Industrial robot health status monitoring method and monitoring system based on digital twin

Abstract

The present invention discloses a method and system for monitoring the health status of an industrial robot based on digital twins, which relates to the technical field of equipment monitoring, including: using multiple sensors to collect real-time status data of an industrial robot, matching and aligning the real-time status data with a three-dimensional simulation model to construct a digital twin model; obtaining twin data through a digital twin model after updating the parameters of an adaptive model, fusing the twin data with the real-time status data to generate a multi-source heterogeneous data set; obtaining feature data characterizing the health status of an industrial robot through feature selection to construct health indicators; constructing a health status assessment model based on a deep belief network, using the current health indicator parameters as model input, and obtaining the health status assessment result of the industrial robot. The present invention integrates multi-source heterogeneous data such as vibration data and electrical data, and combines digital twins to monitor and assess the health status of industrial robots, providing a reliable reference for industrial robot status monitoring and maintenance.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment monitoring, and more specifically, to a health status monitoring method and monitoring system for an industrial robot based on digital twins. Background Art

[0002] High-end CNC equipment such as industrial robots are the foundation for my country to achieve transformation and upgrading of the manufacturing industry and realize intelligent manufacturing. For industrial robots, reliability determines their work efficiency and the quality of processed products. As the service time of the equipment increases, its key components will degrade to varying degrees or even fail, affecting work efficiency and product processing quality. Failure of industrial robots will also lead to long-term downtime for maintenance and parts replacement, reducing workshop production efficiency, and even causing personal safety issues in serious cases. Therefore, in order to avoid economic losses and safety issues caused by equipment failure, it is necessary to obtain the status data of the equipment during operation through condition monitoring technology, and evaluate the health status of the equipment and predict the development trend of the failure by analyzing the data. Before the failure occurs, timely formulate corresponding repair and maintenance strategies, which can not only save costs and avoid the occurrence of failures, but also extend the service life of the equipment.

[0003] At present, digital twins have become a hot topic of research as a means to achieve intelligent manufacturing. Digital twin technology combines physical entities with their high-fidelity virtual models through perception data to carry out operations, monitoring, evaluation, prediction and other tasks, and realize the full life cycle management of CNC equipment. Digital twin technology is not only the key to realizing intelligent manufacturing, but also can carry out related applications such as equipment status monitoring, health assessment, fault warning, etc. on this basis, so as to realize the supervision of the equipment service process. While ensuring the stability of the equipment service process, it can formulate and carry out maintenance in a timely manner, thereby improving the reliability of equipment use and reducing unplanned downtime losses. Therefore, how to comprehensively characterize the health status of industrial robots and improve the monitoring accuracy is an urgent problem to be solved. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes an industrial robot health status monitoring method and monitoring system based on digital twins. The purpose is to conduct a comprehensive and accurate health status assessment of the industrial robot through multi-source heterogeneous data, solve the problems of single data source and incomplete assessment, and make the monitoring results more accurate and reliable.

[0005] A first aspect of the present invention provides a method for monitoring the health status of an industrial robot based on digital twins, comprising the following steps:

[0006] A three-dimensional simulation model is constructed according to the physical entity of the industrial robot, real-time status data of the industrial robot is collected using multiple sensors, and the real-time status data is matched and aligned with the three-dimensional simulation model to construct a digital twin model of the industrial robot;

[0007] Adaptively updating model parameters of the digital twin model, acquiring twin data through the digital twin model after the adaptive model parameter update, fusing the twin data with real-time status data, and generating a multi-source heterogeneous data set;

[0008] Extracting features from the multi-source heterogeneous data set to obtain feature data that can characterize the health status of the industrial robot;

[0009] The characteristic data are used to construct the health index of the industrial robot, and the health status levels are divided. A health status assessment model is constructed based on a deep belief network, and the current health index parameters are obtained as model input to obtain the health status assessment result of the industrial robot.

[0010] In this solution, a digital twin model of an industrial robot is constructed, specifically:

[0011] Acquire the physical entities of the components of the industrial robot to construct a physical space, draw a three-dimensional model of the physical entities of the components, assemble and combine the three-dimensional models to establish a three-dimensional simulation model of the industrial robot, and obtain the real-time status data of the industrial robot in multiple dimensions through multiple preset sensors;

[0012] Use the communication interface to transmit the real-time status data of the physical entity of the industrial robot, convert and encapsulate the acquired real-time status data, and perform data structured preprocessing;

[0013] Acquire historical fault data of the industrial robot, preprocess the historical fault data, and cluster the preprocessed historical fault data using a clustering algorithm to obtain clustering clusters based on the fault type, and generate key monitoring features according to the fault components and fault parameters corresponding to each clustering cluster;

[0014] Training the fault virtual models corresponding to the components of the three-dimensional simulation model through clustering clusters corresponding to key monitoring features, using the fault virtual models to learn the fault mechanism of the industrial robot, and using the three-dimensional simulation model and the fault virtual models to construct a twin space;

[0015] The preprocessed real-time status data is used to connect the physical space with the twin space, so as to achieve matching and alignment between the real-time status data and the three-dimensional simulation model, and the model fusion is carried out through data-driven to realize the digital mapping of the industrial robot and build a digital twin model of the industrial robot.

[0016] In this solution, the fault virtual models corresponding to the components of the three-dimensional simulation model are trained by clustering the key monitoring features, specifically:

[0017] The historical fault data in the clusters corresponding to the key monitoring features are used as training samples, the training samples are imported into the AlexNet network, features are extracted through convolution kernels of different sizes, and activation calculations are performed using activation functions;

[0018] The pooling layer is used to reduce the feature dimension by downsampling, and the reduced feature is imported into the fully connected layer to obtain the probability distribution of multi-class sample labels using the softmax function;

[0019] The model parameters of the AlexNet network are updated through training, and the weight parameters are optimized using Adam with adaptive learning rate. After verifying the classification performance of the AlexNet network, the network parameters are retained and the fault virtual model is output.

[0020] In this solution, the digital twin model is adaptively updated in terms of model parameters, specifically:

[0021] In the constructed digital twin model of the industrial robot, data of key monitoring features are updated, parameter variables corresponding to each key monitoring feature are obtained, a multivariable set is generated, and sparrow algorithm parameters are initialized based on the multivariable set;

[0022] Initialize the sparrow population and the maximum number of iterations, construct an objective function based on the error between the measured value of the parameter variable and the output twin data value of the digital twin model, map the objective function to a fitness function, and calculate the fitness of each sparrow as the current optimal value;

[0023] According to the fitness, sparrows are selected as discoverers based on the preset ratio, and the remaining sparrows are selected as joiners to update their positions. Sentinels are randomly selected based on the preset ratio of vigilants to update their positions. After the update, the fitness of each sparrow is recalculated.

[0024] The chaotic algorithm is introduced to process the fitness of the current sparrow in a chaotic way, and the optimal position moving direction is determined by using the differential idea to guide the sparrow to update its position. The optimal solution of the parameter variable is obtained by iterative position update until the maximum number of iterations is reached.

[0025] The digital twin model is configured according to the optimal values ​​of the parameter variables corresponding to each key monitoring feature to reflect the degradation degree of each component of the industrial robot.

[0026] In this solution, feature extraction is performed in the multi-source heterogeneous data set to obtain feature data that can characterize the health status of the industrial robot, specifically:

[0027] Calling the AlexNet network to perform convolution, pooling and full connection operations on the multi-source heterogeneous data set, obtaining a one-dimensional feature vector sequence of the full connection output, obtaining the Pearson correlation coefficient between each feature vector in the one-dimensional feature vector sequence and the health status of the industrial robot, and selecting a preset number of feature vectors according to the degree of correlation;

[0028] The number of nearest neighbors k is set in the selected feature vector set, the Pearson correlation coefficient between the feature vectors is calculated, the first k feature vectors with the smallest values ​​are selected according to the number of nearest neighbors, and the feature data set corresponding to each feature vector is generated;

[0029] An intersection operation is performed on the feature data sets corresponding to the feature vectors to obtain feature data that can characterize the health status of the industrial robot.

[0030] In this solution, the characteristic data is used to construct the health index of the industrial robot, and the health status level is divided, and a health status assessment model is constructed based on a deep belief network, specifically:

[0031] Constructing the health index of the industrial robot based on the acquired feature data, obtaining the historical operation and maintenance instances of the industrial robot, performing clustering based on the historical operation and maintenance instances, generating a data set of historical operation and maintenance instances in different health states, and dividing the health state levels according to their average health states;

[0032] A health status assessment model is constructed based on a deep belief network, Dropout is introduced into the health status assessment model to prevent overfitting, and a feature data set of health indicators in a normal health state is obtained based on historical operation and maintenance examples to generate training data;

[0033] Initialize the deep belief network, use the particle swarm algorithm to optimize the number of restricted Boltzmann machine layers, and stack the restricted Boltzmann machine layers according to the optimal solution of the number of restricted Boltzmann machine layers;

[0034] The training data is trained layer by layer from the bottom layer to the top layer, and a deep architecture with deep nonlinear mapping capability is generated by iteratively updating the weights. The extracted features are transmitted from the top layer to the bottom layer to obtain the reconstructed data corresponding to the input training data.

[0035] The health status assessment model is trained according to the principle of minimizing the error between the reconstructed data and the input training data, and the average error between the reconstructed data and the training data is obtained to set the health status threshold.

[0036] In this scheme, the health status assessment model is used to obtain the health status assessment results of the industrial robot, specifically:

[0037] Based on the health index, health index parameters of the current industrial robot are collected, the health index parameters are imported into the health status assessment model as model input, and data of the health index parameters are reconstructed;

[0038] The error between the reconstructed data and the health indicator parameter is obtained, and the error is compared with the health status threshold. The health status assessment result of the industrial robot is generated through the comparison result, and when the health status assessment result is an abnormal health status, an abnormal health status alarm is generated.

[0039] A second aspect of the present invention provides an industrial robot health status monitoring system based on digital twins, the system comprising: a data acquisition and preprocessing module, a digital twin module, a multi-source data fusion module, a health status assessment module, and a health status output and visualization module;

[0040] The data acquisition and preprocessing module uses multiple sensors to collect real-time status data of the industrial robot and preprocesses the collected real-time status data;

[0041] The digital twin module constructs a three-dimensional simulation model according to the physical entity of the industrial robot, matches the three-dimensional simulation model with the collected real-time status data and performs data alignment to construct a digital twin model of the industrial robot, and maintains consistency with the degradation state of the physical entity by adaptively updating the model parameters of the digital twin model;

[0042] The multi-source data fusion module fuses the twin data output by the digital twin model with the real-time status data to generate a multi-source heterogeneous data set;

[0043] The health status assessment module uses feature selection to obtain feature data that can characterize the health status of the industrial robot in the multi-source heterogeneous data set; uses the feature data to construct the health index of the industrial robot, divides the health status level, establishes a health status assessment model, and uses the trained health status assessment model to obtain the health status assessment result of the industrial robot;

[0044] The health status output and visualization module performs three-dimensional visualization on the health status evaluation result of the industrial robot output by the health status evaluation module, and issues an abnormal alarm according to the health status evaluation result.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] The present invention integrates multi-source heterogeneous data such as vibration data and electrical data, mines the characteristic relationship between the data, obtains aging parameters that can reflect the health status of the industrial robot, and combines digital twins to monitor and evaluate the health status of the industrial robot, providing a reliable reference basis for industrial robot status monitoring and maintenance.

[0047] The present invention proposes a model adaptive update method based on the improved sparrow algorithm to finely reflect the degradation state of the industrial robot. In addition, a comprehensive and accurate health status assessment of the industrial robot is performed through multi-source heterogeneous data, solving the problem of single data source and incomplete assessment, making the monitoring results more accurate and reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments or exemplary embodiments of the present invention, the drawings required for use in the embodiments or exemplary descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained according to the drawings without paying creative work.

[0049] Figure 1 A flow chart of a method for monitoring the health status of an industrial robot based on digital twins of the present invention is shown;

[0050] Figure 2 A flow chart of an embodiment for adaptively updating model parameters of a digital twin model is shown;

[0051] Figure 3 A flowchart of establishing a health status assessment model according to an embodiment is shown;

[0052] Figure 4 A block diagram of the industrial robot health status monitoring system based on digital twins of the present invention is shown. DETAILED DESCRIPTION

[0053] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0054] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0055] Figure 1 A flow chart of the industrial robot health status monitoring method based on digital twins of the present invention is shown.

[0056] like Figure 1 As shown, in a first embodiment of the present invention, a method for monitoring the health status of an industrial robot based on digital twins is provided, comprising:

[0057] S102, constructing a three-dimensional simulation model according to the physical entity of the industrial robot, using multiple sensors to collect real-time status data of the industrial robot, and aligning the real-time status data with the three-dimensional simulation model to construct a digital twin model of the industrial robot;

[0058] S104, adaptively updating model parameters of the digital twin model, acquiring twin data through the digital twin model after the adaptive model parameters are updated, fusing the twin data with the real-time status data, and generating a multi-source heterogeneous data set;

[0059] S106, extracting features from the multi-source heterogeneous data set to obtain feature data that can characterize the health status of the industrial robot;

[0060] S108, using the characteristic data to construct the health index of the industrial robot, and divide the health status level, constructing a health status assessment model based on a deep belief network, obtaining the current health index parameters as model input, and obtaining the health status assessment result of the industrial robot.

[0061] It should be noted that digital twins refer to the simulation of physical entity systems within a digital platform. With the help of digital twins, the status of physical entity systems can be monitored and even controlled, thereby helping organizations monitor operations, perform predictive maintenance, and improve processes. The physical entities of the components of industrial robots are obtained to construct a physical space, which also includes the peripheral equipment of the industrial robot. The three-dimensional model of the physical entity of the component is drawn using three-dimensional industrial software (NX 12.0), and the three-dimensional model is assembled and combined to establish a three-dimensional simulation model of the industrial robot. In order to accelerate the rapid transmission of data and meet the real-time requirements of digital twins, the simplification process in terms of geometry, information, logic, etc. is carried out in the three-dimensional simulation model to achieve lightweight processing.

[0062] The real-time status data of industrial robots is acquired in multiple dimensions through multiple preset sensors. The intelligent level of fault operation and maintenance and production in the entire life cycle is improved based on the acquired multi-dimensional data. The operating status of industrial robots is monitored, and possible faults that may occur when industrial robots are working can be preliminarily checked. The collection of high-fidelity physical data and fast and high-quality transmission are the prerequisites for building a digital twin system. The real-time status data of the physical entity of the industrial robot, such as motion status, vibration of various parts, electrical data, driver data, temperature data, operating environment data, etc., are transmitted using a communication interface. The acquired real-time status data is converted and packaged, and data structured preprocessing is performed, including eliminating abnormal data, filling in missing data, and intercepting high-correlation feature segments of multi-dimensional data, removing low-frequency trend items from vibration data, and other operations.

[0063] The historical fault data of the industrial robot is obtained, the historical fault data is preprocessed, and the preprocessed historical fault data is clustered using a clustering algorithm to obtain clustering clusters based on the fault type, and the fault components and fault parameters are extracted according to the historical fault data samples in each clustering cluster, and the fault construction and fault parameters are integrated to generate the key monitoring features of the industrial robot, and the key monitoring features characterize the high-frequency fault parts and parameters that the industrial robot needs to focus on monitoring; the fault virtual model corresponding to the clustering cluster corresponding to the key monitoring features is trained and the fault virtual model corresponding to each component part of the three-dimensional simulation model is used to learn the fault mechanism of the industrial robot, and the three-dimensional simulation model and the fault virtual model are used to construct a twin space; the preprocessed real-time state data is used to connect the physical space with the twin space, and the real-time state data is matched and aligned with the three-dimensional simulation model, and the model is fused through data drive to realize the digital number mapping of the industrial robot, and the digital twin model of the industrial robot is constructed. The obtained digital twin model can be used to make a preliminary prediction of the occurrence of various faults of the industrial robot, obtain data from different angles, and obtain diagnostic results according to the characteristics of different faults through intelligent diagnosis methods.

[0064] It should be noted that the historical fault data in the clusters corresponding to the key monitoring features are used as training samples, and the proportion of various fault types is preferably considered. The classification weights of the AlexNet network for different faults are trained according to the proportion of the fault types; the training samples are imported into the AlexNet network, which contains 5 convolutional layers (where the 1st, 2nd, and 5th convolutional layers each contain 1 maximum pooling layer) and 3 fully connected layers, and features are extracted through convolution kernels of different sizes, and activation calculations are performed using activation functions; the pooling layer is used to reduce the feature dimension by downsampling, and the reduced feature is imported into the fully connected layer, and the softmax function is used in the last fully connected layer to obtain the probability distribution of multi-class sample labels; the model parameters of the AlexNet network are updated through training, and the weight parameters are optimized using the learning rate adaptive Adam. After verifying the classification performance of the AlexNet network, the network parameters are retained and the fault virtual model is output. The use of the AlexNet network to construct a fault virtual model improves the problems of gradient vanishing and convergence fluctuations. At the same time, the introduction of the DropOut technology prevents overfitting and improves the accuracy of the digital twin model for preliminary fault diagnosis.

[0065] Figure 3 A flowchart of establishing a health status assessment model according to an embodiment is shown.

[0066] According to an embodiment of the present invention, the digital twin model is adaptively updated in terms of model parameters, specifically:

[0067] S202, in the constructed digital twin model of the industrial robot, updating the data of the key monitoring features, obtaining the parameter variables corresponding to each key monitoring feature, generating a multivariable set, and initializing the sparrow algorithm parameters based on the multivariable set;

[0068] S204, initializing the sparrow population and the maximum number of iterations, constructing an objective function according to the error between the measured value of the parameter variable and the twin data value output by the digital twin model, mapping the objective function to a fitness function, and calculating the fitness of each sparrow as the current optimal value;

[0069] S206, selecting sparrows as discoverers based on a preset ratio according to the fitness, and the remaining sparrows as joiners, and updating the positions, and randomly selecting alerters based on a preset alerter ratio, and updating the positions, and recalculating the fitness of each sparrow after the update;

[0070] S208, introducing a chaotic algorithm to perform chaotic processing on the fitness of the current sparrow, and using the differential idea to determine the optimal position moving direction, guiding the sparrow to update the position, and iterating the position update until the maximum number of iterations is reached to obtain the optimal solution of the parameter variable;

[0071] S210, configuring the digital twin model according to the optimal values ​​of the parameter variables corresponding to each key monitoring feature to reflect the degree of degradation of each component of the industrial robot.

[0072] It should be noted that, due to the wear, loss, aging of components and other reasons, the physical entity of the industrial robot will gradually degrade and its working performance will be reduced during operation. If the digital twin model is to reflect the degradation state of the components, it is necessary to adaptively update the model parameters to keep them consistent with the degradation state of the physical entity. The improved sparrow algorithm is combined with the digital twin model, and the data of the digital twin model is updated through the parameter variables corresponding to each key monitoring feature provided by the physical model. The parameter variables corresponding to the key monitoring features can characterize the degradation state of the components. For example, the parameter variables such as the angular speed, torque and force conditions are obtained in the mechanical arm and reducer, and the parameter variables such as the input torque, speed and current of the servo motor are obtained in the drive system. Obtaining the degradation state of the components of the industrial robot and performing accurate health monitoring avoids repeated experiments to find the optimal value, thereby improving the calculation efficiency and accuracy of the digital twin model. The sparrow algorithm is optimized by using the chaos algorithm, and the Logistic mapping equation is used to construct chaotic variables to disturb the sparrow population, so that it jumps out of the local convergence area. In addition, the competition between sparrows is used with the help of the difference idea to make the new individuals more inclined to the optimal individuals. The three sparrows with the highest fitness are obtained respectively, and the residuals between the second highest fitness sparrow and the highest fitness sparrow and the third highest fitness sparrow and the highest fitness sparrow are calculated respectively. The two residuals are weighted by random weighting, and the two weighted residuals are added to obtain new sparrow individuals. The communication between sparrow individuals is strengthened by difference, and the sparrow individuals at the current position are guided to move to the global optimal position during the iteration process, so as to reduce the model construction time and improve the prediction accuracy.

[0073] It should be noted that the AlexNet network is called to perform convolution, pooling and full connection operations on the multi-source heterogeneous data set, and a one-dimensional feature vector sequence output by the first two fully connected layers is obtained. The Pearson correlation coefficient between each feature vector in the one-dimensional feature vector sequence and the health status of the industrial robot is obtained, and a preset number of feature vectors are selected according to the degree of correlation; the k-nearest neighbor algorithm is introduced in the feature selection, and the number of neighbors k is set in the selected feature vector set, and the Pearson correlation coefficient between the feature vectors is calculated. The first k feature vectors with the smallest value are selected according to the number of neighbors, and the feature data set corresponding to each feature vector is generated. That is, the Pearson correlation coefficient between the feature vectors is calculated to characterize the redundancy between the feature vectors. When the redundancy is too high, the model prediction result is more susceptible to noise and prone to overfitting. Therefore, the k-nearest neighbor algorithm is used to select feature vectors with lower redundancy to improve the subsequent model prediction accuracy. The feature data set corresponding to the feature vector is subjected to an intersection operation to obtain feature data that can characterize the health status of the industrial robot. The preferred feature data include: time series characteristics (root mean square, standard deviation, peak factor) of the vibration signal, average frequency, energy characteristics of the third-layer wavelet packet decomposition signal, approximate entropy characteristics, root mean square characteristics of the driver torque signal and current signal, etc.

[0074] Figure 3 A flowchart of establishing a health status assessment model according to an embodiment is shown.

[0075] According to an embodiment of the present invention, the characteristic data is used to construct the health index of the industrial robot, and the health status level is divided, and a health status assessment model is constructed based on a deep belief network, specifically:

[0076] S302, constructing a health index of the industrial robot according to the acquired feature data, acquiring historical operation and maintenance instances of the industrial robot, performing clustering based on the historical operation and maintenance instances, generating a data set of historical operation and maintenance instances in different health states, and dividing the health state levels according to their average health states;

[0077] S304, constructing a health status assessment model based on a deep belief network, introducing Dropout into the health status assessment model to prevent overfitting, obtaining a feature data set of health indicators in a normal health state based on historical operation and maintenance instances, and generating training data;

[0078] S306, initializing the deep belief network, using the particle swarm algorithm to optimize the number of restricted Boltzmann machine layers, and stacking the restricted Boltzmann machine layers according to the optimal solution of the number of restricted Boltzmann machine layers;

[0079] S308, training the training data layer by layer from the bottom layer to the top layer, generating a deep architecture with deep nonlinear mapping capability by iteratively updating the weights, transmitting the extracted features from the top layer to the bottom layer, and obtaining reconstructed data corresponding to the input training data;

[0080] S310, training a health status assessment model according to the principle of minimizing the error between the reconstructed data and the input training data, and obtaining an average error between the reconstructed data and the training data to set a health status threshold.

[0081] It should be noted that the health status assessment model of industrial robots is constructed based on a deep belief network based on a restricted Boltzmann machine. Each restricted Boltzmann machine can map high-dimensional and complex input data to a low-dimensional feature space by learning the probability distribution of training data, and extract the potential structure of the training data. The feature data set of health indicators in a normal healthy state is used to generate training data for model training, which reduces the computational complexity of the model. The Dropout technology is introduced into the health status assessment model to randomly select neurons and their connections from the restricted Boltzmann machine to prevent overfitting. The loss function is constructed by the mean square error between the input training data and the reconstructed data. The health status assessment model is trained according to the principle of minimum error between the reconstructed data and the input training data, and the weight of the model is adjusted using back propagation. The model generates a deep architecture with deep nonlinear mapping capabilities, and uses the idea of ​​Euclidean distance to measure the distance between the monitoring data and the estimated data in the normal healthy state to obtain the error. When the industrial robot is in an abnormal health state, its error will become larger, that is, the monitoring data gradually deviates from the normal health state and enters an abnormal health state.

[0082] It should be noted that based on the health index, the health index parameters of the current industrial robot are collected, the health index parameters are imported into the health status assessment model as the model input, and the health index parameters are reconstructed; the error between the reconstructed data and the health index parameters is obtained, and the error is compared with the health status threshold, and the health status assessment result of the industrial robot is generated by the comparison result. When the error is less than the health status threshold, it is considered that the industrial robot is currently in a normal health state, and the deviation between the health status threshold and the error is obtained. The smaller the deviation is, the closer it is to the abnormal health state. According to the deviation, an abnormal warning threshold is constructed. When the deviation is less than the abnormal warning threshold, a health status warning of the industrial robot is generated to remind the operation and maintenance personnel to perform operation and maintenance detection of the industrial robot; preferably, the deviation change rate is obtained, and the change rate threshold is set according to the average historical deviation change rate of different stages, and adaptive weights are set for different stages. The deviation change rate is weighted using the adaptive weight. When the weighted deviation change rate is greater than the change rate threshold, a health status warning of the industrial robot is generated.

[0083] When the error is greater than or equal to the health status threshold, the industrial robot is deemed to be in an abnormal health state. The deviation between the error and the health status threshold is obtained. The larger the deviation, the greater the degree of the abnormal health state. Different abnormal health status levels are set according to the deviation. Preferably, they are divided into "mild abnormal health state", "moderate abnormal health state" and "severe abnormal health state". When the health status assessment result is an abnormal health state, different levels of abnormal health status alarms are generated according to different abnormal health status levels.

[0084] According to an embodiment of the present invention, a product production line used by an industrial robot is obtained, operating characteristics of industrial robots at different positions in the product production line are determined, inspection data of products produced by the production line are collected, defect data of the products are obtained based on the inspection data, production traceability is performed using the defect data, operating characteristics that may cause defects are determined through the production traceability results, industrial robots at corresponding positions are determined using the operating characteristics, and the determined industrial robots are marked; in the marked industrial robots, defect characteristics of the defect data are extracted, the defect characteristics are used to search through historical defect data, production parameter labels involved in the defect data with the highest similarity are obtained, standard production parameters in the product production line are extracted based on the production parameter labels, a dynamic time warping algorithm is used to obtain the degree of deviation between actual production parameters and standard production parameters, and when the degree of deviation is greater than a preset threshold, the production parameters of the marked industrial robot are adjusted and updated.

[0085] Figure 4 A block diagram of the industrial robot health status monitoring system based on digital twins of the present invention is shown.

[0086] The second embodiment of the present invention provides an industrial robot health status monitoring system 4 based on digital twins, which includes: a data acquisition and preprocessing module 401, a digital twin module 402, a multi-source data fusion module 403, a health status assessment module 404 and a health status output and visualization module 405;

[0087] The data acquisition and preprocessing module uses multiple sensors to collect real-time status data of the industrial robot and preprocesses the collected real-time status data;

[0088] The digital twin module constructs a three-dimensional simulation model according to the physical entity of the industrial robot, matches the three-dimensional simulation model with the collected real-time status data and performs data alignment to construct a digital twin model of the industrial robot, and maintains consistency with the degradation state of the physical entity by adaptively updating the model parameters of the digital twin model;

[0089] The multi-source data fusion module fuses the twin data output by the digital twin model with the real-time status data to generate a multi-source heterogeneous data set;

[0090] The health status assessment module uses feature selection to obtain feature data that can characterize the health status of the industrial robot in the multi-source heterogeneous data set; uses the feature data to construct the health index of the industrial robot, divides the health status level, establishes a health status assessment model, and uses the trained health status assessment model to obtain the health status assessment result of the industrial robot;

[0091] The health status output and visualization module performs three-dimensional visualization on the health status evaluation result of the industrial robot output by the health status evaluation module, and issues an abnormal alarm according to the health status evaluation result.

[0092] The third embodiment of the present invention provides a computer-readable storage medium, which includes an industrial robot health status monitoring method program based on digital twins. When the industrial robot health status monitoring method program based on digital twins is executed by a processor, the steps of the industrial robot health status monitoring method based on digital twins are implemented.

[0093] In the several embodiments provided in the present application, it should be understood that the disclosed method can be implemented in other ways. The embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as: multiple modules or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0094] In addition, all functional modules in the embodiments of the present invention may be integrated into one processing module, or each module may be a separate module, or two or more modules may be integrated into one module; the above-mentioned integrated modules may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0095] A person of ordinary skill in the art can understand that: all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: a mobile storage device, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0096] Alternatively, if the above-mentioned integrated module of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0097] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for monitoring the health status of an industrial robot based on digital twins, characterized in that: The following steps are involved: A three-dimensional simulation model is constructed according to the physical entity of the industrial robot, real-time status data of the industrial robot is collected using multiple sensors, and the real-time status data is matched and aligned with the three-dimensional simulation model to construct a digital twin model of the industrial robot; Adaptively updating model parameters of the digital twin model, acquiring twin data through the digital twin model after the adaptive model parameter update, fusing the twin data with real-time status data, and generating a multi-source heterogeneous data set; Extracting features from the multi-source heterogeneous data set to obtain feature data that can characterize the health status of the industrial robot; Using the characteristic data to construct the health index of the industrial robot, and dividing the health status level, constructing a health status assessment model based on a deep belief network, obtaining the current health index parameters as model input, and obtaining the health status assessment result of the industrial robot; Build a digital twin model of an industrial robot, specifically: Acquire the physical entities of the components of the industrial robot to construct a physical space, draw a three-dimensional model of the physical entities of the components, assemble and combine the three-dimensional models to establish a three-dimensional simulation model of the industrial robot, and obtain the real-time status data of the industrial robot in multiple dimensions through multiple preset sensors; Use the communication interface to transmit the real-time status data of the physical entity of the industrial robot, convert and encapsulate the acquired real-time status data, and perform data structured preprocessing; Acquire historical fault data of the industrial robot, preprocess the historical fault data, and cluster the preprocessed historical fault data using a clustering algorithm to obtain clustering clusters based on the fault type, and generate key monitoring features according to the fault components and fault parameters corresponding to each clustering cluster; Training the fault virtual models corresponding to the components of the three-dimensional simulation model through clustering clusters corresponding to key monitoring features, using the fault virtual models to learn the fault mechanism of the industrial robot, and using the three-dimensional simulation model and the fault virtual models to construct a twin space; The physical space is connected to the twin space using the preprocessed real-time status data, so as to achieve matching alignment and digital mapping between the real-time status data and the three-dimensional simulation model, and construct a digital twin model of the industrial robot; The digital twin model is adaptively updated in terms of model parameters, specifically: In the constructed digital twin model of the industrial robot, data of key monitoring features are updated, parameter variables corresponding to each key monitoring feature are obtained, a multivariable set is generated, and sparrow algorithm parameters are initialized based on the multivariable set; Initialize the sparrow population and the maximum number of iterations, construct an objective function based on the error between the measured value of the parameter variable and the output twin data value of the digital twin model, map the objective function to a fitness function, and calculate the fitness of each sparrow as the current optimal value; According to the fitness, sparrows are selected as discoverers based on the preset ratio, and the remaining sparrows are selected as joiners to update their positions. Sentinels are randomly selected based on the preset ratio of vigilants to update their positions. After the update, the fitness of each sparrow is recalculated. The chaotic algorithm is introduced to process the fitness of the current sparrow in a chaotic way, and the optimal position moving direction is determined by using the differential idea to guide the sparrow to update its position. The optimal solution of the parameter variable is obtained by iterative position update until the maximum number of iterations is reached. The digital twin model is configured according to the optimal values ​​of the parameter variables corresponding to each key monitoring feature to reflect the degradation degree of each component of the industrial robot.

2. According to claim 1, a method for monitoring the health status of an industrial robot based on digital twins is characterized in that: The fault virtual models corresponding to the components of the three-dimensional simulation model are trained by clustering the key monitoring features, specifically: The historical fault data in the clusters corresponding to the key monitoring features are used as training samples, the training samples are imported into the AlexNet network, features are extracted through convolution kernels of different sizes, and activation calculations are performed using activation functions; The pooling layer is used to reduce the feature dimension by downsampling, and the reduced feature is imported into the fully connected layer to obtain the probability distribution of multi-class sample labels using the softmax function; The model parameters of the AlexNet network are updated through training, and the weight parameters are optimized using Adam with adaptive learning rate. After verifying the classification performance of the AlexNet network, the network parameters are retained and the fault virtual model is output.

3. The method for monitoring the health status of an industrial robot based on digital twin according to claim 1, characterized in that: Feature extraction is performed in the multi-source heterogeneous data set to obtain feature data that can characterize the health status of the industrial robot, specifically: Calling the AlexNet network to perform convolution, pooling and full connection operations on the multi-source heterogeneous data set, obtaining a one-dimensional feature vector sequence of the full connection output, obtaining the Pearson correlation coefficient between each feature vector in the one-dimensional feature vector sequence and the health status of the industrial robot, and selecting a preset number of feature vectors according to the degree of correlation; The number of nearest neighbors k is set in the selected feature vector set, the Pearson correlation coefficient between the feature vectors is calculated, the first k feature vectors with the smallest values ​​are selected according to the number of nearest neighbors, and the feature data set corresponding to each feature vector is generated; An intersection operation is performed on the feature data sets corresponding to the feature vectors to obtain feature data that can characterize the health status of the industrial robot.

4. The method for monitoring the health status of an industrial robot based on digital twins according to claim 1, characterized in that: The characteristic data are used to construct the health index of the industrial robot, and the health status levels are divided. A health status assessment model is constructed based on a deep belief network, specifically: Constructing the health index of the industrial robot based on the acquired feature data, obtaining the historical operation and maintenance instances of the industrial robot, performing clustering based on the historical operation and maintenance instances, generating a data set of historical operation and maintenance instances in different health states, and dividing the health state levels according to their average health states; A health status assessment model is constructed based on a deep belief network, Dropout is introduced into the health status assessment model to prevent overfitting, and a feature data set of health indicators in a normal health state is obtained based on historical operation and maintenance examples to generate training data; Initialize the deep belief network, use the particle swarm algorithm to optimize the number of restricted Boltzmann machine layers, and stack the restricted Boltzmann machine layers according to the optimal solution of the number of restricted Boltzmann machine layers; The training data is trained layer by layer from the bottom layer to the top layer, and a deep architecture with deep nonlinear mapping capability is generated by iteratively updating the weights. The extracted features are transmitted from the top layer to the bottom layer to obtain the reconstructed data corresponding to the input training data. The health status assessment model is trained according to the principle of minimizing the error between the reconstructed data and the input training data, and the average error between the reconstructed data and the training data is obtained to set the health status threshold.

5. The method for monitoring the health status of an industrial robot based on digital twin according to claim 1, characterized in that: The health status assessment model is used to obtain the health status assessment results of industrial robots, specifically: Based on the health index, health index parameters of the current industrial robot are collected, the health index parameters are imported into the health status assessment model as model input, and data of the health index parameters are reconstructed; The error between the reconstructed data and the health indicator parameter is obtained, and the error is compared with the health status threshold. The health status assessment result of the industrial robot is generated through the comparison result, and when the health status assessment result is an abnormal health status, an abnormal health status alarm is generated.

6. An industrial robot health status monitoring system based on digital twins, characterized in that: Implementing the health status monitoring method of an industrial robot based on digital twins as described in any one of claims 1 to 5, the system comprises: a data acquisition and preprocessing module, a digital twin module, a multi-source data fusion module, a health status assessment module, and a health status output and visualization module; The data acquisition and preprocessing module uses multiple sensors to collect real-time status data of the industrial robot and preprocesses the collected real-time status data; The digital twin module constructs a three-dimensional simulation model according to the physical entity of the industrial robot, matches the three-dimensional simulation model with the collected real-time status data and performs data alignment to construct a digital twin model of the industrial robot, and maintains consistency with the degradation state of the physical entity by adaptively updating the model parameters of the digital twin model; The multi-source data fusion module fuses the twin data output by the digital twin model with the real-time status data to generate a multi-source heterogeneous data set; The health status assessment module uses feature selection to obtain feature data that can characterize the health status of the industrial robot in the multi-source heterogeneous data set; uses the feature data to construct the health index of the industrial robot, divides the health status level, establishes a health status assessment model, and uses the trained health status assessment model to obtain the health status assessment result of the industrial robot; The health status output and visualization module performs three-dimensional visualization on the health status evaluation result of the industrial robot output by the health status evaluation module, and issues an abnormal alarm according to the health status evaluation result.

Images