution;

[0089] Feature extraction is performed by the convolutional neural network, and then classification output is performed by the softmax classifier. It maps the input real number to the interval [0, 1] through the sigmoid function as follows:

[0090]

[0091] Its loss function is as follows:

[0092]

[0093] The softmax function is as follows:

[0094]

[0095] S9. Select the defective area map and the non-defective area from the classification output results for segmentation, and label them. Store the labeled pictures in the training library; as the number of detections increases, the generalization ability of the model can be gradually increased, and the recognition accuracy can be improved; according to the corresponding relationship between its feature information and the defect, determine whether there are defects inside the butt joint and the type, position and size of the defects.

[0096] Furthermore, the defect detection module 7 implements S3 to S5, and the defect recognition module 8 implements S6 to S9.

[0097] The advantages of the present invention are as follows:

[0098] 1. Through the movement of the probe, the present invention can achieve omnidirectional scanning and detection of the hot melt butt joint of the high-density polyethylene pipe, and can record the detection data and detection atlas through the ultrasonic phased array. The detection efficiency is high, the operation is simple, and the result is reliable;

[0099] 2. The present invention uses the short-time Fourier transform signal analysis method to extract features from the A-scan information, uses the histogram of oriented gradients image processing algorithm to extract features from the B-scan image and the C-scan image, uses a convolutional neural network to perform deep feature extraction on the signal features and image features, and then classifies and outputs them by a classifier; as the number of detections increases, the generalization ability of the model can be gradually increased, and the recognition accuracy can be improved.

[0100] 3. The present invention is applied to the defect detection of the hot melt butt joint of high-density polyethylene pipes, and can effectively detect various defects of the hot melt butt joint of high-density polyethylene pipes. The requirements for professional technicians are greatly reduced, the interference of human factors is avoided, the defect recognition result is more reliable and the evaluation result is more objective, and the accurate positioning, qualitative and quantitative analysis of the defects of the hot melt butt joint can be realized.

[0101] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A defect detection and identification system for hot melt butt joints of high-density polyethylene pipes, characterized in that, Comprising: An ultrasonic phased array, a probe, and a scanning frame, wherein the probe is respectively connected to the ultrasonic phased array and the scanning frame; The probe acts on the hot melt butt joint of the high-density polyethylene pipe body for: Sending an ultrasonic detection signal to the hot melt butt joint of the high-density polyethylene pipe body; Receiving the ultrasonic reflection signal after passing through the high-density polyethylene pipe body; The ultrasonic phased array is used for: Adding time gain corresponding to the ultrasonic signal for signal synthesis and amplification to generate an A-scan signal; Performing image reconstruction on the A-scan signal to generate a B-scan image; Controlling the probe to scan along the surface of the high-density polyethylene pipe body through the scanning frame. After the A-scan signal and the B-scan image are transformed and calculated, a C-scan image is generated; Using the short-time Fourier transform signal analysis method to extract features from the A-scan information; Using the histogram of oriented gradients image processing algorithm to extract features from the B-scan and C-scan images; Performing deep feature extraction on the signal features and image features using a convolutional neural network, and then classifying and outputting by a classifier.

2. The defect detection and recognition system according to claim 1, characterized in that, The scanning frequency of the ultrasonic phased array is 2.5 - 10 MHz, and the probe is connected to the ultrasonic phased array through a 64-pin shielded wire.

3. The defect detection and recognition system according to claim 1, characterized in that, A coupling agent is provided on the contact surface between the probe and the high-density polyethylene pipe body.

4. The defect detection and recognition system according to claim 1, characterized in that, The short-time Fourier transform is defined as: In the formula, t is the time domain and ω is the frequency domain.

5. The defect detection and recognition system according to claim 1, wherein, The gradient calculation formula is defined as: The amplitude of the gradient is: The direction of the gradient is:

6. The defect detection and recognition system according to claim 1, characterized in that, During the convolution process, the feature map of the previous layer is convolved by a learnable convolution kernel to obtain the output feature map; each output feature map contains the convolution of multiple input feature maps, and its formula is: In the formula, represents the j-th feature map in the l-th layer, is called the activation function, is the convolutional kernel matrix, is the bias of the feature map after convolution.

7. The defect detection and recognition system according to claim 1, characterized in that Classifying and outputting by a softmax classifier, which maps the input real number to the interval [0, 1] through the sigmoid function, as follows: Its loss function is as follows: The softmax function is as follows:

8. The defect detection and recognition system according to claim 1, wherein, Also comprising: Selecting the area map with defects and the area without defects from the classification output result for segmentation, labeling them, and storing the labeled pictures in the training library; According to the corresponding relationship between its feature information and the defects, to determine whether there are defects inside the butt joint, and the type, position, and size of the defects.

9. A defect detection and recognition method based on the defect detection and recognition system according to any one of claims 1 to 8, characterized in that, Including: Sending an ultrasonic detection signal to the hot melt butt joint of the high-density polyethylene pipe body; Receiving the ultrasonic reflection signal after passing through the high-density polyethylene pipe body; Adding time gain corresponding to the ultrasonic signal for signal synthesis and amplification to generate an A-scan signal; Performing image reconstruction on the A-scan signal to generate a B-scan image; Controlling the probe to scan along the surface of the high-density polyethylene pipe body, and after the A-scan signal and the B-scan image are transformed and calculated, a C-scan image is generated; Using the short-time Fourier transform signal analysis method to extract features from the A-scan information; Using the histogram of oriented gradients image processing algorithm to extract features from the B-scan and C-scan images; Performing deep feature extraction on the signal features and image features using a convolutional neural network, and then classifying and outputting by a classifier.

10. The defect detection and recognition method according to claim 9, characterized in that, Also comprising: Selecting the area map with defects and the area without defects from the classification output result for segmentation, labeling them, and storing the labeled pictures in the training library; Based on the correspondence between its characteristic information and defects, to determine whether there are defects inside the butt joint and the type, location and size of the defects.

Images