Robot nondestructive testing method and system based on artificial intelligence

Abstract

The invention relates to the technical field of robots, and discloses a robot nondestructive testing method and system based on artificial intelligence, and the method comprises the steps: obtaining multi-modal sensing data and motion state parameters; constructing a three-dimensional material feature model and analyzing to generate an initial defect feature evaluation model; loading material attributes and detection precision constraints to generate a comprehensive model, and combining motion parameter simulation to obtain path regulation and control data; training the dynamic defect prediction model to obtain a detection path correction model and generating initial detection parameters; obtaining dynamic correction simulation information based on the initial parameters and the like; constructing a multi-objective optimization model, and optimizing to obtain an optimal detection path parameter; and generating a safety detection probability by combining real-time model reasoning, and adjusting a detection strategy to realize dynamic path matching. The system comprises a data acquisition module, a feature modeling and analysis module, a comprehensive evaluation and simulation module and the like. The method improves the detection precision, efficiency and adaptability, and is suitable for nondestructive detection of complex targets.

Description

Technical Field

[0001] The present invention relates to the technical field of robots, and specifically to a non-destructive testing method and system for robots based on artificial intelligence. Background Art

[0002] In the fields of industrial manufacturing, aerospace, energy and transportation, etc., non-destructive testing of materials and components is a key link to ensure product quality and safe operation. Traditional non-destructive testing methods, such as manual visual inspection, ultrasonic testing, ray testing, etc., have problems such as low detection efficiency, large human errors, limited detection range, and insufficient adaptability to complex structures. With the development of industrial automation and intelligence, higher requirements are put forward for the accuracy, efficiency, automation level of non-destructive testing technology and the detection ability for complex targets.

[0003] Although the existing robot non-destructive testing technology has realized automated detection to a certain extent, it still faces many challenges. On the one hand, single-modal sensing data is difficult to comprehensively and accurately reflect the internal defect characteristics of materials, resulting in limited accuracy of defect identification and evaluation. On the other hand, the detection path planning is often based on fixed rules or simple algorithms, and cannot be adaptively adjusted according to the real-time changes of material properties, the dynamic characteristics of defect distribution, and the motion state of the robot, which is prone to problems such as detection blind spots, repeated detections, or unstable detection accuracy. In addition, traditional methods lack effective means for feature fusion and analysis when dealing with multi-dimensional and multi-source heterogeneous data, and it is difficult to construct accurate material feature models and defect evaluation models, thereby affecting the scientificity and effectiveness of detection strategies.

[0004] With the rapid development of artificial intelligence technologies, such as machine learning, deep learning, intelligent optimization algorithms, etc., new ideas and methods are provided to solve the above problems. Combining artificial intelligence technology with robot non-destructive testing is expected to achieve efficient processing and analysis of multi-modal sensing data, construct more accurate material and defect models, realize intelligent planning and dynamic adjustment of detection paths, and improve the automation level of detection and the reliability of detection results. However, at present, the robot non-destructive testing technology based on artificial intelligence is still in the development stage, and there are still many technical problems to be solved in aspects such as effectively fusing multi-modal sensing data, constructing a dynamic defect evaluation model, realizing intelligent optimization of detection paths, and adjusting detection strategies according to real-time feedback.

[0005] Therefore, there is an urgent need for a non-destructive testing method and system for robots based on artificial intelligence to overcome the deficiencies of the existing technology, improve the accuracy, efficiency and adaptability of non-destructive testing, and meet the needs of modern industry for high-quality non-destructive testing. Summary of the Invention

[0006] The object of the present invention is to provide a non-destructive testing method and system for robots based on artificial intelligence to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: A non-destructive testing method for robots based on artificial intelligence, the method includes: Obtain multi-modal sensing data of the target to be detected and the motion state parameters of the end effector of the robot; Perform multi-dimensional scanning on the target and construct a three-dimensional material feature model, and perform dynamic defect feature analysis on the three-dimensional material feature model in combination with multi-modal sensing data to generate an initial defect feature evaluation model; Load the material property distribution data and detection accuracy constraint conditions into the initial defect feature evaluation model to generate a comprehensive defect feature evaluation model, and perform dynamic simulation of the detection path in combination with the motion state parameters to obtain path regulation simulation data; Construct a dynamic defect prediction model and train it through the path regulation simulation data to obtain a detection path correction model, and then generate initial detection parameters, where the initial detection parameters at least include initial detection point coordinates, initial scanning path, and initial correction response parameters; Based on the initial detection parameters, the comprehensive defect feature evaluation model, and the motion state parameters, obtain dynamic correction simulation information; Combine the dynamic correction simulation information with the detection task planning parameters, construct a multi-objective optimization model and iteratively optimize the planning parameters to obtain the optimal detection path parameters; Infer the detection path parameters of the current period based on the real-time material feature model to generate the safety detection probability of the current period, and combine the optimal detection path parameters of the current period with the actual scanning trajectory to adjust the real-time detection strategy of the robot to achieve the goal of dynamic path matching.

[0008] Preferably, the path regulation simulation data at least includes material stress distribution, defect clustering area, safety detection boundary set, and dynamic correction priority sequence; The dynamic correction simulation information at least includes detection point offset, path coverage completeness rate, and accuracy stability evaluation index.

[0009] Preferably, the performing multi-dimensional scanning on the target and constructing a three-dimensional material feature model, and performing dynamic defect feature analysis on the three-dimensional material feature model in combination with multi-modal sensing data to generate an initial defect feature evaluation model includes the following steps: Perform ray and acoustic wave composite scanning on the target to obtain a sensing data sequence of the internal defect distribution of the material; Preprocess the sensing data sequence to obtain normalized feature fusion data, where the preprocessing includes one or more of signal denoising, time domain registration, feature fusion, grid division, and data interpolation; Generate a three-dimensional material feature model based on standardized feature fusion data and combined with three-dimensional point cloud reconstruction technology; Perform dynamic defect feature analysis on the three-dimensional material feature model to generate an initial defect feature evaluation model. Among them, the dynamic defect feature analysis at least includes grid density allocation and feature weight mapping, divides the defect risk level through grid density allocation, and assigns corresponding sensing data parameters to each level area.

[0010] Preferably, load the material property distribution data and detection accuracy constraint conditions into the initial defect feature evaluation model to generate a comprehensive defect feature evaluation model, and perform dynamic simulation of the detection path in combination with the motion state parameters to obtain path regulation simulation data, including the following steps: Load the material property distribution data into the initial defect feature evaluation model to simulate real-time property changes and generate a first defect feature evaluation model; Load the detection accuracy constraint conditions into the first defect feature evaluation model to generate a comprehensive defect feature evaluation model. Among them, the detection accuracy constraint conditions include the maximum detection angle limit, the minimum resolution threshold, and the sensor response redundancy parameter; Based on the comprehensive defect feature evaluation model and the motion state parameters, perform dynamic simulation of the detection path. The specific process includes: Construct a multi-objective path planning equation. The multi-objective path planning equation at least includes a coverage integrity equation, a trajectory coincidence equation, and a detection efficiency equation. In combination with the comprehensive defect feature evaluation model, numerically solve the multi-objective path planning equation through the piecewise gradient descent method to obtain the material stress distribution, the defect clustering area, the set of safety detection boundaries, and the dynamic correction priority sequence; The generation of the set of safety detection boundaries includes the following steps: Based on the defect clustering area, calculate the defect coverage density of each grid cell; Identify the areas in the comprehensive defect feature evaluation model where the defect coverage density is not greater than the preset threshold to generate a set of safety detection boundaries.

[0011] Preferably, construct a dynamic defect prediction model and train it with the path regulation simulation data to obtain a detection path correction model, and then generate initial detection parameters, including the following steps: Construct a dynamic defect prediction model based on the three-dimensional material feature model; Train and verify the dynamic defect prediction model with the path regulation simulation data to obtain a detection path correction model; Input the real-time motion state parameters into the detection path correction model to predict the set of safety detection boundaries; Generate initial detection parameters based on a predicted set of safety detection boundaries, where the initial detection parameters at least include initial detection point coordinates, an initial scanning path, and an initial correction response parameter.

[0012] Preferably, construct a dynamic defect prediction model and train it with path regulation simulation data to obtain a detection path correction model, and then generate initial detection parameters. It also includes data reconstruction of the path regulation simulation data, specifically: Construct an initial multi-dimensional detection input tensor based on material stress distribution, defect clustering regions, and the set of safety detection boundaries; Perform standardization and feature dimensionality reduction processing on the initial multi-dimensional detection input tensor to generate a final multi-dimensional detection input tensor; Construct a detection path correction label tensor based on a dynamic correction priority sequence; Combine the final multi-dimensional detection input tensor and the detection path correction label tensor to form a training sample set.

[0013] Preferably, the generating of initial detection parameters based on the predicted set of safety detection boundaries includes the following steps: Extract the grid cell with the lowest defect coverage density in the predicted set of safety detection boundaries to generate initial detection point coordinates; According to the spatial topological relationship of the predicted set of safety detection boundaries, fit the feasible connection structure of the initial scanning path to generate an initial correction response parameter; Calculate the material stress gradient direction of the predicted set of safety detection boundaries and normalize it into a path reference vector, and the path reference vector is the initial scanning path direction.

[0014] Preferably, the obtaining of dynamic correction simulation information based on the initial detection parameters, the comprehensive defect feature evaluation model, and the motion state parameters includes the following steps: Map the initial detection point coordinates into the comprehensive defect feature evaluation model, match the initial scanning path and the initial correction response parameter, and adjust the path node density in the correction area, update the comprehensive defect feature evaluation model, define dynamic correction trigger conditions, path reconstruction rules, and response parameter increments, and generate a dynamic correction simulation model; Based on the dynamic correction simulation model, iteratively solve the multi-objective path planning equation by the piecewise gradient descent method to obtain dynamic correction simulation information, specifically including: When the dynamic correction trigger condition is satisfied, update the detection point offset, path coverage completeness rate, and accuracy stability evaluation index, and re-solve the multi-objective path planning equation until the simulation termination condition is reached; The dynamic correction trigger condition includes triggering the update of response parameters when the current path coverage completion rate is not greater than a preset completion rate threshold; the path reconstruction rule includes adjusting the initial scanning path based on the material stress gradient direction; and the response parameter increment has a piecewise linear relationship with the current accuracy stability evaluation index.

[0015] Preferably, the steps of constructing a multi-objective optimization model and iteratively optimizing the planning parameters to obtain the optimal detection path parameters include the following: Construct a multi-objective optimization model, where the optimization variables include the path node distribution density, the scanning angle threshold, and the sensor sampling frequency, the optimization objectives include maximizing the path coverage efficiency and minimizing the detection energy consumption, and the constraint conditions include the upper limit of the robot kinematics and the sensor accuracy threshold; Perform a preliminary solution of the multi-objective optimization model through a dynamic programming algorithm to generate an initial set of optimized paths; Based on the initial set of optimized paths, use the ant colony optimization algorithm for global optimization to generate the optimal detection path parameters.

[0016] Preferably, the present invention further includes an artificial intelligence-based robot non-destructive testing system, and the system includes: A data acquisition module for acquiring multi-modal sensing data of the target to be detected and the motion state parameters of the robot end effector; A feature modeling and analysis module for performing multi-dimensional scanning on the target and constructing a three-dimensional material feature model, dynamically analyzing the defect features of the three-dimensional material feature model in combination with the multi-modal sensing data, and generating an initial defect feature evaluation model; A comprehensive evaluation and simulation module for loading the material property distribution data and the detection accuracy constraint conditions into the initial defect feature evaluation model to generate a comprehensive defect feature evaluation model, and performing dynamic simulation of the detection path in combination with the motion state parameters to obtain path regulation simulation data; A parameter generation module for constructing a dynamic defect prediction model and training it through the path regulation simulation data to obtain a detection path correction model, and further generating initial detection parameters, where the initial detection parameters at least include the initial detection point coordinates, the initial scanning path, and the initial correction response parameters; A dynamic correction simulation module for obtaining dynamic correction simulation information based on the initial detection parameters, the comprehensive defect feature evaluation model, and the motion state parameters; A path optimization module for constructing a multi-objective optimization model in combination with the dynamic correction simulation information and the detection task planning parameters and iteratively optimizing the planning parameters to obtain the optimal detection path parameters; A strategy adjustment module is used to infer the detection path parameters in the current period based on the real-time material feature model, generate the safety detection probability in the current period, and combine the optimal detection path parameters in the current period with the actual scanning trajectory to adjust the real-time detection strategy of the robot to achieve the goal of dynamic path matching.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: In terms of detection accuracy, by obtaining multi-modal sensing data of the target to be detected and combining technologies such as composite scanning of rays and sound waves, it is possible to obtain more comprehensive and accurate information on the internal defect distribution of the material. Signal denoising, time-domain registration, feature fusion and other preprocessing are performed on the sensing data to generate standardized feature fusion data, and a three-dimensional material feature model is constructed based on this, realizing the accurate modeling of material features. Through dynamic defect feature analysis, including grid density allocation and feature weight mapping, corresponding sensing data parameters can be assigned to each region according to the defect risk level, making the defect feature evaluation more accurate. Loading the material property distribution data and detection accuracy constraint conditions to generate a comprehensive defect feature evaluation model further considers the real-time changes of material properties and the requirements of detection accuracy, improving the accuracy and reliability of defect evaluation.

[0018] In terms of detection efficiency, a multi-objective path planning equation is constructed, including a coverage integrity equation, a trajectory coincidence equation and a detection efficiency equation. Combining with the comprehensive defect feature evaluation model and performing numerical solution through the piecewise gradient descent method, an efficient detection path can be generated. Using the dynamic programming algorithm and the ant colony optimization algorithm to solve the multi-objective optimization model realizes the iterative optimization of the detection path parameters, obtains the optimal detection path parameters, effectively reduces the redundant paths and repeated detections in the detection process, and improves the detection efficiency. At the same time, through the dynamic simulation of the detection path and the dynamic correction simulation, the detection path can be adjusted in time according to the real-time feedback, avoiding detection blind spots and unreasonable paths, and further improving the detection efficiency.

[0019] In terms of adaptability, the system can adjust the detection strategy in real time according to the motion state parameters of the robot end effector to achieve the goal of dynamic path matching. By constructing a dynamic defect prediction model and training it with path regulation simulation data, the safety detection boundary set can be predicted, and the initial detection parameters suitable for different detection scenarios can be generated. When the dynamic correction trigger condition is met during the detection process, the detection point offset, path coverage integrity rate and accuracy stability evaluation index can be automatically updated, and the multi-objective path planning equation can be solved again, enabling the detection system to adapt to complex situations such as material property changes and dynamic adjustment of defect distribution, improving the adaptability and robustness of the system.

[0020] In terms of data processing and model construction, multi-dimensional scanning and feature fusion are performed on multi-modal sensing data to construct a three-dimensional material feature model and a comprehensive defect feature evaluation model, which can make full use of the advantages of multi-source heterogeneous data, improve the efficiency of data processing and the accuracy of the model. By reconstructing the path regulation simulation data, an initial multi-dimensional detection input tensor is constructed and standardized and feature dimension reduction processing is performed to generate a training sample set, providing high-quality data support for the training of the dynamic defect prediction model and enhancing the performance and generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 FIG. is a working principle diagram of the robot non-destructive testing method based on artificial intelligence according to the present invention; Figure 2 FIG. is a flowchart for training the dynamic defect prediction model and generating initial detection parameters according to the present invention; Figure 3 FIG. is a flowchart for generating initial detection parameters based on the set of safety detection boundaries according to the present invention; Figure 4 FIG. is a flowchart for generating dynamic correction simulation information and path iteration optimization according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0023] Please refer to Figures 1-4 , a robot non-destructive testing method based on artificial intelligence according to the present invention, and the specific implementation steps are as follows: Through a variety of sensors (such as ray sensors, acoustic sensors, visual sensors, etc.) deployed at the end of the robot, multi-modal sensing data of the target to be detected is collected in real time, including but not limited to the internal structure signal of the material, surface feature information, etc.; at the same time, the motion state parameters of the end effector, such as position, speed, acceleration, etc., are obtained through the robot control system.

[0024] Perform multi-dimensional scanning on the target, use the scanned data to construct a three-dimensional material feature model, and combine the multi-modal sensing data to perform dynamic defect feature analysis on the model to generate an initial defect feature evaluation model.

[0025] Load the material property distribution data and detection accuracy constraint conditions into the initial defect feature evaluation model to generate a comprehensive defect feature evaluation model, combine the motion state parameters to perform dynamic simulation of the detection path, and obtain path regulation simulation data.

[0026] Construct a dynamic defect prediction model and train it with the path regulation simulation data to obtain a detection path correction model, and then generate initial detection parameters including the initial detection point coordinates, the initial scanning path, and the initial correction response parameters.

[0027] Based on the initial detection parameters, the comprehensive defect feature evaluation model, and the motion state parameters, obtain dynamic correction simulation information.

[0028] Combine the dynamic correction simulation information with the detection task planning parameters, construct a multi-objective optimization model and iteratively optimize the planning parameters to obtain the optimal detection path parameters.

[0029] Based on the real-time material feature model, infer the detection path parameters for the current period to generate the safety detection probability for the current period. Combine the optimal detection path parameters for the current period with the actual scanning trajectory, and adjust the robot's real-time detection strategy to achieve the dynamic path matching goal.

[0030] The present invention will be further described below in conjunction with Embodiments 1 to 5: Embodiment 1: This embodiment details the specific composition and generation logic of the path regulation simulation data and the dynamic correction simulation information. The following specifically describes from the definitions, constituent elements, generation principles, and roles in the detection process of these two types of data: The path regulation simulation data is the key intermediate data generated during the dynamic simulation of the detection path, and is used to provide a basis for subsequent detection path correction and optimization. It at least includes the material stress distribution, the defect clustering region, the set of safety detection boundaries, and the dynamic correction priority sequence. Among them, the generation of the material stress distribution is based on the simulation calculation of the material properties of the detection target. During the dynamic simulation of the detection path, the system will load the property parameters such as the elastic modulus and Poisson's ratio of the material into the comprehensive defect feature evaluation model, and simulate the stress state of the material during the detection process through methods such as finite element analysis, and then obtain the stress numerical distribution of each region. This distribution takes the grid unit as the basic unit, and each unit corresponds to a stress value, reflecting the distribution of the material's internal resistance to deformation. High-stress regions may have potential structural weak points and need to be focused on in the detection path planning.

[0031] The defect clustering region is determined by statistically analyzing the defect features in the multi-modal sensing data. After generating the initial defect feature evaluation model, the system processes the defect features in the model using a clustering algorithm (such as the DBSCAN algorithm or the K-means algorithm), dividing defects with similar spatial positions and features into the same clustering cluster, thereby identifying the regions where defects are concentrated. Each clustering region contains information such as the position coordinates, size, and type of the defects. This information helps to judge the damage degree and development trend of the detection target. For example, the aggregation of multiple small defects may indicate the impending formation of a larger crack. Therefore, the determination of the defect clustering region can provide direct guidance for the key coverage areas of the detection path.

[0032] The generation of the safety detection boundary set depends on the calculation and analysis of the defect coverage density. First, the system divides the target region in the comprehensive defect feature evaluation model into several grid cells. For each grid cell, the number of defects inside it or the proportion of the defect area to the total area of the grid cell is calculated to obtain the defect coverage density. The preset threshold is set in advance according to the detection standard and the characteristics of the target material. For example, for aerospace materials with high reliability requirements, the preset threshold may be set lower to ensure sufficient detection of low-defect-density regions. When the defect coverage density of a certain grid cell is not greater than the preset threshold, the cell is determined as a low-risk region, and its boundary is extracted and integrated into the safety detection boundary set. This set defines the safe range of the detection path, avoiding over-detection of low-risk regions while ensuring the comprehensiveness of the detection, thereby improving the detection efficiency.

[0033] The dynamic correction priority sequence is the correction order of the detection path determined according to the defect risk level and detection requirements. In the initial defect feature evaluation model, the defect risk levels of each region have been divided through grid density assignment and feature weight mapping (such as high, medium, and low levels). The dynamic correction priority sequence first sets a higher correction priority for the detection path segments corresponding to high-risk level regions, ensuring that the paths in these regions are adjusted first during the detection process to deal with defects that may develop rapidly; secondly for medium-risk level regions, and finally for low-risk level regions. The setting of the priority sequence also takes into account the motion state parameters of the robot end effector. For example, when the robot moves at a relatively high speed in a certain region, to avoid missed detection, the correction priority of the path in this region may be appropriately increased.

[0034] The dynamically corrected simulation information is data generated based on the initial detection parameters, the comprehensive defect feature evaluation model, and the motion state parameters, and is used to guide the dynamic correction of the detection path. It includes at least the detection point offset, the path coverage completion rate, and the accuracy stability evaluation index. The detection point offset refers to the spatial deviation between the position of the detection point of the robot's end effector during the actual detection process and the coordinates of the planned initial detection point. This deviation is obtained by real-time measurement using the robot's position sensor and is compared and calculated with the coordinate system in the comprehensive defect feature evaluation model. The magnitude of the detection point offset reflects the accuracy of the robot's motion control and the execution error of the detection path. When the offset exceeds the preset allowable range, it is necessary to trigger the path correction mechanism to adjust the subsequent detection path.

[0035] The path coverage completion rate is used to measure the coverage degree of the detection path for the target area. Its calculation method is: the ratio of the area of the detected area to the total area of the target area. During the dynamic correction simulation process, the system will track the scanning trajectory of the robot in real time, mark the scanned area in the comprehensive defect feature evaluation model, and calculate the coverage area. The preset completion rate threshold is set according to the requirements of the detection task. For example, for the detection of key components, the threshold may be set to more than 95%. When the path coverage completion rate is not greater than the preset threshold, it indicates that there are coverage gaps in the detection path, and it is necessary to perform supplementary inspections by adjusting the detection point coordinates, scanning path direction, etc. to ensure the comprehensiveness of the detection.

[0036] The accuracy stability evaluation index is obtained by analyzing the fluctuation of sensor data during the detection process. When multi-modal sensors collect data, they may be affected by factors such as environmental noise and sensor drift, resulting in data fluctuations. The system will perform statistical analysis on the real-time data collected by the sensors, calculate statistical quantities such as the standard deviation and coefficient of variation of the data, as the accuracy stability evaluation index. This index reflects the stability of the detection accuracy. When the index value is large, it indicates that the detection accuracy fluctuates greatly, which may affect the accurate identification of defect features. At this time, it is necessary to adjust parameters such as the sampling frequency and detection angle of the sensor, or calibrate the sensor to improve the stability of the detection accuracy.

[0037] In the detection process, the path regulation simulation data and the dynamic correction simulation information are interrelated and interact synergistically. The path regulation simulation data is the basis for generating the dynamic correction simulation information. For example, the safety detection boundary set provides a basis for generating the initial detection point coordinates, and the dynamic correction priority sequence affects the setting of the response parameter increment. The dynamic correction simulation information, on the other hand, is the feedback and adjustment of the path regulation simulation data during the actual detection process. Through indicators such as the detection point offset and the path coverage completeness rate, the system can evaluate the execution effect of the detection path in real time and dynamically correct the detection path based on information such as the material stress distribution and defect clustering area in the path regulation simulation data. This closed-loop feedback mechanism ensures that the non-destructive detection process of the robot can adaptively adjust the detection strategy according to the actual state of the target material and the detection requirements, improving the accuracy and efficiency of the detection.

[0038] Embodiment 2: This embodiment elaborates in detail the specific implementation process of multi-dimensional scanning of the target, constructing a three-dimensional material feature model, and generating an initial defect feature evaluation model in combination with multi-modal sensing data. This process realizes the precise modeling and preliminary evaluation of internal defects of the detection target through the organic combination of composite scanning technology, data preprocessing, three-dimensional reconstruction, and dynamic analysis. The specific content is as follows: First, the target is subjected to a combined ray and acoustic wave scan to obtain a sequence of sensing data. In actual operation, the end effector of the robot needs to carry at least two types of sensors: a ray sensor (such as an X-ray or γ-ray sensor) and an acoustic wave sensor (such as an ultrasonic sensor). The ray sensor emits rays to penetrate the target material and obtains information on the internal density distribution, cracks, pores, and other defects of the material based on the differences in the attenuation degree of the rays after passing through the material. The acoustic wave sensor emits acoustic wave signals (such as longitudinal waves, transverse waves, or surface waves) to the target material and analyzes the internal interface characteristics, defect positions, and sizes of the material by receiving the reflected, refracted, or scattered acoustic wave signals. The two sensors work synchronously to perform multi-dimensional scanning of the target: the ray scan provides the overall density distribution and macroscopic defect profile inside the material, while the acoustic wave scan supplements the fine boundary information of the internal microstructure characteristics and defects of the material. During the scanning process, the position, scanning angle, and moving speed of the sensors are precisely controlled by the robot control system to ensure that the scanning path covers the entire area of the target and avoid missing key detection parts. After the scanning is completed, the two types of sensors respectively output the original sensing data, forming a sequence of sensing data containing information such as time series, spatial coordinates, and signal intensity, which serves as the basic data for subsequent analysis.

[0039] Preprocess the sensing data sequence to obtain normalized feature fusion data. The preprocessing link includes various processing means such as signal denoising, time-domain registration, feature fusion, grid division, and data interpolation. These means are combined according to the data characteristics and detection requirements. Signal denoising aims to remove noise interference in the original data. Common methods include Fourier transform filtering, wavelet denoising, etc. For example, for the random noise generated by environmental radiation in ray sensing data, the wavelet threshold denoising algorithm can be used to retain the effective signal components and suppress the noise; for the periodic noise generated by sensor vibration in acoustic wave sensing data, a band-pass filter can be used for frequency-domain filtering. Time-domain registration is used to solve the problem of asynchronous multi-sensor data in the time dimension. By timestamp alignment or linear interpolation methods, the ray and acoustic wave sensing data sequences are uniformly sampled in the time axis to ensure the time correspondence of data points during subsequent feature fusion. Feature fusion is to integrate the features of multi-modal sensing data to form a unified feature representation vector. Specifically, the attenuation coefficient feature of ray data and the sound velocity and acoustic impedance features of acoustic wave data are normalized (such as standardized to the [0,1] interval), and then concatenated by dimension to generate a feature vector containing multi-source information, which can more comprehensively reflect the internal characteristics of the material. Grid division is to divide the target material into regular or irregular grid cells in three-dimensional space. The size of the grid is determined according to the detection accuracy requirements. For example, for key parts that require high-precision detection, a smaller grid size (such as millimeter level) is used to improve the accuracy of defect location; for non-critical parts, a larger grid size (such as centimeter level) can be used to reduce the data processing volume. Data interpolation is used to fill the data missing areas caused by scanning blind spots or sensor resolution limitations. Common methods include nearest neighbor interpolation, bilinear interpolation, or Kriging interpolation. For example, when the ray scan shows sparse data due to angle limitations in the target edge area, the missing data can be estimated by the Kriging interpolation method using the feature values of adjacent grid cells to improve the spatial continuity and integrity of the data.

[0040] Based on the standardized feature fusion data, the three-dimensional material feature model is generated by combining the three-dimensional point cloud reconstruction technology. The core of the three-dimensional point cloud reconstruction technology is to convert the two-dimensional sensor data into a point cloud set in three-dimensional space, and then construct a three-dimensional structural model of the target material. The specific steps are as follows: First, according to the spatial position and scanning angle of the sensor, the feature fusion data of each grid unit is mapped to the corresponding position in the three-dimensional coordinate system to generate point cloud data containing X, Y, Z coordinates and characteristic attributes (such as density, sound speed). Then, the point cloud filtering algorithm (such as statistical filtering, radius filtering) is used to remove outliers and improve the quality of the point cloud data. Then, the point cloud meshing algorithm (such as Delaunay triangulation or Poisson surface reconstruction) is used to connect the discrete point cloud data into a continuous surface mesh to form the geometric structure of the three-dimensional material feature model. This model not only intuitively displays the external contour of the target material, but also visualizes the distribution of characteristic attributes inside the material through color coding or transparency settings, such as red for high-density areas and blue for low-density areas, which is convenient for operators to quickly locate potential defect areas. In addition, the 3D material feature model supports interactive operations such as rotation, scaling, and cutting, so that the internal structure of the material can be observed from different angles.

[0041] The key link of this embodiment is to perform dynamic defect feature analysis on the three-dimensional material feature model to generate an initial defect feature evaluation model. The dynamic defect feature analysis includes two core steps: grid density allocation and feature weight mapping. Grid density allocation is to dynamically adjust the density of grid units in the three-dimensional model according to the defect risk level. First, the system performs defect detection on each grid unit in the three-dimensional material feature model through a preset defect recognition algorithm (such as a threshold-based segmentation algorithm or a machine learning classification algorithm) to determine whether the unit has defects and the severity of the defects. For grid units that are detected to have defects, the corresponding risk level (such as high, medium, and low) is assigned according to the size, type, and number of defects. High-risk level units (such as units containing large-sized cracks or dense pores) are subdivided using a smaller grid size, for example, dividing an original grid unit into eight subunits to improve the detection resolution of the area; medium-risk level units maintain the original grid size; low-risk level units appropriately increase the grid size to reduce the amount of calculation. Through this adaptive grid density allocation, the system can concentrate computing resources in defect-concentrated areas to improve the accuracy of defect feature analysis, while reducing computational complexity in low-risk areas and improving overall processing efficiency.

[0042] Feature weight mapping assigns corresponding weights to sensing data parameters for different risk-level regions to highlight the importance of features in high-risk regions. Specifically, for grid cells in high-risk regions, the weight values of key features such as the ray attenuation coefficient and the acoustic wave reflection intensity are increased (e.g., the weight coefficient is set to 1.5), so that these features play a greater decision-making role in subsequent defect assessments; for medium-risk regions, the weight coefficient is set to 1.0 to maintain the importance of the original features; for low-risk regions, the weight coefficient is set to 0.8 to appropriately reduce the influence of features. The setting of the weight coefficient is determined by a trained machine learning model (such as a random forest or a neural network), which is trained based on historical detection data and defect assessment results and can automatically learn the importance of different features in defect recognition. Through feature weight mapping, when the system generates an initial defect feature assessment model, it can pay more attention to the feature changes in high-risk regions and improve the ability to identify early defects or potential defects.

[0043] After generating the initial defect feature assessment model, this model will be used as the basic data for subsequent processes to load material property distribution data and detection accuracy constraint conditions, and then generate a comprehensive defect feature assessment model. At the same time, the defect risk level division and feature weight information in the initial model will also provide input parameters for the multi-objective path planning equation in the dynamic simulation of the detection path. For example, the path segments corresponding to high-risk regions will be given higher coverage integrity requirements and detection efficiency priorities.

[0044] Embodiment 3: This embodiment details the specific process of comprehensively processing the initial defect feature assessment model to generate a comprehensive defect feature assessment model and performing dynamic simulation of the detection path in combination with motion state parameters to obtain path regulation simulation data. This process realizes the dynamic simulation and optimization of the detection path through links such as material property simulation, detection accuracy constraint loading, multi-objective path planning, and boundary set generation. The specific content is as follows: Load the material property distribution data into the initial defect feature evaluation model to simulate real-time property changes and generate the first defect feature evaluation model. The material property distribution data covers the physical property parameters of the material, such as elastic modulus, density, coefficient of thermal expansion, Poisson's ratio, etc. These parameters are preset according to the type of target material (such as metal, composite material, ceramic, etc.) and specific working conditions (such as temperature, pressure environment) or obtained by real-time sensor acquisition. During the loading process, the system maps the material property parameters to each grid cell of the initial model according to the spatial distribution. For example, for the area with a large temperature gradient in a metal component, an elastic modulus parameter that varies with temperature is assigned to simulate the property fluctuations of the material under thermal stress. Through this simulation, the first defect feature evaluation model can more realistically reflect the state of the material under actual working conditions, making the subsequent detection path planning more in line with actual needs. For example, in the detection of aeroengine blades, considering the influence of the decrease in the elastic modulus of the material under high-temperature environment on the defect development, by loading the temperature-related property data, the model can predict the expansion trend of the defect under thermal stress, so as to adjust the key detection area.

[0045] Load the detection accuracy constraint conditions into the first defect feature evaluation model to generate a comprehensive defect feature evaluation model. The detection accuracy constraint conditions are key parameters to ensure the reliability of the detection results, including the maximum detection angle limit, the minimum resolution threshold, and the sensor response redundancy parameter. The maximum detection angle limit is set based on the physical characteristics and detection principle of the sensor. For example, an ultrasonic sensor has an optimal detection angle range. When the detection angle exceeds this range, the intensity of the acoustic wave reflection signal will significantly attenuate, resulting in a decrease in the defect recognition rate. Therefore, the detection angle needs to be limited within the effective working range of the sensor. The minimum resolution threshold is determined according to the detection standard and the type of defect. For example, for crack defects, the minimum crack length that the sensor can detect (such as 0.5 mm) needs to be set to ensure that defects smaller than this threshold in the model will not be ignored. The sensor response redundancy parameter is used to cope with possible abnormal responses of the sensor. For example, a redundancy threshold for the sensor signal intensity is set. When the actual detection signal intensity is lower than the redundancy threshold, a mechanism for sensor calibration or switching to a backup sensor is triggered. When loading the detection accuracy constraint conditions, the system checks each grid cell in the first defect feature evaluation model to ensure that the corresponding detection angle, resolution, and sensor response parameters all meet the constraint conditions, marks or adjusts the cells that do not meet the conditions, and finally generates a comprehensive defect feature evaluation model that includes double constraints of material properties and detection accuracy.

[0046] Based on the comprehensive defect feature evaluation model and motion state parameters, the core of dynamic simulation of the detection path is to construct and solve the multi-objective path planning equation. The multi-objective path planning equation includes at least a coverage integrity equation, a trajectory overlap equation, and a detection efficiency equation. Each equation describes different optimization objectives of the detection path through mathematical expressions. The coverage integrity equation aims to maximize the coverage area of the detection path for the target area and avoid undetected blind spots. Its expression is usually calculated based on the coverage status (covered or uncovered) of grid cells; the trajectory overlap equation aims to minimize the repeated scanning area of the detection path. By calculating the overlapping area ratio of the path trajectory, redundant detections are reduced and the detection efficiency is improved; the detection efficiency equation aims to minimize the detection time or energy consumption. Combining parameters such as the motion speed and acceleration of the robot end effector, the continuity and smoothness of the path are optimized.

[0047] When solving the multi-objective path planning equation, the system uses the segmented gradient descent method. The entire detection area is divided into multiple sub-areas, and the optimal path is solved separately in each sub-area, and then the global path is obtained by stitching the paths of each sub-area. The specific steps are as follows: First, the grid cells in the comprehensive defect feature evaluation model are grouped according to the defect risk level (high, medium, low). The grid cells in the high-risk level group are preferentially path-planned to ensure the detection accuracy of key areas; second, in each sub-area, with the coverage integrity equation as the main optimization objective, combined with the trajectory overlap equation and the detection efficiency equation, the coordinates of the path nodes are iteratively updated through the gradient descent algorithm until the local optimal solution is reached; finally, the local optimal paths of each sub-area are smoothed to avoid mutations at the path connection points and form a complete detection path. The path regulation simulation data obtained by solving through this method includes material stress distribution, defect clustering area, safety detection boundary set, and dynamic correction priority sequence. Among them, the material stress distribution is obtained through the material property parameters in the comprehensive defect feature model and the mechanical analysis in the path planning process, reflecting the stress state of each point on the detection path and providing a mechanical basis for path correction.

[0048] The process of generating the safety detection boundary set is as follows: First, based on the distribution of the defect clustering area, the defect coverage density of each grid cell is calculated. The calculation method of the defect coverage density is the ratio of the area or number of defects in the cell to the total area of the cell, which reflects the density of defects in the cell. For example, for a certain grid cell, if it contains 3 pores with sizes of 1mm², 2mm², and 1.5mm² respectively, the defect coverage density is (1 + 2 + 1.5) / cell area. The preset threshold is set according to the detection standard and the allowable defect degree of the target material. For example, for pressure vessel materials, the preset threshold may be set to 5%, that is, when the cell defect coverage density is less than or equal to 5%, the cell is considered a low-risk area.

[0049] After identifying the areas in the comprehensive defect feature evaluation model where the defect coverage density is not greater than the preset threshold, the system extracts the boundaries of these low-risk areas through edge detection algorithms (such as the Canny algorithm or the Sobel algorithm) to generate a set of safety detection boundaries. This set consists of a series of continuous boundary segments or surfaces, clearly demarcating the high-risk areas that require key detection from the low-risk areas where the detection density can be appropriately reduced. In the detection path planning, the set of safety detection boundaries is used to limit the expansion range of the path, that is, the detection path mainly conducts intensive scans within the high-risk areas surrounded by the boundary set, while sparse scans or skipping can be adopted for the low-risk areas outside the boundary set, thereby reducing the detection time and energy consumption while ensuring the detection quality. For example, in pipeline weld detection, the set of safety detection boundaries can be limited to high-risk areas such as the weld and its heat-affected zone, while only periodic spot checks can be carried out for the low-risk areas of the pipeline body.

[0050] The generation of the dynamically corrected priority sequence is related to the defect risk level and the complexity of the detection path. The path segments corresponding to high-risk level areas are assigned the highest priority because the defects develop rapidly and have a great impact on structural safety, and they are preferentially adjusted during the dynamic correction process; the path segments of medium-risk level areas have the second highest priority; the path segments of low-risk level areas have the lowest priority. In addition, the complexity of the path (such as the number of path nodes and the number of turns) also affects the priority. Complex path segments may be assigned a higher correction priority because they are more likely to have execution errors. The dynamically corrected priority sequence is stored in the form of a list, and each element corresponds to the identifier and priority value of a path segment. The system corrects the path segments in sequence according to this sequence to ensure the orderliness and efficiency of the detection process.

[0051] During the dynamic simulation of the detection path, the motion state parameters (such as the position, speed, and acceleration of the robot end effector) are input into the simulation system in real time to verify the feasibility of the path and the reachability of the robot. For example, when there is an angle in the simulation path that the robot joint movement range cannot reach, the system will automatically mark this path segment and trigger the path reconstruction mechanism to adjust the position or angle of the path nodes until the path meets the robot kinematic constraints. At the same time, the motion state parameters are also used to calculate the execution time and energy consumption of the path, providing data support for the solution of the detection efficiency equation.

[0052] Embodiment 4: This embodiment details the complete process of constructing a dynamic defect prediction model, training the model through path regulation simulation data to generate a detection path correction model, and generating initial detection parameters based on the model, and also covers the specific implementation steps of data reconstruction. The following will specifically describe from aspects such as model construction, data processing, and parameter generation: Build a dynamic defect prediction model based on a three-dimensional material feature model. The three-dimensional material feature model has been generated through multi-dimensional scanning and point cloud reconstruction technologies, and includes the geometric structure of the target material, the distribution of internal feature attributes (such as density, sound velocity), and mesh cell division information. The dynamic defect prediction model adopts an artificial intelligence algorithm architecture, such as a convolutional neural network (CNN), a recurrent neural network (RNN), or a graph neural network (GNN). Its input is the mesh cell feature vector of the three-dimensional model (including multi-modal sensing data after standardization), and the output is the defect probability value or defect type label of each mesh cell. The choice of model architecture depends on the spatial distribution characteristics of defect features: for defects with local spatial correlation (such as crack propagation), CNN is more suitable for capturing two-dimensional or three-dimensional spatial features; for the defect development process that changes over time, RNN or a temporal convolutional network (TCN) is more suitable for processing sequence data; for models with complex topological relationships between mesh cells, GNN can better model the dependencies between nodes. During the model construction process, hyperparameters such as the number of network layers, the number of neurons, and activation functions need to be configured according to the characteristics of the detection target. For example, in the detection of aerospace materials, due to the complexity of defect types, a model structure combining a deep CNN with an attention mechanism can be adopted to enhance the ability to capture subtle defect features.

[0053] Train and validate the dynamic defect prediction model with path-regulated simulation data. This process requires prior reconstruction of the data. The first step in data reconstruction is to build an initial multi-dimensional detection input tensor, which integrates the material stress distribution, defect clustering regions, and safety detection boundary set information in the path-regulated simulation data. Specifically, the material stress distribution values, the mesh cell identifiers of the defect clustering regions (such as belonging to the nth clustering cluster), and the mesh cell identifiers of the safety detection boundary set (such as 0 representing non-boundary cells and 1 representing boundary cells) are dimensionally concatenated in the order of mesh cells to form a three-dimensional tensor (dimension: number of mesh cells × feature dimension × 1). For example, each mesh cell corresponds to a feature vector of length 3, which respectively includes the stress value, the clustering cluster number, and the boundary identifier. The shape of the initial tensor is (N, 3, 1), where N is the total number of mesh cells.

[0054] Normalizing and dimension reducing the initial multi-dimensional detection input tensor are key steps in data reconstruction. The normalization process normalizes the data of each dimension to a distribution with zero mean and unit variance by subtracting the feature mean and dividing by the standard deviation, avoiding model training bias caused by differences in feature dimensions. The feature dimension reduction process uses algorithms such as principal component analysis (PCA) or t-distributed stochastic neighbor embedding (t-SNE). For example, reducing a 3D feature vector to 2D reduces the computational complexity while retaining the main information. The shape of the final multi-dimensional detection input tensor after dimension reduction becomes (N, 2, 1). This tensor, as the input data of the model, can effectively improve the training efficiency and the generalization ability of the model.

[0055] Construct a detection path correction label tensor based on the dynamic correction priority sequence. The dynamic correction priority sequence is the pre-set correction order of path segments according to the defect risk level and detection requirements. Each grid cell corresponds to a priority value (e.g., 3 for high-risk areas, 2 for medium-risk areas, and 1 for low-risk areas). The label tensor is constructed as follows: perform one-hot encoding on the priority value of each grid cell, convert it into a binary vector of length 3 (e.g., priority 3 corresponds to [1, 0, 0]), and then concatenate the label vectors of all grid cells in order to form a three-dimensional tensor (dimension N×3×1). This label tensor is used to indicate the detection path correction direction that the model needs to learn during training. For example, the grid cells corresponding to high-priority labels need to focus on path adjustment.

[0056] Combine the final multi-dimensional detection input tensor and the detection path correction label tensor to form a training sample set. Each sample consists of a grid cell feature vector in the input tensor and the corresponding label vector. The sample set is divided into a training set (accounting for 70%), a validation set (accounting for 20%), and a test set (accounting for 10%). During training, use stochastic gradient descent (SGD) or an adaptive optimization algorithm (such as Adam) to minimize the loss function (such as the cross-entropy loss function), and update the model parameters through the backpropagation algorithm. The validation set is used to monitor the overfitting phenomenon during model training. When the validation loss no longer decreases, trigger the early stopping mechanism to avoid overtraining of the model. The test set is used to evaluate the generalization ability of the model, calculate metrics such as accuracy and recall, but does not involve the description of specific experimental effect data.

[0057] Through the trained and verified dynamic defect prediction model, a detection path correction model is further obtained. The detection path correction model is a derivative model of the dynamic defect prediction model, and its output is the detection path correction parameters for each grid cell (such as the scanning angle adjustment amount, path node offset amount). Input the real-time motion state parameters (such as the current position, speed, and acceleration of the robot end effector) into the detection path correction model, and the model combines the characteristic attributes and priority labels of the current grid cell to predict the set of safety detection boundaries corresponding to the cell. For example, when the robot moves to a grid cell in a high-risk area, the model predicts that the safety detection boundary of the cell may shift due to motion inertia according to the real-time speed parameter, and thus outputs the corresponding boundary adjustment value.

[0058] Generate initial detection parameters based on the predicted set of safety detection boundaries, including the following specific steps: First, extract the grid cell with the lowest defect coverage density in the set of safety detection boundaries, and use its geometric center coordinates or characteristic point coordinates (such as the midpoint of the boundary line segment) as the initial detection point coordinates. The area with the lowest defect coverage density is usually considered the "weak link" in the detection path. Setting the detection point here first can increase the probability of discovering potential defects. For example, in a certain area of the set of safety detection boundaries, by traversing the defect coverage density values of all grid cells, find the cell corresponding to the minimum value, and take its center coordinates (x0, y0, z0) as the initial detection point.

[0059] According to the spatial topological relationship of the predicted set of safety detection boundaries, fit the feasible connection structure of the initial scanning path. The spatial topological relationship is represented by the adjacency matrix of the grid cells, which records the connection relationship between each cell and its adjacent cells (such as the adjacent cells in the six directions of up, down, left, right, front, and back). The fitting process uses the shortest path algorithm in graph theory (such as Dijkstra algorithm or A* algorithm), starting from the initial detection point and aiming to cover all grid cells in high-risk areas, searching for the shortest path connecting each detection point while satisfying the kinematic constraints of the robot (such as the maximum turning angle, minimum straight-line moving distance). During the fitting process, generate the initial correction response parameters, including the step size of path adjustment (such as moving 5mm per step), angle increment (such as turning no more than 30° each time), etc. These parameters are used to control the path correction amplitude of the robot during the scanning process.

[0060] Calculate the material stress gradient direction of the predicted safety detection boundary set and normalize it to the path reference vector. The material stress gradient direction is obtained by calculating the gradient of the stress values of each grid cell within the safety detection boundary set. The direction of the gradient vector points to the direction where the stress increases fastest, and its magnitude represents the stress change rate. The normalization process converts the gradient vector into a unit vector, serving as the reference for the initial scanning path direction. For example, if the calculated stress gradient vector is (Δx, Δy, Δz), then the path reference vector is (Δx / |Δr|, Δy / |Δr|, Δz / |Δr|), where |Δr| is the magnitude of the gradient vector. The initial scanning path direction extends along this reference vector to ensure that the detection path can preferentially cover the stress concentration area and improve the detection efficiency of stress-induced defects.

[0061] Throughout the implementation process, the data reconstruction link converts multi-source heterogeneous path regulation simulation data into a standard format suitable for model training through tensor construction and processing; the construction and training of the dynamic defect prediction model and the detection path correction model realize the intelligent mapping from material characteristics to detection path correction; the generation of the initial detection parameters combines geometric analysis, topology optimization, and mechanical properties to ensure the scientificity and feasibility of the parameters. Each link is closely connected to form an artificial intelligence-based detection path pre-planning and correction mechanism, providing key initial conditions for subsequent dynamic correction simulation and multi-objective optimization.

[0062] Example 5: This example details the specific process of obtaining dynamic correction simulation information based on the initial detection parameters and the comprehensive defect feature evaluation model, and constructing a multi-objective optimization model to iteratively optimize the detection path parameters. The following will specifically describe from aspects such as the construction of the dynamic correction simulation model, the iterative solution process, and the application of the multi-objective optimization algorithm: To generate dynamic correction simulation information based on the initial detection parameters, the comprehensive defect feature evaluation model, and the motion state parameters, it is necessary to first construct a dynamic correction simulation model. The specific steps are as follows: Map the initial detection point coordinates to the three-dimensional space coordinate system of the comprehensive defect feature evaluation model, match the initial scanning path with the initial correction response parameters (such as step size, angle increment), and adjust the path node density according to the defect risk level of the correction area - increase the node density in high-risk areas (such as reducing the node spacing to 2 mm), and decrease the node density in low-risk areas (such as expanding the node spacing to 10 mm), thereby updating the path parameters of the comprehensive defect feature evaluation model. Subsequently, define the dynamic correction trigger condition, path reconstruction rule, and response parameter increment: The dynamic correction trigger condition is set as the current path coverage completion rate not greater than the preset completion rate threshold (such as 90%); the path reconstruction rules include adjusting the initial scanning path direction based on the material stress gradient direction so that the new path direction is aligned with The included angle does not exceed ; the response parameter increment and the current precision stability evaluation index show a piecewise linear relationship, that is, when , when , when , (where is the preset increment coefficient, and are the high and low thresholds of precision stability).

[0063] Based on the dynamic correction simulation model, the multi-objective path planning equation is iteratively solved by the piecewise gradient descent method. The multi-objective path planning equation includes coverage integrity , trajectory coincidence , detection efficiency , etc. The objective functions are as follows (only the coverage integrity equation is given as formula 1): (1) Among them, represents the number of covered grid cells, represents the total number of grid cells in the target area. During the iterative solution process, the system monitors the path coverage completion rate in real time: when , the response parameter update is triggered, and the detection point offset , the path coverage completion rate , and the calculated value of the precision stability evaluation index are adjusted according to the current precision stability evaluation index , and the multi-objective path planning equation is re-solved until or the preset number of iterations (such as 50 times) is reached, and the simulation is terminated.

[0064] The steps to construct the multi-objective optimization model and iteratively optimize the planning parameters are as follows: The variables of the optimization model include the path node distribution density , the scanning angle threshold , the sensor sampling frequency , the optimization objective is to maximize the path coverage efficiency and minimize the detection energy consumption , and the constraint conditions include the upper limit of the robot kinematics (such as the maximum joint speed , the maximum acceleration ) and the sensor accuracy threshold (such as the minimum resolution ). First, the multi-objective optimization model is initially solved using a dynamic programming algorithm. The detection area is divided into multiple stages, and each stage corresponds to a set of path node distribution densities and scanning angle thresholds . By recursively calculating the optimal solutions for each stage, an initial optimized path set containing 10 - 20 paths is generated.

[0065] Based on the initial optimized path set, an ant colony optimization algorithm is used for global optimization. The ant colony optimization algorithm searches for the optimal solution in the initial optimized path set by simulating the pheromone transmission mechanism in the ant foraging process. The specific process is as follows: Each ant represents a detection path. When searching for a path, an ant selects the next node according to the pheromone concentration and heuristic information (such as the reciprocal of the path length). The pheromone concentration is updated according to the quality of the path passed by the ant - the pheromone concentration of paths with high coverage efficiency and low energy consumption increases, and vice versa. After iterations (such as 100 times), the optimal detection path parameters are obtained through convergence, including the optimal path node distribution density , scanning angle threshold , sensor sampling frequency . This set of parameters simultaneously satisfies the constraint conditions of coverage efficiency (such as 95%) and energy consumption (such as 500 J).

[0066] During the dynamic correction simulation and multi-objective optimization process, motion state parameters (such as the real-time position of the robot's end effector and speed) are fed back to the system in real time to adjust the spatial coordinates of path nodes and the simulation time step. For example, when the robot's motion speed exceeds the preset threshold , the system automatically increases the path node density to ensure the detection accuracy at high speeds; conversely, when the speed is low, the node density can be appropriately reduced to improve efficiency.

[0067] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0068] Although embodiments of the present invention have been shown and described, those of ordinary skill in the art will appreciate that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A non-destructive testing method for robots based on artificial intelligence, characterized in that, Including the following steps: Obtain multi-modal sensing data of the target to be detected and the motion state parameters of the end effector of the robot; Perform multi-dimensional scanning on the target and construct a three-dimensional material feature model, and perform dynamic defect feature analysis on the three-dimensional material feature model in combination with the multi-modal sensing data to generate an initial defect feature evaluation model; Load the material property distribution data and detection accuracy constraint conditions into the initial defect feature evaluation model to generate a comprehensive defect feature evaluation model, and perform dynamic simulation of the detection path in combination with the motion state parameters to obtain path regulation simulation data; Construct a dynamic defect prediction model and train it with the path regulation simulation data to obtain a detection path correction model, and then generate initial detection parameters, where the initial detection parameters at least include the coordinates of the initial detection points, the initial scanning path, and the initial correction response parameters; Based on the initial detection parameters, the comprehensive defect feature evaluation model, and the motion state parameters, obtain dynamic correction simulation information; Combine the dynamic correction simulation information with the detection task planning parameters, construct a multi-objective optimization model and iteratively optimize the planning parameters to obtain the optimal detection path parameters; Infer the detection path parameters for the current period based on the real-time material feature model, generate the safety detection probability for the current period, and combine the optimal detection path parameters for the current period with the actual scanning trajectory to adjust the real-time detection strategy of the robot to achieve the goal of dynamic path matching.

2. The method for non-destructive testing of a robot based on artificial intelligence according to claim 1, wherein, The path regulation simulation data at least includes material stress distribution, defect clustering regions, safety detection boundary sets, and dynamic correction priority sequences; The dynamic correction simulation information at least includes detection point offset, path coverage completeness rate, and accuracy stability evaluation indicators.

3. The method for non-destructive testing of a robot based on artificial intelligence according to claim 1, wherein The step of performing multi-dimensional scanning on the target and constructing a three-dimensional material feature model, and performing dynamic defect feature analysis on the three-dimensional material feature model in combination with the multi-modal sensing data to generate an initial defect feature evaluation model includes the following steps: Perform combined ray and acoustic wave scanning on the target to obtain a sensing data sequence of the internal defect distribution of the material; Preprocess the sensing data sequence to obtain standardized feature fusion data, where the preprocessing includes one or more of signal denoising, time domain registration, feature fusion, grid division, and data interpolation; Based on the standardized feature fusion data, in combination with the three-dimensional point cloud reconstruction technology, generate a three-dimensional material feature model; Perform dynamic defect feature analysis on the three-dimensional material feature model to generate an initial defect feature evaluation model, where the dynamic defect feature analysis at least includes grid density assignment and feature weight mapping, divides the defect risk levels through grid density assignment, and assigns corresponding sensing data parameters to each level region.

4. The method for non-destructive testing of a robot based on artificial intelligence according to claim 1, wherein The step of loading the material property distribution data and detection accuracy constraint conditions into the initial defect feature evaluation model to generate a comprehensive defect feature evaluation model, and performing dynamic simulation of the detection path in combination with the motion state parameters to obtain path regulation simulation data includes the following steps: Load the material property distribution data into the initial defect feature evaluation model to simulate real-time property changes and generate a first defect feature evaluation model; Load the detection accuracy constraint conditions into the first defect feature evaluation model to generate a comprehensive defect feature evaluation model, where the detection accuracy constraint conditions include the maximum detection angle limit, the minimum resolution threshold, and the sensor response redundancy parameter; Based on the comprehensive defect feature evaluation model and the motion state parameters, perform dynamic simulation of the detection path. The specific process includes: Construct a multi-objective path planning equation, which at least includes a coverage integrity equation, a trajectory coincidence equation, and a detection efficiency equation. Combine the comprehensive defect feature evaluation model and numerically solve the multi-objective path planning equation by the piecewise gradient descent method to obtain the material stress distribution, the defect clustering region, the set of safety detection boundaries, and the dynamic correction priority sequence; The generation of the set of safety detection boundaries includes the following steps: Based on the defect clustering region, calculate the defect coverage density of each grid cell; Identify the regions in the comprehensive defect feature evaluation model where the defect coverage density is not greater than the preset threshold, and generate the set of safety detection boundaries.

5. The method for non-destructive testing of a robot based on artificial intelligence according to claim 1, wherein The steps of constructing a dynamic defect prediction model and training it with the path regulation simulation data to obtain a detection path correction model, and then generating initial detection parameters include: Construct a dynamic defect prediction model based on the three-dimensional material feature model; Train and validate the dynamic defect prediction model with the path regulation simulation data to obtain a detection path correction model; Input the real-time motion state parameters into the detection path correction model to predict the set of safety detection boundaries; Based on the predicted set of safety detection boundaries, generate initial detection parameters, where the initial detection parameters at least include the initial detection point coordinates, the initial scanning path, and the initial correction response parameter.

6. The method for non-destructive testing of a robot based on artificial intelligence according to claim 5, wherein The steps of constructing a dynamic defect prediction model and training it with the path regulation simulation data to obtain a detection path correction model, and then generating initial detection parameters also include data reconstruction of the path regulation simulation data. Specifically: Based on the material stress distribution, the defect clustering region, and the set of safety detection boundaries, construct an initial multi-dimensional detection input tensor; Perform standardization and feature dimensionality reduction processing on the initial multi-dimensional detection input tensor to generate a final multi-dimensional detection input tensor; Based on the dynamic correction priority sequence, construct a detection path correction label tensor; Combine the final multi-dimensional detection input tensor and the detection path correction label tensor to form a training sample set.

7. The method for non-destructive testing of a robot based on artificial intelligence according to claim 6, characterized in that, The steps of generating initial detection parameters based on the predicted set of safety detection boundaries include: Extract the grid cell with the lowest defect coverage density in the predicted set of safety detection boundaries to generate the initial detection point coordinates; According to the spatial topological relationship of the predicted set of safety detection boundaries, fit the feasible connection structure of the initial scanning path to generate the initial correction response parameter; Calculate the material stress gradient direction of the predicted set of safety detection boundaries and normalize it into a path reference vector, and the path reference vector is the initial scanning path direction.

8. The method for non-destructive testing of a robot based on artificial intelligence according to claim 1, characterized in that, The steps of obtaining the dynamic correction simulation information based on the initial detection parameters, the comprehensive defect feature evaluation model, and the motion state parameters include: Map the initial detection point coordinates into the comprehensive defect feature evaluation model, match the initial scanning path and the initial correction response parameters, adjust the path node density of the correction area, update the comprehensive defect feature evaluation model, define the dynamic correction trigger conditions, path reconstruction rules and response parameter increments, and generate a dynamic correction simulation model; Based on the dynamic correction simulation model, use the piecewise gradient descent method to iteratively solve the multi-objective path planning equation to obtain dynamic correction simulation information, specifically including: When the dynamic correction trigger conditions are met, update the detection point offset, path coverage integrity rate and accuracy stability evaluation index, and re-solve the multi-objective path planning equation until the simulation termination conditions are reached; The dynamic correction trigger conditions include triggering the update of the response parameters when the current path coverage integrity rate is not greater than the preset integrity rate threshold; the path reconstruction rules include adjusting the initial scanning path based on the material stress gradient direction; the response parameter increment has a piecewise linear relationship with the current accuracy stability evaluation index.

9. The method for non-destructive testing of a robot based on artificial intelligence according to claim 1, wherein The steps of constructing a multi-objective optimization model and iteratively optimizing the planning parameters to obtain the optimal detection path parameters include: Construct a multi-objective optimization model, where the optimization variables include the path node distribution density, scanning angle threshold and sensor sampling frequency, the optimization objectives include maximizing the path coverage efficiency and minimizing the detection energy consumption, and the constraint conditions include the upper limit of robot kinematics and the sensor accuracy threshold; Use the dynamic programming algorithm to perform a preliminary solution on the multi-objective optimization model to generate an initial set of optimized paths; Based on the initial set of optimized paths, use the ant colony optimization algorithm for global optimization to generate the optimal detection path parameters.

10. An artificial intelligence-based robot non-destructive testing system, characterized in that, Including: A data acquisition module for obtaining multi-modal sensing data of the target to be detected and the motion state parameters of the robot end effector; A feature modeling and analysis module for performing multi-dimensional scanning on the target and constructing a three-dimensional material feature model, and dynamically analyzing the defect features of the three-dimensional material feature model in combination with the multi-modal sensing data to generate an initial defect feature evaluation model; A comprehensive evaluation and simulation module for loading the material property distribution data and detection accuracy constraint conditions into the initial defect feature evaluation model to generate a comprehensive defect feature evaluation model, and performing dynamic simulation of the detection path in combination with the motion state parameters to obtain path regulation simulation data; A parameter generation module for constructing a dynamic defect prediction model and training it with the path regulation simulation data to obtain a detection path correction model, and then generating initial detection parameters, where the initial detection parameters at least include initial detection point coordinates, an initial scanning path and initial correction response parameters; A dynamic correction simulation module for obtaining dynamic correction simulation information based on the initial detection parameters, the comprehensive defect feature evaluation model and the motion state parameters; A path optimization module for constructing a multi-objective optimization model in combination with the dynamic correction simulation information and the detection task planning parameters and iteratively optimizing the planning parameters to obtain the optimal detection path parameters; A strategy adjustment module is used to infer the detection path parameters of the current period based on the real-time material feature model, generate the safety detection probability of the current period, and combine the optimal detection path parameters of the current period with the actual scanning trajectory to adjust the real-time detection strategy of the robot to achieve the goal of dynamic path matching.

Images