A method and system for evaluating the feasibility of programming information for an industrial control system

Abstract

The application provides a kind of industrial control system programming information feasibility evaluation method and system, the method comprises: using multi-source sensor obtains the time series and operating parameter sequence of historical acquisition mechanical arm, constructs synchronous historical mechanical data;The length of motion time cycle is determined to the historical mechanical data, and the historical mechanical data is segmented to obtain a plurality of video image data segments according to the length of motion time cycle;Video image data segment is analyzed to a plurality of graph neural networks, and arm operation feature vector set is obtained;Arm operation feature vector set is analyzed using neural network, and a verification model is constructed;Real-time data of real-time mechanical arm is collected, and the real-time data of mechanical arm is input into the verification model, and operation result is generated.Through the verification model, the operation defects of the mechanical arm are completely evaluated, so that the feasibility of the programming information is evaluated in practice.

Description

Technical Field

[0001] This invention relates to the field of electronic information programming, and in particular to a method and system for assessing the feasibility of programming information in industrial control systems. Background Technology

[0002] With the rapid development of industrial automation, industrial control systems have been widely applied in various manufacturing industries, warehousing and logistics, and automated production lines. Among these, robotic arms, as key equipment, are widely used in assembly, welding, painting, and material handling. To ensure the efficient and stable operation of robotic arms, traditional control methods rely on hard-coded programming and periodic testing. However, as the complexity of robotic arm control tasks increases, these traditional methods are no longer sufficient to meet the high requirements for accuracy, response speed, and maintainability.

[0003] In modern industrial control systems, an increasing number of intelligent technologies (information engineering technologies) are being introduced to optimize the control and monitoring of robotic arms. Digital twin technology, as a cutting-edge simulation and analysis tool, has become an effective means to improve the performance of robotic arm systems. Through digital twins, the behavior and performance of robotic arms can be simulated and optimized in real time in a virtual environment, potential faults can be predicted in advance, and control strategies can be optimized, thereby achieving predictive maintenance and precise control.

[0004] However, to achieve this goal, the core issue that needs to be addressed first is how to extract valuable information from massive amounts of sensor data for effective modeling and analysis. Existing industrial control systems often rely on traditional sensor data processing methods, such as static data analysis and simplified time series processing. However, these methods are inadequate when faced with high-dimensional, complex data with strong spatiotemporal correlations, especially when it comes to the feasibility verification and evaluation of experimental industrial control system programming information. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for feasibility assessment of programming information in industrial control systems, which solves the above-mentioned technical problems pointed out in the prior art.

[0006] This invention provides a method for feasibility assessment of programming information in industrial control systems, comprising the following steps:

[0007] By using multi-source sensors to acquire historical data of the robotic arm, including time series and operating parameter sequences, synchronous historical mechanical data can be constructed.

[0008] The historical mechanical data is used to determine the length of the motion time cycle, and the historical mechanical data is segmented according to the length of the motion time cycle to obtain multiple video image data segments; graph neural network analysis is performed on the multiple video image data segments to obtain a set of arm movement feature vectors; neural network analysis is performed using the set of arm movement feature vectors to construct a verification model;

[0009] Real-time data from the robotic arm is collected and input into the verification model to generate the running results.

[0010] Preferably, the motion time period length is determined from the historical mechanical data, and the specific operation steps are as follows:

[0011] Using the historical mechanical data, the joint angles of key joints of the robotic arm are randomly extracted, and the rotation angle change value of the joint angle is calculated. The delayed time points of joint angle rotation at adjacent time points are filtered by a preset rotation angle change threshold to form a time delay curve. The corresponding delay time point of the maximum delay time point in the time delay curve and the maximum delay time point in the next time point are selected as the motion time cycle length.

[0012] Preferably, historical mechanical data is segmented according to the length of the motion time cycle to obtain multiple video image data segments; graph neural network analysis is performed on the multiple video image data segments to obtain a set of arm movement feature vectors; neural network analysis is then performed using the set of arm movement feature vectors to construct a validation model. The specific operation steps are as follows:

[0013] The historical mechanical data is segmented according to the motion time period length to form multiple video image data segments; the video image data segments are converted into two-dimensional images to form grayscale images, and wavelet packet decomposition is used to calculate the frequency band energy to obtain multiple frequency band energy feature vector sets; all key joints in the historical mechanical data are connected to adjacent key joints to construct a topology graph of the robotic arm kinematic chain; a graph neural network is used to perform convolutional layers on the multiple frequency band energy feature vector sets to obtain a topological constraint feature tensor; the grayscale image is weighted with the multiple frequency band energy feature vector sets and the topological constraint feature tensor to obtain the arm operation feature vector set.

[0014] Preferably, the feature vector set of the arm's movement is used for neural network analysis to construct a verification model. The specific operation steps are as follows:

[0015] A probabilistic defect topology network is constructed using a neural network; the feature vector set of the arm's movement is input into the probabilistic defect topology network to construct a defect triggering probability matrix; singular value decomposition is performed on the defect triggering probability matrix to obtain a defect propagation path; and a verification model is constructed using the defect propagation path.

[0016] Preferably, a probabilistic defect topology network is constructed using a neural network; the feature vector set of the arm's movement is input into the probabilistic defect topology network to construct a defect triggering probability matrix. The specific operation steps are as follows:

[0017] By using a neural network to input the frequency band energy feature vector set corresponding to the length of each motion time cycle into the input layer, and through the time dependence analysis of the progressive deterioration trend of the LSTM hidden layer and Bayesian output layer of the neural network, a probabilistic defect topology network is obtained; the arm running feature vector set is input into the probabilistic defect topology network, and the defect triggering probability matrix is ​​output.

[0018] Preferably, the defect triggering probability matrix is ​​subjected to singular value decomposition to obtain the defect propagation path; the specific steps for constructing a verification model using the defect propagation path are as follows:

[0019] The defect triggering probability matrix is ​​subjected to singular value decomposition to obtain a left singular vector matrix; principal component analysis is performed on the left singular vector matrix to obtain the maximum singular value of the unit element in each column, and the defect propagation path is constructed.

[0020] An initial verification model is constructed; the initial verification model is subjected to multimodal perturbation simulation using the defect propagation path to obtain an updated verification model; virtual simulation data is collected under the verification model to generate a virtual verification defect triggering probability matrix; the defect triggering probability matrix and the virtual verification defect triggering probability matrix are used for verification to obtain the final verification model.

[0021] Preferably, the initial validation model is constructed, and the specific steps are as follows:

[0022] Obtain a 3D CAD model of the robotic arm and input the 3D CAD model of the real robotic arm into the simulation platform; define the key joint types and motion axes of the simulation platform according to the kinematic chain topology diagram of the robotic arm;

[0023] Physics engine parameters are introduced into the simulation platform, and an initial verification model is constructed using these parameters, along with key joint types and motion axes.

[0024] Preferably, the initial verification model is subjected to multimodal perturbation simulation using the defect propagation path to obtain an updated verification model. The specific operation steps are as follows:

[0025] The defect propagation path is input into the initial verification model to perform multimodal disturbance simulation of superimposed torque pulse, spatial current field and communication delay, and the initial verification model is updated to obtain a new verification model.

[0026] Preferably, virtual simulation data is collected under the verification model to generate a virtual verification defect triggering probability matrix; the defect triggering probability matrix and the virtual verification defect triggering probability matrix are then used for verification to obtain the final verification model. The specific operation steps are as follows:

[0027] Virtual sensors are deployed on the verification model, and virtual simulation data is collected using the virtual sensors.

[0028] The complete operation process of obtaining the arm running feature vector set in the above steps is repeated to reconstruct the final virtual verification arm running feature vector set.

[0029] The virtual verification arm utilizes the feature vector set to input the probability defect topology network and generates a virtual verification defect trigger probability matrix.

[0030] Align the defect triggering probability matrix with the virtual verification defect triggering probability matrix by dimension and calculate the KL divergence.

[0031] A preset divergence threshold is set; it is then determined whether the KL divergence is less than the divergence threshold.

[0032] If so, it is determined that the results of the virtual verification defect triggering probability matrix and the defect triggering probability matrix under the verification model are consistent, and the closed-loop operation ends;

[0033] Accordingly, this invention also proposes a feasibility assessment system for programming information in industrial control systems, comprising: a data acquisition module; an analysis module; and an identification module;

[0034] The acquisition module is used to acquire the time series and operating parameter series of the historical acquisition robot arm using multi-source sensors, and to construct synchronized historical mechanical data;

[0035] The analysis module is used to determine the motion time cycle length of the historical mechanical data, segment the historical mechanical data according to the motion time cycle length to obtain multiple video image data segments; perform graph neural network analysis on the multiple video image data segments to obtain a set of arm movement feature vectors; and use the set of arm movement feature vectors to perform neural network analysis to build a verification model.

[0036] The identification module is used to collect real-time data from the robotic arm, input the real-time data from the robotic arm into the verification model, and generate the running results.

[0037] Compared with the prior art, the embodiments of the present invention have at least the following technical advantages:

[0038] Analysis of the above-mentioned feasibility assessment method and system for programming information in industrial control systems provided by this invention reveals that, in specific applications, firstly, multi-source sensors are used to acquire historical data on the time series and operating parameter sequences of the robotic arm, constructing synchronized historical mechanical data. This involves analyzing the time series data of the robotic arm's joint angles, angular velocities, and current to capture abnormalities such as motion delays and response lags during operation. Then, time delay curves and rotation angle change thresholds are used to identify whether the robotic arm exhibits faults or abnormal motion patterns. Simultaneously, the periodicity of the robotic arm's motion signals is analyzed to determine the length of each motion cycle and identify the maximum delay point within each cycle, thereby obtaining the robotic arm's working cycle (i.e., motion rhythm). Next, a probabilistic defect topology network constructed using a neural network is used to identify potential defects in the robotic arm. Singular value decomposition is used to construct defect propagation paths, allowing for early detection of potential problems during the control process (such as communication failures and overloads), enabling effective verification and ensuring accurate identification of the feasibility of the robotic arm's actions.

[0039] Then, the historical mechanical data is segmented according to the length of the motion time cycle to form multiple video image data segments. These video image data segments are then converted into two-dimensional images to form grayscale images, and wavelet packet decomposition is used to calculate the frequency band energy. In this process, the method utilizes multiple data sources such as video image data, frequency band energy features of wavelet packet decomposition, and grayscale images. It combines attention mechanism networks and graph neural networks for fusion analysis to improve the multi-dimensional accuracy of motion monitoring and capture complex motion behaviors, such as coupling effects between joints and current changes, thereby forming a more comprehensive monitoring system. This system uses neural networks for motion recognition and defect prediction.

[0040] Furthermore, by inputting the frequency band energy feature vector of each motion cycle of the robotic arm into a neural network, and then using hidden layers to set memory units to analyze the progressive defect deterioration trend for each motion cycle, the defect types occurring in continuous time cycles are identified, increasing the verification accuracy of the validation model. Next, a Bayesian output layer is used to analyze the failure probability distribution of key joints in the progressive deterioration trend, forming a probabilistic defect topology network. This network reflects the possible defects of each joint, forming defect types. The feature vector set of the arm's operation is input into the probabilistic defect topology network to construct a defect triggering probability matrix, reflecting the combination patterns between nodes, i.e., the defect types that may arise when key joints are paired with actuators and power supplies. The defect triggering probability matrix is ​​then subjected to singular value decomposition to form a left singular vector matrix, extracting the combination patterns between nodes and defect types to construct defect propagation paths. An initial validation model is constructed using existing mechanical system and simulation models as the basis for the validation model. This initial validation model is then subjected to virtual simulation data and multimodal simulation perturbations, and a validation model is constructed in actual engineering to verify its effectiveness and accuracy. Optimization and correction further enhance the model's predictive ability. Attached Figure Description

[0041] Figure 1 This is an overall flowchart of a feasibility assessment method for programming information in an industrial control system, as described in Embodiment 1.

[0042] Figure 2 This is a flowchart of a feasibility assessment method for programming information in an industrial control system, as described in Embodiment 1.

[0043] Figure 3 This is a flowchart illustrating the construction of a verification model using an arm movement feature vector set as part of an industrial control system programming information feasibility assessment method according to Embodiment 1.

[0044] Figure 4 This is a schematic diagram illustrating the network principle and effect of the attention mechanism in the feasibility assessment method for programming information in an industrial control system, as described in Embodiment 1.

[0045] Figure 5 This is a flowchart illustrating the construction and verification model of a neural network for a feasibility assessment method of programming information in an industrial control system, as described in Embodiment 1.

[0046] Figure 6 This is a schematic diagram of the neural network in a feasibility assessment method for programming information in an industrial control system, as described in Embodiment 1.

[0047] Figure 7 This is a flowchart illustrating the verification of a feasibility assessment method for programming information in an industrial control system according to Embodiment 1.

[0048] Figure 8 This is a flowchart of a feasibility assessment system for programming information in an industrial control system, as described in Embodiment 2.

[0049] Labels: Acquisition module 10; Analysis module 20; Identification module 30. Detailed Implementation

[0050] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0052] Example 1

[0053] like Figure 1 , Figure 2 As shown, this embodiment of the invention provides a method for feasibility assessment of programming information in an industrial control system, including the following steps:

[0054] S1: Use multi-source sensors to acquire the time series and operating parameter series of historical robotic arms (i.e., mechanical devices mainly consisting of robotic hands and robotic arms) and construct synchronized historical mechanical data;

[0055] It should be noted that an array of various types of sensors, such as inertial measurement units, torque sensors, current sensors, and vision cameras, is used to collect physical quantities (i.e., parameter sequences) that change over time during the movement of the robotic arm, such as joint angles, angular velocities, current, and end-effector pose. All sensor data are aligned on the same time reference (i.e., time synchronization) and arranged according to spatial topology (i.e., spatial synchronization), which is also known as synchronization. These collected data are then processed into a three-dimensional tensor to form historical mechanical data.

[0056] S2: Determine the motion time cycle length of the historical mechanical data, and segment the historical mechanical data according to the motion time cycle length to obtain multiple video image data segments; perform graph neural network analysis on the multiple video image data segments to obtain an arm operation feature vector set; use the arm operation feature vector set to perform neural network analysis to construct a verification model;

[0057] It should be noted that determining the length of the robotic arm's motion time cycle (i.e., the time required for the robotic arm to complete a complete, repetitive motion cycle, or the time required for the robotic arm to complete a repetitive motion task; it is a key time scale for dividing the continuous operation process into analyzable units, used to support subsequent extraction of the arm's motion feature vector set for state evaluation and building a validation model) involves dividing historical mechanical data into multiple video image data segments according to the motion cycle. Each data block contains the spatiotemporal data of the robotic arm within one motion cycle. Then, the video image data segments are input into a graph neural network (GNN) for analysis to extract the arm's motion feature vector set, representing the key features of the robotic arm's motion. Finally, the arm's motion feature vector set is used to build a validation model, that is, through further analysis and training of the neural network, a validation model that can simulate and predict the performance of the robotic arm in real time is obtained.

[0058] By segmenting the data and applying graph neural networks, we extract the feature vector set of robotic arm operation from complex spatiotemporal data. This feature vector set helps to describe and predict the performance of the robotic arm under different working conditions. Graph neural networks can effectively process data with spatial topological relationships, ensuring the accuracy and efficiency of feature extraction. Building a verification model allows us to simulate and verify various operating conditions of the robotic arm in a virtual environment, thereby providing accuracy for optimizing control strategies and predicting faults.

[0059] S3: Collect real-time data from the robotic arm, input the real-time data from the robotic arm into the verification model, and generate the running results (i.e., defect result evaluation).

[0060] It should be noted that after trainees perform experimental industrial control system programming, the robotic arm executes the programming instructions (including joint rotation angles, rotation speeds, etc.). Subsequently, various forms of real-time data (such as joint loosening and communication signal lights) can be collected from the robotic arm through sensors installed in the system. Regarding the real-time data of the robotic arm (after the final verification model is determined in step S2, real-time data is input in step S3. At this time, the real-time data is essentially a kind of mechanical data to be tested, which needs to be used to test the feasibility of the programming information) after being input into the verification model, dynamic operation status analysis and optimization are performed to detect defects that occur during the execution of the robotic arm's movements (i.e., defect types include current overload, joint wear, joint loosening, communication delay, and angle deviation). The feasibility of the programming information in the robotic arm's movement is evaluated based on the defects found, and the programming information is modified accordingly, thereby teaching the trainees.

[0061] Specifically, such as Figure 3As shown, in step S2, the motion time cycle length of the historical mechanical data is determined, and the historical mechanical data is segmented according to the motion time cycle length to obtain multiple video image data segments; graph neural network analysis is performed on the multiple video image data segments to obtain a set of arm movement feature vectors; the arm movement feature vector set is used for neural network analysis to construct a verification model. The specific operation steps are as follows:

[0062] In steps S21-S22, the joint angles of key joints of the robotic arm are randomly extracted using the historical mechanical data, and the rotation angle change value of the joint angle is calculated. The delayed time points of joint angle rotation at adjacent time points are filtered by a preset rotation angle change threshold to form a time delay curve. The corresponding delay time point of the maximum delay time point in the delayed time point and the maximum delay time point in the next time point are selected by the time delay curve to determine the motion time cycle length.

[0063] S21: Using the historical mechanical data, key joints are randomly extracted from the robotic arm, and the joint angles of the key joints are analyzed in time series to form a joint angle sequence;

[0064] The joint angles at adjacent time points are calculated to obtain the change in rotation angle.

[0065] It should be noted that step S2 above identifies whether there is motion lag or slow response during the execution of instructions by the robotic arm; constructs a time trajectory (i.e., time delay curve) reflecting motion abnormalities, and transforms potential problems such as the robotic arm's motion loss, slow drive response, and excessive load into observable time delay signals; at the same time, it acquires data on the change value of rotation angle to help trainees understand the trend of data changes more intuitively.

[0066] S22: Preset rotation angle change threshold; determine whether the rotation angle change value is less than the rotation angle change threshold;

[0067] If so, it is determined that there is a delay in the joint angle rotation at the adjacent time point, and the delay time point is recorded. If there are multiple delay time points, a time delay curve is formed on the horizontal axis.

[0068] The correct rotation angle changes are generated based on the joint angle sequence of the time series to form the correct rotation angle signal;

[0069] Align the correct rotation angle signal (i.e., the signal is also a curve) with the time delay curve to determine the similarity in time structure (i.e., align the correct rotation angle signal with the time delay curve at each time point, and determine whether the structure of the curve (or signal) at each time point overlaps, thereby determining the similarity of the curve (or signal), which is to indirectly determine the consistency of the curve (or signal), and thus determine the period of the delayed time point); find the maximum delay time point of the time delay curve, and find the next maximum delay time point from the time point of the maximum delay time point when the mechanical action is executed, and determine the corresponding delay time point of the next maximum delay time point as the length of the motion time period; study It was found that communication failures in issuing commands can easily lead to time delays. However, by determining the length of the motion time cycle, a complete motion action of a robotic arm control command can be captured. Moreover, the identification of the motion time cycle length can not only capture the rhythm of the robotic arm's physical movements, but also indirectly sense and reflect communication anomalies in the control system (such as command delays), thereby enabling the monitoring of the health status of the robotic arm's motion chain. At the same time, the maximum delay time point is determined based on the overlap of the curves (or signals). When the two overlap, the time point with the most obvious vertical change (i.e., the delay between the upper and lower time points of the overlap is determined using the x and y axis coordinates) is taken as the maximum delay time point of the time delay curve.

[0070] It should be noted that the determination of whether the change in rotation angle is less than the rotation angle change threshold is as follows: For example, if the command for joint rotation time is 2 seconds and the joint rotation angle is 180° (i.e., the rotation angle change threshold), but after this command is issued, if there is a time delay, such as a one-second delay, the control time is shortened by one second. In this case, the remaining one second can only rotate 90° (i.e., the rotation angle change value). Therefore, it is determined whether the change in rotation angle is less than the rotation angle change threshold. In this case, it is determined that there is a delay in the control time.

[0071] Align the correct rotation angle signal (i.e., the rotation angle signal is also a curve) with the time delay curve, and determine the similarity of the rotation angle at each time point (i.e., the proximity of the peaks or curves, that is, similarity or overlap). Assume that the robotic arm is performing a repetitive action of "grasping-lifting-lowering-returning", which is completed once every 2 seconds; the curve of its first joint angle changing with time is a waveform that repeats once every 2 seconds (i.e., an action angle is presented as a curve). By comparing the waveforms of the correct rotation angle signal and the time delay curve, when the delay time is 0 seconds, the signal alignment is such that the peaks or curves completely overlap, and the maximum peak similarity is the highest (maximum peak). When the delay time is 1 second, the peaks or curves... There will be some misalignment, at which point the maximum peak similarity is low; at the 2nd second, they completely overlap again, with a very high similarity (the second peak); when the time point is the 4th second, they overlap again (after two cycles), and the similarity is also very high (the third peak); therefore, at a delay of 2 seconds, the first significant non-zero peak will appear; this 2 seconds is the time required for the signal to repeat once (i.e., one cycle) – that is, the length of the motion time cycle; therefore, the time delay corresponding to the next most obvious peak is the time difference between the first time the signal is highly similar to itself again, forming the length of a complete cycle; by comparing the actual signal with the theoretical signal, students will understand how to identify control delay through curve comparison;

[0072] Steps S23-S24 mainly describe how to segment the historical mechanical data according to the length of the motion time cycle to form multiple video image data segments (video image data segments refer to the time series, video image data, and spatial data of the robotic arm, and it has been stated in S1 that the historical mechanical data collected can reflect the joint angles, angular velocities, currents, end-effector poses, etc. of the robotic arm); convert the video image data segments into two-dimensional images to form grayscale images, and use wavelet packet decomposition to calculate the frequency band energy, obtaining multiple frequency band energy feature vector sets; and perform analysis on all key joints in the historical mechanical data (key joints are joints with connection relationships, such as: the robotic arm and its connected robotic forearm, the robotic forearm and its connected robotic hand joint). The connection of adjacent key joints constructs a kinematic chain topology graph for the robotic arm. A graph neural network is used to perform convolutional layers on the multiple frequency band energy feature vector sets to obtain a topological constraint feature tensor. A weighted sum of the grayscale image, multiple frequency band energy feature vector sets, and the topological constraint feature tensor is then used to obtain the arm's operational feature vector set (i.e., a weighted sum of the grayscale image, multiple frequency band energy feature vector sets, and the topological constraint feature tensor is used to obtain the arm's operational feature vector set). An attention mechanism network is used to weightedly fuse the grayscale image, multiple frequency band energy feature vector sets, and the topological constraint feature tensor, and then the fused feature vector is subjected to dimensionality reduction. The arm's operational feature vector set is formed by combining the fused features of video image data segments of each motion cycle length.

[0073] S23: The historical mechanical data is segmented using the motion time cycle length to form multiple video image data segments;

[0074] Each video image data segment is converted into a two-dimensional image and then converted to grayscale to form a grayscale image;

[0075] Wavelet packet decomposition is used to calculate the motion time period length of the grayscale image signal (i.e., the curve signal formed by the acquisition of angle, angular velocity, and current according to the time series; by calculating the frequency band energy, the running change state of information such as current can be understood; the frequency band energy reflects the energy distribution generated by the running current and joint angle when the arm is running, that is, the distribution of current, angle, etc.), and multiple frequency band energy feature vector sets are obtained.

[0076] It should be noted that complex multidimensional time-series data is converted into a structured, comparable set of low-dimensional features (i.e., grayscale images); each feature vector reflects the energy distribution characteristics of the system operation within that period (such as whether high-frequency vibrations intensify, or whether excessive current is concentrated in the low-frequency band).

[0077] S24: Connect adjacent key joints in all the historical mechanical data (i.e., the mechanical connections of the joints of the robotic arm, that is, the kinematic chain connections) to construct an adjacency matrix; add edge weights to the adjacency matrix (i.e., the edges represent the lines connecting the joints, and the weights represent the kinematic coupling strength (i.e., the inertial influence of the link mass and length calculation)) to construct the kinematic chain topology of the robotic arm.

[0078] A graph neural network is used to construct a graph convolutional layer by using the multiple frequency band energy feature vector sets as input features for each key joint.

[0079] The graph convolutional layer captures neighbor information at each layer (i.e., capturing neighbor information is represented as learning the mutual influence between joints), forming a topological constraint feature tensor (i.e., each node (joint) has a new feature vector that incorporates neighbor information, which is to extract features with structure awareness).

[0080] The grayscale image is weighted with multiple frequency band energy feature vector sets and topological constraint feature tensors using an attention mechanism network to generate a dimension-compressed arm movement feature vector set (i.e., dimensionality reduction and fusion using an attention mechanism network), specifically including:

[0081] The grayscale image is unified with multiple frequency band energy feature vector sets and topological constraint feature tensors in a feature space (i.e., dimension alignment).

[0082] The grayscale image is encoded in time series to construct a grayscale image matrix. The grayscale feature vector is then formed through the grayscale image matrix (that is, the 50x50 of the grayscale image is encoded to form a grayscale image matrix, and the grayscale feature vector becomes 128-dimensional, which is to unify the dimension alignment of the feature space).

[0083] The multiple frequency band energy feature vector sets are mapped through a fully connected layer to form a high-dimensional multiple frequency band energy feature vector set (i.e., if the multiple frequency band energy feature vector set itself has a dimension of 64, it is mapped to 128 dimensions through a fully connected layer for dimension alignment of the feature space).

[0084] Global average pooling is used to average the feature vectors of adjacent key joints of the topological constraint feature tensor to obtain a high-dimensional topological constraint feature tensor (i.e., a 128-dimensional vector).

[0085] The gray-scale feature vector is horizontally concatenated with a high-dimensional set of multiple frequency band energy feature vectors and a high-dimensional topological constraint feature tensor to form a long vector (i.e., three 128-dimensional vectors are horizontally concatenated to form a 384-dimensional long vector, which is a coarse fused vector).

[0086] See Figure 4 An attention mechanism network is introduced, and the long vector is input into the attention mechanism network for weighted summation to obtain the arm movement feature vector set (that is, the attention mechanism network is introduced to perform weighted summation of the 384-dimensional long vector with assigned weights, which is to reduce the dimensionality of the 384-dimensional long vector to a 128-dimensional fused feature vector, which is the arm movement feature vector. The steps are repeated for video image data segments of each movement time period to obtain the arm movement feature vector set).

[0087] It should be noted that each motion cycle corresponds to a high-order fusion feature vector, which includes both local dynamics and system structure; the features have "structure awareness" and can distinguish between isolated anomalies and propagated anomalies, achieving adaptive fusion of multimodal and multi-level features; students will understand the power of graph structures, especially in their application in multi-joint and complex motion systems.

[0088] S25: Construct a probabilistic defect topology network using a neural network; input the arm running feature vector set into the probabilistic defect topology network to construct a defect triggering probability matrix; perform singular value decomposition on the defect triggering probability matrix to obtain a defect propagation path; construct a verification model using the defect propagation path;

[0089] It should be noted that by inputting the fused feature vector set of the robotic arm operation into a probabilistic defect topology network with uncertainty modeling capabilities, predicting defect probabilities, and then constructing a verifiable virtual environment in reverse, trainees will learn how to use neural networks and probabilistic modeling to improve the fault detection capabilities of robotic arm systems.

[0090] Specifically, such as Figure 5 As shown, in step S25, a probabilistic defect topology network is constructed using a neural network; the feature vector set of the arm's movement is input into the probabilistic defect topology network to construct a defect triggering probability matrix; singular value decomposition is performed on the defect triggering probability matrix to obtain a defect propagation path; and a verification model is constructed using the defect propagation path. The specific operation steps are as follows:

[0091] Steps S251-S253 mainly describe how the frequency band energy feature vector set corresponding to the length of each motion time cycle is input into the input layer of the neural network. After the time dependence analysis of the progressive deterioration trend of the LSTM hidden layer and Bayesian output layer of the neural network, a probabilistic defect topology network is obtained. The arm running feature vector set is input into the probabilistic defect topology network, and the defect triggering probability matrix is ​​output.

[0092] S251: The frequency band energy feature vector set corresponding to each motion time cycle length (i.e., historical mechanical data is divided into video image data segments according to the motion time cycle length, so each video image data segment corresponds to a motion time cycle length) is input into the input layer of the neural network using a neural network. The frequency band energy feature vector set corresponding to each motion time cycle length is mapped through a fully connected layer according to the time sequence to obtain a fixed field vector space.

[0093] It should be noted that in the extension of step S25, the probability of various defects occurring in each "node" (such as joints, actuators, etc.) of the robotic arm system is mainly predicted based on the above-mentioned arm operation feature vector set, and the key paths that defects may propagate in the system are identified; a probabilistic defect topology network is constructed using a neural network as a learning framework, which can model the temporal evolution law and uncertainty of defect occurrence and output the failure probability of each key joint under different defect types.

[0094] See Figure 6 A neural network consists of an input layer, an LSTM (Long Short-Term Memory) hidden layer, and a Bayesian output layer.

[0095] The input layer takes the frequency band energy feature vector set corresponding to each motion time cycle length of the historical mechanical data as input. The frequency band energy feature vector set is calculated in step S23 above. The frequency band energy can reveal the operational changes of information such as current. The frequency band energy reflects the energy distribution generated during the operation of the arm, such as current and joint angles. Therefore, by using the frequency band energy feature vector set as input analysis, it is mapped to a fixed-length vector space (that is, the role of the fixed-length vector space is to uniformly map the original data from different sources, with different structures and different durations to digital vectors of the same length and format, so that they can be input into neural network models for training and inference). Through practical examples, students are helped to understand how to train a model using historical data and use it for prediction.

[0096] S252: Memory units are set up using the hidden layer of the LSTM (Long Short-Term Memory) network of the neural network; the frequency band energy in the fixed field vector space corresponding to the length of each motion time cycle is recorded using the memory units, and the progressive deterioration trend of the key joint anomaly is analyzed; the temporal dependence of the anomaly is captured using the progressive deterioration trend.

[0097] It should be noted that when using LSTM (Long Short-Term Memory) networks to set memory units, such as 64 memory units, the memory unit represents that the layer has a 64-dimensional internal state space, which records the running state of the previous multiple cycles, determines whether there is a "gradual deterioration trend" (such as a continuous increase in current and increased vibration), and captures the time dependence of defect development (e.g., a joint overheats → which leads to abnormal torque in the next cycle).

[0098] S253: The Bayesian output layer of the neural network outputs the failure probability distribution of key joints based on the time dependence of the progressive deterioration trend, as a probabilistic defect topology network.

[0099] The feature vector set of arm movement is input into the probabilistic defect topology network to construct a defect triggering probability matrix;

[0100] It should be noted that the Bayesian output layer outputs a probability distribution, which represents the probability of failure of each joint under each defect type (i.e., current overload, joint wear, joint loosening, communication delay, angle deviation, etc.), generating the failure probability distribution of critical joints.

[0101] The probabilistic defect topology network outputs the failure probability of each joint under different defect types. It constructs a two-dimensional matrix (i.e., the defect trigger probability matrix) from the arm movement feature vector set corresponding to each motion time cycle length (i.e., video image data segment). For example, the defect trigger probability matrix is ​​m×n, where m is the number of nodes (e.g., critical joints: J1, J2, J3; driver 1; power module; communication interface, etc., a total of m=8); n is the number of defect types (e.g., current overload, joint wear, joint loosening, communication delay, angle deviation, etc., a total of n=6); each unit element Pij represents the predicted probability (between 0 and 1) of the i-th node experiencing the j-th defect.

[0102] Steps S254-S255 mainly describe the process of performing singular value decomposition on the defect triggering probability matrix to obtain a left singular vector matrix; and performing principal component analysis on the left singular vector matrix to determine the maximum singular value of the unit element in each column, thereby constructing the defect propagation path.

[0103] S254: Perform singular value decomposition on the defect triggering probability matrix, decomposing the defect triggering probability matrix into a left singular vector matrix, a singular value diagonal matrix, and a right singular vector matrix to obtain f maximum singular values;

[0104] The f maximum singular values ​​are sorted in descending order, and the feature vectors in the defect triggering probability matrix corresponding to the first half of the middle number of h maximum singular values ​​are selected for dimensionality reduction to obtain h principal components (i.e. the most important defect propagation modes).

[0105] It should be noted that the above defect triggering probability matrix is ​​decomposed using singular value decomposition to extract the most important defect propagation patterns (i.e., "principal components"), reducing the complexity of the analysis. The defect triggering probability matrix is ​​decomposed into three matrices through singular value decomposition: the left singular vector matrix (i.e., size m×r, describing the combination patterns between nodes), the singular value diagonal matrix (i.e., size r×r, representing the "energy" or importance of each pattern), and the right singular vector matrix (i.e., size r×n, describing the association patterns between defect types).

[0106] The defect triggering probability matrix is ​​decomposed using singular value decomposition, which is dimensionality reduction. Principal components are then analyzed, and in this scheme, the principal components represent the main defect propagation patterns (i.e., defect types). These defect propagation patterns have already been explained in step S253 and will not be repeated here. Specific examples are used to help trainees master how to use SVD to simplify complex matrices and extract the most important defect propagation patterns.

[0107] S255: Perform principal component analysis on the left singular vector matrix, filter the unit element of each column of the left singular vector matrix to find the maximum singular value, and analyze the time dependency of the unit element according to the topology of the robotic arm kinematic chain to construct the defect propagation path.

[0108] It should be noted that because the left singular vector matrix represents the combination pattern between nodes, that is, the combination of defects between the number of nodes and the number of defect types, it best describes the defect propagation type of the defect triggering probability matrix.

[0109] Furthermore, each column of the left singular vector matrix corresponds to a defect propagation mode. The maximum singular value represents the role played in the defect propagation mode, which is the most obvious defect. It is determined whether the maximum singular value of the unit element before and after is close to the maximum singular value of the column (i.e., by calculating the adjacent difference = and judging by it, which will not be elaborated further). Then, the nodes with the maximum singular value are connected through the topology diagram of the robotic arm kinematic chain in step S23 above (i.e., the defect may start from J1 and propagate to J2 through the drive chain). Then, time dependence (i.e., if a joint overheats, it will cause abnormal torque in the next cycle) analysis is performed, and finally the defect propagation path is formed.

[0110] S256: Construct an initial verification model; perform multimodal perturbation simulation on the initial verification model using the defect propagation path to obtain an updated verification model; collect virtual simulation data under the verification model to generate a virtual verification defect triggering probability matrix; verify the defect triggering probability matrix with the virtual verification defect triggering probability matrix to obtain the final verification model;

[0111] It should be noted that the purpose of this step is to create an accurate digital model based on the structure and behavior of the real system, providing a foundation for further verification and simulation; by simulating different types of disturbances (such as pressure, current, etc.), the digital twin environment is made closer to the potential problems of the real system; disturbance simulation is usually carried out according to the defect propagation path, with the aim of verifying how the system’s performance changes under different operating conditions. Simulating multimodal disturbances can reveal the system’s behavior under abnormal conditions and help identify potential defects or hidden dangers;

[0112] The accuracy and reliability of the verification model are evaluated by comparing the actual generated defect triggering probability matrix with the virtual generated probability matrix. If the two are in good agreement, it means that the verification model can truly reflect the occurrence of defects in the physical system. The purpose of verification is to ensure that the digital twin environment can truly simulate the behavior of the physical system and can correctly predict the probability of defect occurrence. This verifies the effectiveness of the entire verification model. By correcting or optimizing the model, a reliable and accurate verification model is finally obtained.

[0113] Specifically, such as Figure 7 As shown, in step S256, an initial verification model is constructed; multimodal perturbation simulation is performed on the initial verification model using the defect propagation path to obtain an updated verification model; virtual simulation data is collected under the verification model to generate a virtual verification defect triggering probability matrix; the defect triggering probability matrix and the virtual verification defect triggering probability matrix are used for verification to obtain the final verification model. The specific operation steps are as follows:

[0114] S2561: Obtain the 3D CAD model of the robotic arm, input the 3D CAD model of the real robotic arm into the simulation platform (i.e., the simulation system); define the key joint types and motion axes of the simulation platform according to the kinematic chain topology diagram of the robotic arm; and introduce physics engine parameters into the simulation platform, and construct an initial verification model through the physics engine parameters and the key joint types and motion axes.

[0115] It should be noted that the process involves acquiring a 3D CAD model of a real robotic arm, accurately reproducing its 3D geometry in a virtual network environment, and inputting it into a simulation platform (i.e., a simulation system) to ensure consistency in kinematics and spatial interference relationships. Furthermore, the robotic arm's kinematic chain topology diagram is used to define the key joint types and motion axes, indicating whether each joint is a rotational or translational joint, calibrating the rotational / translational axis direction (X / Y / Z axes) of each joint, and setting the motion range (e.g., J1: -180°~+180°), establishing correct parent-child connection relationships. This ensures that the simulation platform allows manual driving of each joint in the virtual environment, observing the naturalness of the overall motion, comparing it to the motion trajectory of a real robotic arm, and ensuring no stuttering, misalignment, or interference anomalies.

[0116] The physics engine parameters in the above steps include the mass properties of each link in the joint (such as the actual mass of each link, the inertia tensor (i.e., its ability to resist rotation)) and joint dynamic parameters (i.e., friction coefficient, damping coefficient, etc.). By introducing physics engine parameters, the initial verification model is given realistic "physical behavior," making its performance under force, acceleration, and collision close to that of a real device. Furthermore, trainees need to understand how to obtain an accurate 3D CAD model from the actual device (robotic arm) and build an accurate initial verification model in the simulation platform. The above system helps trainees understand how to define key parameters in the simulation system and compare the behavior of virtual and real robotic arms.

[0117] The following step S2562 mainly describes the process of inputting the defect propagation path into the initial verification model to perform multi-modal disturbance simulation of superimposed torque pulses, spatial current fields, and communication delays, updating the initial verification model, and obtaining a new verification model.

[0118] S2562: Input the defect propagation path into the initial verification model for analysis and multimodal perturbation simulation;

[0119] The superimposed torque pulses are used to simulate multimodal perturbations in the kinematic chains between key joints using the aforementioned kinematic chain topology diagram of the robotic arm;

[0120] The multimodal perturbation simulation is used to simulate the spatial current field of the environment in which the defect propagation path is simulated; the spatial current field is used to simulate the change of current over time in the robotic arm, and the influence of the change of current over time on the robotic arm is mapped to the key joints of the robotic arm.

[0121] Delay noise is introduced into the kinematic chain between the key joints to simulate the communication delay of the kinematic chain;

[0122] The initial verification model is updated using the superimposed torque pulse, the spatial current field, and the communication delay to form a new verification model;

[0123] It should be noted that the above steps analyze the multimodal disturbance simulation of the defect propagation path through the initial verification model, mainly analyzing and simulating complex and coupled interference factors in the real industrial environment; the multimodal disturbance simulation mainly includes the simulation of superimposed torque pulses, the simulation of the time change of the spatial current field, and the simulation of communication delay.

[0124] The superimposed torque pulses in the multimodal disturbance simulation above simulate collisions and sudden load changes in the kinematic chain between joints, verifying whether the defect of the robotic arm is due to external or internal interference. The aforementioned setting of the spatial current field is to simulate the temperature rise of the joints during long-term operation, mapping the current effect to key components. For example, as the motor resistance increases with the current, the temperature increases, or as the lubricating oil viscosity decreases, friction changes. The delay noise simulation introduces fixed delays, random jitter, and data packet loss into the kinematic chain of the robotic arm to simulate abnormal behavior, such as control command lag, asynchronous arm movement, and increased vibration, thereby testing and verifying whether the model can identify "pseudo-mechanical faults" caused by communication problems. In practical operation, the above steps guide students to conduct step-by-step experiments, simulating different types of disturbances (such as current changes, increased friction, etc.), observing their effects on the movement of the robotic arm, and recording the results.

[0125] S2563: Arrange virtual sensors on the verification model and use the virtual sensors to collect virtual simulation data (that is, this step S2563 is the same as step S1 above, both of which collect the actual operation data of the robotic arm, except that the data is virtual simulation data and is the same as the real data).

[0126] The complete operation process of steps S21-S24 above is reconstructed to finally obtain the virtual verification arm running feature vector set.

[0127] It should be noted that after the verification model is corrected and tested through multimodal perturbation simulation, step S1 is re-implemented under the verification model to collect the sequence of operating parameters of the robotic arm. The entire process of steps S21-S24 (i.e., the complete process of obtaining the feature vector set of the arm's operation) is repeated in the verification model. Finally, the virtual verification arm's operating feature vector set is obtained. This step S2563, which obtains the virtual verification arm's operating feature vector set, is to verify the verification model through the collected virtual simulation data to ensure the accuracy of the results in the real scene. Students manually operate the simulation platform, set up and collect data, and perform real-time data analysis to help them understand the connection between simulation and actual operation.

[0128] S2564: Using the virtual verification arm to run the feature vector set, the virtual verification defect topology network is used to generate a virtual verification defect trigger probability matrix (i.e., this step S2564 is the same as the above steps S251-S253, and will not be repeated).

[0129] It should be noted that the above-mentioned probabilistic defect topology network is used to verify the feature vector set of the virtual verification arm, which is to verify the probability of the occurrence of the defect propagation path. For example, if the friction of the J2 joint is increased in the simulation, and the model significantly increases the probability of the "wear" type defect in the defect propagation path on J2 in the virtual verification defect trigger probability matrix, then the verification is successful.

[0130] In steps S2563-S2564 above, the acquisition of virtual simulation data, the obtained virtual verification arm operation feature vector set, and the final virtual verification defect trigger probability matrix are all completed under the verification model. This ensures that each operation in the operation process can complete the simulation to verify the probability of the defect type being triggered. Based on this, it helps students understand how to use the probabilistic defect topology network to generate the defect trigger probability matrix and verify the virtual behavior of the robotic arm.

[0131] S2565: Align the defect triggering probability matrix with the virtual verification defect triggering probability matrix by dimension and calculate the KL divergence;

[0132] A preset divergence threshold is set; it is then determined whether the KL divergence is less than the divergence threshold.

[0133] If so, it is determined that the virtual verification defect triggering probability matrix under the verification model is consistent with the result of the defect triggering probability matrix finally obtained by collecting the running parameters in step S1, and the closed-loop operation ends (that is, it indicates the accuracy of the verification of the verification model, the closed-loop operation ends, and the model prediction is considered reliable).

[0134] It should be noted that the defect triggering probability matrix is ​​the real operating parameter data collected through step S1, while the virtual verification defect triggering probability matrix is ​​the virtual real operating parameter data collected; both are obtained by outputting through the probabilistic defect topology network, and both can obtain the defect propagation path. The defect propagation path reflects the probability of the type of defect that occurs in the operation of the robotic arm, so the two matrices are verified.

[0135] The alignment dimension ensures that the order of rows (nodes) and columns (defect types) in the two matrices is consistent. If there is a slight mismatch (such as a certain type of defect not being simulated in the virtual environment), interpolation or masking is performed, and finally, the KL divergence is calculated. KL divergence is a mathematical tool that measures the "information difference" between two probability distributions. The larger the KL divergence value, the more inconsistent the results of the virtual verification of the verification model in the real scene are with the actual collected running parameters (i.e., the defect trigger probability matrix finally obtained in step S1). Through the actual calculation examples of the above steps, students are helped to understand how KL divergence is used to verify the credibility of simulation results, and it is demonstrated how to judge the accuracy of the model through the divergence threshold.

[0136] Example 2

[0137] like Figure 8 As shown, the present invention also provides a feasibility assessment system for programming information of industrial control systems, including: a data acquisition module 10; an analysis module 20; and an identification module 30.

[0138] The acquisition module 10 is used to acquire the time series and operating parameter series of the historical acquisition robot arm using multi-source sensors, and to construct synchronized historical mechanical data.

[0139] The analysis module 20 is used to determine the motion time cycle length of the historical mechanical data, segment the historical mechanical data according to the motion time cycle length to obtain multiple video image data segments; perform graph neural network analysis on the multiple video image data segments to obtain a set of arm movement feature vectors; and use the set of arm movement feature vectors to perform neural network analysis to construct a verification model.

[0140] The identification module 30 is used to collect real-time data of the robotic arm, input the real-time data of the robotic arm into the verification model, and generate the running results.

[0141] This invention also provides a method and system for feasibility assessment of programming information in industrial control systems. It extracts frequency band energy features from historical mechanical data (segmented into video image data segments and corresponding motion cycles); constructs a probabilistic defect topology network using a neural network (including fully connected layers, LSTM hidden layers, and Bayesian output layers); outputs a defect trigger probability matrix, and then extracts the main defect propagation modes through singular value decomposition (SVD); finally, it constructs a verification model based on the defect propagation modes and obtains the final verification model through multimodal perturbation simulation. Thus, it utilizes existing mechanical system and simulation models to construct an initial verification model as the basis for the verification model; the initial verification model is then subjected to virtual simulation data and multimodal simulation perturbation to construct a verification model in actual engineering, ensuring the effectiveness and accuracy of the target verification model. This provides a technical foundation and guarantee for subsequent acquisition of real-time data from the robotic arm, inputting the real-time data from the robotic arm into the verification model, and generating operating results.

[0142] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; those skilled in the art can modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some or all of the technical features; however, these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for assessing the feasibility of programming information in an industrial control system, characterized in that, The following steps are included: By using multi-source sensors to acquire historical data on the robotic arm's time series and operating parameter sequences, synchronized historical mechanical data can be constructed. Based on the rotation angle changes of key joints in the historical mechanical data, the motion time cycle length required for the robotic arm to complete one repetitive motion cycle is determined; the historical mechanical data is segmented according to the motion time cycle length to obtain multiple video image data segments; graph neural network analysis is performed on the multiple video image data segments to obtain a set of feature vectors for arm operation; neural network analysis is performed using the set of feature vectors for arm operation to construct a verification model for simulating and evaluating the operating performance of the robotic arm; Real-time data from the robotic arm is collected and input into the verification model to generate operational results for evaluating the feasibility of the programming information. The historical mechanical data is segmented according to the length of the motion time cycle to form multiple video image data segments; The video image data segments are converted into two-dimensional images to form grayscale images, and the frequency band energy is calculated using wavelet packet decomposition to obtain multiple frequency band energy feature vector sets. The arm's operational feature vector set is used for neural network analysis to construct a validation model. The specific steps are as follows: A probabilistic defect topology network is constructed using a neural network; the feature vector set of the arm's movement is input into the probabilistic defect topology network to construct a defect triggering probability matrix; singular value decomposition is performed on the defect triggering probability matrix to obtain the defect propagation path; based on the defect propagation path, a verification model is constructed by performing multimodal perturbation simulation on the initial verification model; A probabilistic defect topology network is constructed using a neural network; the feature vector set of the arm's movement is input into the probabilistic defect topology network to construct a defect triggering probability matrix. The specific operation steps are as follows: By using a neural network to input the frequency band energy feature vector set corresponding to the length of each motion time cycle into the input layer, and through the time dependence analysis of the progressive deterioration trend of the LSTM hidden layer and Bayesian output layer of the neural network, a probabilistic defect topology network is obtained; the arm running feature vector set is input into the probabilistic defect topology network, and the defect triggering probability matrix is ​​output. The defect triggering probability matrix is ​​subjected to singular value decomposition to obtain the defect propagation path. The specific steps for constructing a verification model using the defect propagation path are as follows: The defect triggering probability matrix is ​​subjected to singular value decomposition to obtain a left singular vector matrix; principal component analysis is performed on the left singular vector matrix to obtain the maximum singular value of the unit element in each column, and the defect propagation path is constructed. Build the initial validation model; The initial verification model is subjected to multimodal perturbation simulation using the defect propagation path to obtain an updated verification model; virtual simulation data is collected under the verification model to generate a virtual verification defect triggering probability matrix; the defect triggering probability matrix and the virtual verification defect triggering probability matrix are used for verification to obtain the final verification model.

2. The method for feasibility assessment of programming information in an industrial control system according to claim 1, characterized in that, The specific steps for determining the motion time cycle length from the historical mechanical data are as follows: Using the historical mechanical data, the joint angles of key joints of the robotic arm are randomly extracted, and the rotation angle change value of the joint angle is calculated. The delayed time points of joint angle rotation at adjacent time points are filtered by a preset rotation angle change threshold to form a time delay curve. The corresponding delay time point of the maximum delay time point in the time delay curve and the maximum delay time point in the next time point are selected as the motion time cycle length.

3. The method for feasibility assessment of programming information in an industrial control system according to claim 2, characterized in that, Historical mechanical data is segmented based on the length of the motion time cycle to obtain multiple video image data segments; graph neural network analysis is performed on these multiple video image data segments to obtain a set of arm movement feature vectors; neural network analysis is then performed using the arm movement feature vector set to construct a validation model. The specific operation steps are as follows: Connect adjacent key joints in the historical mechanical data to construct a kinematic chain topology graph of the robotic arm; use a graph neural network to perform graph convolution operation on the multiple frequency band energy feature vector sets to obtain a topological constraint feature tensor. By weighting the grayscale image with multiple frequency band energy feature vector sets and topological constraint feature tensors, the arm operation feature vector set is obtained.

4. The method for feasibility assessment of programming information in an industrial control system according to claim 3, characterized in that, The specific steps for building the initial validation model are as follows: A 3D CAD model of the robotic arm is obtained and input into a simulation platform; key joint types and motion axes are defined on the simulation platform based on the kinematic chain topology diagram of the robotic arm. Physics engine parameters are introduced into the simulation platform, and an initial verification model is constructed using these parameters, along with key joint types and motion axes.

5. The method for feasibility assessment of programming information in an industrial control system according to claim 4, characterized in that, The initial verification model is subjected to multimodal perturbation simulation using the aforementioned defect propagation path to obtain an updated verification model. The specific operation steps are as follows: The defect propagation path is input into the initial verification model to perform multimodal disturbance simulation of superimposed torque pulse, spatial current field and communication delay, and the initial verification model is updated to obtain a new verification model.

6. The method for feasibility assessment of programming information in an industrial control system according to claim 5, characterized in that, Virtual simulation data is collected under the verification model to generate a virtual verification defect triggering probability matrix; the defect triggering probability matrix and the virtual verification defect triggering probability matrix are used for verification to obtain the final verification model. The specific operation steps are as follows: Virtual sensors are deployed on the verification model, and virtual simulation data is collected using the virtual sensors. The complete operation process of obtaining the arm running feature vector set in the above steps is repeated to reconstruct the final virtual verification arm running feature vector set. The virtual verification arm utilizes the feature vector set to input the probability defect topology network and generates a virtual verification defect trigger probability matrix. Align the defect triggering probability matrix with the virtual verification defect triggering probability matrix by dimension and calculate the KL divergence. A preset divergence threshold is set; it is then determined whether the KL divergence is less than the divergence threshold. If so, it is determined that the results of the virtual verification defect trigger probability matrix and the defect trigger probability matrix under the verification model are consistent, and the closed-loop operation ends.

7. A feasibility assessment system for programming information in an industrial control system, characterized in that, include: Data acquisition module; Analysis module; Recognition module; The acquisition module is used to acquire the time series and operating parameter series of the historical acquisition robot arm using multi-source sensors, and to construct synchronized historical mechanical data; The analysis module determines the motion cycle length required for the robotic arm to complete one repetitive motion cycle based on the rotation angle changes of key joints in the historical mechanical data; it segments the historical mechanical data into multiple video image data segments based on the motion cycle length; it performs graph neural network analysis on the multiple video image data segments to obtain a set of arm operation feature vectors; and it uses the set of arm operation feature vectors to perform neural network analysis to construct a verification model for simulating and evaluating the performance of the robotic arm. The identification module is used to collect real-time data of the robotic arm, input the real-time data of the robotic arm into the verification model, and generate running results for evaluating the feasibility of the programming information. The historical mechanical data is segmented according to the length of the motion time cycle to form multiple video image data segments; The video image data segments are converted into two-dimensional images to form grayscale images, and the frequency band energy is calculated using wavelet packet decomposition to obtain multiple frequency band energy feature vector sets. The arm's operational feature vector set is used for neural network analysis to construct a validation model. The specific steps are as follows: A probabilistic defect topology network is constructed using a neural network; the feature vector set of the arm's movement is input into the probabilistic defect topology network to construct a defect triggering probability matrix; singular value decomposition is performed on the defect triggering probability matrix to obtain the defect propagation path; based on the defect propagation path, a verification model is constructed by performing multimodal perturbation simulation on the initial verification model; A probabilistic defect topology network is constructed using a neural network; the feature vector set of the arm's movement is input into the probabilistic defect topology network to construct a defect triggering probability matrix. The specific operation steps are as follows: By using a neural network to input the frequency band energy feature vector set corresponding to the length of each motion time cycle into the input layer, and through the time dependence analysis of the progressive deterioration trend of the LSTM hidden layer and Bayesian output layer of the neural network, a probabilistic defect topology network is obtained; the arm running feature vector set is input into the probabilistic defect topology network, and the defect triggering probability matrix is ​​output. The defect triggering probability matrix is ​​subjected to singular value decomposition to obtain the defect propagation path. The specific steps for constructing a verification model using the defect propagation path are as follows: The defect triggering probability matrix is ​​subjected to singular value decomposition to obtain a left singular vector matrix; principal component analysis is performed on the left singular vector matrix to obtain the maximum singular value of the unit element in each column, and the defect propagation path is constructed. Build the initial validation model; The initial verification model is subjected to multimodal perturbation simulation using the defect propagation path to obtain an updated verification model; virtual simulation data is collected under the verification model to generate a virtual verification defect triggering probability matrix; the defect triggering probability matrix and the virtual verification defect triggering probability matrix are used for verification to obtain the final verification model.

Images