Robot intelligent operation and maintenance system based on digital twinning

Abstract

The application discloses a robot intelligent operation and maintenance system based on digital twinning, relates to the technical field of operation and maintenance management, and comprises the following steps: performing hidden state estimation uncertainty analysis on environment dynamic change data of a robot key component according to initial parameters of a digital twinning model, obtaining hidden state estimation uncertainty data, performing robot health state deviation correlation mining based on the hidden state estimation uncertainty data, and obtaining health state deviation correlation data; performing adaptive adjustment of the digital twinning model parameters according to the health state deviation correlation data, generating digital twinning model parameter adjustment data, performing robot health state evaluation and fault prediction through the digital twinning model parameter adjustment data, and obtaining robot health evaluation and fault prediction data; and performing robot predictive maintenance decision planning based on the robot health evaluation and fault prediction data, and generating robot intelligent operation and maintenance decision data.

Description

Technical Field

[0001] This application relates to the field of operation and maintenance management technology, and in particular to a robot-based intelligent operation and maintenance system based on digital twins. Background Technology

[0002] A robot is an intelligent machine that can work semi-autonomously or fully autonomously. Robots can perform tasks such as jobs or movement through programming and automatic control.

[0003] In related technologies, during robot operation and maintenance, critical states such as internal gear wear and bearing lubrication status cannot be directly measured by sensors. They can only be inferred from indirect observation data (such as motor current and sound signals). Digital twin models need to estimate hidden states based on these partial observation data to achieve accurate health assessment and fault prediction. Moreover, existing methods for estimating hidden states often rely on historical data patterns. When the robot's operating environment or load conditions change dynamically, the estimation results will have high uncertainty. For example, when the load suddenly increases, changes in motor current may be misjudged as a fault when it is actually a normal adjustment, thereby reducing the efficiency of intelligent robot operation and maintenance. There are areas for improvement. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this application provides a robot-based intelligent operation and maintenance system based on digital twins.

[0005] The robot-based intelligent operation and maintenance system based on digital twins provided in this application includes: The acquisition module is used to acquire indirect observation data of the robot and initial parameters of the digital twin model, and to perform dynamic change analysis of the robot's operating environment and load based on the indirect observation data, thereby obtaining dynamic change data of the environment. The analysis module is used to perform hidden state estimation uncertainty analysis on the dynamic environmental change data of the robot based on the initial parameters of the digital twin model, to obtain hidden state estimation uncertainty data, and to perform robot health state deviation correlation mining based on the hidden state estimation uncertainty data to obtain health state deviation correlation data. The adjustment module is used to adaptively adjust the parameters of the digital twin model based on the health status deviation correlation data, generate digital twin model parameter adjustment data, and use the digital twin model parameter adjustment data to perform robot health status assessment and fault prediction, thereby obtaining robot health assessment and fault prediction data. The decision-making module is used to make predictive maintenance decisions and plans for the robot based on the robot health assessment and fault prediction data, and generate intelligent operation and maintenance decision data for the robot.

[0006] Preferably, the indirect observation data includes motor current signals and sound signals, and the initial parameters of the digital twin model include the model network structure, initial weights, hidden state estimation model, model learning rate, and number of network layers.

[0007] Preferably, the process of obtaining dynamic environmental change data specifically includes: Obtain indirect observation data and initial parameters of the digital twin model corresponding to the robot; The indirect observation data is preprocessed to generate preprocessed indirect observation data; The preprocessed indirect observation data is used to extract the robot's operating environment load change data. Based on the initial parameters of the digital twin model, the dynamic changes of the robot's operating environment are analyzed using the load change data of the operating environment to obtain dynamic environmental change data.

[0008] Preferably, based on the hidden state estimation uncertainty data, robot health state deviation correlation mining is performed to obtain health state deviation correlation data, specifically including: Extract the hidden state estimation model corresponding to the initial parameters of the digital twin model; Based on the initial parameters of the digital twin model and the hidden state estimation model, an uncertainty analysis of the hidden state estimation for the robot transmission system is performed on the dynamic environmental change data to obtain the hidden state estimation uncertainty data. The hidden state estimation uncertainty data is fitted with robot health state deviation tolerance data to generate health state deviation tolerance data. Based on the health status deviation tolerance data and the hidden state estimation uncertainty data, the robot health status deviation correlation mining is performed to obtain health status deviation correlation data.

[0009] Preferably, the process of obtaining the uncertainty data for the hidden state estimation specifically includes: Extract the model learning rate and network layer number from the initial parameters of the digital twin model, and calculate the correlation between robot observation data and the hidden state of mechanical parts based on the hidden state estimation model to obtain the correlation coefficient between observation and state; Based on the environmental dynamic change data, the correlation coefficient between the observation and the state is analyzed to determine the impact of robot dynamic load change, and dynamic load change impact data is obtained. Based on the model learning rate and the number of network layers, the hidden state estimation simulation of robot parts under multiple time steps is performed on the dynamic load change impact data to obtain a multi-time step hidden state estimation sequence. The robot state estimation error propagation behavior imbalance analysis is performed on the multi-time-step hidden state estimation sequence to obtain estimation error propagation behavior imbalance data; Perform robot operation and maintenance uncertainty quantification regression analysis on the imbalance data of the estimated error propagation behavior to obtain uncertainty quantification regression data; Based on the imbalance data of the propagation behavior of the estimation error and the uncertainty quantification regression data, the hidden state estimation uncertainty data is obtained by performing hidden state estimation uncertainty analysis on the environmental dynamic change data.

[0010] Preferably, the robot state estimation error propagation behavior imbalance analysis is performed on the multi-time-step hidden state estimation sequence to obtain estimation error propagation behavior imbalance data, specifically including: The robot state estimation error density non-uniformity is identified in the multi-time-step hidden state estimation sequence to obtain the estimation error density non-uniformity vector; Based on the non-uniformity vector of the estimation error density, the time-series variation differential coupling of the robot's hidden state estimation deviation is performed to generate estimation deviation time-series variation differential coupling data; Based on the non-uniformity vector of the estimated error density and the time-series variation differential coupling data of the estimated deviation, a local calculation of the prediction variance of the robot digital twin model is performed to obtain the local data of the model prediction variance. Based on the non-uniformity vector of the estimated error density, the differential coupling data of the temporal variation of the estimated deviation, and the local data of the model prediction variance, the error propagation ratio difference between the hidden states of each robot component is deduced, and the error propagation ratio difference between different hidden states is generated. Based on the difference in the error propagation ratio between the different hidden states, an imbalance analysis of the robot's estimation error propagation behavior is performed to obtain the estimation error propagation behavior imbalance data.

[0011] Preferably, the hidden state estimation uncertainty data is fitted with robot health state deviation tolerance data to generate health state deviation tolerance data, specifically including: The hidden state estimation uncertainty data is subjected to robot operation and maintenance uncertainty network partitioning processing to obtain uncertainty network partitioning data; The geometric feature analysis of robot health state deviation is performed on the uncertain network partitioning data to obtain the geometric feature data of health state deviation. Based on the aforementioned health state deviation geometric feature data, the force offset distribution of key mechanical components of the robot is simulated to generate component force offset distribution data. The stress offset distribution data of the component is subjected to strain rate component decomposition of the robot structure stress distribution to obtain strain rate component decomposition data. Based on the strain rate component decomposition data, the robot health state deviation tolerance is fitted to generate health state deviation tolerance data.

[0012] Preferably, robot health status assessment and fault prediction are performed using the parameter adjustment data of the digital twin model to obtain robot health assessment and fault prediction data, specifically including: Robot operation and maintenance feature learning is performed on the health status deviation correlation data to obtain health status deviation correlation feature data; Based on the health status deviation correlation feature data, the initial parameters of the digital twin model are adjusted for robot health prediction, generating digital twin model parameter adjustment data. The parameter adjustment data of the digital twin model are used to optimize the robot fault prediction accuracy, thereby obtaining optimized model prediction accuracy data; The robot's health status is assessed and faults are predicted by optimizing the model's prediction accuracy data, thus obtaining robot health assessment and fault prediction data.

[0013] Preferably, the initial parameters of the digital twin model are adjusted based on the health status deviation correlation feature data to generate digital twin model parameter adjustment data for robot health prediction, specifically including: Based on the health status deviation correlation feature data, the initial parameters of the digital twin model are matched with the robot operation and maintenance model learning rate to obtain model learning rate matching data; Based on the model learning rate matching data and the health status deviation correlation feature data, the number of network layers in the robot digital twin model is adjusted and controlled to obtain network layer control data; The robot's hidden state estimation convergence speed is matched by performing a process on the network layer control data to obtain hidden state estimation convergence speed matching data. Based on the model learning rate matching data, the network layer number control data, and the hidden state estimation convergence speed matching data, the initial parameters of the digital twin model are adjusted to generate digital twin model parameter adjustment data.

[0014] Preferably, the parameter adjustment data of the digital twin model is used to optimize the robot fault prediction accuracy, thereby obtaining optimized model prediction accuracy data, specifically including: The deviation between the model prediction output and the actual observation is calculated based on the parameter adjustment data of the digital twin model to obtain the deviation data between the model prediction and the actual observation. Based on the deviation data between the model prediction and the actual observation, the model residual distribution data is analyzed by adjusting the parameters of the digital twin model to obtain the model residual distribution data. The model prediction local variance is calculated based on the model residual distribution data to obtain the model prediction local variance data; Based on the deviation data between the model prediction and the actual observation, the model residual distribution data, and the model prediction local variance data, the model prediction accuracy is optimized and adjusted by adjusting the parameter adjustment data of the digital twin model, thereby obtaining optimized model prediction accuracy data.

[0015] Preferably, predictive maintenance decision planning for the robot is performed based on the robot health assessment and fault prediction data to generate intelligent operation and maintenance decision data for the robot, specifically including: The robot health assessment and fault prediction data are used to calculate a health score based on the robot's operating status, thereby generating health score data. Based on the health score data, the robot's failure risk level is classified to obtain failure risk level data; Based on the fault risk level data, predictive maintenance action sequence planning for the robot is performed to generate maintenance action sequence data; Robot operation and maintenance decision-making is executed based on the maintenance action sequence data, thereby obtaining intelligent operation and maintenance decision data for the robot.

[0016] In summary, this application includes at least one of the following beneficial technical effects: 1. This application provides a robot intelligent operation and maintenance system based on digital twins. Through the acquisition module, the system analyzes the dynamic changes of the operating environment and load by indirectly observed data. It can effectively distinguish between normal working condition adjustments and potential faults, and avoid misjudgments caused by dynamic factors such as sudden load changes. The analysis module further combines the initial parameters of the digital twin model to perform hidden state estimation uncertainty analysis and mine the correlation data of health state deviation. Thus, it can still accurately estimate the hidden state of key robot components under environmental change conditions, significantly reduce estimation uncertainty, and improve the reliability of state monitoring. 2. The adjustment module, based on health status deviation correlation data, enables adaptive adjustment of digital twin model parameters, allowing the model to dynamically respond to changes in the robot's operating environment and load. By optimizing the health status assessment and fault prediction process through model parameter adjustment data, it effectively overcomes the shortcomings of traditional methods that rely on historical data patterns and have insufficient generalization ability under dynamic conditions, thereby improving the prediction accuracy and stability of the model under complex working conditions. 3. The decision-making module uses robot health assessment and fault prediction data to make predictive maintenance decisions and generate intelligent operation and maintenance decision data. This enables the early identification of fault risk levels and the planning of maintenance action sequences, realizing the transformation from passive maintenance to proactive predictive maintenance, significantly reducing unexpected downtime, lowering maintenance costs, and improving the overall operation and maintenance efficiency and safety of the robot system. 4. By constructing a complete digital twin-driven operation and maintenance closed loop from data collection, uncertainty analysis, model adjustment to decision planning, and multi-module collaboration, a refined monitoring and management of robot health status is achieved. This overcomes the low operation and maintenance efficiency caused by estimation error propagation and behavioral imbalance in existing technologies, and provides a reliable guarantee for the long-term stable operation of industrial robots. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of a robot-based intelligent operation and maintenance system based on digital twins, according to an embodiment of this application. Detailed Implementation

[0019] The following is in conjunction with the appendix Figure 1 This application will be described in further detail. Example

[0020] This application also discloses a robot-based intelligent operation and maintenance system based on digital twins.

[0021] Reference Figure 1 A robot-based intelligent operation and maintenance system based on digital twins includes: The acquisition module is used to acquire indirect observation data of the robot and initial parameters of the digital twin model, and to perform dynamic change analysis of the robot's operating environment and load based on the indirect observation data, thereby obtaining dynamic change data of the environment. For example, the indirect observation data includes motor current signals and sound signals, and the initial parameters of the digital twin model include the model network structure, initial weights, hidden state estimation model, model learning rate, and number of network layers.

[0022] For example, the process of obtaining dynamic environmental change data specifically includes: Obtain indirect observation data and initial parameters of the digital twin model corresponding to the robot; The indirect observation data is preprocessed to generate preprocessed indirect observation data; The preprocessed indirect observation data is used to extract the robot's operating environment load change data. Based on the initial parameters of the digital twin model, the dynamic changes of the robot's operating environment are analyzed using the load change data of the operating environment to obtain dynamic environmental change data.

[0023] Specifically, the process of obtaining dynamic environmental change data includes: First, acquiring indirect observation data and initial parameters of the digital twin model corresponding to the robot. The indirect observation data originates from the robot's motor current and sound signals, which are collected in real-time by sensors. The initial parameters of the digital twin model include the model network structure, initial weights, hidden state estimation model, model learning rate, and number of network layers. In this embodiment, the model network structure uses a recurrent neural network or a long short-term memory network to process time-series data. The initial weights are a set of small random values ​​set through random initialization or pre-training to define the model connection strength. The hidden state estimation model, as a functional module, is responsible for inferring the robot's internal state, such as gear wear or bearing lubrication, from the observation data. The model learning rate controls the hyperparameter of the update step size, and the number of network layers defines the number of hidden layers to balance complexity and generalization ability. Next, the indirect observation data undergoes signal preprocessing by applying digital filtering techniques and signal standardization methods to eliminate environmental interference and sensor errors, thereby generating preprocessed indirect observation data. The process begins with receiving observational data. Then, the preprocessed indirect observational data is used to extract load changes in the robot's operating environment. Time-domain analysis and frequency-domain transformation are employed, and load fluctuation patterns are identified by calculating signal amplitude, energy spectral density, and statistical characteristics. A sliding window technique is used to segment the data stream to detect sudden increases or decreases in load, thus obtaining load change data for the operating environment. Finally, based on the initial parameters of the digital twin model, dynamic changes in the robot's operating environment are analyzed. This process utilizes the model network structure and initial weights of the hidden state estimation model. Forward propagation is used to calculate the correlation between the indirect observational data and the hidden state, and dynamic time warping or correlation coefficient analysis is applied to assess the impact of load changes on the robot's state. Simultaneously, multi-time-step simulations are performed using the model learning rate and network layer number. Iterative parameter adjustments quantify the dynamic characteristics of environmental changes (such as the rate and duration of load change), outputting dynamic environmental change data to provide a foundation for subsequent steps. This process ensures a coherent data stream, forming a closed-loop processing chain to enhance the accuracy and adaptability of intelligent robot operation and maintenance.

[0024] The analysis module is used to perform hidden state estimation uncertainty analysis on the dynamic environmental change data of the robot based on the initial parameters of the digital twin model, to obtain hidden state estimation uncertainty data, and to perform robot health state deviation correlation mining based on the hidden state estimation uncertainty data to obtain health state deviation correlation data. For example, based on the hidden state estimation uncertainty data, robot health state deviation correlation mining is performed to obtain health state deviation correlation data, specifically including: Extract the hidden state estimation model corresponding to the initial parameters of the digital twin model; Based on the initial parameters of the digital twin model and the hidden state estimation model, an uncertainty analysis of the hidden state estimation for the robot transmission system is performed on the dynamic environmental change data to obtain the hidden state estimation uncertainty data. The hidden state estimation uncertainty data is fitted with robot health state deviation tolerance data to generate health state deviation tolerance data. Based on the health status deviation tolerance data and the hidden state estimation uncertainty data, the robot health status deviation correlation mining is performed to obtain health status deviation correlation data.

[0025] Specifically, based on the hidden state estimation uncertainty data, the robot health state deviation correlation mining is performed to obtain health state deviation correlation data. This includes: First, extracting the corresponding hidden state estimation model from the initial parameters of the digital twin model. This process involves analyzing the model network structure and initial weights to determine the core computational units and state transition functions, thereby constructing a functional module capable of inferring the hidden state within the robot's transmission system from indirect observation data. Next, based on the initial parameters of the digital twin model (including the model learning rate and network layer number) and the extracted hidden state estimation model, performing uncertainty analysis on the hidden state estimation of the robot's transmission system using dynamic environmental change data. This is achieved by applying probabilistic inference methods, where the forward propagation of the model is used to calculate the conditional probability distribution between the observed data and the hidden state, and the uncertainty of the estimation result is quantified by calculating the confidence interval or variance index. Simultaneously, the load fluctuation information in the dynamic environmental change data is combined to evaluate the propagation behavior of uncertainty over time, thereby outputting the hidden state estimate. Uncertainty data is first collected. Then, the robot health status deviation tolerance is fitted to the hidden state estimation uncertainty data. The above steps employ statistical distribution fitting techniques, calculating the tolerance boundary of the health status by analyzing the distribution characteristics of the uncertainty data. Specifically, the deviation threshold is determined through iterative optimization methods, maximizing the probability that the actual state falls within the tolerance range at a given confidence level (e.g., 95%), and considering the operating specifications of the robot's transmission system, thereby generating health status deviation tolerance data. Finally, based on the health status deviation tolerance data and the hidden state estimation uncertainty data, robot health status deviation correlation mining is performed. This involves using association rule mining algorithms to analyze the correlation between uncertainty patterns and health deviations, identifying key correlation features by calculating the covariance matrix or mutual information index, and combining time series alignment techniques to match the deviation tolerance data with uncertainty fluctuations to extract potential factors and dependencies affecting the robot's health status, ultimately obtaining health status deviation correlation data, providing a basis for subsequent model optimization and operation and maintenance decisions. The above process ensures the coherence of the data flow from model extraction to uncertainty analysis, tolerance fitting, and correlation mining, forming a closed-loop processing chain to enhance the accuracy and robustness of robot health assessment.

[0026] For example, the process of obtaining the uncertainty data for the hidden state estimation specifically includes: Extract the model learning rate and network layer number from the initial parameters of the digital twin model, and calculate the correlation between robot observation data and the hidden state of mechanical parts based on the hidden state estimation model to obtain the correlation coefficient between observation and state; Based on the environmental dynamic change data, the correlation coefficient between the observation and the state is analyzed to determine the impact of robot dynamic load change, and dynamic load change impact data is obtained. Based on the model learning rate and the number of network layers, the hidden state estimation simulation of robot parts under multiple time steps is performed on the dynamic load change impact data to obtain a multi-time step hidden state estimation sequence. The robot state estimation error propagation behavior imbalance analysis is performed on the multi-time-step hidden state estimation sequence to obtain estimation error propagation behavior imbalance data; Perform robot operation and maintenance uncertainty quantification regression analysis on the imbalance data of the estimated error propagation behavior to obtain uncertainty quantification regression data; Based on the imbalance data of the propagation behavior of the estimation error and the uncertainty quantification regression data, the hidden state estimation uncertainty data is obtained by performing hidden state estimation uncertainty analysis on the environmental dynamic change data.

[0027] Specifically, the process of obtaining the hidden state estimation uncertainty data includes the following steps: First, extracting the model learning rate and network layer number from the initial parameters of the digital twin model, and calculating the correlation between robot observation data and the hidden state of mechanical parts based on the hidden state estimation model. This step is achieved by applying statistical correlation analysis methods, where the covariance matrix or information entropy index between the model's forward propagation output and the actual observation data is used to quantify the linear or nonlinear dependence between variables, thereby obtaining the correlation coefficient between observation and state. Next, based on the dynamic environmental change data, the impact of robot dynamic load changes on the correlation coefficient between observation and state is analyzed. Weighted regression or sliding window techniques are introduced to adjust the correlation coefficient. Specifically, based on the load mutation points and trend changes in the dynamic environmental change data, the dynamic adjustment factor of the correlation coefficient is calculated, and time series decomposition methods are applied to separate the load-affected components, thus obtaining the dynamic load change impact data. Then, based on the model learning rate and network layer number, multi-time-step simulation analysis of the robot part hidden state estimation is performed on the dynamic load change impact data. This process uses the learning rate to control the parameter update amplitude and combines the network layer number to construct a deep time series model (e.g., through recursive calculation of a recurrent neural network or an attention mechanism). The robot generates a multi-time-step hidden state estimation sequence by simulating state estimation at multiple time points through iterative forward propagation and integrating dynamic load impact data as input perturbations. Subsequently, an imbalance analysis of robot state estimation error propagation behavior is performed on the multi-time-step hidden state estimation sequence. This involves applying error propagation theory to analyze the error diffusion pattern in the time dimension and using statistical hypothesis testing to identify asymmetric or abnormal fluctuation points in the error distribution, thereby obtaining imbalance data on estimation error propagation behavior. Afterwards, a robot operation and maintenance uncertainty quantification regression analysis is performed on the imbalance data of estimation error propagation behavior, using a regression modeling method with the error propagation data as the self-evaluation model. The variable, with the uncertainty index as the dependent variable, is fitted with a regression curve through least squares optimization or maximum likelihood estimation, and the residual distribution is calculated to quantify the uncertainty level, thus obtaining uncertainty quantification regression data. Finally, based on the imbalance data of estimation error propagation behavior and uncertainty quantification regression data, hidden state estimation uncertainty analysis is performed on the environmental dynamic change data. The above steps integrate multi-source data, adjust the uncertainty estimate in combination with environmental change patterns (such as load periodicity), and apply probabilistic graphical models or Bayesian inference to integrate error propagation and regression results, thereby outputting hidden state estimation uncertainty data, providing a foundation for subsequent health status assessment. The above process ensures the coherence of the data flow from parameter extraction to correlation calculation, load impact analysis, multi-time step simulation, error propagation analysis, uncertainty quantification, and final integration, forming a closed-loop processing chain to enhance the robustness and accuracy of robot intelligent operation and maintenance.

[0028] For example, performing robot state estimation error propagation behavior imbalance analysis on the multi-time-step hidden state estimation sequence to obtain estimation error propagation behavior imbalance data specifically includes: The robot state estimation error density non-uniformity is identified in the multi-time-step hidden state estimation sequence to obtain the estimation error density non-uniformity vector; Based on the non-uniformity vector of the estimation error density, the time-series variation differential coupling of the robot's hidden state estimation deviation is performed to generate estimation deviation time-series variation differential coupling data; Based on the non-uniformity vector of the estimated error density and the time-series variation differential coupling data of the estimated deviation, a local calculation of the prediction variance of the robot digital twin model is performed to obtain the local data of the model prediction variance. Based on the non-uniformity vector of the estimated error density, the differential coupling data of the temporal variation of the estimated deviation, and the local data of the model prediction variance, the error propagation ratio difference between the hidden states of each robot component is deduced, and the error propagation ratio difference between different hidden states is generated. Based on the difference in the error propagation ratio between the different hidden states, an imbalance analysis of the robot's estimation error propagation behavior is performed to obtain the estimation error propagation behavior imbalance data.

[0029] Specifically, an imbalance analysis of robot state estimation error propagation behavior is performed on the multi-time-step hidden state estimation sequence to obtain imbalance data of estimation error propagation behavior. This includes: firstly, identifying the non-uniformity of robot state estimation error density in the multi-time-step hidden state estimation sequence. This process involves dividing the time series into multiple continuous time periods using a sliding window technique, and calculating the probability density function of the error distribution within each time period using a kernel density estimation method. Error clustering regions are identified by comparing the peak differences and distribution pattern changes of the density functions between adjacent time periods. Simultaneously, skewness and kurtosis indices are calculated using statistical moment analysis of the error values. The asymmetry and sharpness of the quantified distribution are used to obtain the estimated error density non-uniformity vector, which characterizes the heterogeneity of the error spatial distribution. Then, based on this estimated error density non-uniformity vector, a differential coupling of the temporal variation of the robot's hidden state estimation bias is performed. This is done by constructing a time differential operator to capture the rate of change of the error vector at continuous time points. Numerical differentiation methods are used to calculate the first and second derivatives of each component of the error density vector with respect to time. The derivatives of different components are then weighted and fused using a coupling function, where the weights are dynamically adjusted according to the magnitude of the error density to highlight the influence of high-density regions. This generates an estimated bias that reflects the temporal evolution pattern of the error. The process involves first obtaining differentially coupled data representing the time-series variation of the estimation error density; then, based on the non-uniform vector of the estimation error density and the differentially coupled data representing the time-series variation of the estimation deviation, a local calculation of the prediction variance for the robot's digital twin model is performed. This step establishes a variance propagation model, using the non-uniform vector as a spatial weighting factor and the differentially coupled data as a temporal dynamic factor. A local weighted regression method is used to calculate the prediction variance at each time point and state dimension. An iterative optimization algorithm is used to adjust the smoothing parameter in the variance calculation to adapt to the fluctuation characteristics of different regions, thereby obtaining local model prediction variance data that accurately characterizes the model's prediction uncertainty. Based on this, the estimation error density... The error propagation ratio difference between the hidden states of various robot components is derived by using non-uniformity vectors, differential coupling data of estimation bias temporal variation, and local data of model prediction variance. This is achieved by constructing an error propagation network model, treating the hidden states of different components as network nodes, using non-uniformity vectors to determine the connection strength between nodes, using differential coupling data to calculate the error propagation rate, and combining local variance data to evaluate the sensitivity of nodes to received errors. Finally, the degree of imbalance in error propagation is quantified by calculating the ratio difference between the error input and output of adjacent nodes, thereby generating the error propagation ratio difference between the hidden states that reveal the heterogeneity of error propagation between components.Finally, based on the difference in error propagation ratios among the different hidden states, an imbalance analysis of the robot's estimated error propagation behavior is conducted. This is achieved by establishing an imbalance evaluation function, mapping the ratio difference data to a preset imbalance metric space, using cluster analysis to identify abnormal propagation patterns, and combining this with a time consistency test to screen for persistent imbalance signals. By weighted aggregation of imbalance indicators for each component and setting thresholds, it is determined whether the overall propagation behavior deviates from the normal range, thereby obtaining comprehensive estimated error propagation behavior imbalance data that characterizes the degree of error propagation anomalies.

[0030] For example, the hidden state estimation uncertainty data is fitted with robot health state deviation tolerance data to generate health state deviation tolerance data, specifically including: The hidden state estimation uncertainty data is subjected to robot operation and maintenance uncertainty network partitioning processing to obtain uncertainty network partitioning data; The geometric feature analysis of robot health state deviation is performed on the uncertain network partitioning data to obtain the geometric feature data of health state deviation. Based on the aforementioned health state deviation geometric feature data, the force offset distribution of key mechanical components of the robot is simulated to generate component force offset distribution data. The stress offset distribution data of the component is subjected to strain rate component decomposition of the robot structure stress distribution to obtain strain rate component decomposition data. Based on the strain rate component decomposition data, the robot health state deviation tolerance is fitted to generate health state deviation tolerance data.

[0031] Specifically, the hidden state estimation uncertainty data is fitted with robot health state deviation tolerance data to generate health state deviation tolerance data. This process includes: firstly, performing robot operation and maintenance uncertainty network partitioning on the hidden state estimation uncertainty data. This process involves constructing a graph-based uncertainty propagation network model, using the uncertainty data of each robot component as network nodes, employing a community detection algorithm to cluster the uncertainty based on the similarity of uncertainty values, and determining network connection weights by calculating the information entropy transmission strength between nodes. Simultaneously, a sliding time window technique is used to dynamically segment the uncertainty data, thereby obtaining uncertainty data reflecting the spatial distribution characteristics of uncertainty. The network is partitioned into data. Next, geometric feature analysis of robot health state deviation is performed on this uncertain network partitioned data. This is done by extracting the uncertainty distribution morphology features of each network partition, including calculating the convex hull area, centroid location, and boundary curvature changes of the uncertainty values. Principal component analysis is used for dimensionality reduction to obtain the main geometric feature vectors. Simultaneously, fractal dimension calculation is combined to quantify the distribution complexity, thereby obtaining geometric feature data of health state deviation characterizing the uncertain spatial structure. Then, based on the geometric feature data of health state deviation, the force offset distribution of key mechanical components of the robot is simulated. This step maps the geometric feature data to the robot by establishing a force transmission model for the components. In the kinematic model, the stress concentration region under uncertainty is calculated using finite element analysis. The force offset of the component under different working conditions is solved using numerical integration. Simultaneously, probabilistic simulation of the force distribution is performed using material mechanical property parameters, generating component force offset distribution data reflecting the abnormal force distribution of the mechanical component. Subsequently, the component force offset distribution data is decomposed into strain rate components of the robot structure stress distribution. This is done by constructing a strain rate tensor model, converting the force offset data into a strain energy density distribution, and using eigenvalue decomposition to decompose the strain rate tensor into normal strain rate and shear strain rate components. The stress is then quantified by calculating the time change rate of each component. The evolution rate is varied, and the elastic and plastic strain components are separated by combining the stress-strain constitutive relationship, thus obtaining strain rate component decomposition data that accurately describes the structural deformation characteristics. Finally, based on the strain rate component decomposition data, robot health state deviation tolerance fitting is performed. This is done by establishing a tolerance limit state function, using strain rate component data as input variables, constructing a tolerance boundary surface using the response surface method, generating a large number of random samples through Monte Carlo simulation for reliability analysis, and combining fault thresholds from historical operation and maintenance data for parameter calibration. An optimization algorithm is then used to solve for the maximum allowable deviation range at a given confidence level, thereby generating health state deviation tolerance data to guide operation and maintenance decisions.

[0032] The adjustment module is used to adaptively adjust the parameters of the digital twin model based on the health status deviation correlation data, generate digital twin model parameter adjustment data, and use the digital twin model parameter adjustment data to perform robot health status assessment and fault prediction, thereby obtaining robot health assessment and fault prediction data. For example, robot health status assessment and fault prediction are performed using the parameter adjustment data of the digital twin model to obtain robot health assessment and fault prediction data, specifically including: Robot operation and maintenance feature learning is performed on the health status deviation correlation data to obtain health status deviation correlation feature data; Based on the health status deviation correlation feature data, the initial parameters of the digital twin model are adjusted for robot health prediction, generating digital twin model parameter adjustment data. The parameter adjustment data of the digital twin model are used to optimize the robot fault prediction accuracy, thereby obtaining optimized model prediction accuracy data; The robot's health status is assessed and faults are predicted by optimizing the model's prediction accuracy data, thus obtaining robot health assessment and fault prediction data.

[0033] Specifically, robot health status assessment and fault prediction are performed using the adjusted parameters of the digital twin model to obtain robot health assessment and fault prediction data. This process includes: first, learning robot operation and maintenance features from health status deviation correlation data. Principal component analysis is used to reduce the dimensionality of multidimensional features in the correlation data, extracting the main feature vectors that best represent changes in health status. Simultaneously, an autoencoder neural network is used to perform nonlinear transformation and reconstruction of the feature data. The most discriminative feature combinations are selected by calculating the reconstruction error, and time series analysis is combined to capture the evolution of features over time, thus obtaining health status deviation correlation feature data containing key health information. Next, the initial parameters of the digital twin model are adjusted based on the health status deviation correlation feature data for robot health prediction. A parameter sensitivity analysis model is constructed to calculate the contribution of each layer weight to the health prediction result. Gradient descent algorithm is used to iteratively update the weight parameters along the negative direction of the loss function. Simultaneously, the model learning rate is dynamically adjusted according to the distribution characteristics of different health states in the feature data, and regularization methods are used to control the effective complexity of the network layers. To prevent overfitting, parameter adjustment data for a digital twin model matching the current health status is generated. Subsequently, the robot fault prediction accuracy is optimized using this parameter adjustment data. This process involves establishing a prediction confidence assessment mechanism, calculating the deviation distribution between the predicted results and the actual values ​​under different parameter configurations, employing ensemble learning to fuse the prediction outputs of multiple models to improve stability, determining the optimal parameter combination through cross-validation, and combining prior knowledge from historical fault data to perform Bayesian correction on the prediction results, thereby obtaining more reliable model prediction accuracy optimization data. Finally, the robot's health status is assessed and faults are predicted using the optimized model prediction accuracy data. This is achieved by inputting real-time sensor data into the optimized digital twin model, calculating the probability distribution of the health status of each component using forward propagation, determining the critical value of the health status based on the equipment's operating history using a dynamic threshold segmentation method, constructing a fault propagation graph model to analyze the evolution path of potential faults, and estimating the remaining service life using a time series prediction algorithm. This yields robot health assessment and fault prediction data containing health scores, fault probabilities, and early warning information.

[0034] For example, the initial parameters of the digital twin model are adjusted based on the health status deviation correlation feature data to generate digital twin model parameter adjustment data for robot health prediction, specifically including: Based on the health status deviation correlation feature data, the initial parameters of the digital twin model are matched with the robot operation and maintenance model learning rate to obtain model learning rate matching data; Based on the model learning rate matching data and the health status deviation correlation feature data, the number of network layers in the robot digital twin model is adjusted and controlled to obtain network layer control data; The robot's hidden state estimation convergence speed is matched by performing a process on the network layer control data to obtain hidden state estimation convergence speed matching data. Based on the model learning rate matching data, the network layer number control data, and the hidden state estimation convergence speed matching data, the initial parameters of the digital twin model are adjusted to generate digital twin model parameter adjustment data.

[0035] Specifically, the initial parameters of the digital twin model are adjusted for robot health prediction based on the health status deviation correlation feature data, generating digital twin model parameter adjustment data. This includes: firstly, matching the learning rate of the robot operation and maintenance model to the initial parameters of the digital twin model based on the health status deviation correlation feature data. This process analyzes the fluctuation frequency and magnitude of health status changes in the feature data, uses an adaptive adjustment algorithm to calculate the optimal learning rate value during model training, specifically by evaluating the gradient change trend of the feature data to determine the direction of learning rate adjustment, and uses an exponentially weighted moving average method to smooth the learning rate update process. Simultaneously, the learning rate is dynamically adjusted based on historical training effect feedback, thus obtaining model learning rate matching data that matches the current health status characteristics. Next, the number of network layers in the robot digital twin model is adjusted and controlled based on the model learning rate matching data and the health status deviation correlation feature data. This is achieved by constructing a network depth optimization model, analyzing the representation ability of different network layers for health features, using an inter-layer contribution evaluation method to calculate the relative importance of each hidden layer to the final prediction result, and removing hidden layers using a pruning algorithm. Redundant network layers are used, and the distribution of connection weights between network layers is adjusted based on learning rate matching data to obtain optimized network layer number control data. Subsequently, the robot hidden state estimation convergence speed is matched using the network layer number control data. The above steps involve establishing a convergence performance evaluation function, analyzing the impact of network structure changes on the training process, using an iteration count prediction model to estimate the number of training rounds required to reach the target accuracy, and optimizing the convergence trajectory by adjusting momentum parameters and batch size. An early stopping mechanism is also incorporated to prevent overfitting, thus obtaining hidden state estimation convergence speed matching data that ensures training efficiency. Finally, based on the model learning rate matching data, network layer number control data, and hidden state estimation convergence speed matching data, the initial parameters of the digital twin model are adjusted. A multi-objective optimization framework is constructed, comprehensively considering three dimensions: learning rate stability, network structure complexity, and training efficiency. A parameter coordination algorithm is used to balance the mutual influence between various adjustment factors, and the parameter combination is iteratively optimized using gradient descent. Simultaneously, cross-validation technology is used to verify the performance of the adjusted model, thereby generating digital twin model parameter adjustment data that can effectively improve the accuracy of health prediction.

[0036] For example, the parameter adjustment data of the digital twin model is used to optimize the robot fault prediction accuracy, thereby obtaining optimized model prediction accuracy data, specifically including: The deviation between the model prediction output and the actual observation is calculated based on the parameter adjustment data of the digital twin model to obtain the deviation data between the model prediction and the actual observation. Based on the deviation data between the model prediction and the actual observation, the model residual distribution data is analyzed by adjusting the parameters of the digital twin model to obtain the model residual distribution data. The model prediction local variance is calculated based on the model residual distribution data to obtain the model prediction local variance data; Based on the deviation data between the model prediction and the actual observation, the model residual distribution data, and the model prediction local variance data, the model prediction accuracy is optimized and adjusted by adjusting the parameter adjustment data of the digital twin model, thereby obtaining optimized model prediction accuracy data.

[0037] Specifically, the parameter adjustment data of the digital twin model is used to optimize the robot fault prediction accuracy, thereby obtaining optimized model prediction accuracy data. This includes: firstly, calculating the deviation between the model prediction output and actual observations based on the adjusted parameters. This is done by comparing the model's predicted output after parameter adjustment with indirect observation data (motor current and sound signals) collected during actual robot operation point by point. The absolute value error accumulation method is used to calculate the prediction deviation value at each time point, and the deviation sequence is smoothed using a sliding window technique to eliminate random fluctuations, thus obtaining model prediction deviation data reflecting the model's prediction accuracy. Next, based on this model prediction deviation data, model residual distribution analysis is performed on the adjusted parameters. By constructing a residual statistical model, the deviation data is segmented according to the time series, and the mean, standard deviation, and skewness of the residuals for each segment are calculated. The kernel density estimation method is used to fit the probability distribution curve of the residuals, and hypothesis testing is used to identify whether the distribution deviates from normality, thereby obtaining data characterizing the distribution of prediction errors. The model residual distribution data is used as the basis for calculation. Then, the model prediction local variance is calculated based on the residual distribution data. This is done by dividing the residual distribution data into multiple local intervals, calculating the sum of squared residual values ​​in each interval and dividing by the interval length to obtain the local variance. Simultaneously, an exponentially weighted moving average method is used to smooth the variance between adjacent intervals to capture the trend of variance changes, thus obtaining model prediction local variance data that reflects the spatial distribution of prediction uncertainty. Finally, based on the deviation data between model prediction and actual observation, the model residual distribution data, and the model prediction local variance data, the model prediction accuracy is optimized by adjusting the parameters of the digital twin model. A multi-objective optimization function is established, using the deviation data as the accuracy loss term, the residual distribution data as the stability constraint term, and the local variance data as the uncertainty penalty term. The gradient descent algorithm is used to iteratively adjust the model parameters to minimize the optimization function value. An early stopping mechanism is combined to prevent overfitting, and cross-validation is used to evaluate the generalization ability of the adjusted model on unseen data, thereby obtaining optimized model prediction accuracy data that improves fault prediction accuracy and robustness.

[0038] The decision-making module is used to make predictive maintenance decisions and plans for the robot based on the robot health assessment and fault prediction data, and generate intelligent operation and maintenance decision data for the robot.

[0039] For example, predictive maintenance decision planning for the robot is performed based on the robot health assessment and fault prediction data to generate intelligent operation and maintenance decision data for the robot, specifically including: The robot health assessment and fault prediction data are used to calculate a health score based on the robot's operating status, thereby generating health score data. Based on the health score data, the robot's failure risk level is classified to obtain failure risk level data; Based on the fault risk level data, predictive maintenance action sequence planning for the robot is performed to generate maintenance action sequence data; Robot operation and maintenance decision-making is executed based on the maintenance action sequence data, thereby obtaining intelligent operation and maintenance decision data for the robot.

[0040] Specifically, based on the robot health assessment and fault prediction data, predictive maintenance decision planning for the robot is performed to generate intelligent operation and maintenance decision data. This includes: First, calculating a health score based on the robot's operating status using the robot health assessment and fault prediction data. A multi-dimensional health assessment index system is constructed, and the health indices of each component in the health assessment data are weighted and integrated with the fault probability in the fault prediction data. A linear weighted method is used to calculate the comprehensive health score, where the weights are determined by expert scoring based on the criticality of each component and the severity of the fault consequences. The score is also adjusted for timeliness by considering equipment runtime and maintenance history, thus generating health score data that comprehensively reflects the equipment's health status. Next, based on the health score data, robot fault risk levels are classified. A risk level matrix is ​​established, dividing the health score into five levels: excellent, good, caution, warning, and danger. A dynamic threshold adjustment method is used to set the score range for each level based on equipment type and usage environment. The risk level of equipment nearing fault is increased by combining the time dimension information in the fault prediction data. Considering the cumulative risk effect when multiple components malfunction simultaneously, fault risk level data that accurately characterizes the equipment's risk status is obtained. Then, based on this fault risk level data, predictive maintenance action sequence planning for robots is performed. This involves constructing a maintenance knowledge base, mapping different risk levels to corresponding sets of maintenance measures, using a topological sorting algorithm to determine the order of maintenance actions, identifying essential maintenance items that must be prioritized based on equipment structural dependencies, and considering maintenance resource constraints and downtime limitations. Integer programming methods are used to optimize the timing and resource allocation of maintenance actions, thereby generating scientifically sound maintenance action sequence data. Finally, robot operation and maintenance decision execution is performed based on this maintenance action sequence data. This involves converting the maintenance action sequence into an executable instruction set, using a workflow engine to drive the automatic dispatch and execution of maintenance tasks, monitoring maintenance progress in real time and dynamically adjusting the task sequence according to actual conditions, integrating the equipment control system to automate some maintenance operations, and verifying maintenance effectiveness and updating equipment health status through a data acquisition system. This provides intelligent robot operation and maintenance decision data to guide actual operation and maintenance work.

[0041] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the concept of the invention, they should all fall within the protection scope of the present invention.

[0042] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0043] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A robot-based intelligent operation and maintenance system based on digital twins, characterized in that, include: The acquisition module is used to acquire indirect observation data of the robot and initial parameters of the digital twin model, and to perform dynamic change analysis of the robot's operating environment and load based on the indirect observation data, thereby obtaining dynamic change data of the environment. The analysis module is used to perform hidden state estimation uncertainty analysis on the dynamic environmental change data of the robot based on the initial parameters of the digital twin model, to obtain hidden state estimation uncertainty data, and to perform robot health state deviation correlation mining based on the hidden state estimation uncertainty data to obtain health state deviation correlation data. The adjustment module is used to adaptively adjust the parameters of the digital twin model based on the health status deviation correlation data, generate digital twin model parameter adjustment data, and use the digital twin model parameter adjustment data to perform robot health status assessment and fault prediction, thereby obtaining robot health assessment and fault prediction data. The decision-making module is used to make predictive maintenance decisions and plans for the robot based on the robot health assessment and fault prediction data, and generate intelligent operation and maintenance decision data for the robot.

2. The robot intelligent operation and maintenance system based on digital twins according to claim 1, characterized in that, The indirect observation data includes motor current signals and sound signals, and the initial parameters of the digital twin model include the model network structure, initial weights, hidden state estimation model, model learning rate, and number of network layers.

3. The robot intelligent operation and maintenance system based on digital twins according to claim 1, characterized in that, The process of obtaining dynamic environmental change data specifically includes: Obtain indirect observation data and initial parameters of the digital twin model corresponding to the robot; The indirect observation data is preprocessed to generate preprocessed indirect observation data; The preprocessed indirect observation data is used to extract the robot's operating environment load change data. Based on the initial parameters of the digital twin model, the dynamic changes of the robot's operating environment are analyzed using the load change data of the operating environment to obtain dynamic environmental change data.

4. The robot intelligent operation and maintenance system based on digital twin according to claim 2, characterized in that, Based on the hidden state estimation uncertainty data, robot health state deviation correlation mining is performed to obtain health state deviation correlation data, specifically including: Extract the hidden state estimation model corresponding to the initial parameters of the digital twin model; Based on the initial parameters of the digital twin model and the hidden state estimation model, an uncertainty analysis of the hidden state estimation for the robot transmission system is performed on the dynamic environmental change data to obtain the hidden state estimation uncertainty data. The hidden state estimation uncertainty data is fitted with robot health state deviation tolerance data to generate health state deviation tolerance data. Based on the health status deviation tolerance data and the hidden state estimation uncertainty data, the robot health status deviation correlation mining is performed to obtain health status deviation correlation data.

5. The robot intelligent operation and maintenance system based on digital twins according to claim 4, characterized in that, The process of obtaining the uncertainty data for the hidden state estimation specifically includes: Extract the model learning rate and network layer number from the initial parameters of the digital twin model, and calculate the correlation between robot observation data and the hidden state of mechanical parts based on the hidden state estimation model to obtain the correlation coefficient between observation and state; Based on the environmental dynamic change data, the correlation coefficient between the observation and the state is analyzed to determine the impact of robot dynamic load change, and dynamic load change impact data is obtained. Based on the model learning rate and the number of network layers, the hidden state estimation simulation of robot parts under multiple time steps is performed on the dynamic load change impact data to obtain a multi-time step hidden state estimation sequence. The robot state estimation error propagation behavior imbalance analysis is performed on the multi-time-step hidden state estimation sequence to obtain estimation error propagation behavior imbalance data; Perform robot operation and maintenance uncertainty quantification regression analysis on the imbalance data of the estimated error propagation behavior to obtain uncertainty quantification regression data; Based on the imbalance data of the propagation behavior of the estimation error and the uncertainty quantification regression data, the hidden state estimation uncertainty data is obtained by performing hidden state estimation uncertainty analysis on the environmental dynamic change data.

6. The robot intelligent operation and maintenance system based on digital twin according to claim 5, characterized in that, Perform robot state estimation error propagation behavior imbalance analysis on the multi-time-step hidden state estimation sequence to obtain estimation error propagation behavior imbalance data, specifically including: The robot state estimation error density non-uniformity is identified in the multi-time-step hidden state estimation sequence to obtain the estimation error density non-uniformity vector; Based on the non-uniformity vector of the estimation error density, the time-series variation differential coupling of the robot's hidden state estimation deviation is performed to generate estimation deviation time-series variation differential coupling data; Based on the non-uniformity vector of the estimated error density and the time-series variation differential coupling data of the estimated deviation, a local calculation of the prediction variance of the robot digital twin model is performed to obtain the local data of the model prediction variance. Based on the non-uniformity vector of the estimated error density, the differential coupling data of the temporal variation of the estimated deviation, and the local data of the model prediction variance, the error propagation ratio difference between the hidden states of each robot component is deduced, and the error propagation ratio difference between different hidden states is generated. Based on the difference in the error propagation ratio between the different hidden states, an imbalance analysis of the robot's estimation error propagation behavior is performed to obtain the estimation error propagation behavior imbalance data.

7. The robot intelligent operation and maintenance system based on digital twin according to claim 3, characterized in that, The hidden state estimation uncertainty data is fitted with robot health state deviation tolerance data to generate health state deviation tolerance data, specifically including: The hidden state estimation uncertainty data is subjected to robot operation and maintenance uncertainty network partitioning processing to obtain uncertainty network partitioning data; The geometric feature analysis of robot health state deviation is performed on the uncertain network partitioning data to obtain the geometric feature data of health state deviation. Based on the aforementioned health state deviation geometric feature data, the force offset distribution of key mechanical components of the robot is simulated to generate component force offset distribution data. The stress offset distribution data of the component is subjected to strain rate component decomposition of the robot structure stress distribution to obtain strain rate component decomposition data. Based on the strain rate component decomposition data, the robot health state deviation tolerance is fitted to generate health state deviation tolerance data.

8. The robot intelligent operation and maintenance system based on digital twin according to claim 1, characterized in that, The robot's health status is assessed and fault prediction is predicted by adjusting the parameters of the digital twin model, resulting in robot health assessment and fault prediction data, specifically including: Robot operation and maintenance feature learning is performed on the health status deviation correlation data to obtain health status deviation correlation feature data; Based on the health status deviation correlation feature data, the initial parameters of the digital twin model are adjusted for robot health prediction, generating digital twin model parameter adjustment data. The parameter adjustment data of the digital twin model are used to optimize the robot fault prediction accuracy, thereby obtaining optimized model prediction accuracy data; The robot's health status is assessed and faults are predicted by optimizing the model's prediction accuracy data, thus obtaining robot health assessment and fault prediction data.

9. The robot intelligent operation and maintenance system based on digital twins according to claim 8, characterized in that, Based on the health status deviation correlation feature data, the initial parameters of the digital twin model are adjusted for robot health prediction, generating digital twin model parameter adjustment data, specifically including: Based on the health status deviation correlation feature data, the initial parameters of the digital twin model are matched with the robot operation and maintenance model learning rate to obtain model learning rate matching data; Based on the model learning rate matching data and the health status deviation correlation feature data, the number of network layers in the robot digital twin model is adjusted and controlled to obtain network layer control data; The robot's hidden state estimation convergence speed is matched by performing a process on the network layer control data to obtain hidden state estimation convergence speed matching data. Based on the model learning rate matching data, the network layer number control data, and the hidden state estimation convergence speed matching data, the initial parameters of the digital twin model are adjusted to generate digital twin model parameter adjustment data.

10. The robot intelligent operation and maintenance system based on digital twin according to claim 8, characterized in that, The parameter adjustment data of the digital twin model are used to optimize the robot fault prediction accuracy, thereby obtaining optimized model prediction accuracy data, specifically including: The deviation between the model prediction output and the actual observation is calculated based on the parameter adjustment data of the digital twin model to obtain the deviation data between the model prediction and the actual observation. Based on the deviation data between the model prediction and the actual observation, the model residual distribution data is analyzed by adjusting the parameters of the digital twin model to obtain the model residual distribution data. The model prediction local variance is calculated based on the model residual distribution data to obtain the model prediction local variance data; Based on the deviation data between the model prediction and the actual observation, the model residual distribution data, and the model prediction local variance data, the model prediction accuracy is optimized and adjusted by adjusting the parameter adjustment data of the digital twin model, thereby obtaining optimized model prediction accuracy data.

11. The robot intelligent operation and maintenance system based on digital twin according to claim 1, characterized in that, Based on the robot health assessment and fault prediction data, predictive maintenance decision-making and planning are performed to generate intelligent operation and maintenance decision data for the robot, specifically including: The robot health assessment and fault prediction data are used to calculate a health score based on the robot's operating status, thereby generating health score data. Based on the health score data, the robot's failure risk level is classified to obtain failure risk level data; Based on the fault risk level data, predictive maintenance action sequence planning for the robot is performed to generate maintenance action sequence data; Robot operation and maintenance decision-making is executed based on the maintenance action sequence data, thereby obtaining intelligent operation and maintenance decision data for the robot.

Images