Ultrasonic intelligent detection method for GH4169 bar black spots

Abstract

The invention belongs to the technical field of material nondestructive testing, and discloses an ultrasonic intelligent detection method for GH4169 bar black spots, which comprises the following steps: acquiring a GH4169 bar bottom wave A scanning signal; performing digital signal processing on the bottom wave A-scan signal to obtain a bottom wave A-scan signal characteristic time-frequency graph; preprocessing the data; constructing a bottom wave A-scan signal feature time-frequency graph classification model by using a convolutional neural network technology; model training; and loading the bottom wave A scanning signal feature time-frequency graph of the GH4169 bar to be detected into the trained bottom wave A scanning signal feature time-frequency graph classification model, detecting whether the bar has black spots or not, and finally obtaining an ultrasonic detection result of the black spots of the GH4169 bar. According to the method, special materials are not needed, intelligent detection of the black spots of the GH4169 bar can be rapidly carried out under the condition that the GH4169 bar is not damaged, a reliable qualitative method is provided for an ultrasonic detection technology of the GH4169 bar, operability is high, analysis results are rapid and visual, and quality control and fault analysis can be effectively carried out on the high-temperature alloy bar.

Description

Technical Field

[0001] This invention belongs to the field of nondestructive testing technology for materials, and specifically relates to an ultrasonic intelligent detection method for black spots on GH4169 bars. Background Technology

[0002] With the rapid development of computer technology and artificial intelligence, non-destructive testing (NDT) is rapidly evolving towards automation, digitalization, and intelligence. Ultrasonic testing, in particular, has introduced advanced intelligent NDT analysis methods. In ultrasonic testing of larger billets, the unique ultrasonic water immersion testing method offers higher accuracy and reliability, improving product quality control. Currently, methods for determining black spots on GH4169 bars using ultrasonic testing are lacking, hindering effective quality control and fault analysis of high-temperature alloy bars. Summary of the Invention

[0003] The purpose of this invention is to provide an ultrasonic intelligent detection method for black spots on GH4169 bars. This method requires no special materials and can quickly and intelligently detect black spots on GH4169 bars without damaging them. It provides a reliable qualitative method for ultrasonic testing technology of GH4169 bars, is highly operable, and provides rapid and intuitive analysis results. It can effectively control the quality and analyze faults of high-temperature alloy bars.

[0004] The technical solution adopted in this invention is an ultrasonic intelligent detection method for black spots on GH4169 bars, comprising the following steps:

[0005] S1, acquire the bottom wave A-scan signal of GH4169 bar;

[0006] S2, perform digital signal processing on the bottom wave A-scan signal to obtain the characteristic time-frequency diagram of the bottom wave A-scan signal;

[0007] S3, Data Preprocessing;

[0008] S4. Convolutional neural network technology is used to construct a time-frequency map classification model of the bottom wave A-scan signal features;

[0009] S5, Model Training;

[0010] S6. The time-frequency map of the bottom wave A-scan signal features to be predicted is loaded into the trained bottom wave A-scan signal feature time-frequency map classification model to predict the black spot region, and finally the ultrasonic classification result of the black spot of GH4169 bar is obtained.

[0011] Furthermore, in S1, a water immersion ultrasonic automatic detection system and a 5MHz water immersion detection probe are used to collect bottom wave A-scan signals of GH4169 bars with diameters ranging from Φ250mm to Φ300mm in one area.

[0012] Furthermore, in step S2, the specific steps for performing digital signal processing on the bottom wave A-scan signal to obtain the characteristic time-frequency diagram of the bottom wave A-scan signal include:

[0013] S21, The bottom wave A-scan signal is imported into the designed non-stationary noise reduction processing program for non-stationary noise reduction processing to obtain a new bottom wave A-scan signal.

[0014] S22 combines continuous wavelet transform and short-time Fourier transform techniques to process the non-stationary noise-reduced bottom wave A-scan signal and obtain the characteristic time-frequency diagram of the bottom wave A-scan signal.

[0015] Furthermore, in step S21, the specific steps for performing non-stationary noise reduction processing to obtain a new bottom wave A-scan signal include:

[0016] S211, acquire the bottom wave A-scan signal and perform wavelet packet decomposition;

[0017] S212, Threshold noise reduction is performed after wavelet packet decomposition of the signal;

[0018] S213, perform envelope decomposition on the reconstructed signal after threshold noise reduction;

[0019] S214, acquire the noise-reduced bottom wave A-scan signal.

[0020] Furthermore, in step S22, the specific steps for processing the non-stationary A-scan signal after noise reduction by combining continuous wavelet transform and short-time Fourier transform techniques to obtain the characteristic time-frequency diagram of the A-scan signal include:

[0021] S221, obtain the frequency range of the noise-reduced bottom wave A-scan signal;

[0022] S222, based on the frequency range of S221, the frequency range of the continuous wavelet transform is limited, and the continuous wavelet transform is performed;

[0023] S223 performs a short-time Fourier transform on the signal after continuous wavelet transform processing and plots the characteristic time-frequency diagram of the bottom wavelet A-scan signal.

[0024] Furthermore, in step S3, the specific steps of data preprocessing are as follows:

[0025] The acquired bottom wave A-scan signal feature time-frequency map is imported into the data preprocessing model: the bottom wave A-scan signal feature time-frequency map is converted into the size required for model training, the image types are classified into two types: "black spot" and "non-black spot", and the obtained two types of sample images are inverted to make the total number of samples greater than 400.

[0026] Furthermore, in step S4, the specific steps for constructing a time-frequency map classification model of the bottom-wave A-scan signal features using convolutional neural network technology are as follows:

[0027] Construct convolutional and pooling layers for a convolutional neural network; build a fully connected layer with 64 fully connected neurons; and establish a classification layer.

[0028] Furthermore, in S5, the specific steps of model training include:

[0029] S51, input the preprocessed bottom wave A-scan signal feature time-frequency map into the bottom wave A-scan signal feature time-frequency map classification model;

[0030] S52, training iterations of 100-300 times, establishing two types of model libraries: "black spot" and "non-black spot";

[0031] S53, calculate the accuracy of the model, and adjust the model parameters to make the model accuracy higher than 98%;

[0032] S54, apply the loss function to make the rate of change of the loss function stable;

[0033] S55, model training is complete, ready for application.

[0034] The beneficial effects of this invention are as follows:

[0035] It requires no special materials and can quickly and intelligently detect black spots on GH4169 bars without damaging them. It provides a reliable qualitative method for ultrasonic testing of GH4169 bars, is highly operable, and provides fast and intuitive analysis results. It can effectively control the quality and analyze faults of high-temperature alloy bars. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating the steps of the ultrasonic intelligent detection method for black spots on GH4169 bars according to the present invention.

[0037] Figure 2 This is a time-frequency diagram of the bottom wave A-scan signal characteristics of Embodiment 1 of the present invention.

[0038] Figure 3 This is a schematic diagram of the black spots after dissection in Embodiment 1 of the present invention.

[0039] Figure 4 This is a time-frequency diagram of the bottom wave A-scan signal characteristics in Embodiment 2 of the present invention.

[0040] Figure 5 This is a schematic diagram showing no black spots after dissection in Embodiment 2 of the present invention.

[0041] Figure 6 This is a time-frequency diagram showing the characteristics of the bottom wave A-scan signal in Embodiment 3 of the present invention.

[0042] Figure 7 This is a schematic diagram showing black spots after dissection in Embodiment 3 of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer and more understandable, the technical solutions of this invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the following embodiments are merely illustrative of this invention and are not intended to limit the invention.

[0044] In industry, ultrasonic testing is commonly used to detect defects below the surface. Ultrasonic testing is a method of imaging a cross-section at a certain depth inside the workpiece, parallel to the workpiece surface. Signal acquisition instruments are used to store and analyze the ultrasonic echo signals inside the workpiece, and different analysis methods are employed to image and analyze the material.

[0045] A neural network is an intelligent system that uses mathematical methods to artificially simulate its structure and behavior. Its main functions are learning, training, and execution. Through learning, the neural network can be trained to be highly sensitive to certain useful information. With the rapid development of neural network technology, image classification technology has become increasingly fast and accurate. Image classification currently uses a Convolutional Neural Network (CNN) technique. CNN is a deep feedforward neural network with different analysis layers, including fully connected layers, convolutional layers, pooling layers, and batch normalization layers. The convolutional kernel moves sequentially across the input image, and the number of pixels moved can be set by the stride. After completing one-dimensional convolution calculations, it continues to learn two-dimensional convolution calculations, ultimately enabling efficient and clear identification of image differences.

[0046] An ultrasonic intelligent detection method for black spots on GH4169 bars, such as Figure 1 As shown, it includes the following steps:

[0047] S1, acquire the bottom wave A-scan signal of GH4169 bar;

[0048] A bottom wave A-scan signal was acquired in a region using an automatic water immersion ultrasonic testing system and a 5MHz water immersion probe. The diameter range of GH4169 bars (GH4169 bars with and without black spot segregation) was Φ250mm~Φ300mm.

[0049] S2, perform digital signal processing on the bottom wave A-scan signal to obtain the characteristic time-frequency diagram of the bottom wave A-scan signal. The specific steps include:

[0050] S21. The bottom wave A-scan signal is imported into the designed non-stationary noise reduction processing program for non-stationary noise reduction processing to obtain a new bottom wave A-scan signal, so as to reduce interference factors in subsequent analysis.

[0051] The specific steps for obtaining a new bottom wave A-scan signal through non-stationary noise reduction processing are as follows:

[0052] S211, acquire the bottom wave A-scan signal and perform wavelet packet decomposition;

[0053] S212, Threshold noise reduction is performed after wavelet packet decomposition of the signal;

[0054] S213, perform envelope decomposition on the reconstructed signal after threshold noise reduction;

[0055] S214, acquire the noise-reduced bottom wave A-scan signal.

[0056] S22, combining continuous wavelet transform and short-time Fourier transform techniques, processes the non-stationary noise-reduced bottom wave A-scan signal to obtain the characteristic time-frequency diagram of the bottom wave A-scan signal. Specific steps include:

[0057] S221, obtain the frequency range of the noise-reduced bottom wave A-scan signal;

[0058] S222, based on the frequency range of S221, the frequency range of the continuous wavelet transform is limited, and the continuous wavelet transform is performed;

[0059] S223 performs a short-time Fourier transform on the signal after continuous wavelet transform processing and plots the characteristic time-frequency diagram of the bottom wavelet A-scan signal.

[0060] S3, data preprocessing, includes the following steps:

[0061] The acquired bottom wave A-scan signal feature time-frequency map is imported into the data preprocessing model: the bottom wave A-scan signal feature time-frequency map is converted into the size required for model training, the image types are classified into two types: "black spot" and "non-black spot", and the obtained two types of sample images are inverted to make the total number of samples greater than 400.

[0062] S4. A time-frequency map classification model for bottom-wave A-scan signal features is constructed using convolutional neural network technology. The specific steps are as follows:

[0063] Convolutional and pooling layers of a convolutional neural network are constructed; a fully connected layer is built, which uses 64 fully connected neurons; a classification layer is established (deep features are extracted through multi-layer convolution and pooling operations, and finally classification is performed through the fully connected layer and the classification layer to achieve intelligent detection of black spots on GH4169 rods).

[0064] S5, model training, the specific steps include:

[0065] S51, input the preprocessed bottom wave A-scan signal feature time-frequency map into the bottom wave A-scan signal feature time-frequency map classification model;

[0066] S52, training iterations of 100-300 times, establishing two types of model libraries: "black spot" and "non-black spot";

[0067] S53, calculate the accuracy of the model, and adjust the model parameters to make the model accuracy higher than 98%;

[0068] S54, apply the loss function to make the rate of change of the loss function stable;

[0069] S55, model training is complete, ready for application.

[0070] S6. Load the bottom wave A-scan signal feature time-frequency map of the GH4169 bar to be tested into the trained bottom wave A-scan signal feature time-frequency map classification model to detect whether there are black spots in the bar, and finally obtain the ultrasonic detection results of black spots in the GH4169 bar.

[0071] Example 1

[0072] Using an automated water immersion ultrasonic testing system and a 5MHz water immersion probe, bottom wave A-scan signals were acquired in a region at a sensitivity of 80% of the bottom wave amplitude. The acquired bottom wave A-scan signals were then digitally processed to obtain the characteristic time-frequency diagram of the bottom wave A-scan signals (see...). Figure 2 The bottom wave A-scan signal feature time-frequency map was loaded into the trained bottom wave A-scan signal feature time-frequency map classification model. Ultrasonic detection revealed black spots on the rod. Corresponding dissection confirmed the presence of black spots on the rod (see...). Figure 3 ).

[0073] Example 2

[0074] Using an automated water immersion ultrasonic testing system and a 5MHz water immersion probe, bottom wave A-scan signals were acquired in a region at a sensitivity of 80% of the bottom wave amplitude. The acquired bottom wave A-scan signals were then digitally processed to obtain the characteristic time-frequency diagram of the bottom wave A-scan signals (see...). Figure 4 The bottom wave A-scan signal feature time-frequency map was loaded into the trained bottom wave A-scan signal feature time-frequency map classification model. Ultrasonic testing revealed no black spots on the rod. Corresponding dissection showed that the rod did not have any black spots (see...). Figure 5 ).

[0075] Example 3

[0076] Using an automated water immersion ultrasonic testing system and a 5MHz water immersion probe, bottom wave A-scan signals were acquired in a region at a sensitivity of 80% of the bottom wave amplitude. The acquired bottom wave A-scan signals were then digitally processed to obtain the characteristic time-frequency diagram of the bottom wave A-scan signals (see...). Figure 6 The bottom wave A-scan signal feature time-frequency map was loaded into the trained bottom wave A-scan signal feature time-frequency map classification model. Ultrasonic detection revealed black spots on the rod. Corresponding dissection confirmed the presence of black spots on the rod (see...). Figure 7 ).

[0077] Any content not described in detail in this specification belongs to the prior art in this technical field.

Claims

1. An ultrasonic intelligent detection method for black spots on GH4169 bars, characterized in that, Includes the following steps: S1, acquire the bottom wave A-scan signal of GH4169 bar; S2, perform digital signal processing on the bottom wave A-scan signal to obtain the characteristic time-frequency diagram of the bottom wave A-scan signal; S3, Data Preprocessing; S4. Convolutional neural network technology is used to construct a time-frequency map classification model of the bottom wave A-scan signal features; S5, Model Training; S6. The time-frequency map of the bottom wave A-scan signal features to be predicted is loaded into the trained bottom wave A-scan signal feature time-frequency map classification model to predict the black spot region, and finally the ultrasonic classification result of the black spot of GH4169 bar is obtained.

2. The ultrasonic intelligent detection method for black spots on GH4169 bars according to claim 1, characterized in that, In step S1, a water immersion ultrasonic automatic detection system and a 5MHz water immersion detection probe are used to collect bottom wave A-scan signals of GH4169 bars with diameters ranging from Φ250mm to Φ300mm in one area.

3. The ultrasonic intelligent detection method for black spots on GH4169 bars according to claim 1, characterized in that, In step S2, the specific steps for performing digital signal processing on the bottom wave A-scan signal to obtain the characteristic time-frequency diagram of the bottom wave A-scan signal include: S21, The bottom wave A-scan signal is imported into the designed non-stationary noise reduction processing program for non-stationary noise reduction processing to obtain a new bottom wave A-scan signal. S22 combines continuous wavelet transform and short-time Fourier transform techniques to process the non-stationary noise-reduced bottom wave A-scan signal and obtain the characteristic time-frequency diagram of the bottom wave A-scan signal.

4. The ultrasonic intelligent detection method for black spots on GH4169 bars according to claim 3, characterized in that, In step S21, the specific steps for performing non-stationary noise reduction processing to obtain a new bottom wave A-scan signal include: S211, acquire the bottom wave A-scan signal and perform wavelet packet decomposition; S212, Threshold noise reduction is performed after wavelet packet decomposition of the signal; S213, perform envelope decomposition on the reconstructed signal after threshold noise reduction; S214, acquire the noise-reduced bottom wave A-scan signal.

5. The ultrasonic intelligent detection method for black spots on GH4169 bars according to claim 3, characterized in that, In step S22, the specific steps for processing the non-stationary A-scan signal after noise reduction by combining continuous wavelet transform and short-time Fourier transform techniques to obtain the characteristic time-frequency map of the A-scan signal include: S221, obtain the frequency range of the noise-reduced bottom wave A-scan signal; S222, based on the frequency range of S221, the frequency range of the continuous wavelet transform is limited, and the continuous wavelet transform is performed; S223 performs a short-time Fourier transform on the signal after continuous wavelet transform processing and plots the characteristic time-frequency diagram of the bottom wavelet A-scan signal.

6. The ultrasonic intelligent detection method for black spots on GH4169 bars according to claim 1, characterized in that, In step S3, the specific steps of data preprocessing are as follows: The acquired bottom wave A-scan signal feature time-frequency map is imported into the data preprocessing model: the bottom wave A-scan signal feature time-frequency map is converted into the size required for model training, the image types are classified into two types: "black spot" and "non-black spot", and the obtained two types of sample images are inverted to make the total number of samples greater than 400.

7. The ultrasonic intelligent detection method for black spots on GH4169 bars according to claim 1, characterized in that, In step S4, the specific steps for constructing a time-frequency map classification model of the bottom wave A-scan signal features using convolutional neural network technology are as follows: Construct convolutional and pooling layers for a convolutional neural network; build a fully connected layer with 64 fully connected neurons; and establish a classification layer.

8. The ultrasonic intelligent detection method for black spots on GH4169 bars according to claim 1, characterized in that, In S5, the specific steps for model training include: S51, input the preprocessed bottom wave A-scan signal feature time-frequency map into the bottom wave A-scan signal feature time-frequency map classification model; S52, training iterations of 100-300 times, establishing two types of model libraries: "black spots" and "non-black spots"; S53, calculate the accuracy of the model, and adjust the model parameters to make the model accuracy higher than 98%; S54, apply the loss function to make the rate of change of the loss function stable; S55, model training is complete, ready for application.

Images