Control method based on online monitoring robot and robot

Abstract

The invention relates to the technical field of robot control, in particular to a control method based on an online monitoring robot and the robot, and the method comprises the steps: obtaining a real-time environment monitoring data flow collected by the online monitoring robot for a current inspection node, and extracting a topological identity ID based on the current inspection node; calculating an abnormal deviation index based on a comparison result of the real-time environment monitoring data stream and a preset safety reference value; and if the abnormal deviation index exceeds a preset threshold value, retrieving a preset equipment physical connection adjacency list by taking the topological identity ID as an index. According to the invention, through the linkage of the topological identity ID and the equipment physical connection adjacency list, when the robot detects that a certain equipment node is abnormal, the robot can automatically and quickly lock the adjacent equipment which may be influenced according to the physical connection relationship between the equipment, and preferentially carry out key monitoring on the equipment, so that the safety of the robot is improved. Therefore, missed detection caused by propagation of faults along a physical path is avoided, and the risk of cascading faults is reduced.

Description

Technical Field

[0001] This invention relates to the field of robot control technology, specifically to a control method and robot based on online monitoring. Background Technology

[0002] With the popularization of digital factory technology, intelligent inspection robots have been widely used in high-risk scenarios such as substations, chemical plants, and underground utility tunnels. These robots are typically equipped with various detection modules such as visible light cameras, infrared thermal imagers, and gas sensors. According to pre-set inspection routes and schedules, they periodically monitor the status of on-site equipment and collect data to replace manual labor in completing high-frequency, high-risk daily inspection tasks, ensuring the safe and stable operation of the production environment.

[0003] However, existing inspection robot control systems mostly adopt linear task scheduling and isolated anomaly alarm modes. When the robot detects a data anomaly at a certain node, it usually only triggers an alarm and uploads data for that single node, but cannot automatically search for and lock potential affected devices based on the physical connection relationship between devices. This monitoring method severs the propagation attribute of equipment failure in physical space, causing the robot to fail to discover potential hazards upstream and downstream of the fault source in a timely manner, which can easily lead to missed detection of chain failures.

[0004] Furthermore, when faced with sudden anomalies, traditional robots are often limited by fixed priority logic, making it difficult to plan the optimal focusing path. The lag and rigidity of this response mechanism severely restrict the efficiency of emergency response and the depth of risk perception of robots in complex industrial scenarios. Summary of the Invention

[0005] To achieve the above objectives, the present invention provides the following technical solution: a control method based on an online monitoring robot, comprising: Acquire the real-time environmental monitoring data stream collected by the online monitoring robot for the current inspection node, and extract the topology identity ID based on the current inspection node; Based on the comparison results between real-time environmental monitoring data streams and preset safety benchmark values, an anomaly deviation index is calculated. If the abnormal deviation index exceeds the preset threshold, the preset device physical connection adjacency table is retrieved using the topology identity ID as the index to lock the set of associated nodes that have a physical connection relationship with the current inspection node; Obtain the original priority sorting code of the set of associated nodes in the preset global task queue, perform a cascading priority replacement operation to uniformly promote the priority sorting code of the set of associated nodes to a higher level than the original maximum priority in the global task queue; obtain the spatial coordinates corresponding to the set of associated nodes, calculate the local risk cluster center based on the spatial coordinates, and generate a reconstructed inspection task queue. Based on the reconstructed inspection task queue and local risk cluster center, an emergency focusing path is constructed from the current location to the local risk cluster center; Based on the emergency focus path, the robot is driven to interrupt its original trajectory and prioritize the execution of encircling monitoring operations targeting the set of related nodes according to the reconstructed inspection task queue.

[0006] Preferably, the real-time environmental monitoring data stream includes at least one or more combinations of visible light image data, infrared thermal imaging temperature data, gas concentration data, ambient temperature and humidity data, and acoustic fingerprint monitoring data; Acquire the real-time environmental monitoring data stream collected by the online monitoring robot for the current inspection node, and extract the topology identity ID based on the current inspection node, including: The robot's positioning and recognition module reads the preset location marker beacon at the current inspection node, analyzes the radio frequency or visual features of the location marker beacon, and generates the node's physical location code at the current moment. The node physical location code is used as the query key value. A unique match is performed in the pre-set digital factory topology map database to extract the logical node number bound to the node physical location code, which is used as the topology identity ID. Based on the topology identity ID, the robot invokes a preset sensor configuration strategy to control the sensor modules on the robot to synchronously trigger sampling of the target device and obtain raw data from multiple sources. The topology identity ID and the timestamp information of the acquisition time are written into the header frame of the multi-source sensor raw data, and the multi-source sensor raw data is spatiotemporally aligned to generate a real-time environmental monitoring data stream.

[0007] Preferably, based on the comparison results between real-time environmental monitoring data streams and preset safety benchmark values, an anomaly deviation index is calculated, including: The frame structure of the real-time environmental monitoring data stream is parsed to separate the monitoring values ​​of different sensor channels and the corresponding data type identification codes. Based on the data type identifier, the corresponding parameter threshold range is retrieved from the preset safety benchmark database. The sub-monitoring values ​​are compared and normalized with the parameter threshold range to generate a dimensionless deviation vector that eliminates the influence of physical units. Based on the data type identifier code matching the preset fault propagation sensitivity weight, the fault propagation sensitivity weight is multiplied point-to-point with each component in the dimensionless deviation vector to generate a weighted risk factor sequence that reflects the degree of potential impact of each parameter on the associated node. Perform modulus calculation or weighted summation on the weighted risk factor sequence to generate a single value that represents the overall severity of the anomaly at the current inspection node, namely the anomaly deviation index.

[0008] Preferably, if the abnormal deviation index exceeds a preset threshold, a preset device physical connection adjacency table is retrieved using the topology identity ID as an index to lock the set of associated nodes that have a physical connection relationship with the current inspection node, including: The abnormal deviation index is compared with the system's preset alarm threshold. When the abnormal deviation index is greater than the alarm threshold, a topology search trigger command with the current timestamp is generated. In response to the topology search trigger command, the topology identity ID is used as the unique hash key value to perform a row-level scan of the preset device physical connection adjacency table, locate and lock the target node index row that completely matches the unique hash key value; Read the physical connection pointer data stored in the target node index row, parse out the identity codes of all directly adjacent nodes pointed to by the data, and generate an original adjacency code sequence containing multiple node codes; The original adjacency coding sequence is validated and deduplicated. After removing invalid codes, the remaining valid node codes are identified as the target objects to be inspected and packaged into a set of associated nodes.

[0009] Preferably, the original priority sorting codes of the associated node set in the preset global task queue are obtained, and a cascading priority replacement operation is performed to uniformly raise the priority sorting codes of the associated node set to a burst level higher than the original maximum priority in the global task queue, including: Iterate through the priority fields of all tasks to be executed in the global task queue, filter out the highest priority value in the current queue by numerical comparison, and add a preset burst response weight to generate a cascaded replacement baseline value. Obtain the priority sorting code of each associated node in the set of associated nodes in its original state, calculate the numerical difference of each associated node relative to the lowest priority node in the set, and generate an internal relative position difference sequence. Each difference in the internal relative position difference sequence is superimposed on the cascade permutation benchmark value to generate a new burst reset priority code, and this code is used to cover the corresponding original priority sorting code in the associated node set.

[0010] Preferably, the spatial coordinates corresponding to the set of associated nodes are obtained, and the local risk cluster center is calculated based on the spatial coordinates, including: Using the topological identity ID in the set of associated nodes as an index, the system retrieves the preset 3D digital factory map database and extracts the 3D physical coordinate vector and equipment monitoring face orientation information corresponding to each associated node. Obtain the robot's current real-time position coordinates, calculate the Euclidean distance between the 3D physical coordinate vector and the real-time position coordinates, and define the normalized reciprocal of the Euclidean distance as the spatial proximity weight. By using spatial proximity weights to perform weighted average geometric operations on the three-dimensional physical coordinate vectors, the coordinates of the local risk cluster center representing the physical geometric center of the risk area are synthesized and output.

[0011] Preferably, the process of generating a reconstructed inspection task queue includes: Get all task data after priority replacement and currently interrupted pending task data, sort them in descending order according to the value of priority sort code, and divide the all task data into a high-priority related sequence at the head and a normal pending sequence at the tail. Based on the coordinates of the local risk cluster centers obtained from the above calculation, navigation guidance anchors are constructed, and the navigation guidance anchors are encapsulated and inserted into the head position of the high-quality association sequence to generate a guided high-quality sequence. Establish a linked list pointer index, point the tail pointer of the guiding high-priority sequence to the head pointer of the normal pending sequence, complete the logical concatenation, and generate the final reconstructed inspection task queue that can be parsed by the navigation system.

[0012] Preferably, based on the reconstructed inspection task queue and local risk cluster center, an emergency focusing path is constructed from the current location to the local risk cluster center, including: The robot's current location coordinates are used as the starting point, and the coordinates of the center of the local risk cluster are used as the ending point. A spatial guidance vector connecting the starting point and the ending point is constructed, and a navigation search corridor covering the area along the way is generated based on this vector. The navigation search corridor is mapped onto a pre-set 3D environment grid map. Static obstacles occupying grids and dynamic restricted areas are removed, and the remaining continuous free grids are extracted to generate a collision-free passage topology. Traverse the high-priority associated nodes at the head of the reconstructed inspection task queue, extract the corresponding equipment monitoring face orientation information, map the orientation information to the path nodes of the collision-free passage topology, and generate a set of full-dimensional attitude key point sequences that satisfy both passage safety and early visualization of risk targets. Based on the robot's kinematic constraint parameters, a spline interpolation algorithm is used to smoothly connect the full-dimensional attitude key point sequence to generate an emergency focusing path that includes time, position, velocity, and gimbal angle commands.

[0013] Preferably, based on the emergency focusing path, the robot is driven to interrupt its original trajectory, and the encircling monitoring operation targeting the set of related nodes is prioritized and executed according to the reconstructed inspection task queue, including: Upon receiving the emergency focus path as the highest priority control signal, immediately send a low-level interrupt command to the robot motion controller to clear the currently cached navigation waypoints and resolve the emergency focus path into a millisecond-level preemptive motion control flow; The preemptive motion control flow is sent to the chassis drive unit and gimbal servo unit to drive the robot to quickly approach the risk area along the planned path. When the deviation between the real-time positioning coordinates and the center of the local risk cluster is less than a preset threshold, the area positioning lock signal is triggered. In response to the area positioning and locking signal, the physical coordinate distribution of the associated node set in the reconstructed inspection task queue is retrieved, the minimum external field of view covering all associated nodes is calculated, and it is discretized and decomposed into a set of multi-point surrounding observation matrices for surrounding observation of this range. The multi-point encirclement observation matrix is ​​transformed into a combination sequence of gimbal attitude commands and chassis fine-tuning commands, which drives the robot to continuously perform multi-angle high-frequency sampling of the associated node set while maintaining focus on the center of the local risk cluster, thus completing the encirclement monitoring operation.

[0014] A robot, applicable to the above-described control method for online monitoring robots, includes: The data acquisition module is used to acquire the real-time environmental monitoring data stream collected by the online monitoring robot for the current inspection node, and extract the topology identity ID based on the current inspection node; The deviation calculation module is used to calculate the abnormal deviation index based on the comparison results between the real-time environmental monitoring data stream and the preset safety benchmark value. The topology locking module is used to retrieve the preset device physical connection adjacency table using the topology identity ID as the index when the abnormal deviation index exceeds the preset threshold, and lock the set of associated nodes that have a physical connection relationship with the current inspection node. The spatiotemporal reconstruction module is used to obtain the original priority sorting code of the set of associated nodes in the preset global task queue, perform a cascading priority replacement operation, and uniformly promote the priority sorting code of the set of associated nodes to a burst level higher than the original maximum priority in the global task queue; obtain the spatial coordinates corresponding to the set of associated nodes, calculate the local risk cluster center based on the spatial coordinates, and generate a reconstruction inspection task queue. The focused pathfinding module is used to construct an emergency focused path from the current location to the local risk cluster center based on the reconstructed inspection task queue and the local risk cluster center. The surround monitoring module is used to drive the robot to interrupt its original trajectory according to the emergency focusing path, and to prioritize the execution of surround monitoring operations on the set of related nodes according to the reconstructed inspection task queue.

[0015] Compared with the prior art, the beneficial effects of the present invention are: (1) By linking the topology identity ID with the device physical connection adjacency table, when the robot detects an abnormality in a device node, it can automatically and quickly lock the adjacent devices that may be affected based on the physical connection relationship between the devices, and prioritize the monitoring of these devices, thus avoiding missed detections caused by the propagation of faults along physical paths such as pipes and cables, and significantly reducing the risk of cascading failures. (2) The present invention adopts a cascaded priority replacement operation for the task queue. After an anomaly occurs, the robot can immediately interrupt the original inspection route and rush to the risk area for centralized inspection. The response is faster and avoids the delay of traditional methods that only alarm a single node and wait for manual intervention. This improves the efficiency of emergency response in case of an emergency. By intelligently calculating the center position of the risk area and planning a dedicated focusing path, the robot can reach the most needed inspection position faster and safer. At the same time, it automatically adjusts the angle during the inspection to observe multiple related devices from all angles, making the monitoring more comprehensive and thorough. Attached Figure Description

[0016] Figure 1 This is a schematic flowchart of the overall method in one embodiment of the present invention; Figure 2 This is a schematic diagram of a structure according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a hardware entity in one embodiment of the present invention. Detailed Implementation

[0017] 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 some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1, please refer to Figure 1 This invention provides a technical solution: a control method based on an online monitoring robot, comprising: S1. Obtain the real-time environmental monitoring data stream collected by the online monitoring robot for the current inspection node, and extract the topology identity ID based on the current inspection node; S2. Calculate the anomaly deviation index based on the comparison results between the real-time environmental monitoring data stream and the preset safety benchmark value; S3. If the abnormal deviation index exceeds the preset threshold, the preset device physical connection adjacency table is retrieved using the topology identity ID as the index to lock the set of associated nodes that have a physical connection relationship with the current inspection node. S4. Obtain the original priority sorting code of the associated node set in the preset global task queue, perform a cascading priority replacement operation, and uniformly promote the priority sorting code of the associated node set to a higher level than the original maximum priority in the global task queue; obtain the spatial coordinates corresponding to the associated node set, calculate the local risk cluster center based on the spatial coordinates, and generate a reconstructed inspection task queue. S5. Based on the reconstructed inspection task queue and local risk cluster center, construct an emergency focusing path from the current location to the local risk cluster center; S6. Drive the robot to interrupt its original trajectory according to the emergency focus path, and prioritize the execution of encirclement monitoring operations for the set of related nodes according to the reconstructed inspection task queue.

[0019] In this invention, the technical essence of steps S1 and S2 is to transform multi-dimensional physical environmental parameters into standardized digital risk measurement indicators. Through real-time data acquisition and threshold comparison analysis, quantitative data support is provided for subsequent logical judgments. For example, when the robot inspects a valve node and collects temperature and gas concentration data, the system calculates through an algorithm that the parameters of the node deviate significantly from the standard operating baseline, indicating that the node is in an unsteady state. Based on this, the system determines that abnormal handling logic needs to be initiated. The core logic of steps S3 and S4 lies in identifying potential risk propagation paths based on predefined industrial topology relationships and using a dynamic priority scheduling algorithm to reorganize the execution sequence of the task queue at the logical level, constructing a work sequence centered on the risk area. For example, after determining the above-mentioned valve abnormality, the system searches the database to lock the upstream pump and downstream flow meter that are physically connected to it. Subsequently, the system uses an algorithm to increase the priority value of these three devices from the normal 50 to the emergency level of 1050, and calculates their spatial geometric center of gravity, thereby completing the scheduling conversion from routine inspection to emergency investigation at the data structure level. Steps S5 and S6 aim to transform the logical task reconstruction into physical motion control commands. By generating optimal paths that satisfy kinematic constraints, the robot is driven to execute hardware interrupts and trajectory replanning to achieve efficient coverage of the target area. For example, when the robot receives the reconstructed command, it immediately terminates its current movement to the power distribution room and loads an emergency focusing path pointing to the center point of the valve area. After reaching the target location, the robot controls the chassis and pan-tilt unit to continuously acquire images of the valve, pump, and flow meter from multiple angles, completing targeted monitoring of specific risk areas.

[0020] In an optional embodiment, the real-time environmental monitoring data stream includes at least one or more combinations of visible light image data, infrared thermal imaging temperature data, gas concentration data, ambient temperature and humidity data, and acoustic signature monitoring data; Acquire the real-time environmental monitoring data stream collected by the online monitoring robot for the current inspection node, and extract the topology identity ID based on the current inspection node, including: The robot's positioning and recognition module reads the preset location marker beacon at the current inspection node, analyzes the radio frequency or visual features of the location marker beacon, and generates the node's physical location code at the current moment. It should be noted that location identification beacons refer to physical media deployed at key monitoring points in industrial sites to assist robots in accurate positioning and identification. These include, but are not limited to, passive RFID tags, NFC near-field communication patches, industrial QR codes, or AprilTag visual tags. Their function is to provide a verification benchmark for the absolute physical position of the moving robot, eliminating the cumulative positioning errors caused by relying solely on odometers or SLAM algorithms. For example, when a robot travels to the area of ​​the No. 1 main transformer in a substation, its radio frequency reader at the bottom senses the RFID tag buried underground, or its camera scans the anti-fouling QR code affixed to the equipment cabinet door, thereby resolving a unique hexadecimal physical code "LOC_Tag_A7F2". This code represents the robot's current exact physical location. The node physical location code is used as the query key value. A unique match is performed in the pre-set digital factory topology map database to extract the logical node number bound to the node physical location code, which is used as the topology identity ID. It should be noted that the digital factory topology map database is a middleware system that stores the correspondence between physical locations and logical assets. It maps semantically meaningless physical codes to logical identities with business attributes. The topology identity ID is an index key used within the system to uniquely identify the role of the tested equipment in the power grid or pipeline topology, containing semantic information such as equipment type, bay to which it belongs, and connection relationship. For example, after the system receives the physical code "LOC_Tag_A7F2", it searches in the database and finds that the location is bound to "#1 main transformer - high voltage side bushing A phase". Then it extracts the corresponding logical ID "Trans_01_HV_Bush_A". This ID not only represents the location, but also implies that the equipment belongs to the high voltage circuit and has a parallel relationship with other B and C phase bushings. Based on the topology identity ID, the robot invokes a preset sensor configuration strategy to control the sensor modules on the robot to synchronously trigger sampling of the target device and obtain raw data from multiple sources. It should be noted that the sensor configuration strategy is a set of predefined intelligent acquisition rules for different types of devices. Based on the device attributes indicated by the topology identity ID, it determines which sensors should be enabled at a specific node, what sampling frequency to set, and what pan-tilt zoom parameters to configure. This avoids indiscriminate full data acquisition from all nodes, thereby saving storage space and improving the targeting of data processing. For example, when the topology identity ID is identified as "sulfur hexafluoride circuit breaker", the strategy will instruct the "infrared thermal imager" and "gas leak detector" to be turned on to monitor temperature and airtightness, and turn off the irrelevant "partial discharge detector". When the ID is identified as "transformer oil conservator", the strategy will instruct the "visible light camera" to be turned on to identify the oil level scale and automatically adjust the pan-tilt zoom to 5x to obtain a clear local instrument image. The topology identity ID and the timestamp information of the acquisition time are written into the header frame of the multi-source sensor raw data, and the multi-source sensor raw data is spatiotemporally aligned to generate a real-time environmental monitoring data stream. It should be noted that spatiotemporal alignment processing refers to the process of uniformly mapping asynchronous data streams from different sensor channels (such as image sensors, gas sensors, and acoustic microphones) to the same topological object and the same time window. By adding a unified ID and microsecond-level timestamp to the header frame of the data packet, time deviations caused by differences in sensor response speeds are eliminated, ensuring the multidimensional correlation of data in subsequent analysis. For example, if a visible light camera takes a picture at 10:00:00.050, while the gas sensor completes a reading at 10:00:00.055, there is a slight time difference between the two. Through alignment processing, the system uniformly encapsulates these two data points in a standardized data packet with "ID: Trans_01_HV_Bush_A, Time: 10:00:00". This allows the algorithm model to combine the image appearance and gas concentration at that moment for comprehensive diagnosis when judging faults, avoiding misjudgments caused by data asynchrony.

[0021] In an optional embodiment, based on the comparison result between the real-time environmental monitoring data stream and the preset safety benchmark value, an anomaly deviation index is calculated, including: The frame structure of the real-time environmental monitoring data stream is parsed to separate the monitoring values ​​of different sensor channels and the corresponding data type identification codes. It should be noted that frame structure parsing refers to the process of decomposing a continuous binary data stream into data fields with specific physical meanings according to the bit mapping rules defined by the underlying communication protocol (such as Modbus, CAN bus, or custom TCP / IP protocol). The data type identifier is a metadata tag attached to the numerical value, used to inform the system what physical quantity the value represents, such as temperature, pressure, or voltage. For example, when the system receives a hexadecimal data frame "0x01_03E8_0x02_0032", the parser decomposes it into two sets of data according to preset rules: the first set of identifiers "0x01" represents infrared temperature, and the value "03E8" corresponds to 1000 in decimal, i.e., 100.0℃; the second set of identifiers "0x02" represents ambient humidity, and the value "0032" corresponds to 50 in decimal, i.e., 50%RH, thus achieving the structured separation of mixed data streams. Based on the data type identifier, the corresponding parameter threshold range is retrieved from the preset safety benchmark database. The sub-monitoring values ​​are compared and normalized with the parameter threshold range to generate a dimensionless deviation vector that eliminates the influence of physical units. It should be noted that the preset safety benchmark database is a standard library storing the operating ranges of various physical parameters of various industrial equipment under normal operating conditions, covering national standard values ​​and historical operating experience values. Normalization aims to eliminate the differences in magnitude between different physical dimensions, such as Celsius, Pascal, and decibels, and unify them into deviation factors relative to the safety benchmark values, constructing a dimensionless deviation vector to facilitate subsequent comprehensive evaluation on the same mathematical plane. For example, assuming the upper limit of safe operating temperature for a switchgear is 80℃ and the upper limit of safe internal air pressure is 1.0MPa, if the current monitoring values ​​collected by the robot are 100℃ and 1.2MPa, directly comparing 100 and 1.2 is meaningless. However, after normalization calculations, for example: The temperature deviation is The air pressure deviation is The resulting dimensionless vector is [0.25, 0.20], which indicates that although the absolute value of the air pressure is very small, in terms of relative risk, the degree of temperature exceeding the standard at 0.25 is more serious than that at air pressure at 0.20. Based on the data type identifier code matching the preset fault propagation sensitivity weight, the fault propagation sensitivity weight is multiplied point-to-point with each component in the dimensionless deviation vector to generate a weighted risk factor sequence that reflects the degree of potential impact of each parameter on the associated node. It should be noted that the fault propagation sensitivity weight is a coefficient value (usually between 0 and 1) set based on an expert experience base. It characterizes the likelihood that different types of fault parameters will cause cascading damage to physically connected neighboring nodes. A higher weight means that once the parameter is abnormal, it is more likely to trigger cascading failures in related equipment. For example, in a substation scenario, "SF6 gas leakage" may rapidly spread, leading to insulation breakdown and affecting the entire switchgear group; its propagation sensitivity weight is set to 0.9. On the other hand, "minor corrosion on the cabinet surface" usually only affects the lifespan of the cabinet itself and is unlikely to instantly affect neighboring equipment; its weight is set to 0.2. Assuming that the normalization deviation of both is 0.5, after weighted calculation, the risk factor for gas leakage is... The risk factor for corrosion is This highlights the impact weight of high-risk parameters on subsequent path planning at the algorithm level. Perform modulus calculation or weighted summation on the weighted risk factor sequence to generate a single value that represents the overall severity of the anomaly at the current inspection node, namely the anomaly deviation index. It should be noted that the anomaly deviation index is a mathematical aggregation algorithm that compresses multi-dimensional risk factors into a one-dimensional scalar. It is typically calculated using the Euclidean norm of the vector. A higher value indicates that the current node is not only in a worse state itself, but also poses a greater threat to surrounding nodes. This is a direct criterion for triggering subsequent topological searches and emergency responses. For example, if the above steps yield a risk factor sequence for a node containing [temperature risk 0.4, air pressure risk 0.3], the system calculates the magnitude of this vector. If the preset system alarm threshold is 0.45, the calculated result of 0.5 exceeds the threshold. The system determines that the node is in a sudden and spreading abnormal state, and then triggers the association locking and surrounding monitoring process for its physical connected neighbors.

[0022] In an optional embodiment, if the abnormal deviation index exceeds a preset threshold, a preset device physical connection adjacency table is retrieved using the topology identity ID as an index to lock the set of associated nodes that have a physical connection relationship with the current inspection node, including: The abnormal deviation index is compared with the system's preset alarm threshold. When the abnormal deviation index is greater than the alarm threshold, a topology search trigger command with the current timestamp is generated. It should be noted that the topology search trigger command is a high-priority soft interrupt signal generated internally by the system. It marks the switch of the robot's working mode from routine linear inspection to emergency correlation investigation. The timestamp carried by this command is used to mark the start time of the fault event, so as to facilitate subsequent task backtracking and log auditing. For example, if the alarm threshold is set to 0.7, when the robot calculates that the abnormal deviation of the current transformer is 0.85, the logic judgment unit confirms the abnormality and immediately generates the command code "CMD_Topo_Search_20231027_143005". Once this command is generated, it will directly activate the search thread of the background database. In response to the topology search trigger command, the topology identity ID is used as the unique hash key value to perform a row-level scan of the preset device physical connection adjacency table, locate and lock the target node index row that completely matches the unique hash key value; It should be noted that the device physical connection adjacency list is a graph data structure pre-stored in the robot's onboard controller or cloud server. It does not record the spatial distance between devices, but specifically records the hard connections between devices, such as current paths, fluid pipelines, and mechanical transmission chains. The unique hash key is an index identifier used to achieve direct addressing in the data structure. Compared with traditional sequential traversal queries, the hash retrieval mechanism can establish a direct mapping relationship between IDs and storage addresses. For example, in a chemical plant database with tens of thousands of device nodes, if a conventional search is used, locking a device may require scanning the entire list, and the time consumption increases with the number of devices, making it difficult to meet the real-time control requirements of sudden failures. However, in this embodiment, by using the topology identity ID as the hash key, the system can directly jump to the corresponding adjacency list storage address in memory. Regardless of the size of the factory, the retrieval time remains within a constant microsecond range (i.e., constant time complexity), thereby ensuring zero-latency response of the robot to sudden anomalies. Read the physical connection pointer data stored in the target node index row, parse out the identity codes of all directly adjacent nodes pointed to by the data, and generate an original adjacency code sequence containing multiple node codes; It should be noted that the physical connection pointer data stores the IDs of upstream and downstream devices that have a direct physical coupling relationship with the current device, representing the physical path that a fault may propagate. The original adjacency encoding sequence is a list of uncleaned associated devices initially extracted by the system. For example, in the data row of "Water Pump No. 1", the pointer points to "VALVE_IN_01 (inlet valve)" at the inlet end, "VALVE_OUT_01 (outlet valve)" at the outlet end, and "INVERTER_01 (frequency converter)" at the power supply end. The system parses these pointers and generates an original sequence [VALVE_IN_01, VALVE_OUT_01, INVERTER_01] containing these three device IDs. The original adjacency coding sequence is validated and deduplicated. After removing invalid codes, the remaining valid node codes are identified as the target objects to be inspected and packaged into a set of associated nodes. It should be noted that validity verification and deduplication are performed to prevent "zombie links" of obsolete devices or "duplicate references" caused by loop topology in the database, ensuring that the tasks to be generated by the robot are truly executable. For example, suppose the original sequence contains an old sensor ID "SENSOR_OLD" that has been removed, and "VALVE_IN_01" appears twice due to network loops. The verification program will query the asset status table and find that "SENSOR_OLD" is in the "obsolete" state, so it will be removed. At the same time, duplicate "VALVE_IN_01" will be merged. Finally, the system outputs the cleaned set of associated nodes [VALVE_IN_01, VALVE_OUT_01, INVERTER_01], which serves as the exact target for subsequent path planning.

[0023] In an optional embodiment, the original priority sorting codes of the associated node set in a preset global task queue are obtained, and a cascading priority replacement operation is performed to uniformly raise the priority sorting codes of the associated node set to a burst level higher than the original maximum priority in the global task queue, including: Iterate through the priority fields of all tasks to be executed in the global task queue, filter out the highest priority value in the current queue by numerical comparison, and add a preset burst response weight to generate a cascaded replacement baseline value. It should be noted that the global task queue is a dynamic data structure residing in the robot's main control computer memory. It is used to store and manage all pending inspection instruction task blocks in real time. Each task block encapsulates key information such as the target device ID, action type, preset priority, and status bits. The priority field is a numerical label stored in the task control block to quantify the urgency of task execution; generally, the larger the value, the higher the priority. The system uses this field to determine the CPU scheduling order. The burst response weight is a preset, large step value, such as 100, that far exceeds the normal task priority range. Logically, it enforces a distinction between normal and emergency states; while the cascade replacement benchmark value is a dynamic water level calculated based on the above two, namely: the current maximum priority + emergency weight. It constructs an emergency channel starting point above all normal tasks; for example: assuming the highest priority of normal tasks in the current queue is 50, when a failure occurs, the system adds an emergency response weight of 1000, generating a cascade replacement benchmark value of 1050. This ensures that any emergency task generated subsequently (>1050) absolutely outweighs all the original tasks (≤50) in terms of value, thereby achieving unconditional task queue jumping and millisecond-level preemption.

[0024] Obtain the priority sorting code of each associated node in the set of associated nodes in its original state, calculate the numerical difference of each associated node relative to the lowest priority node in the set, and generate an internal relative position difference sequence. It should be noted that the priority sorting code is a digital weight pre-assigned to each inspection node or task, used to characterize the relative importance of the node in the normal inspection state, usually determined by the criticality level of the equipment; while the internal relative position difference sequence is the relative offset calculated based on these sorting codes. Its core purpose is to maintain order, that is, to ensure that while the overall priority is increased to jump the queue, the original order of maintenance logic is still maintained within the associated set; for example, assuming that the associated node set includes "auxiliary cooling fan" (priority sorting code is 12) and "main control unit" (priority sorting code is 15), the system identifies the lowest value as 12, and the calculated differences are 12−12=0 and 15−12=3, respectively. The generated sequence [0,3] records the relative fact that "the main control unit is 3 levels more important than the fan", ensuring that even in an emergency, the robot will check the main control unit first; Each difference in the internal relative position difference sequence is superimposed on the cascade permutation benchmark value to generate a new burst reset priority code, and this code is used to cover the corresponding original priority sorting code in the associated node set. It should be noted that the sudden reset priority code is the execution token ultimately assigned to the associated node. It combines the dual attributes of absolute high priority and relative internal order, ensuring that the emergency response is both rapid and orderly. For example, based on the above baseline value of 1050 and the difference sequence [0,3], the new priority of switchgear A becomes 1050+0=1050, and the new priority of main transformer B becomes 1050+3=1053. The results show that both are much higher than the regular tasks (50) in the original queue, so they will be immediately inserted into the queue for execution. At the same time, B 1053 is still higher than A 1050, ensuring that the more important transformer will be checked first, realizing intelligent scheduling with overall insertion into the queue and internal order.

[0025] In an optional embodiment, obtaining the spatial coordinates corresponding to the set of associated nodes and calculating the local risk cluster center based on the spatial coordinates includes: Using the topological identity ID in the set of associated nodes as an index, the system retrieves the preset 3D digital factory map database and extracts the 3D physical coordinate vector and equipment monitoring face orientation information corresponding to each associated node. It should be noted that the 3D digital factory map database is a digital twin foundation built on the robot control system. It accurately stores the absolute spatial position (X, Y, Z coordinates) of every piece of equipment in the factory, as well as the geometric attributes of its key components. The equipment monitoring face orientation information is a normal vector data used to indicate the facing direction of the equipment's instrument panel, level gauge, or infrared observation window, ensuring that the robot can plan an observation path facing the equipment rather than going around to the back of the equipment. For example, for a valve with the ID "VALVE_05", the database returns its center coordinates as (15.0, 20.0, 1.5) meters and its orientation vector as (0, 1, 0) (i.e., due north). Based on this, the robot knows that it needs to move to the south side of the valve for observation. Obtain the robot's current real-time position coordinates, calculate the Euclidean distance between the 3D physical coordinate vector and the real-time position coordinates, and define the normalized reciprocal of the Euclidean distance as the spatial proximity weight. It's important to note that Euclidean distance refers to the straight-line physical distance between two coordinate points in three-dimensional space, i.e., the length of the line connecting them. It's typically calculated based on the Pythagorean theorem and used to precisely quantify the spatial span between a robot and a target device. Spatial proximity weight, on the other hand, is based on the principle that closer points have higher weights. It's a numerical coefficient generated by taking the reciprocal of the Euclidean distance and normalizing it, aiming to prioritize the robot's attention to nearby risk points. For example, assuming the robot's current coordinates are the origin (0,0,0), and the coordinates of a related valve are (3,4,0), according to the Euclidean distance formula... The straight-line distance between the two is calculated as follows: If the distance to another device is 10 meters, then the device at 5 meters is closer. The system takes the reciprocal for calculation (1 / 5=0.2, 1 / 10=0.1). After normalization, the former has a higher weight, thus guiding the algorithm to shift the center of the risk cluster towards the valve at 5 meters. By using spatial proximity weights to perform weighted average geometric operations on the three-dimensional physical coordinate vectors, the coordinates of the local risk cluster center representing the physical geometric center of the risk area are synthesized and output. It should be noted that the local risk cluster center is not a simple multi-point geometric center, but a virtual gravitational center that incorporates spatial distance weights. It represents the optimal observation point for the group of fault-related devices, and can maximize the balance between the robot's observation perspective and movement cost for each risk node. For example, continuing the above example, the cluster center X coordinate is calculated as 10×0.83+50×0.17=8.3+8.5=16.8 meters based on the weighted average of the coordinates (A:0.83, B:0.17). Compared to the simple geometric midpoint of 30 meters ((10+50) / 2), the result of 16.8 meters is closer to the robot's current position and node A. The robot will navigate to this coordinate, so that it can quickly observe A, while also keeping an eye on B at a distance.

[0026] In an optional embodiment, generating a reconstructed inspection task queue includes: Get all task data after priority replacement and currently interrupted pending task data, sort them in descending order according to the value of priority sort code, and divide the all task data into a high-priority related sequence at the head and a normal pending sequence at the tail. It should be noted that suspended task data refers to a "snapshot" of currently executing but incomplete tasks taken by the system at the moment of a sudden interruption. This includes not only the task ID but also the instruction pointer position, register state, and collected semi-finished data at the breakpoint. High-priority related sequences refer to the set of all related fault nodes whose priority values ​​have jumped above the sudden interruption baseline (e.g., 1000+) after cascading priority replacement operations, representing urgent tasks that must be executed immediately. Regular pending sequences include the remaining regular inspection points in the original global queue. And suspended task data that are forced to pause due to sudden interruption, that is, tasks that are currently being executed but not completed, their priority is maintained at a low level (such as 0-100); for example: when the robot is executing the inspection of point A with a priority of 50, it is interrupted, that is, point A becomes a suspended task, the system reorders all tasks, and the two related fault points with priorities of 1053 and 1050 are placed at the top, forming a high priority related sequence; while the original point A (50) and the subsequent point B (30) are placed at the bottom, forming a normal pending sequence, thus forming a clear execution echelon of urgent and slow tasks; Based on the coordinates of the local risk cluster centers obtained from the above calculation, navigation guidance anchors are constructed, and the navigation guidance anchors are encapsulated and inserted into the head position of the high-quality association sequence to generate a guided high-quality sequence. It should be noted that the navigation guide anchor point is a special virtual task node. It does not correspond to a specific equipment detection action, but rather to a pure move-to-position instruction. Its purpose is to drive the robot's chassis to move to the optimal observation position, i.e., the center of the local risk cluster. For example, if the calculated risk cluster center coordinates are (15,20), the system creates a task with the code "NAV_ANCHOR_01" with the instruction to quickly move to (15,20). This task is inserted at the beginning of the high-priority sequence, forming a structure of [move anchor point -> check fault point 1 -> check fault point 2]. This ensures that the robot first runs to the optimal position, stands firmly, and then starts turning around to check the equipment, avoiding low-quality observations during movement. Establish a linked list pointer index, point the tail pointer of the guiding high-priority sequence to the head pointer of the normal pending sequence, complete the logical concatenation, and generate the final reconstructed inspection task queue that can be parsed by the navigation system; It's important to note that linked list pointer indexing is a highly efficient data structure connection method. It achieves sequence merging by modifying memory addresses rather than moving data blocks, significantly saving system resources. The reconstructed inspection task queue is the final linear execution flow containing all instructions for both emergency and routine tasks. For example, assuming the last task in the guided high-priority sequence is "inspect the frequency converter (node ​​Y)," and the first task in the routine pending sequence is "inspect the water meter (node ​​A)," the system directly modifies the "Next_Task_Ptr" (next task pointer) in node Y's data structure to point to the memory address of node A. Thus, after the robot completes task Y, it automatically reads and executes task A following the pointer, achieving a seamless automatic switch from emergency mode to routine mode without manual intervention.

[0027] In an optional embodiment, based on the reconstructed inspection task queue and local risk cluster center, an emergency focusing path is constructed from the current location to the local risk cluster center, including: The robot's current location coordinates are used as the starting point, and the coordinates of the center of the local risk cluster are used as the ending point. A spatial guidance vector connecting the starting point and the ending point is constructed, and a navigation search corridor covering the area along the way is generated based on this vector. It should be noted that the spatial guidance vector refers to the directed line segment in a three-dimensional coordinate system that directly connects the robot's current point and the target point (the center of the risk cluster), representing the theoretical shortest straight path. The navigation search corridor, on the other hand, is a tubular or cuboid virtual search space formed by extending a certain width (e.g., 3 meters) to the left, right, up, and down with this vector as the central axis. For example, if the robot is located at (0,0) and the target is located at (100,100), searching the entire map would require a huge amount of computation. The system constructs a diagonal vector connecting these two points and sets the path planning calculation to be performed only within a corridor area centered on the diagonal and 5 meters wide. This shields the data interference from other irrelevant areas of the factory and improves the path calculation efficiency by several times. The navigation search corridor is mapped onto a pre-set 3D environment grid map. Static obstacles occupying grids and dynamic restricted areas are removed, and the remaining continuous free grids are extracted to generate a collision-free passage topology. It should be noted that the 3D environment grid map is a digital map that discretizes the physical space of the factory into small cubic units. Each grid is marked as occupied (with walls / pillars) or free. The collision-free passage topology refers to the set of all interconnected free grids remaining within the corridor area mentioned above after removing grids occupied by walls, equipment, and temporary maintenance barriers. For example, in the corridor mentioned above, there is a load-bearing column at coordinates (50,50). The system will mark the grid at this location as impassable and search for free grids around it at (49,50) or (51,50) to build a connected network that bypasses the column, ensuring that the path generated by the robot is absolutely safe. Traverse the high-priority associated nodes at the head of the reconstructed inspection task queue, extract the corresponding equipment monitoring face orientation information, map the orientation information to the path nodes of the collision-free passage topology, and generate a set of full-dimensional attitude key point sequences that satisfy both passage safety and early visualization of risk targets. It should be noted that the equipment monitoring surface orientation information refers to the normal direction vector of the key observation part of the target equipment; the full-dimensional attitude key point sequence is different from ordinary two-dimensional coordinate points (x,y), it includes position coordinates and gimbal orientation angle. The combined data aims to achieve "observation in motion"; for example, a robot is moving north along a corridor to carry out emergency repairs, while the associated faulty equipment is located on the east wall of the corridor. When planning the path points, the system not only sets the robot chassis to move north, but also adds the action of "rotating the gimbal 90 degrees to the right" to the key point instructions. In this way, before the robot reaches the destination, the camera has already been pointed at the faulty equipment halfway, realizing early visualization and continuous tracking of risk targets. Based on the robot's kinematic constraint parameters, a spline interpolation algorithm is used to smoothly connect the full-dimensional attitude key point sequence to generate an emergency focusing path that includes time, position, velocity and gimbal angle commands; It should be noted that the kinematic constraint parameters of a robot refer to the limit values ​​set due to physical mechanical characteristics, such as maximum linear velocity, maximum angular acceleration, and minimum turning radius. Spline interpolation algorithms (such as B-Spline or Bézier curves) are mathematical smoothing techniques used to fit discrete key points in a polygonal pattern into a continuous and smooth curve. For example, the original key point sequence may be a right-angle turn (the robot cannot physically achieve an instantaneous 90-degree turn). The algorithm, combined with the parameter of "minimum turning radius of 0.5 meters", fits the right-angle connection into a circular arc trajectory and generates corresponding timestamp control commands (such as: reaching position A at 5 seconds, speed of 2 m / s, and maintaining a 30-degree elevation angle on the gimbal). The final emergency focusing path is not only a line, but also a complete control script that evolves with the time axis and conforms to the laws of physical motion.

[0028] In an optional embodiment, the robot is driven to interrupt its original trajectory according to the emergency focusing path, and a surround monitoring operation targeting the set of associated nodes is prioritized and executed according to the reconstructed inspection task queue, including: Upon receiving the emergency focus path as the highest priority control signal, immediately send a low-level interrupt command to the robot motion controller to clear the currently cached navigation waypoints and resolve the emergency focus path into a millisecond-level preemptive motion control flow; It should be noted that the robot motion controller is the core computing unit embedded in the robot's underlying layer. It is responsible for converting the trajectory coordinates generated by the upper-layer navigation algorithm into specific motor pulse signals in real time, and for handling millisecond-level safety logic such as emergency stops and fall prevention. The underlying interrupt instruction is a forced stop signal that directly acts on the robot motion control firmware layer. Its authority is higher than all application-layer navigation logic. It is used to ensure that the motor power output can be cut off and the instruction cache can be cleared instantly in critical moments. The preemptive motion control flow is an instruction set that does not require queuing and directly overwrites the controller register. For example, if the robot is currently executing the instruction "go to point A" and points B and C are still in the cache, once this signal is received, the controller will forcibly clear all data of A, B, and C within 10 milliseconds, set the motor speed to zero, and immediately fill in the first coordinate point of the emergency path. The preemptive motion control flow is sent to the chassis drive unit and gimbal servo unit to drive the robot to quickly approach the risk area along the planned path. When the deviation between the real-time positioning coordinates and the center of the local risk cluster is less than a preset threshold, the area positioning lock signal is triggered. It should be noted that the chassis drive unit and gimbal servo unit are the robot's actuators, responsible for position movement and viewing angle adjustment, respectively; the area positioning lock signal is a flag used to confirm that the robot has physically reached the optimal observation range, preventing the robot from taking pictures before it has come to a complete stop; for example, if the planned local risk cluster center coordinates are (20,20) and the system's error threshold is set to 0.5 meters, when the robot's GPS or laser positioning shows that the current position coordinates have entered the circular range of (20±0.5,20±0.5), the system automatically triggers the "Target_Locked" signal, notifying the upper-level software to stop the high-speed movement mode and prepare to enter the fine observation mode; In response to the area positioning and locking signal, the physical coordinate distribution of the associated node set in the reconstructed inspection task queue is retrieved, the minimum external field of view covering all associated nodes is calculated, and it is discretized and decomposed into a set of multi-point surrounding observation matrices for surrounding observation of this range. It should be noted that the minimum external field of view is a virtual cone or rectangle calculated based on computer vision geometry principles, aiming to encompass all scattered fault-related devices with the fewest possible camera shots. The multi-point surrounding observation matrix is ​​a sequence list composed of multiple specific gimbal shooting parameters (e.g., horizontal angles of 0°, 30°, and -30°), used to achieve multi-angle coverage of the field of view. For example, if the set of related nodes includes a "leaking valve" and a "smoking controller" that are 2 meters apart, the robot calculates that a single photo cannot capture them all, so it generates a minimum external field of view that covers both points and decomposes it into 3 shooting points: [Point 1: aligned with the left valve; Point 2: aligned with the middle area; Point 3: aligned with the right controller], ensuring no blind spots. The multi-point surrounding observation matrix is ​​transformed into a combination sequence of gimbal attitude commands and chassis fine-tuning commands, driving the robot to continuously perform multi-angle high-frequency sampling of the associated node set while maintaining focus on the center of the local risk cluster, thus completing the surrounding monitoring operation. It should be noted that the combined sequence of gimbal attitude commands and chassis fine-tuning commands refers to a complex action flow in which, in order to obtain the best shooting angle, the robot not only rotates the camera but also moves the vehicle body slightly forward, backward, left, and right, i.e., chassis fine-tuning. Multi-angle high-frequency sampling refers to continuously taking multiple high-definition images or infrared thermal images in a short period of time. For example, when executing the above matrix, the robot first stays in place, and the gimbal turns 30 degrees to the left to take a picture (command 1); then the chassis moves 0.2 meters to the right to avoid reflection, and at the same time the gimbal turns 30 degrees to the right to take a picture (command 2). This collaborative method simulates the action of a human maintenance worker leaning to the side to look, ensuring that the collected data is clear and comprehensive, and can accurately reflect the details of the fault.

[0029] Example 2, please refer to Figure 2The present invention provides a technical solution: a robot applicable to the above-mentioned control method for online monitoring robots, comprising: Data acquisition module 1 is used to acquire the real-time environmental monitoring data stream collected by the online monitoring robot for the current inspection node, and extract the topology identity ID based on the current inspection node; Deviation calculation module 2 is used to calculate the abnormal deviation index based on the comparison results between the real-time environmental monitoring data stream and the preset safety benchmark value. Topology locking module 3 is used to retrieve the preset device physical connection adjacency table with the topology identity ID as the index when the abnormal deviation index exceeds the preset threshold, and lock the set of associated nodes that have a physical connection relationship with the current inspection node. The spatiotemporal reconstruction module 4 is used to obtain the original priority sorting code of the associated node set in the preset global task queue, perform a cascading priority replacement operation, and uniformly promote the priority sorting code of the associated node set to a burst level higher than the original maximum priority in the global task queue; obtain the spatial coordinates corresponding to the associated node set, calculate the local risk cluster center based on the spatial coordinates, and generate a reconstruction inspection task queue. Focusing Path Module 5 is used to construct an emergency focusing path from the current location to the local risk cluster center based on the reconstructed inspection task queue and the local risk cluster center. The surround monitoring module 6 is used to drive the robot to interrupt its original trajectory according to the emergency focusing path, and to prioritize the execution of surround monitoring operations for the set of associated nodes according to the reconstructed inspection task queue.

[0030] Example 3, please refer to Figure 3 This invention provides a schematic diagram of the hardware entity of a robot according to Embodiment 2. The hardware entity of the robot includes: a processor 7, a communication interface 8, and a memory 9, wherein:

[0031] The processor typically controls the overall operation of the robot; the communication interface 8 enables the robot to communicate with other terminals or servers via a network; the memory 9 is configured to store instructions and applications executable by the processor, and can also cache data to be processed or already processed by the processor 7 and various modules in the robot, which can be implemented through flash memory or random access memory. Data can be transferred between the processor 7, communication interface 8, and memory 9 via a bus.

[0032] In embodiment four, this invention also provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the control method for an online monitoring robot as described in embodiment one. The storage medium may include various media capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0033] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A control method for an online monitoring robot, characterized in that, include: Acquire the real-time environmental monitoring data stream collected by the online monitoring robot for the current inspection node, and extract the topology identity ID based on the current inspection node; Based on the comparison results between real-time environmental monitoring data streams and preset safety benchmark values, an anomaly deviation index is calculated. If the abnormal deviation index exceeds the preset threshold, the preset device physical connection adjacency table is retrieved using the topology identity ID as the index to lock the set of associated nodes that have a physical connection relationship with the current inspection node; Obtain the original priority sorting code of the set of associated nodes in the preset global task queue, perform a cascading priority replacement operation to uniformly promote the priority sorting code of the set of associated nodes to a higher level than the original maximum priority in the global task queue; obtain the spatial coordinates corresponding to the set of associated nodes, calculate the local risk cluster center based on the spatial coordinates, and generate a reconstructed inspection task queue. Based on the reconstructed inspection task queue and local risk cluster center, an emergency focusing path is constructed from the current location to the local risk cluster center; Based on the emergency focus path, the robot is driven to interrupt its original trajectory and prioritize the execution of encircling monitoring operations targeting the set of related nodes according to the reconstructed inspection task queue.

2. The control method based on an online monitoring robot according to claim 1, characterized in that, The real-time environmental monitoring data stream includes at least one or more combinations of visible light image data, infrared thermal imaging temperature data, gas concentration data, ambient temperature and humidity data, and acoustic signature monitoring data. Acquire the real-time environmental monitoring data stream collected by the online monitoring robot for the current inspection node, and extract the topology identity ID based on the current inspection node, including: The robot's positioning and recognition module reads the preset location marker beacon at the current inspection node, analyzes the radio frequency or visual features of the location marker beacon, and generates the node's physical location code at the current moment. The node physical location code is used as the query key value. A unique match is performed in the pre-set digital factory topology map database to extract the logical node number bound to the node physical location code, which is used as the topology identity ID. Based on the topology identity ID, the robot invokes a preset sensor configuration strategy to control the sensor modules on the robot to synchronously trigger sampling of the target device and obtain raw data from multiple sources. The topology identity ID and the timestamp information of the acquisition time are written into the header frame of the multi-source sensor raw data, and the multi-source sensor raw data is spatiotemporally aligned to generate a real-time environmental monitoring data stream.

3. The control method based on an online monitoring robot according to claim 2, characterized in that, Based on the comparison results between real-time environmental monitoring data streams and preset safety benchmark values, an anomaly deviation index is calculated, including: The frame structure of the real-time environmental monitoring data stream is parsed to separate the monitoring values ​​of different sensor channels and the corresponding data type identification codes. Based on the data type identifier, the corresponding parameter threshold range is retrieved from the preset safety benchmark database. The sub-monitoring values ​​are compared and normalized with the parameter threshold range to generate a dimensionless deviation vector that eliminates the influence of physical units. Based on the data type identifier code matching the preset fault propagation sensitivity weight, the fault propagation sensitivity weight is multiplied point-to-point with each component in the dimensionless deviation vector to generate a weighted risk factor sequence that reflects the degree of potential impact of each parameter on the associated node. Perform modulus calculation or weighted summation on the weighted risk factor sequence to generate a single value that represents the overall severity of the anomaly at the current inspection node, namely the anomaly deviation index.

4. The control method based on an online monitoring robot according to claim 3, characterized in that, If the abnormal deviation index exceeds a preset threshold, the preset device physical connection adjacency table is retrieved using the topology identity ID as an index to lock the set of associated nodes that have a physical connection relationship with the current inspection node, including: The abnormal deviation index is compared with the system's preset alarm threshold. When the abnormal deviation index is greater than the alarm threshold, a topology search trigger command with the current timestamp is generated. In response to the topology search trigger command, the topology identity ID is used as the unique hash key value to perform a row-level scan of the preset device physical connection adjacency table, locate and lock the target node index row that completely matches the unique hash key value; Read the physical connection pointer data stored in the target node index row, parse out the identity codes of all directly adjacent nodes pointed to by the data, and generate an original adjacency code sequence containing multiple node codes; The original adjacency coding sequence is validated and deduplicated. After removing invalid codes, the remaining valid node codes are identified as the target objects to be inspected and packaged into a set of associated nodes.

5. The control method based on an online monitoring robot according to claim 4, characterized in that, Obtain the original priority sorting codes of the associated node set in the preset global task queue, perform a cascading priority replacement operation, and uniformly raise the priority sorting codes of the associated node set to a burst level higher than the original highest priority in the global task queue, including: Iterate through the priority fields of all tasks to be executed in the global task queue, filter out the highest priority value in the current queue by numerical comparison, and add a preset burst response weight to generate a cascaded replacement baseline value. Obtain the priority sorting code of each associated node in the set of associated nodes in its original state, calculate the numerical difference of each associated node relative to the lowest priority node in the set, and generate an internal relative position difference sequence. Each difference in the internal relative position difference sequence is superimposed on the cascade permutation benchmark value to generate a new burst reset priority code, and this code is used to cover the corresponding original priority sorting code in the associated node set.

6. The control method based on an online monitoring robot according to claim 5, characterized in that, Obtain the spatial coordinates corresponding to the set of associated nodes, and calculate the local risk cluster center based on the spatial coordinates, including: Using the topological identity ID in the set of associated nodes as an index, the system retrieves the preset 3D digital factory map database and extracts the 3D physical coordinate vectors and equipment monitoring face orientation information corresponding to each associated node. Obtain the robot's current real-time position coordinates, calculate the Euclidean distance between the 3D physical coordinate vector and the real-time position coordinates, and define the normalized reciprocal of the Euclidean distance as the spatial proximity weight. By using spatial proximity weights to perform weighted average geometric operations on the three-dimensional physical coordinate vectors, the coordinates of the local risk cluster center representing the physical geometric center of the risk area are synthesized and output.

7. The control method based on an online monitoring robot according to claim 6, characterized in that, Generate a refactoring inspection task queue, including: Get all task data after priority replacement and currently interrupted pending task data, sort them in descending order according to the value of priority sort code, and divide the all task data into a high-priority related sequence at the head and a normal pending sequence at the tail. Based on the coordinates of the local risk cluster centers obtained from the above calculation, navigation guidance anchors are constructed, and the navigation guidance anchors are encapsulated and inserted into the head position of the high-quality association sequence to generate a guided high-quality sequence. Establish a linked list pointer index, point the tail pointer of the guiding high-priority sequence to the head pointer of the normal pending sequence, complete the logical concatenation, and generate the final reconstructed inspection task queue that can be parsed by the navigation system.

8. The control method based on an online monitoring robot according to claim 7, characterized in that, Based on the reconstructed inspection task queue and local risk cluster centers, an emergency focusing path is constructed from the current location to the local risk cluster center, including: The robot's current location coordinates are used as the starting point, and the coordinates of the center of the local risk cluster are used as the ending point. A spatial guidance vector connecting the starting point and the ending point is constructed, and a navigation search corridor covering the area along the way is generated based on this vector. The navigation search corridor is mapped onto a pre-set 3D environment grid map. Static obstacles occupying grids and dynamic restricted areas are removed, and the remaining continuous free grids are extracted to generate a collision-free passage topology. Traverse the high-priority associated nodes at the head of the reconstructed inspection task queue, extract the corresponding equipment monitoring face orientation information, map the orientation information to the path nodes of the collision-free passage topology, and generate a set of full-dimensional attitude key point sequences that satisfy both passage safety and early visualization of risk targets. Based on the robot's kinematic constraint parameters, a spline interpolation algorithm is used to smoothly connect the full-dimensional attitude key point sequence to generate an emergency focusing path that includes time, position, velocity, and gimbal angle commands.

9. The control method based on an online monitoring robot according to claim 8, characterized in that, Based on the emergency focus path, the robot interrupts its original trajectory and prioritizes the execution of encircling monitoring operations targeting the set of related nodes according to the reconstructed inspection task queue, including: Upon receiving the emergency focus path as the highest priority control signal, immediately send a low-level interrupt command to the robot motion controller to clear the currently cached navigation waypoints and resolve the emergency focus path into a millisecond-level preemptive motion control flow; The preemptive motion control flow is sent to the chassis drive unit and gimbal servo unit to drive the robot to quickly approach the risk area along the planned path. When the deviation between the real-time positioning coordinates and the center of the local risk cluster is less than a preset threshold, the area positioning lock signal is triggered. In response to the area positioning and locking signal, the physical coordinate distribution of the associated node set in the reconstructed inspection task queue is retrieved, the minimum external field of view covering all associated nodes is calculated, and it is discretized and decomposed into a set of multi-point surrounding observation matrices for surrounding observation of this range. The multi-point encirclement observation matrix is ​​transformed into a combination sequence of gimbal attitude commands and chassis fine-tuning commands, which drives the robot to continuously perform multi-angle high-frequency sampling of the associated node set while maintaining focus on the center of the local risk cluster, thus completing the encirclement monitoring operation.

10. A robot, applicable to the control method for an online monitoring robot as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire the real-time environmental monitoring data stream collected by the online monitoring robot for the current inspection node, and extract the topology identity ID based on the current inspection node; The deviation calculation module is used to calculate the abnormal deviation index based on the comparison results between the real-time environmental monitoring data stream and the preset safety benchmark value. The topology locking module is used to retrieve the preset device physical connection adjacency table using the topology identity ID as the index when the abnormal deviation index exceeds the preset threshold, and lock the set of associated nodes that have a physical connection relationship with the current inspection node. The spatiotemporal reconstruction module is used to obtain the original priority sorting code of the set of associated nodes in the preset global task queue, perform a cascading priority replacement operation, and uniformly promote the priority sorting code of the set of associated nodes to a burst level higher than the original maximum priority in the global task queue; obtain the spatial coordinates corresponding to the set of associated nodes, calculate the local risk cluster center based on the spatial coordinates, and generate a reconstruction inspection task queue. The focused pathfinding module is used to construct an emergency focused path from the current location to the local risk cluster center based on the reconstructed inspection task queue and the local risk cluster center. The surround monitoring module is used to drive the robot to interrupt its original trajectory according to the emergency focusing path, and to prioritize the execution of surround monitoring operations on the set of related nodes according to the reconstructed inspection task queue.

Images