An intelligent single-axis robot abnormal monitoring system

Abstract

This invention relates to the field of single-axis robot technology, and more particularly to an intelligent single-axis robot anomaly monitoring system. The system first acquires mechanism dynamics model data and simultaneously collects real-time monitoring data from the transmission system and the multi-force sensor nodes at the motion execution head. Second, it performs collision identification processing on the data, generating a collision anomaly feature distribution map matching the model's spatial position. Subsequently, combining the map with long-stroke motion constraints, it dynamically and visually reconstructs the dynamics model, outputting an interface with anomaly level markers. Finally, it associates and maps the visualization interface with the real-time data to generate emergency protection and maintenance strategies, triggering a robot protection response through an industrial control port. This invention achieves high-precision positioning and hierarchical visualization of collisions at the motion execution head of a long-stroke single-axis robot, constructs a closed-loop intelligent protection mechanism, and significantly improves equipment safety protection capabilities and operational efficiency.

Description

Technical Field

[0001] This invention relates to the field of single-axis robot technology, and in particular to an intelligent single-axis robot anomaly monitoring system. Background Technology

[0002] With the increasing demand for high-precision and high-efficiency processing in modern manufacturing, long-stroke single-axis robots are widely used in various mechanical equipment, automated equipment, and intelligent equipment in intelligent manufacturing (smart factories), including the manufacturing of 3C semiconductor equipment, display panels, photovoltaic energy storage equipment, new energy equipment, testing and inspection equipment, and so on. During equipment manufacturing, single-axis robots can perform stable laser cutting on the aforementioned equipment components to execute precision cutting tasks on complex contours. However, in actual operation, accidental collisions between the laser cutting head nozzle and the workpiece or fixture are one of the main failures affecting system stability and processing quality. Such collisions can not only damage expensive optical components and reduce cutting accuracy, but in severe cases, they can also cause equipment downtime, resulting in production interruptions and economic losses.

[0003] Traditional collision detection methods often rely on indirect sensing techniques such as motor current monitoring or encoder feedback, which struggle to accurately identify minor or oblique collisions and lack spatial localization capabilities, resulting in high false alarm and false negative rates. Furthermore, existing systems typically only perform simple emergency stops after an anomaly occurs, lacking detailed analysis of the collision location, type, and impact range, thus failing to provide effective support for subsequent emergency protection and maintenance decisions. Especially in long-stroke motion scenarios, factors such as uneven guide rail stiffness distribution and nonlinear friction in the transmission system further increase the complexity of collision identification.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide an intelligent single-axis robot anomaly monitoring system, which aims to solve the technical problems of existing long-stroke single-axis robots that rely on indirect sensing methods, resulting in low collision recognition accuracy, lack of spatial positioning capabilities, and inability to construct a closed-loop protection and maintenance decision-making mechanism based on refined analysis.

[0006] To achieve the above objectives, the present invention provides an intelligent single-axis robot anomaly monitoring system, the system comprising: The data synchronization module is used to acquire the dynamic model data of the motion actuator of the long-stroke single-axis robot, and simultaneously collect the real-time monitoring data set of multiple force sensing nodes of the single-axis robot transmission system and motion execution head. The map generation module is used to perform collision anomaly state identification processing on the real-time monitoring data set and generate a collision anomaly feature distribution map that matches the spatial location of the dynamic model data. The visual reconstruction module is used to perform dynamic visual reconstruction processing on the dynamic model data based on the collision anomaly feature distribution map and the preset single-axis long-stroke motion constraints, and generate a visual output interface with anomaly level markings. The strategy triggering module is used to associate and map the visualization output interface with the real-time monitoring data set, generate a collision emergency protection and maintenance feedback strategy, and trigger the protection response operation of the single-axis robot through the industrial control port.

[0007] Optionally, the step of acquiring the dynamic model data of the long-stroke single-axis robot motion actuator and simultaneously collecting real-time monitoring data sets of multiple force sensing nodes of the single-axis robot transmission system and motion execution head includes: The design parameter database of the long-stroke single-axis robot is called to extract the structural stiffness parameters, motion inertia parameters, and nonlinear friction parameters that constitute the dynamic model data; The structural stiffness parameters are used to describe the stress deformation characteristics of the long-stroke guide rail and the transmission mechanism. The nonlinear friction parameters include the friction compensation coefficients at different stroke positions and the installation coordinates and configuration parameters associated with the sensing nodes. According to the configuration parameters, an activation command is sent to each sensing node, and the end force sensing data, servo motor torque and current data and encoder position and speed data of each sensing node are collected according to the set sampling period. The end effector force sensing data, the servo motor torque and current data, and the encoder position and speed data are spatiotemporally aligned based on the sensor node installation coordinates and the stroke hierarchy of the dynamic model to generate a hierarchical real-time monitoring data set. The segmented structure of the hierarchical real-time monitoring data set is mapped to the stroke stiffness hierarchy of the long-stroke single-axis robot, and the stroke stiffness hierarchy is a hierarchical structure divided according to the force characteristics of the long-stroke structure.

[0008] Optionally, the step of performing collision anomaly state identification processing on the real-time monitoring data set to generate a collision anomaly feature distribution map that matches the spatial location of the dynamic model data includes: The end force sensing data in the hierarchical real-time monitoring data set is processed to detect abrupt change points, identify collision force abrupt change regions that exceed the dynamic fluctuation threshold, and extract the change rate and abrupt change amplitude of the collision force abrupt change regions. The trend deviation analysis is performed on the servo motor torque and current data after inertia compensation to determine the deviation between the actual torque and the expected torque of the preset model, and to mark the abnormal torque area where the deviation exceeds the dynamic threshold. The spatial coordinates of the collision force mutation region and the torque anomaly region are mapped to the corresponding positions in the dynamic model data to generate an initial collision anomaly feature distribution map containing collision type labels and anomaly levels. Based on the position offset alarm information in the encoder position and speed data, the initial collision anomaly feature distribution map is cross-validated to eliminate false alarm anomaly areas and correct the anomaly level, thereby generating the collision anomaly feature distribution map.

[0009] Optionally, the step of dynamically visually reconstructing the dynamic model data based on the collision anomaly feature distribution map and preset single-axis long-stroke motion constraints to generate a visual output interface with anomaly level labels includes: The connection relationship data of the transmission mechanism and the servo control parameter data are extracted from the dynamic model data to construct the motion constraint relationship network of the long-stroke single-axis robot. Based on the collision type label in the collision anomaly feature distribution map, assign a corresponding color code and dynamic flashing frequency to each anomaly region; The color code and dynamic flashing frequency are superimposed onto the corresponding spatial position of the dynamic model data to generate a first visualization structure; Based on the motion constraint relationship network, the abnormal regions in the first visualization structure are subjected to positioning accuracy impact diffusion simulation processing to generate a second visualization structure containing potential accuracy offset region markers. The second visualization structure is fused with the position and velocity data in the hierarchical real-time monitoring dataset to generate the visualization output interface.

[0010] Optionally, the step of associating and mapping the visualization output interface with the real-time monitoring data set to generate a collision emergency protection and maintenance feedback strategy includes: Extract the spatial coordinate set and collision type label of the abnormal region from the visualization output interface to generate a collision anomaly feature vector; Based on the collision anomaly feature vector, the corresponding protection priority scoring rules are matched from the preset strategy library to calculate the protection priority score for each anomaly region; Based on the protection priority score and the transmission connection tightness in the motion constraint relationship network, a collision emergency protection and maintenance feedback strategy is generated, which includes emergency response sequence, torque control scheme and resource allocation scheme; the emergency response sequence is a sequence sorted according to the priority score and transmission influence tightness.

[0011] Optionally, the protection response operation triggered by the industrial control port of the single-axis robot includes: The collision emergency protection and maintenance feedback strategy is converted into a real-time control instruction set, which includes a single-axis emergency stop instruction, a reverse torque unloading instruction, a data acquisition frequency adjustment instruction, and a maintenance work order generation instruction. The real-time control instruction set is sent to the servo control unit of the single-axis robot via an encrypted communication link, instructing the servo control unit to perform reverse torque unloading and emergency stop operations; The maintenance work order generation instruction is simultaneously pushed to the preset equipment maintenance management platform, and the work order execution status data returned by the maintenance management platform is received. The abnormal status markers in the visualization output interface are updated based on the work order execution status data to generate a maintenance progress tracking layer.

[0012] Optionally, the step of performing abrupt change point detection processing on the end force sensing data in the hierarchical real-time monitoring data set to identify collision force abrupt change regions exceeding the dynamic fluctuation threshold includes: The end force sensing data is divided into continuous force data segments according to a preset time window, and a first-order difference calculation is performed on each force data segment to obtain a force change rate sequence. Sliding window standard deviation analysis was performed on the force change rate sequence to identify abrupt change time points where the standard deviation exceeded the dynamic threshold; Extract the corresponding end force monitoring sub-data based on the mutation time point, and calculate the percentage offset between the end force monitoring sub-data and the model predicted force data; If the offset percentage exceeds a preset fluctuation threshold, the corresponding area of ​​the end force monitoring sub-data is marked as a collision force mutation area.

[0013] Optionally, the trend deviation analysis processing of the servo motor torque and current data after inertia compensation to determine the deviation between the actual torque and the expected torque of the preset model, and to mark the abnormal torque region where the deviation exceeds the dynamic threshold, includes: Obtain preset torque reference curves for the long-stroke single-axis robot at different stroke positions and different movement speeds. The preset torque reference curves include a segmented expected torque range that takes into account inertial force and friction. The servo motor torque and current data are categorized by stroke segment and speed segment, and the categorized torque and current data are subjected to trend deviation analysis based on the preset torque reference curve to generate a real-time torque trend line. Calculate the deviation between the real-time torque trend line and the corresponding preset torque reference curve, wherein the deviation includes the duration of deviation and the cumulative amount of deviation; If the duration of the deviation exceeds the first dynamic threshold and the cumulative amount of the deviation exceeds the second dynamic threshold, the corresponding travel segment spatial coordinates are mapped to the dynamic model data, and the abnormal torque region is marked according to the mapped spatial coordinates.

[0014] Optionally, the process of establishing the encrypted communication link includes: Assign a unique device identifier and dynamic encryption key to the servo control unit of each single-axis robot; Before sending the real-time control command set, a command header verification code is generated based on the identity identifier, and the command content is segmented and encrypted using the dynamic encryption key; the segmented and encrypted command content is combined with the command header verification code to generate an encrypted data packet, which is then sent to the servo control unit via a multipath transmission protocol. Receive the instruction confirmation signal returned by the target servo control unit, and update the effective status of the dynamic encryption key according to the instruction confirmation signal.

[0015] Optionally, the step of extracting the corresponding end force monitoring sub-data based on the mutation time point and calculating the percentage offset between the end force monitoring sub-data and the model predicted force data includes: The model predicted force data of the long-stroke single-axis robot in the current motion state is obtained. The model predicted force data is calculated and generated based on the structural stiffness parameters, nonlinear friction parameters and current encoder position and velocity data in the dynamic model data, and reflects the theoretical force state under no collision interference. A dynamic error compensation window is constructed, and the time width of the dynamic error compensation window is adaptively set according to the response lag characteristics of the long-stroke single-axis robot and the width of the preset time window. Within the dynamic error compensation window, the end force monitoring data is filtered to remove high-frequency noise components caused by mechanical vibration, resulting in smoothed measured force data. Calculate the absolute value of the difference between the smoothed measured force data and the model predicted force data at the corresponding time point, and then divide the absolute value of the difference by the model predicted force data to obtain the offset percentage.

[0016] In this invention's intelligent single-axis robot anomaly monitoring system, by integrating the dynamic model of a long-stroke single-axis robot with multi-node force sensing data, high-sensitivity and high-precision identification and spatial localization of head-end collision anomalies during single-axis robot motion execution are achieved. By generating a collision anomaly feature distribution map that matches the model space, the accuracy and interpretability of anomaly judgment are improved. Dynamic visualization reconstruction combined with motion constraints visually presents different levels of anomaly regions and their potential impact range, enhancing the monitorability of the system status. Finally, by associating the visualization interface with real-time data to generate a closed-loop emergency protection and maintenance feedback strategy and triggering corresponding control responses, the system significantly improves the safety protection capability, fault response efficiency, and intelligent operation and maintenance level of the equipment under sudden collision conditions, effectively reducing the risk of equipment damage and downtime, and ensuring the stability and processing quality of the single-axis robot's motion process. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the structure of the first embodiment of the intelligent single-axis robot anomaly monitoring system of the present invention; Figure 2 This is a flowchart illustrating the specific steps involved in generating a collision anomaly feature distribution map in the intelligent single-axis robot anomaly monitoring system of this invention.

[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] In one embodiment, such as Figure 1 As shown, an intelligent single-axis robot anomaly monitoring system is provided, the system comprising: The data synchronization module 10 is used to acquire the dynamic model data of the motion actuator of the long-stroke single-axis robot, and simultaneously collect the real-time monitoring data set of multiple force sensing nodes of the single-axis robot transmission system and motion execution head. The force sensing node can be a physical sensor unit deployed at the end of the motion execution head and the transmission system to directly measure contact force or torque. It can provide direct, high-sensitivity perception of collision events, overcoming the limitations of indirect monitoring methods. In one specific embodiment, the force sensing node detects mechanical deformation or force through strain gauges, piezoelectric effect, or optical interference principles. The dynamic model data can be the output of a mathematical model describing the physical characteristics of a long-stroke single-axis robot during motion, such as force, acceleration, and stiffness. It can provide a theoretical benchmark for anomaly identification and support spatial coordinate alignment and influence range assessment. Furthermore, the dynamic model data can be used to establish a parameterized model through offline identification and real-time table lookup, or to dynamically update model parameters through an online recursive algorithm. The long-stroke single-axis robot can be an automated actuator that achieves a large range of linear motion along a single axis. For example, in laser cutting, it can be used to carry the laser cutting head to complete complex contour processing tasks. In laser cutting, the motion execution mechanism can be a processing execution unit integrating a laser generator, focusing lens group, and nozzle, which can be used to perform high-precision material cutting. A single-axis robot transmission system can be a collection of mechanical transmission components that convert drive source power into linear motion. It can be used to transmit motion and force, affecting the system's stiffness and dynamic response characteristics. The motion execution head is the end effector of the motion execution mechanism that directly faces the workpiece. For example, in laser cutting, it may include a nozzle and protective cover, used to guide auxiliary gas and maintain optical path stability; it is a high-risk collision area. The real-time monitoring data set can be a stream of raw or pre-processed force / torque data synchronously collected by multiple force sensing nodes during operation. It can be used as direct input for anomaly identification.

[0021] The map generation module 20 is used to identify and process the collision anomaly state of the real-time monitoring data set and generate a collision anomaly feature distribution map that matches the spatial location of the dynamic model data. Among them, collision anomaly states can be deviations from normal operation caused by unexpected contact between the nozzle and the workpiece or fixture, and can be used as the basis for triggering system protection and maintenance logic. Spatial location can be the specific coordinate point or interval corresponding to the robot in the motion coordinate system, and can be used to geometrically align force data with the dynamic model. Collision anomaly feature distribution map can be a two-dimensional or three-dimensional feature map representing the intensity, direction, and confidence level of collision anomalies within the robot's motion space, and can be used to assign spatial semantics to anomaly events, supporting localization and classification.

[0022] The visual reconstruction module 30 is used to perform dynamic visual reconstruction processing on the dynamic model data based on the collision anomaly feature distribution map and the preset single-axis long-stroke motion constraints, and generate a visual output interface with anomaly level markings. The preset single-axis long-stroke motion constraints can be a set of boundary conditions describing the physical feasibility of the robot in long-stroke motion. These constraints can be used to limit the physical rationality of the visual reconstruction and avoid rendering invalid states. For example, the preset single-axis long-stroke motion constraints can include, but are not limited to, stroke limit constraints, maximum acceleration constraints, and guide rail stiffness partitioning constraints. Dynamic visual reconstruction processing can be the process of online updating and rendering of the robot model based on real-time anomaly information and motion constraints. This can be used to intuitively express the evolution of the state over time. The visual output interface can be a human-computer interactive graphical interface that presents the anomaly level, location, and impact range. This can be used to assist operators in quickly understanding the system state and supporting decision-making. Anomaly level markers can be symbols or color codes used to classify and identify the severity of collision anomalies. This can be used to distinguish response priorities and guide differentiated processing strategies.

[0023] The strategy triggering module 40 is used to associate and map the visualization output interface with the real-time monitoring data set, generate a collision emergency protection and maintenance feedback strategy, and trigger the protection response operation of the single-axis robot through the industrial control port.

[0024] The industrial control port can be a standard communication interface connecting the upper-level monitoring system and the lower-level actuators, used to transmit control commands to execute protection response operations. Protection response operations can be automated control actions performed to prevent equipment damage or escalation of accidents, and can be used to reduce the negative impact of collision consequences. For example, protection response operations can include, but are not limited to, local deceleration, path reversal, and zoned emergency stop. The collision emergency protection and maintenance feedback strategy can be a set of response measures automatically generated based on the type and level of the anomaly, used to achieve closed-loop management from detection to handling. Furthermore, the single-axis robot has a built-in 4G IoT card, which can alert the relevant after-sales service team via WeChat official account and SMS when an abnormal signal occurs, and via email, to perform pre-maintenance on the single-axis robot, avoiding equipment downtime or operation with defects. Associating the visual output interface with the real-time monitoring data set can establish a bidirectional index relationship between graphical elements and underlying sensor data. Furthermore, the visualization output interface can be associated and mapped with the real-time monitoring data set by binding the interface object and data record with a unique ID and using spatial coordinates to query the corresponding sensor node in reverse. This allows for the querying of the original data from the interface operation, forming a closed-loop feedback.

[0025] Taking the handling of sudden collision events in a sheet metal laser cutting production line as an example, the intelligent single-axis robot anomaly monitoring system in this embodiment can detect a sudden increase in tangential force when a ten-meter-long single-axis robot is performing stainless steel plate contour cutting, and the cutting head nozzle scrapes obliquely due to the loosening of the clamp. The force sensing nodes deployed on the nozzle base and guide rail slider synchronously detect the sudden increase in tangential force, and the data synchronization module aligns it with the output of the current dynamic model. The map generation module identifies the event as a moderate oblique collision and generates a high-confidence anomaly mark at X=3.2 meters. The visual reconstruction module combines the travel limit and stiffness partition constraints to highlight the area in yellow in the 3D interface and mark the potential guide rail wear risk. The strategy triggering module associates the mark with the real-time data, generates a command to locally decelerate and retreat to a safe point, sends it to the servo system via the EtherCAT port, and simultaneously pops up a maintenance prompt to check the clamp tightness to avoid a global shutdown.

[0026] In one embodiment, dynamic model data of the long-stroke single-axis robot motion actuator is acquired, and real-time monitoring data sets of multiple force sensing nodes of the single-axis robot transmission system and motion execution head are simultaneously collected, including: The design parameter database of the long-stroke single-axis robot is accessed to extract the structural stiffness parameters, motion inertia parameters, and nonlinear friction parameters that constitute the dynamic model data; The structural stiffness parameters are used to describe the stress deformation characteristics of the long-stroke guide rail and the transmission mechanism. The nonlinear friction parameters include the friction compensation coefficients at different stroke positions and the installation coordinates and configuration parameters associated with the sensing nodes. According to the configuration parameters, an activation command is sent to each sensing node, and the end force sensing data, servo motor torque and current data and encoder position and speed data of each sensing node are collected according to the set sampling period. The end effector force sensing data, the servo motor torque and current data, and the encoder position and speed data are spatiotemporally aligned based on the sensor node installation coordinates and the stroke hierarchy of the dynamic model to generate a hierarchical real-time monitoring data set. The segmented structure of the hierarchical real-time monitoring data set is mapped to the stroke stiffness hierarchy of the long-stroke single-axis robot, and the stroke stiffness hierarchy is a hierarchical structure divided according to the force characteristics of the long-stroke structure.

[0027] The design parameter database can be a set of engineering parameters storing the structural and kinematic characteristics of a long-stroke single-axis robot, providing a high-fidelity parameter source for building dynamic models. Structural stiffness parameters can be physical quantities describing the degree of deformation of the long-stroke guide rail and transmission mechanism under stress at different positions, characterizing the non-uniformity of the system's spatial stiffness and supporting accurate mechanical modeling. Motion inertia parameters can be physical quantities reflecting the mass distribution of the robot's moving parts and its influence on acceleration response, used to calculate dynamic loads and predict motion responses. Nonlinear friction parameters can be a set of compensation coefficients characterizing the changes in friction behavior of the transmission system at different stroke positions, used to correct dynamic deviations caused by nonlinear friction. The long-stroke guide rail can be a high-precision guiding structure supporting and guiding the linear motion of the single-axis robot, used to determine the system's stiffness distribution and motion stability. The transmission mechanism can be a mechanical transmission component that converts the output of the drive motor into linear displacement, used to transmit power and influence the system's dynamic characteristics. Stress-induced deformation characteristics can be the elastic or plastic deformation laws of the structure under external loads, used as the physical basis for stiffness parameters, affecting the accuracy of collision response modeling. The friction compensation coefficient can be a preset correction factor used to offset nonlinear friction effects, and can be used to improve the accuracy of the dynamic model under low-speed or reversing conditions. The sensor node mounting coordinates can be the fixed mounting position of the force sensor node in the robot's body coordinate system, and can be used to provide a geometric reference for the spatial alignment of multi-source data. Configuration parameters can be metadata describing the force sensor node's operating mode, communication address, and sampling settings, and can be used to support the initialization and cooperative control of the sensor network.

[0028] Accessing the design parameter database of a long-stroke single-axis robot can be achieved by accessing a local or remotely stored engineering parameter library to retrieve data related to the robot's structure and dynamics. Furthermore, accessing the design parameter database can be accomplished via API queries of structured databases or loading pre-stored XML or JSON format parameter files, thereby obtaining the fundamental parameters required for high-fidelity modeling. Extracting the structural stiffness parameters, motion inertia parameters, and nonlinear friction parameters that constitute the dynamic model data can be done by filtering and combining key physical parameters from the design parameter database to construct the dynamic model input. For example, the extraction of structural stiffness parameters, motion inertia parameters, and nonlinear friction parameters that constitute the dynamic model data can be achieved by extracting segmented stiffness parameters by stroke position index and dynamically selecting inertia parameter combinations according to load conditions, thereby generating a dynamic representation highly consistent with the actual physical system.

[0029] Sending activation commands to each sensor node according to configuration parameters can be achieved by parsing the communication address and working mode in the configuration parameters and issuing a start command to the corresponding node. In one specific embodiment, sending activation commands to each sensor node according to configuration parameters can be achieved by activating all nodes via Modbus broadcast or by activating specific nodes by addressing them one by one via CAN bus, thereby achieving accurate initialization and collaborative preparation of the sensor network. Collecting end-effector force sensing data, servo motor torque and current data, and encoder position and speed data from each sensor node according to a set sampling period can be achieved by periodically reading the outputs of multiple types of sensors under unified scheduling. Furthermore, collecting end-effector force sensing data, servo motor torque and current data, and encoder position and speed data from each sensor node according to a set sampling period can be achieved by using FPGA hardware timers to trigger synchronous sampling or by using RTOS task scheduling to achieve software periodic acquisition, thereby forming a multi-dimensional fused raw monitoring data stream.

[0030] End-effector force sensing data can be raw data of contact force or torque directly measured by the force sensing node at the action execution head, providing direct sensing evidence for collision events. Servo motor torque and current data can be converted current signal values ​​reflecting the motor's output torque, used to assist in judging internal anomalies or load changes in the transmission system. Encoder position and velocity data can be real-time displacement and velocity information calculated by the position encoder, providing motion state references for spatiotemporal alignment and model verification. Spatiotemporal alignment processing is the process of unifying multi-source heterogeneous sensor data into the dynamic model coordinate system in time and space, eliminating data bias caused by differences in sensor deployment and ensuring consistency in subsequent analysis. Hierarchical real-time monitoring datasets can be structured sets of multi-source sensor data organized according to stroke stiffness levels, enabling anomaly identification to have spatial partitioning semantics and improving positioning accuracy.

[0031] The end-effector force sensing data, servo motor torque and current data, and encoder position and speed data are spatiotemporally aligned based on the sensor node installation coordinates and the stroke hierarchy of the dynamic model. This can be achieved by using spatial transformations based on the installation coordinates and dividing the time-series data into segments according to the stroke stiffness hierarchy. For example, this spatiotemporal alignment can be achieved by mapping the force data to a global coordinate system using a homogeneous transformation matrix and segmenting the current and position data according to stiffness zones, thus maintaining consistency between heterogeneous data in physical space and model hierarchy. Generating a hierarchical real-time monitoring dataset can be achieved by organizing the spatiotemporally aligned multi-source data into structured data packets according to the stroke stiffness hierarchy. In one specific embodiment, generating a hierarchical real-time monitoring dataset can be achieved by constructing a JSON data structure with hierarchical labels and generating a hierarchical HDF5 dataset, thus providing a spatially semantic data foundation for subsequent anomaly identification.

[0032] The stroke stiffness hierarchy can be a hierarchical structure of stiffness intervals divided according to the stress characteristics of long-stroke structures, which can be used to provide a physical partitioning basis for data alignment and anomaly assessment. The interval hierarchy can be a set of ordered sub-intervals with different mechanical properties, dividing a continuous stroke into an ordered set of intervals, which can be used to support hierarchical modeling and partition diagnosis. Dividing the stroke stiffness hierarchy according to the stress characteristics of long-stroke structures can be achieved by automatically segmenting based on stiffness thresholds set according to finite element simulation results and dividing intervals based on wave-forming of measured deformation curves, thereby establishing an analysis partitioning framework that matches the physical system.

[0033] For example, in the scenario of parameter initialization and data acquisition during the startup phase of a long-stroke laser cutting equipment, the intelligent single-axis robot anomaly monitoring system of this embodiment can, after the system is powered on, first call the design parameter database to extract the structural stiffness parameters of the guide rail in three segments: 0-3 meters, 3-7 meters, and 7-10 meters, as well as the nonlinear friction compensation coefficients at the corresponding positions; according to the pre-stored configuration parameters, send activation commands to the three force sensing nodes installed on the nozzle base, the slider sidewall, and the motor output end respectively; then, synchronously acquire the three-dimensional force, servo current, and encoder speed data of the end effector at a sampling period set of 1 kHz; the spatiotemporal alignment processing module uses the installation coordinates of each node to convert the force data to the robot base coordinate system, and classifies it into the corresponding stiffness level interval according to the current stroke position; the finally generated hierarchical real-time monitoring data set contains independent data subsets of the three stiffness segments, and each subset integrates force, current, and motion state information, providing precise partitioned input for subsequent map generation.

[0034] In one embodiment, collision anomaly state identification processing is performed on the real-time monitoring dataset to generate a collision anomaly feature distribution map that matches the spatial location of the dynamic model data, including: The system performs abrupt change detection processing on the end force sensing data in the hierarchical real-time monitoring dataset, identifies collision force abrupt change regions that exceed the dynamic fluctuation threshold, and extracts the change rate and abrupt change amplitude of the collision force abrupt change regions.

[0035] The dynamic fluctuation threshold can be an adaptively adjusted force mutation judgment boundary based on real-time background noise or system operating status. This can be used to avoid misjudgments or missed judgments caused by stiffness or disturbance differences in different segments of a long stroke due to a fixed threshold. In this embodiment, the dynamic fluctuation threshold can be updated online in conjunction with the local statistical characteristics of the current motion phase. The collision force mutation region can be a continuous time-space segment in the end-effector force sensing data that meets the mutation condition, and can be used to identify the interval where potential collision events occur. Furthermore, the collision force mutation region can reflect the spatiotemporal performance of external contact events on the force signal. The rate of change can be the speed at which the collision force rises or falls per unit time, and can be used to distinguish between different types of contact, such as instantaneous hard collisions and slow compression. In a specific embodiment, the rate of change can be obtained by differentiating the mutation region. The mutation amplitude can be the peak increment of the collision force relative to the background value, and can be used to reflect the collision energy intensity and assist in anomaly level determination. Furthermore, the mutation amplitude can be normalized based on the local steady-state background.

[0036] Abrupt change detection processing of end-effector force sensing data in a hierarchical real-time monitoring dataset can be performed by applying mathematical transformations or statistical methods to identify non-stationary abrupt changes in the force signal. Furthermore, this operation can be achieved by using wavelet transform to detect abrupt changes and employing the CUSUM algorithm for online change point detection, thereby capturing the instantaneous mechanical response caused by minor or oblique collisions. Identifying collision force abrupt change regions exceeding a dynamic fluctuation threshold can be achieved by comparing the amplitude of the abrupt change point with an adaptive threshold to filter valid abnormal intervals. Further, this operation can be achieved by setting a dynamic threshold based on local standard deviation and loading a preset threshold table according to stiffness hierarchy, thereby excluding normal disturbance interference and retaining true collision candidates. Extracting the rate of change and amplitude of collision force abrupt change regions can be achieved by performing differentiation and extreme value analysis on the abrupt change regions to obtain dynamic characteristic parameters. Further, this operation can be achieved by estimating the rate of change using the five-point difference method and extracting the amplitude of the abrupt change using a peak hold circuit, thereby supporting collision type differentiation and severity assessment.

[0037] Trend deviation analysis is performed on the servo motor torque and current data after inertia compensation to determine the deviation between the actual torque and the expected torque of the preset model, and to mark the abnormal torque areas where the deviation exceeds the dynamic threshold.

[0038] Inertia compensation can be the process of subtracting the inertial component caused by acceleration and deceleration from the servo motor torque and current data. This can be used to separate external collision loads from motion inertia effects, improving the accuracy of torque anomaly identification. In one specific embodiment, inertia compensation can calculate and subtract the inertial torque based on the current acceleration and known moment of inertia. The preset model expected torque can be the theoretical motor output torque calculated based on the dynamic model under the current motion state, and can be used as a benchmark reference for judging whether the actual torque is abnormal. Furthermore, the preset model expected torque can be solved in real time based on joint position, velocity, and acceleration. Deviation can be the quantitative difference between the actual torque and the preset model expected torque, and can be used to characterize the intensity of the impact of external disturbances or collisions on the transmission system. In one specific embodiment, the deviation can be calculated using a statistical difference index within a sliding time window. The torque anomaly region can be the spatiotemporal interval where the servo motor torque deviates from the expected value beyond a dynamic threshold, and can be used to reveal unexpected load changes occurring inside or at the far end of the transmission system. Furthermore, the torque anomaly region can be mapped to a specific segment on the robotic arm's spatial path.

[0039] Trend deviation analysis of servo motor torque and current data after inertia compensation can be performed by first deducting the inertial component and then analyzing the long-term deviation trend between the remaining torque and the model's expectations. Further, this operation can be achieved by using recursive least squares to estimate the inertial term online and by identifying pre-compensated inertia in the model offline, thereby identifying hidden collisions or jamming in the upstream of the transmission chain. Determining the deviation between the actual torque and the torque expected by the preset model can be achieved by calculating the difference index between the two within a time window. Further, this operation can be achieved by calculating the root mean square deviation and then subtracting after dynamic time warping alignment, thereby quantifying the impact of external disturbances on the system. Marking torque anomaly regions where the deviation exceeds a dynamic threshold can be achieved by comparing the deviation sequence with an adaptive threshold and marking the abnormal time period and corresponding spatial location. Further, this operation can be achieved by using moving percentiles to set the dynamic threshold and adjusting the threshold sensitivity in conjunction with stiffness levels, thereby forming an anomaly spatial mapping in the torque dimension.

[0040] The spatial coordinates of the collision force mutation region and torque anomaly region are mapped to the corresponding positions in the dynamic model data to generate an initial collision anomaly feature distribution map containing collision type labels and anomaly levels.

[0041] Spatial coordinate mapping can be a process of transforming the physical location corresponding to the sensing data to a unified coordinate system of the dynamic model, which can be used to achieve spatial alignment and fusion of multi-source anomalies. In one specific embodiment, spatial coordinate mapping can transform the end effector or joint coordinates to the base coordinate system based on the robot's forward kinematics. Collision type labels can be classification identifiers of the physical properties of collision events, which can be used to support the generation of differentiated response strategies. Furthermore, collision type labels can be automatically assigned based on the dynamic characteristic combination of force and torque. The initial collision anomaly feature distribution map can be a preliminary anomaly spatial distribution map that integrates force mutation and torque anomaly information but has not been verified, which can be used as the input basis for cross-validation. In one specific embodiment, the initial collision anomaly feature distribution map expresses the anomaly spatial distribution in a gridded or node-based form.

[0042] Mapping the spatial coordinates of collision force mutation regions and torque anomaly regions to their corresponding positions in the dynamic model data can be achieved by using the sensor node mounting coordinates and the robot's current position to project both types of anomalies into the model space. Furthermore, this operation can be implemented by mapping to the base coordinate system using a homogeneous transformation matrix and directly indexing the model mesh using travel parameters, thus enabling the fusion of multi-source anomalies in physical space. Generating an initial collision anomaly feature distribution map containing collision type labels and anomaly levels can be achieved by integrating the mapped force and torque anomaly information and assigning types and levels based on feature combinations. Further, this operation can be achieved by matching type labels based on a rule engine and outputting anomaly levels through a lightweight classification model, thus forming a preliminary but structured spatial representation of anomalies.

[0043] Based on the position offset alarm information in the encoder position and speed data, the initial collision anomaly feature distribution map is cross-validated to eliminate false alarm anomaly areas and correct the anomaly level, thereby generating the collision anomaly feature distribution map.

[0044] The position offset alarm information can be an abnormal signal triggered when the actual position detected by the encoder deviates significantly from the command trajectory, and can be used to provide a third-dimensional verification basis independent of force and torque. In one specific embodiment, the position offset alarm information can be activated when the position error exceeds a preset tolerance or the cumulative deviation exceeds the limit. False alarm anomaly regions can be areas marked in the initial map but determined to be non-real collisions through cross-validation, and can be used to remove them to improve the reliability of the final map. Furthermore, false alarm anomaly regions usually originate from sensor signal disturbances caused by non-collision factors. Cross-validation processing can be a mechanism that uses multi-source heterogeneous data to verify the consistency of preliminary anomaly results, and can be used to significantly reduce the false alarm rate and improve the accuracy of anomaly levels. In one specific embodiment, cross-validation processing requires that the force, torque, and position signals have coordinated response characteristics in time and space.

[0045] Based on the position offset alarm information in the encoder position and velocity data, cross-validation is performed on the initial collision anomaly feature distribution map. This can involve checking whether each anomaly region in the initial map is accompanied by a position change signal. Furthermore, this operation can be implemented by setting time windows to align the position offset and force change, and using logical AND gates to determine valid anomalies, thus introducing an independent validation dimension to improve robustness. False alarm anomaly regions are removed and anomaly levels are corrected. This can involve removing anomaly markers without position offset support and adjusting the level of supported regions. Further, this operation can be achieved by completely removing regions without validation support and downgrading some supported regions, thus outputting a highly reliable final anomaly map. The collision anomaly feature distribution map is generated, which can output the final anomaly spatial distribution results after cross-validation. Further, this operation can be achieved by generating a map with confidence weights and outputting a structured JSON format anomaly description, thus providing reliable input for the visualization and strategy modules.

[0046] Taking the accurate identification and false alarm filtering of oblique scraping collisions as an example, the intelligent single-axis robot anomaly monitoring system in this embodiment can detect that when the cutting head slightly scrapes the edge of the fixture at X=5.1 meters, the nozzle base force sensor detects that the tangential force increases by 8N within 2 milliseconds, and the rate of change exceeds the dynamic fluctuation threshold of this stiffness range; at the same time, the servo motor torque shows a continuous positive deviation of 0.3N·m after inertial compensation, exceeding the dynamic threshold; the system maps the two anomalies to the same spatial position in the dynamic model, initially marking them as lateral scraping, with an anomaly level of medium; however, the encoder does not report a significant positional shift, and the cross-validation module determines that the event is high-frequency vibration interference and removes the anomaly area; when a real hard collision occurs at X=2.8 meters, the force change, torque deviation, and position jump occur simultaneously, the system retains this area and raises the anomaly level to high, and the final generated collision anomaly feature distribution map only contains the latter, effectively suppressing false alarms.

[0047] In one embodiment, based on the collision anomaly feature distribution map and preset single-axis long-stroke motion constraints, the dynamic model data is dynamically visualized and reconstructed to generate a visualization output interface with anomaly level labels, including: Data on the connection relationships of the transmission mechanism and servo control parameters are extracted from the dynamic model data to construct a motion constraint network for a long-stroke single-axis robot.

[0048] The transmission mechanism connection relationship data can be topological information describing the physical connections and motion transmission relationships between various mechanical components of a long-stroke single-axis robot, which can be used to support the construction of a constraint network reflecting the real mechanical coupling path. In an exemplary embodiment, the transmission mechanism connection relationship data can be obtained directly by reading the connection relationships through the model metadata interface. The servo control parameter data can be a set of configuration parameters characterizing the response characteristics and control logic of the servo system, which can be used to model the suppression or amplification effect of the control loop on abnormal disturbances. For example, the servo control parameter data can be obtained by back-deriving servo parameters from symbolic dynamic equations. The motion constraint relationship network can be a graph structure model composed of transmission connection relationships and servo control parameters, characterizing the feasible region of system motion and disturbance propagation path, which can be used to provide a topological and dynamic basis for the simulation of accuracy impact diffusion. In this embodiment, the motion constraint relationship network can be constructed using a graph database to store node and edge relationships.

[0049] Extracting transmission mechanism connection data and servo control parameter data from dynamic model data can involve analyzing the internal structure of the dynamic model and separating the mechanical topology and subset of control parameters. Furthermore, this operation can be achieved by directly reading connection relationships through the model metadata interface and inferring servo parameters from symbolic dynamic equations, thus providing the necessary input for constructing a motion constraint network. Constructing a motion constraint network for a long-stroke single-axis robot can be achieved by integrating transmission connection relationships and servo control parameters into an attributed graph structure. Further, this operation can be achieved by using a graph database to store node and edge relationships and constructing an adjacency matrix to represent the constraint network, thus achieving the technical effect of establishing a system-level constraint model required for disturbance propagation analysis.

[0050] Based on the collision type label in the collision anomaly feature distribution map, assign a corresponding color code and dynamic flashing frequency to each anomaly region; Color coding can be a visual color identifier mapped according to collision type or anomaly level, which can be used to intuitively distinguish the severity of anomalies. In one specific embodiment, color coding can be assigned based on HSV color space mapping level and type. Dynamic flashing frequency can be the time frequency of the anomaly area flashing in the visualization interface, which can be used to convey the urgency of the event or the priority of intervention. For example, dynamic flashing frequency can be achieved by controlling the independent flashing of different areas through a timer. According to the collision type label in the collision anomaly feature distribution map, a corresponding color code and dynamic flashing frequency are assigned to each anomaly area, which can be achieved by converting the type label into visual parameters through table lookup or rule matching. Furthermore, this operation can be achieved by mapping level and type based on HSV color space and controlling the independent flashing of different areas through a timer, thereby achieving the technical effect of giving the anomaly area a human-computer readable semantic expression. Superimposing the color code and dynamic flashing frequency onto the corresponding spatial position of the dynamic model data can be achieved by applying visual attributes to a specified coordinate area in a 3D model or 2D projection. Furthermore, this operation can be achieved by dynamically rendering colors and flashing using a shader program and binding area attributes through UI controls, thereby achieving the technical effect of generating anomaly visualization with basic semantics.

[0051] The color code and dynamic flashing frequency are superimposed onto the corresponding spatial position of the dynamic model data to generate a first visualization structure; The first visualization structure can be a preliminary visualization model that only overlays color coding and flicker frequency without considering the diffusion of effects. This model can be used to present the spatial distribution and basic semantics of the original anomaly. In an exemplary embodiment, the first visualization structure can be generated by exporting an OBJ format material model. Generating the first visualization structure can be achieved by outputting a preliminary visualization model with applied color and flicker. Furthermore, this operation can be implemented by exporting an OBJ format material model and generating a WebGL renderable scene, thereby achieving the technical effect of providing an intuitive display of the anomaly's location and type.

[0052] Based on the motion constraint relationship network, the abnormal regions in the first visualization structure are subjected to positioning accuracy impact diffusion simulation processing to generate a second visualization structure containing potential accuracy offset region markers. The simulation of the diffusion of positioning accuracy impacts can be a simulation process that extrapolates the impact range of collision disturbances on downstream processing accuracy based on a motion constraint relationship network. It can be used to identify indirect impact areas that may cause processing errors even without direct collisions. In one specific embodiment, the simulation of the diffusion of positioning accuracy impacts can be implemented using the finite element substructure method to simulate deformation transmission. Potential accuracy offset area markers can be non-collision area identifiers that may cause the actual trajectory to deviate from the command due to structural deformation or control disturbances. These markers can be used to expand the scope of anomaly impact assessment and support preventative maintenance. For example, potential accuracy offset area markers can be presented by marking the offset areas with semi-transparent color blocks. The second visualization structure can be an enhanced visualization model that integrates the original anomaly and the potential accuracy offset area, and can be used to comprehensively reflect the direct and indirect consequences of collisions. In one exemplary embodiment, the second visualization structure can be constructed by delineating the impact range with dashed boundaries.

[0053] Based on a motion constraint network, the impact of abnormal regions in the first visualization structure on positioning accuracy is simulated. This can be achieved by propagating disturbances along the constraint network path and calculating the expected offset of downstream locations. Furthermore, this operation can be implemented by simulating deformation transmission using the finite element substructure method and simulating servo disturbance propagation using a state-space model, thereby achieving the technical effect of identifying indirectly affected regions and expanding the scope of anomaly assessment. A second visualization structure containing markers of potential accuracy offset regions is generated, which can be achieved by adding visual identifiers for indirectly affected regions to the first structure. Further, this operation can be achieved by marking offset regions with semi-transparent color blocks and delineating the impact range with dashed boundaries, thereby achieving the technical effect of comprehensively presenting the combined impact of collisions on machining accuracy.

[0054] The second visualization structure is fused with the position and velocity data in the hierarchical real-time monitoring dataset to generate the visualization output interface.

[0055] Layer fusion processing can be a rendering method that overlays and synthesizes static visualization structures and dynamic motion state data as layers, and can be used to jointly express abnormal states and real-time operational situations. In a specific embodiment, layer fusion processing can be performed by using arrow icons to represent the direction and magnitude of velocity. Layer fusion processing of the second visualization structure and the position and velocity data in the hierarchical real-time monitoring dataset can be performed by overlaying the current cutting head position and velocity as a dynamic layer onto the static anomaly model. Furthermore, this operation can be achieved by using arrow icons to represent the direction and magnitude of velocity and using trajectory prediction lines to display the path for the next few seconds, thereby achieving the technical effect of correlated display of abnormal risks and real-time motion states. Generating a visualization output interface can be achieved by integrating all layers and rendering a human-computer interaction interface. Furthermore, this operation can be achieved by generating an embedded HMI interface and outputting it to a remote monitoring web terminal, thereby achieving the technical effect of providing a comprehensive situational awareness view that is monitorable and decision-making-ready.

[0056] Taking the risk assessment and dynamic monitoring of machining accuracy after lateral scraping as an example, the intelligent single-axis robot anomaly monitoring system in this embodiment can be as follows: When lateral scraping occurs at X=4.2 meters, the system assigns a yellow code and mid-frequency flashing according to the collision type label; the motion constraint relationship network simulation shows that the local bending of the guide rail at this point will cause a decrease in positioning stiffness within 1.5 meters downstream, and the system marks X=4.2 to 5.7 meters as a potential accuracy offset area and covers it with light orange semi-transparent; at the same time, the layer fusion module shows the current cutting head with a green arrow indicating that it is moving towards the high-risk area at a speed of 0.8 m / s; the operator can not only see the "collision position" in the visualization output interface, but also clearly identify the "about to enter the affected area", thereby triggering path replanning in advance and avoiding continued processing in the area of ​​impaired accuracy.

[0057] In one embodiment, the visualization output interface is mapped to a real-time monitoring data set to generate a collision emergency protection and maintenance feedback strategy, including: Extract the spatial coordinate set and collision type label of the abnormal area from the visualization output interface to generate a collision anomaly feature vector; Based on the collision anomaly feature vector, the corresponding protection priority scoring rules are matched from the preset strategy library to calculate the protection priority score for each anomaly area; Based on the protection priority score and the transmission connection tightness in the motion constraint relationship network, a collision emergency protection and maintenance feedback strategy is generated, which includes emergency response sequence, torque control scheme and resource allocation scheme; the emergency response sequence is a sequence after being sorted by priority score and transmission influence tightness.

[0058] The set of spatial coordinates of the abnormal regions can be a list of the position coordinates of all abnormal regions marked in the visualization output interface within the robot's motion space, providing precise spatial positioning for policy generation. In an exemplary embodiment, the set of spatial coordinates of the abnormal regions can obtain a structured abnormal description by parsing the data binding layer or graphical metadata of the visualization interface. The collision anomaly feature vector can be a structured numerical vector composed of spatial coordinates, collision type labels, and other semantic attributes, used to transform visualization information into computational input that can be processed by the policy engine. For example, the collision anomaly feature vector can be implemented by encoding the extracted spatial coordinates and collision type labels into a vector of a unified format. The preset policy library can be a knowledge base storing emergency rules and response templates corresponding to various collision scenarios, supporting rapid matching and policy generation. The protection priority scoring rule can be a set of quantitative rules for calculating the processing priority of abnormal regions based on the collision feature vector, used to achieve objective ranking of anomaly severity.

[0059] Protection priority score, calculated using scoring rules, represents the urgency of handling abnormal areas and can serve as a core criterion for emergency response prioritization. Transmission connection tightness is a quantitative indicator of the mechanical or control coupling strength between nodes in a motion constraint network, reflecting the breadth and depth of an anomaly's potential impact on other parts of the system. Emergency response sequence is an execution sequence for handling anomalies determined by combining priority scores and transmission connection tightness, guiding orderly handling procedures under multi-point anomalies. Torque control scheme is a servo motor torque adjustment strategy tailored to a specific anomaly area, used to achieve local load suppression or disturbance compensation. Resource allocation scheme is a scheduling plan for system computing, power, or maintenance resources during multiple concurrent anomalies, optimizing emergency response efficiency and system availability.

[0060] Extracting the spatial coordinate set and collision type labels of the abnormal region from the visualization output interface can be achieved by parsing the data binding layer or graphic metadata of the visualization interface to obtain a structured anomaly description. Furthermore, this operation can be implemented by retrieving the original anomaly record using the graphic object ID and extracting attributes by traversing the marked region using spatial indexing, thus achieving the conversion of graphic semantics into machine-readable data. Generating a collision anomaly feature vector can be achieved by encoding the extracted spatial coordinates and collision type labels into a vector of a unified format. Further, this operation can be achieved by using one-hot encoding to represent the type labels and concatenating coordinate values, and using embedded vector mapping to map type semantics, thus providing standardized input for policy matching. Matching the corresponding protection priority scoring rule from a pre-defined policy library based on the collision anomaly feature vector can be achieved by using the feature vector as the key to retrieve applicable scoring rules from the policy library. Further, this operation can be achieved by matching based on a rule condition tree and using similarity retrieval to return the closest rule, thus achieving context-adaptive rule invocation.

[0061] Calculating the protection priority score for each anomalous region can be achieved by substituting feature vectors into a matching scoring rule and outputting a numerical priority. Furthermore, this operation can be implemented by executing a weighted summation formula to calculate the score or by calling a lightweight neural network to infer the score, thereby quantifying the urgency of anomaly handling. Based on the protection priority score and the tightness of transmission connections in the motion constraint relationship network, a collision emergency protection and maintenance feedback strategy including emergency response sequence, torque control scheme, and resource allocation scheme is generated. This can be achieved by fusing priority and tightness for multi-objective optimization, outputting a structured strategy package. Further, this operation can be implemented by using the analytic hierarchy process (AHP) to determine the response sequence and generate control schemes by looking up tables, and by simultaneously optimizing the three sub-strategies using a constraint programming solver, thereby achieving a precise, coordinated, and orderly emergency response.

[0062] Taking intelligent emergency dispatching under multiple concurrent anomalies as an example, the intelligent single-axis robot anomaly monitoring system in this embodiment can simultaneously detect a frontal hard collision (high priority score) at X=2.1 meters and a slight scratch (low score) at X=8.5 meters. However, the motion constraint relationship network shows that X=2.1 meters is close to the servo motor, with extremely high transmission connection tightness, while X=8.5 meters is at the end of the stroke, with low tightness. The strategy generation module integrates both and still prioritizes X=2.1 meters as the first response order, and assigns it a torque control scheme that immediately reduces torque to zero and a resource scheme that prioritizes power supply to the drive. For X=8.5 meters, delayed processing is arranged, and only low-speed reversal is activated. The final collision emergency protection and maintenance feedback strategy avoids a global emergency stop, only intervenes in high-risk points locally, and ensures that the remaining areas can continue processing.

[0063] In one embodiment, triggering a protection response operation of a single-axis robot via an industrial control port includes: The collision emergency protection and maintenance feedback strategy is converted into a real-time control instruction set, which includes single-axis emergency stop instruction, reverse torque unloading instruction, data acquisition frequency adjustment instruction, and maintenance work order generation instruction. The real-time control instruction set can be a structured set of instructions converted from collision emergency protection and maintenance feedback strategies and executed by the underlying controller. It can be used to accurately translate high-level strategies into physical actions. In this embodiment, the real-time control instruction set serves as an intermediate carrier for strategy execution, forming a collaborative control link with the servo control unit and equipment maintenance management platform. The single-axis emergency stop instruction can be a safety control signal that forces the single-axis robot to stop immediately, preventing further escalation of collision consequences. In an exemplary embodiment, the single-axis emergency stop instruction acts on the motion system by cutting off servo enable or triggering an electrical motor shutdown. The reverse torque unloading instruction can be a control command that instructs the servo motor to apply torque opposite to the current load direction to release residual stress, preventing secondary damage or deformation accumulation due to continuous stress on the structure. In a specific embodiment, the reverse torque unloading instruction achieves active unloading by adjusting the current vector direction. The data acquisition frequency adjustment instruction can be a configuration command that dynamically modifies the sampling rate of force sensing nodes or encoders, used to increase local data density after an anomaly to support fine-grained diagnosis. In this embodiment, the data acquisition frequency adjustment instruction is associated with the spatial location of the abnormal area for targeted enhancement of perception capabilities. A maintenance work order generation instruction can be a data packet that triggers the equipment maintenance management platform to create a structured maintenance task, and can be used to automate the initiation of manual or automated maintenance processes. In one exemplary embodiment, the maintenance work order generation instruction encapsulates the fault type, location coordinates, and suggested handling measures.

[0064] Converting collision emergency protection and maintenance feedback strategies into real-time control instruction sets can be achieved by parsing the emergency response sequence, torque control scheme, and resource allocation scheme within the strategy and mapping them to specific instruction formats. Furthermore, this conversion can be accomplished by using an instruction template engine to populate parameters and generate the instruction set, and by calling a protocol adapter to serialize the strategy object into a binary instruction stream. This allows for a structured transformation from strategy to executable actions.

[0065] The real-time control command set is sent to the servo control unit of the single-axis robot through an encrypted communication link, instructing the servo control unit to perform reverse torque unloading and emergency stop operations; The encrypted communication link can be a communication channel that uses industrial-grade security protocols to ensure the security and integrity of control command transmission, preventing commands from being interfered with, tampered with, or lost. In this embodiment, the encrypted communication link serves as a reliable medium for command transmission, forming a secure closed loop with the servo control unit. The servo control unit can be a low-level motor drive controller that receives and executes motion and protection commands, and can be used to directly control the servo motor to achieve physical actions such as emergency stop and torque adjustment. In a specific embodiment, the servo control unit integrates current loop, speed loop, and position loop control logic. The emergency stop operation can be a rapid braking process executed by the servo control unit after responding to an emergency stop command, used to terminate dangerous movements in the shortest possible time. In this embodiment, the emergency stop operation and the reverse torque unloading command are executed in tandem to balance safety and structural protection.

[0066] Sending real-time control commands to the servo control unit of a single-axis robot via an encrypted communication link can be achieved by securely transmitting the command set using an industrial safety protocol and then transmitting it to the servo driver via a real-time bus. Furthermore, sending real-time control commands to the servo control unit of the single-axis robot via an encrypted communication link can be achieved by transmitting EtherCATAoE messages encapsulated in TLS and using a secure PLC to relay encrypted commands, thereby ensuring reliable delivery of control commands in complex electromagnetic environments. Instructing the servo control unit to perform reverse torque unloading and emergency stop operations can be achieved by the servo control unit parsing the commands and simultaneously executing reverse torque output and motion cut-off. Furthermore, instructing the servo control unit to perform reverse torque unloading and emergency stop operations can be achieved by first performing a 50ms reverse unloading operation and then triggering an emergency stop, or by executing unloading and stopping in parallel to shorten the response time, thereby proactively alleviating structural stress and terminating dangerous movements.

[0067] The maintenance work order generation instruction is pushed to the preset equipment maintenance management platform in a synchronous manner, and the work order execution status data returned by the maintenance management platform is received. The equipment maintenance management platform can be an information system used to receive, dispatch, track, and archive equipment maintenance tasks, and can be used to achieve data integration between OT and IT systems. In this embodiment, the equipment maintenance management platform serves as the management center for the work order lifecycle, establishing a bidirectional data channel with the industrial control port. The work order execution status data can be structured feedback information returned by the maintenance platform regarding the current processing stage of the work order, which can be used to support dynamic updates and closed-loop verification of abnormal states. In an exemplary embodiment, the work order execution status data includes a timestamp, responsible person, and current step identifier.

[0068] The maintenance work order generation command is synchronously pushed to the preset equipment maintenance management platform. This can be achieved by sending the work order data to the maintenance system via API or message queue. Furthermore, synchronously pushing the maintenance work order generation command to the preset equipment maintenance management platform can be achieved by calling a RESTful API to submit a JSON work order or publishing an MQTT message to a subscribed topic on the maintenance platform, thereby automatically initiating subsequent manual or automated maintenance processes. Receiving the work order execution status data returned by the maintenance management platform can be achieved by listening to the platform's callback interface or polling its status interface to obtain the latest progress. Furthermore, receiving the work order execution status data returned by the maintenance management platform can be achieved by receiving status updates in real time via a WebSocket long connection or by periodically pulling work order status snapshots, thereby establishing a closed-loop feedback channel for operation and maintenance.

[0069] Update the abnormal status markers in the visualization output interface based on the work order execution status data, and generate a maintenance progress tracking layer.

[0070] The abnormal status marker can be a graphic identifier in the visualization interface that marks the current progress of abnormal handling, reflecting the entire lifecycle status from "occurrence" to "resolution". In this embodiment, the color and icon of the abnormal status marker change dynamically with the work order status. The maintenance progress tracking layer can be a dynamic information layer superimposed on the visualization output interface to display the work order execution progress, which can be used to enhance the transparency and collaboration efficiency of operation and maintenance. In one embodiment, the maintenance progress tracking layer is semi-transparently overlaid on the 3D model of the equipment. Updating the abnormal status marker in the visualization output interface according to the work order execution status data can be achieved by mapping the work order status to the corresponding color, icon, or text label and refreshing the interface. Furthermore, updating the abnormal status marker in the visualization output interface according to the work order execution status data can be achieved by using a state machine to drive the marker style switching and binding the data model to automatically trigger UI redrawing, thereby realizing real-time visualization of the abnormal handling status. Generating the maintenance progress tracking layer can be achieved by superimposing a transparent layer containing the work order person in charge, the estimated completion time, and the current step on the original visualization structure. Furthermore, the generation of maintenance progress tracking layers can be achieved by displaying progress details in a floating panel and marking progress badges next to the 3D model, thereby providing the contextual information required for operation and maintenance collaboration.

[0071] For example, in scenarios involving closed-loop maintenance and transparent status after a collision, the intelligent single-axis robot anomaly monitoring system in this embodiment can generate a real-time control instruction set after the system detects a hard collision with the nozzle, including reverse torque unloading, emergency stop, high-frequency sampling, and work order creation. The instructions are sent to the servo control unit via EtherCAT through a safety protocol, and the motor completes unloading and stops within 8ms. At the same time, the maintenance work order generation instruction is pushed to the cloud CMMS platform via HTTPS and automatically assigned to the on-duty technician. After 30 minutes, the platform returns the work order status as "processing," and the system immediately updates the red alarm marker in the abnormal area to an orange "maintenance in progress" marker, and overlays a maintenance progress tracking layer with the technician's name and estimated completion time at X=3.7 meters. The operator can see "responded, under maintenance, expected to recover in 15 minutes" on the main control interface without the need for telephone confirmation, significantly improving collaboration efficiency.

[0072] In one embodiment, abrupt change point detection processing is performed on the end force sensing data in the hierarchical real-time monitoring dataset to identify collision force abrupt change regions exceeding the dynamic fluctuation threshold, including: The force sensing data at the end point is divided into continuous force data segments according to a preset time window, and a first-order difference calculation is performed on each force data segment to obtain a force change rate sequence. The preset time window can be a fixed or adjustable time interval used to segment the end-effector force sensing data. It can be used to divide a continuous signal into analyzable local segments, facilitating the capture of transient characteristics. The force data segment can be a continuous segment of end-effector force sensing data captured within the preset time window, serving as a basic processing unit for first-order difference and wave analysis. For example, the force data segment can include a single-axis force data segment, a multi-dimensional force synthesis data segment, or a filtered force data segment. First-order difference calculation is a mathematical operation that performs difference operations on adjacent sampling points in the force data segment to estimate the rate of change. It can be used to transform the original force signal into a rate sequence reflecting dynamic abrupt changes. The force change rate sequence can be a data sequence characterizing the rate of force change over time, obtained from first-order difference calculation, and can be used to explicitly reveal the dynamic characteristics of instantaneous impact or contact events.

[0073] The end-effector force sensing data is divided into continuous force data segments according to a preset time window. This can be achieved by framing the continuous force signal at fixed time intervals. Furthermore, this operation can be implemented using non-overlapping equal-length windows or 50% overlapping sliding windows, thus providing structured input for subsequent differential and fluctuation analysis. First-order difference calculation is performed on each force data segment to obtain a force change rate sequence. This can be achieved by calculating the force value difference between adjacent sampling points to generate the change rate sequence. The formula for calculating the force change rate can be as follows: In the formula, The end force sensing data collected at time t serves as the basis for calculation, reflecting the magnitude of the force at the current moment; The end force sensing data collected at time t-1 is used as a reference value for the previous time step. A preset time window or sampling period is used to standardize time intervals and calculate the rate of change; The force change rate is calculated using first-order difference to identify abrupt changes in force data, aiding in determining the timing and severity of collisions. Furthermore, this operation can be improved by using central difference to enhance accuracy and by employing forward difference to support real-time processing, thereby achieving the technical effect of highlighting transient change characteristics and suppressing steady-state background interference.

[0074] Sliding window standard deviation analysis was performed on the force change rate sequence to identify abrupt time points where the standard deviation exceeded the dynamic threshold. Sliding window standard deviation analysis is a statistical analysis method that slides a fixed-length window across a force rate of change sequence and calculates the local standard deviation. It can be used to quantify local fluctuation intensity and identify anomalous abrupt change points. The abrupt change time point can be the timestamp corresponding to the sliding window standard deviation exceeding a dynamic threshold, which can be used to pinpoint the precise moment a potential collision occurs. Performing sliding window standard deviation analysis on a force rate of change sequence involves sliding a window across the rate sequence and calculating the standard deviation of the data within each window. Furthermore, this operation can be implemented by using a fixed-width window to calculate the standard deviation and employing exponentially weighted moving standard deviation, thereby achieving the technical effect of dynamically assessing the intensity of local fluctuations and adapting to noise levels in different travel segments. Identifying abrupt change time points where the standard deviation exceeds the dynamic threshold can be achieved by comparing the standard deviation of each window with an adaptive threshold and marking the exceeding position. Further, this operation can be implemented by setting a dynamic threshold based on the local mean plus three times the standard deviation and loading a preset threshold table according to the stiffness level, thereby achieving the technical effect of initially screening possible collision occurrence times.

[0075] Extract the corresponding end force monitoring sub-data based on the mutation time point, and calculate the percentage deviation between the end force monitoring sub-data and the model predicted force data; The end-force monitoring sub-data can be a segment of local end-force sensing data extracted around the abrupt change time point, which can be used to verify the deviation from the model prediction value. The model predicted force data can be the theoretical end-force value calculated based on the dynamic model under the same motion state, which can be used as a benchmark reference to determine whether the measured force is abnormal. The offset percentage can be the relative deviation ratio between the end-force monitoring sub-data and the model predicted force data, which can be used to measure the degree to which the measured value deviates from the theoretical expectation and assist in false alarm filtering. Extracting the corresponding end-force monitoring sub-data based on the abrupt change time point can be achieved by extracting a segment of raw force data centered on the abrupt change time point. Furthermore, this operation can be achieved by extracting 10ms of data before and after the abrupt change point, and extracting the complete pulse segment from the initial rise to the fall back to the baseline, thereby achieving the technical effect of obtaining local measured samples for model verification. Calculating the offset percentage between the end-force monitoring sub-data and the model predicted force data can be achieved by comparing the output of the measured sub-data and the model at the same time period and calculating the relative deviation. Furthermore, this operation can be achieved by calculating the percentage deviation between the sub-data mean and the predicted mean, and by using the integral area ratio to calculate the overall deviation. This achieves the technical effect of introducing a model consistency criterion and improving the reliability of anomaly detection. The formula for calculating the percentage deviation is as follows: In the formula, The data represents the actual force monitoring data collected at the end of the line, indicating the actual force situation detected. Force data predicted by the dynamic model, representing the theoretically expected value of normal force; The offset percentage is used to quantify the degree of deviation between the actual force data and the model prediction, and is used as the core basis for determining whether a collision force abrupt change region is identified.

[0076] If the offset percentage exceeds the preset fluctuation threshold, the corresponding area of ​​the end force monitoring sub-data will be marked as a collision force mutation area.

[0077] The preset fluctuation threshold can be a fixed or configurable threshold used to determine whether the percentage of offset constitutes a valid collision. It can be used in conjunction with the dynamic fluctuation threshold to form a static verification boundary in the dual criteria. If the percentage of offset exceeds the preset fluctuation threshold, the corresponding area of ​​the end force monitoring sub-data is marked as a collision force mutation area. This can be achieved by combining the mutation intensity and model deviation as dual conditions to confirm the valid collision area. Furthermore, this operation can be implemented by marking a collision only when both conditions are met simultaneously, and marking areas that partially meet the conditions as suspicious areas to be verified, thereby achieving a significant reduction in false alarms caused by vibration or inertia.

[0078] Taking the high-sensitivity identification of weak oblique scraping as an example, the intelligent single-axis robot anomaly monitoring system in this embodiment can be as follows: When the cutting head slightly scrapes the fixture at X=6.3 meters, it generates a tangential force pulse of about 5N, lasting for 4ms. The system segments the end force data into 2ms windows and obtains a sharp rate peak through first-order difference. The sliding window standard deviation analysis detects that the standard deviation suddenly increases to 4 times the normal value during this period, exceeding the dynamic threshold set based on the current stiffness section, and marks the abrupt change time point. Then, the force data of 3ms before and after this point is extracted and compared with the near-zero tangential force predicted by the dynamic model. The offset percentage reaches 98%, far exceeding the preset fluctuation threshold of 30%. The system finally confirms this area as the collision force abrupt change area, while the 2N vibration caused by the guide rail joint at the same time has a sudden increase in rate, but the offset percentage is only 15%, which is judged as a non-collision disturbance and filtered out.

[0079] In one embodiment, trend deviation analysis is performed on the servo motor torque and current data after inertia compensation to determine the deviation between the actual torque and the expected torque of the preset model, and abnormal torque regions where the deviation exceeds a dynamic threshold are marked, including: Obtain preset torque reference curves for long-stroke single-axis robots at different stroke positions and different motion speeds. The preset torque reference curves include a segmented expected torque range that takes into account inertial force and friction.

[0080] The preset torque reference curve can be a set of expected torque ranges, including the effects of inertial force and friction, pre-calculated based on a dynamic model under different stroke positions and motion speed combinations. This can be used to provide an adaptive judgment benchmark for measured torque, improving the accuracy of anomaly identification. In this embodiment, the preset torque reference curve can be obtained through offline modeling or online parameter identification, combining the structural stiffness distribution and friction characteristics of the long-stroke single-axis robot. The segmented expected torque range can be the upper and lower limits or distribution range of the theoretical torque within each interval after dividing the continuous working space into several intervals. This can be used to reflect the influence of non-uniform mechanical characteristics on torque under long stroke. In an exemplary embodiment, the segmented expected torque range can be divided according to the guide rail joint position, speed gradient change point, or friction coefficient abrupt change point. Obtaining the preset torque reference curves for different stroke positions and motion speeds of the long-stroke single-axis robot can be achieved by loading from a dynamic model database or calculating the expected torque range for each working condition combination online. Furthermore, this operation can be achieved by loading a pre-stored two-dimensional lookup table reference curve and generating a segmented benchmark in real time based on the current stiffness and friction parameters, thereby establishing a high-fidelity judgment benchmark with comprehensive working condition coverage.

[0081] The torque and current data of the servo motor are categorized by stroke segment and speed segment, and the categorized torque and current data are processed by trend deviation analysis based on the preset torque reference curve to generate a real-time torque trend line.

[0082] The stroke segment can be a robot motion range divided according to long-stroke structure or stiffness characteristics, and can be used as a spatial unit for torque data classification and benchmark matching. In one embodiment, the stroke segment can be defined based on differences in mechanical support strength or structural stiffness. The speed segment can be an operating condition range divided according to the magnitude of motion speed, and can be used to distinguish torque behavior patterns under different dynamic loads. Furthermore, the speed segment can be divided into different categories according to the motion state. Labeling and classification can be the process of tagging the servo motor torque current data with the stroke segment and speed segment to achieve accurate matching of data with corresponding preset benchmark curves. In this embodiment, labeling and classification can be completed by reading the robot's current position and speed information in real time. Generating a real-time torque trend line can be achieved by smoothing the raw torque data to extract the main trend. Furthermore, this operation can be achieved by applying a Savitzky-Golay filter to generate the trend line and using an exponential smoothing algorithm to update the trend in real time, thereby suppressing measurement noise and highlighting the actual load changes.

[0083] A real-time torque trend line can be a continuous torque change trajectory formed after smoothing or filtering the classified torque and current data. It can be used to eliminate high-frequency noise and highlight load trend deviations. In one specific embodiment, the real-time torque trend line can be converted from the original current signal using a specific filtering algorithm. Trend deviation analysis of the classified torque and current data based on a preset torque reference curve can be performed by comparing the measured torque with the corresponding operating condition reference in a time series to identify systematic deviations. Furthermore, this operation can be achieved by using residual sequence analysis and control chart statistical methods to detect out-of-control points, thereby separating normal dynamic response from abnormal load disturbances.

[0084] Calculate the deviation between the real-time torque trend line and the corresponding preset torque reference curve. The deviation includes the duration of the deviation and the cumulative amount of the deviation.

[0085] The duration of deviation can be the length of time the measured torque continuously exceeds a preset reference curve, which can be used to distinguish between transient disturbances and persistent abnormal loads. In an exemplary embodiment, the determination of the duration of deviation can introduce a fault-tolerant mechanism to avoid false triggering. The cumulative deviation can be the total integral or weighted sum of the torque exceeding the reference value during the deviation period, which can be used to quantify the energy intensity of the anomaly and assist in the severity assessment. Furthermore, the cumulative deviation can reflect the potential impact of the anomaly on system heat accumulation or structural stress. Calculating the deviation between the real-time torque trend line and the corresponding preset torque reference curve can be done by calculating the deviation between the trend line and the upper and lower limits of the reference curve point by point or segment by segment, and statistically analyzing the duration and cumulative amount. For example, the torque deviation calculation formula can be as follows: In the formula, The real-time torque trend line value at the i-th sampling point is used to reflect the current actual torque current state of the servo motor. The value of the preset torque reference curve corresponding to the i-th sampling point includes the theoretical expected torque range for compensation of inertial force and friction. The number of sampling points corresponding to the duration of deviation, or the integration time window, is used to define the time span for calculating the cumulative deviation. The cumulative deviation is used to measure the degree of torque anomaly by calculating the cumulative value of the deviation between the actual torque and the baseline curve, and is used to mark the torque anomaly region. Furthermore, this operation can be achieved by calculating the duration and integral area of ​​exceeding the upper limit, while simultaneously monitoring deviations on both the upper and lower sides, thereby quantifying the spatiotemporal characteristics of the anomaly.

[0086] If the duration of the deviation exceeds the first dynamic threshold and the cumulative amount of the deviation exceeds the second dynamic threshold, the spatial coordinates of the corresponding stroke segment are mapped to the dynamic model data, and the abnormal torque area is marked according to the mapped spatial coordinates.

[0087] The first dynamic threshold can be an adaptive time threshold used to determine whether the duration of the deviation constitutes an anomaly, and can be used to prevent short-term fluctuations from being misjudged as collisions. In one specific embodiment, the first dynamic threshold can be dynamically adjusted according to the stability of the current operating condition. The second dynamic threshold can be an adaptive amplitude threshold used to determine whether the cumulative deviation constitutes an anomaly, and can be used to ensure that the anomaly has sufficient energy significance. Furthermore, the second dynamic threshold can be associated with physical risk indicators. The spatial coordinates of the travel segment can be the position interval corresponding to the travel segment in the robot's global coordinate system, used as a geometric bridge for mapping to the dynamic model. The mapped spatial coordinates can be the coordinate representation of the travel segment corresponding to the torque anomaly in the unified space of the dynamic model, which can be used to achieve accurate positioning of the torque anomaly on the physical structure. In one specific embodiment, the mapped spatial coordinates can be obtained through coordinate transformation or interpolation algorithms.

[0088] If the duration of the deviation exceeds a first dynamic threshold and the cumulative deviation exceeds a second dynamic threshold, the spatial coordinates of the corresponding travel segment are mapped to the dynamic model data. This can be achieved by transforming the location range of the anomaly to the unified coordinate system of the model when both conditions are met. Furthermore, this operation can be implemented by directly using the start and end coordinates of the travel segment for mapping and by interpolation to obtain more refined anomaly center coordinates, thus providing a geometric basis for spatial positioning. Torque anomaly regions are marked based on the mapped spatial coordinates, which can be done by labeling the coordinate range as a torque anomaly region in the dynamic model space. Further, this operation can be achieved by highlighting the anomaly segment on the 3D model and inserting anomaly records with coordinates into the graph data structure, thereby forming a spatial anomaly representation that can be used for multi-source fusion.

[0089] Taking the detection of continuous torque anomalies caused by internal jamming in the transmission chain as an example, the intelligent single-axis robot anomaly monitoring system in this embodiment can detect situations where, when the robot reaches X = 4.5 meters (a low-stiffness transition zone), the slider experiences slight jamming due to insufficient lubrication of the guide rail, causing the servo motor output torque to remain 0.25 N·m higher than the baseline curve for 120 ms. The system categorizes this data into the corresponding stroke and speed segments based on the current position and speed (0.8 m / s), generating a real-time torque trend line. The system calculates that the deviation duration of 120 ms exceeds the first dynamic threshold of 100 ms under this condition, and the cumulative deviation of 30 mN·m·s exceeds the second dynamic threshold of 25 mN·m·s. The system maps the spatial coordinates of the stroke segment X = 4.3-4.7 meters to the dynamic model and marks this interval as a torque anomaly area. After cross-verification with information from the force sensing module that shows no sudden changes but slight positional shifts, the system determines that the anomaly is due to internal mechanical resistance rather than nozzle collision, triggering a maintenance prompt instead of an emergency stop.

[0090] In one embodiment, the process of establishing an encrypted communication link includes: Assign a unique device identifier and dynamic encryption key to the servo control unit of each single-axis robot; The device identifier can be a unique identification code assigned to the servo control unit of a single-axis robot. It can be used to achieve device-level authentication and command routing, preventing unauthorized node access or mis-sent commands. In an exemplary embodiment, the device identifier can be issued during system initialization or device registration in conjunction with a security server or local key management module, thereby establishing a one-to-one correspondence between the device and the identifier. The dynamic encryption key can be a symmetric encryption key that is periodically updated or changed based on events. It can be used to ensure the confidentiality of command content and resist replay attacks. For example, the dynamic encryption key can be a preset key read from a trusted platform module through a secure boot process, or a temporary key can be issued online by a central key distribution center. Assigning a unique device identifier and dynamic encryption key to each servo control unit of a single-axis robot can be achieved by issuing a unique identifier and an initial key through a security server or local key management module during system initialization or device registration. Furthermore, this operation can be implemented by reading a preset key from a trusted platform module through a secure boot process and issuing a temporary key online by a central key distribution center, thereby establishing a one-to-one binding relationship between the device identity and the key, laying the foundation for secure communication.

[0091] Before sending the real-time control command set, a command header verification code is generated based on the identity identifier, and the command content is encrypted in segments using a dynamic encryption key; The instruction header verification code can be an integrity check code generated based on the device identifier and instruction metadata, used to verify the legitimacy of the instruction source and that the header has not been tampered with. In this embodiment, the instruction header verification code can be generated using a hash or signature algorithm, combining the identifier and instruction metadata as input. The instruction content can be the core operation parameters and commands that need to be encrypted and transmitted in the real-time control instruction set, used to carry the execution semantics of the actual protection actions. Segmented encryption processing can be a method of dividing the instruction content into multiple data blocks and encrypting them separately, which can be used to reduce the impact of single-point decryption failure on the overall instruction and improve fault tolerance. Furthermore, segmented encryption processing can combine the current dynamic encryption key to independently apply encryption operations to each data block.

[0092] Before sending real-time control commands, a command header verification code is generated based on the identity identifier. This can be achieved by using the identity identifier and command metadata as input, and generating the verification code through a hash or signature algorithm. Furthermore, this operation can be implemented by using the HMAC algorithm combined with a shared key to generate the verification code, or by using a lightweight Ed25519 signature to generate a short verification code, thus ensuring the trustworthiness and tamper-proof nature of the command header. The command content is then segmented and encrypted using a dynamic encryption key. This can be done by dividing the command content into multiple blocks and encrypting each block using the current dynamic key. Further, this operation can be achieved by using AES-CTR mode to encrypt each segment independently, or by using ChaCha20 stream cipher to encrypt segments by offset, thereby improving encryption robustness and limiting the impact of single-point failures.

[0093] The segmented encrypted instruction content is combined with the instruction header verification code to generate an encrypted data packet, which is then sent to the servo control unit via a multipath transmission protocol. The encrypted data packet can be a transmission unit composed of a header verification code and segmented encrypted instruction content, and can be used as a basic payload unit for secure communication. In this embodiment, the encrypted data packet can be combined with a predefined protocol format to concatenate the verification code and the encrypted segment to form a complete transmission unit. The multipath transmission protocol can be a communication mechanism that redundantly sends the same data packet simultaneously through two or more physical or logical channels, which can be used to improve the reliability of instruction delivery under electromagnetic interference or link interruption. In one embodiment, the multipath transmission protocol can combine multiple independent communication channels to synchronously send the same encrypted data packet.

[0094] The segmented encrypted instruction content is combined with the instruction header verification code to generate an encrypted data packet. This can be achieved by concatenating the verification code and encrypted segment according to a predefined protocol format to form a complete transmission unit. Furthermore, this operation can be implemented using TLV (Type-Length-Value) format encapsulation and embedding with an industry-standard protocol frame structure, thereby constructing a structured and verifiable secure payload. The encrypted data packet is then sent to the servo control unit via a multipath transmission protocol, potentially simultaneously through multiple independent communication channels. Furthermore, this operation can be achieved by parallel transmission from the EtherCAT master station and a backup Modbus TCP channel, utilizing dual network interface card (NIC) binding for link redundancy, thus significantly improving the reliable delivery rate of instructions in harsh industrial environments.

[0095] Receive the instruction confirmation signal returned by the target servo control unit, and update the effective status of the dynamic encryption key according to the instruction confirmation signal.

[0096] The instruction confirmation signal can be a response message returned by the servo control unit after successfully receiving and parsing the instruction. It can be used to verify instruction delivery and trigger key status updates. In one specific embodiment, the instruction confirmation signal can be combined with a designated port or bus listening mechanism to capture the ACK or NACK response returned by the servo. The effective status of the dynamic encryption key can be a status marker indicating whether the current dynamic encryption key is available, awaiting update, or expired. It can be used to support key lifecycle management and secure rotation. Furthermore, the effective status of the dynamic encryption key can be switched by combining key usage records and confirmation feedback.

[0097] Receiving the instruction confirmation signal returned by the target servo control unit can be achieved by listening to a designated port or bus and capturing the ACK or NACK response returned by the servo. Furthermore, this operation can be implemented by polling the confirmation status register in real time via the bus and receiving UDP callback confirmation packets, thereby verifying whether the instruction has been correctly received and parsed. The validity status of the dynamic encryption key is updated based on the instruction confirmation signal. If the confirmation signal is valid, the current key is marked as used and preparation is made to switch to the next key. Furthermore, this operation can be implemented by using a counter mode to increment the key index with each confirmation, and by jointly determining key validity based on a time window and confirmation, thereby achieving closed-loop key rotation and replay protection.

[0098] For example, in scenarios requiring reliable and secure transmission of emergency commands in a high-interference environment, the intelligent single-axis robot anomaly monitoring system of this embodiment can be as follows: In a sheet metal workshop with strong electromagnetic interference, the system needs to send emergency stop and reverse unloading commands to the X-axis servo control unit; this unit is pre-configured with a unique device identifier DEV-X07 and an initial dynamic key K1; when the command is generated, the system calculates the HMAC verification code based on DEV-X07 and encrypts the command content into three segments using K1; the encrypted data packet is sent simultaneously through the EtherCAT main channel and the backup Ethernet channel; after the servo unit successfully receives the data on either channel, it verifies the header and ciphertext, executes the command, and returns an acknowledgment signal; after receiving the acknowledgment, the main control system immediately marks K1 as used and activates the next key K2 for subsequent commands; even if EtherCAT loses packets due to interference, the Ethernet channel still ensures command delivery, and the key rotation mechanism prevents attackers from replaying old commands.

[0099] In one embodiment, the corresponding end force monitoring sub-data is extracted based on the mutation time point, and the percentage offset between the end force monitoring sub-data and the model predicted force data is calculated, including: The model predicts the force data of the long-stroke single-axis robot under the current motion state. The model predicts the force data based on the structural stiffness parameters, nonlinear friction parameters and current encoder position and velocity data in the dynamic model data, and reflects the theoretical force state under no collision interference.

[0100] The theoretical force state can be the force distribution that the end effector of a long-stroke single-axis robot should exhibit based on its current motion state under conditions without external collision interference. This can serve as the physical basis for the model's predicted force data and as a benchmark reference for anomaly detection. Furthermore, obtaining the model's predicted force data for the long-stroke single-axis robot in its current motion state can be achieved by calling a dynamic model, inputting the current motion parameters, and outputting the theoretical end effector force. For example, this operation can be implemented through online real-time simulation to calculate the model's predicted force or by interpolation to obtain a pre-calculated predicted force, thereby generating a collision-free benchmark force that matches the actual working conditions. The model's predicted force data can be generated based on the structural stiffness parameters, nonlinear friction parameters, and current encoder position and velocity data in the dynamic model data. This can be achieved by substituting stiffness, friction, and other parameters along with the real-time position and velocity into the dynamic equations to solve for the theoretical force. In a specific embodiment, this operation can be achieved by using a recursive Newton-Euler algorithm to calculate or by using a state-space model to output the predicted force, ensuring that the predicted force reflects the physical characteristics of the current stroke position.

[0101] A dynamic error compensation window is constructed, and the time width of the dynamic error compensation window is adaptively set according to the response lag characteristics of the long-stroke single-axis robot and the width of the preset time window.

[0102] The dynamic error compensation window can be an adaptive time interval constructed around the abrupt change time point for noise suppression and signal alignment. It can be used to match the physical response hysteresis characteristics of the system, improving the comparability of measured and predicted data. The response hysteresis characteristics can be the time delay and dynamic transition features of the robot system from command input to the generation of a stable force output, and can be used to determine the appropriate scale of the dynamic error compensation window. The time width can be the duration of the dynamic error compensation window on the time axis, and can be used to control the data range for filtering and alignment.

[0103] Constructing a dynamic error compensation window can be achieved by expanding an adaptive time width forward and backward from the abrupt change time point to form an analysis interval. In an exemplary embodiment, this operation can be implemented by symmetrically expanding the window or using a forward-biased window to match the causal system response, thereby providing a physically reasonable data range for filtering and alignment. The time width of the dynamic error compensation window is adaptively set based on the response hysteresis characteristics of the long-stroke single-axis robot and the width of the preset time window. This can be achieved by combining the system's step response time constant with the original analysis window length to calculate the compensation width. For example, this operation can be implemented by making the time width equal to the maximum value of an empirical coefficient multiplied by the response time constant and the preset window width, or by retrieving a recommended width from a stiffness hierarchy database, ensuring that the window scale is consistent with the system's dynamic characteristics and avoiding over- or under-compensation.

[0104] Within the dynamic error compensation window, the end force monitoring data is filtered to remove high-frequency noise components caused by mechanical vibration, resulting in smoothed measured force data.

[0105] The filtering process involves frequency-selective smoothing of the end-effector force monitoring data. This can remove high-frequency vibration interference and retain valid collision signals. Mechanical vibration, caused by non-collision periodic or random force disturbances from the transmission system, guide rail joints, or external excitation, can constitute a major source of false alarms and needs to be suppressed through filtering. High-frequency noise components are interference elements in the force signal with frequencies significantly higher than the collision characteristic frequency band. These can be used to mask the true collision signal and reduce the signal-to-noise ratio. The smoothed measured force data, after filtering to remove high-frequency noise, can provide force measurements that more closely approximate actual physical contact.

[0106] Within the dynamic error compensation window, the end-force monitoring data is filtered. This can be achieved by applying a digital filter to smooth the data in the frequency or time domain. Further, this operation can be implemented using a second-order low-pass IIR filter or a 5-point moving average filter, thereby suppressing high-frequency vibrations and improving the signal-to-noise ratio of the collision signal. Removing high-frequency noise components caused by mechanical vibration can be achieved by setting the filter cutoff frequency to filter out components higher than the collision characteristic frequency band. In a specific embodiment, this operation can be achieved by setting the cutoff frequency to 200Hz to retain millisecond-level collision pulses or by using wavelet decomposition to remove high-frequency detail coefficients, thus preserving the low-frequency or transient force changes caused by the actual collision. The smoothed measured force data can then be output as a comparison object. For example, this operation can be achieved by directly outputting the filter output sequence or by extracting the envelope of the filtering result, thereby providing a more reliable measured benchmark.

[0107] Calculate the absolute value of the difference between the smoothed measured force data and the model predicted force data at the corresponding time points, and then divide the absolute value of the difference by the model predicted force data to obtain the offset percentage.

[0108] The absolute value of the difference can be the absolute amount of the numerical deviation between the smoothed measured force and the model predicted force at a corresponding time point. It can be used to quantify the degree of deviation and serve as the basis for calculating the relative deviation. Calculating the absolute value of the difference between the smoothed measured force data and the model predicted force data at a corresponding time point can be achieved by calculating the absolute deviation of the two sequences at corresponding time points point by point. Furthermore, this operation can be achieved by calculating the maximum absolute value of the difference within the entire window or by calculating the root mean square difference, thereby obtaining the unnormalized deviation. The percentage of offset is obtained by dividing the absolute value of the difference by the model predicted force data. This can be achieved by dividing the absolute value of the difference between the measured and predicted forces by the absolute value of the predicted force at each time point. In an exemplary embodiment, this operation can be achieved by adding an offset denominator to prevent division by zero when the predicted force is close to zero, or by using the average percentage of offset within the sliding window as the final indicator, thus achieving a normalized deviation measurement and making the threshold condition-adaptive.

[0109] Taking micro-collision recognition during high-speed cutting as an example, the intelligent single-axis robot anomaly monitoring system in this embodiment can detect a slight scratch when the robot is running at a speed of 1.2 m / s to a distance of X = 4.5 meters. The encoder feeds back the position and velocity data and inputs it into the dynamic model in real time. Combining the high stiffness parameters and nonlinear friction coefficient of this section, the theoretical tangential force is calculated to be 0.3 N. After the system detects the sudden change in time, based on the measured response lag of about 8 ms in this travel segment, the dynamic error compensation window is set to 12 ms. Within this window, the original 5 N pulse force data is low-pass filtered to remove vibration noise above 300 Hz, resulting in a smoothed peak value of 4.7 N. The absolute value of the difference between this and the predicted value of 0.3 N is 4.4 N, with an offset percentage of 1467%, far exceeding the threshold, thus confirming a valid collision. Meanwhile, the 1.5 N vibration caused by the guide rail joint in the low-speed section (X = 1.2 meters) is filtered out after the same process, with an offset percentage of only 80%, which is below the threshold.

[0110] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. An intelligent single-axis robot anomaly monitoring system, characterized in that, The system includes: The data synchronization module is used to acquire the dynamic model data of the motion actuator of the long-stroke single-axis robot, and simultaneously collect the real-time monitoring data set of multiple force sensing nodes of the single-axis robot transmission system and motion execution head. The map generation module is used to perform collision anomaly state identification processing on the real-time monitoring data set and generate a collision anomaly feature distribution map that matches the spatial location of the dynamic model data. The visual reconstruction module is used to perform dynamic visual reconstruction processing on the dynamic model data based on the collision anomaly feature distribution map and the preset single-axis long-stroke motion constraints, and generate a visual output interface with anomaly level markings. The strategy triggering module is used to associate and map the visualization output interface with the real-time monitoring data set, generate a collision emergency protection and maintenance feedback strategy, and trigger the protection response operation of the single-axis robot through the industrial control port.

2. The intelligent single-axis robot anomaly monitoring system as described in claim 1, characterized in that, The process of acquiring the dynamic model data of the long-stroke single-axis robot motion actuator and simultaneously collecting real-time monitoring data sets from multiple force sensing nodes of the single-axis robot transmission system and motion execution head includes: The design parameter database of the long-stroke single-axis robot is called to extract the structural stiffness parameters, motion inertia parameters, and nonlinear friction parameters that constitute the dynamic model data; The structural stiffness parameters are used to describe the stress deformation characteristics of the long-stroke guide rail and the transmission mechanism. The nonlinear friction parameters include the friction compensation coefficients at different stroke positions and the installation coordinates and configuration parameters associated with the sensing nodes. According to the configuration parameters, an activation command is sent to each sensing node, and the end force sensing data, servo motor torque and current data and encoder position and speed data of each sensing node are collected according to the set sampling period. The end effector force sensing data, the servo motor torque and current data, and the encoder position and speed data are spatiotemporally aligned based on the sensor node installation coordinates and the stroke hierarchy of the dynamic model to generate a hierarchical real-time monitoring data set. The segmented structure of the hierarchical real-time monitoring data set is mapped to the stroke stiffness hierarchy of the long-stroke single-axis robot, and the stroke stiffness hierarchy is a hierarchical structure divided according to the force characteristics of the long-stroke structure.

3. The intelligent single-axis robot anomaly monitoring system as described in claim 2, characterized in that, The step of performing collision anomaly state identification processing on the real-time monitoring data set to generate a collision anomaly feature distribution map that matches the spatial location of the dynamic model data includes: The end force sensing data in the hierarchical real-time monitoring data set is processed to detect abrupt change points, identify collision force abrupt change regions that exceed the dynamic fluctuation threshold, and extract the change rate and abrupt change amplitude of the collision force abrupt change regions. The trend deviation analysis is performed on the servo motor torque and current data after inertia compensation to determine the deviation between the actual torque and the expected torque of the preset model, and to mark the abnormal torque area where the deviation exceeds the dynamic threshold. The spatial coordinates of the collision force mutation region and the torque anomaly region are mapped to the corresponding positions in the dynamic model data to generate an initial collision anomaly feature distribution map containing collision type labels and anomaly levels. Based on the position offset alarm information in the encoder position and speed data, the initial collision anomaly feature distribution map is cross-validated to eliminate false alarm anomaly areas and correct the anomaly level, thereby generating the collision anomaly feature distribution map.

4. The intelligent single-axis robot anomaly monitoring system as described in claim 3, characterized in that, The dynamic visualization reconstruction process, based on the collision anomaly feature distribution map and preset single-axis long-stroke motion constraints, performs dynamic visualization reconstruction on the dynamic model data to generate a visualization output interface with anomaly level labels, including: The connection relationship data of the transmission mechanism and the servo control parameter data are extracted from the dynamic model data to construct the motion constraint relationship network of the long-stroke single-axis robot. Based on the collision type label in the collision anomaly feature distribution map, assign a corresponding color code and dynamic flashing frequency to each anomaly region; The color code and dynamic flashing frequency are superimposed onto the corresponding spatial position of the dynamic model data to generate a first visualization structure; Based on the motion constraint relationship network, the abnormal regions in the first visualization structure are subjected to positioning accuracy impact diffusion simulation processing to generate a second visualization structure containing potential accuracy offset region markers. The second visualization structure is fused with the position and velocity data in the hierarchical real-time monitoring dataset to generate the visualization output interface.

5. The intelligent single-axis robot anomaly monitoring system as described in claim 4, characterized in that, The step of associating and mapping the visualization output interface with the real-time monitoring data set to generate a collision emergency protection and maintenance feedback strategy includes: Extract the spatial coordinate set and collision type label of the abnormal region from the visualization output interface to generate a collision anomaly feature vector; Based on the collision anomaly feature vector, the corresponding protection priority scoring rules are matched from the preset strategy library to calculate the protection priority score for each anomaly region; Based on the protection priority score and the transmission connection tightness in the motion constraint relationship network, a collision emergency protection and maintenance feedback strategy is generated, which includes emergency response sequence, torque control scheme and resource allocation scheme; the emergency response sequence is a sequence sorted according to the priority score and transmission influence tightness.

6. The intelligent single-axis robot anomaly monitoring system as described in claim 5, characterized in that, The protection response operation triggered by the industrial control port of the single-axis robot includes: The collision emergency protection and maintenance feedback strategy is converted into a real-time control instruction set, which includes a single-axis emergency stop instruction, a reverse torque unloading instruction, a data acquisition frequency adjustment instruction, and a maintenance work order generation instruction. The real-time control instruction set is sent to the servo control unit of the single-axis robot via an encrypted communication link, instructing the servo control unit to perform reverse torque unloading and emergency stop operations; The maintenance work order generation instruction is simultaneously pushed to the preset equipment maintenance management platform, and the work order execution status data returned by the maintenance management platform is received. The abnormal status markers in the visualization output interface are updated based on the work order execution status data to generate a maintenance progress tracking layer.

7. The intelligent single-axis robot anomaly monitoring system as described in claim 3, characterized in that, The step of performing abrupt change point detection processing on the end force sensing data in the hierarchical real-time monitoring data set to identify collision force abrupt change regions exceeding the dynamic fluctuation threshold includes: The end force sensing data is divided into continuous force data segments according to a preset time window, and a first-order difference calculation is performed on each force data segment to obtain a force change rate sequence. Sliding window standard deviation analysis was performed on the force change rate sequence to identify abrupt change time points where the standard deviation exceeded the dynamic threshold; Extract the corresponding end force monitoring sub-data based on the mutation time point, and calculate the percentage offset between the end force monitoring sub-data and the model predicted force data; If the offset percentage exceeds a preset fluctuation threshold, the corresponding area of ​​the end force monitoring sub-data is marked as a collision force mutation area.

8. The intelligent single-axis robot anomaly monitoring system as described in claim 7, characterized in that, The trend deviation analysis processing of the servo motor torque and current data after inertia compensation determines the deviation between the actual torque and the expected torque of the preset model, and marks the abnormal torque regions where the deviation exceeds the dynamic threshold, including: Obtain preset torque reference curves for the long-stroke single-axis robot at different stroke positions and different movement speeds. The preset torque reference curves include a segmented expected torque range that takes into account inertial force and friction. The servo motor torque and current data are categorized by stroke segment and speed segment, and the categorized torque and current data are subjected to trend deviation analysis based on the preset torque reference curve to generate a real-time torque trend line. Calculate the deviation between the real-time torque trend line and the corresponding preset torque reference curve, wherein the deviation includes the duration of deviation and the cumulative amount of deviation; If the duration of the deviation exceeds the first dynamic threshold and the cumulative amount of the deviation exceeds the second dynamic threshold, the corresponding travel segment spatial coordinates are mapped to the dynamic model data, and the abnormal torque region is marked according to the mapped spatial coordinates.

9. The intelligent single-axis robot anomaly monitoring system as described in claim 6, characterized in that, The process of establishing the encrypted communication link includes: Assign a unique device identifier and dynamic encryption key to the servo control unit of each single-axis robot; Before sending the real-time control instruction set, an instruction header verification code is generated based on the identity identifier, and the instruction content is segmented and encrypted using the dynamic encryption key. The segmented encrypted instruction content is combined with the instruction header verification code to generate an encrypted data packet, which is then sent to the servo control unit via a multipath transmission protocol. Receive the instruction confirmation signal returned by the target servo control unit, and update the effective status of the dynamic encryption key according to the instruction confirmation signal.

10. The intelligent single-axis robot anomaly monitoring system as described in claim 7, characterized in that, The step of extracting the corresponding end force monitoring sub-data based on the mutation time point and calculating the percentage offset between the end force monitoring sub-data and the model predicted force data includes: The model predicted force data of the long-stroke single-axis robot in the current motion state is obtained. The model predicted force data is calculated and generated based on the structural stiffness parameters, nonlinear friction parameters and current encoder position and velocity data in the dynamic model data, and reflects the theoretical force state under no collision interference. A dynamic error compensation window is constructed, and the time width of the dynamic error compensation window is adaptively set according to the response lag characteristics of the long-stroke single-axis robot and the width of the preset time window. Within the dynamic error compensation window, the end force monitoring data is filtered to remove high-frequency noise components caused by mechanical vibration, resulting in smoothed measured force data. Calculate the absolute value of the difference between the smoothed measured force data and the model predicted force data at the corresponding time point, and then divide the absolute value of the difference by the model predicted force data to obtain the offset percentage.

Images