Inspection method and system for quadruped robot

Abstract

The invention discloses an inspection method and system for a quadruped robot, and relates to the technical field of robot intelligent control, and the method comprises the steps: collecting the data of an inspection region where the quadruped robot is located in real time, integrating the data into a data vector, and carrying out the preprocessing; planning an inspection path of the quadruped robot by using a deep reinforcement learning model, and optimizing the inspection path in real time by using an MPC; calculating the fault risk probability of the inspected equipment in the inspection area according to a fault risk prediction formula, and setting the task priority of the inspection task in the inspection area; and a quadruped robot task allocation and dynamic scheduling mechanism is constructed.

Description

Technical Field

[0001] The present invention relates to the technical field of robot intelligent control, and in particular to an inspection method and system of a quadruped robot. Background Art

[0002] In the field of industrial automation, quadruped robot inspection systems play a key role, especially in performing equipment status monitoring and troubleshooting tasks in complex or dangerous environments. Currently, these systems mainly rely on preset fixed paths for inspection, use fixed thresholds to trigger fault warnings, and rely on human decision-making to achieve task allocation and scheduling among multiple robots.

[0003] However, this traditional method has obvious limitations. First, the inspection method that relies on fixed paths cannot adapt to environmental changes, such as new obstacles or changes in the position of the inspected equipment, which may cause the inspection task to fail to be completed smoothly or the efficiency to drop significantly. Secondly, the fixed fault warning threshold is often no longer applicable when the environmental conditions fluctuate greatly or the types of inspected equipment are diverse, making it impossible to flexibly adjust the priority of the robot's task execution to adapt to the actual situation. In addition, the task allocation and scheduling process with manual participation is not only inefficient, but also easily affected by personal subjective judgment, which may cause the execution of inspection tasks to be not timely and accurate enough. Therefore, the existing inspection system urgently needs a more adaptive solution to solve the above problems. Summary of the invention

[0004] In view of the problems existing in the above-mentioned existing inspection methods and systems of quadruped robots, the present invention is proposed.

[0005] Therefore, the purpose of the present invention is to improve the adaptability of the quadruped robot inspection method.

[0006] In order to solve the above technical problems, the first aspect of the present invention provides a quadruped robot inspection method, comprising:

[0007] Collect data of the inspection area where the quadruped robot is located in real time, and integrate the data into data vectors for preprocessing;

[0008] Using a deep reinforcement learning model to plan the inspection path of the quadruped robot, and using MPC to optimize the inspection path in real time;

[0009] Calculate the failure risk probability of the inspected equipment in the inspection area according to the failure risk prediction formula and set the task priority of the inspection task in the inspection area;

[0010] Construct a task allocation and dynamic scheduling mechanism for quadruped robots.

[0011] A second aspect of the present invention provides a quadruped robot inspection system, comprising:

[0012] A data acquisition module is used to collect data of the inspection area where the quadruped robot is located in real time, and integrate the data into a data vector for pre-processing;

[0013] A path planning module, used to plan the inspection path of the quadruped robot using a deep reinforcement learning model, and to optimize the inspection path in real time using MPC;

[0014] A fault prediction module, used to calculate the fault risk probability of the inspected equipment in the inspection area according to the fault risk prediction formula and set the task priority of the inspection task in the inspection area;

[0015] The task allocation module is used to build the task allocation and dynamic scheduling mechanism of the quadruped robot.

[0016] A computer inspection device comprises: a memory and a processor; the memory stores a computer program, and the processor implements the steps of the inspection method of the quadruped robot in the first aspect when executing the computer program.

[0017] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the inspection method for a quadruped robot according to the first aspect.

[0018] The beneficial effects of the present invention are as follows: combining a deep reinforcement learning model to plan the inspection path of the quadruped robot, and using model predictive control (MPC) to optimize the path in real time, so that the flexibility and efficient adjustment of the robot's inspection path can be achieved in a dynamic environment. At the same time, by constructing a fault risk prediction formula, the failure probability of the inspected equipment in the inspected area can be calculated, and the priority of the inspection task can be set accordingly. This method can provide a more accurate fault risk assessment, ensure that high-risk inspected equipment is inspected first, and thus significantly improve the early warning capability of the system. On this basis, a task allocation and dynamic scheduling mechanism is established, which can achieve more adaptive task allocation results according to the specific conditions of the current environment. This mechanism not only improves the inspection efficiency, but also ensures the timeliness and accuracy of the inspection tasks, so that the quadruped robot inspection system can better adapt to the complex and changeable industrial environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0020] Figure 1 A schematic flow chart of a quadruped robot inspection method provided by an embodiment of the present invention;

[0021] Figure 2 A schematic diagram of the process of intelligent inspection by a quadruped robot provided by an embodiment of the present invention;

[0022] Figure 3 A schematic structural diagram of a quadruped robot inspection system provided by an embodiment of the present invention;

[0023] Figure 4 A schematic structural diagram of another quadruped robot inspection system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0025] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0026] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0027] Example 1

[0028] Reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and which provides a quadruped robot inspection method, the quadruped robot inspection method comprising:

[0029] S1. Collect data from the inspection area where the quadruped robot is located in real time, and integrate the data into data vectors for preprocessing.

[0030] Specifically, real-time data collection in the inspection area includes using sensors pre-installed on the inspected equipment to collect real-time status data of the inspected equipment and using sensors in the inspection area to collect real-time environmental data in the inspection area. This process realizes comprehensive monitoring of the inspected equipment and environment in the inspection area, provides an accurate and real-time data basis for subsequent path planning, risk prediction and intelligent early warning, and effectively improves the intelligence and response speed of the inspection system.

[0031] The environmental data includes the ambient temperature and humidity, and the status data includes the current of the inspected equipment, the temperature of the inspected equipment, and the vibration amplitude of the inspected equipment.

[0032] It should be noted that the above sensors collect environmental data and status data according to a preset collection frequency.

[0033] Furthermore, the collected status data and environmental data are integrated into data vectors and then preprocessed, including using statistical anomaly detection algorithms to identify and remove noise and erroneous data, using linear interpolation to complete missing data, and standardizing and normalizing the completed data. This process can effectively improve the quality and consistency of the data, ensure that subsequent analysis and decision-making are based on reliable data information, and thus improve the overall accuracy and robustness of the system.

[0034] S2. Use the deep reinforcement learning model to plan the inspection path of the quadruped robot, and use MPC to optimize the inspection path in real time.

[0035] Specifically, using a deep reinforcement learning model to plan the inspection path of the quadruped robot means obtaining the position of the quadruped robot and the position of the nearest obstacle to the quadruped robot in real time through GPS, and then integrating the position of the quadruped robot and the position of the nearest obstacle to the quadruped robot into a state vector, where the position is a three-dimensional coordinate.

[0036] The state vector includes the position of the quadruped robot at different time steps corresponding to each action and the position of the nearest obstacle to the quadruped robot;

[0037] According to the movement operations taken by the quadruped robot in three-dimensional space, the action space of the quadruped robot during the inspection process is defined. For example, if the quadruped robot can move in six directions: up, down, left, right, forward, and backward, the action space is defined as six discrete actions.

[0038] Up: [0,0,+1]——Move 1 unit along the positive direction of the z axis;

[0039] Down: [0,0,-1] - Move 1 unit along the negative z-axis;

[0040] Left: [0,-1,0] - Move 1 unit along the negative y-axis.

[0041] Right: [0,+1,0]——Move 1 unit along the positive y-axis;

[0042] Before: [+1,0,0]——Move 1 unit along the positive x-axis;

[0043] After: [-1,0,0]——Move 1 unit along the negative direction of the x-axis;

[0044] The target position of the quadruped robot is defined as the precise position coordinates of the inspected equipment, that is, the quadruped robot needs to reach the target position through path planning;

[0045] Use the deep reinforcement learning model for strategy learning, build an adaptive instant reward function, and calculate the instant reward value of the state vector at different time steps. The expression is:

[0046]

[0047] Among them, st represents the state vector at the tth time step, R(s t ) represents the instantaneous reward value obtained by the state vector st at the t-th time step, x t and x g They represent the x-coordinate and y-coordinate of the quadruped robot at the tth time step and the target position of the quadruped robot respectively. t and g They represent the y coordinates of the quadruped robot at the tth time step and the z coordinates of the target position of the quadruped robot. t and z g They represent the z coordinates of the quadruped robot at the tth time step and the target position of the quadruped robot, respectively. α represents the penalty coefficient of the obstacle, which is used to control the impact of obstacles on the path. The larger the value, the more likely the quadruped robot will avoid obstacles. t,o represents the distance between the quadruped robot and the nearest obstacle at the tth time step;

[0048] It is the distance term in the adaptive instant reward function, which represents the Euclidean distance between the quadruped robot and the target position. The negative sign indicates that the smaller the distance, the higher the reward. This term is used to guide the quadruped robot to approach the target position.

[0049] Construct an adaptive instant reward function. By calculating the Euclidean distance of the target position and the obstacle distance, the robot is given a positive reward for approaching the target and a penalty coefficient is imposed on obstacles. This reward function design enables the robot to obtain higher rewards when approaching the target position while avoiding obstacles to ensure the safety and effectiveness of the path.

[0050] Based on the adaptive instant reward function and discount factor, the policy value function is constructed to calculate the cumulative reward value of the state vector at different time steps. The formula is:

[0051]

[0052] Among them, V(s t ) is the policy value function, calculated on the state vector s t This function represents the cumulative reward of taking a series of actions under a given state vector. T represents the maximum number of time steps preset in the path planning process of the quadruped robot. γ represents the discount factor, which is used to gradually attenuate the influence of future rewards on the current decision. The smaller the discount factor, the more attention is paid to recent rewards, and the larger the discount factor, the more attention is paid to future rewards.

[0053] Based on this immediate reward function and discount factor, a policy value function is constructed to calculate the cumulative rewards at different time steps, thereby measuring the overall effect of each path;

[0054] Construct a deep Q network and transform the state vector s t As input, the Q value of each action is output, and the Q value is updated at each time step using the Q-learning update formula according to the cumulative reward value obtained by the policy value function. The formula is:

[0055] Q(s t ,a t )←Q(s t ,a t )+[R(s t )+γV(s t+1 )-Q(s t ,a t )](Formula 3)

[0056] Among them, Q(s t ,a t ) represents the Q value of the state vector selecting action at the tth time step, V(s t+1 ) represents the cumulative reward value starting from the state vector at the t+1th time step;

[0057] For each state vector s t , select the action a with the largest Q value t As the optimal path action for the current time step;

[0058] By using a deep reinforcement learning model for policy learning, the state vector s is constructed t and action space a t , and by generating a strategy value function and an adaptive immediate reward function, the quadruped robot can obtain the corresponding reward at each time step.

[0059] A deep reinforcement learning model is used to plan the inspection path of the quadruped robot. By constructing a state vector, action space, adaptive immediate reward function, strategy value function and deep Q network, efficient path planning and real-time obstacle avoidance in complex environments are achieved. The state vector is used as input, the Q value of each action is output, and the Q value is updated through the Q-learning update formula. At each time step, the action with the largest Q value is selected as the current optimal path action, and the robot position is updated to generate a new state vector, which is input into the deep Q network again for the next path planning.

[0060] Furthermore, when the inspection path is optimized in real time by using MPC, the following steps are included:

[0061] Among them, MPC is an advanced control method that is widely used in the control of dynamic systems, especially for control tasks with constraints and complex dynamic environments. Its core idea is to dynamically calculate the optimal control input at each time step by predicting the future state of the system, so as to adjust the control strategy in real time during the actual operation process;

[0062] The control time domain means that only the first control input obtained by calculation is executed in each time step, and then it is rolled over to the next time step for recalculation to ensure the real-time adjustability of the path;

[0063] During the inspection process of the quadruped robot, based on the current state vector s t As initial conditions, a linear prediction model is used to predict the step size.

[0064] The optimal control output sequence for the current time step is calculated using the MPC cost function and minimizing it using gradient descent to obtain the optimal solution.

[0065] Among them, the optimal control output sequence refers to the control input set calculated by minimizing the cost function. MPC optimizes at each time step and predicts the optimal path action for multiple time steps in the future prediction step. The prediction step refers to the maximum number of time steps T that MPC predicts the quadruped robot to be pre-set during the path planning process.

[0066] Among them, the MPC cost function includes state error terms and control input terms.

[0067] The first control input calculated at the current time step is sent to the quadruped robot controller to update the position of the quadruped robot and obtain the new state vector s t+1 , and the new state vector s t+1 The deep Q network is input again to make the next path decision, and the MPC optimization is re-executed at the next time step until the target position is reached.

[0068] The MPC module is added to the step of the deep Q network. By defining the prediction step size and control time domain, MPC is based on the current state vector s t The rolling optimization control input is minimized by the cost function to achieve dynamic adjustment of the path. At each time step, MPC generates the optimal control sequence and executes the first control input to ensure that the quadruped robot can adjust the path in real time. This rolling optimization feature effectively makes up for the lack of real-time performance of deep reinforcement learning models in dynamic environments.

[0069] Combined with MPC for real-time path optimization, the concepts of prediction step size and control time domain are introduced for the path planning of the quadruped robot. By predicting and rolling adjustments to the future path, efficient response to complex dynamic environments is achieved. The application of MPC uses the current state vector as the initial condition, combined with the linear prediction model and the minimization of the cost function, to generate a control output sequence for multiple time steps in the future. The state error term of the cost function ensures that the robot gradually approaches the target position, and the control input term smoothes the control changes and avoids drastic adjustments to the path, thereby maintaining the stability and accuracy of the path. MPC executes only the first control input at each time step through rolling optimization, and then recalculates the path based on the updated state to achieve dynamic adjustment. Compared with traditional deep reinforcement learning path planning, MPC makes up for the lack of real-time performance in the deep Q network, allowing the robot to flexibly adjust the path according to real-time environmental changes, improving its adaptability and robustness in complex and uncertain environments.

[0070] S3. Calculate the failure risk probability of the inspected equipment in the inspection area according to the failure risk prediction formula and set the task priority of the inspection tasks in the inspection area.

[0071] Specifically, constructing a fault risk prediction formula to calculate the fault risk probability of the inspected equipment and setting the task priority refers to defining the environmental adaptability index at the tth time step according to the data vector. The formula is:

[0072] Et=Zt′+Ht′(Formula 4)

[0073] Among them, E t It represents the environmental fitness index of the inspection area at the tth time step, Z t ′ represents the normalized ambient temperature of the inspection area at the tth time step, H t ′ represents the normalized ambient humidity of the inspection area at the tth time step;

[0074] A fault risk prediction formula is constructed using a logistic regression function to convert the environmental adaptability and the status of the inspected equipment into a fault risk probability. The formula is:

[0075]

[0076] Among them, P t represents the failure risk probability of the inspected equipment at the tth time step, I t ′,A t ′,F t ′ respectively represent the normalized current of the inspected equipment, the temperature of the inspected equipment and the vibration amplitude of the inspected equipment at the tth time step;

[0077] According to the failure risk probability of the inspected equipment, set the failure risk threshold R 1 and R 2 , determine the warning triggering conditions;

[0078] If P t <R 1 , it is determined that the inspected equipment is in a low fault risk range, and there is no need to trigger an early warning. The inspected equipment is in normal status;

[0079] If R 1 ≤P t <R 2 , then the inspected equipment is judged to be in the medium fault risk range, triggering a low-level warning, and the quadruped robot's task priority for handling the inspected equipment is low;

[0080] If P t ≥R 2 , it is determined that the inspected equipment is in a high-risk range, triggering a high-level warning, and the quadruped robot has a high priority in handling the task of inspecting the equipment.

[0081] By constructing a fault risk prediction formula, the environmental adaptability and the status of the inspected equipment are converted into fault risk probability, and the task priority is set according to the probability, thus realizing the hierarchical management of the fault risk of the inspected equipment. By setting different fault risk thresholds, the system can distinguish between low-, medium-, and high-risk inspected equipment, trigger the corresponding warning level and adjust the task priority of the quadruped robot. In this way, the system can maintain a normal state when the risk is low, issue a warning prompt when the risk is medium, and prioritize the processing tasks when the risk is high, thereby effectively improving the responsiveness of the inspection system and the safety of the inspected equipment.

[0082] S4. Construct a task allocation and dynamic scheduling mechanism for quadruped robots.

[0083] Specifically, constructing a quadruped robot task allocation and dynamic scheduling mechanism refers to defining a quadruped robot node set by integrating the state vectors of all quadruped robots at the current time step into a grid graph G, with each quadruped robot serving as a node in the grid graph.

[0084] Based on the communication distance between each quadruped robot, connecting edges are defined for other quadruped robots that can communicate directly.

[0085] Define the target state table, initialize all pending tasks to the "unassigned" state, and set task priorities based on the failure risk probability of the inspected equipment.

[0086] Among them, the target state table includes the target position, the number of the assigned quadruped robot, the task status, the task priority and the task timestamp.

[0087] Configure the real-time update mechanism of the target status table so that the task status and allocation information can be adjusted dynamically.

[0088] When each quadruped robot receives a new task request, it can query the target status table to confirm the task status of the task. If the target position already has a number assigned to the quadruped robot and the task status is "in progress", the task is determined to be repeated and the quadruped robot waits to receive a new task. Otherwise, the task starts.

[0089] After determining that the task is repeated, it can be reallocated according to the task priority recorded in the target state table. In addition, if the priority of the repeated task is high, the quadruped robot closest to the task location will be kept to continue to perform the task. If the priority of the repeated task is low, the current task allocation will be canceled and wait for other quadruped robots to be allocated.

[0090] By constructing a task allocation and dynamic scheduling mechanism for quadruped robots, the states of all robots are integrated into a grid graph, and efficient communication and collaboration are achieved using nodes and connecting edges. By defining the target state table and setting task priorities, the system can update task status and allocation information in real time to ensure that high-risk tasks receive priority response. This mechanism effectively avoids task conflicts and realizes intelligent allocation and dynamic adjustment of tasks, thereby improving the flexibility and efficiency of the quadruped robot inspection system.

[0091] Example 2

[0092] Reference Figure 3 , which is a second embodiment of the present invention, is different from the previous embodiment and provides a quadruped robot inspection system, comprising:

[0093] A data acquisition module is used to collect data of the inspection area where the quadruped robot is located in real time, and integrate the data into a data vector for pre-processing;

[0094] A path planning module, used to plan the inspection path of the quadruped robot using a deep reinforcement learning model, and to optimize the inspection path in real time using MPC;

[0095] A fault prediction module, used to calculate the fault risk probability of the inspected equipment in the inspection area according to the fault risk prediction formula and set the task priority of the inspection task in the inspection area;

[0096] The task allocation module is used to build the task allocation and dynamic scheduling mechanism of the quadruped robot.

[0097] Reference Figure 4 , which is the third embodiment of the present invention, provides another inspection system of a quadruped robot, comprising:

[0098] A data acquisition module is used to collect data of the inspection area where the quadruped robot is located in real time, and integrate the data into a data vector for pre-processing;

[0099] A path planning module, used to plan the inspection path of the quadruped robot using a deep reinforcement learning model, and to optimize the inspection path in real time using MPC;

[0100] A fault prediction module, used to calculate the fault risk probability of the inspected equipment in the inspection area according to the fault risk prediction formula and set the task priority of the inspection task in the inspection area;

[0101] The task allocation module is used to build the task allocation and dynamic scheduling mechanism of the quadruped robot.

[0102] In an optional implementation, the data acquisition module includes:

[0103] A status data acquisition submodule, used to collect status data of the inspected device in real time using sensors pre-installed on the inspected device, wherein the sensors collect data at a preset acquisition frequency, and the status data includes the current of the inspected device, the temperature of the inspected device, and the vibration amplitude of the inspected device;

[0104] The environmental data collection submodule is used to collect environmental data in the inspection area in real time using sensors in the inspection area, and the environmental data includes environmental temperature and environmental humidity.

[0105] In an optional embodiment, the preprocessing in the data acquisition module includes using a statistical-based anomaly detection algorithm to identify and eliminate noise and erroneous data in the data vector, using linear interpolation to complete the missing data in the data vector, and standardizing and normalizing the completed data.

[0106] In an optional implementation, the path planning module uses a deep reinforcement learning model to plan the inspection path of the quadruped robot, specifically for:

[0107] Acquire the position of the quadruped robot and the position of the nearest obstacle to the quadruped robot in real time through GPS;

[0108] Integrate the position of the quadruped robot and the position of the nearest obstacle to the quadruped robot into a state vector, wherein the state vector includes the position of the quadruped robot and the position of the nearest obstacle to the quadruped robot at different time steps corresponding to each action;

[0109] According to the moving operations taken by the quadruped robot in three-dimensional space, the action space of the quadruped robot during the inspection process is defined;

[0110] Define the target position of the quadruped robot as the precise position coordinates of the inspected equipment;

[0111] Use the deep reinforcement learning model for strategy learning, build an adaptive instant reward function, and calculate the instant reward value of the state vector at different time steps. The expression is:

[0112]

[0113] Among them, st represents the state vector at the tth time step, R(s t ) represents the instantaneous reward value obtained by the state vector st at the t-th time step, x t and x g They represent the x-coordinate and y-coordinate of the quadruped robot at the tth time step and the target position of the quadruped robot respectively. t and g They represent the y coordinates of the quadruped robot at the tth time step and the z coordinates of the target position of the quadruped robot. t and z g denote the z coordinates of the quadruped robot at the tth time step and the target position of the quadruped robot, α denotes the penalty coefficient of the obstacle, and d t,o represents the distance between the quadruped robot and the nearest obstacle at the tth time step;

[0114] Based on the adaptive instant reward function and discount factor, the policy value function is constructed to calculate the cumulative reward value of the state vector at different time steps. The formula is:

[0115]

[0116] Among them, V(s t ) is the policy value function, calculated on the state vector s t The cumulative reward value obtained, T represents the number of steps of the maximum time step pre-set in the path planning process of the quadruped robot, and γ represents the discount factor;

[0117] Construct a deep Q network and transform the state vector s tAs input, the Q value of each action is output, and the Q value is updated at each time step using the Q-learning update formula according to the cumulative reward value obtained by the policy value function. The formula is:

[0118] Q(s t ,a t )←Q(s t ,a t )+[R(s t )+γV(s t+1 )-Q(s t ,a t )]

[0119] Among them, Q(s t ,a t ) represents the Q value of the state vector selecting action at the tth time step, V(s t+1 ) represents the cumulative reward value starting from the state vector at the t+1th time step;

[0120] For each state vector s t , select the action a with the largest Q value t As the optimal path action for the current time step.

[0121] In an optional implementation, the path planning module uses MPC to optimize the inspection path in real time, specifically for:

[0122] During the inspection process of the quadruped robot, based on the current state vector s t As an initial condition, a linear prediction model is used to predict the step length, wherein the predicted step length refers to the maximum time step number T that is pre-set in the path planning process of the quadruped robot predicted by the MPC;

[0123] Use the MPC cost function and use gradient descent minimization to obtain the optimal solution, calculate the optimal control output sequence for the current time step, the optimal control output sequence refers to the control input set calculated by minimizing the cost function, MPC is optimized at each time step, and the optimal path action for multiple time steps within the future prediction step is predicted, and the MPC cost function includes a state error term and a control input term;

[0124] The first control input calculated at the current time step is sent to the quadruped robot controller to update the position of the quadruped robot and obtain the new state vector s t+1 ;

[0125] The new state vector s t+1 The deep Q network is input again to make the next path decision, and the MPC optimization is re-executed at the next time step until the target position is reached.

[0126] In an optional implementation manner, the fault prediction module refers to an environmental fitness index defined at the tth time step according to the data vector, and the formula is:

[0127] E t =Z t ′+H t '

[0128] Among them, E t It represents the environmental fitness index of the inspection area at the tth time step, Z t ′ represents the normalized ambient temperature of the inspection area at the tth time step, H t ′ represents the normalized ambient humidity of the inspection area at the tth time step;

[0129] A fault risk prediction formula is constructed using a logistic regression function to convert the environmental adaptability and the status of the inspected equipment into a fault risk probability. The formula is:

[0130]

[0131] Among them, P t represents the failure risk probability of the inspected equipment at the tth time step, I t ′,A t ′,F t ′ respectively represent the normalized current of the inspected equipment, the temperature of the inspected equipment and the vibration amplitude of the inspected equipment at the tth time step;

[0132] According to the failure risk probability of the inspected equipment, set the failure risk threshold R 1 and R 2 , determine the warning triggering conditions;

[0133] If P t <R 1 , it is determined that the inspected equipment is in a low fault risk range, and there is no need to trigger an early warning. The inspected equipment is in normal status;

[0134] If R 1 ≤P t <R 2 , then the inspected equipment is judged to be in the medium fault risk range, triggering a low-level warning, and the quadruped robot's task priority for handling the inspected equipment is low;

[0135] If P t ≥R 2 , it is determined that the inspected equipment is in a high-risk range, triggering a high-level warning, and the quadruped robot has a high priority in handling the task of inspecting the equipment.

[0136] In an optional embodiment, the task assignment module refers to defining a set of quadruped robot nodes by integrating the state vectors of the current time step of all quadruped robots into the grid graph G, and each quadruped robot is a node in the grid graph;

[0137] According to the communication distance between each quadruped robot, a connection edge is defined for other quadruped robots that can directly communicate;

[0138] Define the target state table, initialize all pending tasks to the unassigned state, and set task priorities based on the failure risk probability of the inspected equipment;

[0139] When each quadruped robot receives a new task request, it queries the target status table to confirm the task status of the task;

[0140] If the number assigned to the quadruped robot already exists at the target location and the task status is in progress, the task is determined to be repeated and the quadruped robot waits to receive a new task;

[0141] If the task status is not in progress, the quadruped robot starts the task;

[0142] After determining that the task is repeated, the method further includes:

[0143] reallocate tasks of the quadruped robot according to the task priorities recorded in the target state table;

[0144] If the priority of a repeated task is high, the quadruped robot closest to the task location will be kept to continue executing the task;

[0145] If the priority of the repeated task is low, the current task assignment is canceled and waits for other quadruped robots to assign it.

[0146] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0147] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0148] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0149] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0150] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A quadruped robot inspection method, characterized in that: include, Collect data of the inspection area where the quadruped robot is located in real time, and integrate the data into data vectors for preprocessing; Using a deep reinforcement learning model to plan the inspection path of the quadruped robot, and using MPC to optimize the inspection path in real time; Calculate the failure risk probability of the inspected equipment in the inspection area according to the failure risk prediction formula and set the task priority of the inspection task in the inspection area; Construct a task allocation and dynamic scheduling mechanism for quadruped robots.

2. The inspection method of a quadruped robot according to claim 1, characterized in that: Real-time data collection of the inspection area where the quadruped robot is located, including: Using sensors pre-installed on the inspected equipment to collect status data of the inspected equipment in real time, wherein the sensors collect data at a preset collection frequency, and the status data includes the current of the inspected equipment, the temperature of the inspected equipment, and the vibration amplitude of the inspected equipment; The environmental data in the inspection area are collected in real time using sensors in the inspection area, and the environmental data include environmental temperature and environmental humidity.

3. The inspection method of a quadruped robot as claimed in claim 2, characterized in that: The preprocessing includes using a statistically based anomaly detection algorithm to identify and remove noise and erroneous data in the data vector, using a linear interpolation method to complete the missing data in the data vector, and standardizing and normalizing the completed data.

4. The inspection method of a quadruped robot as claimed in claim 3, characterized in that: The inspection path of the quadruped robot is planned using a deep reinforcement learning model, including: Acquire the position of the quadruped robot and the position of the nearest obstacle to the quadruped robot in real time through GPS; Integrate the position of the quadruped robot and the position of the nearest obstacle to the quadruped robot into a state vector, wherein the state vector includes the position of the quadruped robot and the position of the nearest obstacle to the quadruped robot at different time steps corresponding to each action; According to the moving operations taken by the quadruped robot in three-dimensional space, the action space of the quadruped robot during the inspection process is defined; Define the target position of the quadruped robot as the precise position coordinates of the inspected equipment; Use the deep reinforcement learning model for strategy learning, build an adaptive instant reward function, and calculate the instant reward value of the state vector at different time steps. The expression is: Among them, st represents the state vector at the tth time step, R(s t ) represents the instantaneous reward value obtained by the state vector st at the t-th time step, x t and x g They represent the x-coordinate and y-coordinate of the quadruped robot at the tth time step and the target position of the quadruped robot respectively. t and g They represent the y coordinates of the quadruped robot at the tth time step and the target position of the quadruped robot, respectively. t and z g denote the z coordinates of the quadruped robot at the tth time step and the target position of the quadruped robot, α denotes the penalty coefficient of the obstacle, and d t,o represents the distance between the quadruped robot and the nearest obstacle at the tth time step; Based on the adaptive instant reward function and discount factor, the policy value function is constructed to calculate the cumulative reward value of the state vector at different time steps. The formula is: Among them, V(s t ) is the policy value function, calculated on the state vector s t The cumulative reward value obtained, T represents the number of steps of the maximum time step pre-set in the path planning process of the quadruped robot, and γ represents the discount factor; Construct a deep Q network and transform the state vector s t As input, the Q value of each action is output, and the Q value is updated at each time step using the Q-learning update formula according to the cumulative reward value obtained by the policy value function. The formula is: Q(s t ,a t )←Q(s t ,a t )+[R(s t )+γV(s t+1 )-Q(s t ,a t )] Among them, Q(s t ,a t ) represents the Q value of the state vector selecting action at the tth time step, V(s t+1 ) represents the cumulative reward value starting from the state vector at the t+1th time step; For each state vector s t , select the action a with the largest Q value t As the optimal path action for the current time step.

5. The inspection method of a quadruped robot as claimed in claim 4, characterized in that: The inspection path is optimized in real time using MPC, including: During the inspection process of the quadruped robot, based on the current state vector s t As an initial condition, a linear prediction model is used to predict the step length, wherein the predicted step length refers to the number of steps T of the maximum time step pre-set in the path planning process predicted by MPC for the quadruped robot; Use the MPC cost function and use gradient descent minimization to obtain the optimal solution, calculate the optimal control output sequence for the current time step, the optimal control output sequence refers to the control input set calculated by minimizing the cost function, MPC is optimized at each time step, and the optimal path action for multiple time steps within the future prediction step is predicted, and the MPC cost function includes a state error term and a control input term; The first control input calculated at the current time step is sent to the quadruped robot controller to update the position of the quadruped robot and obtain the new state vector s t+1 ; The new state vector s t+1 The deep Q network is input again to make the next path decision, and the MPC optimization is re-executed at the next time step until the target position is reached.

6. The inspection method of a quadruped robot as claimed in claim 5, characterized in that: Calculating the failure risk probability of the inspected equipment in the inspection area according to the failure risk prediction formula and setting the task priority of the inspection task in the inspection area refers to defining the environmental adaptability index at the tth time step according to the data vector, and the formula is: E t =Z t ′+H t ′ Among them, E t It represents the environmental fitness index of the inspection area at the tth time step, Z t ′ represents the normalized ambient temperature of the inspection area at the tth time step, H t ′ represents the normalized ambient humidity of the inspection area at the tth time step; A fault risk prediction formula is constructed using a logistic regression function to convert the environmental adaptability and the status of the inspected equipment into a fault risk probability. The formula is: Among them, P t represents the failure risk probability of the inspected equipment at the tth time step, I t ′,A t ′,F t ′ respectively represent the normalized current of the inspected equipment, the temperature of the inspected equipment and the vibration amplitude of the inspected equipment at the tth time step; According to the failure risk probability of the inspected equipment, set the failure risk thresholds R1 and R2 to determine the early warning triggering conditions; If P t < is less than R1, it is determined that the equipment under inspection is in the low fault risk range, and there is no need to trigger an early warning, and the status of the equipment under inspection is normal; If R1 ≤ P t < and R2, it is determined that the device under inspection is in the medium fault risk range, triggering a low-level warning, and the task priority of the quadruped robot to process the device under inspection is low; If P t ≥R2, the inspected equipment is judged to be in a high-risk range, triggering a high-level warning, and the quadruped robot has a high priority in handling the task of inspected equipment.

7. The inspection method of a quadruped robot as claimed in claim 6, characterized in that: The construction of the quadruped robot task allocation and dynamic scheduling mechanism refers to defining a set of quadruped robot nodes by integrating the state vectors of the current time step of all quadruped robots into the grid graph G, with each quadruped robot serving as a node in the grid graph; According to the communication distance between each quadruped robot, a connection edge is defined for other quadruped robots that can directly communicate; Define the target state table, initialize all pending tasks to the unassigned state, and set task priorities based on the failure risk probability of the inspected equipment; When each quadruped robot receives a new task request, it queries the target status table to confirm the task status of the task; If the number assigned to the quadruped robot already exists at the target location and the task status is in progress, the task is determined to be repeated and the quadruped robot waits to receive a new task; If the task status is not in progress, the quadruped robot starts the task; After determining that the task is repeated, the method further includes: reallocate tasks of the quadruped robot according to the task priorities recorded in the target state table; If the priority of a repeated task is high, the quadruped robot closest to the task location will be kept to continue executing the task; If the priority of the repeated task is low, the current task assignment is canceled and waits for other quadruped robots to assign it.

8. A quadruped robot inspection system based on the quadruped robot inspection method according to any one of claims 1 to 7, characterized in that: include: A data acquisition module is used to collect data of the inspection area where the quadruped robot is located in real time, and integrate the data into a data vector for pre-processing; A path planning module, used to plan the inspection path of the quadruped robot using a deep reinforcement learning model, and to optimize the inspection path in real time using MPC; A fault prediction module, used to calculate the fault risk probability of the inspected equipment in the inspection area according to the fault risk prediction formula and set the task priority of the inspection task in the inspection area; The task allocation module is used to build the task allocation and dynamic scheduling mechanism of the quadruped robot.

9. A computer inspection device, comprising: Memory and processor; The memory stores a computer program, wherein the processor implements the steps of the inspection method of the quadruped robot according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the inspection method of a quadruped robot according to any one of claims 1 to 7 are implemented.

Images