Robot health state assessment method and system based on artificial intelligence

Abstract

The invention discloses a robot health state assessment method and system based on artificial intelligence. The method comprises the steps of data acquisition and fusion, digital twin health modeling, multi-scale predictive maintenance, fault tolerance control and health state assessment. The invention relates to the technical field of digital robot health analysis, and the method comprises the following steps: constructing an optimized feature set by fusing multi-source sensing data, realizing digital twin health modeling by adopting a dynamic dual-data flow neural differential equation combining physical constraint and wear evolution, and outputting a robot health degree parameter; quantum annealing optimization and a graph attention mechanism are introduced, a multi-level predictive maintenance architecture is constructed, fault propagation prediction and maintenance optimization from a part level to a system level are realized, and fault-tolerant control is realized in combination with model prediction control. And the accuracy, interpretability and practicability of health assessment of the robot are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital robot health analysis, and specifically refers to a method and system for evaluating the health status of a robot based on artificial intelligence. Background Art

[0002] A method and system for evaluating the health status of a robot based on artificial intelligence refers to an intelligent technical system that uses machine learning, computer vision, and sensor data analysis technologies to perform real-time monitoring, fault diagnosis, and performance prediction on key components such as the mechanical structure, electronic components, and software operation of the robot; by collecting multiple information such as the motion parameters, energy consumption data, and operation logs of the robot, and combining deep learning and anomaly detection algorithms to establish a health status model, it realizes automatic fault identification and remaining life prediction. Its core function is to improve the operation and maintenance efficiency, reduce the risk of sudden failures, and extend the service life of the robot through pre-maintenance strategies, thereby ensuring the stable operation of equipment in scenarios such as industrial automation and service robots.

[0003] However, in the existing methods for evaluating the health status of robots, there are limitations in the narrow collection dimension of robot health data and the lack of multi-modal fusion. Usually, only a single type of sensor (such as temperature, current, or vibration) is relied on, making it difficult to comprehensively reflect the true health status during the operation of the robot. Moreover, the existing intelligent methods cannot be linked with maintenance strategy generation or control logic, making it difficult to be used in actual engineering; in the existing process of digital modeling of robot health, traditional methods mostly adopt pure data-driven machine learning models, lacking consideration of the dynamics of the robot body and the law of energy conservation. Although the modeling results can fit the sample data, they lack physical interpretability and are difficult to be popularized and used in complex tasks or unknown working conditions; in the existing multi-scale predictive maintenance process of robots, there are technical problems that the existing methods often have a split in the life cycle dimension, and the predictive algorithms at different levels are independent of each other and difficult to cooperate. Summary of the Invention

[0004] In view of the above situation, to overcome the defects of the existing technology, the present invention provides a method and system for evaluating the health status of a robot based on artificial intelligence. In the existing methods for evaluating the health status of a robot, there are limitations such as narrow dimensions in collecting health data of the robot and lack of multimodal fusion. Usually, only a single type of sensor (such as temperature, current or vibration) is relied on, which is difficult to comprehensively reflect the true health status during the operation of the robot. Moreover, the existing intelligent methods cannot be linked with maintenance strategy generation or control logic, making it difficult to be used in actual engineering. The present solution creatively adopts an overall intelligent health evaluation and processing method combining digital twin health modeling, multi-scale hierarchical predictive maintenance and fault-tolerant control optimization, realizing full-cycle, hierarchical and high-precision intelligent diagnosis and dynamic regulation of the operating state of the robot, and significantly improving the evaluation accuracy. In the existing process of digital twin health modeling of a robot, traditional methods mostly adopt pure data-driven machine learning models, lacking consideration of the dynamics and energy conservation laws of the robot body. Although the modeling results can fit the sample data, they lack physical interpretability and are difficult to be popularized and used under complex tasks or unknown working conditions. The present solution creatively adopts a dynamic double-data-stream neural ordinary differential equation modeling method for digital twin health modeling, realizing joint energy evolution modeling and wear trend prediction under physical consistency constraints, and effectively improving the modeling accuracy and fitting degree of the actual scenario. In the existing multi-scale predictive maintenance process of a robot, there are technical problems such as the disconnection of the life cycle dimension and the independence and difficulty in coordination of different-level prediction algorithms in the existing methods. The present solution creatively adopts a multi-time-scale prediction architecture combining quantum annealing optimization and graph attention network for multi-scale predictive maintenance, realizing cross-scale collaborative prediction and dynamic maintenance plan generation from the part level, system level to the machine level, and improving the maintenance timeliness and resource scheduling optimization ability.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides a method for evaluating the health status of a robot based on artificial intelligence, and the method includes the following steps:

[0006] Step S1: Data acquisition and fusion;

[0007] Step S2: Digital twin health modeling;

[0008] Step S3: Multi-scale predictive maintenance;

[0009] Step S4: Fault-tolerant control;

[0010] Step S5: Health status evaluation.

[0011] Further, in step S1, the data acquisition and fusion is used to collect raw data, extract spatio-temporal features and perform feature fusion. Specifically, through sensor data collection and dynamic sampling coordination, an original data set for robot health state assessment is obtained, and through data preprocessing and process feature fusion, optimized data for robot health state assessment is obtained;

[0012] The original data set for robot health state assessment includes vibration data, robot joint position and speed data, joint force data, joint torque data, voltage and current data, temperature data, control instruction data and maintenance log data;

[0013] The optimized data for robot health state assessment specifically includes vibration signal time-domain feature data, frequency-domain feature data, state variable feature data, robot health derivative feature data, load feature data, maintenance history feature data and robot health label data.

[0014] Further, in step S2, the digital twin health modeling is used to construct a virtual digital twin model of the robot health state. Specifically, based on the optimized data for robot health state assessment, the dynamic dual-data-stream neural ordinary differential equation modeling method is used to perform digital twin health modeling to obtain robot health degree parameters, including the following steps:

[0015] Step S21: Physical constraint embedding modeling. Specifically, based on the optimized data for robot health state assessment, the Hamiltonian neural network combined with the physical constraints of the dynamic system is used to perform dynamic consistency physical constraint model modeling to obtain robot joint health assessment data;

[0016] The Hamiltonian neural network combined with the physical constraints of the dynamic system includes the definition of generalized momentum, the definition of Hamiltonian dynamic system and the definition of physical modeling loss function. The definition of generalized momentum is used to map the actually observed robot joint angular velocity into generalized momentum and serve as a prediction index;

[0017] The definition of the Hamiltonian dynamic system is used to construct a physical evolution model of the robot joint state;

[0018] Step S22: Wear evolution modeling. Specifically, based on the optimized data for robot health state assessment, by constructing a wear evolution neural ordinary differential equation modeling and constructing a standard multi-layer perceptron structure for dynamic evolution of wear state and future wear prediction modeling, the robot joint wear rate and predicted wear value data are calculated;

[0019] Step S23: Adversarial verification. Specifically, based on the predicted wear value data and the true wear value data, standard generative adversarial network training is performed to verify the authenticity of the digital twin health modeling, and health modeling authenticity assessment data is obtained;

[0020] Step S24: Digital twin health modeling, specifically, through the above-mentioned physical constraint embedding modeling, wear evolution modeling, and adversarial verification, comprehensive digital twin health modeling is carried out to obtain robot health parameters;

[0021] The robot health parameters include robot joint simulation consistency indicators, robot joint wear degree estimates, robot health prediction residuals, health modeling authenticity assessment data, and robot comprehensive health scores.

[0022] Furthermore, in step S3, the multi-scale predictive maintenance is used to achieve cross-scale prediction from micro wear to system-level faults. Specifically, based on the robot health parameters and the robot health status evaluation and optimization data, a multi-time scale prediction architecture combining quantum annealing optimization and graph attention network is adopted to perform multi-scale predictive maintenance to obtain hierarchical predictive maintenance reference data, including the following steps:

[0023] Step S31: Component-level life prediction. Based on the robot health status evaluation and optimization data, and using the robot data in the robot health parameters that meet the requirements for preventive maintenance as the original data input, a long short-term memory neural network combined with quantum annealing optimization is used to perform component-level life prediction to obtain remaining life estimate data;

[0024] The long short-term memory neural network combined with quantum annealing optimization specifically performs component-level life prediction by constructing an improved quantum annealing optimization algorithm and combining it with a standard long short-term memory neural network, and calculates the remaining life estimate data by modeling the output layer as a Bayesian linear regression;

[0025] The improved quantum annealing optimization algorithm specifically improves the standard quantum annealing optimization algorithm by introducing a third-order parameter interaction relationship modeling term and a test accuracy term;

[0026] Step S32: System-level fault propagation graph construction. Specifically, based on the robot health parameters and the robot health status evaluation and optimization data, by constructing a robot system topology graph and using the standard graph attention network method, fault propagation modeling is carried out, and through fault diffusion probability modeling, fault propagation prediction data is obtained, including fault propagation probability matrix data and propagation path prediction data;

[0027] Step S33: Machine-level maintenance plan optimization. Specifically, based on the robot health parameters and the robot health status evaluation and optimization data, a deep reinforcement learning method combining an improved multi-objective reward function and quantum annealing strategy distillation is used to optimize the maintenance plan of the robot to obtain optimized maintenance strategy prediction data;

[0028] The deep reinforcement learning method combining an improved multi-objective reward function and quantum annealing policy distillation includes constructing a multi-objective reward function and improving quantum annealing policy distillation;

[0029] The improvement of quantum annealing policy distillation is specifically to optimize the weights of the reward terms in the multi-objective reward function according to the improved quantum annealing optimization algorithm, and specifically by replacing the test set verification accuracy function term in the multi-objective reward function formula with the reward value;

[0030] The deep reinforcement learning method constructs the robot maintenance state space parameters and the robot maintenance action space parameters, and performs standard reinforcement learning training according to the multi-objective reward function;

[0031] Step S34: Multi-scale integration, specifically integrating the remaining life estimation data, the fault propagation prediction data, and the optimized maintenance strategy prediction data, performing part-level, system-level, and machine-level multi-scale robot health predictive maintenance, and constructing a dynamic update mechanism for prediction optimization to obtain multi-scale integrated prediction data;

[0032] Step S35: Hierarchical predictive maintenance, specifically performing hierarchical predictive maintenance according to the multi-scale integrated prediction data to obtain hierarchical predictive maintenance reference data;

[0033] The hierarchical predictive maintenance reference data includes part-level maintenance parameters, system-level maintenance parameters, and machine-level maintenance decisions;

[0034] The part-level maintenance parameters include part number, remaining service life of the part, prediction uncertainty estimation, health status parameters, and temperature outliers;

[0035] The system-level maintenance parameters include fault propagation path number, fault propagation probability, fault path criticality evaluation value, and risk path chain position;

[0036] The machine-level maintenance decisions include machine number, optional maintenance action set, optimal recommended maintenance action, maintenance action cost, maintenance downtime prediction time, maintenance improvement effect prediction value, and maintenance strategy weight;

[0037] The part-level maintenance parameters are used to assist maintenance technicians in part repair or replacement; the system-level maintenance parameters are used to assist engineers in fault propagation path control; the machine-level maintenance decisions are used to assist robot managers in balancing maintenance planning strategies.

[0038] Further, in step S4, the fault-tolerant control is used to maintain the basic functions of the robot when a fault occurs. Specifically, based on the hierarchical predictive maintenance reference data and the robot health status evaluation and optimization data, a model predictive controller with health limits is constructed by dynamically mapping the health status of the key components of the robot into control constraints for fault-tolerant control to obtain fault occurrence control instruction data;

[0039] The fault occurrence control instruction data includes control quantity adjustment data and safety action limit data.

[0040] Further, in step S5, the health status evaluation is used to generate a comprehensive health status evaluation result. Specifically, by combining the robot health degree parameters, the hierarchical predictive maintenance reference data, and the fault occurrence control instruction data, a comprehensive evaluation of the robot health status is performed to obtain robot health status comprehensive evaluation reference data.

[0041] The robot health status evaluation system based on artificial intelligence provided by the present invention includes a data acquisition and fusion module, a digital twin health modeling module, a multi-scale predictive maintenance module, a fault-tolerant control module, and a health status evaluation module;

[0042] The data acquisition and fusion module is used for data acquisition and fusion. Through data acquisition and fusion, robot health status evaluation and optimization data are obtained, and the robot health status evaluation and optimization data are sent to the digital twin health modeling module, the multi-scale predictive maintenance module, and the fault-tolerant control module;

[0043] The digital twin health modeling module is used for digital twin health modeling. Through digital twin health modeling, robot health degree parameters are obtained, and the robot health degree parameters are sent to the multi-scale predictive maintenance module and the health status evaluation module;

[0044] The multi-scale predictive maintenance module is used for multi-scale predictive maintenance. Through multi-scale predictive maintenance, hierarchical predictive maintenance reference data are obtained, and the hierarchical predictive maintenance reference data are sent to the fault-tolerant control module and the health status evaluation module;

[0045] The fault-tolerant control module is used for fault-tolerant control. Through fault-tolerant control, fault occurrence control instruction data are obtained, and the fault occurrence control instruction data are sent to the health status evaluation module;

[0046] The health status evaluation module is used for health status evaluation. Through health status evaluation, robot health status comprehensive evaluation reference data are obtained.

[0047] The beneficial effects achieved by the present invention by adopting the above solution are as follows:

[0048] (1) In the existing robot health status assessment methods, there are limitations such as narrow dimensions of robot health data collection and lack of multimodal fusion. Usually, only relying on a single type of sensor (such as temperature, current, or vibration), it is difficult to comprehensively reflect the true health status during the operation of the robot. Moreover, the existing intelligent methods cannot be linked with maintenance strategy generation or control logic, making it difficult to be used in actual engineering. This solution creatively adopts an overall intelligent health assessment and processing method that combines digital twin health modeling, multi-scale hierarchical predictive maintenance, and fault-tolerant control optimization, achieving full-cycle, hierarchical, and high-precision intelligent diagnosis and dynamic regulation of the robot's operating state, significantly improving the assessment accuracy.

[0049] (2) In the existing process of robot health digital modeling, traditional methods mostly adopt pure data-driven machine learning models, lacking consideration of the robot's ontology dynamics and energy conservation laws. Although the modeling results can fit the sample data, they lack physical interpretability and are difficult to be promoted and used in complex tasks or unknown working conditions. This solution creatively adopts a dynamic double-data-stream neural ordinary differential equation modeling method for digital twin health modeling, realizing joint energy evolution modeling and wear trend prediction under physical consistency constraints, effectively improving the modeling accuracy and the fitting degree of the actual scenario.

[0050] (3) In the existing process of robot multi-scale predictive maintenance, there are technical problems such as the disconnection of the life cycle dimension and the independence and difficulty in coordination of different-level prediction algorithms in the existing methods. This solution creatively adopts a multi-time-scale prediction architecture that combines quantum annealing optimization and graph attention network for multi-scale predictive maintenance, realizing cross-scale collaborative prediction and dynamic maintenance plan generation from the part level, system level to the machine level, improving the maintenance timeliness and resource scheduling optimization ability. Brief Description of the Drawings

[0051] Figure 1 It is a schematic flow chart of the robot health status assessment method based on artificial intelligence provided by the present invention;

[0052] Figure 2 It is a schematic diagram of the robot health status assessment system based on artificial intelligence provided by the present invention;

[0053] Figure 3 It is a schematic flow chart of the digital twin health modeling in step S2;

[0054] Figure 4 It is a schematic flow chart of the multi-scale predictive maintenance in step S3.

[0055] The accompanying drawings are used to provide a further understanding of the present invention and form a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation on the present invention. Detailed Embodiments

[0056] 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.

[0057] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation on the present invention.

[0058] Embodiment 1. Refer to Figure 1 , the method for evaluating the health status of a robot based on artificial intelligence provided by the present invention includes the following steps:

[0059] Step S1: Data acquisition and fusion;

[0060] Step S2: Digital twin health modeling;

[0061] Step S3: Multi-scale predictive maintenance;

[0062] Step S4: Fault tolerance control;

[0063] Step S5: Health status evaluation.

[0064] By performing the above operations, in the existing method for evaluating the health status of a robot, there are limitations such as narrow dimensions in collecting health data of the robot and lack of multi-modal fusion. Usually, it only relies on a single type of sensor (such as temperature, current or vibration), which is difficult to comprehensively reflect the true health status during the operation of the robot. Moreover, the existing intelligent methods cannot be linked with maintenance strategy generation or control logic, making it difficult to be used in actual engineering. The present solution creatively adopts an overall intelligent health evaluation and processing method combining digital twin health modeling, multi-scale hierarchical predictive maintenance and fault tolerance control optimization, realizing full-cycle, hierarchical and high-precision intelligent diagnosis and dynamic regulation of the running state of the robot, and significantly improving the evaluation accuracy.

[0065] Embodiment 2. Refer toFigure 1 and Figure 2 In step S1, the data acquisition and fusion is used to collect raw data, extract spatio-temporal features and perform feature fusion. Specifically, through the coordination of sensing data collection and dynamic sampling, the original dataset for robot health state assessment is obtained, and through data preprocessing and process feature fusion, the optimized data for robot health state assessment is obtained;

[0066] The original dataset for robot health state assessment includes vibration data, robot joint position and speed data, joint force data, joint torque data, voltage and current data, temperature data, control instruction data and maintenance log data;

[0067] The optimized data for robot health state assessment specifically includes vibration signal time-domain feature data, frequency-domain feature data, state variable feature data, robot health derivative feature data, load feature data, maintenance history feature data and robot health label data;

[0068] The robot health derivative feature data specifically includes wear rate feature, cumulative wear feature and temperature change feature;

[0069] The load feature data specifically includes load ratio feature and drive current peak feature;

[0070] The maintenance history feature data includes the time interval feature since the last maintenance, the cumulative running time feature of the robot and the part replacement frequency feature;

[0071] The robot health label data specifically refers to part-level fault type labels, system-level fault propagation labels and machine-level maintenance action labels.

[0072] Embodiment 3, refer to Figure 1 、 Figure 2 and Figure 3 Based on the above embodiment, in step S2, the digital twin health modeling is used to construct a virtual digital twin model of the robot health state. Specifically, according to the optimized data for robot health state assessment, the dynamic double-data-stream neural ordinary differential equation modeling method is adopted to perform digital twin health modeling to obtain robot health parameters, including the following steps:

[0073] Step S21: Physical constraint embedding modeling. Specifically, according to the optimized data for robot health state assessment, the Hamiltonian neural network combined with the physical constraints of the dynamic system is adopted to perform dynamic consistency physical constraint model modeling to obtain robot joint health assessment data;

[0074] The Hamiltonian neural network combined with the physical constraints of the dynamic system includes the definition of generalized momentum, the definition of the Hamiltonian dynamic system, and the definition of the physical modeling loss function. The definition of generalized momentum is used to map the actually observed angular velocity of the robot joints to the generalized momentum and use it as a prediction index. The calculation formula is:

[0075] p = M(q)·q 0 ;

[0076] In the formula, p is the value of the generalized momentum, M(·) is the mass matrix calculation function, q is the joint angle value, and q 0 is the joint angular velocity value;

[0077] The definition of the Hamiltonian dynamic system is used to construct a physical evolution model of the robot joint state. The calculation formula is:

[0078] ;

[0079] In the formula, The whole is the differential value of the joint angle, is the Hamiltonian function fitted by the neural network, is the momentum gradient operator, q is the joint angle value, p is the generalized momentum value, The whole is the differential value of the momentum, is the joint angle gradient operator, is the external disturbance torque parameter;

[0080] The calculation formula of the definition of the physical modeling loss function is:

[0081] ;

[0082] In the formula, is the physical modeling loss function, The whole is the angular velocity output loss. Among them, is the gradient value of the neural network estimated value of the Hamiltonian function with respect to the momentum p, which is used to represent the theoretical system speed of the physical modeling, is the actual angular velocity of the robot joints collected by the sensor, The whole is the constrained momentum evolution loss, is the physical consistency weight. Among them, is the gradient value of the neural network estimated value of the Hamiltonian function with respect to the joint angle q, which is used to represent the potential energy gradient of the physical modeling. M -1 (·) is the inverse function of the mass matrix calculation function;

[0083] Preferably, the specific value of the physical consistency weight is 0.5;

[0084] Step S22: Wear evolution modeling. Specifically, based on the optimized data for robot health state assessment, a wear evolution neural ordinary differential equation is constructed for modeling, and a standard multi-layer perceptron structure is built for dynamic evolution of the wear state and future wear prediction modeling, and the robot joint wear rate and predicted wear value data are calculated;

[0085] The calculation formula for the wear evolution neural ordinary differential equation modeling is:

[0086] ;

[0087] In the formula, The whole is the modeling output of the robot joint wear evolution rate at the current moment, which is used to represent the growth rate of joint wear. is the function represented by the wear evolution differential equation established by the standard multi-layer perceptron. w is the robot joint wear amount, which is used to represent the cumulative loss of the robot bearing and joint parts. x is the original signal of the vibration sensor. is the multi-layer perceptron function, concat(·) is the vector concatenation function, and FFT(x) is the frequency domain feature extraction function;

[0088] The calculation formula for the loss function of the standard multi-layer perceptron structure for dynamic evolution of the wear state and future wear prediction modeling is:

[0089] ;

[0090] In the formula, is the wear evolution modeling loss function, w pred (·) is the joint wear amount prediction function, which is used to represent the predicted robot joint wear value based on the wear evolution neural ordinary differential equation modeling. w real is the actually measured wear amount. is the regularization coefficient. is the gradient of the neural network parameters, and ||·|| Fro is the Frobenius norm operator;

[0091] Step S23: Adversarial verification. Specifically, based on the predicted wear value data and the true wear value data, standard generative adversarial network training is carried out to verify the authenticity of the digital twin health modeling, and the health modeling authenticity evaluation data is obtained;

[0092] Step S24: Digital twin health modeling. Specifically, through the physical constraint embedding modeling, the wear evolution modeling, and the adversarial verification, digital twin health comprehensive modeling is carried out to obtain the robot health degree parameters;

[0093] The robot health parameters include the robot joint simulation consistency index, the robot joint wear degree estimation, the robot health prediction residual, the health modeling authenticity evaluation data, and the robot comprehensive health score;

[0094] The calculation formula for the robot comprehensive health score is:

[0095] ;

[0096] In the formula, is the robot comprehensive health score, is the Sigmoid function, is the joint energy weight, is the robot joint simulation consistency index, specifically used to represent the angular velocity output loss value, is the cumulative wear weight, w(t) is the robot joint wear degree estimation, is the health prediction residual weight, is the robot health prediction residual, specifically representing the residual between the predicted value and the actual value, is the health modeling authenticity evaluation weight, is the health modeling authenticity evaluation data, specifically representing the difference between the robot health state generated by the adversarial verification and the real robot health state;

[0097] Preferably, Table 1 is the table of the meaning of the robot comprehensive health score value. As shown in the table, when , it indicates that the robot health state is good. When , it indicates that the robot needs maintenance and inspection. When , it indicates that the robot health state is poor and preventive maintenance needs to be implemented.

[0098] Table 1 Table of the meaning of the robot comprehensive health score value

[0099]

[0100] By performing the above operations, aiming at the technical problem that in the existing robot health digital modeling process, traditional methods mostly adopt pure data-driven machine learning models, lacking consideration of the robot ontology dynamics and the law of energy conservation. Although the modeling results can fit the sample data, they lack physical interpretability and are difficult to be popularized and used in complex tasks or unknown working conditions. This solution creatively adopts the dynamic double data stream neural ordinary differential equation modeling method for digital twin health modeling, realizes the joint energy evolution modeling and wear trend prediction under physical consistency constraints, and effectively improves the modeling accuracy and the fitting degree of the actual scenario.

[0101] Example 4, refer to Figure 1 、 Figure 2and Figure 4 This embodiment is based on the above embodiment. In step S3, the multi-scale predictive maintenance is used to achieve cross-scale prediction from micro wear to system-level faults. Specifically, according to the robot health parameter and the optimized data for robot health state evaluation, a multi-time scale prediction architecture combining quantum annealing optimization and graph attention network is adopted to perform multi-scale predictive maintenance, and hierarchical predictive maintenance reference data is obtained, including the following steps:

[0102] Step S31: Part-level life prediction. According to the optimized data for robot health state evaluation, and taking the robot data in the robot health parameter that meets the need for preventive maintenance as the original data input, a long short-term memory neural network combined with quantum annealing optimization is used to perform part-level life prediction, and the remaining life estimation data is obtained;

[0103] The long short-term memory neural network combined with quantum annealing optimization specifically performs part-level life prediction by constructing an improved quantum annealing optimization algorithm and combining it with a standard long short-term memory neural network, and calculates the remaining life estimation data by modeling the output layer as Bayesian linear regression;

[0104] The improved quantum annealing optimization algorithm specifically improves the standard quantum annealing optimization algorithm by introducing a third-order parameter interaction relationship modeling term and a test accuracy term, and the calculation formula is:

[0105] ;

[0106] In the formula, H(·) is the output of the improved quantum annealing optimization algorithm, z is the set of hyperparameters of the standard long short-term memory neural network, The overall is a bivariate coupling relationship modeling term between hyperparameters, which is used to represent the interaction strength between model parameters. Among them, i is the first hyperparameter index, j is the second hyperparameter index, J ij is the hyperparameter interaction strength identifier, z i is the model hyperparameter corresponding to the first hyperparameter i, z j is the model hyperparameter corresponding to the second hyperparameter j, The overall hyperparameter individual influence modeling term, h i is the preference parameter of the model hyperparameter corresponding to the first hyperparameter i, The overall is a third-order interaction relationship modeling term, is the third-order regulation weight, k is the third hyperparameter index, is the influence value parameter of the three hyperparameters on the model performance when all are jointly activated, z k is the model hyperparameter corresponding to the third hyperparameter k, is the feedback coefficient, and Acc(·) is the test set validation accuracy function;

[0107] The calculation formula for modeling the output layer as Bayesian linear regression is as follows:

[0108] ;

[0109] In the formula, is the predicted remaining life estimation data, is the normal distribution probability density function, is the predicted mean mapping function, h T is the output hidden state of the standard long short-term memory neural network, is the predicted variance mapping function;

[0110] Step S32: Construction of the system-level fault propagation graph. Specifically, based on the robot health parameters and the robot health state evaluation and optimization data, by constructing a robot system topology graph and using the standard graph attention network method, fault propagation modeling is carried out, and through fault diffusion probability modeling, fault propagation prediction data is obtained, including fault propagation probability matrix data and propagation path prediction data;

[0111] The calculation formula for the fault diffusion probability modeling is as follows:

[0112] ;

[0113] In the formula, P(·) is the fault propagation probability prediction function, is the edge parameter between node and node , is the node index of the robot system topology graph, is the adjacent node index of the robot system topology graph, u is the learnable weight value, is the output vector of the node corresponding to the L-th layer of the graph attention network, is the element-wise multiplication operator, is the output vector of the node corresponding to the L-th layer of the graph attention network, L is the total number of layers of the graph attention network, and the specific value is 3;

[0114] Step S33: Optimization of the machine-level maintenance plan. Specifically, based on the robot health parameters and the robot health state evaluation and optimization data, a deep reinforcement learning method combining an improved multi-objective reward function and quantum annealing strategy distillation is used to optimize the robot's maintenance plan, and optimized maintenance strategy prediction data is obtained;

[0115] The deep reinforcement learning method combining an improved multi-objective reward function and quantum annealing strategy distillation includes constructing a multi-objective reward function and improving the quantum annealing strategy distillation;

[0116] The calculation formula of the multi-objective reward function is as follows:

[0117] ;

[0118] In the formula, r(s,a) is the multi-objective reward function, where s is the robot maintenance state space parameter, a is the robot maintenance action space parameter, cost(a) is the maintenance cost reward term, downtime(a) is the predicted downtime reward term, reliability_gain(a) is the performance improvement reward term, safety_improvement(a) is the safety improvement reward term, remaining_life_boost(a) is the positive improvement reward term for the life of key parts, and resource_usage(a) is the resource consumption reward term;

[0119] The improvement of the quantum annealing strategy distillation is specifically to optimize the weights of each reward term in the multi-objective reward function according to the improved quantum annealing optimization algorithm, and specifically by replacing the test set verification accuracy function term in the multi-objective reward function formula with the reward value;

[0120] The deep reinforcement learning method constructs the robot maintenance state space parameter and the robot maintenance action space parameter, and performs standard reinforcement learning training according to the multi-objective reward function;

[0121] The robot maintenance state space parameter specifically includes the remaining life estimation, vibration index, temperature monitoring value, fault sequence, time interval since the last maintenance, and load ratio;

[0122] The robot maintenance action space parameter specifically includes routine inspection actions, part replacement actions, fault calibration actions, emergency shutdown maintenance actions, and load scheduling actions;

[0123] Step S34: Multi-scale integration, specifically integrating the remaining life estimation data, the fault propagation prediction data, and the optimized maintenance strategy prediction data, performing part-level, system-level, and machine-level multi-scale robot health predictive maintenance, and constructing a dynamic update mechanism to perform prediction optimization to obtain multi-scale integrated prediction data;

[0124] Step S35: Hierarchical predictive maintenance, specifically performing hierarchical predictive maintenance according to the multi-scale integrated prediction data to obtain hierarchical predictive maintenance reference data;

[0125] The hierarchical predictive maintenance reference data includes part-level maintenance parameters, system-level maintenance parameters, and machine-level maintenance decisions;

[0126] The part-level maintenance parameters include part number, remaining service life of the part, prediction uncertainty estimate, health status parameter, and temperature outlier value;

[0127] The system-level maintenance parameters include fault propagation path number, fault propagation probability, criticality evaluation value of the fault path, and position of the risk path chain;

[0128] The machine-level maintenance decision includes machine number, set of optional maintenance actions, optimal recommended maintenance action, cost of the maintenance action, predicted maintenance downtime, predicted value of the maintenance improvement effect, and maintenance strategy weight;

[0129] The part-level maintenance parameters are used to assist maintenance technicians in part repair or replacement; the system-level maintenance parameters are used to assist engineers in fault propagation path control; the machine-level maintenance decision is used to assist robot managers in balancing maintenance planning strategies.

[0130] By performing the above operations, in the existing multi-scale predictive maintenance process of robots, there are often technical problems such as the fragmentation of the life cycle dimension and the independence and difficulty in coordination of different-level prediction algorithms in the existing methods. This solution creatively adopts a multi-time-scale prediction architecture combining quantum annealing optimization and graph attention network for multi-scale predictive maintenance, realizing cross-scale collaborative prediction and dynamic maintenance plan generation from the part level, system level to the machine level, and improving the maintenance timeliness and resource scheduling optimization ability.

[0131] Example Five, refer to Figure 1 and Figure 2 In this example, based on the above example, in step S4, the fault tolerance control is used to maintain the basic functions of the robot when a fault occurs. Specifically, according to the hierarchical predictive maintenance reference data and the robot health status evaluation and optimization data, by dynamically mapping the health status of the key components of the robot into control constraints, a model predictive controller with health limits is constructed for fault tolerance control to obtain fault occurrence control instruction data;

[0132] The fault occurrence control instruction data includes control quantity adjustment data and safety action limit data.

[0133] Example Six, refer to Figure 1 and Figure 2 In this example, based on the above example, in step S5, the health status evaluation is used to generate a comprehensive health status evaluation result. Specifically, by combining the robot health parameters, the hierarchical predictive maintenance reference data, and the fault occurrence control instruction data, a comprehensive evaluation of the robot health status is performed to obtain the robot health status comprehensive evaluation reference data.

[0134] Example Seven, refer toFigure 1 and Figure 2 Based on the above embodiments, the robot health status evaluation system based on artificial intelligence provided by the present invention includes a data acquisition and fusion module, a digital twin health modeling module, a multi-scale predictive maintenance module, a fault tolerance control module, and a health status evaluation module;

[0135] The data acquisition and fusion module is used for data acquisition and fusion. Through data acquisition and fusion, optimized data for robot health status evaluation is obtained, and the optimized data for robot health status evaluation is sent to the digital twin health modeling module, the multi-scale predictive maintenance module, and the fault tolerance control module;

[0136] The digital twin health modeling module is used for digital twin health modeling. Through digital twin health modeling, robot health degree parameters are obtained, and the robot health degree parameters are sent to the multi-scale predictive maintenance module and the health status evaluation module;

[0137] The multi-scale predictive maintenance module is used for multi-scale predictive maintenance. Through multi-scale predictive maintenance, hierarchical predictive maintenance reference data is obtained, and the hierarchical predictive maintenance reference data is sent to the fault tolerance control module and the health status evaluation module;

[0138] The fault tolerance control module is used for fault tolerance control. Through fault tolerance control, fault occurrence control instruction data is obtained, and the fault occurrence control instruction data is sent to the health status evaluation module;

[0139] The health status evaluation module is used for health status evaluation. Through health status evaluation, comprehensive evaluation reference data for robot health status is obtained.

[0140] 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 terms "include", "comprise" or any other variant thereof are 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 expressly listed, or also includes elements inherent to such process, method, article or device.

[0141] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention.

[0142] The above describes the present invention and its embodiments. Such description is not restrictive. What is shown in the drawings is only one of the embodiments of the present invention, and the actual structure is not limited thereto. In summary, if those of ordinary skill in the art are inspired by it and, without departing from the gist of the present invention, design similar structural modes and embodiments to this technical solution without creative efforts, they shall fall within the protection scope of the present invention.

Claims

1. A robot health status assessment method based on artificial intelligence, characterized by: The method comprises the following steps: Step S1: Data collection and fusion to obtain robot health status assessment optimization data; Step S2: Digital twin health modeling, using the dynamic dual data stream neural ordinary differential equation modeling method to perform digital twin health modeling and obtain the robot health parameters, including the following steps: Step S21: Physical constraint embedding modeling, using the Hamiltonian neural network combined with the physical constraints of the dynamic system to perform dynamic consistency physical constraint modeling; Step S22: Wear evolution modeling; Step S23: Adversarial verification; Step S24: Digital twin health modeling; Step S3: Multi-scale predictive maintenance, using a multi-time scale prediction architecture combined with quantum annealing optimization and graph attention network to perform multi-scale predictive maintenance and obtain hierarchical predictive maintenance reference data, including the following steps: Step S31: Part-level life prediction, using a long short-term memory neural network combined with quantum annealing optimization to perform part-level life prediction; Step S32: System-level fault propagation graph construction; Step S33: Machine-level maintenance plan optimization; Step S34: Multi-scale integration; Step S35: Hierarchical predictive maintenance; The quantum annealing optimization specifically improves the standard quantum annealing optimization algorithm by introducing a third-order parameter interaction relationship modeling term and a test accuracy term; Step S4: Fault tolerance control, obtaining fault occurrence control instruction data; Step S5: Health status assessment, obtaining comprehensive assessment reference data of the robot's health status.

2. The method for evaluating the health status of a robot based on artificial intelligence according to claim 1, characterized in that: In step S1, the data acquisition and fusion is used to collect raw data, extract spatiotemporal features and perform feature fusion, specifically, to obtain the original data set for robot health status assessment through sensor data collection and dynamic sampling coordination, and to obtain robot health status assessment optimization data through data preprocessing and process feature fusion; The robot health status assessment optimization data specifically includes vibration signal time domain feature data, frequency domain feature data, state variable feature data, robot health derived feature data, load feature data, maintenance history feature data and robot health label data.

3. The method for evaluating the health status of a robot based on artificial intelligence according to claim 2, characterized in that: In step S2, the digital twin health modeling is used to construct a virtual digital twin model of the robot health status, specifically, based on the robot health status evaluation optimization data, a dynamic dual data stream neural ordinary differential equation modeling method is used to perform digital twin health modeling to obtain the robot health parameters, including the following steps: Step S21: physical constraint embedding modeling, specifically, based on the robot health status assessment optimization data, a Hamiltonian neural network combined with the physical constraints of the dynamic system is used to model a dynamic consistency physical constraint model to obtain the robot joint health assessment data; The Hamiltonian neural network combined with the physical constraints of the dynamic system includes a generalized momentum definition, a Hamiltonian dynamic system definition and a physical modeling loss function definition, wherein the generalized momentum definition is used to map the actually observed robot joint angular velocity into a generalized momentum and use it as a prediction indicator; The Hamiltonian dynamics system definition is used to construct a physical evolution model of the robot joint state; Step S22: Wear evolution modeling, specifically, based on the robot health status evaluation optimization data, by constructing a wear evolution neural ordinary differential equation modeling, and constructing a standard multi-layer perceptron structure to perform dynamic evolution of the wear state and future wear prediction modeling, and calculate the robot joint wear rate and predicted wear value data; Step S23: adversarial verification, specifically, performing standard generative adversarial network training based on the predicted wear value data and the actual wear value data, verifying the authenticity of the digital twin health modeling, and obtaining health modeling authenticity evaluation data; Step S24: digital twin health modeling, specifically, performing digital twin health comprehensive modeling through the physical constraint embedding modeling, the wear evolution modeling and the adversarial verification to obtain robot health parameters; The robot health parameters include robot joint simulation consistency index, robot joint wear degree estimation, robot health prediction residual, health modeling authenticity assessment data and robot comprehensive health score.

4. The method for evaluating the health status of a robot based on artificial intelligence according to claim 3, characterized in that: In step S3, the multi-scale predictive maintenance is used to achieve cross-scale prediction from micro-wear to system-level failures. Specifically, based on the robot health parameter and the robot health status evaluation optimization data, a multi-time scale prediction architecture combining quantum annealing optimization and graph attention network is used to perform multi-scale predictive maintenance to obtain hierarchical predictive maintenance reference data, including the following steps: Step S31: Part-level life prediction, based on the robot health status evaluation optimization data, and taking the robot data that meets the need for preventive maintenance in the robot health parameters as the original data input, using the long short-term memory neural network combined with quantum annealing optimization to perform part-level life prediction, and obtain remaining life estimation data; The long short-term memory neural network combined with quantum annealing optimization specifically constructs an improved quantum annealing optimization algorithm and combines it with a standard long short-term memory neural network to perform part-level life prediction, and calculates the remaining life estimation data by modeling the output layer as a Bayesian linear regression; Step S32: constructing a system-level fault propagation graph, specifically, constructing a robot system topology graph and adopting a standard graph attention network method to perform fault propagation modeling based on the robot health parameter and the robot health status evaluation optimization data, and obtaining fault propagation prediction data through fault diffusion probability modeling, including fault propagation probability matrix data and propagation path prediction data; Step S33: optimizing the machine-level maintenance plan, specifically, optimizing the robot maintenance plan based on the robot health parameter and the robot health status evaluation optimization data, using a deep reinforcement learning method combining an improved multi-objective reward function and quantum annealing strategy distillation to obtain optimized maintenance strategy prediction data; Step S34: multi-scale integration, specifically integrating the remaining life estimation data, the fault propagation prediction data and the optimized maintenance strategy prediction data, performing multi-scale robot health predictive maintenance at the part level, system level and machine level, and building a dynamic update mechanism to perform prediction optimization to obtain multi-scale integrated prediction data; Step S35: hierarchical predictive maintenance, specifically, performing hierarchical predictive maintenance based on the multi-scale integrated prediction data to obtain hierarchical predictive maintenance reference data; The hierarchical predictive maintenance reference data includes part-level maintenance parameters, system-level maintenance parameters and machine-level maintenance decisions; The part-level maintenance parameters include part number, remaining useful life of the part, prediction uncertainty estimate, health status parameters, and temperature anomaly values; The system-level maintenance parameters include fault propagation path number, fault propagation probability, fault path criticality assessment value and risk path chain position; The machine-level maintenance decision includes machine number, optional maintenance action set, optimal recommended maintenance action, maintenance action cost, maintenance downtime prediction time, maintenance improvement effect prediction value and maintenance strategy weight; The part-level maintenance parameters are used to assist maintenance technicians in repairing or replacing parts; the system-level maintenance parameters are used to assist engineers in controlling fault propagation paths; and the machine-level maintenance decisions are used to assist robot managers in balancing maintenance planning strategies.

5. The method for evaluating the health status of a robot based on artificial intelligence according to claim 4, characterized in that: In step S33, the deep reinforcement learning method combining the improved multi-objective reward function and the quantum annealing strategy distillation includes constructing the multi-objective reward function and the quantum annealing strategy distillation improvement; The quantum annealing strategy distillation improvement is specifically to optimize the weight of each reward item in the multi-objective reward function according to the improved quantum annealing optimization algorithm, and specifically to replace the test set verification accuracy function item in the multi-objective reward function formula with the reward value; The deep reinforcement learning method constructs robot maintenance state space parameters and robot maintenance action space parameters, and performs standard reinforcement learning training according to the multi-objective reward function.

6. The method for evaluating the health status of a robot based on artificial intelligence according to claim 5, characterized in that: In step S4, the fault-tolerant control is used to maintain the basic functions of the robot when a fault occurs, specifically by dynamically mapping the health status of key components of the robot into control constraints based on the hierarchical predictive maintenance reference data and the robot health status assessment optimization data, constructing a model predictive controller with health limits, performing fault-tolerant control, and obtaining fault occurrence control instruction data; The fault occurrence control instruction data includes control amount adjustment data and safety action limitation data.

7. The method for evaluating the health status of a robot based on artificial intelligence according to claim 6, characterized in that: In step S5, the health status assessment is used to generate a comprehensive health status assessment result, specifically by combining the robot health parameters, the hierarchical predictive maintenance reference data and the fault occurrence control instruction data to perform a comprehensive assessment of the robot health status and obtain the robot health status comprehensive assessment reference data.

8. A robot health status assessment system based on artificial intelligence, used to implement the robot health status assessment method based on artificial intelligence as described in any one of claims 1 to 7, characterized in that: It includes data acquisition and fusion module, digital twin health modeling module, multi-scale predictive maintenance module, fault-tolerant control module and health status assessment module.

9. The robot health status assessment system based on artificial intelligence according to claim 8, characterized in that: The data acquisition and fusion module is used for data acquisition and fusion, and obtains robot health status evaluation and optimization data through data acquisition and fusion, and sends the robot health status evaluation and optimization data to the digital twin health modeling module, the multi-scale predictive maintenance module and the fault-tolerant control module; The digital twin health modeling module is used for digital twin health modeling, obtains robot health parameters through digital twin health modeling, and sends the robot health parameters to the multi-scale predictive maintenance module and the health status assessment module; The multi-scale predictive maintenance module is used for multi-scale predictive maintenance, obtains hierarchical predictive maintenance reference data through multi-scale predictive maintenance, and sends the hierarchical predictive maintenance reference data to the fault tolerance control module and the health status assessment module; The fault tolerance control module is used for fault tolerance control, obtains fault occurrence control instruction data through fault tolerance control, and sends the fault occurrence control instruction data to the health status assessment module; The health status assessment module is used for health status assessment, and obtains robot health status comprehensive assessment reference data through health status assessment.

Images