CN-121798640-B - Discrete manufacturing-oriented digital twin mechanical arm control method and system

CN121798640BCN 121798640 BCN121798640 BCN 121798640BCN-121798640-B

Abstract

A digital twin mechanical arm control method and system for discrete manufacturing is characterized in that a virtual mechanical arm is built for a physical mechanical arm based on a digital twin technology, the physical mechanical arm and the virtual mechanical arm conduct bidirectional interaction through an uplink and a downlink, the control method comprises the following steps of receiving multimode sensing data of the physical mechanical arm in real time through the uplink, conducting physical consistency verification on the multimode sensing data based on a physical information neural network, identifying real physical parameters of the physical mechanical arm in real time through online learning of the physical information neural network when the data are normal, conducting multimode causal logic verification on the multimode sensing data based on a space-time diagram neural network, and the control method and system can deeply fuse physical mechanism and deep learning characteristics and have a digital twin interaction safety method with virtual-real parameter synchronization and anti-exercise capability so as to guarantee high robustness and high reliability of the discrete manufacturing system in an uncertain environment.

Inventors

ZHANG MINGCHUAN
ZHAO XUHUI
YANG LEI
WANG DONG
LIU MUHUA
WU QINGTAO
Yang meiyi
ZHU JUNLONG
WANG LIN
FENG JIAMEI

Assignees

河南科技大学

Dates

Publication Date: 20260508
Application Date: 20260309

Claims (10)

1. A digital twin mechanical arm control method for discrete manufacturing is characterized in that a virtual mechanical arm is built for a physical mechanical arm based on a digital twin technology, the physical mechanical arm and the virtual mechanical arm perform bidirectional interaction through an uplink and a downlink, and the control method comprises the following steps: The physical consistency check is carried out on the multi-modal sensing data based on the physical information neural network PINNs, and when the data are normal, the real physical parameters of the physical mechanical arm are recognized in real time through the online learning of the physical information neural network PINNs Meanwhile, based on the result of the physical consistency check and the causal logic check, the credibility of the uplink data is judged, and the credible real physical parameters are identified Writing into a dynamic physical parameter pool; nominal control instruction to be issued When the dynamic physical parameter pool is used, the real physical parameters of the dynamic physical parameter pool are called Reconstructing physical properties of the Unity sandbox to obtain a high-fidelity simulation environment, and executing nominal control instructions in the high-fidelity simulation environment Performing a challenge robustness exercise and calculating a cumulative risk score If the cumulative risk score Exceeding a safety threshold Then the nominal control instruction is based on the control barrier function CBF Performing optimization correction to generate safety control instructions And the safety control instruction is sent to the computer Or cumulative risk score Does not exceed a safety threshold Is controlled by the nominal control instruction of (1) And transmitting the data to the physical mechanical arm through a downlink.
2. The discrete manufacturing-oriented digital twin mechanical arm control method according to claim 1, wherein the step of performing physical consistency verification on the multi-modal sensing data based on a physical information neural network PINNs and recognizing real physical parameters of the physical mechanical arm in real time through online learning of the physical information neural network PINNs when the data is normal comprises: Constructing a physical information neural network PINNs model, wherein a loss function of the model comprises a physical constraint term based on a Lagrange dynamics equation; joint state data in the multi-modal awareness data Inputting the physical information neural network PINNs model to obtain a predicted moment ; Calculating the predicted moment And the measured moment obtained by conversion of the measured current of the motor in the multi-mode sensing data Physical coherence residuals between ; According to statistics Criterion calculation of first physical threshold First physical threshold A dynamic threshold for distinguishing model fitting errors from significant physical anomalies; if the physical consistency residual error Exceeding a first physical threshold Judging that the physical abnormality exists and intercepting the data; if the physical consistency residual error Does not exceed the first physical threshold Then use the physical consistency residual Back propagation updating is carried out on the leachable physical parameters in the physical information neural network PINNs model, and the updated parameter values are used as the real physical parameters 。
3. The method of claim 2, wherein the learnable physical parameters include joint friction coefficient and link load mass parameter.
4. The discrete manufacturing-oriented digital twin mechanical arm control method according to claim 1, wherein the step of performing multi-modal causal logic verification on the multi-modal sensory data based on a space-time diagram neural network ST-GNN comprises: Constructing a graph structure, and mapping the multi-modal awareness data to predefined graph structure nodes, wherein the graph structure nodes represent different sensors, and edges represent physical causal links among the sensors; Computing causal attention coefficients between nodes through graph attention mechanism ; Taking causal attention coefficient of current moment Reference attention profile under normal conditions As a logical anomaly score ; Setting logic threshold based on statistical distribution characteristics of historical operation data under normal working conditions ; If the logic abnormality score Exceeding a logic threshold Determining that the logic is abnormal, and if the logic is abnormal, scoring Does not exceed a logic threshold And judging that the logic is normal.
5. A discrete manufacturing-oriented digital twin mechanical arm control method as defined in claim 1 wherein the multi-modal sensing data comprises joint states from an encoder Motor current from the drive Contact force from tip force sensor With moment, and end space coordinates from the vision system And all data is time synchronized before entering the physical information neural network PINNs.
6. The method for controlling the digital twin mechanical arm for discrete manufacturing of claim 1 is characterized in that when the credibility of uplink data is judged, if the physical consistency check is not abnormal and the multi-mode causal logic check is not abnormal, the data is judged to be credible, and if any check report is abnormal, the data is judged to be not credible and is intercepted.
7. A digital twin mechanical arm control method oriented to discrete manufacturing as defined in claim 1, wherein the nominal control command is given in a high fidelity simulation environment The step of performing an antagonistic robustness exercise comprises: the nominal control instruction As a measured object, ADVERSARIAL RL intelligent agents integrated in the Unity sandbox are used as attackers; The object to be tested is controlled along the nominal control instruction Moving, wherein an attacker simulates at least one of applied network delay, moment noise and virtual external force to interfere and induce the instability of the measured object; Real-time monitoring of collision state of virtual mechanical arm in high-fidelity simulation environment Moment of articulation ; Based on the collision state And joint moment The degree of exceeding the rated value is calculated to obtain the accumulated risk score 。
8. A digital twin mechanical arm control method for discrete manufacturing according to claim 1, characterized in that the safety threshold value Calibrating by the following method: Running a reference track in the Unity sandbox, applying environmental noise, and recording a reference risk peak value And multiplied by a safety margin coefficient Obtaining a safety threshold 。
9. A discrete manufacturing-oriented digital twin mechanical arm control method according to claim 1, characterized in that the nominal control instruction is based on a control obstacle function CBF The step of performing optimization correction includes: Based on the dynamic model of the physical mechanical arm and the real physical parameters migrated from the dynamic physical parameter pool Construction of control obstacle function ; To minimize correction instructions And the nominal control instruction Targeting the deviation of the control obstacle function The defined safety conditions are constraints, and a quadratic programming model OSQP is constructed; solving the quadratic programming model OSQP to obtain the nominal control instruction The safety control command with minimum Euclidean distance 。
10. A digital twin mechanical arm control system for discrete manufacturing, comprising: the physical entity layer comprises a physical mechanical arm with a sensor and a physical controller, wherein the sensor is used for acquiring multi-mode sensing data of the physical mechanical arm in real time and transmitting the multi-mode sensing data to the edge computing layer; the edge computing layer comprises an edge gateway and an edge server, wherein the edge gateway is used for synchronizing time of the multi-mode sensing data transmitted by the physical entity layer and uploading the multi-mode sensing data to the edge server; The digital twin interaction layer comprises an uplink reasoning engine with a built-in physical information neural network PINNs and a space-time diagram neural network ST-GNN, a memory area with a dynamic physical parameter pool and a downlink defense engine with a built-in Unity sandbox and a CBF corrector, and is used for receiving the multi-mode sensing data uploaded by the edge computing layer and executing the digital twin mechanical arm control method according to any one of claims 1-9.

Description

Discrete manufacturing-oriented digital twin mechanical arm control method and system Technical Field The invention relates to the technical field of mechanical arm control, in particular to a discrete manufacturing-oriented digital twin mechanical arm control method and system. Background Digital twin technology has become a key driving force for discrete manufacturing to intelligentize, and the core is to construct a closed-loop interaction system between a physical entity and a virtual model. In particular, in the scenes of precise assembly of aviation, automatic welding of automobiles and the like, a six-degree-of-freedom industrial mechanical arm is used as a key execution unit, and the operation of the six-degree-of-freedom industrial mechanical arm depends on high-frequency closed-loop interaction between a physical entity and a virtual model. The real-time state of the physical equipment is perceived through the uplink, and the downlink feedback control instruction is utilized, so that the accurate mapping of the manufacturing process is realized. An efficient and reliable digital twin system must be based on the absolute trust and security of the bi-directional data stream of "perceptron-decision-enforcement". However, in a complex industrial field environment, ensuring a virtual-real two-way interactive system faces serious security challenges and trust crisis. An attacker may input false data which accords with a numerical range but violates a physical law to a twin body through man-in-the-middle attack or sensor fault injection, such as replay attack and data tampering, so that the twin body can misjudge a physical state and further induce decision failure. And in the long-term operation process of the mechanical arm, physical parameters such as joint friction coefficient, connecting rod damping, transmission clearance and the like of the mechanical arm can change time along with the aging and lubrication state change of equipment. If the digital twin cannot sense and synchronize these tiny physical property changes in real time, virtual-real gaps are generated, resulting in overshooting, concussion and even collision of the issued control instructions when executed on the aged physical equipment. The existing control instruction generation is based on algorithm planning in an ideal environment, and unstructured interference such as network random delay, instantaneous load disturbance and the like is not fully considered. Once subjected to a malicious network attack or extreme conditions, instructions that appear to be compliant may evolve instantaneously into abnormal instructions that damage the device. The existing general solution has the main limitation that firstly, the existing uplink detection multi-reliance fixed rigid body dynamic model calculates residual errors, the method adapts to parameter drift caused by equipment aging, and a large number of false positives are easy to generate. Pure data driven methods, although not requiring precise modeling, lack physical interpretability, are susceptible to countersample spoofing where statistical laws are normal but causal logic is violated. Secondly, the prior art fusion technology only focuses on value alignment and ignores physical causal topological relation among sensors, so that the system is difficult to identify hidden semantic level attacks which only tamper with a single mode. Thirdly, the existing downlink verification mostly adopts a static threshold value or simple collision detection under an ideal simulation environment, and the robustness of the network jitter, malicious disturbance and other real scenes cannot be estimated. When a dangerous instruction is detected, only a passive strategy of interception or sudden stop is adopted, and an active mechanism automatically corrected to be a safe instruction is lacked, so that frequent stop is caused, and the continuity of a production line and the comprehensive efficiency of equipment are seriously influenced. In summary, the prior art has significantly short plates in terms of physical parameter adaptive migration, cross-modal causal verification, and command antagonism defense and correction, facing the increasingly complex industrial security threats and equipment full life cycle management needs. Therefore, it is necessary to provide a digital twin mechanical arm control method and system for discrete manufacturing. Disclosure of Invention The invention aims to provide a digital twin mechanical arm control method and a system for discrete manufacturing, which can deeply fuse a physical mechanism and a deep learning characteristic, and have a digital twin interaction safety method with virtual-real parameter synchronization and anti-exercise capability so as to ensure high robustness and high reliability of a discrete manufacturing system in an uncertain environment. The technical scheme adopted by the invention is that the digital twin mechanical arm control method for discr