A method for automatically identifying the fixed length of defect signals based on TOFD detection technology

Abstract

The present invention discloses a method for automatically identifying the length of defect signals based on TOFD detection technology, comprising the following steps: step 1, extracting features from the collected original signal samples using a feature extraction algorithm to obtain a signal feature set; step 2, inputting the features obtained in step 1 into a pre-designed neural network for training so that it can learn the feature information of the signal; step 3, identifying the signal using the trained neural network to obtain a recognition result set, and arranging the elements in the result set in sequence according to the corresponding sequence of the probe scan; step 4, using a connected domain solution algorithm to solve the recognition result set and obtain the length of the defect. The fixed length result obtained in the present invention is accurate and objective, and does not vary with the change of the operator. This accuracy and objectivity are derived from the effectiveness of feature extraction and the accuracy of intelligent algorithm recognition, and can be effectively used for TOFD detection of fixed length.

Description

Technical Field

[0001] The invention relates to the technical field of ultrasonic nondestructive testing, in particular to a method for automatically identifying the fixed length of a defect signal based on TOFD testing technology. Background Art

[0002] Time of flight diffraction (TOFD) is an ultrasonic nondestructive testing technology that characterizes defects by emitting ultrasonic waves into the workpiece and receiving diffraction signals generated by the end angles of the defects. Compared with conventional reflection ultrasonic testing technology, it has the characteristics of fast testing speed, high quantitative accuracy and high detection rate, so it is widely used in on-site testing of industrial parts.

[0003] In actual TOFD testing, when the probe detects a defect, the defect forms a crescent-shaped morphology with an arc-shaped tail on the edge on the D-scan image, such as Figure 1 As shown. The traditional quantitative method uses the parabola fitting method, that is, two parabolas are used to fit the tails on the left and right ends of the defect, and the distance between the two parabola vertices (AB) is the length of the defect. When using this method for length determination, the following difficulties are encountered: (1) The arc tail of each defect is different, so the fitting parabola used must be adjusted at any time according to the actual situation, while the parabola in actual operation is fixed, resulting in poor fitting effect; (2) The fitting process is relatively subjective, and different people will produce different defect lengths; (3) Strong noise background sometimes submerges the arc tail, and it is difficult to use the above method at this time. In summary, there is an urgent need to develop an automatic TOFD length determination method.

[0004] The methods of the prior art have at least the following technical problems: (1) The current manual analysis of ultrasonic TOFD signals has problems such as cumbersome workflow and the evaluation results relying on the inspection personnel's experience in analyzing defect images; (2) The accuracy of defect length determination is poor; (3) Under certain working conditions, such as when the signal contains strong noise, the prior art is difficult to implement. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method for automatically identifying the length of defect signals based on TOFD detection technology, perform a series of processing operations on the collected original signal data, and thus realize automatic, objective and accurate length determination.

[0006] In order to solve the above technical problems, the present invention provides a method for automatically identifying the fixed length of defect signals based on TOFD detection technology, comprising the following steps:

[0007] Step 1: Use a feature extraction algorithm to extract features from the collected original signal samples to obtain a signal feature set;

[0008] Step 2: Input the features obtained in step 1 into a pre-designed neural network to train it so that it can learn the characteristic information of the signal;

[0009] Step 3: Use the trained neural network to identify the signal, obtain an identification result set, and arrange the elements in the result set in the order corresponding to the probe scanning;

[0010] Step 4: Use the connected domain solving algorithm to solve the identification result set and obtain the defect length.

[0011] Preferably, in step 1, the original signal sample is a set of ultrasonic A-scan signals collected by scanning a test block.

[0012] Preferably, in step 1, the extracted features are signal slope, signal fluctuation range, signal fluctuation variance, signal energy and signal maximum energy frequency.

[0013] Preferably, in step 1, a feature extraction algorithm is used to extract features from the collected original signal samples, and the signal feature set obtained is specifically:

[0014] (a) Signal slope;

[0015]

[0016] Among them, S n , S m are the maximum and minimum amplitude values ​​of the signal segment between the through wave and the bottom wave, respectively, and n and m are the corresponding moments or sampling points;

[0017] (b) Signal fluctuation range;

[0018] ΔS=S n -S m

[0019] (c) signal fluctuation variance;

[0020]

[0021] Among them, S i , are the amplitude and average amplitude of the signal at a certain moment, respectively, and l is the length of the signal;

[0022] (d) signal energy;

[0023]

[0024] (e) Signal maximum energy frequency f maxE , that is, the frequency corresponding to the maximum energy is obtained through wavelet transform.

[0025] Preferably, in step 2, the designed neural network is a three-layer fully connected neural network with one hidden layer, in the form of: 5-30-2, indicating 5 neurons in the input layer, 30 neurons in the hidden layer, and 2 neurons in the output layer.

[0026] Preferably, in step 3, the signal is identified using a trained neural network to obtain an identification result set, and the elements in the result set are arranged in sequence according to the order corresponding to the probe scan, specifically: the order of the identification results is consistent with the signal emission order.

[0027] Preferably, in step 4, the connected domain solving algorithm is used to solve the recognition result set and then obtain the defect length as follows: traverse the elements in the Recog set, when Recog(i)=1 and the following Recog(j)=0 appear, extract the scanning position coordinates i, j, then the defect length is equal to ji.

[0028] The beneficial effects of the present invention are as follows: the present invention performs a series of processing operations (feature extraction, intelligent recognition, connected domain solution) on the collected original signal data, thereby realizing automatic, objective and accurate length determination; the obtained length determination result is accurate and objective, and does not vary with changes in operators. This accuracy and objectivity are derived from the effectiveness of feature extraction and the accuracy of intelligent algorithm recognition, and can be effectively used for TOFD detection length determination. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A schematic diagram showing length measurement using the traditional parabola lamination method on a D-scan image.

[0030] Figure 2 Schematic diagram of the neural network structure of the present invention.

[0031] Figure 3 It is a schematic diagram of the defect identification result in the present invention.

[0032] Figure 4 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION

[0033] like Figure 4 As shown, a method for automatically identifying the fixed length of defect signals based on TOFD detection technology includes the following steps:

[0034] Step 1: Use the feature extraction algorithm to extract features from the collected original signal samples to obtain a signal feature set; the original signal samples are a set of ultrasonic A-scan signals collected by scanning a test block, and the extracted features are signal slope, signal fluctuation range, signal fluctuation variance, signal energy and signal maximum energy frequency.

[0035] The feature extraction algorithm is used to extract features from the collected original signal samples, and the signal feature set is obtained as follows:

[0036] (a) Signal slope;

[0037]

[0038] Among them, S n , S m are the maximum and minimum amplitude values ​​of the signal segment between the through wave and the bottom wave, respectively, and n and m are the corresponding moments or sampling points;

[0039] (b) Signal fluctuation range;

[0040] ΔS=S n -S m

[0041] (c) signal fluctuation variance;

[0042]

[0043] Among them, S i , are the amplitude and average amplitude of the signal at a certain moment, respectively, and l is the length of the signal.

[0044] (d) signal energy;

[0045]

[0046] (e) Signal maximum energy frequency f maxE , that is, the frequency corresponding to the maximum energy is obtained through wavelet transform.

[0047] Step 2: Input the features obtained in step 1 into a pre-designed neural network and train it to learn the feature information of the signal; the designed neural network is a three-layer fully connected neural network with one hidden layer in the form of 5-30-2, which means that there are 5 neurons in the input layer, 30 neurons in the hidden layer, and 2 neurons in the output layer. Figure 2 shown.

[0048] Step 3: Use the trained neural network to identify the signal, obtain a set of identification results, and arrange the elements in the result set in the order corresponding to the probe scanning; that is, the order of the identification results is consistent with the signal emission order.

[0049] Step 4: Use the connected domain solution algorithm to solve the recognition result set and obtain the defect length; traverse the elements in the Recog set, and when Recog(i)=1 and Recog(j)=0 appears, extract the scanning position coordinates i, j, then the defect length is equal to ji.

[0050] The embodiments of the present invention are described in detail below by way of examples with reference to specific examples.

[0051] Specific implementation method: This implementation method takes a carbon steel weld with a thickness of 90 mm as the detection object.

[0052] Step 1: Signal acquisition. The tested block is a 90mm thick carbon steel flat plate butt weld with 4 crack defects arranged in the weld. A 5MHz TOFD probe is used to perform non-parallel scanning of the weld (along the weld direction), with a step size of 1mm, a scanning distance of 400mm, a system gain of 60dB, a sampling frequency of 100MHz, and a sampling depth of 896. Finally, a 400×896 signal matrix is ​​obtained; feature extraction, using the relevant feature calculation formula, the feature data of all signals are extracted respectively to obtain the feature data set of the signal.

[0053] Step 2: Network training: Input the training samples in the feature data set into the designed fully connected neural network for training.

[0054] Step 3: Input the test samples in the feature data set into the trained fully connected neural network for recognition, and arrange the recognition results in sequence according to the probe scanning order. Figure 3 The recognition results are shown. As you can see, the defects are well identified.

[0055] Step 4: Use the connected domain solving algorithm to perform connected domain analysis on the identification result set, automatically obtain the defect length, and give the obtained fixed length result.

[0056] The present invention performs a series of processing operations (feature extraction, intelligent identification, connected domain solution) on the collected raw signal data, thereby realizing automatic, objective and accurate defect identification and length determination. Specifically, for a set of collected signal data, the method of the present invention first extracts its data features, and then based on these feature data, uses a neural network to intelligently identify the signal, and finally uses a connected domain solution algorithm to automatically determine the length. Therefore, the length determination result obtained by the present invention is accurate and objective, and does not vary with changes in the operator. This accuracy and objectivity stems from the effectiveness of feature extraction and the accuracy of intelligent algorithm identification, and can be effectively used for TOFD detection length determination.

Claims

1. A method for automatically identifying the fixed length of defect signals based on TOFD detection technology, characterized in that: The steps include: Step 1: Use a feature extraction algorithm to extract features from the collected original signal samples to obtain a signal feature set; Step 2: Input the features obtained in step 1 into a pre-designed neural network to train it so that it can learn the characteristic information of the signal; Step 3: Use the trained neural network to identify the signal, obtain an identification result set, and arrange the elements in the result set in the order corresponding to the probe scanning; Step 4: Use the connected domain solution algorithm to solve the recognition result set and obtain the defect length. Traverse the elements in the Recog set. When Recog(i)=1 and Recog(j)=0 appears, extract the scanning position coordinates i, j, then the defect length is equal to ji.

2. The method for automatically identifying the fixed length of defect signals based on TOFD detection technology according to claim 1, characterized in that: In step 1, the original signal sample is a set of ultrasonic A-scan signals collected by scanning a test block.

3. The method for automatically identifying the fixed length of defect signals based on TOFD detection technology according to claim 1, characterized in that: In step 1, the extracted features are signal slope, signal fluctuation range, signal fluctuation variance, signal energy and signal maximum energy frequency.

4. The method for automatically identifying the fixed length of defect signals based on TOFD detection technology according to claim 3, characterized in that: In step 1, the feature extraction algorithm is used to extract features from the collected original signal samples, and the signal feature set obtained is specifically: (a) Signal slope; Among them, S n , S m are the maximum and minimum amplitude values ​​of the signal segment between the through wave and the bottom wave, respectively, and n and m are the corresponding moments or sampling points; (b) Signal fluctuation range; ΔS=S n -S m (c) signal fluctuation variance; Among them, S i , are the amplitude and average amplitude of the signal at a certain moment, respectively, and l is the length of the signal; (d) signal energy; (e) Signal maximum energy frequency f maxE , that is, the frequency corresponding to the maximum energy is obtained through wavelet transform.

5. The method for automatically identifying the fixed length of defect signals based on TOFD detection technology according to claim 1, characterized in that: In step 2, the designed neural network is a three-layer fully connected neural network with one hidden layer in the form of: 5-30-2, which means 5 neurons in the input layer, 30 neurons in the hidden layer, and 2 neurons in the output layer.

6. The method for automatically identifying the fixed length of defect signals based on TOFD detection technology according to claim 1, characterized in that: In step 3, the trained neural network is used to recognize the signal to obtain a recognition result set Recog, and the elements in the result set are arranged in sequence according to the sequence corresponding to the probe scanning: that is, the sequence of the recognition results is consistent with the signal emission sequence.

Images