Unmanned aerial vehicle-mounted X-ray flaw detection method

Abstract

The invention discloses an unmanned aerial vehicle-mounted X-ray flaw detection method, and relates to the technical field of X-ray flaw detection, and the method comprises the following steps: establishing a multi-source disturbance coupling dynamics model, and according to the multi-source disturbance coupling dynamics model, obtaining the current flight estimation state and interference prediction quantity of an unmanned aerial vehicle; based on the flight estimation state and the interference prediction quantity, obtaining a composite driving instruction; controlling the emission state of an X-ray emission source according to the attitude error, the angular velocity and the fuzzy length prediction of the multi-axis stabilized platform; when the emission state is started, the X-ray emission source emits a pulse to the to-be-detected equipment to obtain a real-time image of the to-be-detected equipment; and obtaining a flaw detection result of the to-be-detected equipment based on processing and defect identification of the real-time image. According to the method, the stability, imaging quality and real-time diagnosis capability of the unmanned aerial vehicle-mounted X-ray flaw detection system can be improved, the detection cost is reduced, and the service life of equipment is prolonged.

Description

Technical Field

[0001] This application generally relates to the field of X-ray flaw detection technology, and specifically to an unmanned aerial vehicle (UAV)-borne X-ray flaw detection method. Background Technology

[0002] Existing UAV-borne X-ray flaw detection technology faces several challenges in practical applications. First, the UAV platform is affected by external disturbances and its own vibrations during flight, making it difficult to maintain the stability of the flaw detection payload, which in turn affects the clarity and accuracy of X-ray imaging. Second, while traditional passive vibration isolation and long-exposure imaging strategies can partially suppress dynamic interference, they often come at the cost of inspection efficiency and are difficult to meet the needs of real-time diagnosis.

[0003] The aforementioned problems are even more pronounced in complex application scenarios such as flaw detection of high-altitude power transmission and transformation equipment. External factors such as wind load, electromagnetic interference, and structural vibration significantly impact the design and control of the UAV, making platform stabilization extremely difficult. Furthermore, high-quality imaging in dynamic environments requires precise coordination of multiple aspects, including platform stabilization and image processing, which places higher demands on the system's real-time performance and synchronization. Therefore, we propose an UAV-borne X-ray flaw detection method to address these issues. Summary of the Invention

[0004] In view of the above-mentioned defects or deficiencies in the prior art, it is desirable to provide an unmanned aerial vehicle (UAV) X-ray flaw detection method to solve the above problems.

[0005] This application provides an unmanned aerial vehicle (UAV)-borne X-ray flaw detection method. The flaw detection method is applied to a UAV-payload coupling system, which is built into the UAV. The UAV-payload coupling system is used to perform flaw detection on the equipment to be inspected. The method includes: A multi-source disturbance coupling dynamic model is established, and based on the multi-source disturbance coupling dynamic model, the current flight estimation state and disturbance prediction of the UAV are obtained; Based on the estimated flight state and predicted interference, the control commands of the multi-axis stabilization platform in the UAV-payload coupling system are modified to obtain composite drive commands; the multi-axis stabilization platform is equipped with an X-ray emission source, which is used to perform flaw detection imaging of the equipment to be inspected; The attitude error, angular velocity, and fuzzy length prediction of the multi-axis stabilization platform are obtained, and the emission state of the X-ray emission source is controlled based on the attitude error, angular velocity, and fuzzy length prediction. When the emission state is activated, the X-ray emission source emits pulses toward the device under test to obtain a real-time image of the device under test; based on the processing of the real-time image and defect identification, the flaw detection result of the device under test is obtained.

[0006] According to the technical solution provided in this application, based on the estimated flight state and predicted interference, the control commands of the multi-axis stabilization platform in the UAV-payload coupling system are modified to obtain composite drive commands, including: The difference between the estimated flight state and the desired flight state is obtained. Based on the difference, proportional-integral control is applied to the control command to obtain the feedback correction control command. The feedforward gain matrix is ​​used to compensate for the predicted disturbance to obtain the feedforward compensation control command. The composite drive command is calculated based on the feedforward compensation control command and the feedback correction control command.

[0007] According to the technical solution provided in this application, the emission state of the X-ray emission source is controlled based on the attitude error, angular velocity, and fuzzy length prediction, including: When the attitude error is less than or equal to a preset attitude error threshold, the angular velocity is less than or equal to a preset angular velocity threshold, and the fuzzy length prediction is less than or equal to a preset fuzzy length threshold, the emission state of the X-ray emission source is controlled to start.

[0008] According to the technical solution provided in this application, based on the processing of the real-time image and defect identification, the flaw detection result of the device to be inspected is obtained, including: The real-time image is acquired and preprocessed to obtain a set of potential defects of the device to be detected; the set of potential defects includes at least one type of defect. The confidence score of each potential defect type in the potential defect set is determined. If the confidence score of at least one defect type is greater than a preset judgment threshold, the corresponding structured defect entry information is generated as the flaw detection result. If there is no defect type with a confidence score greater than the preset judgment threshold, a real-time diagnostic report is generated as the flaw detection result based on all identified potential defects.

[0009] According to the technical solution provided in this application, the real-time image is acquired, and the real-time image is preprocessed to obtain a set of potential defects in the device to be detected, including: Receive the real-time image and preprocess the real-time image; The preprocessed real-time image is input into the defect recognition model, which is used to extract features from the real-time image. The defect identification model outputs a set of potential defects, which includes: preliminary classification information of potential defects, and the bounding box and confidence score corresponding to each type of potential defect.

[0010] According to the technical solution provided in this application, the method further includes: The imaging quality of real-time images is acquired, and an imaging diagnostic result is obtained based on the imaging quality; the imaging diagnostic result includes at least: control factor result and imaging factor result; Based on the obtained imaging diagnostic results, a first adjustment strategy corresponding to the control cause result or a second adjustment strategy corresponding to the imaging cause result is selected to adjust the imaging quality until the imaging quality meets the preset imaging requirements.

[0011] According to the technical solution provided in this application, controlling the emission state of the X-ray emission source to be activated includes: Under preset timing constraints, the emission state of the X-ray emission source is controlled to switch from off to on; wherein, the timing constraints are based on the X-ray pulse exposure time, the attitude change of the multi-axis stabilization platform at adjacent time points, and a preset attitude change threshold.

[0012] According to the technical solution provided in this application, the state-space expression of the multi-source disturbance coupled dynamics model is as follows: (1) Formula (1); Wherein, the state vector It includes platform location Attitude angle linear velocity angular velocity Electromagnetic interference status and field velocity components ; For control input; The perturbation vector; This is process noise; For measuring noise; It is a nonlinear state transition function. For the observation function, The disturbance effect matrix, This represents the external disturbance vector. This is the observation vector.

[0013] In summary, this technical solution specifically discloses an unmanned aerial vehicle (UAV) X-ray flaw detection method, including: establishing a multi-source disturbance coupling dynamic model, and obtaining the current flight estimation state and disturbance prediction quantity of the UAV based on the multi-source disturbance coupling dynamic model; modifying the control commands of the multi-axis stabilization platform in the UAV-payload coupling system based on the flight estimation state and disturbance prediction quantity to obtain composite drive commands; the multi-axis stabilization platform is equipped with an X-ray emission source, which is used to perform flaw detection imaging of the device under test; obtaining the attitude error, angular velocity, and fuzzy length prediction quantity of the multi-axis stabilization platform, and controlling the emission state of the X-ray emission source based on the attitude error, angular velocity, and fuzzy length prediction quantity; when the emission state is activated, the X-ray emission source emits pulses to the device under test to obtain a real-time image of the device under test; and obtaining the flaw detection result of the device under test based on the processing of the real-time image and defect identification.

[0014] In existing high-altitude power transmission and transformation equipment flaw detection, UAVs encounter numerous external factors that affect their flight status. Therefore, achieving high-quality imaging in dynamic environments requires overcoming even more challenges. This application establishes a dynamic model to predict and compensate for various disturbances during flight, obtaining composite drive commands to actively drive a multi-axis stabilized platform for reverse compensation. This cancels out external interference and lays the foundation for intelligently determining the optimal imaging timing, ensuring that X-ray emission is only initiated when the platform is in a highly stable state. Ultimately, this enhances the UAV's ability to acquire clear and accurate X-ray flaw detection images under complex conditions, enabling efficient and reliable identification of defects in the inspected equipment and improving the efficiency, accuracy, and intelligence level of industrial non-destructive testing. Attached Figure Description

[0015] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating an unmanned aerial vehicle (UAV)-borne X-ray flaw detection method. Detailed Implementation

[0016] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] Example 1 To make the technical solutions of the embodiments of this application clearer and easier to understand, the application background of the embodiments of this application is introduced below.

[0019] Existing UAV-borne X-ray flaw detection technology faces several challenges in practical applications. First, the UAV platform is affected by external disturbances and its own vibrations during flight, making it difficult to maintain stability of the flaw detection payload, thus affecting the clarity and accuracy of X-ray imaging. Second, while traditional passive vibration isolation and long-exposure imaging strategies can partially suppress dynamic interference, they often come at the cost of inspection efficiency and are insufficient for real-time diagnostic needs. Furthermore, the offline image processing methods commonly used in existing technologies suffer from computational delays, failing to meet the requirements of rapid diagnostics in industrial settings and limiting the technology's practical application value.

[0020] In complex applications such as flaw detection of high-altitude power transmission and transformation equipment, the aforementioned problems are even more pronounced. The coupling effect of multiple sources of disturbance, such as wind load, electromagnetic interference, and structural vibration, makes the design and control of a stable platform extremely difficult. Simultaneously, high-quality imaging in dynamic environments requires precise coordination of multiple aspects, including platform stability, exposure timing, and image processing, which places higher demands on the system's real-time performance and synchronization. Furthermore, traditional methods often suffer from low efficiency and high safety risks when handling inspection tasks in high-altitude, hazardous, or inaccessible areas.

[0021] In view of this, this application proposes an unmanned aerial vehicle (UAV)-borne X-ray flaw detection method, which includes: establishing a multi-source disturbance coupling dynamic model, and obtaining the current flight estimation state and disturbance prediction quantity of the UAV based on the multi-source disturbance coupling dynamic model; modifying the control command of the multi-axis stabilization platform in the UAV-payload coupling system based on the flight estimation state and disturbance prediction quantity to obtain a composite drive command; the multi-axis stabilization platform is equipped with an X-ray emission source, which is used to perform flaw detection imaging of the device under test; obtaining the attitude error, angular velocity, and fuzzy length prediction quantity of the multi-axis stabilization platform, and controlling the emission state of the X-ray emission source based on the attitude error, angular velocity, and fuzzy length prediction quantity; when the emission state is activated, the X-ray emission source emits pulses to the device under test to obtain a real-time image of the device under test; and obtaining the flaw detection result of the device under test based on the processing of the real-time image and defect identification. Therefore, this application aims to achieve high-precision and stable control of an unmanned aerial vehicle (UAV)-borne X-ray flaw detection system in complex dynamic environments, ensuring the acquisition of high-quality X-ray images. It establishes a coupled dynamic model capable of sensing and predicting multi-source disturbances such as wind force and fuselage sway, and generates composite drive commands to actively drive a multi-axis stabilization platform for reverse compensation, thereby physically canceling interference. Furthermore, it determines the optimal imaging timing by monitoring the platform's attitude error, angular velocity, and fuzzy length prediction, ensuring that X-ray emission can be initiated the instant the platform is in a highly stable state. This application organically integrates the above technologies to construct a highly autonomous and intelligent UAV-borne flaw detection system adaptable to various complex working conditions, enabling precise and efficient automated non-destructive testing operations in complex application scenarios such as flaw detection of high-altitude power transmission and transformation equipment, thereby effectively improving the efficiency, accuracy, and intelligence level of industrial non-destructive testing.

[0022] To make the technical solution of this application clearer and easier to understand, the flaw detection method provided in the embodiments of this application will be described below with reference to the accompanying drawings. Figure 1 As shown, this figure is a flowchart of a flaw detection method provided in an embodiment of this application. This method is applied to a drone or a drone-payload coupling system, wherein the drone-payload coupling system (used for flaw detection of the device to be inspected) is built into the drone. Specifically, the UAV-payload coupling system in this embodiment mainly consists of the following parts: a multi-axis stabilization platform, an X-ray emission source, an image acquisition device, an airborne FPGA processing unit, and a vision-inertial fusion sensing system. The multi-axis stabilization platform isolates the UAV's attitude sway, providing a stable pointing and platform for the imaging device. The X-ray emission source can be a pulsed X-ray source, used to generate brief, high-energy X-ray pulses to penetrate the object under test for imaging. The image acquisition device receives the X-rays after they penetrate the object and converts them into digital image signals. The airborne FPGA processing unit has a built-in dedicated convolutional neural network model for real-time intelligent analysis of the acquired images. The vision-inertial fusion sensing system provides the UAV with high-precision self-positioning, attitude, and three-dimensional perception of the surrounding environment. Finally, the aforementioned dynamic freeze-frame imaging describes the ability to obtain clear X-ray images, as if captured in a static state, by utilizing the X-ray emission source and the multi-axis stabilization platform in a coordinated manner, even when both the UAV and the target may be in motion.

[0023] Furthermore, the following example uses a UAV-payload coupling system as the processing device to illustrate this method, which includes the following steps: S100. Establish a multi-source disturbance coupling dynamic model, and based on the multi-source disturbance coupling dynamic model, obtain the current flight estimation state and disturbance prediction of the UAV. Since the embodiments of this application are mainly aimed at the task of flaw detection of equipment under test in the high-altitude operation environment of power transmission and transformation, in order to avoid interference from many external factors, a multi-source disturbance dynamic model is first established. This model considers the nonlinear coupling characteristics of wind load, electromagnetic interference and structural vibration, and its state space expression is as follows (1): Formula (1); Wherein, the state vector It includes platform location Attitude angle linear velocity angular velocity Electromagnetic interference status and field velocity components ; For control input; The perturbation vector; This is process noise; For measuring noise; It is a nonlinear state transition function. For the observation function, The disturbance effect matrix, This represents the external disturbance vector. This is the observation vector.

[0024] It is evident that a multi-source disturbance dynamics model can integrate various disturbance factors such as wind resistance, load sway, and self-vibration during UAV flight, clarifying the coupling relationship between each disturbance. In applications, the multi-source disturbance dynamics model can be used to calculate the current flight state of the UAV (such as position, speed, and attitude) and the upcoming disturbances (such as airflow change prediction) in real time, allowing the UAV-load coupling system to grasp the dynamic environment in advance, so that the multi-axis stabilization platform can adjust its attitude in time, thereby accurately performing flaw detection on the equipment to be inspected.

[0025] Based on the above dynamic model, in order to facilitate timely attitude adjustment of the multi-axis stabilized platform, this application embodiment designs a predictive feedforward compensation controller. This predictive feedforward compensation controller uses an extended Kalman filter (EKF) to fuse visual and inertial data to realize state estimation and disturbance prediction of the multi-axis stabilized platform. For details, please refer to the following formula (2): Formula (2); in, For time step Based on the previous time step The prior flight state is obtained from the information. The prior error covariance matrix; In time step Control inputs applied to the system; The perturbation input matrix; In time step The estimated amount of disturbance prediction; This is the transpose of the state transition Jacobian matrix; The process noise covariance matrix; Let be the state transition Jacobian matrix.

[0026] S200: Based on the estimated flight state and predicted interference, the control commands of the multi-axis stabilization platform in the UAV-payload coupling system are modified to obtain composite drive commands; the multi-axis stabilization platform is equipped with an X-ray emission source, which is used to perform flaw detection imaging of the equipment to be inspected; When drones operate at high altitudes, the X-ray emission source they carry can shift or vibrate due to airflow disturbances, their own attitude adjustments, and load swaying, resulting in blurred flaw detection images and missed lesions. The core function of a multi-axis stabilization platform is to isolate the dynamic interference from the drone, ensuring that the X-ray emission source is always aligned with the equipment being inspected (such as insulators and welds in power transmission lines). To counteract the dynamic interference from the drone, this application employs a method of modifying control commands to obtain composite drive commands, thereby addressing both predictable interference and sudden deviations.

[0027] Therefore, in order to counteract the dynamic interference of the UAV and ensure that the X-ray emission source can remain stable on the multi-axis stabilization platform, the predictive feedforward compensation controller designed based on the embodiments of this application can obtain the control law of the multi-axis stabilization platform as shown in the following formula (3): Formula (3); in, This is a compound driver instruction; This is the feedforward gain matrix; These are the proportional and integral gain matrices, respectively; For interference prediction; Desired flight state; The prior flight estimated state; Therefore, based on the application of the control law, and based on the estimated flight state and the predicted disturbance, the process of modifying the control commands of the multi-axis stabilization platform in the UAV-payload coupled system to obtain the composite drive command is as follows: obtain the difference between the estimated flight state and the desired flight state; perform proportional-integral control on the control commands based on the difference to obtain the feedback correction control command; use the feedforward gain matrix to compensate for the predicted disturbance to obtain the feedforward compensation control command; and calculate the composite drive command based on the feedforward compensation control command and the feedback correction control command.

[0028] Specifically, as can be seen from formula (3), the predictive feedforward compensation controller corrects the control command of the multi-axis stabilized platform in two ways: feedforward modification and feedback correction. Feedforward correction uses a multi-source disturbance coupled dynamic model and input data to predict the disturbance motion of the multi-axis stabilized platform in the next control cycle. The feedforward gain matrix is ​​used to compensate for the predicted disturbance in advance so as to offset the predictable disturbance. Feedback correction is performed by proportional-integral control. The difference between the expected flight state of the multi-axis stabilized platform and the actual predicted flight estimate state is obtained to obtain the error between the two, i.e., the residual error. The proportional gain matrix directly generates the correction amount based on the current error, while the integral gain matrix is ​​used to accumulate historical errors and eliminate long-term small deviations. Finally, this embodiment of the application achieves high-precision control of the multi-axis stabilized platform by first using feedforward to offset the predictable disturbance and then using feedback to correct the residual error.

[0029] In addition, in a preferred embodiment, the extended Kalman filter can be replaced with an unscented Kalman filter (UKF) or an H∞ robust filter in relation to the structure of the predictive feedforward compensation controller.

[0030] For example, if an unscented Kalman filter is used instead, then in formula (2) It can also be rewritten as the following formula (4): Formula (4); in, These are the sigma points generated through the UT transformation. For the corresponding weighting coefficients, this method has better estimation accuracy in strongly nonlinear scenarios; If replaced with an H∞ robust filter, then in formula (2) It can also be rewritten as the following formula (5): Formula (5); in, This is the measurement matrix, used to map the system state vector to the measurement space; for The measurement vector at time, i.e., the system at... The actual observation data collected at any given time; This is the Kalman gain matrix.

[0031] In a preferred embodiment, other methods can also be used for the control law of the multi-axis stabilized platform, such as using a model predictive control (MPC) architecture or a sliding mode control (SMC) structure for control; In the MPC architecture, the control input can be solved through rolling time-domain optimization, as shown in the following formula (6): Formula (6); in, The control input sequence to be optimized. To predict the time domain length (i.e., to consider the state and control over the next N time steps during optimization); For the system in the future The actual state of the step; To control the weighted norm of the input; It is the weighted norm of the state tracking error.

[0032] Based on formula (6), in order for the optimization result to conform to the physical laws of the system, it also needs to satisfy the system dynamics constraints, which are the following formula (7): Formula (7); in, For the system in the future The state at each time step is used to recursively describe the system state in the time domain. Let this be the system's state transition function; For the system in the future The actual state of the step; For the system in the future Step control input.

[0033] Specifically, under the sliding mode control (SMC) structure, please refer to the following formula (8): Formula (8); Among them, state error It is used to characterize the error between the desired attitude and the actual attitude; Represented as a sliding surface; These are design parameters used to adjust the weights of the integral term, ensuring finite-time convergence through the convergence law and enhancing the ability to resist disturbances.

[0034] S300: Obtain the attitude error, angular velocity, and fuzzy length prediction of the multi-axis stabilized platform, and control the emission state of the X-ray emission source based on the attitude error, angular velocity, and fuzzy length prediction. To avoid blurry X-ray imaging caused by instability of the multi-axis stabilization platform, in addition to controlling its state, the timing of X-ray emission is also extremely important. In this embodiment, the stability of the multi-axis stabilization platform and the imaging effect are judged by real-time monitoring of the changes in the attitude error, angular velocity and blur length prediction of the multi-axis stabilization platform, thereby selecting the timing to control and adjust the emission state of the X-ray emission source; here, the emission state can be defined as the start state and the shut-off state.

[0035] Specifically, the imaging triggering conditions for the X-ray emission source are designed as follows (8): Formula (8); in, This is an imaging triggering condition, meaning that the emission state of the X-ray emission source is the condition for activation; "Conditional selection" indicates either "if" or "condition". This refers to attitude error; The preset attitude error threshold is used; Angular velocity; The preset angular velocity threshold; For fuzzy length prediction; Set a preset fuzzy length threshold; The opposite of the above conditions; 1 and 0 represent the emission state of the X-ray emission source, with 1 indicating that the X-ray emission source is in the activated state and 0 indicating that the X-ray emission source is in the deactivated state.

[0036] Furthermore, based on the above formula (8), the attitude error, angular velocity and fuzzy length prediction of the multi-axis stabilization platform are obtained, and the specific process of controlling the emission state of the X-ray emission source based on the attitude error, angular velocity and fuzzy length prediction is as follows: when the attitude error is less than or equal to the preset attitude error threshold, the angular velocity is less than or equal to the preset angular velocity threshold, and the fuzzy length prediction is less than or equal to the preset fuzzy length threshold, the emission state of the X-ray emission source is controlled to start.

[0037] Attitude refers to the three-dimensional spatial attitude of the multi-axis stabilized platform of the UAV payload, including roll angle, pitch angle, and yaw angle. The actual attitude can be obtained by fusing the roll angle, pitch angle, and yaw angle through the inertial measurement unit (IMU) and the vision sensor. Angular velocity refers to the angular motion rate of the multi-axis stabilized platform. The actual angular velocity corresponds to the three-axis angular velocity components measured by the IMU, namely roll angular velocity, pitch angular velocity, and yaw angular velocity. Blur length prediction refers to the degree of image blur caused by the attitude disturbance and translation speed of the multi-axis stabilized platform within the exposure time window. It is obtained by predicting the motion trajectory of pixels in the field of view during the imaging process and reflects the estimated index of image sharpness.

[0038] It should be noted that attitude error refers to the difference between the actual attitude and the desired attitude, while angular velocity refers to the difference between the actual angle and the desired angular velocity. The desired attitude, desired angular velocity, and the aforementioned desired flight state are all directly issued by the UAV's upper-level control system (such as the system's mission planning and control center). The preset attitude error threshold, preset angular velocity threshold, and preset fuzzy length threshold can all be set according to the actual situation, and no special restrictions are imposed here.

[0039] In a preferred embodiment, in addition to the imaging triggering conditions described above, the embodiments of this application may be replaced with fuzzy logic control or a triggering strategy based on multi-objective optimization.

[0040] For example, taking the use of a fuzzy controller to fuzzify attitude error and angular velocity as an example, its triggering rule can be expressed as the following formula (9): Formula (9); in, These are logical functions. For example, when the input is below or above a certain threshold, it returns "true", that is, it returns "true" when the condition is met. This logical function is used to directly determine whether the attitude error, angular velocity and fuzzy length prediction of the multi-axis stabilization platform meet the set conditions. Let be the norm of the attitude error of the multi-axis stabilized platform; Let be the norm of the angular velocity of the multi-axis stabilized platform.

[0041] Alternatively, the NSGA-II multi-objective optimization algorithm can be used, with imaging quality indicators and energy consumption as optimization objectives, to dynamically adjust the combination of threshold parameters. That's fine; no special restrictions are imposed here.

[0042] Next, to achieve precise coordination of multi-axis stabilization platform stability, imaging triggering, and computational processing, this application embodiment also designs a timing coordination mechanism. This mechanism is used to ensure that the X-ray pulse is triggered within the time window of attitude stabilization and that image acquisition and processing are completed within milliseconds. Correspondingly, the process of controlling the emission state of the X-ray emission source to be on also includes: under a preset timing constraint, controlling the emission state of the X-ray emission source to switch from off to on; wherein, the timing constraint is based on the X-ray pulse exposure time, the attitude change of the multi-axis stabilization platform at adjacent times, and a preset attitude change threshold.

[0043] Specifically, the timing constraint expression is shown in the following formula (10): Formula (10); in, For multi-axis stabilization platform moments t The attitude angle vectors include roll, pitch, and yaw angles; The time interval corresponding to the attitude change; This refers to the X-ray pulse exposure time; The preset attitude change threshold is defined as follows: the attitude change of the multi-axis stabilized platform at adjacent time points is defined as... .

[0044] S400. When the emission state is activated, the X-ray emission source emits pulses towards the device under test to obtain a real-time image of the device under test; based on the processing of the real-time image and defect identification, the flaw detection result of the device under test is obtained.

[0045] Specifically, once the attitude error, angular velocity, and fuzzy length prediction all meet the set conditions, the X-ray emission source will start according to the set timing sequence, and the X-ray emission source will instantaneously emit X-rays to the device under test for image acquisition. This ensures that the X-ray emission timing is when the attitude of the multi-axis stabilized platform is stable, thus guaranteeing the quality of the acquired real-time images.

[0046] Furthermore, based on the processing of real-time images and defect identification, the flaw detection results of the device to be inspected are obtained by the following steps: Step A1: Acquire real-time images and preprocess the real-time images to obtain a set of potential defects of the device to be inspected; the set of potential defects includes at least one type of defect. First, in the flaw detection of high-altitude power transmission and transformation equipment, obtaining high-quality real-time images is the foundation for ensuring the accuracy of flaw detection results. After obtaining the corresponding real-time images, it is also necessary to preprocess the real-time images to further improve the imaging quality of the real-time images, provide favorable conditions for initially obtaining the set of potential defects, and accelerate the defect identification speed.

[0047] Furthermore, the specific process of step A1 includes: receiving a real-time image and preprocessing the real-time image; inputting the preprocessed real-time image into a defect recognition model, which is used to extract features from the real-time image; and outputting a set of potential defects from the defect recognition model, which includes: preliminary classification information of potential defects, and bounding boxes and confidence scores corresponding to each type of potential defect.

[0048] It should be explained in detail that after receiving the real-time image generated after the X-ray emission source penetrates the device under test, the preprocessing of the real-time image includes, but is not limited to, noise reduction, contrast enhancement, and ROI extraction, etc., which are not specifically limited here. Then, the preprocessed real-time image can be directly input into the airborne FPGA processing unit for processing. The airborne FPGA processing unit has a built-in dedicated defect recognition model (such as a CNN model). The CNN model can perform forward inference and finally output the preliminary defect analysis results, i.e., the potential defect set, based on which efficient feature extraction and defect pattern matching are achieved.

[0049] In practical applications, the following information is used: Preliminary classification information: This is used to label the type of each potential defect (such as cracks, voids, inclusions, incomplete penetration, etc.), determined based on the matching results of high-level features extracted by the model and training samples; Bounding boxes: Rectangular boxes are used to mark the specific location and range of potential defects in the image, making it convenient for staff to quickly locate the corresponding area of ​​the defect on the equipment; Confidence score: A value of 0-1 or 0-100 is used to represent the model's confidence level in "this area is a certain type of defect" (e.g., a confidence score of 0.95 means a 95% probability of a crack). Generally, a confidence threshold (e.g., 0.7) is set, and results below the threshold are filtered out to reduce misjudgments.

[0050] Step A2: Determine the confidence score of each potential defect type within the potential defect set. If the confidence score of at least one defect type is greater than the preset judgment threshold, generate the corresponding structured defect entry information as the flaw detection result. If there is no defect type with a confidence score greater than the preset judgment threshold, generate a real-time diagnostic report as the flaw detection result based on all identified potential defects.

[0051] Following on from the previous point, the potential defects within the potential defect set carry corresponding confidence scores. Therefore, by judging the confidence scores of various defect types, real defects can be generated. Specifically, if the confidence scores of one or more defect types are greater than a preset judgment threshold, the system can directly generate structured defect entry information as the flaw detection result. For example, among potential defects, cracks, voids, and inclusions have confidence scores of 95, 97, and 98 respectively, and the system's preset judgment threshold is 95. The final generated structured defect entry information will then include relevant information about cracks, voids, and inclusions. This relevant information includes at least: image slices, coordinates, type, size, and grade, used for quickly filtering potential defects. Of course, if no defect type with a confidence score greater than the preset judgment threshold is found, then based on all identified potential defects, a real-time diagnostic report can be generated as the flaw detection result. This real-time diagnostic report can be provided to technicians for further investigation.

[0052] It should be noted that the relevant information in the structured defect entry information is also obtained by the defect identification model. The specific process is as follows: after obtaining the processed real-time image, the defect identification model performs precise calculations of geometric parameters and grade assessments to ensure the quantification and classification of defect characteristics. Specifically, the system quantitatively calculates and classifies defects into different grades based on key parameters such as defect size (length, width), shape characteristics, spatial distribution, and potential impact on structural integrity, referring to industry non-destructive testing standards (such as DL / T or relevant ASTM standards). It is also evident that the clarity of the real-time image is particularly important in this stage. Only by obtaining high-quality, low-blur X-ray images under dynamic freeze-frame imaging can the boundary, size, and geometric morphology of defects be accurately extracted; otherwise, the grading results will be biased. Furthermore, accurate grade assessment has direct significance for subsequent engineering decisions (such as whether immediate repair is needed or whether operational safety is affected).

[0053] As can be seen, this method achieves a closed-loop processing chain from high-resolution image acquisition, intelligent recognition, reliable judgment to diagnostic output, effectively improving the automation level of defect detection and the real-time performance and reliability of diagnosis. Furthermore, this embodiment employs a lightweight convolutional neural network model embedded in an FPGA for real-time image processing and defect recognition. The network structure is optimized through channel pruning and quantization-aware training to meet the real-time requirements of edge computing, and its processing latency satisfies: .

[0054] Alternatively, for the airborne real-time intelligent diagnostic architecture, the airborne FPGA processing unit in this embodiment can also adopt a SoC (System-on-Chip) architecture or a lightweight model design based on Transformer. Specifically, by using a heterogeneous SoC to integrate FPGA and GPU acceleration cores, its processing latency constraints can be reconfigured as follows: .

[0055] In a preferred embodiment, the imaging quality of a real-time image is acquired, and an imaging diagnostic result is obtained based on the imaging quality; the imaging diagnostic result includes at least: control factor result and imaging factor result; Based on the obtained imaging diagnostic results, a first adjustment strategy corresponding to the control cause result or a second adjustment strategy corresponding to the imaging cause result is selected to adjust the imaging quality until the imaging quality meets the preset imaging requirements.

[0056] In this embodiment, the imaging quality of the real-time image is continuously optimized, forming an adaptive closed-loop control method for imaging quality feedback. Specifically, after a complete stabilization-imaging-analysis process is executed in the UAV-payload coupled system, the imaging quality of the real-time image needs to be directly obtained. The imaging quality can be obtained by extracting and evaluating its imaging quality indicators, which include, but are not limited to, signal-to-noise ratio, sharpness, and average confidence level. After obtaining the imaging quality, it can be compared with the optimal quality threshold to obtain the imaging diagnosis result, which can be divided into meeting or failing the standard.

[0057] Here, the calculation of image quality can be performed by weighted summation based on various image quality indicators. First, each indicator needs to be standardized to eliminate the influence of dimensions. Then, according to the influence of each indicator on image quality, corresponding weights are assigned. Finally, the corresponding data of each indicator obtained from the analysis are weighted and calculated to obtain a value that characterizes the image quality. Relatedly, the optimal quality threshold can also be set as a set value by technicians according to engineering requirements.

[0058] Next, if the imaging diagnosis result shows that it does not meet the standard, it means that the imaging effect of the system is poor at this time, and then adaptive adjustment is required. This application embodiment includes two adjustment strategies (first adjustment strategy and second adjustment strategy), and the two adjustment strategies correspond to different situations, making the imaging effect adjustment more targeted and faster.

[0059] This section explains the first and second adjustment strategies. When the imaging diagnostic results show that the quality is not up to standard, the system needs to determine whether the quality degradation is caused by the excessive residual of the stability control, and accordingly distinguish which adjustment strategy to implement. The judgment criteria are as follows: when the overall brightness of the image is insufficient or the contrast is low, and the signal-to-noise ratio (SNR) is lower than the preset threshold (e.g., 30dB), but the attitude error and angular velocity are still within the threshold range, it is determined that the imaging parameters are insufficient. This usually means that the pulse width is too short or the exposure energy is insufficient. At this time, it is necessary to send a command to the X-ray source to adjust the pulse energy or duration, which is equivalent to implementing the second adjustment strategy. When the image brightness and contrast are normal, but the edge contours are blurred and the sharpness value is low, and the attitude error or angular velocity exceeds the corresponding threshold, it is determined that the stability control accuracy is insufficient. This means that the multi-axis stabilization platform has not maintained sufficient stability within the exposure window. At this time, it is necessary to send a command to the stability control system to correct the issue by adjusting the prediction compensation weight or feedback gain, which is equivalent to implementing the first adjustment strategy.

[0060] Based on the aforementioned distinguishing criteria, the system can accurately identify the root cause of image quality degradation and select different control paths, thereby achieving adaptive closed-loop optimization and improving the reliability of system imaging and diagnosis. When insufficient imaging parameters are detected during diagnosis, instructions are sent to the X-ray source to adaptively adjust the pulse width or exposure energy; when insufficient stabilization control precision is confirmed, instructions are sent to the stabilization control system to adjust control parameters, including increasing prediction weights or improving feedback gain; simultaneously, the system re-executes the task according to the adjusted parameters until the real-time image quality reaches the optimal quality threshold. If the quality meets the requirements, the process enters an end or standby state, awaiting subsequent instructions. This method, through bidirectional feedback control of imaging quality and control precision, achieves adaptive optimization of the imaging and stabilization stages, effectively improving the reliability and intelligence level of the system imaging.

[0061] Based on the above description, this application proposes an unmanned aerial vehicle (UAV) X-ray flaw detection method, comprising: firstly, establishing a multi-source disturbance coupled dynamic model to obtain the current flight estimation state and disturbance prediction of the UAV; secondly, performing proportional-integral control based on the difference between the flight estimation state and the expected flight state to obtain a feedback correction control command, and simultaneously using a feedforward gain matrix to compensate for the disturbance prediction to obtain a feedforward compensation control command; and thirdly, combining the two to calculate a composite drive command for correcting the control command of the multi-axis stabilized platform; subsequently, monitoring the attitude error, angular velocity, and fuzzy length prediction of the multi-axis stabilized platform, and when all three are not greater than the corresponding preset thresholds, controlling the X-ray emission source to start and emit pulses under timing constraints to obtain the equipment under inspection. The system obtains real-time images; after preprocessing the real-time images, it inputs them into the defect recognition model. After feature extraction, it outputs a set of potential defects containing preliminary classification information, bounding boxes, and confidence scores. If there are defect types with confidence scores greater than a preset judgment threshold, structured defect entry information is generated as the flaw detection result; otherwise, a real-time diagnostic report is generated based on all potential defects as the flaw detection result. In addition, the system also obtains the imaging quality of the real-time images to obtain imaging diagnostic results containing control inducement results (caused by the platform control link) and imaging inducement results (caused by the imaging or X-ray emission link). The corresponding first adjustment strategy or second adjustment strategy is selected to adjust the imaging quality until the preset imaging requirements are met.

[0062] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for X-ray flaw detection carried by an unmanned aerial vehicle (UAV), characterized in that, The flaw detection method is applied to a UAV-payload coupling system, which is built into the UAV and used to perform flaw detection on the equipment to be inspected. The method includes: A multi-source disturbance coupling dynamic model is established, and based on the multi-source disturbance coupling dynamic model, the current flight estimation state and disturbance prediction of the UAV are obtained; Based on the estimated flight state and predicted interference, the control commands of the multi-axis stabilization platform in the UAV-payload coupling system are modified to obtain composite drive commands; the multi-axis stabilization platform is equipped with an X-ray emission source, which is used to perform flaw detection imaging of the equipment to be inspected; The attitude error, angular velocity, and fuzzy length prediction of the multi-axis stabilization platform are obtained, and the emission state of the X-ray emission source is controlled based on the attitude error, angular velocity, and fuzzy length prediction. When the emission state is activated, the X-ray emission source emits pulses toward the device under test to obtain a real-time image of the device under test; based on the processing of the real-time image and defect identification, the flaw detection result of the device under test is obtained.

2. The UAV-borne X-ray flaw detection method according to claim 1, characterized in that, Based on the estimated flight state and predicted disturbance, the control commands of the multi-axis stabilization platform within the UAV-payload coupled system are modified to obtain composite drive commands, including: The difference between the estimated flight state and the desired flight state is obtained. Based on the difference, proportional-integral control is applied to the control command to obtain the feedback correction control command. The feedforward gain matrix is ​​used to compensate for the predicted disturbance to obtain the feedforward compensation control command. The composite drive command is calculated based on the feedforward compensation control command and the feedback correction control command.

3. The UAV-borne X-ray flaw detection method according to claim 1, characterized in that, Based on the attitude error, angular velocity, and fuzzy length prediction, the emission state of the X-ray emission source is controlled, including: When the attitude error is less than or equal to a preset attitude error threshold, the angular velocity is less than or equal to a preset angular velocity threshold, and the fuzzy length prediction is less than or equal to a preset fuzzy length threshold, the emission state of the X-ray emission source is controlled to start.

4. The UAV-borne X-ray flaw detection method according to claim 1, characterized in that, Based on the processing of the real-time image and defect identification, the flaw detection results of the device under inspection are obtained, including: The real-time image is acquired and preprocessed to obtain a set of potential defects of the device to be detected; the set of potential defects includes at least one type of defect. The confidence score of each potential defect type in the potential defect set is determined. If the confidence score of at least one defect type is greater than a preset judgment threshold, the corresponding structured defect entry information is generated as the flaw detection result. If there is no defect type with a confidence score greater than the preset judgment threshold, a real-time diagnostic report is generated as the flaw detection result based on all identified potential defects.

5. The UAV-borne X-ray flaw detection method according to claim 4, characterized in that, The real-time image is acquired and preprocessed to obtain a set of potential defects in the device to be detected, including: Receive the real-time image and preprocess the real-time image; The preprocessed real-time image is input into the defect recognition model, which is used to extract features from the real-time image. The defect identification model outputs a set of potential defects, which includes: preliminary classification information of potential defects, and the bounding box and confidence score corresponding to each type of potential defect.

6. The UAV-borne X-ray flaw detection method according to claim 1, characterized in that, The method further includes: The imaging quality of real-time images is acquired, and an imaging diagnostic result is obtained based on the imaging quality; the imaging diagnostic result includes at least: control factor result and imaging factor result; Based on the obtained imaging diagnostic results, a first adjustment strategy corresponding to the control cause result or a second adjustment strategy corresponding to the imaging cause result is selected to adjust the imaging quality until the imaging quality meets the preset imaging requirements.

7. The UAV-borne X-ray flaw detection method according to claim 1, characterized in that, Controlling the emission state of the X-ray emission source to "on" includes: Under preset timing constraints, the emission state of the X-ray emission source is controlled to switch from off to on; wherein, the timing constraints are based on the X-ray pulse exposure time, the attitude change of the multi-axis stabilization platform at adjacent time points, and a preset attitude change threshold.

8. The UAV-borne X-ray flaw detection method according to claim 1, characterized in that, The state-space expression of the multi-source disturbance coupled dynamics model is as follows: (1) Formula (1); Wherein, the state vector It includes platform location Attitude angle linear velocity angular velocity Electromagnetic interference status and field velocity components ; For control input; The perturbation vector; This is process noise; For measuring noise; It is a nonlinear state transition function. For the observation function, The disturbance effect matrix, This represents the external disturbance vector. This is the observation vector.

Images