Explosion-proof area autonomous inspection method and system for robot dog

Abstract

The invention discloses an explosion-proof area autonomous inspection method and system for a robot dog, and relates to the technical field of intelligent inspection of machines. The invention discloses an explosion-proof area autonomous inspection system for a robot dog. The explosion-proof area autonomous inspection system comprises a map building and sorting module, a global path planning module, a global path smoothing module, a local path optimization module and a state detection and alarm module. According to the method, the optimal inspection path planning of the two-dimensional space is carried out between the two adjacent starting points or inspection points through the improved A * algorithm, the inspection path can be dynamically adjusted according to the width of the channel, namely, a wider channel is preferentially selected to effectively reduce the collision risk, and it is ensured that the robot dog keeps a safe distance from the channel wall in the narrow channel; therefore, the safety and reliability of robot dog inspection are improved, the passing ability and inspection efficiency of an inspection path can be improved in combination with the kinematic characteristics of the robot dog, unnecessary detour is reduced, and the safety of the autonomous inspection method and system for the anti-explosion area of the robot dog is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine intelligence inspection, in particular to a method and system for autonomous inspection of an explosion-proof area for a robot dog. BACKGROUND

[0002] An explosion-proof area refers to an area in which an explosive gas environment or a combustible dust environment may occur during production, such as an oil exploitation platform, a chemical plant, a gas station, and a coal mine, etc. These areas contain flammable and explosive gases, liquids, dust, or fibers, etc., which are prone to cause explosion accidents due to static electricity, electric sparks, and high temperature, etc., and pose a serious threat to the safety of personnel and equipment. Therefore, in order to timely discover and handle potential safety hazards in the explosion-proof area, such as equipment leakage, electrical faults, and static electricity accumulation, etc., so as to prevent the development of a fire or explosion accident and cause casualties or damage to the workers and equipment in the explosion-proof area, relevant enterprises need to conduct regular inspections of the explosion-proof area, and in order to reduce the safety risks of the inspection and improve the inspection efficiency, the relevant enterprises usually use robots or robot dogs to replace manual inspection of the explosion-proof area.

[0003] The existing robots or robot dogs usually determine the inspection order according to the positions of multiple inspection points when inspecting the explosion-proof area, then plan a global path between two adjacent inspection points in the inspection order according to the distribution of obstacles between the two points, and finally monitor the positions of dynamic obstacles during the inspection through various sensors and devices and dynamically adjust the local path accordingly. However, when the number of dynamic obstacles is large and the moving speed is fast, the robot or robot dog is likely to react slowly and fail to avoid obstacles, and in the conventional path planning method, the width of the path and the stability of the robot or robot dog on the path are not considered, which may cause the robot or robot dog to tip over or even collide with the surrounding obstacles during the inspection, thereby reducing the safety of the existing autonomous inspection method and system for the explosion-proof area.

[0004] Based on the above situation, the present application provides a method and system for autonomous inspection of an explosion-proof area for a robot dog with high safety. SUMMARY

[0005] In order to overcome the existing robot or robot dog in the explosion-proof area when the inspection, if the number of dynamic obstacles is more and the moving speed is fast, the reaction is slow and the obstacle avoidance fails, and in the conventional path planning method, the path width and the stability of the robot or robot dog on the path are not considered, so that the robot or robot dog may fall down or collide with the surrounding obstacles during the inspection, thereby causing the safety of the existing explosion-proof area autonomous inspection method and system to be low, the present application provides a kind of safety high for robot dog's explosion-proof area autonomous inspection method and system.

[0006] A kind of for robot dog's explosion-proof area autonomous inspection method, comprising the following steps: The explosion-proof area is scanned and data is collected by laser radar, camera and sensor, the two-dimensional plane information of the explosion-proof area is acquired and the corresponding two-dimensional map is constructed, the position information of the starting point and multiple inspection points is acquired, the inspection sequence and original path of the robot dog are acquired by ant colony algorithm; For any adjacent two starting points or inspection points in the inspection sequence of the robot dog, improved A* algorithm is used for optimal inspection path planning in two-dimensional space, to obtain the global inspection path between the adjacent two starting points or inspection points; Based on improved B-spline curve fitting, the global inspection path between all adjacent two starting points or inspection points is smoothed, to obtain the global smooth inspection path between the adjacent two starting points or inspection points; The explosion-proof area is scanned in real time, the real-time position of the dynamic obstacle in the explosion-proof area is acquired and the two-dimensional map is updated in real time, the future state information of the dynamic obstacle is predicted by motion prediction model based on convolutional neural network, based on the prediction result and combined with the motion state of the robot dog, the local obstacle avoidance path is optimized by model predictive control algorithm; The environment and equipment around the inspection point are collected by sensor and camera, the collected data is analyzed and processed, and the running state of the equipment is evaluated by using fault diagnosis model, and an alarm is issued when the equipment has abnormal state.

[0007] As a preferred aspect of the application, the two-dimensional plane information includes static obstacle position information, channel width information and equipment layout information in the explosion-proof area, and the future state information includes position information, motion direction information, motion speed information and acceleration information of the dynamic obstacle.

[0008] As a preferred aspect of the application, the specific steps of using improved A* algorithm for optimal inspection path planning in two-dimensional space to obtain the global inspection path between the adjacent two starting points or inspection points are as follows: discretize the two-dimensional map of the explosion-proof area into a grid, wherein each grid represents a node in the explosion-proof area, and divide the state of each node into unobstructed or obstructed according to the two-dimensional plane information of the explosion-proof area; determine the position coordinates of the two adjacent starting points or inspection points as the starting point and target point of the robot dog respectively, initialize the open list and the closed list, and add the starting point to the open list, wherein the open list is used to store the nodes to be evaluated, the closed list is used to store the nodes that have been evaluated, and the heuristic function is set as the Manhattan distance calculation formula; define the heuristic cost function wherein represents the actual cost from the starting point to the current node, represents the heuristic estimated cost from the current node to the target point, represents the channel width constraint term, and , represents the channel width, represents the weight coefficient of the channel width constraint term; take out the node with the minimum heuristic cost from the open list generate all adjacent nodes of the node for each adjacent node , judge whether the state of the node is obstructed, if yes, discard it, otherwise, judge whether the node is in the open list and the closed list, if it is in the closed list, skip it, if it is neither in the open list nor in the closed list, add it to the open list, update the actual cost , the heuristic estimated cost , the channel width constraint term and the heuristic cost of the node and record the parent node ; if the node is already in the open list, judge whether the path through the current node to the adjacent node is better, if yes, update the actual cost , the heuristic estimated cost , the channel width constraint term and the heuristic cost of the node and record the parent node ; add the current node to the closed list and remove it from the open list, continue to select the next node with the minimum heuristic cost from the open list for evaluation until the target point is selected; After the target point is found, the global inspection path between the two adjacent starting points or inspection points is constructed and output by tracing back the parent node information of each node from the target point to the starting point.

[0009] As a preferred aspect of the application, the global inspection path between the two adjacent starting points or inspection points is smoothed by the improved B-spline curve fitting, and the specific steps for obtaining the global smooth inspection path between the two adjacent starting points or inspection points are as follows: Obtain the size parameters of the robot dog, including length and width, and set the minimum safety distance between the robot dog and the static obstacle according to the size of the robot dog and the inspection task requirements Uniformly select control points from the global inspection path, and ensure that the starting point and the ending point of the global inspection path are selected as control points; For each control point , calculate the minimum distance between it and the static obstacle If , move the control point outward so that the distance between it and the static obstacle satisfies The moving direction is away from the static obstacle or along the normal direction of the global inspection path, and the specific position expression of the moved control point is as follows:

[0010] Where is the normal direction vector at the control point , and the moved control point replaces the control point ; Select the control point as the control point of the cubic B-spline curve, and the mathematical expression of the cubic B-spline curve is as follows:

[0011] Where represents the cubic B-spline basis function, represents the total number of control points; Calculate the points on the B-spline curve by the formula of the cubic B-spline curve , and form a smooth curve, then expand the maximum width of the robot dog to both sides based on the smooth curve to form a detection curve, calculate the signed distance between the points on the detection curve and the static obstacle, and if there is a case less than zero, adjust the position of the corresponding control point and regenerate the B-spline curve, and repeat the above steps until the generated B-spline curve meets the detection requirements, and the global smooth inspection path between the two adjacent starting points or inspection points is obtained.

[0012] As a preferred aspect of the application, the specific steps of using the model predictive control algorithm to optimize the local obstacle avoidance path based on the prediction results and in combination with the motion state of the robot dog are as follows: Obtain the current motion state information of the robot dog, including position, velocity, acceleration and orientation, set the MPC parameters, including prediction time domain length, control time domain length and sampling time; Construct an optimization problem, define state variables and control variables, wherein the state variables include the position, velocity, acceleration and orientation of the robot dog, and the control variables include the acceleration and angular velocity of the robot dog, and construct a state transition equation according to the kinematic model of the robot dog; Define the objective function based on the safety distance, inspection efficiency and motion stability, and the specific expression of the objective function is as follows:

[0013] Wherein represents the predicted safety distance of the robot dog and the dynamic obstacle at time , represents the velocity of the robot dog at time , represents the acceleration of the robot dog at time , , and are the weight coefficients of the safety distance, inspection efficiency and motion stability, respectively; Set the constraint conditions, including the safety distance constraint, the robot dog dynamics constraint and the path feasibility constraint; At each control period, solve the optimization problem according to the current motion state of the robot dog and the future state information of the dynamic obstacle to obtain the optimal control output, apply the optimal control input to the robot dog, and update the motion state of the robot dog in real time to realize dynamic obstacle avoidance of the robot dog and obtain the optimized local obstacle avoidance path.

[0014] As a preferred aspect of the application, the safety distance constraint is , wherein represents the preset minimum safety distance; the robot dog dynamics constraint is , and , wherein , and represent the minimum velocity, minimum acceleration and minimum turning angle of the robot dog, respectively, , and respectively represent the maximum speed, maximum acceleration and maximum steering angle of the robot dog, and the path feasibility constraint requires that the optimized local obstacle avoidance path and the global smooth inspection path maintain continuity within a preset range, wherein the continuity is quantitatively constrained by the path angle and curvature.

[0015] An explosion-proof area autonomous inspection system for a robot dog comprises: A map construction and sorting module is configured to perform omnidirectional scanning and data acquisition on the explosion-proof area by using a laser radar, a camera and a sensor, to obtain two-dimensional plane information of the explosion-proof area and construct a corresponding two-dimensional map, to obtain position information of a starting point and a plurality of inspection points, and to obtain an inspection sequence and an original path of the robot dog by using an ant colony algorithm. A global path planning module is configured to perform optimal inspection path planning in a two-dimensional space for any adjacent two starting points or inspection points in the inspection sequence of the robot dog by using an improved A* algorithm, to obtain a global inspection path between the adjacent two starting points or inspection points. A global path smoothing module is configured to perform smoothing processing on the global inspection path between the adjacent two starting points or inspection points by using an improved B-spline curve fitting, to obtain a global smooth inspection path between the adjacent two starting points or inspection points. A local path optimization module is configured to perform omnidirectional real-time scanning on the explosion-proof area, to obtain real-time positions of dynamic obstacles in the explosion-proof area and to perform real-time updating on the two-dimensional map, to predict future state information of the dynamic obstacles by using a motion prediction model based on a convolutional neural network, to optimize a local obstacle avoidance path based on the prediction result and in combination with a motion state of the robot dog, and to use a model predictive control algorithm. A state detection and alarm module is configured to perform data acquisition on an environment and equipment around an inspection point by using a sensor and a camera, to analyze and process the acquired data, and to evaluate a running state of the equipment by using a fault diagnosis model, and to issue an alarm when the equipment has an abnormal state.

[0016] The present application has the following advantages: 1. The present application performs optimal inspection path planning in a two-dimensional space for adjacent two starting points or inspection points by using an improved A* algorithm, which not only can dynamically adjust the inspection path according to a channel width, i.e., preferentially select a wider channel to effectively reduce collision risk and ensure that the robot dog maintains a safe distance from a channel wall in a narrow channel, thereby improving safety and reliability of robot dog inspection, but also can improve passability and inspection efficiency of the inspection path in combination with kinematic characteristics of the robot dog, to reduce unnecessary detours and improve safety of the explosion-proof area autonomous inspection method and system for the robot dog.

[0017] 2. The present invention smoothes the global inspection path between all two adjacent starting points or inspection points through improved B-spline curve fitting, which not only makes the global inspection path smoother, thereby improving the fluency and stability of the robot dog's inspection, but also can adjust the global inspection path based on the length and width of the robot dog and the width of the path, thereby effectively avoiding the robot dog from colliding with obstacles during inspection, ensuring the smooth completion of the inspection task, and improving the safety of this autonomous inspection method and system for explosion-proof areas for robot dogs.

[0018] 3. The present invention predicts the future state information of dynamic obstacles through a motion prediction model based on a convolutional neural network, and can obtain the movement trend of obstacles in advance, thereby providing prior information for the local obstacle avoidance of the robot dog. By combining the prediction results with the robot dog's own motion state and using the model predictive control algorithm to optimize the local obstacle avoidance path, it can effectively avoid dynamic obstacles while ensuring a safe distance, while taking into account inspection efficiency and motion smoothness. The objective function of the algorithm comprehensively considers the safe distance, inspection efficiency and motion smoothness, so that the robot dog can quickly and smoothly adjust the path in a dynamic environment to avoid collisions with dynamic obstacles, thereby improving the execution efficiency and reliability of the inspection task, and enhancing the safety and practicality of this autonomous inspection method and system for explosion-proof areas for robot dogs. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a flow chart of a method for autonomous inspection of explosion-proof areas by a robot dog, as adopted in an embodiment of the present invention.

[0020] Figure 2 This is a structural diagram of an autonomous inspection system for explosion-proof areas for a robot dog used in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0022] Example 1, a method for autonomous inspection of explosion-proof areas for a robot dog, such as Figure 1 As shown, the following steps are included: The explosion-proof area is scanned and data collected comprehensively using lidar, cameras, and sensors. Two-dimensional plane information of the explosion-proof area is obtained and a corresponding two-dimensional map is constructed. The location information of the starting point and multiple inspection points is obtained. The inspection sequence and original path of the robot dog are determined using an ant colony algorithm. For any two adjacent starting points or inspection points in the inspection sequence of the robot dog, an improved A* algorithm is used for optimal inspection path planning in two-dimensional space to obtain a global inspection path between the two adjacent starting points or inspection points; Based on the improved B-spline curve fitting, the global inspection path between any two adjacent starting points or inspection points is smoothed to obtain a global smooth inspection path between the two adjacent starting points or inspection points; The explosion-proof area is scanned in all directions in real time to obtain the real-time position of the dynamic obstacle in the explosion-proof area and update the two-dimensional map in real time. The future state information of the dynamic obstacle is predicted by a motion prediction model based on a convolutional neural network. Based on the prediction result and combined with the motion state of the robot dog, a model predictive control algorithm is used to optimize the local obstacle avoidance path. The environment and equipment around the inspection point are data collected by sensors and cameras. The collected data is analyzed and processed, and a fault diagnosis model is used to evaluate the running state of the equipment. When the equipment has an abnormal state, an alarm is issued.

[0023] It should be noted that the inspection sequence and the original path of the robot dog based on the position information of the starting point and the plurality of inspection points and obtained by the ant colony algorithm belong to mature prior art, and therefore will not be described in detail here. The motion prediction model based on a convolutional neural network is used to predict the future state information of the dynamic obstacle. A large amount of motion trajectory data of high-speed dynamic obstacles (such as forklifts and automated guided vehicles) in different scenarios needs to be collected in advance, including position, speed, acceleration, and turning angle information as training samples, and the motion prediction model needs to be trained. In actual inspection, the robot dog uses the motion prediction model to extract and analyze the real-time motion state of the high-speed dynamic obstacle, and can predict the position and speed change trend in the next few seconds. The analysis and processing of the collected data and the use of the fault diagnosis model to evaluate the running state of the equipment include but are not limited to combining gas concentration data with temperature and humidity data by a data fusion algorithm (such as a Bayesian fusion algorithm and a Kalman filter fusion algorithm), considering the influence of environmental temperature and humidity on the measurement results of the gas sensor, correcting the gas concentration data to obtain more accurate actual gas concentration values, and issuing an alarm when the actual gas concentration value exceeds the preset threshold, and spatially registering and fusing image data and infrared thermal imaging data, and superimposing temperature distribution information on the equipment appearance image to generate an intuitive equipment state visualization image, which facilitates remote monitoring personnel to quickly and accurately determine the equipment running state. For example, the remote monitoring personnel can clearly see that the temperature of a certain part of the equipment abnormally rises, and accurately locate the part on the image, so as to timely discover potential fault points and issue an alarm.

[0024] The two-dimensional plane information includes static obstacle position information, channel width information, and equipment layout information within the explosion-proof area, and the future state information includes dynamic obstacle position information, movement direction information, movement speed information, and acceleration information.

[0025] The specific steps of using the improved A* algorithm to perform optimal inspection path planning in two-dimensional space and obtain a global inspection path between two adjacent starting points or inspection points are as follows: The two-dimensional map of the explosion-proof area is discretized into grids, where each grid represents a node in the explosion-proof area, and the state of each node is classified as barrier-free or barrier-exposed based on the two-dimensional plane information of the explosion-proof area; Two adjacent starting points or inspection points are used as the starting point and target point of the robot dog respectively and their position coordinates are determined. The open list and closed list are initialized, and the starting point is added to the open list. The open list is used to store the nodes to be evaluated, and the closed list is used to store the evaluated nodes. The heuristic function is set to the Manhattan distance calculation formula. Define the heuristic cost function ,in Indicates the actual cost from the starting point to the current node, Represents the heuristic estimated cost from the current node to the target point, that is, the Manhattan distance from the current node to the target point, represents the channel width constraint, and , represents the channel width, Represents the weight coefficient of the channel width constraint; Take the node with the minimum heuristic cost from the open list , generate nodes All adjacent nodes of , judgment node Is the state of the node blocked? If so, discard it. Otherwise, judge the node Is it in the open list and closed list? If it is in the closed list, skip it. If it is neither in the closed list nor in the open list, add it to the open list and update the node. The actual cost , heuristic cost estimation , channel width constraint and heuristic cost And record its parent node ; If the node If already in the open list, determine through the current node Reach the adjacent node Is the path better? If so, update the node The actual cost heuristic cost channel width constraint term and heuristic cost and record its parent node ; add the current node to the closed list and remove it from the open list, continue to select the next node with the minimum heuristic cost from the open list for evaluation until the target point is selected; After the target point is found, the parent node information of each node is traced back from the target point to the starting point, and the global inspection path between the two adjacent starting points or inspection points is constructed and output.

[0026] It should be noted that when planning the global inspection path, the high-density obstacle region (i.e. the region where the number density of obstacles exceeds the preset threshold) can be identified based on the obstacle density information in the map, and then the state of all nodes contained in the entire region is regarded as having obstacles, so as to avoid these regions from a macroscopic point of view when planning the path, thereby effectively simplifying the global inspection path.

[0027] The above steps plan the optimal inspection path in two-dimensional space between two adjacent starting points or inspection points by using the improved A* algorithm, which not only dynamically adjusts the inspection path according to the channel width, i.e. preferentially selects wider channels to effectively reduce the risk of collision, and ensures that the robot dog maintains a safe distance from the channel wall in narrow channels, thereby improving the safety and reliability of the robot dog inspection, but also improves the passability and inspection efficiency of the inspection path in combination with the kinematic characteristics of the robot dog, thereby reducing unnecessary detours and improving the safety of the autonomous inspection method and system for the robot dog in the explosion-proof area.

[0028] The specific steps for smoothing the global inspection path between the two adjacent starting points or inspection points based on the improved B-spline curve fitting are as follows: Obtain the size parameters of the robot dog, including length and width, and set the minimum safety distance between the robot dog and static obstacles according to the size of the robot dog and the inspection task requirements , uniformly select control points from the global inspection path, and ensure that the starting point and the ending point of the global inspection path are selected as control points; For each control point , calculate the minimum distance between it and the static obstacle If , move the control point outward to make the distance between it and the static obstacle satisfy, the moving direction is the direction away from the static obstacle or the direction along the normal of the global inspection path, the specific position expression of the control point after moving is:

[0029] wherein is the normal direction vector at the control point , the control point after moving is substituted for the control point ; the selected control point is taken as the control point of the cubic B-spline curve, and the mathematical expression of the cubic B-spline curve is:

[0030] wherein represents the cubic B-spline basis function, represents the total number of control points; the points on the B-spline curve are calculated through the formula of the cubic B-spline curve , and a smooth curve is formed, the maximum width of the robot dog is expanded to both sides based on the smooth curve, a detection curve is formed, the signed distance between the points on the detection curve and the static obstacle is calculated, if there is a case less than zero, the position of the corresponding control point is adjusted and the B-spline curve is regenerated, and the cycle is repeated until the generated B-spline curve meets the detection requirements, and the global smooth inspection path between the two adjacent starting points or inspection points is obtained.

[0031] The above steps can not only make the global inspection path more smooth, thereby improving the smoothness and stability of the robot dog inspection, but also adjust the global inspection path based on the length and width of the robot dog and the width of the path, thereby effectively avoiding the collision between the robot dog and the obstacle during the inspection, so as to ensure the smooth completion of the inspection task and improve the safety of the anti-explosion area autonomous inspection method and system for the robot dog.

[0032] The specific steps of optimizing the local obstacle avoidance path based on the prediction result and combining the motion state of the robot dog are as follows: The current motion state information of the robot dog is obtained, including position, speed, acceleration and orientation, and MPC parameters are set, including prediction time domain length, control time domain length and sampling time; An optimization problem is constructed, and state variables and control variables are defined, wherein the state variables include the position, speed, acceleration and orientation of the robot dog, the control variables include the acceleration and angular velocity of the robot dog, and the state transition equation is constructed according to the kinematic model of the robot dog; A target function is defined based on the safety distance, the inspection efficiency and the motion stability, and a specific expression of the target function is:

[0033] wherein represents a predicted safety distance of the robot dog and the dynamic obstacle at time , represents a speed of the robot dog at time , represents an acceleration of the robot dog at time , , and are weight coefficients of the safety distance, the inspection efficiency and the motion stability, respectively; constraint conditions are set, including a safety distance constraint, a robot dog dynamics constraint and a path feasibility constraint; In each control cycle, an optimization problem is solved according to the current motion state of the robot dog and the future state information of the dynamic obstacle, an optimal control output is obtained, the optimal control input is applied to the robot dog, and the motion state of the robot dog is updated in real time, so that the dynamic obstacle avoidance of the robot dog is realized and an optimized local obstacle avoidance path is obtained.

[0034] The safety distance constraint is wherein represents a preset minimum safety distance; the robot dog dynamics constraint is , and wherein , and respectively represent a minimum speed, a minimum acceleration and a minimum steering angle of the robot dog, , and respectively represent a maximum speed, a maximum acceleration and a maximum steering angle of the robot dog; and the path feasibility constraint requires that the optimized local obstacle avoidance path and the global smooth inspection path maintain continuity within a preset range, wherein the continuity is quantitatively constrained by a path included angle and a curvature.

[0035] The above step can obtain the motion trend of the dynamic obstacle in advance by predicting the future state information of the dynamic obstacle based on the motion prediction model of the convolutional neural network, thereby providing prior information for local obstacle avoidance of the robot dog. The prediction result is combined with the motion state of the robot dog, and the model predictive control algorithm is used to optimize the local obstacle avoidance path, so that the dynamic obstacle can be effectively avoided under the premise of ensuring the safety distance, while the patrol efficiency and motion stability are taken into account. The objective function of the algorithm comprehensively considers the safety distance, the patrol efficiency and the motion stability, so that the robot dog can quickly and smoothly adjust the path in the dynamic environment to avoid collision with the dynamic obstacle, thereby improving the execution efficiency and reliability of the patrol task, and improving the safety and practicality of the autonomous patrol method and system for the explosion-proof area of the robot dog.

[0036] Embodiment 2, an autonomous patrol system for an explosion-proof area of a robot dog, as shown in Figure 2 comprises: a map construction and sorting module, configured to scan and collect data of the explosion-proof area in all directions by a laser radar, a camera and a sensor, obtain two-dimensional plane information of the explosion-proof area and construct a corresponding two-dimensional map, obtain position information of a starting point and a plurality of patrol points, and obtain a patrol order and an original path of the robot dog by an ant colony algorithm; a global path planning module, configured to plan an optimal patrol path in a two-dimensional space for any two adjacent starting points or patrol points in the patrol order of the robot dog by an improved A* algorithm, and obtain a global patrol path between the two adjacent starting points or patrol points; a global path smoothing module, configured to perform smoothing processing on the global patrol path between the two adjacent starting points or patrol points by an improved B-spline curve fitting, and obtain a global smooth patrol path between the two adjacent starting points or patrol points; a local path optimization module, configured to perform real-time scanning of the explosion-proof area in all directions, obtain real-time positions of dynamic obstacles in the explosion-proof area and perform real-time updating of the two-dimensional map, predict future state information of the dynamic obstacles by a motion prediction model based on a convolutional neural network, optimize a local obstacle avoidance path based on the prediction result and in combination with a motion state of the robot dog, and use a model predictive control algorithm; a state detection and alarm module, configured to collect data of an environment and equipment around the patrol point by a sensor and a camera, analyze and process the collected data, evaluate an operation state of the equipment by using a fault diagnosis model, and issue an alarm when the equipment has an abnormal state.

[0037] It is to be understood that all of the above modifications and alterations can be made to the above-described arrangements and that all such modifications and alterations are intended to be included within the scope of the present application. Those skilled in the art will readily appreciate that other modifications and alterations can be made to the present application without departing from the scope of the application.

Claims

1. A method for autonomous inspection of explosion-proof areas by a robot dog, characterized in that: The following steps are involved: The explosion-proof area is scanned and data collected comprehensively using lidar, cameras, and sensors. Two-dimensional plane information of the explosion-proof area is obtained and a corresponding two-dimensional map is constructed. The location information of the starting point and multiple inspection points is obtained. The inspection sequence and original path of the robot dog are determined using an ant colony algorithm. For any two adjacent starting points or inspection points in the inspection sequence of the robot dog, the improved A* algorithm is used to plan the optimal inspection path in two-dimensional space to obtain the global inspection path between the two adjacent starting points or inspection points; Based on the improved B-spline curve fitting, the global inspection path between all two adjacent starting points or inspection points is smoothed to obtain the global smooth inspection path between the two adjacent starting points or inspection points; Perform a full range of real-time scanning of the explosion-proof area, obtain the real-time position of dynamic obstacles within the explosion-proof area, and update the two-dimensional map in real time. Use a motion prediction model based on a convolutional neural network to predict the future state information of dynamic obstacles. Based on the prediction results and the robot dog's own motion state, the model predictive control algorithm is used to optimize the local obstacle avoidance path. Sensors and cameras are used to collect data on the environment and equipment around the inspection points, which are then analyzed and processed. Fault diagnosis models are used to evaluate the operating status of the equipment, and alarms are issued when the equipment is in an abnormal state.

2. The method for autonomous inspection of explosion-proof areas for a robot dog according to claim 1, characterized in that: The two-dimensional plane information includes static obstacle position information, channel width information, and equipment layout information within the explosion-proof area, and the future state information includes dynamic obstacle position information, movement direction information, movement speed information, and acceleration information.

3. The method for autonomous inspection of explosion-proof areas for a robot dog according to claim 2, characterized in that: The specific steps of using the improved A* algorithm to perform optimal inspection path planning in two-dimensional space and obtain a global inspection path between two adjacent starting points or inspection points are as follows: The two-dimensional map of the explosion-proof area is discretized into grids, where each grid represents a node in the explosion-proof area, and the state of each node is classified as barrier-free or barrier-exposed based on the two-dimensional plane information of the explosion-proof area; Two adjacent starting points or inspection points are used as the starting point and target point of the robot dog respectively and their position coordinates are determined. The open list and closed list are initialized, and the starting point is added to the open list. The open list is used to store the nodes to be evaluated, and the closed list is used to store the evaluated nodes. The heuristic function is set to the Manhattan distance calculation formula. Define the heuristic cost function ,in Indicates the actual cost from the starting point to the current node, represents the heuristic estimated cost from the current node to the target point, represents the channel width constraint, and , represents the channel width, Represents the weight coefficient of the channel width constraint; Take the node with the minimum heuristic cost from the open list , generate nodes All adjacent nodes of , judgment node Is the state of the node blocked? If so, discard it. Otherwise, judge the node Is it in the open list and closed list? If it is in the closed list, skip it. If it is neither in the closed list nor in the open list, add it to the open list and update the node. The actual cost , heuristic cost estimation , channel width constraint and heuristic cost And record its parent node ; If the node If already in the open list, determine through the current node Reach the adjacent node Is the path better? If so, update the node The actual cost , heuristic cost estimation , channel width constraint and heuristic cost And record its parent node ; The current node Add it to the closed list and remove it from the open list. Continue to select the next node with the smallest heuristic cost from the open list for evaluation until the target point is selected. After finding the target point, the parent node information of each node is traced back from the target point to the starting point, and the global inspection path between the two adjacent starting points or inspection points is constructed and output.

4. The method for autonomous inspection of explosion-proof areas for a robot dog according to claim 3, characterized in that: The specific steps of smoothing the global inspection path between all two adjacent starting points or inspection points based on the improved B-spline curve fitting to obtain the global smooth inspection path between the two adjacent starting points or inspection points are as follows: Get the size parameters of the robot dog, including length and width, and set the minimum safe distance between the robot dog and static obstacles according to the size of the robot dog and the inspection task requirements ,control points are evenly selected from the global inspection path, and,ensure that both the starting point and the end point of the global,inspection path are selected as control points; For each control point , calculate the minimum distance between it and the static obstacle ,if , then move the control point outward so that the distance between it and the static obstacle satisfies The moving direction is away from the static obstacle or outward along the normal of the global inspection path. The control point after moving The specific position expression is: ； in It is a control point The normal direction vector at , using the moved control point Replace control point ; Select the control points As the control points of the cubic B-spline curve, the mathematical expression of the cubic B-spline curve is: ； in represents the cubic B-spline basis function, Indicates the total number of control points; Through the cubic B-spline curve The formula is used to calculate the points on the B-spline curve and form a smooth curve. Based on the smooth curve, the maximum width of the robot dog is extended to both sides to form a detection curve. The signed distance between the points on the detection curve and the static obstacle is calculated. If the signed distance is less than zero, the position of the corresponding control point is adjusted and the B-spline curve is regenerated. This cycle is repeated until the generated B-spline curve meets the detection requirements, and a global smooth inspection path between two adjacent starting points or inspection points is obtained.

5. The method for autonomous inspection of explosion-proof areas for a robot dog according to claim 4, characterized in that: The specific steps of using the model predictive control algorithm to optimize the local obstacle avoidance path based on the prediction results and the robot dog's own motion state are as follows: Obtain the robot dog's current motion state information, including position, velocity, acceleration, and orientation, and set MPC parameters, including prediction time domain length, control time domain length, and sampling time; Construct an optimization problem and define state variables and control variables. State variables include the robot's position, velocity, acceleration, and orientation, and control variables include its acceleration and angular velocity. Construct a state transition equation based on the robot's kinematic model. The objective function is defined based on safety distance, inspection efficiency and motion smoothness. The specific expression of the objective function is: ； in Indicates the time between the robot dog and the dynamic obstacle The predicted safety distance, Indicates that the robot dog is at time speed, Indicates that the robot dog is at time The acceleration of 、 and are the weight coefficients of safety distance, inspection efficiency and motion smoothness respectively; Set constraints, including safety distance constraints, robot dog dynamics constraints, and path feasibility constraints; In each control cycle, the optimization problem is solved based on the current motion state of the robot dog and the future state information of dynamic obstacles to obtain the optimal control output. The optimal control input is applied to the robot dog, and the motion state of the robot dog is updated in real time to achieve dynamic obstacle avoidance of the robot dog and obtain an optimized local obstacle avoidance path.

6. The method for autonomous inspection of explosion-proof areas for a robot dog according to claim 5, characterized in that: The safety distance constraint is ,in Represents the preset minimum safety distance; the robot dog dynamics constraint is 、 and ,in 、 and They represent the minimum speed, minimum acceleration and minimum steering angle of the robot dog respectively. 、 and They represent the maximum speed, maximum acceleration and maximum steering angle of the robot dog respectively; the path feasibility constraint requires that the optimized local obstacle avoidance path and the global smooth inspection path maintain consistency within a preset range, where the consistency is quantified by the path angle and curvature.

7. An autonomous inspection system for explosion-proof areas for a robot dog, applied to the autonomous inspection method for explosion-proof areas for a robot dog according to any one of claims 1 to 6, characterized in that: Includes: The map construction and sorting module is used to perform comprehensive scanning and data collection of the explosion-proof area using lidar, cameras, and sensors. This module obtains two-dimensional planar information of the explosion-proof area and constructs a corresponding two-dimensional map. This module also obtains the location information of the starting point and multiple inspection points, and uses an ant colony algorithm to determine the inspection order and original path of the robot dog. The global path planning module is used to plan the optimal inspection path in two-dimensional space for any two adjacent starting points or inspection points in the inspection sequence of the robot dog using the improved A* algorithm, and obtain the global inspection path between the two adjacent starting points or inspection points; A global path smoothing module is used to smooth the global inspection path between all two adjacent starting points or inspection points based on improved B-spline curve fitting to obtain a global smooth inspection path between the two adjacent starting points or inspection points; The local path optimization module is used to perform a full-scale real-time scan of the explosion-proof area, obtain the real-time position of dynamic obstacles within the explosion-proof area, and update the two-dimensional map in real time. The future state information of dynamic obstacles is predicted through a motion prediction model based on a convolutional neural network. Based on the prediction results and the robot dog's own motion state, the model predictive control algorithm is used to optimize the local obstacle avoidance path; The status detection and alarm module is used to collect data on the environment and equipment around the inspection point through sensors and cameras, analyze and process the collected data, and use fault diagnosis models to evaluate the operating status of the equipment, and issue an alarm when the equipment is in an abnormal state.

Images