Automatic Identification of Interfacial Bonding Quality of Layered Structures and C-Scan Imaging Method and System

Abstract

The present invention provides a method and system for automatically identifying the bonding quality of a layered structure interface and C-scan imaging, specifically including: collecting ultrasonic echo signals of an artificial defect calibration specimen and the interface of a layered bonding structure to be detected, and respectively constructing a training set array and a test set array; preprocessing the ultrasonic echo signals respectively, and establishing an upper envelope characterization model of the ultrasonic echo signals; extracting the exponential features and time-domain features of the upper envelope of the ultrasonic echo signals; inputting the exponential features and time-domain features extracted from the training set array into a BP neural network for training to obtain a debonding defect recognition model; inputting the exponential features and time-domain features extracted from the test set array into the debonding defect recognition model, outputting the debonding defect recognition result and performing two-dimensional C-scan imaging. In view of the characteristics of the interface debonding detection signal, the present invention realizes the automatic identification and imaging of the debonding defects of the layered bonding structure interface by extracting the morphological features of the interface signal and building an effective neural network structure.

Description

Technical Field

[0001] The present invention relates to the technical field of interface defect scanning imaging of layered bonding structures, and particularly to a method and system for automatically identifying the bonding quality of the interface of a layered structure and C-scan imaging. Background Art

[0002] Layered bonding structures are widely used in the national defense and aerospace industries, such as the bonding between the propellant grain and the cladding layer in a solid rocket motor, the bonding between the insulation layer and the outer shell, the heterogeneous interfaces of composite materials and multiple materials, and the bonding in polycrystals. Due to the performance differences of the materials on both sides of the bonding interface, the stress action at the interface, and external interference factors during the construction process, debonding defects may occur in composite material bonding components during manufacturing and use, posing a great threat to the safety of the components. Therefore, it is very necessary to effectively detect the debonding defects of the bonding interface.

[0003] Traditional ultrasonic interface bonding quality detection is usually based on ultrasonic C-scan imaging, in which the imaging is achieved by a single amplitude feature, and the recognition and extraction of the defect area mainly adopt threshold segmentation imaging of the imaging result. The manual selection of the threshold and the quality of the threshold selection directly affect the detection and recognition results. With the rapid development of artificial intelligence and machine learning, the automatic recognition and imaging technology of interface debonding defects will gradually replace the traditional detection technology and become a new trend in the field of non-destructive testing. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for automatically identifying the bonding quality of the interface of a layered structure and C-scan imaging. According to the characteristics of the interface debonding detection signal, by extracting the morphological features of the interface signal and building an effective neural network structure, the automatic recognition and imaging of the debonding defects of the layered bonding structure interface are realized.

[0005] To achieve the above object, the present invention is realized through the following technical solutions:

[0006] On the one hand, the present invention provides a method for automatically identifying the bonding quality of the interface of a layered structure and C-scan imaging, including the following steps:

[0007] Step 1: Establish an artificial defect calibration specimen for the interface of the layered bonding structure, use ultrasonic array detection technology to collect the ultrasonic echo signals of the debonding defect area and the non-debonding defect area of the artificial defect calibration specimen, and construct a training set array;

[0008] Step 2: Use ultrasonic array detection technology to collect the ultrasonic echo signals of the interface of the layered bonding structure to be detected, and construct a test set array;

[0009] Step 3: Based on the training set array and the test set array constructed in Step 1 and Step 2, preprocess the ultrasonic echo signals respectively;

[0010] Step 4: Establish an upper envelope characterization model for the ultrasonic echo signal;

[0011] Step 5: Extract the exponential features and time-domain features of the upper envelope of the ultrasonic echo signal;

[0012] Step 6: Input the exponential features and time-domain features extracted from the training set array into a BP neural network for training to obtain a debonding defect recognition model;

[0013] Step 7: Input the exponential features and time-domain features extracted from the test set array into the debonding defect recognition model, and the output layer outputs the recognition result of the debonding defect at the interface of the layered bonding structure;

[0014] Step 8: Perform two-dimensional C-scan imaging on the recognition result obtained in Step 7.

[0015] Furthermore, the data acquisition modes in the ultrasonic array detection technology include but are not limited to linear scanning, one transmit and multiple receive, and full matrix scanning.

[0016] Furthermore, the preprocessing is specifically to first normalize the collected ultrasonic echo signal, and then use a fourth-order Butterworth filter to filter out the low-frequency information below 0.6 MHz. The purpose is to eliminate the differences in waveform, amplitude order of magnitude, and noise in the ultrasonic echo signal during the experiment.

[0017] Furthermore, the expression of the upper envelope characterization model of the ultrasonic echo signal is:

[0018] A(t) = exp(-t / T)u(t)

[0019] where t is time, u(t) is the unit step signal, and T is the exponential feature parameter related to the specific signal.

[0020] Based on the actual morphological characteristics of the collected signal, that is, the amplitude of the upper envelope of the waveform decays with time in sequence, and the decay curve basically decays exponentially, which conforms to the shape characteristics of the descending part of the waveform in the exponential model. Therefore, the present invention proposes the above exponential model to approximate the upper envelope curve of the waveform.

[0021] Further, an improved particle swarm algorithm is used to extract the exponential features of the upper envelope of the ultrasonic echo signal, specifically including: setting the dimension of the target search space to 1, the spatial domain to [E1, E2], the number of particles to n, and the particles are evenly distributed in the space [E1, E2]. Calculate the fitness values of the n particles for the objective function, and only retain and record the positions [P1, P2] of the optimal particle and the sub-optimal particle. Set the velocity of the optimal particle, and iteratively search for the optimal position along the direction of P1→P2 to obtain the optimal characterization parameter, which is the exponential feature. In the present invention, the optimization dimension of the exponential feature parameter of the upper envelope of the ultrasonic echo signal is 1 dimension, and the objective function is a unimodal function. If the standard particle swarm algorithm is used, there will be problems such as a long search process, large computational volume, and slow convergence speed. The proposed improved particle swarm algorithm can realize the rapid search for the optimal characteristic parameters of the upper envelope of the ultrasonic echo signal.

[0022] Further, the time-domain features of the upper envelope of the ultrasonic echo signal include signal mean, variance, standard deviation, kurtosis, waveform factor, peak factor, and margin factor, so as to extract more shape features characterizing the interface echo signal and effectively realize the features for distinguishing the debonding and non-debonding states.

[0023] Further, the BP neural network is a three-layer network structure composed of 1 input layer, 1 hidden layer, and 1 output layer; the number of nodes in the input layer is 7, and the input vector includes the exponential features and time-domain features of the upper envelope of the ultrasonic echo signal; the number of nodes in the hidden layer is 12, and its transfer function is the Sigmoid function; the number of nodes in the output layer is 1, and the output vector is the situation of whether there is a debonding defect; the learning algorithm of the BP neural network is the quasi-Newton backpropagation algorithm.

[0024] On the other hand, the present invention provides an automatic identification and C-scan imaging system for the bonding quality of the interface of a layered structure, including:

[0025] Training set array construction unit: used to collect the ultrasonic echo signals of the debonding defect area and the non-debonding defect area of the artificial defect calibration specimen of the layered bonding structure interface, and construct a training set array;

[0026] Test set array construction unit: used to collect the ultrasonic echo signals of the interface of the layered bonding structure to be detected, and construct a test set array;

[0027] Ultrasonic echo signal preprocessing unit: used to normalize the collected ultrasonic echo signals, and then use a fourth-order Butterworth filter to filter out the low-frequency information below 0.6 MHz;

[0028] Exponential feature and time-domain feature extraction unit: used to establish the corresponding upper envelope characterization model of the ultrasonic echo signal for the constructed training set array and test set array, and extract the exponential features and time-domain features of the upper envelope of the ultrasonic echo signal;

[0029] Debonding defect identification model construction unit: used to input the exponential features and time-domain features extracted from the training set array into a BP neural network for training to obtain a debonding defect identification model;

[0030] Debonding defect identification and imaging unit: used to input the exponential features and time-domain features extracted from the test set array into the debonding defect identification model, output the identification result of the debonding defect at the interface of the layered bonding structure, and perform two-dimensional C-scan imaging on the identification result.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0032] (1) The present invention uses ultrasonic array detection technology to realize data acquisition of the interface of the layered bonding structure. Compared with the traditional ultrasonic detection technology, it can achieve large-range and rapid scanning, and the present invention is applicable to the detection signal arrays obtained by various acquisition methods;

[0033] (2) According to the morphological characteristics of the actually acquired signals of the layered bonding structure, the present invention proposes an improved upper envelope characterization model of the ultrasonic echo signal, which can effectively characterize the upper envelope characteristics of the ultrasonic acquisition signals;

[0034] (3) Based on the fact that the feature parameter optimization dimension of the upper envelope characterization model of the ultrasonic echo signal is 1 dimension and the objective function is a unimodal function, the present invention proposes an improved particle swarm algorithm to realize the rapid search for the optimal feature parameters of the upper envelope characterization model;

[0035] (4) The exponential feature parameters and time-domain feature parameters of the ultrasonic echo signals extracted by the present invention can be used as effective features for characterizing the ultrasonic acquisition signals and as effective inputs for the neural network;

[0036] (5) The overall scheme proposed by the present invention can realize the automatic identification and imaging of the debonding defects at the interface of the layered bonding structure, thereby improving the intelligent level of the interface defect detection of such components. Brief Description of the Drawings

[0037] Figure 1 It is a flowchart of the automatic identification and C-scan imaging method for the bonding quality of the interface of the layered structure;

[0038] Figure 2 It is a schematic diagram of the automatic identification and C-scan imaging system for the bonding quality of the interface of the layered structure;

[0039] Figure 3 It is the C-scan imaging result of the traditional interface debonding defect;

[0040] Figure 4 It is the C-scan imaging result of the debonding defect at the interface of the layered bonding structure of the present invention. Detailed Embodiments

[0041] The object of the present invention is to provide a method and system for automatically identifying the bonding quality of the interface of a layered structure and C-scan imaging. By extracting the exponential feature parameters and time-domain feature parameters of the interface ultrasonic echo signal, an effective BP neural network structure is built for the automatic identification model of debonding defects, so as to realize the automatic identification and imaging method and system for the debonding defects of the interface of the layered bonding structure.

[0042] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] The following examples are implemented on the premise of the technical solution of the present invention, and give detailed implementation manners and specific operation processes, but do not limit the protection scope of the present invention. Any technical solutions obtained by using equivalent replacements or equivalent transformations shall fall within the protection scope of the present invention.

[0044] Example 1

[0045] This example proposes a method for automatically identifying the bonding quality of the interface of a layered structure and C-scan imaging, as Figure 1 shown, including the following steps:

[0046] Step 1: Establish an artificial defect calibration specimen for the interface of the layered bonding structure, use ultrasonic array linear scanning to collect the ultrasonic echo signals of the debonding defect area and the non-debonding defect area of the artificial defect calibration specimen, and construct a training set array;

[0047] Step 2: Use ultrasonic array linear scanning to collect the ultrasonic echo signals of the interface of the layered bonding structure to be detected, and construct a test set array;

[0048] Step 3: Based on the training set array and the test set array constructed in Step 1 and Step 2, preprocess the ultrasonic echo signals respectively;

[0049] Step 4: Establish an upper envelope characterization model for the ultrasonic echo signal;

[0050] Step 5: Extract the exponential features and time-domain features of the upper envelope of the ultrasonic echo signal;

[0051] Step 6: Input the exponential features and time-domain features extracted from the training set array into the BP neural network for training to obtain a debonding defect recognition model;

[0052] Step 7: Input the exponential features and time-domain features extracted from the test set array into the debonding defect recognition model, and output the debonding defect recognition result of the interface of the layered bonding structure;

[0053] Step 8: Perform two-dimensional C-scan imaging on the recognition result obtained in Step 7.

[0054] Furthermore, in the ultrasonic array line scan, it is assumed that the ultrasonic array is a one-dimensional linear array with N array elements, the aperture size of the linear array is D, and the area to be detected is a rectangular area with a length of M*D and a width of W. To complete the data acquisition of the entire rectangular area, each array element is sequentially excited in the direction of the row where the array element is located to complete a linear scan, obtaining N sets of echo signals, completing the scan of the area with a length of D. Next, continue to scan the area with a length of 2*D in the row direction until the scan of the area with a length of M*D in one row is completed. Then repeat the above acquisition process for the next row, sequentially scanning the rectangular area to be detected, and obtaining M*N*W sets of echo signals h ij (h ij represents the detection signal obtained by the j-th acquisition in the i-th row), constituting a linear scan data set;

[0055]

[0056] Furthermore, the expression of the upper envelope characterization model of the ultrasonic echo signal is:

[0057] A(t) = exp(-t / T)u(t)

[0058] where t is time, u(t) is the unit step signal, and T is the exponential characteristic parameter related to the specific signal.

[0059] Furthermore, an improved particle swarm algorithm is used to extract the exponential characteristics of the upper envelope of the ultrasonic echo signal, specifically including: setting the dimension of the target search space to 1, the spatial domain definition to [E1, E2], the number of particles to n, the particles are evenly distributed in the space [E1, E2], calculating the fitness values of the n particles for the objective function, only retaining and recording the positions [P1, P2] of the optimal particle and the sub-optimal particle, setting the velocity of the optimal particle, and iteratively searching for the optimal position along the direction of P1→P2 to obtain the optimal characterization parameter, which is the exponential characteristic.

[0060] Furthermore, the time-domain characteristics of the upper envelope of the ultrasonic echo signal include signal mean, variance, standard deviation, kurtosis, waveform factor, peak factor, and margin factor.

[0061] Furthermore, the BP neural network is a three-layer network structure composed of 1 input layer, 1 hidden layer, and 1 output layer; the number of nodes in the input layer is 7, and the input vector includes the exponential characteristics and time-domain characteristics of the upper envelope of the ultrasonic echo signal; the number of nodes in the hidden layer is 12, and its transfer function is the Sigmoid function; the number of nodes in the output layer is 1, and the output vector is the situation of whether there is a debonding defect; the learning algorithm of the BP neural network is the quasi-Newton backpropagation algorithm.

[0062] Embodiment 2

[0063] This embodiment proposes a system for automatically identifying the bonding quality of a layered structure interface and C-scan imaging, which is applied to the method for automatically identifying the bonding quality of a layered structure interface and C-scan imaging proposed in Embodiment 1, as Figure 2 shown, and specifically includes:

[0064] Training set array construction unit: used to collect the ultrasonic echo signals of the debonding defect area and the non-debonding defect area of the artificial defect calibration specimen of the layered bonding structure interface by using ultrasonic array detection technology, and construct a training set array;

[0065] Test set array construction unit: used to collect the ultrasonic echo signals of the interface of the layered bonding structure to be detected by using ultrasonic array detection technology, and construct a test set array;

[0066] Exponential feature and time-domain feature extraction unit: used to establish a corresponding ultrasonic echo signal upper envelope characterization model for the constructed training set array and test set array, and extract the exponential features and time-domain features of the ultrasonic echo signal upper envelope;

[0067] Ultrasonic echo signal preprocessing unit: used to normalize the collected ultrasonic echo signals, and then use a fourth-order Butterworth filter to filter out low-frequency information below 0.6 MHz;

[0068] Debonding defect recognition model construction unit: used to input the exponential features and time-domain features extracted from the training set array into a BP neural network for training to obtain a debonding defect recognition model;

[0069] Debonding defect recognition and imaging unit: used to input the exponential features and time-domain features extracted from the test set array into the debonding defect recognition model, output the debonding defect recognition result of the layered bonding structure interface, and perform two-dimensional C-scan imaging on the recognition result.

Claims

1. An automatic recognition method for the bonding quality of a layered structure interface and a C-scan imaging method, characterized in that: It includes the following steps: Step 1: Establish an artificial defect calibration specimen for the interface of the layered bonding structure, collect the ultrasonic echo signals in the debonding defect area and the non-debonding defect area of the artificial defect calibration specimen, and construct a training set array; Step 2: Collect the ultrasonic echo signals of the interface of the layered bonding structure to be detected, and construct a test set array; Step 3: Based on the training set array and the test set array constructed in Step 1 and Step 2, preprocess the ultrasonic echo signals respectively; Step 4: Establish an upper envelope characterization model for the ultrasonic echo signals; Step 5: Extract the exponential features and time-domain features of the upper envelope of the ultrasonic echo signals; Use an improved particle swarm optimization algorithm to extract the exponential features of the upper envelope of the ultrasonic echo signals, specifically including: setting the dimension of the target search space to 1, the spatial domain definition to [E1, E2], the number of particles to n, the particles are evenly distributed in the space [E1, E2], calculate the fitness values of the n particles for the objective function, only retain and record the positions [P1, P2] of the optimal particle and the sub-optimal particle, set the velocity of the optimal particle, and iteratively search for the optimal position along the direction of P1→P2 to obtain the optimal characterization parameter, which is the exponential feature; Step 6: Input the exponential features and time-domain features extracted from the training set array into a BP neural network for training to obtain a debonding defect recognition model; Step 7: Input the exponential features and time-domain features extracted from the test set array into the debonding defect recognition model, and output the debonding defect recognition result of the interface of the layered bonding structure; Step 8: Perform two-dimensional C-scan imaging on the recognition result obtained in Step 7.

2. The automatic recognition method for the bonding quality of a layered structure interface and the C-scan imaging method according to claim 1, characterized in that: The preprocessing is specifically to first normalize the collected ultrasonic echo signals, and then use a fourth-order Butterworth filter to filter out the low-frequency information below 0.6 MHz.

3. The automatic recognition method for the bonding quality of a layered structure interface and the C-scan imaging method according to claim 1, characterized in that: The expression of the upper envelope characterization model of the ultrasonic echo signals is: A(t) = exp(-t / T)u(t) where t is time, u(t) is the unit step signal, and T is the exponential feature parameter related to the specific signal.

4. The automatic recognition method for the bonding quality of a layered structure interface and the C-scan imaging method according to claim 1, wherein: The time-domain features of the upper envelope of the ultrasonic echo signals include signal mean, variance, standard deviation, kurtosis, waveform factor, peak factor, and margin factor.

5. The automatic recognition method for the bonding quality of a layered structure interface and the C-scan imaging method according to claim 1, wherein: The BP neural network is a three-layer network structure composed of 1 input layer, 1 hidden layer, and 1 output layer; the number of nodes in the input layer is 7, and the input vector includes the exponential features and time-domain features of the upper envelope of the ultrasonic echo signals; the number of nodes in the hidden layer is 12, and its transfer function is the Sigmoid function; the number of nodes in the output layer is 1, and the output vector is the situation of whether there is a debonding defect; the learning algorithm of the BP neural network is the quasi-Newton backpropagation algorithm.

6. An automatic recognition system for the bonding quality of a layered structure interface and a C-scan imaging system, which is applied to the automatic recognition method for the bonding quality of a layered structure interface and the C-scan imaging method according to any one of claims 1 to 5, and is characterized in that: It includes: Training set array construction unit: used to collect the ultrasonic echo signals in the debonding defect area and the non-debonding defect area of the artificial defect calibration specimen of the interface of the layered bonding structure, and construct a training set array; Test set array construction unit: used to collect the ultrasonic echo signals of the interface of the layered bonding structure to be detected, and construct a test set array; Ultrasonic echo signal preprocessing unit: used to normalize the collected ultrasonic echo signals, and then use a fourth-order Butterworth filter to filter out the low-frequency information below 0.6 MHz; Exponential Feature and Time Domain Feature Extraction Unit: It is used to establish a corresponding ultrasonic echo signal upper envelope characterization model for the constructed training set array and test set array, and extract the exponential features and time domain features of the ultrasonic echo signal upper envelope; Debonding Defect Recognition Model Construction Unit: It is used to input the exponential features and time domain features extracted from the training set array into a BP neural network for training to obtain a debonding defect recognition model; Debonding Defect Recognition and Imaging Unit: It is used to input the exponential features and time domain features extracted from the test set array into the debonding defect recognition model, output the debonding defect recognition result of the layered bonding structure, and perform two-dimensional C-scan imaging on the recognition result.

Images