Fusion control method and system for mechanical arm flaw detection operation during robot dog climbing

Abstract

The invention discloses a fusion control method and system for mechanical arm flaw detection operation during robot dog climbing, and relates to the related field of robot control, and the method comprises the steps: building a climbing feature sequence when the position of a robot dog meets a detection working position; performing multi-dimensional injury condition prediction on a to-be-detected target to obtain target injury condition prediction distribution, and performing arm-out operation control analysis on the mechanical arm; performing multiple rounds of propagation evolution optimization on the arm-out operation control space to generate an arm-out control optimization strategy; interference triggering prediction is carried out on the robot dog, optimization compensation is carried out on the robot dog according to an interference triggering prediction result, and an interference compensation strategy is obtained; and performing flaw detection fusion control on the to-be-detected target, and executing flaw detection control closed-loop correction according to the flaw detection process induction data. The technical problems that in existing flaw detection work, climbing and flaw detection are lack of a dynamic cooperation mechanism, the environmental adaptability is poor, and the flaw detection precision is low are solved, and the technical effect that the environmental adaptability and the flaw detection precision are improved through climbing and flaw detection cooperation optimization is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of robot control, in particular to a fusion control method and system for robot dog climbing and robot arm defect detection operation. BACKGROUND

[0002] The defect detection operation of the robot dog in a complex environment is crucial for industrial detection and disaster rescue, and its efficiency and accuracy directly affect the task execution effect and safety. The main method to solve this problem at present is to use a phased control strategy, that is, first complete the robot dog climbing through fixed path planning, and then independently control the robot arm to perform defect detection operation. Since the coupling between climbing and defect detection operation is not fully considered, the existing method is prone to cause the robot dog to be disturbed by the movement of the robot arm in a complex environment, resulting in a decrease in defect detection accuracy, and cannot adapt to environmental changes in real time, resulting in control delay and error accumulation.

[0003] In the related art at present, the defect detection operation of the robot dog in a complex environment lacks a dynamic coordination mechanism between climbing and defect detection, has poor environmental adaptability, and has low defect detection accuracy. SUMMARY

[0004] The present application provides a fusion control method and system for robot dog climbing and robot arm defect detection operation. The climbing characteristics sequence of the robot dog is constructed after the robot dog is positioned, the target defect is predicted synchronously, and the robot arm operation space is analyzed. The optimal arm strategy is generated through multi-round evolution optimization, the movement disturbance of the robot arm is predicted in combination with the climbing characteristics, and real-time compensation is performed. Finally, the coordinated adaptive control of climbing and defect detection is realized through defect data closed-loop feedback, etc. Technical means solve the technical problems of the existing robot dog in a complex environment, such as lack of dynamic coordination mechanism between climbing and defect detection, poor environmental adaptability, and low defect detection accuracy. The technical effect of optimizing climbing and defect detection is achieved to improve environmental adaptability and defect detection accuracy.

[0005] The application provides a fusion control method for robot dog climbing and robot arm defect detection operation, comprising the following steps: when the position of the robot dog meets the detection working position of a to-be-detected target, a climbing feature sequence of the robot dog is constructed; multi-dimensional defect condition prediction is performed on the to-be-detected target to obtain a target defect condition prediction distribution, and arm operation control analysis is performed on the robot arm based on the target defect condition prediction distribution to obtain an arm operation control space; multi-round reproduction evolution optimization is performed on the arm operation control space according to a defect detection operation evaluation model to generate an arm control optimization strategy; interference trigger prediction is performed on the robot dog according to the arm control optimization strategy based on the climbing feature sequence to obtain an interference trigger prediction result, and optimization compensation is performed on the robot dog according to the interference trigger prediction result to obtain an interference compensation strategy; defect detection fusion control is performed on the to-be-detected target based on the arm control optimization strategy and the interference compensation strategy to obtain defect detection process sensing data, and defect detection control closed-loop correction is performed according to the defect detection process sensing data.

[0006] In possible implementation manners, the to-be-detected target is subjected to multi-dimensional defect condition prediction to obtain a target defect condition prediction distribution, and the following processing is performed: defect detection category mining is performed on the to-be-detected target to obtain a defect detection category set; defect condition prediction model of each category is obtained through defect condition record training based on the defect detection category set; a multi-dimensional defect condition prediction channel satisfying a distillation loss constraint is built through distillation loss iterative training according to the defect condition prediction model of each category; and defect condition prediction fusion is performed on the to-be-detected target according to the multi-dimensional defect condition prediction channel to generate the target defect condition prediction distribution.

[0007] In possible implementation manners, multi-round reproduction evolution optimization is performed on the arm operation control space according to a defect detection operation evaluation model to generate an arm control optimization strategy, and the following processing is performed: evaluation constraint optimization is performed on the arm operation control space according to the defect detection operation evaluation model to generate an arm control optimization space; a defect detection operation fitness analysis model is established through weight configuration of a defect detection operation evaluation multi-dimensional index of the defect detection operation evaluation model, the defect detection operation evaluation multi-dimensional index including defect detection signal quality, defect detection contact stability and defect detection contact smoothness; N arm control evolution spaces are generated through N rounds of reproduction evolution according to the arm control optimization space based on the defect detection operation fitness analysis model and the defect detection operation evaluation model, N being a positive integer greater than 1; joint optimization is performed on the arm control optimization space and the N arm control evolution spaces according to the defect detection operation fitness analysis model to obtain the arm control optimization strategy.

[0008] In a possible implementation, the out-arm operation control space is evaluated and constrained for optimization according to the flaw detection operation evaluation model, an out-arm control optimization space is generated, and the following processing is performed: an out-arm operation control first decision is extracted according to the out-arm operation control space, and flaw detection operation simulation is performed based on the out-arm operation control first decision to obtain first flaw detection fitting data; the first flaw detection fitting data is input into the flaw detection operation evaluation model to obtain a first flaw detection operation evaluation result; a flaw detection operation evaluation constraint condition is constructed based on the flaw detection operation evaluation multi-dimensional index; it is judged whether the first flaw detection operation evaluation result satisfies the flaw detection operation evaluation constraint condition; and if the first flaw detection operation evaluation result satisfies the flaw detection operation evaluation constraint condition, the out-arm operation control first decision is added to the out-arm control optimization space.

[0009] In a possible implementation, N out-arm control evolution spaces are generated by performing N rounds of reproduction evolution on the out-arm control optimization space based on the flaw detection operation fitness analysis model and the flaw detection operation evaluation model, and the following processing is performed: flaw detection operation fitness calculation is performed on the out-arm control optimization space according to the flaw detection operation fitness analysis model to establish a flaw detection operation fitness distribution; reproduction capacity distribution is allocated to each out-arm operation control decision in the out-arm control optimization space according to the flaw detection operation fitness distribution based on a reproduction capacity constraint to obtain a first round of reproduction capacity distribution; the out-arm control optimization space is subjected to adaptive mutation based on the first round of reproduction capacity distribution to generate an out-arm control first round of reproduction space; the out-arm control first round of reproduction space is evaluated and constrained for optimization according to the flaw detection operation evaluation model to generate a first out-arm control evolution space; and the first out-arm control evolution space is used as a basis to continue performing N-1 rounds of reproduction evolution based on the flaw detection operation fitness analysis model and the flaw detection operation evaluation model until the N out-arm control evolution spaces are obtained.

[0010] In a possible implementation, interference trigger prediction of the robot dog is performed according to the out-arm control optimization strategy based on the climbing feature sequence to obtain an interference trigger prediction result, and the following processing is performed: an out-arm control time window is determined based on the out-arm control optimization strategy; a posture feature prediction distribution is generated by performing posture prediction on the robot dog according to the climbing feature sequence based on the out-arm control time window; interference impact trigger analysis is performed on the posture feature prediction distribution according to the out-arm control optimization strategy to obtain an interference impact trigger analysis result; interference risk trigger analysis is performed on the posture feature prediction distribution according to the out-arm control optimization strategy to obtain an interference risk trigger analysis result, and the interference trigger prediction result is generated in combination with the interference impact trigger analysis result.

[0011] In a possible implementation, the method further includes: performing abnormality detection according to the detection process sensing data to obtain a detection process abnormality feature; performing correlation influence analysis on the robot dog according to the detection process abnormality feature to generate an abnormality correlation influence first feature; performing correlation influence analysis on the mechanical arm according to the detection process abnormality feature to generate an abnormality correlation influence second feature; and performing collaborative optimization on the arm control optimization strategy and the interference compensation strategy according to the abnormality correlation influence first feature and the abnormality correlation influence second feature to generate a detection fusion control optimization strategy.

[0012] In a possible implementation, when the robot dog position meets a detection work position of a target to be detected, a climbing feature sequence of the robot dog is constructed, and the following processing is performed: when the robot dog position meets the detection work position, multi-modal feature collection is performed on the robot dog to obtain a robot dog feature set; and the robot dog feature set is cleaned and combed to generate the climbing feature sequence.

[0013] In a possible implementation, a climbing feature sequence of the robot dog is constructed, and the following processing is performed: instability risk evaluation is performed on the robot dog according to the climbing feature sequence to generate an instability risk coefficient; and if the instability risk coefficient is greater than or equal to an instability risk threshold, a robot dog instability early warning signal is generated.

[0014] The application also provides a fusion control system for mechanical arm detection operation when a robot dog is climbing, including: a climbing feature sequence construction module, configured to construct a climbing feature sequence of the robot dog when a robot dog position meets a detection work position of a target to be detected; an arm operation control analysis module, configured to perform multi-dimensional damage prediction on the target to be detected to obtain a target damage prediction distribution, and perform arm operation control analysis on a mechanical arm based on the target damage prediction distribution to obtain an arm operation control space; an arm control optimization module, configured to perform multi-round reproduction evolution optimization on the arm operation control space according to a detection operation evaluation model to generate an arm control optimization strategy; an interference compensation module, configured to perform interference trigger prediction on the robot dog according to the arm control optimization strategy based on the climbing feature sequence to obtain an interference trigger prediction result, and perform optimization compensation on the robot dog according to the interference trigger prediction result to obtain an interference compensation strategy; and a detection fusion control module, configured to perform detection fusion control on the target to be detected based on the arm control optimization strategy and the interference compensation strategy to obtain detection process sensing data, and perform detection control closed-loop correction according to the detection process sensing data.

[0015] The machine dog climbing mechanical arm flaw detection operation fusion control method and system provided in the application first constructs a climbing feature sequence of the machine dog when the position of the machine dog meets the detection work position of the target to be detected, then performs multi-dimensional damage prediction on the target to be detected to obtain a target damage prediction distribution, and performs out-arm operation control analysis on the mechanical arm based on the target damage prediction distribution to obtain an out-arm operation control space. Next, the out-arm operation control space is subjected to multi-round reproduction evolution optimization according to a flaw detection operation evaluation model to generate an out-arm control optimization strategy. Then, the machine dog is subjected to interference trigger prediction according to the out-arm control optimization strategy based on the climbing feature sequence to obtain an interference trigger prediction result, and the machine dog is subjected to optimization compensation according to the interference trigger prediction result to obtain an interference compensation strategy. Finally, the target to be detected is subjected to flaw detection fusion control based on the out-arm control optimization strategy and the interference compensation strategy to obtain flaw detection process sensing data, and flaw detection control closed-loop correction is performed according to the flaw detection process sensing data. The technical effect of improving environmental adaptability and flaw detection precision through climbing and flaw detection collaborative optimization is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings of the embodiments of the application will be briefly introduced below. In the present application, a flowchart is used to illustrate the operations performed by the system according to the embodiments of the application. It should be understood that the foregoing or the following operations are not necessarily performed in sequence. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can be added to these processes, or a step or several steps of operation can be removed from these processes.

[0017] Figure 1 The flowchart of the machine dog climbing mechanical arm flaw detection operation fusion control method provided by the embodiments of the application.

[0018] Figure 2 The structural diagram of the machine dog climbing mechanical arm flaw detection operation fusion control system provided by the embodiments of the application.

[0019] The reference signs are explained as follows: a climbing feature sequence construction module 10, an out-arm operation control analysis module 20, an out-arm control optimization module 30, an interference compensation module 40, and a flaw detection fusion control module 50. DETAILED DESCRIPTION

[0020] The above description is only a summary of the technical solutions of the application. In order to more clearly understand the technical means of the application, the embodiments of the application can be implemented according to the content of the description, and in order to make the above and other purposes, features and advantages of the application more obvious and easy to understand, the specific embodiments of the application are described as follows.

[0021] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings, and the described embodiments should not be regarded as limiting the present application. All other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0022] In the following description, "some embodiments" are referred to, which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. The term "first\second" referred to only distinguishes similar objects, and does not represent a specific order of the objects. The terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or modules not clearly listed or inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by those skilled in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application.

[0023] The embodiments of the present application provide a fusion control method for robot dog climbing and robot arm defect detection operation, as shown in Figure 1 The method comprises the following steps. Step S100, when the position of the robot dog meets the detection work position of the to-be-detected target, a climbing feature sequence of the robot dog is constructed.

[0024] Specifically, the motion data of the robot dog in the climbing process is collected in real time by using various sensors (such as an accelerometer, a gyroscope, a visual sensor, etc.) installed on the robot dog. The collected sensor data is preprocessed by a signal processing algorithm (such as filtering, noise reduction), and then the features capable of representing the climbing state of the robot dog, such as speed, acceleration, attitude angle, etc., are extracted from the preprocessed data by using a feature extraction algorithm (such as wavelet transform, principal component analysis, etc.). The extracted features are arranged in time sequence to form a climbing feature sequence.

[0025] For example, after the acceleration data collected by the accelerometer in the climbing process is filtered, the peak value features of the acceleration are extracted by using wavelet transform, and these peak value features are arranged in time sequence to form a climbing feature sequence.

[0026] In one possible implementation, when the position of the robot dog satisfies the detection working position of the target to be detected, a climbing feature sequence of the robot dog is constructed, and step S100 further includes step S110: when the position of the robot dog satisfies the detection working position, multi-modal feature collection is performed on the robot dog to obtain a robot dog feature set. Specifically, a plurality of sensors are installed on the robot dog, including attitude sensors (such as a gyroscope and an accelerometer), force sensors (such as foot end pressure sensors), visual sensors (such as a camera), surface contact sensors, and the like, for collecting multi-modal features of the robot dog. All sensors on the robot dog are started, and when the position of the robot dog satisfies the detection working position, a data collection module is triggered to synchronously collect multi-modal data such as attitude angles, center of gravity positions, foot end contact forces, climbing surface inclinations, material types, and surface structure features, and the collected data is stored as a robot dog feature set.

[0027] For example, during the climbing process of the robot dog, the attitude sensors synchronously collect attitude angles and center of gravity positions of the robot dog, the foot end pressure sensors collect foot end contact forces, the visual sensors collect images of the climbing surface, and the surface contact sensors collect material types and surface structure features of the climbing surface. These data are synchronously collected and stored as a robot dog feature set.

[0028] Step S120: cleaning the robot dog feature set to generate the climbing feature sequence. Specifically, abnormal value detection is performed on the collected robot dog feature set to remove abnormal data, interpolation processing is performed on missing data to ensure the integrity of the data. Key features are extracted from the cleaned data by a feature extraction algorithm (such as principal component analysis PCA), and the climbing feature sequence is generated in time sequence.

[0029] For example, the foot end contact force of some data points in the collected robot dog feature set may be abnormally high, which is removed by an abnormal value detection algorithm. For missing data points, a linear interpolation method is used for filling. Then, the PCA algorithm is used to extract key features such as attitude angles, center of gravity positions, and foot end contact forces from the cleaned data, and the climbing feature sequence is generated in time sequence.

[0030] This implementation ensures that various state data of the robot dog during the climbing process can be completely collected through multi-modal feature collection, providing rich information for subsequent control and decision-making. The data cleaning and feature extraction process effectively removes noise and abnormal data, improves the quality and accuracy of the data, and makes the generated climbing feature sequence more truly reflect the climbing state of the robot dog. Through the collection of the material type and surface structure feature of the climbing surface, the robot dog can better adapt to different climbing environments, improving the adaptability and robustness of the robot dog.

[0031] In a possible implementation, the sequence of climbing features of the robot dog is constructed, and step S100 further includes step S130 of evaluating the instability risk of the robot dog according to the sequence of climbing features to generate an instability risk coefficient. Specifically, a risk evaluation model based on machine learning or statistical analysis is constructed, the sequence of climbing features is input, and the instability risk coefficient is output. The model can be trained based on historical data to identify key features and patterns that lead to instability. According to the actual application scenario and safety requirements, an instability risk threshold is set to determine whether the robot dog is in an instability risk state.

[0032] For example, the instability risk evaluation model can be a support vector machine (SVM) based classifier that calculates the instability risk coefficient based on features such as posture angle, center of gravity position, and foot contact force in the sequence of climbing features. If the instability risk coefficient exceeds the set threshold (e.g., 0.8), the robot dog is considered to be in a high-risk state.

[0033] Step S140, if the instability risk coefficient is greater than or equal to the instability risk threshold, a robot dog instability warning signal is generated. Specifically, it is checked whether the instability risk coefficient is greater than or equal to the instability risk threshold. If the condition is met, the warning signal generation module is triggered to generate an instability warning signal, and the signal is transmitted to the control system or operator through a suitable communication method (such as a wireless communication module).

[0034] For example, when the instability risk coefficient reaches 0.8, the warning signal generation module is triggered to generate a red warning signal, and the signal is sent to the operator of the robot dog through the wireless communication module, prompting the operator to take measures to avoid instability of the robot dog.

[0035] This implementation can detect the instability risk of the robot dog in time by calculating the instability risk coefficient and generating a warning signal in real time, take measures in advance to avoid accidents, enhance the safety of the robot dog in complex climbing environments, and improve the reliability and stability of the robot dog.

[0036] Step S200, performing multi-dimensional injury prediction on the target to be detected to obtain a target injury prediction distribution, and performing out-arm operation control analysis on the robot arm based on the target injury prediction distribution to obtain an out-arm operation control space.

[0037] Specifically, a multi-dimensional damage prediction model is constructed based on a machine learning algorithm (such as a convolutional neural network CNN in deep learning), and the damage condition (such as cracks, corrosion, etc.) of the target to be detected is predicted in multiple dimensions, including the position, size, depth, etc. of the damage. Collect multi-source data (such as images, ultrasonic data) of the target to be detected, input the collected data into the multi-dimensional damage prediction model, and obtain the target damage prediction distribution. Using kinematics and dynamics models, combined with the target damage prediction distribution, the arm operation control space of the robot arm is calculated, that is, the set of motion trajectories and postures that the robot arm can execute.

[0038] For example, the image of the surface of the target to be detected is analyzed using a CNN model, and the crack position and depth distribution of the surface of the target to be detected are predicted. Then, according to the crack distribution, the position and posture of each detection point that the robot arm needs to reach are calculated using the dynamics model of the robot arm, forming the arm operation control space.

[0039] In one possible implementation, the multi-dimensional damage prediction of the target to be detected is performed to obtain the target damage prediction distribution, and step S200 further includes step S210. The damage detection category mining of the target to be detected is performed to obtain a damage detection category set. Specifically, historical damage detection data is collected, including damage detection images, ultrasonic data, and X-ray detection data. Different damage detection categories are mined from the historical damage detection data using clustering analysis (such as K-means) and association rule mining, and the mined categories are audited and supplemented in combination with the experience and knowledge of experts in the field, to form a damage detection category set. For example, the categories mined from the historical damage detection data include cracks, corrosion, and holes, which are classified by clustering analysis algorithm and audited in combination with expert knowledge, to finally form a damage detection category set.

[0040] Step S220, based on the damage detection category set, the damage record training is performed to obtain a damage prediction model for each category. Specifically, the historical damage detection data is labeled to clearly indicate the damage category and features corresponding to each data sample. A supervised learning algorithm (such as a support vector machine SVM and a convolutional neural network CNN) is used to train the damage record of each damage detection category, and a damage prediction model for each category is generated. For example, CNN is used to train the crack category, and SVM is used to train the corrosion category. Through the labeled historical damage detection data, damage prediction models for cracks and corrosion are trained.

[0041] Step S230, according to the various types of injury prediction model, the distillation loss iterative training is carried out, and a multi-dimensional injury prediction channel meeting the distillation loss constraint is built. Specifically, knowledge distillation is a model compression technique that migrates the knowledge of a complex model (teacher model) to a simple model (student model) to improve the performance of the student model. In this step, the various types of injury prediction model is taken as the teacher model, and the learning of the student model is guided by the distillation loss function. A lightweight neural network architecture, such as MobileNet, is selected as the student model. The distillation loss function consists of two parts: the cross-entropy loss between the predicted output of the student model and the true label, and the KL divergence loss between the predicted output of the student model and the predicted output of the teacher model. In each iteration, the student model is trained using the training dataset, the distillation loss function is calculated, and the parameters of the student model are updated through backpropagation. This process is repeated until the performance of the student model converges or the preset number of iterations is reached.

[0042] For example, assume there are three teacher models corresponding to crack, corrosion, and hole injury prediction. MobileNet is selected as the student model, the distillation loss function is defined, and iterative training is performed. In each iteration, the KL divergence loss between the predicted output of the student model and the predicted output of the teacher model is minimized, while the cross-entropy loss between the predicted output of the student model and the true label is also minimized. Through multiple iterations, the student model gradually learns the knowledge of the teacher model, and finally forms a multi-dimensional injury prediction channel that can comprehensively predict multiple types of injury.

[0043] Step S240, according to the multi-dimensional injury prediction channel, the injury prediction fusion of the target to be detected is performed, and the target injury prediction distribution is generated. Specifically, the multi-source data (such as images, ultrasonic data, etc.) of the target to be detected is input into the multi-dimensional injury prediction channel, and the multi-dimensional injury prediction channel outputs the prediction results of multiple types of injury, each prediction result of injury includes probability distribution and specific type. The output of the multi-dimensional injury prediction channel is fused using a fusion algorithm, where the fusion algorithm can be based on weighted average, voting mechanism or other more complex fusion strategies. For each position, the specific injury type is determined according to the fused probability distribution. For example, the injury type with the highest probability is selected as the final prediction result of that position, and a comprehensive target injury prediction distribution is finally output, which contains not only the probability of injury but also the specific injury type.

[0044] For example, assume there are three injury categories: crack, corrosion, and hole. The output of the multi-dimensional injury prediction channel is shown in Table 1.

[0045] Table 1: Output example of multi-dimensional injury prediction channel

[0046] The selected weighted average method is used for fusion, and the weights are 0.4, 0.3 and 0.3, respectively. The fusion probability at position (10, 10) is: 0.4x0.8+0.3x0.6+0.3x0.4=0.62, the fusion probability at position (20, 20) is: 0.4x0.7+0.3x0.5+0.3x0.3=0.52, and the fusion probability at position (30, 30) is: 0.4x0.6+0.3x0.4+0.3x0.2=0.42. At position (10, 10), the predicted probability of crack is the highest (0.8), so the final prediction result of this position is crack. At position (20, 20), the predicted probability of crack is the highest (0.7), so the final prediction result of this position is crack. At position (30, 30), the predicted probability of crack is the highest (0.6), so the final prediction result of this position is crack. Therefore, the final output of the comprehensive injury prediction distribution is shown in Table 2.

[0047] Table 2: Example of comprehensive injury prediction distribution

[0048] This implementation migrates the knowledge of multiple teacher models to a lightweight student model through knowledge distillation technology, reducing the computational complexity of the model and improving the performance of the student model. By fusing the outputs of the multi-dimensional injury prediction channels through a fusion algorithm, various injury characteristics can be considered to generate a more accurate target injury prediction distribution.

[0049] Step S300, according to the flaw detection operation evaluation model, the out-arm operation control space is subjected to multi-round reproduction evolution optimization, and an out-arm control optimization strategy is generated.

[0050] Specifically, an evolutionary algorithm such as genetic algorithm (GA) is used to perform multi-round reproduction evolution optimization on the out-arm operation control space to generate an out-arm control optimization strategy. The out-arm control optimization strategy is the optimal robot motion control strategy found from the out-arm operation control space.

[0051] First, the out-arm control strategy population is initialized, and then a flaw detection operation evaluation model is constructed to evaluate the pros and cons of different out-arm control strategies, such as detection accuracy, operation time, etc. The flaw detection operation evaluation model is used to evaluate each strategy in the population and calculate the fitness. According to the fitness, selection, crossover and mutation operations are performed to generate a new population. Repeat the above process for multiple rounds until the convergence condition is met, and the optimal out-arm control strategy is obtained.

[0052] For example, a plurality of different arm control strategies are included in the initial population, each corresponding to a motion trajectory of the robot arm. The detection accuracy and operation time of these strategies are evaluated by the flaw detection operation evaluation model, and the strategies with high fitness are selected for crossover and mutation. After multiple iterations, the optimal arm control strategy is obtained.

[0053] In one possible implementation, the arm operation control space is subjected to multiple rounds of reproductive evolution optimization according to the flaw detection operation evaluation model, and an arm control optimization strategy is generated. Step S300 further includes step S310 of evaluating and constraining optimization of the arm operation control space according to the flaw detection operation evaluation model, and generating an arm control optimization space. Specifically, the arm operation control space is initialized, and the parameter range of the arm operation control space is defined, such as the motion trajectory, speed, and force of the robot arm. The flaw detection operation evaluation model is applied, which evaluates each candidate solution in the arm operation control space according to the multi-dimensional indicators of the flaw detection operation (flaw detection signal quality, flaw detection contact stability, and flaw detection contact smoothness), and calculates the flaw detection signal quality, flaw detection contact stability, and flaw detection contact smoothness. According to actual requirements, the threshold of each evaluation indicator is set, for example, the evaluation constraints are set as follows: the flaw detection signal quality is not less than 0.7, the flaw detection contact stability is not less than 0.6, and the flaw detection contact smoothness is not less than 0.5. An optimization algorithm (such as a genetic algorithm) is used to find solutions that satisfy the evaluation constraints in the arm operation control space, and an arm control optimization space is generated.

[0054] Step S320, the flaw detection operation evaluation multi-dimensional indicators of the flaw detection operation evaluation model are configured with weights, a flaw detection operation fitness analysis model is established, and the flaw detection operation evaluation multi-dimensional indicators include flaw detection signal quality, flaw detection contact stability, and flaw detection contact smoothness. Specifically, according to actual requirements and experience, the flaw detection signal quality, flaw detection contact stability, and flaw detection contact smoothness are assigned weights, for example, the weights are 0.5, 0.3, and 0.2 respectively. A comprehensive evaluation model is established, and the weighted sum of the multi-dimensional indicators is taken as the fitness of the flaw detection operation.

[0055] For example, for a candidate solution, the flaw detection signal quality is 0.8, the flaw detection contact stability is 0.7, and the flaw detection contact smoothness is 0.6. Then the fitness of the candidate solution is: fitness = 0.5 × 0.8 + 0.3 × 0.7 + 0.2 × 0.6 = 0.73.

[0056] Step S330, based on the flaw detection operation optimization and resolution model and the flaw detection operation evaluation model, N rounds of breeding evolution are performed according to the out-arm control optimization space to generate N out-arm control evolution spaces, and N is a positive integer greater than 1. Specifically, a population is initialized in the out-arm control optimization space, and a genetic algorithm or other evolution algorithm is used to perform multiple rounds of breeding evolution on the out-arm control optimization space. Each round includes selection, crossover, and mutation operations. After each round of breeding evolution, a new out-arm control evolution space is generated.

[0057] Step S340, according to the flaw detection operation optimization and resolution model, joint optimization is performed on the out-arm control optimization space and the N out-arm control evolution spaces to obtain the out-arm control optimization strategy. Specifically, all candidate solutions in the out-arm control optimization space and the multiple out-arm control evolution spaces are comprehensively evaluated, the optimization degree of each candidate solution is calculated, and the candidate solution with the highest optimization degree is selected as the out-arm control optimization strategy. For example, assuming that 3 rounds of breeding evolution are performed to generate 3 out-arm control evolution spaces, each space containing 10 candidate solutions. These candidate solutions and the candidate solutions in the initial out-arm control optimization space are comprehensively evaluated, and the candidate solution with the highest optimization degree is selected as the out-arm control optimization strategy.

[0058] This implementation generates the optimal out-arm control strategy through multiple rounds of breeding evolution and joint optimization, improving the efficiency and accuracy of flaw detection operation. Through the flaw detection operation optimization and resolution model, the flaw detection signal quality, flaw detection contact stability, and flaw detection contact smoothness are comprehensively considered to ensure the comprehensiveness and reliability of the control strategy.

[0059] In one possible implementation, according to the flaw detection operation evaluation model, the out-arm operation control space is evaluated and constrained to optimize, and the out-arm control optimization space is generated. Step S310 further includes step S311, an out-arm operation control first decision is extracted from the out-arm operation control space, and based on the out-arm operation control first decision, a flaw detection operation simulation is performed to obtain first flaw detection fitting data. Specifically, the feasible range of parameters such as the motion trajectory, speed, and force of the robot arm is defined, and an initial out-arm operation control decision is randomly selected or selected based on a certain strategy from the out-arm operation control space. Using simulation software or models, flaw detection operation simulation is performed according to the extracted out-arm operation control decision to generate first flaw detection fitting data.

[0060] Step S312, the first flaw detection fitting data is input into the flaw detection operation evaluation model to obtain the first flaw detection operation evaluation result. Specifically, the first flaw detection fitting data is input into the flaw detection operation evaluation model, which evaluates the flaw detection operation according to multiple-dimensional indicators such as flaw detection signal quality, flaw detection contact stability, and flaw detection contact smoothness.

[0061] Step S313, constructing a flaw detection operation evaluation constraint condition based on the multi-dimensional index evaluation of the flaw detection operation; step S314, judging whether the first flaw detection operation evaluation result meets the flaw detection operation evaluation constraint condition; step S315, if the first flaw detection operation evaluation result meets the flaw detection operation evaluation constraint condition, adding the first decision of the out-arm operation control to the out-arm control optimization space.

[0062] Specifically, according to the actual demand, a threshold is set for each evaluation index as the flaw detection operation evaluation constraint condition. The scores of flaw detection signal quality, flaw detection contact stability and flaw detection contact smoothness are compared one by one to see whether they meet the set threshold. If all the indexes meet the threshold, it is considered that the evaluation result meets the constraint condition; otherwise, it does not meet the constraint condition. If the first flaw detection operation evaluation result meets the evaluation constraint condition, the corresponding out-arm operation control decision is added to the out-arm control optimization space as the basis for subsequent optimization.

[0063] This implementation gradually extracts the out-arm operation control decision, simulates, evaluates and judges, gradually filters out the out-arm operation control decision that meets the evaluation constraint condition, and provides a basis for subsequent optimization.

[0064] In a possible implementation, based on the flaw detection operation optimization fitness analysis model and the flaw detection operation evaluation model, N rounds of breeding evolution are performed according to the out-arm control optimization space to generate N out-arm control evolution spaces, and step S330 further includes step S331, performing flaw detection operation optimization fitness calculation on the out-arm control optimization space according to the flaw detection operation optimization fitness analysis model to establish a flaw detection operation optimization fitness distribution. Specifically, for each out-arm operation control decision in the out-arm control optimization space, the flaw detection operation optimization fitness analysis model is used to calculate its optimization fitness, and the optimization fitness of all out-arm operation control decisions is recorded to form a flaw detection operation optimization fitness distribution.

[0065] For example, assume that there are 3 out-arm operation control decisions in the out-arm control optimization space, and their flaw detection signal quality, flaw detection contact stability, and flaw detection contact smoothness are as follows: out-arm operation control decision 1: flaw detection signal quality 0.8, flaw detection contact stability 0.7, flaw detection contact smoothness 0.6; out-arm operation control decision 2: flaw detection signal quality 0.7, flaw detection contact stability 0.6, flaw detection contact smoothness 0.5; out-arm operation control decision 3: flaw detection signal quality 0.9, flaw detection contact stability 0.8, flaw detection contact smoothness 0.7. The weights are 0.5, 0.3, and 0.2 respectively, and the calculation of the fitness is as follows: out-arm operation control decision 1: 0.5*0.8+0.3*0.7+0.2*0.6=0.73, out-arm operation control decision 2: 0.5*0.7+0.3*0.6+0.2*0.5=0.63, out-arm operation control decision 3: 0.5*0.9+0.3*0.8+0.2*0.7=0.83. Therefore, the flaw detection operation fitness distribution is shown in Table 3.

[0066] Table 3: Example of flaw detection operation fitness distribution

[0067] In step S332, based on the breeding capacity constraint, the breeding capacity distribution of each out-arm operation control decision in the out-arm control optimization space is allocated according to the flaw detection operation fitness distribution, and the first round of breeding capacity distribution is obtained. Specifically, the upper limit of the breeding capacity of each out-arm operation control decision is set to ensure reasonable allocation of resources, for example, each out-arm operation control decision can have at most 3 offspring. According to the flaw detection operation fitness distribution, the breeding capacity of each out-arm operation control decision is allocated, and the out-arm operation control decision with higher flaw detection operation fitness is allocated more breeding capacity.

[0068] For example, assume that the upper limit of the breeding capacity is 3, and the breeding capacity is allocated according to the flaw detection operation fitness distribution as follows: out-arm operation control decision 1: fitness 0.73, 2 breeding capacities are allocated; out-arm operation control decision 2: fitness 0.63, 1 breeding capacity is allocated; out-arm operation control decision 3: fitness 0.83, 3 breeding capacities are allocated.

[0069] In step S333, the out-arm control first round of breeding space is generated by adaptively mutating the out-arm control optimization space based on the first round of breeding capacity distribution. Specifically, according to the first round of breeding capacity distribution, the out-arm operation control decisions that need to be mutated are selected, and a new out-arm operation control decision is generated by performing mutation operation on each selected out-arm operation control decision. The mutation operation can include adjusting the parameters such as motion trajectory, speed, and force. The out-arm control first round of breeding space contains these new out-arm operation control decisions.

[0070] Step S334, according to the flaw detection operation evaluation model, the first out-arm control breeding space is evaluated and constrained optimization is performed to generate a first out-arm control evolution space. Specifically, according to the flaw detection operation evaluation model, each out-arm operation control decision in the first out-arm control breeding space is evaluated, and the flaw detection signal quality, flaw detection contact stability, and flaw detection contact smoothness are calculated. According to the set threshold, the out-arm operation control decisions that meet the evaluation constraints are screened.

[0071] Step S335, based on the flaw detection operation optimization solution analysis model and the flaw detection operation evaluation model, the first out-arm control evolution space is continued to be bred for N-1 rounds until the N out-arm control evolution spaces are obtained. Specifically, steps S331 to S334 are repeated to perform flaw detection operation optimization calculation, breeding capacity allocation, adaptive mutation, and evaluation constraint optimization on the first out-arm control evolution space to generate a second out-arm control evolution space. The above process is repeated until N rounds of breeding evolution are completed to generate N out-arm control evolution spaces.

[0072] This implementation performs adaptive mutation according to the flaw detection operation optimization distribution, ensuring the diversity and adaptability of the out-arm operation control decisions. Through the flaw detection operation evaluation model and the evaluation constraints, the scientificity and reliability of each out-arm operation control decision are ensured.

[0073] Step S400, based on the climbing feature sequence, the out-arm control optimization strategy is used to predict the interference trigger of the robot dog, obtain an interference trigger prediction result, and perform optimization compensation on the robot dog according to the interference trigger prediction result to obtain an interference compensation strategy.

[0074] Specifically, based on the climbing feature sequence of the robot dog and the out-arm control optimization strategy, an interference trigger prediction model is constructed to predict the interference that the robot dog may receive when performing the out-arm operation, such as posture deviation and motion instability. According to the interference trigger prediction result, an optimization algorithm (such as particle swarm optimization PSO) is used to generate an interference compensation strategy to compensate for the motion of the robot dog to reduce the impact of interference.

[0075] For example, the interference trigger prediction model predicts that when the robot arm performs the out-arm operation, the robot dog may deviate due to the unevenness of the climbing surface. According to this prediction result, the PSO algorithm is used to calculate the interference compensation strategy to adjust the motion parameters of the robot dog, such as adjusting the leg posture of the robot dog to reduce the deviation.

[0076] In a possible implementation, based on the climbing feature sequence, interference trigger prediction is performed on the robot dog according to the arm control optimization strategy, and an interference trigger prediction result is obtained, and step S400 further includes step S410. Based on the arm control optimization strategy, an arm control time window is determined. Specifically, parameters in the arm control optimization strategy are obtained, such as motion trajectory, speed, and strength. According to the parameters in the arm control optimization strategy, the start time and the end time of the arm operation are calculated. For example, if the speed of the arm operation is 10 cm / s and the motion distance is 100 cm, the time window of the arm operation is 10 seconds.

[0077] Step S420, based on the arm control time window, posture prediction is performed on the robot dog according to the climbing feature sequence, and a posture feature prediction distribution is generated. Specifically, a posture prediction model is established using a machine learning algorithm (such as LSTM), and the posture features of the robot dog in the arm control time window are predicted according to the climbing feature sequence, such as the posture angle and the center of gravity position. The predicted posture features are recorded to form the posture feature prediction distribution.

[0078] Step S430, according to the arm control optimization strategy, interference influence trigger analysis is performed on the posture feature prediction distribution, and an interference influence trigger analysis result is obtained. Specifically, parameters in the arm control optimization strategy are obtained, such as motion trajectory, speed, and strength. The specific influence of arm operation on the posture and center of gravity of the robot dog is calculated using a physical model or a simulation tool. The specific interference influence obtained by analysis is recorded to form the interference influence trigger analysis result.

[0079] Step S440, according to the arm control optimization strategy, interference risk trigger analysis is performed on the posture feature prediction distribution, and an interference risk trigger analysis result is obtained, and the interference trigger prediction result is generated in combination with the interference influence trigger analysis result. Specifically, parameters in the arm control optimization strategy are obtained, such as the motion trajectory, speed, and strength of the robot arm. The risk assessment model is used to evaluate the severity of the interference risk according to the influence of the arm operation. The risk level is set, for example, the posture angle change exceeding 10° or the center of gravity position change exceeding 10 cm is regarded as high risk. The interference risk obtained by evaluation is recorded to form the interference risk trigger analysis result. The final interference trigger prediction result is generated in combination with the interference influence trigger analysis result and the interference risk trigger analysis result.

[0080] This implementation can accurately predict the interference that the robot dog may receive in the arm control time window through interference influence trigger analysis and interference risk trigger analysis, and can evaluate the risk that the robot dog may exist in the arm control process, so that measures can be taken in advance to avoid accidents.

[0081] Step S500, based on the out-arm control optimization strategy and the disturbance compensation strategy, the detected target is controlled for defect detection fusion control, the defect detection process sensing data is obtained, and the defect detection control closed loop correction is performed according to the defect detection process sensing data.

[0082] Specifically, defect detection fusion control refers to combining the out-arm control optimization strategy and the disturbance compensation strategy to cooperatively control the robot arm and the robot dog to realize the defect detection operation process. Specifically, the out-arm control optimization strategy and the disturbance compensation strategy are input into the control module, and the control module cooperatively controls the robot arm and the robot dog according to these strategies to start the defect detection operation. The sensor is used to collect the sensing data in the defect detection process in real time, such as the contact force between the probe and the target surface, the defect detection signal, etc., and the defect detection operation is corrected in real time according to these data.

[0083] For example, in the defect detection process, the contact force sensor between the probe and the target surface collects the contact force data in real time. If the contact force is too large, the control module adjusts the movement speed and force of the robot arm according to the PID control algorithm to ensure the stability and accuracy of the defect detection operation.

[0084] In one possible implementation, according to the defect detection process sensing data, the defect detection control closed loop correction is performed, and step S500 further includes step S510, according to the defect detection process sensing data, the abnormality is detected, and the defect detection process abnormality feature is obtained. Specifically, the data collected by the sensor in the defect detection process is obtained, such as the defect detection signal strength, the contact force, the probe position, etc. The statistical analysis, machine learning or deep learning algorithm is used to detect the abnormality of the defect detection process sensing data, and the abnormal data points are identified. The abnormal features are extracted from the abnormal data points to form the defect detection process abnormality features, such as low defect detection signal strength, high contact force, etc.

[0085] Step S520, according to the defect detection process abnormality feature, the robot dog is analyzed for associated influence, and the abnormal associated influence first feature is generated. Specifically, the extracted defect detection process abnormality features are obtained, including the defect detection signal strength, the contact force, the probe position, etc. According to the dynamics and kinematics model of the robot dog, a physical model is established, and the influence of the abnormality feature on the posture and movement of the robot dog is analyzed. The features describing the influence of the robot dog caused by the abnormality are extracted from the analysis results, such as the posture offset angle, the movement speed change amount, etc. The extracted features are recorded to form the abnormal associated influence first feature.

[0086] Step S530: Analyze the impact of the abnormality on the robotic arm based on the abnormality characteristics of the flaw detection process to generate a second feature of the abnormality. Specifically, extract the abnormality characteristics of the flaw detection process, including the flaw detection signal strength, contact force, probe position, etc. Based on the dynamic and kinematic models of the robotic arm, a physical model is established to analyze the impact of the abnormality on the movement and operation of the robotic arm, such as motion trajectory offset and change in operating force. From the analysis results, features describing the impact of the abnormality on the robotic arm are extracted, such as motion trajectory offset and change in operating force. The extracted features are recorded to form the second feature of the abnormality.

[0087] Step S540, based on the first feature of the abnormal correlation effect and the second feature of the abnormal correlation effect, the arm control optimization strategy and the interference compensation strategy are collaboratively optimized to generate a flaw detection fusion control optimization strategy. Specifically, the first feature and the second feature of the abnormal correlation effect generated in step S520 and step S530 are obtained. Define the optimization goal, such as minimizing the posture deviation and movement speed change of the robot dog, as well as the movement trajectory deviation and operation force change of the robot arm. According to the optimization goal, design the parameters of the PID controller (proportional coefficient K p , integral coefficient K i , differential coefficient K d ). The PID control algorithm is used to adjust the arm control optimization strategy and interference compensation strategy to generate an optimized flaw detection fusion control strategy.

[0088] This approach uses anomaly detection algorithms and correlation impact analysis to accurately identify abnormal characteristics during the flaw detection process and their impact on the robot dog and arm. A collaborative optimization algorithm optimizes the arm control optimization strategy and interference compensation strategy, improving the stability and accuracy of flaw detection operations.

[0089] The embodiment of the present application adopts a technical means of constructing a climbing feature sequence of the robot dog after it is in place, simultaneously predicting the target injury and analyzing the working space of the robot arm, generating the optimal arm extension strategy through multiple rounds of evolutionary optimization, and then predicting the robot arm motion interference in combination with the climbing features and compensating in real time, and finally realizing the coordinated adaptive control of climbing and flaw detection through closed-loop feedback of flaw detection data. It solves the technical problems of the existing robot dog's flaw detection operations in complex environments, such as the lack of dynamic coordination mechanism between climbing and flaw detection, poor environmental adaptability, and low flaw detection accuracy, and achieves the technical effect of improving environmental adaptability and flaw detection accuracy through coordinated optimization of climbing and flaw detection.

[0090] In the above, refer to Figure 1 The fusion control method of the robot arm flaw detection operation when the robot dog climbs according to the embodiment of the present invention is described in detail. Figure 2 The following describes a fusion control system for a robot arm flaw detection operation when a robot dog climbs according to an embodiment of the present invention.

[0091] The machine dog climbing mechanical arm flaw detection operation fusion control system according to the embodiment of the present application is used to solve the technical problems of the lack of dynamic coordination mechanism between climbing and flaw detection, poor environmental adaptability, and low flaw detection precision of the existing machine dog in the flaw detection operation in a complex environment, and achieves the technical effect of optimizing climbing and flaw detection coordination, improving environmental adaptability and flaw detection precision. The machine dog climbing mechanical arm flaw detection operation fusion control system comprises a climbing feature sequence construction module 10, an arm operation control analysis module 20, an arm control optimization module 30, an interference compensation module 40, and a flaw detection fusion control module 50.

[0092] The climbing feature sequence construction module 10 is configured to construct a climbing feature sequence of the machine dog when the position of the machine dog satisfies a detection work position of a target to be detected. The arm operation control analysis module 20 is configured to perform multi-dimensional flaw condition prediction on the target to be detected, obtain a target flaw condition prediction distribution, and perform arm operation control analysis on the mechanical arm based on the target flaw condition prediction distribution to obtain an arm operation control space. The arm control optimization module 30 is configured to perform multi-round reproduction evolution optimization on the arm operation control space according to a flaw detection operation evaluation model to generate an arm control optimization strategy. The interference compensation module 40 is configured to perform interference trigger prediction on the machine dog according to the arm control optimization strategy based on the climbing feature sequence to obtain an interference trigger prediction result, and perform optimization compensation on the machine dog according to the interference trigger prediction result to obtain an interference compensation strategy. The flaw detection fusion control module 50 is configured to perform flaw detection fusion control on the target to be detected based on the arm control optimization strategy and the interference compensation strategy, obtain flaw detection process sensing data, and perform flaw detection control closed-loop correction according to the flaw detection process sensing data.

[0093] In the following, the specific configuration of the arm operation control analysis module 20 will be described in detail. As described above, the target to be detected is subjected to multi-dimensional flaw condition prediction to obtain a target flaw condition prediction distribution. The arm operation control analysis module 20 can further comprise a flaw category mining unit configured to mine flaw categories of the target to be detected to obtain a flaw category set; a flaw condition record training unit configured to train flaw condition records based on the flaw category set to obtain a plurality of flaw condition prediction models; a distillation loss iterative training unit configured to perform distillation loss iterative training according to the plurality of flaw condition prediction models to build a multi-dimensional flaw condition prediction channel satisfying a distillation loss constraint; and a flaw condition prediction fusion unit configured to perform flaw condition prediction fusion on the target to be detected according to the multi-dimensional flaw condition prediction channel to generate the target flaw condition prediction distribution.

[0094] In the following, the specific configuration of the out-arm control optimization module 30 will be described in detail. As described above, the out-arm operation control space is subjected to multi-round reproduction evolution optimization according to the flaw detection operation evaluation model to generate an out-arm control optimization strategy. The out-arm control optimization module 30 can further include: an evaluation constraint optimization unit configured to perform evaluation constraint optimization on the out-arm operation control space according to the flaw detection operation evaluation model to generate an out-arm control optimization space; a flaw detection operation optimization fitness analysis model establishing unit configured to configure weights of flaw detection operation evaluation multi-dimensional indexes of the flaw detection operation evaluation model to establish a flaw detection operation optimization fitness analysis model, the flaw detection operation evaluation multi-dimensional indexes including flaw detection signal quality, flaw detection contact stability, and flaw detection contact smoothness; a reproduction evolution unit configured to perform N rounds of reproduction evolution on the out-arm control optimization space based on the flaw detection operation optimization fitness analysis model and the flaw detection operation evaluation model to generate N out-arm control evolution spaces, N being a positive integer greater than 1; and a joint optimization unit configured to perform joint optimization on the out-arm control optimization space and the N out-arm control evolution spaces according to the flaw detection operation optimization fitness analysis model to obtain the out-arm control optimization strategy.

[0095] In the evaluation constraint optimization on the out-arm operation control space according to the flaw detection operation evaluation model, the evaluation constraint optimization unit can further include: a flaw detection operation simulation subunit configured to extract an out-arm operation control first decision from the out-arm operation control space and perform flaw detection operation simulation based on the out-arm operation control first decision to obtain first flaw detection fitting data; a flaw detection operation evaluation subunit configured to input the first flaw detection fitting data into the flaw detection operation evaluation model to obtain a first flaw detection operation evaluation result; a flaw detection operation evaluation constraint condition constructing subunit configured to construct a flaw detection operation evaluation constraint condition based on the flaw detection operation evaluation multi-dimensional indexes; a judging subunit configured to judge whether the first flaw detection operation evaluation result satisfies the flaw detection operation evaluation constraint condition; and an out-arm control optimization space generating subunit configured to add the out-arm operation control first decision to the out-arm control optimization space if the first flaw detection operation evaluation result satisfies the flaw detection operation evaluation constraint condition.

[0096] The N-round breeding evolution subunit is configured to continue N-1 rounds of breeding evolution according to the first out-arm control evolution space based on the flaw detection operation optimization analytical model and the flaw detection operation evaluation model, until the N out-arm control evolution spaces are obtained.

[0097] The interference compensation module 40 can further include an out-arm control time window determination unit configured to determine an out-arm control time window based on the out-arm control optimization strategy; a posture prediction unit configured to perform posture prediction on the robot dog according to the climbing feature sequence based on the out-arm control time window, and generate a posture feature prediction distribution; an interference influence trigger analysis unit configured to perform interference influence trigger analysis on the posture feature prediction distribution according to the out-arm control optimization strategy, and obtain an interference influence trigger analysis result; and an interference risk trigger analysis unit configured to perform interference risk trigger analysis on the posture feature prediction distribution according to the out-arm control optimization strategy, obtain an interference risk trigger analysis result, and combine the interference influence trigger analysis result to generate the interference trigger prediction result.

[0098] Below, the specific configuration of the defect detection fusion control module 50 will be described in detail. As described above, the defect detection fusion control module 50 can further include: an anomaly detection unit for performing anomaly detection according to the defect detection process sensing data to obtain a defect detection process anomaly feature; a first correlation influence analysis unit for performing correlation influence analysis on the robot dog according to the defect detection process anomaly feature to generate an anomaly correlation influence first feature; a second correlation influence analysis unit for performing correlation influence analysis on the mechanical arm according to the defect detection process anomaly feature to generate an anomaly correlation influence second feature; and a collaborative optimization unit for performing collaborative optimization on the arm control optimization strategy and the interference compensation strategy according to the anomaly correlation influence first feature and the anomaly correlation influence second feature to generate a defect detection fusion control optimization strategy, according to the defect detection control closed-loop correction performed according to the defect detection process sensing data.

[0099] Below, the specific configuration of the climbing feature sequence construction module 10 will be described in detail. As described above, when the robot dog position meets the detection work position of the to-be-detected target, the climbing feature sequence of the robot dog is constructed, and the climbing feature sequence construction module 10 can further include: a multi-modal feature acquisition unit for acquiring multi-modal features of the robot dog when the robot dog position meets the detection work position to obtain a robot dog feature set; and a cleaning and combing unit for cleaning and combing the robot dog feature set to generate the climbing feature sequence.

[0100] In the process of constructing the climbing feature sequence of the robot dog, the climbing feature sequence construction module 10 can further include: a loss of stability risk evaluation unit for evaluating the loss of stability risk of the robot dog according to the climbing feature sequence to generate a loss of stability risk coefficient; and a robot dog loss of stability early warning signal generation unit for generating a robot dog loss of stability early warning signal if the loss of stability risk coefficient is greater than or equal to a loss of stability risk threshold.

[0101] The fusion control system for robot dog climbing mechanical arm defect detection operation provided in the embodiments of the present application can perform the fusion control method for robot dog climbing mechanical arm defect detection operation provided in any of the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0102] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of each functional unit are only for easy mutual differentiation, and do not limit the protection scope of the present application.

[0103] The foregoing DETAILED DESCRIPTION, including the above section titled "Detailed Description," is not to be taken as limiting the scope of the application. Various modifications, combinations, and equivalents can be apparent to those skilled in the art and can be made once the nature of the application is understood. Any modification, combination, or equivalent, which falls within the principles and the scope of the present application, is intended to be included in the present application. In some instances, the actions or steps can be performed in different order from those described herein, and still achieve desirable results. Additionally, the process depicted in the figures can not necessarily require the particular order shown or sequential order to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous.

Claims

1. A fusion control method for the flaw detection operation of a robot arm when a robot dog is climbing, characterized in that: include: When the robot dog's position meets the detection working position of the target to be detected, a climbing feature sequence of the robot dog is constructed; Performing multi-dimensional injury prediction on the target to be detected to obtain a target injury prediction distribution, and performing arm-out operation control analysis on the robotic arm based on the target injury prediction distribution to obtain an arm-out operation control space; Perform multiple rounds of evolutionary optimization on the arm-extension operation control space according to the flaw detection operation evaluation model to generate an arm-extension control optimization strategy; Based on the climbing feature sequence, performing interference trigger prediction on the robot dog according to the arm control optimization strategy to obtain an interference trigger prediction result, and performing optimal compensation on the robot dog according to the interference trigger prediction result to obtain an interference compensation strategy; Based on the arm control optimization strategy and the interference compensation strategy, flaw detection fusion control is performed on the target to be detected, flaw detection process sensing data is obtained, and flaw detection control closed-loop correction is performed according to the flaw detection process sensing data.

2. The fusion control method for the robot arm flaw detection operation when the robot dog climbs as claimed in claim 1, characterized in that: Performing multi-dimensional injury prediction on the target to be detected to obtain target injury prediction distribution, including: Performing flaw detection category mining on the target to be detected to obtain a flaw detection category set; Perform injury record training based on the flaw detection category set to obtain injury prediction models for each category; Perform iterative training of distillation loss based on the injury prediction models of various categories to build a multi-dimensional injury prediction channel that meets the distillation loss constraints; The target to be detected is subjected to injury prediction fusion according to the multi-dimensional injury prediction channel to generate the target injury prediction distribution.

3. The fusion control method for the robot arm flaw detection operation when the robot dog climbs as claimed in claim 1, characterized in that: According to the flaw detection operation evaluation model, multiple rounds of reproduction evolution optimization are performed on the arm operation control space to generate an arm control optimization strategy, including: Performing evaluation and constraint optimization on the arm-extrusion operation control space according to the flaw detection operation evaluation model to generate an arm-extrusion control optimization space; Weighting the multidimensional indicators of the flaw detection operation evaluation model to establish a flaw detection operation optimization analytical model, wherein the multidimensional indicators include flaw detection signal quality, flaw detection contact stability, and flaw detection contact smoothness; Based on the flaw detection operation optimality analytical model and the flaw detection operation evaluation model, N rounds of reproduction evolution are performed according to the arm control optimization space to generate N arm control evolution spaces, where N is a positive integer greater than 1; According to the optimal analytical model of the flaw detection operation, the arm control optimization space and the N arm control evolution spaces are jointly optimized to obtain the arm control optimization strategy.

4. The fusion control method for the robot arm flaw detection operation when the robot dog climbs as claimed in claim 3, characterized in that: The arm-outlet operation control space is evaluated and constrained to be optimized according to the flaw detection operation evaluation model to generate an arm-outlet control optimization space, including: extracting a first arm operation control decision according to the arm operation control space, and performing a flaw detection operation simulation based on the first arm operation control decision to obtain first flaw detection fitting data; inputting the first flaw detection fitting data into the flaw detection operation evaluation model to obtain a first flaw detection operation evaluation result; Based on the multi-dimensional indicators for flaw detection operation evaluation, constructing flaw detection operation evaluation constraint conditions; Determining whether the first flaw detection operation evaluation result satisfies the flaw detection operation evaluation constraint condition; If the first flaw detection operation evaluation result satisfies the flaw detection operation evaluation constraint condition, the first arm outward operation control decision is added to the arm outward control optimization space.

5. The fusion control method for the robot arm flaw detection operation when the robot dog climbs as claimed in claim 3, characterized in that: Based on the flaw detection operation optimality analytical model and the flaw detection operation evaluation model, N rounds of reproduction evolution are performed according to the arm control optimization space to generate N arm control evolution spaces, including: Calculate the detection operation optimum in the arm control optimization space according to the detection operation optimum analytical model, and establish the detection operation optimum distribution; Based on the reproduction capacity constraint, the reproduction capacity is allocated to each arm operation control decision in the arm control optimization space according to the optimal distribution of the flaw detection operation, and the first round of reproduction capacity distribution is obtained; Adaptively mutate the arm control optimization space based on the first round of breeding capacity distribution to generate the first round of arm control breeding space; Performing evaluation and constraint optimization on the first round of the arm control breeding space according to the flaw detection operation evaluation model to generate a first arm control evolution space; Based on the flaw detection operation optimality analytical model and the flaw detection operation evaluation model, N-1 rounds of reproduction evolution are continued according to the first arm control evolution space until the N arm control evolution spaces are obtained.

6. The fusion control method for the robot arm flaw detection operation when the robot dog climbs as claimed in claim 1, characterized in that: Based on the climbing feature sequence, interference trigger prediction is performed on the robot dog according to the arm control optimization strategy to obtain an interference trigger prediction result, including: Determining an arm-out control time window based on the arm-out control optimization strategy; Based on the arm control time window, the robot dog is predicted according to the climbing feature sequence to generate a posture feature prediction distribution; Performing interference impact trigger analysis on the posture feature prediction distribution according to the arm control optimization strategy to obtain an interference impact trigger analysis result; An interference risk trigger analysis is performed on the posture feature prediction distribution according to the arm control optimization strategy to obtain an interference risk trigger analysis result, and the interference trigger prediction result is generated in combination with the interference impact trigger analysis result.

7. The fusion control method for the robot arm flaw detection operation when the robot dog climbs as claimed in claim 1, characterized in that: Performing closed-loop correction of flaw detection control according to the flaw detection process sensing data includes: Performing anomaly detection based on the sensing data of the flaw detection process to obtain abnormal characteristics of the flaw detection process; Performing correlation impact analysis on the robot dog according to the abnormal characteristics of the flaw detection process to generate a first abnormal correlation impact feature; Performing correlation impact analysis on the robotic arm according to the abnormal characteristics of the flaw detection process to generate a second abnormal correlation impact feature; According to the first feature of the abnormal correlation impact and the second feature of the abnormal correlation impact, the arm control optimization strategy and the interference compensation strategy are collaboratively optimized to generate a flaw detection fusion control optimization strategy.

8. The fusion control method for the robot arm flaw detection operation when the robot dog climbs as claimed in claim 1, characterized in that: When the robot dog's position meets the detection working position of the target to be detected, the robot dog's climbing feature sequence is constructed, including: When the position of the robot dog satisfies the detection working position, multimodal feature collection is performed on the robot dog to obtain a robot dog feature set; The robot dog feature set is cleaned and sorted to generate the climbing feature sequence.

9. The fusion control method for the robot arm flaw detection operation when the robot dog climbs as claimed in claim 1, characterized in that: Construct a climbing feature sequence for the robot dog, including: Performing an instability risk assessment on the robot dog according to the climbing feature sequence to generate an instability risk coefficient; If the instability risk coefficient is greater than or equal to the instability risk threshold, an instability warning signal of the robot dog is generated.

10. The fusion control system of the robot arm flaw detection operation when the robot dog climbs is characterized by: The system is used to implement the fusion control method for the flaw detection operation of the robotic arm when the robot dog climbs as described in any one of claims 1 to 9, and the system includes: A climbing feature sequence construction module is used to construct a climbing feature sequence of the robot dog when the robot dog's position meets the detection working position of the target to be detected; An arm-extracting operation control analysis module is used to perform multi-dimensional injury prediction on the target to be detected, obtain a target injury prediction distribution, and perform arm-extracting operation control analysis on the robotic arm based on the target injury prediction distribution to obtain an arm-extracting operation control space; An arm control optimization module is used to perform multiple rounds of evolutionary optimization on the arm operation control space according to the flaw detection operation evaluation model to generate an arm control optimization strategy; an interference compensation module, configured to perform interference trigger prediction on the robot dog based on the climbing feature sequence and the arm control optimization strategy to obtain an interference trigger prediction result, and perform optimization compensation on the robot dog based on the interference trigger prediction result to obtain an interference compensation strategy; The flaw detection fusion control module is used to perform flaw detection fusion control on the target to be detected based on the arm control optimization strategy and the interference compensation strategy, obtain flaw detection process sensing data, and perform flaw detection control closed-loop correction according to the flaw detection process sensing data.

Images