Park operation and maintenance robot inspection path planning system based on human-computer cooperation

Abstract

The application provides a park operation robot inspection path planning system based on human-computer cooperation, relates to the technical field of robot path planning, and comprises the following steps: constructing a park operation scene model; collecting robot state data; converting operation intention into path planning constraints; constructing an equipment abnormality causal diagnosis graph; calculating the uncertainty reduction of each candidate inspection point on abnormality root cause judgment to obtain the diagnosis value of each candidate inspection point; generating an inspection robot inspection path; updating the abnormality root cause probability, and adjusting the next inspection path according to the update result. Through the evidence value evaluation mechanism based on information entropy, the contribution degree of each candidate inspection point to the reduction of root cause uncertainty is quantitatively calculated, and the diagnosis value is corrected by comprehensively considering the multi-dimensional costs such as driving distance, time, energy consumption and safety risk, so that the problems of strong blindness and excessive invalid inspection of traditional inspection are solved, and the core target of "obtaining maximum root cause certainty with minimum inspection cost" is realized.

Description

Technical Field

[0001] This invention relates to the field of robot path planning technology, and in particular to a park operation and maintenance robot inspection path planning system based on human-machine collaboration. Background Technology

[0002] With the rapid development of industrial intelligence, power digitalization, and large-scale data centers, the operation and maintenance scenarios of various sites, computer rooms, and industrial equipment are becoming increasingly complex. Equipment deployment density continues to increase, and the scope of operation and maintenance coverage is constantly expanding. The demand for refined and routine operation and maintenance tasks, such as equipment operation status monitoring, environmental inspection, fault diagnosis, and daily monitoring, is continuously growing. Current operation and maintenance robots integrate multi-dimensional sensing devices, enabling comprehensive collection of operation and maintenance environment data, equipment operating parameters, and abnormal status information. Simultaneously, combined with edge data processing, edge computing, and IoT transmission technologies, operation and maintenance robots can complete standardized operation and maintenance tasks.

[0003] Typical park maintenance robot inspection path planning methods mainly focus on optimization goals such as inspection point coverage, path distance, energy consumption, and obstacle avoidance safety, which are difficult to support the problem of rapid root cause localization of abnormal events. Summary of the Invention

[0004] This invention provides a human-machine collaborative inspection path planning system for park operation and maintenance robots, which solves the problem that existing park operation and maintenance robot inspection path planning methods are unable to support the rapid root cause localization of abnormal events.

[0005] On one hand, this invention provides a human-machine collaborative park maintenance robot inspection path planning system, including: a park information modeling module, a robot state perception module, a human-machine collaborative interaction module, a cause-effect diagnostic graph construction module, an evidence value assessment module, a path planning module, and a dynamic update module; the park information modeling module is used to collect park spatial maps, road topology, equipment distribution, inspection points, restricted areas, charging locations, and historical maintenance data to construct a park maintenance scenario model; the robot state perception module is used to collect robot state data including the inspection robot's current location, battery status, motion status, sensor status, and environmental perception data; the human-machine collaborative interaction module is used to receive input from maintenance personnel. The system is designed to: 1) define the inspection tasks and maintenance intentions, and transform these intentions into path planning constraints; 2) construct a causal diagnostic graph for equipment anomalies based on the relationships between park equipment, abnormal phenomena, candidate fault causes, and observable evidence points; 3) evaluate the evidence value based on the causal diagnostic graph for equipment anomalies, calculating the reduction in uncertainty of each candidate inspection point in determining the root cause of the anomaly, thus obtaining the diagnostic value of each candidate inspection point; 4) generate the inspection path for the inspection robot based on robot state data, diagnostic value, and path planning constraints; and 5) update the probability of the root cause of the anomaly based on real-time evidence data collected during the robot's inspection process, and adjust the next inspection path based on the update results.

[0006] According to the human-machine collaboration-based park operation and maintenance robot inspection path planning system provided by the present invention, the step of the human-machine collaboration interaction module converting operation and maintenance intentions into path planning constraints includes: The system receives maintenance intentions input by maintenance personnel. These intentions include key inspection areas, key inspection equipment, types of anomalies to be verified, inspection time requirements, avoidance areas, inspection priorities, and security levels. Semantic parsing is performed on the operation and maintenance intent to extract intent keywords, target objects, spatial scope, time scope, and priority information; Match the intent keywords with the equipment type, inspection point, regional attributes and historical operation and maintenance data in the park information modeling module to determine the set of target inspection points corresponding to the operation and maintenance intent; Generate path planning constraints based on the target inspection point set; When there are conflicts among multiple path planning constraints, the conflicts are resolved based on the priority of operation and maintenance intentions, the level of abnormal risk, and robot state data to obtain the final path planning constraints.

[0007] According to the human-machine collaborative park operation and maintenance robot inspection path planning system provided by the present invention, the causal diagnosis graph construction module includes an abnormal node construction unit, a fault cause node construction unit, an evidence point node construction unit, a correlation relationship construction unit, and a spatial mapping unit. The abnormal node construction unit is used to construct abnormal phenomenon nodes based on the operating status of park equipment and historical abnormal records. The fault cause node construction unit is used to construct candidate fault cause nodes based on equipment structure, operation and maintenance rules, and historical fault records. The evidence point node construction unit is used to construct observable evidence point nodes based on the data types that the robot can collect and the park inspection points. The correlation relationship construction unit is used to establish causal correlation relationships between abnormal phenomenon nodes, candidate fault cause nodes, and observable evidence point nodes, and assign causal correlation weights to the causal correlation relationships. The spatial mapping unit is used to map the observable evidence point nodes to their actual locations on the park spatial map, so that the equipment abnormality causal diagnosis graph is associated with the park spatial map.

[0008] The human-machine collaborative park operation and maintenance robot inspection path planning system provided by the present invention includes an evidence value assessment module comprising a root cause probability initialization unit, an uncertainty calculation unit, an evidence impact assessment unit, a diagnostic value calculation unit, and a value correction unit. The root cause probability initialization unit determines the initial probability of candidate fault causes based on historical operation and maintenance data, current anomalies, equipment importance, and operation and maintenance intentions. The uncertainty calculation unit calculates the uncertainty of candidate fault causes under the current known evidence conditions. The evidence impact assessment unit assesses the influence of candidate inspection points on the probability distribution of candidate fault causes based on the type of evidence collected at the candidate inspection points, evidence reliability, causal association weight, and historical validity. The diagnostic value calculation unit calculates the diagnostic value of candidate inspection points based on the change in uncertainty of candidate fault causes before and after the evidence at the candidate inspection points. The value correction unit corrects the diagnostic value based on the distance, time, energy consumption, safety risks, and communication status of the robot reaching the candidate inspection point, obtaining the corrected diagnostic value of the candidate inspection point.

[0009] According to the human-machine collaborative park operation and maintenance robot inspection path planning system provided by the present invention, the step of the evidence impact assessment unit evaluating the degree of influence of candidate inspection points on the probability distribution of candidate fault causes includes: Obtain the types of collectable evidence corresponding to the candidate inspection points, and determine the causal association weights between the types of collectable evidence and the candidate causes of failure. The reliability of the evidence data corresponding to the candidate inspection points is determined based on the robot's sensor status, acquisition accuracy, historical acquisition success rate, and degree of environmental interference. The historical validity of candidate inspection points is determined based on the evidence hit rate of candidate inspection points under similar abnormal scenarios in historical operation and maintenance data. Based on causal association weights, evidence reliability, and historical validity, the probability correction coefficients for each candidate fault cause of the candidate inspection point are calculated. Based on the probability correction coefficient, the degree of change in the probability distribution of candidate fault causes after collecting evidence from candidate inspection points is predicted, and this change is used as the degree of influence of candidate inspection points on the probability distribution of candidate fault causes.

[0010] According to the human-machine collaboration-based park maintenance robot inspection path planning system provided by the present invention, the path planning module includes a diagnostic layer path planning unit, a motion layer path planning unit, a cost calculation unit, a comprehensive optimization unit, a path verification unit, and a replanning unit. The diagnostic layer path planning unit is used to determine the set of target inspection points for verifying the root causes of anomalies from candidate inspection points based on diagnostic value, probability of anomalies, and path planning constraints. The motion layer path planning unit uses the set of target inspection points and robot state data to generate a feasible motion path for the robot to reach the set of target inspection points. The cost calculation unit is used to calculate the risk index of the feasible motion path. The comprehensive optimization unit is used to select the target inspection path from multiple feasible motion paths based on the risk index and with the optimization objective of maximizing the diagnostic value obtained per unit travel cost. The path verification unit is used to verify whether the target inspection path meets the constraints of power, time, safety, restricted area, and human-machine collaboration. The replanning unit is used to redetermine the set of target inspection points and generate the next inspection path when an abnormal event occurs during the robot inspection process.

[0011] According to the human-machine collaborative park operation and maintenance robot inspection path planning system provided by the present invention, the step of the diagnostic layer path planning unit determining the target inspection point set for verifying the root cause of anomalies includes: Obtain the set of candidate fault causes corresponding to the current abnormal phenomenon and the current probability of each candidate fault cause; Based on the equipment anomaly cause-and-effect diagnosis diagram, determine the set of candidate inspection points associated with each candidate fault cause; Obtain the diagnostic value of each candidate inspection point in the candidate inspection point set, and filter the candidate inspection point set in combination with path planning constraints; remove candidate inspection points that do not meet safety constraints, restricted area constraints, power constraints, time constraints, or human-machine collaboration constraints; Calculate the comprehensive verification priority of candidate inspection points based on their diagnostic value, probability of candidate fault causes, and priority of maintenance intentions. Candidate inspection points are selected from high to low according to the comprehensive verification priority to form a target inspection point set for verifying the root cause of the anomaly; when the probability of any candidate fault cause reaches the preset root cause confirmation threshold, the selection of candidate inspection points unrelated to that candidate fault cause is stopped.

[0012] According to the human-machine collaborative park operation and maintenance robot inspection path planning system provided by the present invention, the step of the cost calculation unit calculating the risk index of feasible movement paths includes: Based on the park's spatial map and road topology, calculate the path distance of feasible movement paths; The travel time of feasible movement paths is calculated based on the robot's motion state, road conditions, and dwell time at candidate inspection points. Calculate the energy consumption of feasible motion paths based on path distance, robot load, battery status, movement speed, and ground slope. Safety risks are calculated based on historical anomaly records, population density, vehicle flow, distribution of hazardous areas, and environmental perception data of the areas through which feasible movement routes pass. Calculate communication risks based on the communication coverage strength of the areas traversed by the feasible movement path, historical communication interruption records, and remote control requirements. The robot calculates the obstacle passage risk based on the number of dynamic obstacles, the direction of obstacle movement, the obstacle density, and the road width as perceived in real time. The path distance, travel time, energy consumption, safety risk, communication risk, and obstacle passage risk are normalized and then weighted and fused according to preset weights to obtain the risk index of feasible movement paths.

[0013] The human-machine collaborative park maintenance robot inspection path planning system provided by the present invention includes a dynamic update module comprising a real-time evidence receiving unit, an evidence validity judgment unit, a root cause probability update unit, a diagnostic map weight update unit, and a next path adjustment unit. The real-time evidence receiving unit receives real-time evidence data collected by the robot during inspection. The evidence validity judgment unit determines whether the real-time evidence data meets preset validity conditions based on sensor status, acquisition distance, acquisition angle, environmental interference level, and data integrity. The root cause probability update unit updates the probability of candidate fault causes based on the real-time evidence data when it meets the preset validity conditions. The diagnostic map weight update unit updates the association weights between abnormal phenomenon nodes, candidate fault cause nodes, and observable evidence point nodes in the equipment abnormality causal diagnostic map based on the changes in candidate fault cause probabilities. The next path adjustment unit recalculates the diagnostic value of the remaining candidate inspection points based on the updated candidate fault cause probabilities and updated association weights, and adjusts the next inspection path.

[0014] According to the human-machine collaborative park operation and maintenance robot inspection path planning system provided by the present invention, the step of updating the associated weights in the diagnostic graph weight update unit includes: Acquire real-time evidence data collected by the robot, results of evidence validity assessment, and root cause probability update results; Identify observable evidence point nodes corresponding to real-time evidence data, and identify abnormal phenomenon nodes and candidate fault cause nodes associated with the observable evidence point nodes. When real-time evidence data increases the probability of candidate fault causes, the association weight between the corresponding observable evidence point node and the candidate fault cause node is increased. When real-time evidence data reduces the probability of candidate fault causes, the association weight between the corresponding observable evidence point node and the candidate fault cause node is reduced. When real-time evidence data is determined to be invalid or low-reliability evidence, the evidence reliability weight of the corresponding observable evidence point node is reduced. After the operation and maintenance personnel manually confirm the root cause of the anomaly, they adjust the correlation weight between the anomaly node, the candidate fault cause node, and the observable evidence point node based on the manual confirmation result. The corrected correlation weights are written back to the equipment anomaly cause-effect diagnostic graph for subsequent candidate inspection point diagnostic value calculation and next inspection path adjustment.

[0015] This invention provides a human-machine collaborative park maintenance robot inspection path planning system. By constructing a device anomaly causal diagnosis graph mapped to the physical space, it establishes a quantitative causal relationship between abnormal phenomena, candidate fault causes, and observable evidence points, achieving unification of fault diagnosis logic and park spatial location. Through an evidence value assessment mechanism based on information entropy, it quantitatively calculates the contribution of each candidate inspection point to reducing root cause uncertainty, and comprehensively considers multiple dimensions such as travel distance, time, energy consumption, and safety risks to correct the diagnostic value. This solves the problems of strong blindness and many ineffective inspections in traditional inspections, achieving the core objective of "obtaining the greatest root cause certainty with the least inspection cost." It effectively solves the pain points of traditional park maintenance robot inspections, such as fixed task modes, delayed anomaly response, low fault location efficiency, and inability to adapt to dynamic environmental changes. It significantly improves the speed of detecting park anomalies and the accuracy of root cause location, reduces the intensity and cost of manual maintenance, and enhances the intelligence, automation, and standardization of park maintenance. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the structure of the park operation and maintenance robot inspection path planning system based on human-machine collaboration provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating how the human-computer collaborative interaction module, as provided in this embodiment of the invention, transforms operational intentions into path planning constraints. Figure 3 This is a flowchart of the diagnostic layer path planning unit determining the set of target inspection points for verifying the root causes of anomalies, provided in an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0019] The following is combined with Figures 1-3 This invention describes a human-machine collaborative campus maintenance robot inspection path planning system.

[0020] like Figures 1-3 As shown in the figure, the park operation and maintenance robot inspection path planning system based on human-machine collaboration provided in this embodiment of the invention includes: a park information modeling module, a robot state perception module, a human-machine collaborative interaction module, a cause-effect diagnostic graph construction module, an evidence value assessment module, a path planning module, and a dynamic update module.

[0021] The park information modeling module is used to collect park spatial maps, road topology, equipment distribution, inspection points, restricted areas, charging locations, and historical operation and maintenance data to construct a park operation and maintenance scenario model. The park spatial map adopts a layered modeling approach, including a basic geographic layer, a road infrastructure layer, an equipment asset layer, and an environmental perception layer. The basic geographic layer stores static geographic information such as park topography, elevation, and building outlines. The road infrastructure layer stores road attribute information such as road width, turning radius, slope, pavement material, traffic signs, and speed bump locations. The equipment asset layer stores equipment lifecycle information such as unique equipment identifiers, equipment models, installation locations, installation heights, manufacturers, commissioning times, maintenance cycles, and associated inspection points. Historical operation and maintenance data includes historical anomaly records, root cause records, inspection completion rates, equipment mean time between failures (MTBF), repair response times, and historical robot inspection trajectory data. The park operation and maintenance scenario model is built using digital twin technology, supporting real-time data-driven dynamic updates. It can achieve data interoperability with the park's existing asset management system, video surveillance system, access control system, and fire protection system through standard interfaces.

[0022] The robot state perception module is used to collect robot state data, including the current location, battery status, motion status, sensor status, and environmental perception data of the inspection robot. Current location is achieved through multi-source fusion positioning, integrating GPS / BeiDou satellite positioning data, laser SLAM positioning data, visual odometry data, and UWB indoor positioning data, with a positioning accuracy better than ±5cm. Battery status is collected in real-time by the battery management system (BMS), including battery voltage, current, remaining capacity (SOC), state of health (SOH), and charge / discharge cycle count. Motion status includes the robot's current speed, acceleration, steering angle, mileage, and drive motor operating current. Sensor status includes the operating voltage, operating temperature, acquisition frame rate, data transmission rate, and fault alarm status of each sensor. Sensors include at least a LiDAR, a high-definition visible light camera, an infrared thermal imager, a sound acquisition microphone, a gas sensor, a temperature and humidity sensor, and a vibration sensor. Environmental perception data includes obstacle point cloud data within a 360° range around the robot, dynamic obstacle trajectory prediction data, and ambient light intensity data.

[0023] The human-machine collaborative interaction module is used to receive inspection tasks and maintenance intentions input by maintenance personnel, and to transform the maintenance intentions into path planning constraints.

[0024] The steps by which the human-machine collaborative interaction module transforms operational intentions into path planning constraints include: It receives maintenance intentions input by maintenance personnel, which include key inspection areas, key inspection equipment, types of anomalies to be verified, inspection time requirements, avoidance areas, inspection priorities, and security levels.

[0025] Semantic parsing is performed on the operation and maintenance intentions to extract intent keywords, target objects, spatial scope, temporal scope, and priority information. Semantic parsing uses a pre-trained language model based on the Transformer architecture. First, the input text is segmented, part-of-speech tagging is performed, and named entity recognition is conducted to identify device entities, region entities, time entities, and anomaly type entities. Then, a syntactic dependency tree is constructed through dependency parsing to extract semantic relationships between entities. Finally, an intent classifier determines the core types of operation and maintenance intentions, including routine inspections, anomaly troubleshooting, fault confirmation, and equipment verification.

[0026] The intent keywords are matched with equipment types, inspection points, regional attributes, and historical maintenance data in the park information modeling module to determine the target inspection point set corresponding to the maintenance intent. The matching process combines fuzzy matching and exact matching. Exact matching is used for precise entities such as equipment names and regional names, while fuzzy matching based on semantic similarity is used for fuzzy descriptions such as equipment types and anomaly types. When the maintenance intent only specifies the equipment type, all equipment of that type and its corresponding inspection points in the park are automatically retrieved. When the maintenance intent specifies the anomaly type to be verified, all equipment and associated inspection points that have historically experienced that type of anomaly are automatically retrieved.

[0027] Path planning constraints are generated based on the set of target inspection points. These constraints include spatial, temporal, resource, and safety constraints. Spatial constraints include mandatory inspection points, prohibited areas, and priority routes. Temporal constraints include the start and end times of the inspection task and the maximum dwell time at a single inspection point. Resource constraints include the robot's remaining battery power and charging location. Safety constraints include the robot's maximum speed limit, access rights to hazardous areas, and pedestrian avoidance priority.

[0028] When conflicts exist among multiple path planning constraints, the conflicts are resolved based on the priority of operational intentions, the level of anomaly risk, and robot status data to obtain the final path planning constraints. Conflict resolution employs the Analytic Hierarchy Process (AHP) to construct a constraint priority matrix. First, the priority of operational intentions is divided into four levels: urgent, high, medium, and low; the level of anomaly risk is divided into four levels: extremely high, high, medium, and low; and the robot status data is divided into three levels: normal, warning, and abnormal. Then, the weight coefficients of each constraint factor are determined through pairwise comparisons. Finally, the conflicting constraints are sorted according to the weight coefficients, retaining constraints with higher priority and appropriately relaxing or adjusting constraints with lower priority. When there are unresolvable conflicts, a conflict prompt is sent to the operational personnel through the human-machine collaborative interaction module, providing multiple feasible constraint adjustment schemes for the operational personnel to choose and confirm.

[0029] The causal diagnosis graph construction module is used to construct a causal diagnosis graph of equipment anomalies based on the correlation between park equipment, abnormal phenomena, candidate fault causes, and observable evidence points.

[0030] The cause-effect diagnostic graph construction module includes an anomaly node construction unit, a fault cause node construction unit, an evidence point node construction unit, a correlation construction unit, and a spatial mapping unit. The anomaly node construction unit is used to construct anomaly phenomenon nodes based on the operating status of equipment in the park and historical anomaly records. Anomaly phenomenon nodes include at least one of the following: temperature anomaly, smoke anomaly, vibration anomaly, noise anomaly, instrument reading anomaly, equipment offline anomaly, and environmental parameter anomaly.

[0031] The fault cause node construction unit is used to construct candidate fault cause nodes based on equipment structure, operation and maintenance rules, and historical fault records. These nodes include fault description, fault occurrence probability, fault impact range, fault handling method, and associated abnormal phenomenon attributes. Candidate fault causes are classified according to the location of the fault, such as electrical faults, mechanical faults, software faults, and environmental faults. For each type of fault, it is broken down into more granular sub-fault causes based on the equipment structure. For example, electrical faults can be broken down into short circuits, open circuits, poor contact, component damage, and power supply failures. The fault occurrence probability is statistically calculated based on historical fault records and is dynamically updated according to the equipment's operating years and maintenance status.

[0032] The evidence point node construction unit is used to construct observable evidence point nodes based on the types of data that the robot can collect and the inspection points in the park. The observable evidence point nodes include at least one of the following: image acquisition point, infrared temperature measurement point, instrument recognition point, sound acquisition point, gas detection point, access control status detection point, and environmental parameter detection point.

[0033] The correlation construction unit is used to establish causal relationships between abnormal phenomenon nodes, candidate fault cause nodes, and observable evidence point nodes, and to assign causal relationship weights to these relationships. Causal relationships include a positive causal relationship of "fault cause → abnormal phenomenon" and a negative observation relationship of "abnormal phenomenon → observable evidence point." Causal relationship weights are determined using a combination of expert knowledge and data-driven methods. First, domain experts provide initial weights based on equipment principles and operational experience. Then, the initial weights are iteratively optimized using Bayesian network parameter learning methods based on historical fault case data. The range of causal relationship weights is [0,1], with larger weight values ​​indicating a stronger causal relationship between the two.

[0034] The spatial mapping unit is used to map observable evidence point nodes to their actual locations on the park's spatial map, thus linking the equipment anomaly cause-and-effect diagnostic diagram with the park's spatial map. The spatial mapping process first establishes a one-to-one correspondence between observable evidence point nodes and inspection points in the park's operation and maintenance scenario model, recording the 3D spatial coordinates, orientation angle, and altitude of each observable evidence point. Then, the location and collection range of the observable evidence points are visually marked on the park's digital twin map. When the robot moves to a specific inspection point, the system automatically loads all observable evidence point nodes corresponding to that point, along with associated anomaly nodes and candidate fault cause nodes.

[0035] The evidence value assessment module is used to calculate the reduction in uncertainty of each candidate inspection point in the judgment of the root cause of the anomaly based on the equipment anomaly cause-effect diagnostic diagram, and to obtain the diagnostic value of each candidate inspection point.

[0036] The evidence value assessment module includes a root cause probability initialization unit, an uncertainty calculation unit, an evidence impact assessment unit, a diagnostic value calculation unit, and a value correction unit. The root cause probability initialization unit determines the initial probability of candidate fault causes based on historical maintenance data, current anomalies, equipment importance, and maintenance intentions. The initialization process first retrieves the equipment anomaly causal diagnostic graph based on the current anomaly, obtaining a set of all possible candidate fault causes. Then, it calculates the prior probability based on the frequency of occurrence of each candidate fault cause leading to the anomaly in historical maintenance data. Next, it weights and adjusts the prior probabilities according to equipment importance, appropriately increasing the probability of fault causes corresponding to important equipment. Finally, it corrects the probabilities based on the anomaly type specified in the maintenance intention, appropriately increasing the probability of fault causes that maintenance personnel are focusing on.

[0037] The uncertainty calculation unit is used to calculate the uncertainty of candidate fault causes under the current known evidence conditions. It uses information entropy for quantification, and the formula is expressed as: in, Let be the information entropy value, representing the uncertainty of the candidate fault cause set X given the known evidence set E; X is the candidate fault cause set, and E is the currently known evidence set. Candidate fault causes, Candidate fault causes under known evidence E conditions The posterior probability; the larger the information entropy value, the higher the uncertainty of the candidate fault cause, and the more evidence is needed to reduce the uncertainty.

[0038] The evidence impact assessment unit is used to evaluate the degree of influence of candidate inspection points on the probability distribution of candidate failure causes based on the type of evidence collected at the candidate inspection points, the reliability of the evidence, the causal association weight, and the historical validity. The steps include: Obtain the collectable evidence types corresponding to candidate inspection points and determine the causal association weights between collectable evidence types and candidate fault causes. The causal association weights between collectable evidence types and candidate fault causes are obtained from the equipment anomaly causal diagnosis diagram. For a candidate inspection point corresponding to multiple collectable evidence types, calculate the causal association weight between each evidence type and each candidate fault cause, and then take the maximum value as the causal association weight between the candidate inspection point and the corresponding candidate fault cause.

[0039] Based on the robot sensor status, acquisition accuracy, historical acquisition success rate, and degree of environmental interference, the evidentiary reliability R of the evidence data corresponding to the candidate inspection points is determined, and the calculation formula is expressed as follows: in, This is the sensor state coefficient, with a value range of [0,1]. It is 1 when the sensor is working normally, 0.5 when the sensor issues a warning, and 0 when the sensor malfunctions. The accuracy coefficient is determined based on the ratio of the sensor's nominal accuracy to the required accuracy for inspection. The historical data collection success rate coefficient is determined based on the proportion of times valid evidence was successfully collected in the past 100 inspections of this inspection point. The environmental interference coefficient is determined based on the light intensity, electromagnetic interference intensity, background noise intensity, and obstruction at the inspection point.

[0040] The historical validity of candidate inspection points is determined based on the evidence hit rate of candidate inspection points under similar anomaly scenarios in historical operation and maintenance data. Similar anomaly scenarios are matched by calculating the cosine similarity of the anomaly phenomenon vector, equipment type vector, and operating environment vector, with a similarity threshold set to 0.8. The evidence hit rate is the proportion of times the true root cause is successfully located after collecting evidence for the inspection point under similar anomaly scenarios out of the total number of times. For newly established inspection points without historical data, the historical validity coefficient is initially set to 0.5 and is gradually updated as inspection data accumulates.

[0041] Based on causal association weights, evidence reliability, and historical validity, the probability correction coefficients for each candidate fault cause by the candidate inspection points are calculated. The formula is expressed as: in, The weights of the causal relationships between candidate inspection point i and candidate fault cause j are given. To assess the reliability of evidence for candidate inspection point i, The historical validity of candidate inspection point i; The value range is [0,1]. The larger the value, the greater the impact of the evidence collected from candidate inspection point i on the probability correction of candidate fault cause j.

[0042] Based on the probability correction coefficient, the degree of change in the probability distribution of candidate fault causes after collecting evidence from candidate inspection points is predicted, and this change is taken as the degree of influence of the candidate inspection point on the probability distribution of candidate fault causes. The degree of change in probability distribution is quantified using KL divergence. First, the posterior probability of each candidate fault cause is predicted according to Bayes' theorem when positive and negative evidence are collected, respectively. Then, the KL divergence between the posterior probability distribution and the current prior probability distribution is calculated. Finally, the weighted average of the KL divergences under the positive and negative evidence cases is taken as the degree of influence of the candidate inspection point on the probability distribution of candidate fault causes.

[0043] The diagnostic value calculation unit is used to calculate the diagnostic value of a candidate inspection point based on the change in uncertainty of the candidate fault cause before and after the candidate inspection point evidence. The formula is expressed as follows: in, The diagnostic value of candidate inspection points. To gather evidence for candidate inspection point i The expected information entropy indicates that a positive diagnostic value means that collecting this evidence can reduce uncertainty, and the larger the value, the higher the diagnostic value; a negative diagnostic value means that collecting this evidence will increase uncertainty, and the inspection point should be avoided.

[0044] The value correction unit is used to correct the diagnostic value based on the distance, time, energy consumption, safety risks, and communication status of the robot reaching the candidate inspection point, resulting in the corrected diagnostic value of the candidate inspection point, expressed by the formula: in, The diagnostic value of candidate inspection point i after correction. The cost coefficient is obtained by weighted fusion of distance cost, time cost, energy cost, security risk cost and communication risk cost; each cost item is normalized and the value range is [0,1]; the weight coefficient is dynamically adjusted according to the operation and maintenance needs. For example, for emergency abnormal inspection tasks, the weight of time cost is appropriately increased; for low-battery robots, the weight of energy cost is appropriately increased.

[0045] The path planning module is used to generate inspection paths for the inspection robot based on robot state data, diagnostic value, and path planning constraints.

[0046] The dynamic update module is used to update the probability of abnormal root causes based on real-time evidence data collected during the robot's inspection process, and adjust the next inspection path based on the update results.

[0047] The dynamic update module includes a real-time evidence receiving unit, an evidence validity judgment unit, a root cause probability update unit, a diagnostic map weight update unit, and a next path adjustment unit. The real-time evidence receiving unit receives real-time evidence data collected by the robot during inspections. This real-time evidence data is transmitted to the system server via a 5G / 4G wireless communication network using the MQTT protocol, supporting low-latency and highly reliable data transmission. The system has a data buffer; when a network interruption occurs, the robot caches the evidence data locally, and automatically resumes transmission after network recovery. The real-time evidence data includes a timestamp, robot ID, inspection point ID, evidence type, raw data, and pre-processed data.

[0048] The evidence validity judgment unit is used to determine whether real-time evidence data meets preset validity conditions based on sensor status, acquisition distance, acquisition angle, environmental interference level, and data integrity. Preset validity conditions include: normal sensor operation, acquisition distance within the optimal acquisition distance range, acquisition angle deviation not exceeding ±15°, environmental interference intensity below a preset threshold, and data integrity reaching 100%. For image evidence, sharpness detection, blur detection, and occlusion detection are also required. Sharpness is calculated using the Laplacian variance algorithm; a variance value below the threshold indicates image blur. For infrared thermal imaging evidence, temperature calibration and dead pixel detection are also required. For audio evidence, signal-to-noise ratio (SNR) detection is required; a SNR below the threshold indicates excessive background noise.

[0049] The root cause probability update unit is used to update the probability of candidate fault causes based on real-time evidence data when the real-time evidence data meets preset validity conditions. The formula is expressed as: in, The probability of the updated candidate fault causes. Candidate fault causes The prior probability, Candidate fault causes Evidence observed at the time of occurrence The conditional probability is determined by the causal association weights in the equipment anomaly causal diagnosis diagram. As evidence The marginal probability; after each update, the probability of all candidate fault causes is normalized to ensure that the sum of the probabilities is 1.

[0050] The diagnostic graph weight update unit is used to update the association weights between abnormal phenomenon nodes, candidate fault cause nodes, and observable evidence point nodes in the equipment anomaly causal diagnostic graph based on the probability changes of candidate fault causes. The steps include: The system acquires real-time evidence data collected by the robot, evidence validity assessment results, and root cause probability update results. It also obtains basic information about the current inspection task, including task type, anomaly level, equipment model, and operating environment information, for subsequent fine-grained updates of the associated weights.

[0051] Identify the observable evidence point nodes corresponding to the real-time evidence data, and determine the abnormal phenomenon nodes and candidate fault cause nodes associated with the observable evidence point nodes. Query all abnormal phenomenon nodes and candidate fault cause nodes that have a direct causal relationship with the observable evidence point node using the adjacency matrix of the equipment anomaly causal diagnosis graph, and simultaneously query the indirectly associated secondary nodes.

[0052] When real-time evidence data increases the probability of candidate fault causes, the association weight between the corresponding observable evidence point node and the candidate fault cause node is increased, as expressed by the formula: in, The updated causal association weights between observable evidence point nodes and candidate fault cause nodes. The causal association weights before the update. The learning rate has a value range of [0.01, 0.1]. This represents the increase in the probability of a candidate fault cause. As an incremental limiting factor, it ensures that the weight will not exceed 1: when When the probability increases by more than 1, the increment approaches 0, and the weights will smoothly converge to 1 to avoid weight overflow. When the probability increases by more than 0.5, the learning rate automatically increases to 0.1 to accelerate the convergence of the weights.

[0053] When real-time evidence data reduces the probability of a candidate fault cause, the association weight between the corresponding observable evidence point node and the candidate fault cause node is reduced, as expressed by the formula: in, This represents the reduction in the probability of a candidate cause of failure. When the probability drops to 0, the corresponding association weight is set to 0, indicating that there is no causal relationship between the evidence and the cause of failure.

[0054] When real-time evidence data is determined to be invalid or low-reliability evidence, the evidence reliability weight of the corresponding observable evidence point node is reduced, as expressed by the formula: in, The updated evidence reliability weights for the observable evidence point nodes. The reliability weight of the evidence before the update is 0.1, which is a fixed attenuation coefficient, meaning that the reliability weight is reduced by 10% for each invalid evidence collection. When the reliability weight is lower than 0.3, the system automatically sends a sensor calibration prompt to the maintenance personnel, suggesting that the sensor corresponding to the inspection point be calibrated or replaced.

[0055] After maintenance personnel manually confirm the root cause of an anomaly, the association weights between the anomaly node, candidate fault cause node, and observable evidence point node are adjusted based on the manual confirmation results. For manually confirmed true root causes, the association weight between them and their corresponding anomaly node and observable evidence point node is increased to above 0.9; for fault causes that are manually ruled out, the association weight between them and their corresponding anomaly node and observable evidence point node is reduced to below 0.1. The manual confirmation results are used as high-quality training data for periodic global optimization and updates to the equipment anomaly cause-effect diagnostic graph.

[0056] The corrected correlation weights are written back to the equipment anomaly causal diagnosis graph for subsequent candidate inspection point diagnostic value calculation and next inspection path adjustment. Simultaneously, the corrected correlation weights and corresponding inspection cases are stored in the historical maintenance database, forming a closed-loop learning mechanism. The system periodically uses all historical inspection cases to batch train and update the equipment anomaly causal diagnosis graph, continuously improving the accuracy of causal diagnosis.

[0057] The next path adjustment unit is used to recalculate the diagnostic value of the remaining candidate inspection points based on the updated candidate fault cause probabilities and the updated association weights, and adjust the next inspection path. When the probability of any candidate fault cause reaches the preset root cause confirmation threshold (default 0.9), the system automatically determines that the abnormal root cause has been located, terminates the diagnostic inspection of the remaining inspection points, and retains only the necessary routine inspection points; when the probability of all candidate fault causes is less than 0.1, the system automatically sends a root cause not located prompt to the maintenance personnel, suggesting expanding the inspection scope or conducting manual investigation; the path adjustment adopts an incremental replanning method, only replanning the unexecuted path part, while the executed path part remains unchanged, in order to reduce the time and energy consumption loss caused by path adjustment.

[0058] The path planning module includes a diagnostic layer path planning unit, a motion layer path planning unit, a cost calculation unit, a comprehensive optimization unit, a path verification unit, and a replanning unit. The diagnostic layer path planning unit determines the set of target inspection points for verifying the root causes of anomalies from candidate inspection points based on diagnostic value, probability of root causes, and path planning constraints. The steps include: Obtain the set of candidate fault causes corresponding to the current anomaly and the current probability of each candidate fault cause. At the same time, obtain the fault impact range and urgency of each candidate fault cause. Fault causes with a wide impact range and high urgency are given higher priority weight when selecting subsequent inspection points.

[0059] Based on the equipment anomaly cause-effect diagnostic diagram, a set of candidate inspection points associated with each candidate fault cause is determined. For each candidate fault cause, the nodes with the highest observable evidence points (default 5) are selected as candidate inspection points, sorted from high to low according to their correlation weight; at the same time, inspection points that have already been inspected and inspection points that do not meet the path planning constraints are excluded.

[0060] The diagnostic value of each candidate inspection point in the candidate inspection point set is obtained, and the candidate inspection point set is filtered in conjunction with path planning constraints. Candidate inspection points that do not meet safety constraints, restricted area constraints, power constraints, time constraints, or human-robot collaboration constraints are eliminated. The power constraint adopts a conservative estimation method to calculate the total energy consumption required for the robot to travel from its current position to the inspection point and back to the nearest charging position. If the total energy consumption exceeds 80% of the robot's remaining power, it is determined that the power constraint is not met. The time constraint calculates the total time required to complete the inspection point and all subsequent necessary inspection points. If the total time exceeds the end time required for the inspection task, it is determined that the time constraint is not met.

[0061] Calculate the comprehensive verification priority of candidate inspection points based on their diagnostic value, probability of candidate fault causes, and priority of maintenance intentions. The formula is expressed as: in, The highest probability among all candidate fault causes associated with a candidate inspection point. The priority weights for operation and maintenance intentions are set as follows: 1.5 for emergency priority, 1.2 for high priority, 1.0 for medium priority, and 0.8 for low priority.

[0062] Candidate inspection points are selected from high to low priority based on comprehensive verification criteria to form a target inspection point set for verifying the root causes of anomalies. The size of the set is dynamically adjusted based on the inspection task time and robot battery level, with a default maximum of 20 inspection points. When the probability of any candidate fault cause reaches the preset root cause confirmation threshold (i.e., exceeding 80% of the available time), the selection of candidate inspection points unrelated to that fault cause is stopped.

[0063] The motion layer path planning unit uses the target inspection point set and robot state data to generate a feasible motion path for the robot to reach the target inspection point set. First, a globally optimal path is generated, considering road topology, restricted areas, and static obstacles. Then, local path planning is performed using the dynamic window method to avoid dynamic obstacles in real time. When generating the path, the robot's kinematic constraints are considered, including minimum turning radius, maximum acceleration, and maximum deceleration, to ensure that the generated path is a smooth path that the robot can actually execute.

[0064] The cost calculation unit is used to calculate risk indicators for feasible travel paths, including path distance, travel time, energy consumption, safety risk, communication risk, and obstacle crossing risk. The steps include: Based on the park's spatial map and road topology, the path distances of feasible travel routes are calculated. The path distance calculation uses the road centerline distance from a high-precision map, while also considering the actual driving distance at turns and elevation changes at inclines and declines. For paths composed of multiple road segments, the distances of each segment are added together to obtain the total path distance.

[0065] The travel time for feasible paths is calculated based on the robot's motion state, road conditions, and dwell time at candidate inspection points. Travel time includes both robot travel time and inspection point dwell time. The travel time is calculated based on the minimum of the maximum permissible speed for each road segment and the robot's current maximum speed, taking into account deceleration during turns and uphill / downhill sections. Inspection point dwell time is determined by the type and quantity of evidence that can be collected at that point; the default dwell time for a single inspection point is 10 seconds, with appropriately extended dwell times for inspection points involving multiple evidence collection tasks.

[0066] Calculate the energy consumption of feasible motion paths based on path distance, robot load, battery status, movement speed, and ground slope. The formula is expressed as: Where d is the path distance. Let v be the ground slope angle and v be the average driving speed. For average acceleration, , , The energy consumption model coefficients are calibrated through real vehicle testing; the energy consumption at the inspection point includes sensor operating energy consumption and computing energy consumption, calculated based on the dwell time and the power of each sensor.

[0067] Safety risks are calculated based on historical anomaly records, population density, vehicle traffic flow, hazardous area distribution, and environmental perception data of the areas traversed by feasible movement paths. Safety risks are categorized into four dimensions: personnel collision risk, vehicle collision risk, fall risk, and environmental hazard risk. Personnel collision risk is calculated based on the historical number of personnel collisions and real-time population density in the area; vehicle collision risk is calculated based on the historical number of vehicle collisions and real-time vehicle traffic flow in the area; fall risk is calculated based on the distribution of potholes, steps, and falling objects near the path; and environmental hazard risk is calculated based on the distribution of flammable and explosive areas, high-pressure areas, and high-temperature areas. The total safety risk is obtained by weighted fusion of risks from each dimension.

[0068] Communication risk is calculated based on the communication coverage strength, historical communication interruption records, and remote control requirements of the areas traversed by feasible movement paths. Communication coverage strength is obtained through a heat map of the park's communication signals; areas with signal strength below -100dBm are identified as weak communication areas. Historical communication interruption records are compiled by analyzing the number of communication interruptions and the average interruption duration in the area over the past 30 days. Remote control requirements are determined based on the type of inspection task; abnormal emergency inspection tasks have higher remote control requirements. The formula for calculating communication risk C is: Where S is the normalized communication signal strength, N is the number of historical communication interruptions, and W is the weight of remote control demand.

[0069] Based on the robot's real-time perception of the number of dynamic obstacles, their direction of movement, obstacle density, and road width, the obstacle passage risk is calculated. The obstacle passage risk is determined by predicting the future trajectory of the dynamic obstacles based on their motion state and calculating the time to collision (TTC) between the robot and the obstacle. When the TTC is less than a safe threshold (default 5 seconds), a collision risk is considered to exist. Simultaneously, the passage difficulty is calculated based on obstacle density and road width; higher obstacle density and narrower road width result in a higher passage risk.

[0070] The path distance, travel time, energy consumption, safety risk, communication risk, and obstacle passage risk are normalized and then weighted and fused according to preset weights to obtain the risk index of the feasible movement path. The normalization process adopts the min-max standardization method, mapping each cost term to the [0,1] interval; the preset weights can be adjusted by the operation and maintenance personnel according to actual needs, and the default weights are: path distance 0.2, travel time 0.2, energy consumption 0.2, safety risk 0.2, communication risk 0.1, and obstacle passage risk 0.1; the smaller the risk index value, the better the path.

[0071] The integrated optimization unit is used to optimize based on risk indicators, with the goal of maximizing the diagnostic value obtained per unit of driving cost. ,in It is the sum of the corrected diagnostic values ​​of all target inspection points along the path. The objective function is defined as the total risk index of the path, and the target inspection path is selected from multiple feasible movement paths. A genetic algorithm is used to solve the objective function, with a population size of 50, an iteration count of 100, a crossover probability of 0.8, and a mutation probability of 0.1. The optimal solution obtained is the target inspection path.

[0072] The path verification unit is used to verify whether the target inspection path meets the constraints of power, time, safety, restricted area, and human-machine collaboration. It simulates the entire process of the robot traveling along the target inspection path in the park's digital twin environment, monitoring the robot's remaining power, travel time, and safety status in real time. If a constraint violation occurs during the simulation, the path is marked as invalid, and the system returns to the comprehensive optimization unit to select a suboptimal path. If all feasible paths fail to meet the constraints, a constraint adjustment prompt is sent to the maintenance personnel.

[0073] The replanning unit is used to redefine the target inspection point set and generate the next inspection path when abnormal events occur during robot inspection, including changes in obstacles, changes in the probability of root causes of anomalies, changes in manual constraints, or abnormal robot states. Triggering conditions include: the LiDAR detecting an impassable static obstacle ahead; dynamic obstacles making the original path unsafe to traverse; changes in the probability of root causes of anomalies exceeding a preset threshold (default 0.3); maintenance personnel issuing new inspection constraints through the human-machine collaborative interaction module; the robot's battery level falling below a warning threshold (default 20%); and abnormal states such as sensor or drive malfunctions in the robot. The replanning process uses an incremental replanning method, replanning only the portion of the path affected by the abnormal event to improve replanning efficiency.

[0074] In summary, the human-machine collaborative park maintenance robot inspection path planning system provided by this invention deeply integrates causal reasoning theory with human-machine collaborative technology to construct an intelligent inspection path planning system guided by anomaly root cause diagnosis, breaking through the technical bottleneck of traditional inspection path planning that only focuses on point coverage and driving efficiency. By constructing a device anomaly causal diagnosis map mapped to physical space, a quantitative causal relationship between abnormal phenomena, candidate fault causes, and observable evidence points is established, realizing the unification of fault diagnosis logic and park spatial location. Through an evidence value assessment mechanism based on information entropy, the contribution of each candidate inspection point to reducing root cause uncertainty is quantitatively calculated, and the diagnostic value is corrected by comprehensively considering multiple dimensions such as driving distance, time, energy consumption, and safety risks. This solves the problems of strong blindness and many ineffective inspections in traditional inspections, achieving the core objective of "obtaining the greatest root cause certainty with the least inspection cost". By employing a hierarchical path planning architecture that separates the diagnostic and motion layers, the optimal set of target inspection points is first determined based on diagnostic value. Then, feasible motion paths that satisfy kinematic and safety constraints are generated. Global optimization is performed with the goal of maximizing diagnostic value per unit travel cost, resulting in an optimal inspection path that balances diagnostic efficiency, travel safety, and resource utilization. This improves the speed of detecting abnormal faults and the accuracy of root cause localization within the park, reduces the intensity and cost of manual maintenance, and enhances the intelligence, automation, and standardization of park operations and maintenance.

[0075] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A park operation and maintenance robot inspection path planning system based on human-machine collaboration, characterized in that, include: The park information modeling module is used to collect park spatial maps, road topology, equipment distribution, inspection points, restricted areas, charging locations, and historical operation and maintenance data to build a park operation and maintenance scenario model. The robot state perception module is used to collect robot state data, including the current position, battery status, motion status, sensor status, and environmental perception data of the inspection robot. The human-machine collaborative interaction module is used to receive inspection tasks and maintenance intentions input by maintenance personnel, and to convert the maintenance intentions into path planning constraints. The causal diagnosis graph construction module is used to construct a causal diagnosis graph of equipment anomalies based on the correlation between park equipment, abnormal phenomena, candidate fault causes and observable evidence points. The evidence value assessment module is used to calculate the amount of uncertainty reduction in the judgment of the root cause of the abnormality of each candidate inspection point based on the equipment abnormality cause-effect diagnostic diagram, and to obtain the diagnostic value of each candidate inspection point. The path planning module is used to generate an inspection path for the inspection robot based on the robot state data, the diagnostic value, and the path planning constraints. The dynamic update module is used to update the probability of abnormal root causes based on real-time evidence data collected during the robot's inspection process, and adjust the next inspection path based on the update results.

2. The park operation and maintenance robot inspection path planning system based on human-machine collaboration according to claim 1, characterized in that, The steps by which the human-machine collaborative interaction module transforms the operation and maintenance intention into path planning constraints include: The system receives maintenance intentions input by maintenance personnel, which include key inspection areas, key inspection equipment, types of anomalies to be verified, inspection time requirements, avoidance areas, inspection priorities, and security levels. The operational intent is semantically parsed to extract intent keywords, target objects, spatial scope, time scope, and priority information; The intent keywords are matched with the equipment type, inspection point, regional attributes and historical operation and maintenance data in the park information modeling module to determine the target inspection point set corresponding to the operation and maintenance intent; Generate path planning constraints based on the target inspection point set; When there are conflicts among multiple path planning constraints, the conflicts among the multiple path planning constraints are resolved according to the priority of operation and maintenance intention, the level of abnormal risk and robot status data, so as to obtain the final path planning constraints.

3. The park operation and maintenance robot inspection path planning system based on human-machine collaboration according to claim 1, characterized in that, The causal diagnosis graph construction module includes an anomaly node construction unit, a fault cause node construction unit, an evidence point node construction unit, a correlation relationship construction unit, and a spatial mapping unit. The anomaly node construction unit constructs anomaly phenomenon nodes based on the operating status of park equipment and historical anomaly records. The fault cause node construction unit constructs candidate fault cause nodes based on equipment structure, maintenance rules, and historical fault records. The evidence point node construction unit constructs observable evidence point nodes based on the types of data that the robot can collect and the park inspection points. The correlation relationship construction unit establishes causal relationships between anomaly phenomenon nodes, candidate fault cause nodes, and observable evidence point nodes, and assigns causal relationship weights to these relationships. The spatial mapping unit maps the observable evidence point nodes to their actual locations on the park spatial map, thus linking the equipment anomaly causal diagnosis graph with the park spatial map.

4. The park operation and maintenance robot inspection path planning system based on human-machine collaboration according to claim 3, characterized in that, The evidence value assessment module includes a root cause probability initialization unit, an uncertainty calculation unit, an evidence impact assessment unit, a diagnostic value calculation unit, and a value correction unit. The root cause probability initialization unit determines the initial probability of candidate fault causes based on historical maintenance data, current anomalies, equipment importance, and the maintenance intent. The uncertainty calculation unit calculates the uncertainty of candidate fault causes under the current known evidence conditions. The evidence impact assessment unit assesses the influence of candidate inspection points on the probability distribution of candidate fault causes based on the type of evidence collected at the candidate inspection points, evidence reliability, causal association weight, and historical validity. The diagnostic value calculation unit calculates the diagnostic value of the candidate inspection point based on the change in uncertainty of candidate fault causes before and after the evidence at the candidate inspection point. The value correction unit corrects the diagnostic value based on the distance, time, energy consumption, safety risks, and communication status of the robot reaching the candidate inspection point, obtaining the corrected diagnostic value of the candidate inspection point.

5. The park operation and maintenance robot inspection path planning system based on human-machine collaboration according to claim 4, characterized in that, The steps of the evidence impact assessment unit in evaluating the degree of influence of candidate inspection points on the probability distribution of candidate fault causes include: Obtain the collectable evidence types corresponding to candidate inspection points, and determine the causal association weight between the collectable evidence types and candidate fault causes; The reliability of the evidence data corresponding to the candidate inspection points is determined based on the robot's sensor status, acquisition accuracy, historical acquisition success rate, and degree of environmental interference. The historical validity of the candidate inspection points is determined based on the evidence hit rate of the candidate inspection points under similar abnormal scenarios in historical operation and maintenance data. Based on the aforementioned causal association weights, evidence reliability, and historical validity, the probability correction coefficients for each candidate fault cause of the candidate inspection point are calculated. Based on the probability correction coefficient, the degree of change in the probability distribution of candidate fault causes after collecting evidence from the candidate inspection points is predicted, and this change is taken as the degree of influence of the candidate inspection points on the probability distribution of candidate fault causes.

6. The park operation and maintenance robot inspection path planning system based on human-machine collaboration according to claim 1, characterized in that, The path planning module includes a diagnostic layer path planning unit, a motion layer path planning unit, a cost calculation unit, a comprehensive optimization unit, a path verification unit, and a replanning unit; the diagnostic layer path planning unit is used to determine the set of target inspection points for verifying the root causes of anomalies from the candidate inspection points based on the diagnostic value, the probability of the root cause of anomalies, and the path planning constraints. The motion layer path planning unit is used to generate a feasible motion path for the robot to reach the target inspection point set based on the target inspection point set and the robot state data. The cost calculation unit is used to calculate the risk index of the feasible movement path; The comprehensive optimization unit is used to select a target inspection path from multiple feasible movement paths based on the risk indicators, with the optimization objective of maximizing the diagnostic value obtained per unit driving cost. The path verification unit is used to verify whether the target inspection path meets the constraints of power, time, safety, restricted area and human-machine collaboration; the replanning unit is used to redetermine the target inspection point set and generate the next inspection path when an abnormal event occurs during the robot inspection process.

7. The park operation and maintenance robot inspection path planning system based on human-machine collaboration according to claim 6, characterized in that, The steps of the diagnostic layer path planning unit in determining the target inspection point set for verifying the root causes of anomalies include: Obtain the set of candidate fault causes corresponding to the current abnormal phenomenon and the current probability of each candidate fault cause; Based on the equipment anomaly cause-and-effect diagnosis diagram, determine the set of candidate inspection points associated with each candidate fault cause; Obtain the diagnostic value of each candidate inspection point in the candidate inspection point set, and filter the candidate inspection point set in combination with the path planning constraints; remove candidate inspection points that do not meet the safety constraints, restricted area constraints, power constraints, time constraints, or human-machine collaboration constraints. Calculate the comprehensive verification priority of candidate inspection points based on their diagnostic value, probability of candidate fault causes, and priority of maintenance intentions. Candidate inspection points are selected from high to low according to the comprehensive verification priority to form a target inspection point set for verifying the root cause of the anomaly; when the probability of any candidate fault cause reaches the preset root cause confirmation threshold, the selection of candidate inspection points unrelated to the candidate fault cause is stopped.

8. The park operation and maintenance robot inspection path planning system based on human-machine collaboration according to claim 6, characterized in that, The steps of the cost calculation unit in calculating the risk index of the feasible movement path include: Based on the park's spatial map and road topology, calculate the path distance of the feasible movement path; The travel time of the feasible movement path is calculated based on the robot's motion state, road conditions, and dwell time at candidate inspection points. The energy consumption of the feasible motion path is calculated based on the path distance, robot load, battery status, movement speed, and ground slope. Based on historical anomaly records, population density, vehicle flow, distribution of dangerous areas, and environmental perception data of the areas traversed by the feasible movement path, calculate the safety risks; Based on the communication coverage strength of the areas traversed by the feasible movement path, historical communication interruption records, and remote control requirements, calculate the communication risk; The robot calculates the obstacle passage risk based on the number of dynamic obstacles, the direction of obstacle movement, the obstacle density, and the road width as perceived in real time. The path distance, travel time, energy consumption, safety risk, communication risk, and obstacle passage risk are normalized and then weighted and fused according to preset weights to obtain the risk index of the feasible movement path.

9. The park operation and maintenance robot inspection path planning system based on human-machine collaboration according to claim 1, characterized in that, The dynamic update module includes a real-time evidence receiving unit, an evidence validity judgment unit, a root cause probability update unit, a diagnostic map weight update unit, and a next path adjustment unit. The real-time evidence receiving unit receives real-time evidence data collected by the robot during inspection. The evidence validity judgment unit determines whether the real-time evidence data meets preset validity conditions based on sensor status, acquisition distance, acquisition angle, environmental interference level, and data integrity. The root cause probability update unit updates the probability of candidate fault causes based on the real-time evidence data when it meets the preset validity conditions. The diagnostic map weight update unit updates the association weights between abnormal phenomenon nodes, candidate fault cause nodes, and observable evidence point nodes in the equipment anomaly causal diagnostic map based on the changes in the probability of candidate fault causes. The next path adjustment unit is used to recalculate the diagnostic value of the remaining candidate inspection points based on the updated candidate fault cause probability and the updated association weight, and adjust the next inspection path.

10. The park operation and maintenance robot inspection path planning system based on human-machine collaboration according to claim 9, characterized in that, The steps for updating the associated weights in the diagnostic graph weight update unit include: Acquire real-time evidence data collected by the robot, results of evidence validity assessment, and root cause probability update results; Determine the observable evidence point nodes corresponding to the real-time evidence data, and determine the abnormal phenomenon nodes and candidate fault cause nodes associated with the observable evidence point nodes; When the real-time evidence data increases the probability of candidate fault causes, the association weight between the corresponding observable evidence point node and the candidate fault cause node is increased. When the real-time evidence data reduces the probability of a candidate fault cause, the association weight between the corresponding observable evidence point node and the candidate fault cause node is reduced. When the real-time evidence data is determined to be invalid or low-reliability evidence, the evidence reliability weight of the corresponding observable evidence point node is reduced. After the operation and maintenance personnel manually confirm the root cause of the anomaly, they adjust the correlation weight between the anomaly node, the candidate fault cause node, and the observable evidence point node based on the manual confirmation result. The corrected correlation weights are written back to the device anomaly causal diagnosis graph for subsequent candidate inspection point diagnostic value calculation and next inspection path adjustment.

Images