An explosion-proof robot intelligent inspection method, device and medium

Abstract

The application relates to the technical field of industrial robots, and particularly provides an intelligent inspection method and device of an explosion-proof robot, and a medium, the method comprising the following steps: S1, acquiring real-time multi-modal data of a to-be-inspected node to be passed through; S2, generating a real-time comprehensive risk value according to a preset importance weight and the real-time multi-modal data; S3, when the real-time comprehensive risk value is greater than a pre-labeled comprehensive risk value corresponding to a next to-be-inspected node, generating a first weighted moving cost corresponding to each path edge according to the real-time comprehensive risk values or the pre-labeled comprehensive risk values corresponding to all the to-be-inspected nodes; S4, generating a second inspection path passing through all the to-be-inspected nodes and having a minimum sum of the first weighted moving costs; and S5, controlling the explosion-proof robot to move according to the second inspection path, and returning to step S2; the method can perform dynamic risk assessment and path optimization according to real-time environmental data.

Description

Technical Field

[0001] This application relates to the field of industrial robot technology, and more specifically, to an intelligent inspection method, equipment, and medium for explosion-proof robots. Background Technology

[0002] In the high-risk environment of petrochemical terminals, traditional manual inspection methods are not only inefficient but also expose workers to potential dangers of flammable, explosive, toxic, and hazardous substances. In recent years, with the development of robotics technology, explosion-proof inspection robots have been gradually applied to petrochemical terminals to replace manual labor in daily inspections. Existing technologies largely rely on pre-programmed fixed inspection paths, combined with simple sensors (such as visible light cameras, infrared thermal imagers, and gas sensors) for data collection and anomaly detection. However, these solutions have significant shortcomings in terms of real-time performance, intelligence, and emergency response capabilities.

[0003] Existing explosion-proof robot inspection technologies suffer from the following main shortcomings: 1. Isolated multi-sensor information, failing to achieve effective fusion. Specifically, although current systems are equipped with multiple sensors, each sensor typically operates independently, lacking deep correlation and collaborative analysis between data. For example, when a gas sensor detects a leak, the video sensor cannot automatically locate the leak source and track it. Environmental perception data and image recognition results are fragmented, making it difficult to form a comprehensive and accurate understanding of the on-site environment; 2. Static risk assessment, lacking dynamic adaptability. Specifically, existing technologies mostly rely on preset fixed inspection routes and... The existing inspection algorithms fail to dynamically adjust their inspection strategies based on real-time environmental changes and risk factors (such as sudden changes in gas concentration and abnormal temperature increases). This results in the robot's inability to identify and prioritize high-risk events, leading to delays in responding to major hazards. Furthermore, the path planning is severely disconnected from the importance of the equipment. Specifically, the criticality of equipment within the petrochemical terminal (such as storage tanks, oil conveying arms, and fire protection systems) varies significantly, and the consequences of their failures differ greatly. However, existing inspection algorithms often prioritize the shortest path or full coverage, failing to consider the different risk levels of different equipment. This may result in critical equipment not being inspected in a timely and sufficient manner, leaving safety hazards unaddressed.

[0004] There is currently no effective technical solution to the above problems. Summary of the Invention

[0005] The purpose of this application is to provide an intelligent inspection method, equipment and medium for explosion-proof robots, which can perform dynamic risk assessment and path optimization based on real-time environmental data.

[0006] In a first aspect, this application provides an intelligent inspection method for explosion-proof robots, applied to explosion-proof robots equipped with a sensor array. The intelligent inspection method for explosion-proof robots includes the following steps:

[0007] S1. When the explosion-proof robot moves along the pre-planned first inspection path, it uses a sensor group to acquire real-time multimodal data of the nodes to be inspected along the way.

[0008] S2. Generate a real-time comprehensive risk value based on the preset importance weights of the nodes to be inspected along the route and real-time multimodal data;

[0009] S3. When the real-time comprehensive risk value is greater than the pre-calibrated comprehensive risk value corresponding to the next node to be inspected in the first inspection path, generate the first weighted movement cost corresponding to each path edge according to the real-time comprehensive risk value or pre-calibrated comprehensive risk value corresponding to all nodes to be inspected.

[0010] S4. Generate a second inspection path that passes through all nodes to be inspected and has the minimum sum of the first weighted movement costs based on all the first weighted movement costs.

[0011] S5. Control the explosion-proof robot to move along the second inspection path and return to step S2.

[0012] This application provides an intelligent inspection method for explosion-proof robots, which realizes intelligent risk response by dynamically generating inspection paths based on real-time comprehensive risk values. Since this application can perform dynamic risk assessment and path optimization based on real-time environmental data, it can effectively improve inspection efficiency and safety, thereby effectively solving the problems of insufficient multi-sensor information fusion, static risk assessment, and disconnect between path planning and equipment importance.

[0013] Optionally, the pre-planning process for the first inspection path includes:

[0014] A1. Control the explosion-proof robot to move along the preset path that passes through all inspection nodes, and use the sensor group to obtain the calibration multimodal data corresponding to each inspection node.

[0015] A2. For each inspection node, generate a pre-calibrated comprehensive risk value based on the preset importance weight and calibrated multimodal data corresponding to that node;

[0016] A3. Generate the predefined second weighted movement cost for each path edge based on the pre-calibrated comprehensive risk value corresponding to all nodes to be inspected.

[0017] A4. Generate the first inspection path that passes through all inspection nodes and has the smallest sum of second-weighted movement costs based on all second-weighted movement costs.

[0018] This technical solution introduces a pre-planning process for the first inspection path into the intelligent inspection method for explosion-proof robots. This ensures that the initial inspection path used by the robot is no longer simply the shortest path or full coverage, but rather fully considers the preset importance weights of each inspection node and the risk level assessed through calibrated multimodal data. This mechanism, combining pre-planning with real-time adjustment, allows the explosion-proof robot to perform inspections on a risk-optimized initial path from the start. This provides a solid foundation and reference for subsequent real-time risk perception and dynamic path adjustment, thereby significantly improving the intelligence level and emergency response capability of the entire inspection system.

[0019] Optionally, the process of generating the pre-calibrated comprehensive risk value is the same as the process of generating the real-time comprehensive risk value. The calibration multimodal data includes calibration sensor data corresponding to multiple modes. The process of generating the pre-calibrated comprehensive risk value includes:

[0020] B1. For each mode, obtain the standard deviation of the calibration sensor data corresponding to that mode within a preset time window;

[0021] B2. Calculate the percentage of standard deviation for each mode based on all standard deviations;

[0022] B3. For each mode, calculate the fusion weight corresponding to that mode based on the proportion of the standard deviation corresponding to that mode;

[0023] B4. For each inspection node, the data from all calibrated sensors corresponding to that node are weighted and fused according to all the fusion weights of that node to obtain a comprehensive environmental index.

[0024] B5. Obtain the cumulative running time of each inspection node and the cumulative number of abnormal events in its area;

[0025] B6. For each inspection node, calculate the dynamic risk value based on the comprehensive environmental indicators, cumulative operating time, and cumulative number of abnormal events corresponding to that node. Then, calculate the pre-calibrated comprehensive risk value based on the preset importance weight, preset dynamic risk weight, preset basic risk coefficient, and dynamic risk value corresponding to that node.

[0026] Optionally, step B3 includes:

[0027] B31. For each mode, calculate the entropy value corresponding to that mode based on the proportion of the standard deviation corresponding to that mode, and then calculate the fusion weight corresponding to that mode based on the entropy values ​​corresponding to all modes.

[0028] Alternatively, the formula for calculating the percentage of standard deviation is as follows:

[0029] ;

[0030] Where, p i σ represents the proportion of the standard deviation corresponding to the i-th mode. i σ represents the standard deviation of the calibration sensor data corresponding to the i-th mode within a preset time window. j The standard deviation of the calibration sensor data corresponding to the j-th mode is within a preset time window, and N represents the number of modes contained in the calibration multimodal data corresponding to the inspection node.

[0031] The formula for calculating entropy is shown below:

[0032] ;

[0033] Among them, e i This represents the entropy value corresponding to the i-th mode, where k is a constant;

[0034] The formula for calculating the fusion weight is as follows:

[0035] ;

[0036] Where, w i e represents the fusion weight corresponding to the i-th mode. j This represents the entropy value corresponding to the j-th mode;

[0037] The formula for calculating the comprehensive environmental indicators is as follows:

[0038] ;

[0039] Where E represents the comprehensive environmental index, f i This represents the calibration sensor data for the i-th mode in the calibration multimodal data corresponding to the inspection node.

[0040] Optionally, the formula for calculating the dynamic risk value is as follows:

[0041] ;

[0042] Among them, D j T represents the dynamic risk value corresponding to the j-th inspection node, α, β, and γ are the first preset coefficient, the second preset coefficient, and the third preset coefficient, respectively. j E represents the cumulative running time corresponding to the j-th inspection node. j A represents the dynamic risk value corresponding to the j-th inspection node. j This represents the cumulative number of abnormal times corresponding to the j-th inspection node;

[0043] The formula for calculating the pre-calibrated comprehensive risk value is shown below:

[0044] ;

[0045] Among them, R j S represents the pre-calibrated comprehensive risk value corresponding to the j-th inspection node. j λ represents the preset importance weight corresponding to the j-th inspection node. j M represents the preset dynamic risk weight corresponding to the j-th inspection node. j This represents the preset basic risk coefficient corresponding to the j-th inspection node.

[0046] Optionally, the formula for calculating the first weighted movement cost is the same as the formula for calculating the second weighted movement cost, and the formula for calculating the second weighted movement cost is as follows:

[0047] ;

[0048] in, This represents the second weighted movement cost corresponding to the predefined path edge connecting the k-th inspection node and the l-th inspection node. R represents the predefined basic movement cost corresponding to the path edge connecting the k-th inspection node and the l-th inspection node. k R represents the pre-calibrated comprehensive risk value corresponding to the k-th inspection node. l R represents the pre-calibrated comprehensive risk value corresponding to the l-th inspection node. avg η represents the average of all pre-calibrated comprehensive risk values, and η represents the pre-defined risk weighting factor.

[0049] Optionally, step S1 includes:

[0050] S11. When the explosion-proof robot moves along the pre-planned first inspection path, it uses a sensor group to acquire real-time multimodal data of the nodes to be inspected along the way.

[0051] S12. Preprocess the real-time multimodal data; preprocessing includes denoising, standardization, and image registration.

[0052] Secondly, this application provides an electronic device including a processor and a memory, the memory storing computer-readable instructions, which, when executed by the processor, perform the steps of the method provided in the first aspect above.

[0053] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the method provided in the first aspect above.

[0054] As can be seen from the above, the intelligent inspection method, equipment and medium for explosion-proof robots provided by this application realizes intelligent risk response by dynamically generating inspection paths based on real-time comprehensive risk values. Since this application can perform dynamic risk assessment and path optimization based on real-time environmental data, it can effectively improve inspection efficiency and safety, thereby effectively solving the problems of insufficient multi-sensor information fusion, static risk assessment and disconnect between path planning and equipment importance. Attached Figure Description

[0055] Figure 1 A flowchart of an intelligent inspection method for an explosion-proof robot provided in an embodiment of this application.

[0056] Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0057] Reference numerals: 101, processor; 102, memory; 103, communication bus. Detailed Implementation

[0058] The technical solutions of the embodiments of this application 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 this application, and not all of the embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0059] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0060] In a first aspect, this application provides an intelligent inspection method for explosion-proof robots, applied to explosion-proof robots equipped with a sensor array. The intelligent inspection method for explosion-proof robots includes the following steps:

[0061] S1. When the explosion-proof robot moves along the pre-planned first inspection path, it uses a sensor group to acquire real-time multimodal data of the nodes to be inspected along the way.

[0062] S2. Generate a real-time comprehensive risk value based on the preset importance weights of the nodes to be inspected along the route and real-time multimodal data;

[0063] S3. When the real-time comprehensive risk value is greater than the pre-calibrated comprehensive risk value corresponding to the next node to be inspected in the first inspection path, generate the first weighted movement cost corresponding to each path edge according to the real-time comprehensive risk value or pre-calibrated comprehensive risk value corresponding to all nodes to be inspected.

[0064] S4. Generate a second inspection path that passes through all nodes to be inspected and has the minimum sum of the first weighted movement costs based on all the first weighted movement costs.

[0065] S5. Control the explosion-proof robot to move along the second inspection path and return to step S2.

[0066] For ease of understanding, some key terms in this embodiment are explained below. The explosion-proof robot in this embodiment is a robot specifically designed to perform tasks in flammable and explosive environments. Its structure and electrical system are specially treated to prevent the generation of sparks or overheating, thereby avoiding the ignition of surrounding flammable materials. This explosion-proof robot is capable of autonomous movement, data acquisition, and environmental monitoring. The sensor group in this embodiment is a series of different types of sensors integrated on the explosion-proof robot to acquire multimodal data from the environment. These sensors may include, but are not limited to, visible light cameras, infrared thermal imagers, gas sensors (such as methane sensors, hydrogen sulfide sensors, and carbon monoxide sensors), temperature sensors, humidity sensors, and acoustic sensors. The various sensors in this embodiment work together to provide the explosion-proof robot with comprehensive and real-time environmental perception capabilities of the inspection area. The nodes to be inspected in this embodiment refer to key locations or equipment points pre-set on the inspection path of the explosion-proof robot. These nodes typically correspond to equipment, pipelines, valves, and storage tanks that require key monitoring, or areas with high environmental risks. During the inspection process, the explosion-proof robot will collect multimodal data when stopping or passing through these nodes. The real-time multimodal data in this embodiment refers to environmental data (such as images, videos, gas concentrations, temperature, humidity, and sound) acquired in real time by the explosion-proof robot through its sensor array during inspection. This data is dynamically changing and reflects the environmental conditions of the nodes to be inspected that the explosion-proof robot passes through at the current moment. The preset importance weight in this embodiment is a pre-set value based on factors such as the criticality of the equipment or area represented by the node to be inspected within the entire industrial system, its potential risk level, and the possible consequences of failure. Specifically, a higher importance weight indicates a more critical node that requires greater attention. The real-time comprehensive risk value in this embodiment is a quantitative indicator reflecting the current real-time risk level of the node, calculated using a specific algorithm by combining the preset importance weight of the node to be inspected and the real-time multimodal data. This risk value dynamically reflects environmental changes and potential hazards at the node. The pre-calibrated comprehensive risk value in this embodiment refers to the comprehensive risk value calculated as a benchmark or reference by collecting calibration multimodal data and combining it with preset importance weights during the initial inspection or periodic calibration of the explosion-proof robot. This value represents the inherent risk level of the inspection node under normal or stable conditions. The first weighted movement cost in this embodiment refers to the cost used to evaluate moving from one node to another, calculated based on the real-time comprehensive risk value or pre-calibrated comprehensive risk value of the node to be inspected and the preset basic movement cost when dynamically adjusting the inspection path. Specifically, the higher the risk value, the lower the corresponding first weighted movement cost, so that the path planning algorithm preferentially selects the path edge passing through the high-risk node.The second inspection path in this embodiment refers to the inspection path calculated (in the prior art) by a path planning algorithm based on the dynamically generated first weighted movement cost in the intelligent inspection method for explosion-proof robots. This path is the result of dynamic adjustment and is designed to prioritize the handling of high-risk areas.

[0067] This application proposes an intelligent inspection method for explosion-proof robots. The method is applied to explosion-proof robots equipped with a sensor array. The method aims to solve the problems of isolated sensor data, static risk assessment, and path planning that do not consider the differences in the importance of equipment in existing explosion-proof robot inspections, so as to achieve dynamic intelligent inspection.

[0068] Specifically, in step S1, as the explosion-proof robot moves along the pre-planned first inspection path, the system uses a sensor array to acquire real-time multimodal data of the nodes to be inspected along the way. The sensor array can include various types such as visible light cameras, infrared thermal imagers, gas sensors, and temperature sensors. For example, a visible light camera can acquire video images of the scene to identify abnormalities in the appearance of equipment; an infrared thermal imager can detect the temperature distribution on the surface of equipment to identify hot spots; and a gas sensor can monitor the concentration of flammable or toxic gases in the environment in real time. These sensors can work independently or collect data through a data interface and aggregate it to the processing unit of the explosion-proof robot for preliminary integration. In step S2, the system generates a real-time comprehensive risk value based on the preset importance weights of the nodes to be inspected along the route and real-time multimodal data. The preset importance weights can be manually set or evaluated by an expert system based on factors such as the type of equipment, its criticality in the production process, and historical failure rate. For example, the importance weight of a valve in a main oil pipeline may be much higher than that of an ordinary lighting fixture. It should be understood that the setting of preset importance weights is existing technology, and its working principle and workflow will not be discussed in detail here. Preferably, this embodiment can first extract multimodal features (such as instantaneous peak values ​​of gas concentration, abnormal temperature rise, smoke or liquid leakage identified in the image, etc.) from the real-time multimodal data, and then calculate a comprehensive risk value reflecting the real-time risk level of the nodes to be inspected along the route based on the preset importance weights and multimodal features. In step S3, when the real-time comprehensive risk value is greater than the pre-calibrated comprehensive risk value corresponding to the next node to be inspected in the first inspection path, a predefined first weighted movement cost is generated for each path edge based on the real-time comprehensive risk value or the pre-calibrated comprehensive risk value corresponding to all nodes to be inspected. Specifically, when the real-time risk value of the node to be inspected that the explosion-proof robot passes through is higher than the pre-calibrated comprehensive risk value corresponding to the next node to be inspected, it indicates that there may be an abnormal situation at the node to be inspected that the robot needs to handle first. At this time, the system will dynamically adjust the movement cost of the path edges connecting each node based on the risk value of all nodes (including the real-time comprehensive risk value and the pre-calibrated comprehensive risk value). For example, the path cost of the path edge between nodes with high risk values ​​will be reduced to guide the robot to go to that path edge first.In step S4, the system generates a second inspection path that passes through all nodes to be inspected and minimizes the sum of the first weighted movement costs, based on all the first weighted movement costs. After generating the dynamically adjusted weighted movement costs, the explosion-proof robot uses a path planning algorithm (such as Dijkstra's algorithm, A* algorithm, genetic algorithm, etc.) to recalculate a new inspection path. This new path comprehensively considers the risk status and movement costs of all nodes to be inspected, aiming to find a path that effectively covers all nodes to be inspected while minimizing the total movement cost. Here, minimization means, under the premise of preferentially reaching high-risk nodes to be inspected, minimizing inspection time and energy consumption as much as possible. In step S5, the system controls the explosion-proof robot to move according to the second inspection path and returns to step S2. Specifically, once the new second inspection path is generated, the explosion-proof robot receives the instruction and moves according to this new path. During the movement, the explosion-proof robot continuously uses the sensor group to acquire real-time multimodal data of the nodes it passes through and repeats step S2 and subsequent steps, forming a closed-loop dynamic inspection process. This cyclical mechanism enables the explosion-proof robot to continuously adapt to environmental changes and adjust its inspection strategy in real time, ensuring timely response and effective monitoring of high-risk areas.

[0069] The following is a more specific example to illustrate the above technical solution in more detail: Suppose that in a petrochemical terminal area, an explosion-proof robot needs to inspect multiple storage tanks, oil conveying arms and valves. These devices are defined as nodes to be inspected and are assigned different preset importance weights according to their criticality in production. For example, the main oil conveying arm has the highest importance weight, followed by large storage tanks, while ordinary valves have a relatively low importance weight.

[0070] First, the system pre-plans a primary inspection path, based on historical data and experience, designed to cover all nodes to be inspected. As the explosion-proof robot moves along this path, its onboard sensor array (including a visible light camera, an infrared thermal imager, and a methane gas sensor) acquires real-time multimodal data of each node it passes. For example, as the robot moves towards tank B, it passes tank A. Upon passing tank A, the camera captures its appearance, the infrared thermal imager scans its surface temperature, and the gas sensor detects the methane concentration around tank A. Next, for each node passed, the system generates a real-time comprehensive risk value based on its preset importance weight and the real-time multimodal data. For instance, if the methane concentration near tank A suddenly increases, or the infrared thermal imager detects a localized temperature anomaly, even if tank A's importance weight is not the highest, its real-time comprehensive risk value will significantly increase due to these abnormal data. Subsequently, the system compares the real-time comprehensive risk value of the current node with the pre-calibrated comprehensive risk value corresponding to tank B (the next node to be inspected in the first inspection path). The pre-calibrated comprehensive risk value is a risk baseline obtained under normal operating conditions through historical data or benchmark tests. If the real-time comprehensive risk value of tank A (e.g., reaching 80 due to increased methane concentration) is greater than the pre-calibrated comprehensive risk value of tank B (e.g., 50), it indicates that there is a potential risk in the area of ​​tank A. At this time, the system generates a predefined first weighted movement cost corresponding to each path edge based on the current real-time comprehensive risk value of all nodes to be inspected (for nodes that have passed through) or the pre-calibrated comprehensive risk value (for nodes that have not been reached or passed through). For example, if the risk value of tank A is high, the movement cost of the path edge connected to tank A will be reduced to guide the robot to go there first. Then, based on all these dynamically generated first weighted movement costs, the system uses a path planning algorithm to calculate a second inspection path that passes through all the nodes to be inspected and minimizes the sum of the first weighted movement costs. This second inspection path directs the explosion-proof robot to inspect tank A, and then to inspect tank B. Finally, the system controls the explosion-proof robot to move along this newly generated second inspection path. As the robot moves along the second inspection path, it continuously acquires real-time multimodal data of the nodes it passes through and repeats the risk assessment and path planning steps described above. This closed-loop dynamic adjustment mechanism enables the explosion-proof robot to continuously adapt to environmental changes, respond to emergencies in real time, and ensure the efficiency and safety of the inspection task.

[0071] The intelligent inspection method for explosion-proof robots proposed in this application significantly improves the inspection capabilities of explosion-proof robots in high-risk environments by introducing real-time multimodal data acquisition, dynamic risk assessment, and a risk-based path optimization mechanism. Compared with traditional inspection methods that rely on pre-programmed fixed paths, this method overcomes the shortcomings of isolated sensor information, static risk assessment, and the disconnect between path planning and equipment importance. Specifically, in traditional solutions, multi-sensor data is often processed independently, making it difficult to form a comprehensive understanding of the on-site environment. This method utilizes a sensor array to acquire real-time multimodal data and uses it to generate a real-time comprehensive risk value, achieving effective fusion of multi-source information. This enables the robot to perceive environmental changes more comprehensively and accurately. For example, when a gas sensor detects an abnormal gas concentration, combining the temperature data from an infrared thermal imager and the image from a visible light camera can more accurately determine the leak source and degree, which is difficult to achieve with traditional solutions.

[0072] Furthermore, most existing technologies rely on preset fixed inspection routes and cycles, failing to dynamically adjust inspection strategies based on real-time changes in environmental risk factors. This method compares the real-time comprehensive risk value with the pre-calibrated comprehensive risk value. Once it is found that the real-time comprehensive risk value of the node to be inspected is higher than the pre-calibrated comprehensive risk value of the next node to be inspected, the path will be dynamically adjusted. This mechanism enables the explosion-proof robot to identify and prioritize high-risk events to avoid delays in responding to major hazards. For example, when the real-time risk value of a storage tank area traversed by the explosion-proof robot increases due to a sudden situation, the explosion-proof robot of this application can immediately adjust its path and prioritize going to that area for detailed inspection, instead of continuing the inspection according to the original plan.

[0073] Furthermore, traditional path planning often focuses on finding the shortest path or achieving full coverage, failing to adequately consider the varying risk levels of different equipment. This method, when generating the first weighted movement cost, fully considers the preset importance weights and real-time comprehensive risk values ​​of the nodes to be inspected. This allows path planning to prioritize critical equipment and high-risk areas, ensuring timely and thorough inspection of critical equipment and effectively reducing potential safety hazards. In this way, this method achieves a deep integration of inspection strategy and actual risk conditions, making the inspection work of explosion-proof robots more targeted and efficient.

[0074] In some preferred embodiments, the pre-planning process for the first inspection path includes:

[0075] A1. Control the explosion-proof robot to move along the preset path that passes through all inspection nodes, and use the sensor group to obtain the calibration multimodal data corresponding to each inspection node.

[0076] A2. For each inspection node, generate a pre-calibrated comprehensive risk value based on the preset importance weight and calibrated multimodal data corresponding to that node;

[0077] A3. Generate the predefined second weighted movement cost for each path edge based on the pre-calibrated comprehensive risk value corresponding to all nodes to be inspected.

[0078] A4. Generate the first inspection path that passes through all inspection nodes and has the smallest sum of second-weighted movement costs based on all second-weighted movement costs.

[0079] In step A1, the system controls the explosion-proof robot to move along a preset path passing through all inspection nodes, and uses a sensor array to acquire calibration multimodal data corresponding to each inspection node, aiming to provide basic data for subsequent risk assessment. The preset path in this embodiment can be an initial route determined by human experience, covering all inspection areas and equipment, or a fully covered path generated after preliminary environmental modeling. The calibration multimodal data in this embodiment refers to the multimodal sensor data collected at each inspection node during the robot's movement along the preset path. This data reflects the environmental characteristics of the inspection nodes under relatively stable conditions.

[0080] In step A2, for each inspection node, the system generates a pre-calibrated comprehensive risk value based on the preset importance weight and calibrated multimodal data corresponding to that node. This step aims to quantify the risk level of each inspection node. The method for generating the pre-calibrated comprehensive risk value in this embodiment is preferably the same as the method for generating the real-time comprehensive risk value, and will not be repeated here.

[0081] In step A3, the system generates a predefined second weighted movement cost for each path edge based on the pre-calibrated comprehensive risk value corresponding to all inspection nodes. This aims to integrate the risk information of the nodes into the cost calculation of path planning. In this embodiment, a path edge refers to a virtual or actual path segment connecting two inspection nodes. The generation method of the second weighted movement cost in this embodiment is preferably the same as the generation method of the first weighted movement cost, and will not be repeated here.

[0082] In step A4, the system generates a first inspection path that passes through all inspection nodes and minimizes the sum of the second-weighted movement costs based on all second-weighted movement costs. This step aims to determine an optimal initial inspection route. In this embodiment, the first inspection path is the initial inspection route that the explosion-proof robot will follow. The process of generating this path can employ classic graph theory algorithms, such as variations of the Traveling Salesman Problem (TSP), Dijkstra's algorithm, A* algorithm, or genetic algorithms. The goal of these algorithms is to find a path that minimizes the sum of the second-weighted movement costs corresponding to all path edges, while ensuring that the robot can access all inspection nodes. In this way, the first inspection path generated in this application not only considers movement efficiency but also incorporates the risk information of each inspection node, thus achieving a balance between risk and efficiency.

[0083] This embodiment solves the problem of static initial path planning by refining the process of pre-planning the first inspection path, achieving risk-based dynamic optimization. Specifically, step A1 controls the explosion-proof robot to move along the preset path and acquire calibrated multimodal data to provide comprehensive basic data for risk assessment and ensure that subsequent calculations are based on actual environmental indicators. Step A2 generates a pre-calibrated comprehensive risk value based on preset importance weights and calibrated multimodal data. This step combines equipment criticality (reflected by preset importance weights) and environmental changes (reflected by calibrated multimodal data) to achieve dynamic risk quantification. Step A3 generates a second weighted movement cost corresponding to each path edge based on all pre-calibrated comprehensive risk values, so that the path edge cost reflects the risk level and guides the path to prioritize high-risk inspection nodes. Step A4 generates the first inspection path with the minimum total risk, optimizing inspection efficiency and reducing safety hazards. The entire process emphasizes data-driven and risk-adaptive approaches, improving the intelligence and responsiveness of path planning.

[0084] This embodiment introduces a pre-planning process for the first inspection path into the intelligent inspection method for explosion-proof robots. This ensures that the initial inspection path used by the robot is no longer simply the shortest path or full coverage, but rather fully considers the preset importance weights of each inspection node and the risk level assessed through calibrated multimodal data. This mechanism, combining pre-planning with real-time adjustment, allows the explosion-proof robot to perform inspections on a risk-optimized initial path from the start. This provides a solid foundation and reference for subsequent real-time risk perception and dynamic path adjustment, thereby significantly improving the intelligence level and emergency response capability of the entire inspection system.

[0085] In some preferred embodiments, the process of generating the pre-calibrated comprehensive risk value is the same as the process of generating the real-time comprehensive risk value. The calibration multimodal data includes calibration sensor data corresponding to multiple modes. The process of generating the pre-calibrated comprehensive risk value includes:

[0086] B1. For each mode, obtain the standard deviation of the calibration sensor data corresponding to that mode within a preset time window;

[0087] B2. Calculate the percentage of standard deviation for each mode based on all standard deviations;

[0088] B3. For each mode, calculate the fusion weight corresponding to that mode based on the proportion of the standard deviation corresponding to that mode;

[0089] B4. For each inspection node, the data from all calibrated sensors corresponding to that node are weighted and fused according to all the fusion weights of that node to obtain a comprehensive environmental index.

[0090] B5. Obtain the cumulative running time of each inspection node and the cumulative number of abnormal events in its area;

[0091] B6. For each inspection node, calculate the dynamic risk value based on the comprehensive environmental indicators, cumulative operating time, and cumulative number of abnormal events corresponding to that node. Then, calculate the pre-calibrated comprehensive risk value based on the preset importance weight, preset dynamic risk weight, preset basic risk coefficient, and dynamic risk value corresponding to that node.

[0092] In step B1, when obtaining the standard deviation of the calibration sensor data corresponding to the mode within a preset time window, various methods can be used. For example, the sliding window method can be used to perform statistical calculations on the sensor data of each mode within the preset time window to obtain its standard deviation in order to capture the volatility of the data. Alternatively, the exponentially weighted moving average (EWMA) method can be used to assign different weights to historical data and calculate the weighted standard deviation to more sensitively reflect recent data changes and thus identify potential anomalies.

[0093] In step B2, when calculating the proportion of the standard deviation corresponding to each mode based on all standard deviations, the system can normalize the fluctuation contribution of each mode to facilitate a fair comparison of the importance of different modes. For example, the standard deviation of a single mode can be divided by the sum of the standard deviations of all modes to obtain its relative contribution to the overall fluctuation. Alternatively, the maximum-minimum normalization method can be used to map all standard deviations to the 0-1 interval, and then the proportion of each standard deviation in all standard deviations can be calculated.

[0094] When calculating the fusion weight for each mode based on the proportion of the standard deviation corresponding to that mode, step B3 aims to give higher weights to modes with large fluctuations. For example, a linear mapping can be performed directly based on the proportion of the standard deviation so that the larger the proportion, the greater the weight.

[0095] When performing weighted fusion of all calibrated sensor data corresponding to each inspection node based on all fusion weights corresponding to that node to obtain a comprehensive environmental index, step B4 can effectively integrate multi-source information to improve the comprehensiveness and consistency of environmental assessment. For example, the sensor data of each modality can be multiplied by its corresponding fusion weight and then summed to obtain the comprehensive environmental index of that node.

[0096] In step B5, when obtaining the cumulative running time of each inspection node and the cumulative number of abnormal events in its area, the system introduces historical dynamic factors to make up for the shortcomings of static evaluation. For example, it reads the cumulative running time of the equipment from the equipment management system and queries the historical abnormal event records of the area where the inspection node is located from the abnormal event database and performs cumulative statistics.

[0097] In step B6, for each inspection node, a dynamic risk value is calculated based on the corresponding comprehensive environmental indicators, cumulative operating time, and cumulative number of abnormal events. Then, a pre-calibrated comprehensive risk value is calculated based on the node's preset importance weight, preset dynamic risk weight, preset basic risk coefficient, and dynamic risk value. This ensures that the risk value is not only based on the current environmental state but also incorporates historical dynamics, thereby achieving more comprehensive and reliable risk quantification. For example, the dynamic risk value can be obtained by linearly weighted summing the comprehensive environmental indicators, cumulative operating time, and cumulative number of abnormal events; the pre-calibrated comprehensive risk value is calculated by multiplication or weighting based on this, combined with the preset importance weight, preset dynamic risk weight, and preset basic risk coefficient.

[0098] This embodiment addresses the problems of isolated multi-sensor information and static risk assessment in traditional inspection methods by deeply fusing multi-modal data and performing dynamic risk assessment. First, by acquiring the standard deviation of each modality's sensor data and calculating the fusion weight based on the percentage of standard deviation, the volatility of different modal data is quantified. Since greater volatility in sensor data indicates more drastic changes in the data monitored by that sensor and a greater amount of information contained in the data, this embodiment can assign higher weights to highly volatile modalities, ensuring that modal information that better reflects environmental anomalies or potential risks receives sufficient attention during data fusion. Subsequently, the system uses these fusion weights to weightedly fuse the calibrated sensor data of each modality, generating a comprehensive environmental index, thereby integrating scattered sensor information into a unified and comprehensive environmental assessment metric. Furthermore, this application introduces the cumulative operating time of inspection nodes and the cumulative number of abnormal events in their respective areas. These historical dynamic factors reflect the long-term health status of equipment and the cumulative risk of the area, effectively compensating for the shortcomings of relying solely on real-time environmental data for assessment. Ultimately, the system calculates a dynamic risk value by combining comprehensive environmental indicators, cumulative operating time, and the cumulative number of abnormal events. This value is then combined with preset importance weights, preset dynamic risk weights, and preset basic risk coefficients to generate a pre-calibrated comprehensive risk value. This process makes risk assessment no longer static and isolated, but dynamic and comprehensive. Therefore, the pre-calibrated comprehensive risk value and real-time comprehensive risk value in this embodiment can more accurately reflect the actual risk status of the inspection node. Since the pre-calibrated comprehensive risk value in this embodiment is directly applied to the pre-planning process of the first inspection path, and the pre-calibrated comprehensive risk value and real-time comprehensive risk value are jointly applied to path adjustment during real-time inspection, this embodiment can enable the explosion-proof robot to generate a more optimized inspection path based on the actual risk level of the equipment and dynamic changes in the environment by integrating more accurate and comprehensive pre-calibrated comprehensive risk values ​​and real-time comprehensive risk values ​​into path planning and adjustment. This effectively improves the intelligence level of inspection and emergency response capabilities.

[0099] In some preferred embodiments, step B3 includes:

[0100] B31. For each mode, calculate the entropy value corresponding to that mode based on the proportion of the standard deviation corresponding to that mode, and then calculate the fusion weight corresponding to that mode based on the entropy values ​​corresponding to all modes.

[0101] After acquiring the standard deviation of the calibration sensor data corresponding to each modality within a preset time window and calculating the proportion of the standard deviation for each modality, this embodiment calculates the entropy value corresponding to each modality based on its corresponding standard deviation proportion. The entropy value in this embodiment is an indicator used in information theory to measure the amount of information; here, it is used to quantify the richness or dispersion of information contained in each modality's data. This embodiment, by using the standard deviation proportion to calculate the entropy value, can more precisely evaluate the contribution and importance of each modality's data in the overall data. Subsequently, this application calculates the fusion weight corresponding to each modality based on the entropy values ​​corresponding to all modalities. This fusion weight is used to determine the contribution ratio of each modality's data in the final comprehensive index during multimodal data fusion. This embodiment, by using the entropy value to calculate the fusion weight, ensures that modalities with greater information content receive higher weights, thereby making the fusion result more accurate and representative.

[0102] This embodiment optimizes the determination process of fusion weights by introducing entropy calculation, making the weight allocation more objective and scientific, thereby improving the accuracy of the comprehensive environmental indicators. Specifically, after obtaining the standard deviation of the calibration sensor data corresponding to each modality within a preset time window and calculating the proportion of the standard deviation corresponding to each modality, this application calculates the entropy value corresponding to each modality based on its corresponding standard deviation proportion. As an indicator of information content, the entropy value can more precisely quantify the information contribution of each modality's data in the overall data. Subsequently, this application calculates the fusion weight corresponding to each modality based on the entropy values ​​corresponding to all modalities. This entropy-based weight calculation method allows modalities with greater information content to obtain higher fusion weights, thereby more accurately reflecting the true state of the environment when subsequently weighting and fusing the calibration sensor data to obtain comprehensive environmental indicators. This solution, combined with the multimodal data fusion mechanism in the aforementioned intelligent inspection method for explosion-proof robots, makes the generated pre-calibrated comprehensive risk value and real-time comprehensive risk value more accurate and reliable by performing a more refined information content assessment of sensor data. This improvement ensures that the environmental risks of each inspection node can be considered more effectively when generating the first and second inspection paths, thereby enhancing the intelligence level and the accuracy of risk assessment of the entire inspection method and helping explosion-proof robots to identify and respond to potential risks more timely and accurately.

[0103] In some preferred embodiments, the formula for calculating the percentage of standard deviation is as follows:

[0104] ;

[0105] Where, p i σ represents the proportion of the standard deviation corresponding to the i-th mode. i σ represents the standard deviation of the calibration sensor data corresponding to the i-th mode within a preset time window.j The standard deviation of the calibration sensor data corresponding to the j-th mode is within a preset time window, and N represents the number of modes contained in the calibration multimodal data corresponding to the inspection node.

[0106] The formula for calculating entropy is shown below:

[0107] ;

[0108] Among them, e i This represents the entropy value corresponding to the i-th mode, where k is a constant;

[0109] The formula for calculating the fusion weight is as follows:

[0110] ;

[0111] Where, w i e represents the fusion weight corresponding to the i-th mode. j This represents the entropy value corresponding to the j-th mode;

[0112] The formula for calculating the comprehensive environmental indicators is as follows:

[0113] ;

[0114] Where E represents the comprehensive environmental index, f i This represents the calibration sensor data for the i-th mode in the calibration multimodal data corresponding to the inspection node.

[0115] In some preferred embodiments, the formula for calculating the dynamic risk value is as follows:

[0116] ;

[0117] Among them, D j T represents the dynamic risk value corresponding to the j-th inspection node, α, β, and γ are the first preset coefficient, the second preset coefficient, and the third preset coefficient, respectively. j E represents the cumulative running time corresponding to the j-th inspection node. j A represents the dynamic risk value corresponding to the j-th inspection node. j This represents the cumulative number of abnormal times corresponding to the j-th inspection node;

[0118] The formula for calculating the pre-calibrated comprehensive risk value is shown below:

[0119] ;

[0120] Among them, R j S represents the pre-calibrated comprehensive risk value corresponding to the j-th inspection node. jλ represents the preset importance weight corresponding to the j-th inspection node. j M represents the preset dynamic risk weight corresponding to the j-th inspection node. j This represents the preset basic risk coefficient corresponding to the j-th inspection node.

[0121] The sum of the first, second, and third preset coefficients in this embodiment is preferably 1 to eliminate the interference of absolute values ​​and to place all factors on the same relative scale for fair comparison and synthesis. The preset dynamic risk weights in this embodiment are preferably set by those skilled in the art based on the dynamic risk values ​​of each inspection node before the explosion-proof robot performs its inspection task. Specifically, the preset dynamic risk weight corresponding to an inspection node with a higher dynamic risk value is less than the preset dynamic risk weight corresponding to an inspection node with a lower dynamic risk value. The preset basic risk coefficients in this embodiment preferably reflect the inherent risk attributes of the inspection node that are independent of its real-time state (such as the hazard of the medium or design pressure). For example, for the three types of equipment—storage tanks, oil conveying arms, and fire protection systems—the preset basic risk coefficients are set to 0.6, 0.4, and 0.1, respectively, based on their inherent risk attributes.

[0122] In some preferred embodiments, the formula for calculating the first weighted movement cost is the same as the formula for calculating the second weighted movement cost, and the formula for calculating the second weighted movement cost is as follows:

[0123] ;

[0124] in, This represents the second weighted movement cost corresponding to the predefined path edge connecting the k-th inspection node and the l-th inspection node. R represents the predefined basic movement cost corresponding to the path edge connecting the k-th inspection node and the l-th inspection node. k R represents the pre-calibrated comprehensive risk value corresponding to the k-th inspection node. l R represents the pre-calibrated comprehensive risk value corresponding to the l-th inspection node. avg η represents the average of all pre-calibrated comprehensive risk values, and η represents the pre-defined risk weighting factor.

[0125] The preset basic movement cost in this embodiment can be the distance along the path or the estimated travel time. Setting the preset basic movement cost is prior art, and its working principle and workflow will not be discussed in detail here. The preset risk weight factor in this embodiment is used to control the degree of influence of risk factors on path cost. Specifically, the larger the preset risk weight factor, the more the path planning tends to prioritize high-risk inspection nodes; the smaller the preset risk weight factor, the more the path planning tends to prioritize the path with the lowest basic movement cost.

[0126] In some preferred embodiments, step S1 includes:

[0127] S11. When the explosion-proof robot moves along the pre-planned first inspection path, it uses a sensor group to acquire real-time multimodal data of the nodes to be inspected along the way.

[0128] S12. Preprocess the real-time multimodal data; preprocessing includes denoising, standardization, and image registration.

[0129] Preprocessing in this embodiment refers to a series of operations performed on the raw data before data analysis. Preprocessing aims to improve data quality, making it more suitable for subsequent analysis and processing. Its function is to eliminate noise, inconsistencies, and redundant information in the data, and to standardize data format and scale. Denoising in this embodiment aims to eliminate random errors, interference, or outliers in sensor data. Common denoising methods may include: mean filtering, median filtering, Gaussian filtering, and other filtering algorithms for image or time-series data; or signal processing techniques such as wavelet transform and empirical mode decomposition (EMD). Standardization in this embodiment aims to eliminate differences in dimensions and numerical ranges between different sensor data, making them comparable. Common standardization methods may include: Z-score standardization, which converts the data into a distribution with a mean of 0 and a standard deviation of 1; or Min-Max standardization, which scales the data to a specific range, such as [0, 1]. The image registration in this embodiment aims to align image data from different times, different viewpoints, or different modalities (such as visible light images and infrared images) to the same coordinate system. Common image registration methods may include: feature point-based registration, such as SIFT and SURF feature matching; or region-based registration, such as mutual information and normalized cross-correlation.

[0130] In this embodiment, after the explosion-proof robot moves along a pre-planned first inspection path and acquires real-time multimodal data of the nodes to be inspected along the way using its sensor array, the data is immediately preprocessed. Specifically, as the explosion-proof robot moves along the preset path, its onboard sensor array continuously collects environmental data. This raw data may contain noise, inconsistent dimensions, or misaligned images due to environmental interference, sensor characteristics, or acquisition conditions. To ensure the accuracy of subsequent risk assessment, this application introduces preprocessing steps such as denoising, standardization, and image registration. Denoising effectively filters out random interference in the data, ensuring its authenticity; standardization unifies the scale of different modal data, enabling effective fusion and comparison; image registration solves the spatial misalignment problem between multimodal image data, ensuring spatial consistency between visual and non-visual information. The preprocessed real-time multimodal data has higher quality and consistency, providing a reliable basis for generating the real-time comprehensive risk value in the subsequent step S2. This improvement in data quality makes the real-time comprehensive risk value generated based on real-time multimodal data more accurate. This allows for a more precise assessment of the difference between the current risk and the pre-defined risk in step S3, leading to the generation of a more reasonable and optimized second inspection path in step S4. Ultimately, the explosion-proof robot can move along the second inspection path that better reflects the actual risk situation, achieving dynamic adaptability and intelligence in the inspection strategy.

[0131] In some preferred embodiments, this embodiment preferably uses an improved A * The algorithm performs path planning, and the improved A * The preferred workflow of the algorithm is as follows: 1. Initialization: Set the starting point (e.g., a charging pile), the ending point (e.g., the starting point (charging pile) or a specified ending point), and a list of all nodes that need to be inspected; 2. Open and closed lists: Use an open list to store nodes to be evaluated and a closed list to store nodes that have been evaluated; 3. Construct the cost function: f(n) = g(n) + h(n); where f(n) represents the cost from the starting point to the nth inspected node, g(n) represents the actual total weighted cost from the starting point to the nth inspected node (the sum of the first or second weighted movement costs corresponding to all path edges from the starting point to the nth inspected node), and h(n) represents the estimated cost from the nth inspected node to the target point (preferably the ending point) (preferably Euclidean distance or Manhattan distance); 4. Node expansion: Select the inspected node with the smallest f(n) from the open list for expansion until the target point is added to the closed list or the open list is empty.

[0132] As can be seen from the above, the intelligent inspection method for explosion-proof robots provided in this application realizes intelligent risk response by dynamically generating inspection paths based on real-time comprehensive risk values. Since this application can perform dynamic risk assessment and path optimization based on real-time environmental data, it can effectively improve inspection efficiency and safety, thereby effectively solving the problems of insufficient multi-sensor information fusion, static risk assessment, and disconnect between path planning and equipment importance.

[0133] Please refer to Figure 2 , Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device includes a processor 101 and a memory 102. The processor 101 and the memory 102 are interconnected and communicate with each other via a communication bus 103 and / or other forms of connection mechanisms (not shown). The memory 102 stores computer-readable instructions executable by the processor 101. When the electronic device is running, the processor 101 executes these computer-readable instructions to perform the method in any optional implementation of the above embodiments, thereby achieving the following function: Step S1: When the explosion-proof robot moves along a pre-planned first inspection path, the sensor group acquires the areas to be inspected along the path. Step S2: Generate a real-time comprehensive risk value based on the preset importance weights of the nodes to be inspected along the route and the real-time multimodal data; Step S3: When the real-time comprehensive risk value is greater than the preset calibrated comprehensive risk value of the next node to be inspected in the first inspection path, generate a predefined first weighted movement cost corresponding to each path edge based on the real-time comprehensive risk value or the preset calibrated comprehensive risk value of all nodes to be inspected; Step S4: Generate a second inspection path that passes through all nodes to be inspected and has the smallest sum of the first weighted movement costs based on all the first weighted movement costs; Step S5: Control the explosion-proof robot to move according to the second inspection path and return to step S2.

[0134] This application embodiment also provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it executes the method in any optional implementation of the above embodiments to achieve the following functions: Step S1: When the explosion-proof robot moves along a pre-planned first inspection path, it uses a sensor group to acquire real-time multimodal data of the nodes to be inspected along the way; Step S2: It generates a real-time comprehensive risk value based on the preset importance weights corresponding to the nodes to be inspected along the way and the real-time multimodal data; Step S3: When the real-time comprehensive risk value is greater than the pre-calibrated comprehensive risk value corresponding to the next node to be inspected in the first inspection path, it generates a predefined first weighted movement cost corresponding to each path edge based on the real-time comprehensive risk value or the pre-calibrated comprehensive risk value corresponding to all nodes to be inspected; Step S4: It generates a second inspection path that passes through all nodes to be inspected and has the smallest sum of the first weighted movement costs based on all the first weighted movement costs; Step S5: It controls the explosion-proof robot to move along the second inspection path and returns to step S2. The computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0135] As can be seen from the above, the intelligent inspection method, equipment and medium for explosion-proof robots provided by this application realizes intelligent risk response by dynamically generating inspection paths based on real-time comprehensive risk values. Since this application can perform dynamic risk assessment and path optimization based on real-time environmental data, it can effectively improve inspection efficiency and safety, thereby effectively solving the problems of insufficient multi-sensor information fusion, static risk assessment and disconnect between path planning and equipment importance.

[0136] In the embodiments provided in this application, it should be understood that relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0137] The above are merely embodiments of this application and are not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. An intelligent inspection method for an explosion-proof robot, applied to an explosion-proof robot, wherein the explosion-proof robot is equipped with a sensor array, characterized in that, The intelligent inspection method for explosion-proof robots includes the following steps: S1. When the explosion-proof robot moves along the pre-planned first inspection path, the sensor group is used to acquire real-time multimodal data of the nodes to be inspected along the way. S2. Generate a real-time comprehensive risk value based on the preset importance weights of the nodes to be inspected along the route and real-time multimodal data; S3. When the real-time comprehensive risk value is greater than the pre-calibrated comprehensive risk value corresponding to the next node to be inspected in the first inspection path, generate the first weighted movement cost corresponding to each path edge according to the real-time comprehensive risk value or pre-calibrated comprehensive risk value corresponding to all the nodes to be inspected. S4. Generate a second inspection path that passes through all the nodes to be inspected and has the smallest sum of the first weighted movement costs based on all the first weighted movement costs. S5. Control the explosion-proof robot to move along the second inspection path and return to step S2; The pre-planning process for the first inspection path includes: A1. Control the explosion-proof robot to move along a preset path that passes through all inspection nodes, and use the sensor group to acquire the calibration multimodal data corresponding to each inspection node. A2. For each of the inspection nodes, a pre-calibrated comprehensive risk value is generated based on the preset importance weight and calibrated multimodal data corresponding to the node; A3. Generate the predefined second weighted movement cost for each path edge based on the pre-calibrated comprehensive risk value corresponding to all the nodes to be inspected. A4. Generate a first inspection path that passes through all the inspection nodes and has the smallest sum of the second weighted movement costs based on all the second weighted movement costs.

2. The intelligent inspection method for explosion-proof robots according to claim 1, characterized in that, The process for generating the pre-calibrated comprehensive risk value is the same as the process for generating the real-time comprehensive risk value. The calibration multimodal data includes calibration sensor data corresponding to multiple modes. The process for generating the pre-calibrated comprehensive risk value includes: B1. For each of the modes, obtain the standard deviation of the calibration sensor data corresponding to that mode within a preset time window; B2. Calculate the percentage of the standard deviation corresponding to each mode based on all the standard deviations mentioned; B3. For each mode, calculate the fusion weight corresponding to that mode based on the proportion of the standard deviation corresponding to that mode; B4. For each inspection node, the data of all calibrated sensors corresponding to the node are weighted and fused according to all the fusion weights corresponding to the node to obtain the comprehensive environmental index. B5. Obtain the cumulative running time of each inspection node and the cumulative number of abnormal events in its area; B6. For each inspection node, calculate the dynamic risk value based on the comprehensive environmental indicators, cumulative operating time, and cumulative number of abnormal events corresponding to the node. Then, calculate the pre-calibrated comprehensive risk value based on the preset importance weight, preset dynamic risk weight, preset basic risk coefficient, and the dynamic risk value corresponding to the node.

3. The intelligent inspection method for explosion-proof robots according to claim 2, characterized in that, Step B3 includes: B31. For each mode, calculate the entropy value corresponding to the mode based on the standard deviation ratio of the mode, and then calculate the fusion weight corresponding to the mode based on the entropy values ​​of all modes.

4. The intelligent inspection method for explosion-proof robots according to claim 3, characterized in that, The formula for calculating the percentage of standard deviation is as follows: ； Where, p i σ represents the proportion of the standard deviation corresponding to the i-th mode. i σ represents the standard deviation of the calibration sensor data corresponding to the i-th mode within a preset time window. j The standard deviation of the calibration sensor data corresponding to the j-th mode is within a preset time window, and N represents the number of modes contained in the calibration multimodal data corresponding to the inspection node. The formula for calculating the entropy value is as follows: ； Among them, e i This represents the entropy value corresponding to the i-th mode, where k is a constant; The formula for calculating the fusion weight is as follows: ； Where, w i e represents the fusion weight corresponding to the i-th mode. j This represents the entropy value corresponding to the j-th mode; The formula for calculating the comprehensive environmental index is as follows: ； Where E represents the comprehensive environmental index, f i This represents the calibration sensor data for the i-th mode in the calibration multimodal data corresponding to the inspection node.

5. The intelligent inspection method for explosion-proof robots according to claim 2, characterized in that, The formula for calculating the dynamic risk value is shown below: ； Among them, D j T represents the dynamic risk value corresponding to the j-th inspection node, α, β, and γ are the first preset coefficient, the second preset coefficient, and the third preset coefficient, respectively. j E represents the cumulative running time corresponding to the j-th inspection node. j A represents the dynamic risk value corresponding to the j-th inspection node. j This represents the cumulative number of abnormal times corresponding to the j-th inspection node; The formula for calculating the pre-calibrated comprehensive risk value is as follows: ； Among them, R j S represents the pre-calibrated comprehensive risk value corresponding to the j-th inspection node. j λ represents the preset importance weight corresponding to the j-th inspection node. j M represents the preset dynamic risk weight corresponding to the j-th inspection node. j This represents the preset basic risk coefficient corresponding to the j-th inspection node.

6. The intelligent inspection method for explosion-proof robots according to claim 1, characterized in that, The formula for calculating the first weighted movement cost is the same as the formula for calculating the second weighted movement cost, which is shown in the following formula: ； in, This represents the second weighted movement cost corresponding to the predefined path edge connecting the k-th inspection node and the l-th inspection node. R represents the predefined basic movement cost corresponding to the path edge connecting the k-th inspection node and the l-th inspection node. k R represents the pre-calibrated comprehensive risk value corresponding to the k-th inspection node. l R represents the pre-calibrated comprehensive risk value corresponding to the l-th inspection node. avg η represents the average of all pre-calibrated comprehensive risk values, and η represents the pre-defined risk weighting factor.

7. The intelligent inspection method for explosion-proof robots according to claim 1, characterized in that, Step S1 includes: S11. When the explosion-proof robot moves along the pre-planned first inspection path, the sensor group is used to acquire real-time multimodal data of the nodes to be inspected along the way. S12. Preprocess the real-time multimodal data; the preprocessing includes denoising, standardization and image registration.

8. An electronic device, characterized in that, It includes a processor and a memory, the memory storing computer-readable instructions that, when executed by the processor, perform the steps of the method as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it performs the steps of the method as described in any one of claims 1-7.

Images