Data-driven mechanical arm ultrasonic guided wave multi-defect detection path planning method

Abstract

This invention relates to a data-driven robotic arm ultrasonic guided wave multi-defect detection path planning method, belonging to the field of non-destructive testing and automated inspection technology, and solving the technical problem of the lack of automated defect detection for large plate structures and large curved surface structures in the prior art. The applicable system includes a control and testing system and the test piece. This invention uses a robotic arm carrying electromagnetic acoustic sensors to scan large plate structures and large curved surface structures, thereby realizing the detection, location, and identification of multiple different defects; it can reduce the use of coupling agent, improve detection efficiency, improve detection accuracy, and enhance the accuracy of defect location and identification; the scanning path of the robotic arm is optimized to improve detection efficiency; the collected data is used as the raw dataset to construct a deep learning model and an ensemble CNN model, realizing robotic arm scanning detection path planning for multiple defects inside large components and large curved surface structures.

Description

Technical Field

[0001] This invention relates to the fields of nondestructive testing and automated testing, and in particular to a data-driven method for planning the ultrasonic guided wave multi-defect detection path for robotic arms. Background Technology

[0002] Large plate structures and large curved surface structures are widely used in automobiles, high-speed rail, aerospace, and other fields. Defects such as delamination and cracks generated during the manufacturing process, as well as microcracks and corrosion that develop during service, can severely affect the performance of these large structures, and in severe cases, cause the entire device or equipment to fail. Currently, the detection of defects in these large structures mostly uses traditional ultrasonic non-destructive testing (NDT). The application of coupling agent and the contact between the sensor and the test piece can significantly affect the accuracy of the detection. Furthermore, most traditional ultrasonic NDT methods rely on manual inspection, which is time-consuming and labor-intensive for large structures, and makes it difficult to guarantee positioning and detection accuracy. Therefore, there is a need to develop automated defect detection methods for large components to improve efficiency, ensure detection accuracy, and guarantee precise positioning. Summary of the Invention

[0003] This invention aims to address the technical problem that there is no automated defect detection method for large plate structures and large curved surface structures in the prior art, and provides a data-driven robotic arm ultrasonic guided wave multi-defect detection path planning method.

[0004] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0005] A data-driven method for planning the ultrasonic guided wave multi-defect detection path of a robotic arm includes the following steps:

[0006] First, based on the reference defect signal and noise signal, the signals acquired N times are post-processed and analyzed to distinguish between noise signals and defect signals. At the same time, signal amplitude thresholds are set for different defects present on the test piece. N is the number of times the signal is acquired, and its value can be any positive integer.

[0007] Secondly, the robotic arm, controlled by a computer, carries an electromagnetic acoustic sensor to perform a global scan of the test piece according to a preset scanning path, and records the collected signals.

[0008] The recorded signal is compared and analyzed with the noise signal and defect signal distinguished in the previous step to determine whether the recorded signal is a defect signal. If it is, the detection position of the defect is recorded, and multi-position detection is performed with the detection position as the center and a certain distance between the defect and the detection position as the radius. Otherwise, the test piece is scanned according to the preset scanning path, and the above steps are repeated.

[0009] Determine whether the robotic arm has completed the global scan. If so, record the location and type of all defects and the time taken for scanning and inspection. Otherwise, continue scanning the test piece according to the preset scanning path and repeat the above steps until the global scan is completed.

[0010] Determine whether all alternative scanning paths have been scanned. If so, compare the scanning time to determine the optimal scanning path. Otherwise, select a different scanning path and repeat the scanning detection process.

[0011] The recorded noise and defect signals are processed to form the original dataset, which is then divided into a training set, a validation set, and a test set.

[0012] Construct a deep learning model and an ensemble CNN model, train the ensemble CNN model using training set data, apply test set data to verify the trained deep learning model, and compare the results with the results of the original dataset to determine whether the defect location and type are accurate.

[0013] In the above technical solution, the detection location distribution is specifically as follows: based on the defect location... Defect detection location to Located in The distance is Arranged at equal intervals on the circumference, either clockwise or counterclockwise. to Represents the defect detection location number, where The value of can be any positive integer;

[0014] The deep learning model outputs a classification prediction result, which is the probability of whether there is a defect in a specific area, the specific location of the defect, and its type.

[0015] The ensemble CNN model comprises two parts: a CNN model and a regression model. The CNN model is used to extract data features of defects, while the regression model is used to establish the relationship between data features and defect locations. The ensemble CNN model is trained using training set data.

[0016] The trained deep learning model is validated using test set data. The results are compared with those of the original dataset to determine whether the location and type of defects are accurate. If so, the proposed deep learning model is applied to other test pieces for scanning and detection. Otherwise, the proposed deep learning model and the integrated CNN model should be optimized to improve the output accuracy.

[0017] In the above technical solution, the preset scanning path of the robotic arm is as follows: the scanning direction is initially the positive horizontal direction; after reaching the preset scanning distance, it scans in the positive vertical direction. Distance; then, the scanning direction is the opposite of the horizontal direction; after reaching the preset scanning distance, scanning proceeds in the forward direction of the vertical direction. Distance; then, the scanning direction is adjusted back to the horizontal positive direction, and the above steps are repeated until the global scan is completed, wherein the preset scanning distance in the positive and negative directions should be less than or equal to the size of the test piece along the scanning direction.

[0018] In the above technical solution, the preset scanning path of the robotic arm is specifically as follows: the robotic arm scans in a clockwise or counterclockwise direction, taking the endpoint of the specified scanning area as the scanning start position, and using a scanning step distance. Perform a horizontal forward scan with step increments. When the scan distance reaches the length of the specified scan area, turn 90 degrees to the right and continue scanning with step increments. A vertical reverse scan is performed to increment the step size. When the scan distance reaches the width of the specified scan area, the camera turns 90 degrees to the right for the second time to scan the step size. A horizontal reverse scan is performed to increment the step size. When the scan distance reaches the length of the specified scan area, the camera turns 90 degrees to the right for the third time to increment the step size. Vertical forward scanning is performed to increment the step size. A scan step is defined as the distance from the endpoint of the specified scan area to the scan distance. Then, it turns 90 degrees to the right for the fourth time and repeats the above trajectory, with the scanning distance before each turn decreasing by the scanning step size. ,in For the scan step, The value can be any number.

[0019] In the above technical solution, the preset scanning path of the robotic arm is specifically as follows: Y positions are randomly selected within a specified scanning area to test the workpiece, and the collected signals are recorded; where Y is the number of selected positions, which is any positive integer. In the above technical solution, the deep learning model specifically includes: an input layer, a convolutional layer, a pooling layer, an activation layer, and an output layer.

[0020] The input layer serves as the data entry point for the model, receiving raw data and converting it into a tensor format that the model can process, providing initial input for subsequent layers. The convolutional layer extracts local features from the input data using filters, capturing local correlations and reducing model parameters through parameter sharing, thus improving computational efficiency. The pooling layer samples the feature maps output by the convolutional layers, reducing feature dimensionality and computational cost by preserving the maximum or average values ​​of local regions, while enhancing the model's robustness to minor deformations in the input data. The activation layer introduces nonlinear transformations to overcome the limitations of linear models, enabling the model to learn complex nonlinear relationships and improving its ability to express complex patterns. The output layer serves as the model's output, providing the final result based on the task type.

[0021] In the above technical solution, the test piece is a plate structure or a large curved surface structure.

[0022] In the above technical solution, the test piece is a large-sized metal sheet specimen with specific dimensions of 2 m × 2 m × 3 mm.

[0023] In the above technical solutions, the types of defects include: regular types and irregular types; among which, regular types include: rectangles, circles and polygons.

[0024] In the above technical solution, the test system to which the method of the present invention is applicable includes: a control and testing system, and a test piece;

[0025] The control and testing system includes: a computer, a robotic arm, electromagnetic acoustic sensors, a signal generator / receiver, a power amplifier, and a signal amplifier; among which, the electromagnetic acoustic sensors include: a transmitting sensor and a receiving sensor;

[0026] The transmitting sensor is used to emit the first ultrasonic guided wave; the receiving sensor is used to receive the second ultrasonic guided wave.

[0027] The computer is connected to the signal generator / receiver and the robotic arm; the signal generator / receiver is connected to the power amplifier and the signal amplifier; the electromagnetic acoustic sensor is connected to the power amplifier, the signal amplifier, and the robotic arm.

[0028] The computer is used to control the operation of the signal generator / receiver and the robotic arm;

[0029] The robotic arm is used to control the movement of the electromagnetic acoustic sensor; the electromagnetic acoustic sensor is used to detect the test piece.

[0030] Signal generators / receivers are used to generate and / or receive signals;

[0031] Power amplifiers are used to amplify the power of signals generated by signal generators / receivers;

[0032] A signal amplifier is used to amplify the signals detected and returned by an electromagnetic acoustic sensor.

[0033] The present invention has the following beneficial effects:

[0034] This invention presents a data-driven robotic arm ultrasonic guided wave multi-defect detection path planning method. The method utilizes a robotic arm carrying electromagnetic acoustic sensors to scan large sheet metal and large curved surface structures, thereby enabling the detection, localization, and identification of various defects. This method reduces the use of coupling agent, significantly improves detection efficiency, enhances detection accuracy, and increases the precision of defect localization and identification. The robotic arm's scanning path is optimized to improve detection efficiency. The collected data is used as the raw dataset to construct deep learning models and ensemble CNN models to achieve robotic arm scanning detection path planning for multiple defects within large components and large curved surface structures. Attached Figure Description

[0035] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0036] Figure 1 This is a schematic diagram of the test system applicable to the data-driven robotic arm ultrasonic guided wave multi-defect detection path planning method of the present invention.

[0037] Figure 2 This is a schematic diagram of a robotic arm scanning path.

[0038] Figure 3 This is a schematic diagram of another type of robotic arm scanning path.

[0039] Figure 4 This is a schematic diagram of another type of robotic arm scanning path.

[0040] Figure 5 This is a schematic diagram of the detection location distribution.

[0041] Figure 6 This is a schematic diagram of a noise signal.

[0042] Figure 7 This is a schematic diagram of a defect signal.

[0043] Figure 8 This is a schematic diagram of the detection process of the data-driven robotic arm ultrasonic guided wave multi-defect detection path planning method of the present invention.

[0044] Figure 9 This is a flowchart of the deep learning model workflow.

[0045] The reference numerals in the figure are:

[0046] 1: Control and testing system;

[0047] 11: Computer; 12: Robotic arm; 13: Electromagnetic acoustic sensor; 14: Signal generator / receiver; 15: Power amplifier; 16: Signal amplifier;

[0048] 131: Transmitting sensor; 132: Receiving sensor; 133: First ultrasonic waveguide; 134: Second ultrasonic waveguide;

[0049] 2: Large-size metal sheet specimen; 21: Defects. Detailed Implementation

[0050] The present invention will now be described in detail with reference to the accompanying drawings.

[0051] The data-driven robotic arm ultrasonic guided wave multi-defect detection path planning method of the present invention is applicable to the test system, such as... Figure 1 As shown, it includes: a control and testing system 1, and a test piece; the test piece is a large-size metal plate specimen 2, with specific dimensions of 2 m × 2 m × 3 mm.

[0052] in:

[0053] The control and testing system 1 includes: a computer 11, a robotic arm 12, an electromagnetic acoustic sensor 13, a signal generator / receiver 14, a power amplifier 15, and a signal amplifier 16. The electromagnetic acoustic sensor 13 mainly consists of two parts: a transmitting sensor 131 and a receiving sensor 132. The transmitting sensor 131 is used to emit a first ultrasonic guided wave 133; the receiving sensor 132 is used to receive a second ultrasonic guided wave 134.

[0054] Computer 11 is connected to signal generator / receiver 14 and robotic arm 12 respectively; signal generator / receiver 14 is connected to power amplifier 15 and signal amplifier 16 respectively; electromagnetic acoustic sensor 13 is connected to power amplifier 15, signal amplifier 16, and robotic arm 12 respectively.

[0055] Computer 11 is used to control the operation of signal generator / receiver 14 and robotic arm 12;

[0056] The robotic arm 12 is used to control the movement of the electromagnetic acoustic sensor 13; the electromagnetic acoustic sensor 13 is used to inspect the large-size metal plate specimen 2.

[0057] Signal generator / receiver 14 is used to generate and / or receive signals;

[0058] Power amplifier 15 is used to amplify the signal generated by signal generator / receiver 14;

[0059] Signal amplifier 16 is used to amplify the signal detected and returned by electromagnetic acoustic sensor 13;

[0060] Specifically, the signal generated by the signal generator / receiver 14 is amplified by the power amplifier 15 and then emitted by the transmitting sensor 131 to form the first ultrasonic guided wave 133; the second ultrasonic guided wave 134 is received by the receiving sensor 132, amplified by the signal amplifier 16, and then received by the signal generator / receiver 14.

[0061] Defect 21 exists on large-size metal sheet specimen 2. Defect 21 can be of various types, for example... Figure 1 The system can draw regular types such as rectangles, circles, triangles, and polygons, as well as irregular types, such as those created during erosion and other processes.

[0062] The present invention relates to a data-driven path planning method for ultrasonic guided wave multi-defect detection in robotic arms, such as... Figure 8 As shown (only a summary of the method steps is shown in the figure), the method includes the following steps:

[0063] First, based on the reference defect signal and noise signal, the signals acquired N times are post-processed and analyzed to distinguish noise signals (such as...). Figure 6 (as shown) and defect signals (such as) Figure 7 As shown in the figure, signal amplitude thresholds are set for different defects 21, and signal amplitudes are calibrated for different types of defects 21, such as corrosion, cracks and delamination; N is the number of times the signal is collected, and its value can be any positive integer. In this embodiment, N = 1000 is used as an example.

[0064] Secondly, the computer 11 controls the robotic arm 12 to carry the electromagnetic acoustic sensor 13 according to the scanning path diagram (such as...). Figure 2 (As shown) A global scan was performed on the large-size metal sheet specimen 2 under test; wherein, the scan step size was... , The value of can be any number, here we take as . Taking 2 mm as an example, the collected signals were recorded;

[0065] The recorded signal is compared and analyzed with the calibrated signals of different types of defects 21 to determine whether the recorded signal is a defect signal; if so, the detection location of the defect 21 is recorded, and a range of values ​​between that detection location and the defect 21 is calculated. The distance is the radius, based on the distribution of detection locations (e.g., Figure 5 As shown) Perform multi-position ( (location) detection, here taking = 9 as an example; the specific distribution of detection locations is as follows: taking defect 21 as the location of the defect. Detection location to Located in The distance is Arranged at equal intervals on the circumference, either clockwise or counterclockwise. to This represents the detection location number of defect 21, where The value of can be any positive integer, here we take as . = 9 for example; otherwise, the robotic arm 12 follows the preset scanning path (such as Figure 2 (As shown) Continue scanning the large-size metal sheet specimen 2 and repeat the above steps; as shown Figure 2 As shown, the preset scanning path of the robotic arm 12 is as follows: the scanning direction is initially in the horizontal positive direction; after reaching the preset scanning distance, it scans in the vertical positive direction. Distance; then, the scanning direction is the opposite of the horizontal direction; after reaching the preset scanning distance, scanning proceeds in the forward direction of the vertical direction. Distance; then, the scanning direction is adjusted back to the horizontal positive direction, and the above steps are repeated until the global scan is completed, wherein the preset scanning distance in the positive and negative directions should be less than or equal to the dimension of the test piece along the scanning direction; wherein The value of satisfies the following: it can be any positive integer, and The range of values ​​should be ≥ wavelength to ensure a clear defect 21 reflected signal is obtained.

[0066] Determine whether the robotic arm 12 has completed the global scan. If so, record the locations of all defects 21 and the scanning time; otherwise, follow the preset scanning path (e.g., ...). Figure 2 (As shown) Continue scanning the test piece and repeat the above steps until the global scan is complete.

[0067] Determine whether all alternative scan paths have been completed. Figure 2-4 If the scan is positive, compare the scan time to determine the optimal scan path; otherwise, select a different scan path and repeat the scan detection process.

[0068] Next, we will build deep learning models and ensemble CNN models;

[0069] The workflow of building a deep learning model, such as Figure 9 As shown in the figure (only a summary of the method steps is shown), the specific steps include:

[0070] First, the recorded noise and defect signals are organized and used as the raw dataset. The raw dataset is then divided into three parts: training set, validation set, and test set, and the proportions are set according to requirements.

[0071] A deep learning model is constructed, specifically comprising an input layer, convolutional layers, pooling layers, activation layers, and an output layer. The input layer serves as the model's data entry point, receiving raw data and converting it into a tensor format that the model can process, providing initial input for subsequent layers. The convolutional layer primarily extracts local features from the input data using filters, capturing local correlations and reducing model parameters through parameter sharing, thus improving computational efficiency. The pooling layer samples the feature maps output by the convolutional layer, reducing feature dimensionality and computational load by preserving local maximum or average values, while enhancing the model's robustness to minor deformations in the input data. The activation layer introduces nonlinear transformations to overcome the limitations of linear models, enabling the model to learn complex nonlinear relationships and improving its ability to express complex patterns. The output layer serves as the model's output, providing the final result based on the task type. It can also provide classification predictions based on the model's output, such as the probability of the presence or absence of defect 21 in a specific region, the specific location of defect 21, and the specific type of defect 21.

[0072] The ensemble CNN model is constructed, which mainly consists of two parts: a CNN model and a regression model. The CNN model is used to extract the data features of defect 21, and the regression model is used to establish the relationship between the data features and the location of defect 21. The ensemble CNN model is trained using the training set data.

[0073] The trained model is validated using test set data. The results are compared with those of the original dataset to determine whether the location and type of defect 21 are accurate. If so, the scanning path efficiency is compared, and the optimal scanning path is selected. The proposed deep learning model can then be applied to other test samples for scanning detection. Otherwise, the proposed deep learning model and the integrated CNN model should be optimized to improve their output accuracy.

[0074] In the data-driven robotic arm ultrasonic guided wave multi-defect detection path planning method of the present invention, the preset and alternative scanning paths include: serpentine structure, loop structure, and scattered structure; specifically, such as... Figure 2 As shown in the above specific embodiment, the scanning path of the robotic arm 12 in the data-driven robotic arm ultrasonic guided wave multi-defect detection path planning method of the present invention is a serpentine structure, wherein the scanning step distance is... , The value can be any number. The preset or alternative scanning paths for the robotic arm 12 also include... Figure 3 and Figure 4 The scan path shown. Specifically, Figure 3 The scanning path shown is as follows: the robotic arm scans in a clockwise or counterclockwise direction, starting from the endpoint of the specified scanning area, and using a scanning step distance. Perform a horizontal forward scan with step increments. When the scan distance reaches the length of the specified scan area, turn 90 degrees to the right and continue scanning with step increments. A vertical reverse scan is performed to increment the step size. When the scan distance reaches the width of the specified scan area, the camera turns 90 degrees to the right for the second time to scan the step size. A horizontal reverse scan is performed to increment the step size. When the scan distance reaches the length of the specified scan area, the camera turns 90 degrees to the right for the third time to increment the step size. Vertical forward scanning is performed to increment the step size. A scan step is defined as the distance from the endpoint of the specified scan area to the scan distance. Then, it turns 90 degrees to the right for the fourth time and repeats the above trajectory, with the scanning distance before each turn decreasing by the scanning step size. ,in For the scan step, The value of can be any number. Figure 4 The scanning path shown is as follows: multiple test positions are randomly selected, and Y positions are randomly selected within the specified scanning area to test the test piece and record the acquired signals; where Y is the number of selected positions, which is any positive integer.

[0075] In order to accurately determine the location of defect 21, in other specific embodiments, The value of should satisfy: It is a positive integer, and ≥6.

[0076] In the above specific embodiments, the test specimen is a large-size metal plate specimen 2; in other specific embodiments, the test specimen may also be a plate structure or a large curved surface structure made of other materials.

[0077] In the above specific embodiments, the sensor involved is an electromagnetic acoustic sensor 13; in other specific embodiments, the sensor may also be an eddy current sensor, a traditional piezoelectric ceramic sensor, etc.

[0078] This invention presents a data-driven robotic arm ultrasonic guided wave multi-defect detection path planning method. The method utilizes a robotic arm carrying electromagnetic acoustic sensors to scan large sheet metal and large curved surface structures, thereby enabling the detection, localization, and identification of various defects. This method reduces the use of coupling agent, significantly improves detection efficiency, enhances detection accuracy, and increases the precision of defect localization and identification. The robotic arm's scanning path is optimized to improve detection efficiency. The collected data is used as the raw dataset to construct deep learning models and ensemble CNN models to achieve robotic arm scanning detection path planning for multiple defects within large components and large curved surface structures.

[0079] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A data-driven method for planning the ultrasonic guided wave multi-defect detection path of a robotic arm, characterized in that, Includes the following steps: First, based on the reference defect signal and noise signal, the signals collected N times are post-processed and analyzed to distinguish between noise signals and defect signals. At the same time, the signal amplitude thresholds for different defects (21) existing on the test piece are set; N is the number of times the signal is collected, and its value can be any positive integer. Secondly, the robotic arm (12) is controlled by computer (11) to carry electromagnetic acoustic sensors (13) to perform a global scan of the test piece according to the preset scanning path, and the collected signals are recorded. The recorded signal is compared and analyzed with the noise signal and defect signal after the previous step to determine whether the recorded signal is a defect signal. If it is, the detection position of the detected defect (21) is recorded, and multi-position detection is performed based on the detection position distribution with the detection position as the center and a certain distance between the detection position and the defect (21) as the radius. Otherwise, the test piece is scanned according to the preset scanning path, and the above steps are repeated. Determine whether the robotic arm (12) has completed the global scan. If yes, record the location and type of all defects (21) and the time taken for scanning and detection. Otherwise, continue to scan the test piece according to the preset scanning path and repeat the above steps until the global scan is completed. Determine whether all alternative scanning paths have been scanned. If so, compare the scanning time to determine the optimal scanning path. Otherwise, select a different scanning path and repeat the scanning detection process. The recorded noise and defect signals are processed to form the original dataset, which is then divided into a training set, a validation set, and a test set. Construct a deep learning model and an ensemble CNN model, and train the ensemble CNN model using training set data; apply test set data to verify the trained deep learning model, and compare the results with the original dataset results to determine whether the location and type of the defect (21) are accurate. The specific distribution of detection locations is as follows: The location of defect (21) is taken as Defect (21) detection location to Located in The distance is Arranged at equal intervals on the circumference, either clockwise or counterclockwise. to The defect (21) is represented by the detection location number, where ; The deep learning model outputs a classification prediction result, which is the probability of whether there is a defect (21) in a specific region, the specific location of the defect (21), and its type. The ensemble CNN model consists of two parts: a CNN model and a regression model. The CNN model is used to extract the data features of the defect (21), and the regression model is used to establish the relationship between the data features and the location of the defect (21). The ensemble CNN model is trained using the training set data. The test set data is used to verify the trained deep learning model. The results are compared with the results of the original dataset to determine whether the location and type of the defect (21) are accurate. If so, the proposed deep learning model is applied to other test pieces for scanning detection. Otherwise, the proposed deep learning model and the integrated CNN model should be optimized to improve the output accuracy.

2. The data-driven method for planning the ultrasonic guided wave multi-defect detection path of a robotic arm according to claim 1, characterized in that, The preset scanning path of the robotic arm (12) is as follows: The scanning direction is initially horizontal; once the preset scanning distance is reached, it then scans vertically. Distance; then, the scanning direction is the opposite of the horizontal direction; after reaching the preset scanning distance, scanning proceeds in the forward direction of the vertical direction. Distance; then, the scanning direction is adjusted back to the positive horizontal direction, and the above steps are repeated until a global scan is completed; where The scanning step distance should be less than or equal to the size of the test piece along the scanning direction, both in the forward and reverse directions.

3. The data-driven method for planning the ultrasonic guided wave multi-defect detection path of a robotic arm according to claim 1, characterized in that, The preset scanning path of the robotic arm (12) is as follows: The robotic arm (12) scans in a clockwise or counterclockwise direction, with the endpoint of the specified scanning area as the starting position and the scanning step distance as the starting position. Perform a horizontal forward scan with step increments. When the scan distance reaches the length of the specified scan area, turn 90 degrees to the right and continue scanning with step increments. A vertical reverse scan is performed to increment the step size. When the scan distance reaches the width of the specified scan area, the camera turns 90 degrees to the right for the second time to scan the step size. A horizontal reverse scan is performed to increment the step size. When the scan distance reaches the length of the specified scan area, the camera turns 90 degrees to the right for the third time to increment the step size. Vertical forward scanning is performed to increment the step size. A scan step is defined as the distance from the endpoint of the specified scan area to the scan distance. Then, it turns 90 degrees to the right for the fourth time and repeats the above trajectory, with the scanning distance before each turn decreasing by the scanning step size. ,in For the scan step, The value can be any number.

4. The data-driven method for planning the multi-defect detection path of ultrasonic guided waves for robotic arms according to claim 1, characterized in that, The preset scanning path of the robotic arm (12) is as follows: Within the specified scanning area, Y positions are randomly selected to test the test piece and the acquired signals are recorded; where Y is the number of selected positions and is any positive integer.

5. The data-driven method for planning the multi-defect detection path of ultrasonic guided waves for robotic arms according to claim 1, characterized in that, Deep learning models specifically include: input layer, convolutional layer, pooling layer, activation layer, and output layer; The input layer serves as the data entry point for the model, receiving raw data and converting it into a tensor format that the model can process, providing initial input for subsequent layers. The convolutional layer extracts local features from the input data using filters, capturing local correlations and reducing model parameters through parameter sharing, thus improving computational efficiency. The pooling layer samples the feature maps output by the convolutional layers, reducing feature dimensionality and computational cost by preserving the maximum or average values ​​of local regions, while enhancing the model's robustness to minor deformations in the input data. The activation layer primarily introduces nonlinear transformations to overcome the limitations of linear models, enabling the model to learn complex nonlinear relationships and improving its ability to express complex patterns. The output layer serves as the model's output, providing the final result based on the task type.

6. The data-driven method for planning the ultrasonic guided wave multi-defect detection path of a robotic arm according to claim 1, characterized in that, The test piece is a sheet metal structure or a large curved surface structure.

7. The data-driven method for planning the ultrasonic guided wave multi-defect detection path of a robotic arm according to claim 6, characterized in that, The test specimen is a large-size metal plate specimen (2), with specific dimensions of 2 m × 2 m × 3 mm.

8. The data-driven method for multi-defect detection path planning of ultrasonic guided waves for robotic arms according to claim 1, characterized in that, The types of defects (21) include: regular types and irregular types; regular types include: rectangles, circles and polygons.

9. The data-driven method for multi-defect detection path planning of ultrasonic guided waves for robotic arms according to any one of claims 1-8, characterized in that, The test systems to which this method is applicable include: a control and testing system (1), and the test piece; The control and testing system (1) includes: a computer (11), a robotic arm (12), an electromagnetic acoustic sensor (13), a signal generator / receiver (14), a power amplifier (15), and a signal amplifier (16); wherein the electromagnetic acoustic sensor (13) includes: a transmitting sensor (131) and a receiving sensor (132). The transmitting sensor (131) is used to emit the first ultrasonic guided wave (133); the receiving sensor (132) is used to receive the second ultrasonic guided wave (134). The computer (11) is connected to the signal generator / receiver (14) and the robotic arm (12) respectively; the signal generator / receiver (14) is connected to the power amplifier (15) and the signal amplifier (16) respectively; the electromagnetic acoustic sensor (13) is connected to the power amplifier (15), the signal amplifier (16), and the robotic arm (12) respectively; The computer (11) is used to control the operation of the signal generator / receiver (14) and the robotic arm (12); The robotic arm (12) is used to control the movement of the electromagnetic acoustic sensor (13); the electromagnetic acoustic sensor (13) is used to detect the test piece; The signal generator / receiver (14) is used to generate and / or receive signals; The power amplifier (15) is used to amplify the signal generated by the signal generator / receiver (14); The signal amplifier (16) is used to amplify the signal detected and returned by the electromagnetic acoustic sensor (13).

Images