ROS-based data center equipment inspection robot

Abstract

The invention discloses a data center equipment inspection robot based on an ROS, and relates to the technical field of data monitoring. The robot comprises a high-definition camera, a laser radar, an encoder, Mecanum wheels, a direct-current motor, a bottom-layer driving plate and an industrial personal computer configured with an ROS development environment, and through the communication connection relation and mutual cooperation of the high-definition camera, the laser radar, the encoder, the Mecanum wheels, the direct-current motor, the bottom-layer driving plate and the industrial personal computer, an environment map can be firstly built according to received point cloud data in the full-coverage moving process; the method comprises the following steps of: finding inspection stop nodes of a data center device, adding the found inspection stop nodes into a map, planning an inspection path by applying a Q-Learning algorithm to obtain a recommended route which is shortest in total distance and sequentially passes through all the inspection stop nodes, and finally driving a robot to move according to the recommended route and stop at the inspection stop nodes to obtain the indicator light inspection data of the data center device. Therefore, the time and the cost of manual inspection can be reduced, the working efficiency is improved, and the purposes of reducing the error rate and quickly responding can be achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of data monitoring, and particularly relates to a data center equipment inspection robot based on ROS. Background Art

[0002] With the rapid development of technologies such as big data, cloud computing, and the Internet of Things, data centers have become important infrastructure to support modern digital society. A large number of high-value devices and systems are integrated in the data center, such as servers, storage devices, power systems, and network facilities. The stability of these devices directly affects the continuity and security of business. Therefore, the operation and maintenance (referred to as operation and maintenance) work of the data center is crucial.

[0003] At present, although the traditional manual inspection method for data center equipment was effective in the early stage, with the increase in the types and quantities of equipment, manual inspection gradually exposes problems such as low efficiency, high error rate, and slow response, and cannot meet the requirements of modern data centers for high availability, high reliability, and high efficiency. Summary of the Invention

[0004] The purpose of the present invention is to provide a data center equipment inspection robot based on ROS to solve the problems of low efficiency, high error rate, and slow response existing in the existing manual inspection method for data center equipment.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions: The present invention provides a data center equipment inspection robot based on ROS, including a high-definition camera, a lidar, an encoder, Mecanum wheels, a DC motor, a bottom drive board, and an industrial computer configured with a ROS development environment. Among them, the high-definition camera is used to collect and transmit video images of the environment around the robot to the industrial computer in real time, the lidar is used to collect and transmit point cloud data of the environment around the robot to the industrial computer in real time, the encoder is used to collect and transmit displacement speed information during the walking process of the robot to the industrial computer in real time, and the bottom drive board is used to drive the DC motor under the control of the industrial computer to make the Mecanum wheels and the entire inspection robot move; The industrial computer is used to perform the following steps: Apply the keyboard control module in the ROS development environment to move the data center equipment layout site in a full-coverage manner under control; During the full-coverage movement, according to the received point cloud data, apply the RBPF-SLAM algorithm to construct an environmental map, and perform real-time data center equipment recognition processing on the received video images based on the target detection algorithm. When a data center equipment is recognized, add the position where the robot is located as an inspection stop node to the environmental map; After the full-coverage movement ends, starting from the current position of the robot, the Q-Learning algorithm is applied in the ROS development environment for path planning to obtain a recommended route with the shortest total distance that sequentially passes through all the inspection stop nodes in the environmental map; According to the received displacement speed information, the PID algorithm is applied to control the underlying drive board in real time, so as to drive the robot to move along the recommended route through the DC motor and Mecanum wheels and stop at the inspection stop nodes; During the stop interval, the current indicator light picture of the data center device is intercepted from the received video image, and then computer vision technology is applied to identify the number and color of the indicator lights in the current indicator light picture to obtain the inspection data of the data center device.

[0006] Based on the above invention content, a new robot solution for autonomous inspection of data center devices based on the ROS development environment is provided, including a high-definition camera, a lidar, an encoder, Mecanum wheels, a DC motor, an underlying drive board, and an industrial computer configured with the ROS development environment. Through their communication connection relationships and mutual cooperation, during the full-coverage movement process, the environmental map can be constructed according to the received point cloud data, and the discovered inspection stop nodes can be added to the environmental map. Then, the Q-Learning algorithm is applied for inspection path planning to obtain a recommended route with the shortest total distance that sequentially passes through all the inspection stop nodes in the environmental map. Finally, the robot is driven to move along the recommended route and stop at the inspection stop nodes to obtain the indicator light inspection data of the data center device. In this way, not only can the inspection task be automated, reducing the time and cost of manual inspection and improving work efficiency, but also the shortest inspection path can be autonomously planned to quickly and cyclically monitor the device operation status, thereby facilitating the timely discovery of potential faulty devices, early warning, avoiding large-scale failures caused by neglecting small faults in the devices, achieving the purpose of reducing the error rate and fast response, and being convenient for practical application and promotion.

[0007] In a possible design, starting from the current position of the robot, the Q-Learning algorithm is applied in the ROS development environment for path planning to obtain a recommended route with the shortest total distance that sequentially passes through all the inspection stop nodes in the environmental map, including: Taking the current position of the robot as the starting point, and aggregating this starting point and all the inspection stop nodes in the environmental map to obtain a node set; For each pair of nodes in the node set, with one corresponding node as the starting point and the other corresponding node as the ending point, the Q-Learning algorithm is applied in the ROS development environment for path planning between the corresponding nodes to obtain the corresponding shortest path; According to the shortest paths of each pair of nodes, a recommended route with the shortest total distance and passing through all the inspection stop nodes in the environmental map in sequence is spliced.

[0008] In a possible design, for each pair of nodes in the node set, with one corresponding node as the starting point and the other corresponding node as the ending point, the Q-Learning algorithm is applied in the ROS development environment for path planning between the corresponding nodes to obtain the corresponding shortest path, including: For a certain pair of nodes in the node set, with one corresponding node as the starting point and the other corresponding node as the ending point, the Q-Learning algorithm is applied in the ROS development environment for path planning between the corresponding nodes to obtain the corresponding globally feasible path; The beetle antennae search algorithm is applied to optimize the globally feasible path to obtain the shortest path of the certain pair of nodes.

[0009] In a possible design, the Q-Learning algorithm adopts any one or any combination of the following improvement points (A) to (C): (A) An RBF neural network is used to obtain the estimated result of the Q-value function in the Q-Learning algorithm. Among them, the RBF neural network includes an input layer, a hidden layer, and an output layer. The input layer includes input nodes, the hidden layer includes hidden nodes, the output layer includes one output node, and the input nodes are used to input the -dimensional finite state variable and one-dimensional action variable in the Q-Learning algorithm one by one. represents a positive integer greater than or equal to 2. represents a positive integer and has , and each of the hidden nodes in the hidden layer is a -dimensional Gaussian function and the output expression of the -th hidden node is , represents a positive integer greater than or equal to 2. represents a positive integer less than or equal to . represents a positive integer less than or equal to . represents the natural exponential function. represents the input value of the -th input node among the input nodes. represents the -th radial basis function at the the central value in the dimension indicating the th radial basis function in the dimension, and the output expression of the output node is , indicating the weight value between the th hidden node and the output node; (B) The greedy strategy in the Q-Learning algorithm is executed by dynamically adjusting the greedy factor, where the dynamically adjusted greedy factor is expressed as follows:

[0010] In the formula, represents the maximum value function, represents the preset maximum value of the greedy factor, represents the preset minimum value of the greedy factor, represents the decay amount of the greedy factor in each training round, represents the current training round number; (C) The action set in the Q-Learning algorithm includes forward action, backward action, left action, right action, left front action, right front action, left rear action, and right rear action.

[0011] In a possible design, the beetle antenna search algorithm is applied to optimize the path of the global feasible path to obtain the shortest path between a pair of nodes, including the following steps S3221 to S3229: S3221. Initialize the required parameters of the beetle antenna search algorithm, and then execute step S3222, where the required parameters include the number of beetle individuals , the initial value of the beetle movement step size, the initial value of the beetle antenna sensing length, and the maximum update times; S3222. Initialize the integer variable to 1, and then execute step S3223; S3223. Place beetles at the th non-starting and ending points on the global feasible path and along the direction from the starting point to the ending point, and use the th non-starting and ending points on the global feasible path and along the direction from the starting point to the ending point as the target position, and then execute step S3224; S3224. Create random vectors corresponding one-to-one to the beetles, and calculate according to the target position in the The fitness values of the left and right antennae of each longhorn beetle among the only longhorn beetles, and then step S3225 is executed; S3225. For each longhorn beetle among the only longhorn beetles, update the corresponding position according to the fitness values of the corresponding left and right antennae, and calculate the fitness value of the corresponding new position according to the target position, and then step S3226 is executed; S3226. Determine the best longhorn beetle according to the fitness values of the new positions of the longhorn beetles, and use the new position of the best longhorn beetle as the best position, and then step S3227 is executed; S3227. Determine whether the best position is better than the position where the th non-start and end point is located. If so, update the position where the th non-start and end point is located to the best position, and then step S3228 is executed. Otherwise, step S3228 is executed; S3228. Let be incremented by 1, and then determine whether is less than the total number of points of the global feasible path. If so, return to execute step S3223. Otherwise, update the longhorn beetle movement step length and the longhorn beetle antenna perception length, and then execute step S3229; S3229. Determine whether the number of path point updates has reached the maximum number of updates. If so, generate the shortest path between the pair of nodes according to the start point, end point and the current positions of all non-start and end points. Otherwise, return to execute step S3222.

[0012] In a possible design, the fitness value is calculated using the following fitness function as follows:

[0013] where represents the position variable, represents the distance penalty coefficient, represents the obstacle collision penalty coefficient, represents the distance cost from the position variable to the target position, represents the obstacle penalty cost, represents the shortest distance from the obstacle in the environmental map to the path segment from the position variable to the target position, represents the preset minimum safety distance.

[0014] In a possible design, updating the longhorn beetle movement step length and the longhorn beetle antenna perception length includes: Update the longhorn beetle movement step length according to the following formula:

[0015] In the formula, represents the current search times of the longhorn beetle, represents at the th search of the longhorn beetle, the movement step size, and when there is represents the initial value of the movement step size of the longhorn beetle, represents at the th search of the longhorn beetle, the movement step size, represents the preset minimum step size, represents the preset step size attenuation coefficient; Update the antenna sensing length of the longhorn beetle according to the following formula:

[0016] In the formula, represents the antenna sensing length of the longhorn beetle at the th search, represents the preset antenna length factor.

[0017] In a possible design, according to the shortest paths of the pairs of nodes, splice to obtain a recommended route with the shortest total distance and passing through all the inspection stop nodes in the environmental map in sequence, including: Perform combinatorial permutation on the shortest paths of the pairs of nodes to obtain multiple path queues. Among them, each path queue in the multiple path queues conforms to the following rules: the starting point of the first shortest path in the path queue is the starting point, and the end point of the adjacent previous shortest path in the path queue is the same as the starting point of the adjacent subsequent shortest path, and the end points of all the shortest paths in the path queue include all the inspection stop nodes in the environmental map; For each path queue in the multiple path queues, calculate the corresponding total distance by superimposing according to the corresponding all shortest paths; Perform splicing processing on all the shortest paths of a path queue corresponding to the shortest total distance to obtain a recommended route with the shortest total distance and passing through all the inspection stop nodes in the environmental map in sequence.

[0018] In a possible design, it further includes a wireless communication module communicatively connected to the industrial control computer; The industrial control computer is further configured to, after identifying the number and color of the indicator lights in the current indicator light picture by using computer vision technology, perform the following steps: Based on the recognition result, the indicator light data and color of the data center device in the normal state, it is determined whether the data center device is abnormal. If so, an alarm message for the data center device is sent to the data center device inspection platform or the data center device inspector through the wireless communication module.

[0019] In a possible design, after the full-coverage movement is completed, the map storage package in the ROS development environment is called to save the environment map.

[0020] Beneficial effects of the above solution: (1) The present invention provides a new solution for a robot to autonomously inspect data center devices based on the ROS development environment, including a high-definition camera, a lidar, an encoder, Mecanum wheels, a DC motor, a bottom drive board, and an industrial computer configured with the ROS development environment. Through their communication connection relationships and mutual cooperation, during the full-coverage movement, an environment map can be constructed according to the received point cloud data, and the inspection stop nodes found are added to the environment map. Then, the Q-Learning algorithm is applied to plan the inspection path, and a recommended route with the shortest total distance and passing through all the inspection stop nodes in the environment map in sequence is obtained. Finally, the robot is driven to move along the recommended route and stop at the inspection stop nodes to obtain the indicator light inspection data of the data center devices. In this way, not only can the inspection task be automated, reducing the time and cost of manual inspection and improving work efficiency, but also the shortest inspection path can be autonomously planned to quickly and cyclically monitor the device operation status, which is conducive to timely discovering potential faulty devices, giving early warnings, avoiding large-scale failures caused by neglecting small faults, achieving the purpose of reducing the error rate and rapid response, and facilitating practical application and promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0022] Figure 1 It is a schematic structural diagram of a ROS-based data center device inspection robot provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the present invention in combination with the accompanying drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other embodiments can be obtained based on these embodiments. It should be noted here that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation to the present invention.

[0024] It should be understood that although terms such as first and second may be used herein to describe various objects, these objects should not be limited by these terms. These terms are only used to distinguish one object from another. For example, the first object can be called the second object, and similarly, the second object can be called the first object, without departing from the scope of the exemplary embodiments of the present invention.

[0025] It should be understood that for the term "and / or" that may appear in this document, it is merely a description of the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, B exists alone, or A and B exist simultaneously, etc.; another example, A, B, and / or C can represent any one of A, B, and C or any combination of them; for the term " / and" that may appear in this document, it is a description of another association object relationship, indicating that two relationships can exist. For example, A / and B can represent: A exists alone or A and B exist simultaneously, etc.; in addition, for the character " / " that may appear in this document, generally, it represents that the front and back associated objects are an "or" relationship.

[0026] Embodiment As Figure 1 shown, the data center equipment inspection robot provided in this embodiment and based on ROS includes, but is not limited to, a high-definition camera, a lidar, an encoder, Mecanum wheels, a DC motor, a bottom drive board, and an industrial computer configured with a ROS (Robot Operating System, a framework designed for programming robots) development environment. Among them, the high-definition camera is used to collect and transmit the video images of the robot's surrounding environment to the industrial computer in real time, the lidar is used to collect and transmit the point cloud data of the robot's surrounding environment to the industrial computer in real time, the encoder is used to collect and transmit the displacement speed information during the robot's walking process to the industrial computer in real time, and the bottom drive board is used to drive the DC motor under the control of the industrial computer to make the Mecanum wheels and the entire inspection robot move. The industrial computer is used to execute the following steps S1 to S5.

[0027] S1. Apply the keyboard control module in the ROS development environment to controllably perform full-coverage movement on the layout site of the data center equipment.

[0028] In step S1, specifically, the inspection personnel can conventionally operate the keyboard control module to control the data center equipment inspection robot to perform full-coverage movement on the layout site of the data center equipment. In addition, the layout site of the data center equipment is exemplified as a site where equipment such as servers, storage devices, power systems, and network facilities are arranged.

[0029] S2. During the full-coverage movement, according to the received point cloud data, apply the RBPF-SLAM algorithm to construct an environmental map, and perform real-time data center equipment recognition processing on the received video images based on the object detection algorithm. When a data center equipment is recognized, add the position where the robot is located as an inspection stop node to the environmental map.

[0030] In step S2, the RBPF-SLAM (Rao-Blackwellized Particle Filter SLAM) algorithm is a positioning and mapping method based on particle filtering. It uses a particle swarm to describe the likelihood of estimating the robot's pose and map. Each particle contains a possible historical trajectory of the robot and its associated map. Through continuous update and resampling processes, the particle swarm can converge to a few particles with higher weight coefficients, thereby determining the position of the robot and constructing an environmental map. Therefore, it can be applied to this step to obtain the environmental map. The environmental map is preferably a grid map. The object detection algorithm is an important branch in the field of computer vision. It aims to identify and locate target objects in images. Therefore, it can be based on the object detection algorithm to detect whether there is a data center equipment bounding box in the video image. If so, it can be determined that a data center equipment is recognized. In addition, after the full-coverage movement is completed, call the map storage package in the ROS development environment to save the environmental map.

[0031] S3. After the full-coverage movement is completed, starting from the current position of the robot, apply the Q-Learning algorithm in the ROS development environment for path planning to obtain a recommended route with the shortest total distance and passing through all the inspection stop nodes in the environmental map in sequence.

[0032] In the step S3, the Q-Learning algorithm is a reinforcement learning algorithm based on Temporal Difference (TD). Its core idea is to evaluate the expected utility of taking a certain action in a given state by learning an action value function (Q function). The Q-learning algorithm belongs to the model-free prediction algorithm, which does not require the dynamic model of the environment (i.e., transition probability and reward distribution), but learns through interaction with the environment. Its basic concepts and core ideas include the following: State: a specific situation or configuration of the environment; Action: possible behaviors that can be taken in a given state; Reward: the immediate return obtained from the environment after taking a certain action; Policy: the mapping from state to action, guiding how to select actions according to the current state; Q function (Q-value): representing the expected return of taking action a in state s. Considering that there are multiple different data center devices arranged in the data center device layout site, such as servers, storage devices, power systems, and network facilities, etc., so the number of inspection stop nodes will be multiple, and the Q-Learning algorithm is often used for path planning for a pair of starting points and ending points. In order to ensure a recommended route with the shortest total distance and passing through all inspection stop nodes in the environmental map in sequence for multiple inspection stop nodes, preferably, starting from the current position of the robot, apply the Q-Learning algorithm in the ROS development environment for path planning to obtain a recommended route with the shortest total distance and passing through all inspection stop nodes in the environmental map in sequence, including but not limited to the following steps S31 to S33.

[0033] S31. Take the current position of the robot as the starting point, and summarize this starting point and all inspection stop nodes in the environmental map to obtain a node set.

[0034] S32. For each pair of nodes in the node set, take one corresponding node as the starting point and the other corresponding node as the ending point, and apply the Q-Learning algorithm in the ROS development environment for path planning between the corresponding nodes to obtain the corresponding shortest path.

[0035] In the step S32, there are two problems with the traditional Q-learning algorithm: First, it initializes the Q-value table with equal values or random numbers, that is, it learns without prior environmental information, which makes the exploration of the robot in the initial stage too blind, resulting in a large invalid iteration space for the algorithm and a long path planning time. Second, the algorithm selects the four positions of up, down, left, and right adjacent to the current state as the possible states at the next moment, resulting in a large number of path corners, which is inconvenient for the robot to move. In order to solve the problems that the path obtained based on the Q-learning algorithm still has a large number of corners and a large cumulative turning angle, it is necessary to optimize the path obtained based on the Q-learning algorithm by combining common intelligent bionic algorithms such as genetic algorithms, particle swarm algorithms, and ant colony algorithms. Considering that the Beetle Antennae Search (BAS) algorithm has the characteristics of small computational complexity and strong search ability, further preferably, for each pair of nodes in the node set, with a corresponding node as the starting point and a corresponding another node as the ending point, apply the Q-Learning algorithm in the ROS development environment to perform path planning between the corresponding nodes to obtain the corresponding shortest path, including but not limited to the following steps S321 to S322.

[0036] S321. For a pair of nodes in the node set, with a corresponding node as the starting point and a corresponding another node as the ending point, apply the Q-Learning algorithm in the ROS development environment to perform path planning between the corresponding nodes to obtain the corresponding global feasible path.

[0037] In the step S321, in order to solve the problems of slow convergence speed, low exploration and exploitation efficiency, and poor planned path when the Q-Learning algorithm performs path planning tasks, it is necessary to improve the Q-Learning algorithm from aspects such as the initial value of the Q-value table, the robot exploration mechanism, and the exploration step size. That is, in detail and preferably, the Q-Learning algorithm adopts any one or any combination of the following improvement points (A) to (C).

[0038] (A) Use an RBF neural network to obtain the Q-value function estimation result in the Q-Learning algorithm. Among them, the RBF neural network includes an input layer, a hidden layer, and an output layer. The input layer includes input nodes, the hidden layer includes hidden nodes, the output layer includes one output node, and the input nodes are used to input the -dimensional finite state variable and one-dimensional action variable in the Q-Learning algorithm one by one. represents a positive integer greater than or equal to 2, represents a positive integer and has , where each of the hidden nodes among the hidden nodes is a dimensional Gaussian function, and the output expression of the th hidden node is represents a positive integer greater than or equal to 2, represents a positive integer less than or equal to , represents a positive integer less than or equal to , represents the natural exponential function, represents the input value of the th input node among the input nodes, represents the th radial basis function's center value in the th dimension, represents the width value of the th radial basis function in the th dimension. The output expression of the output node is , represents the weight value between the th hidden node and the output node. Considering in path planning, most of the robot's states are continuous, the dimension of the state space is too high, it is difficult to use the traditional table representation, and the Q-Learning algorithm will have the "curse of dimensionality" problem. Also, considering that the RBF neural network has many advantages, including simple structure, easy training, and fast convergence speed, etc., its core ability lies in being able to approximate any non-linear function. Therefore, this improvement point (A) is to use the local approximation ability of the RBF neural network to approximate the Q-value function in the Q-Learning algorithm, which can effectively handle the high-dimensional state space and obtain accurate Q-value function estimation results, thus improving the algorithm performance.

[0039] (B) Adopt dynamic adjustment of the greedy factor to execute the greedy strategy in the Q-Learning algorithm, where the dynamic adjustment of the greedy factor is expressed as follows:

[0040] In the formula, represents the maximum value function, represents the preset maximum value of the greedy factor, represents the preset minimum value of the greedy factor, represents the decay amount of the greedy factor in each training round, Indicates the current training round. By dynamically adjusting the greedy factor as described above value, the proportion of exploration can be gradually reduced and the proportion of exploitation increased during the training process. The improved strategy described above will conduct more exploration in the initial stage to better understand the environment and obtain more information; while in the subsequent stage, it will make more use of the known information as prior knowledge to obtain higher rewards.

[0041] (C)The action set in the Q-Learning algorithm includes forward action, backward action, left action, right action, left-front action, right-front action, left-back action, and right-back action. By increasing the exploration in the diagonal direction as described above, the exploration efficiency of the algorithm can be further improved.

[0042] S322. Apply the beetle antenna search algorithm to optimize the path of the global feasible path to obtain the shortest path between the pair of nodes.

[0043] In the step S322, the beetle antenna search algorithm is an efficient intelligent bionic algorithm, which was proposed by Jiang et al. in 2017 inspired by the foraging behavior of beetles. When foraging, beetles use two antennae to sense the food odor in the air. Since the distances between the left and right antennae and the food are different, the intensities of the odors sensed by the two antennae also differ, and the antenna closer to the food senses a stronger food odor. The beetle adjusts its movement direction based on this difference and always moves towards the side with a stronger food odor. The BAS algorithm is a single-agent search algorithm that only requires one individual to achieve efficient optimization. Considering that the original BAS algorithm uses one beetle to search for the optimal solution and directly updates the position of the beetle after calculating the fitness values of the left and right antennae of the beetle, its search and update process is relatively blind and random, and it is prone to ineffective iterative updates or cause the beetle to get stuck in a dead end and unable to reach the target position in a complex environment. To enable the BAS algorithm to search and update positions more effectively, specifically, applying the beetle antenna search algorithm to optimize the path of the global feasible path to obtain the shortest path between the pair of nodes includes, but is not limited to, the following steps S3221 to S3229.

[0044] S3221. Initialize the required parameters of the beetle antenna search algorithm, and then execute step S3222, where the required parameters include, but are not limited to, the number of beetle individuals 、the initial value of the beetle movement step size, the initial value of the beetle antenna sensing length, and the maximum number of updates, etc.

[0045] S3222. Initialize the integer variable to 1, and then execute step S3223.

[0046] S3223. On the global feasible path and in the direction from the starting point to the ending point at the A non-starting and ending point placement stag beetles, and use the th non-starting and ending point on the global feasible path and along the direction from the starting point to the ending point as the target position, and then execute step S3224.

[0047] S3224. Create random vectors corresponding one-to-one to the stag beetles, and calculate the fitness values of the left and right antennae of each stag beetle among the stag beetles according to the target position, and then execute step S3225.

[0048] In step S3224, considering that in this embodiment, the BAS algorithm is used to optimize the path corners in a two-dimensional environment with obstacles, it is necessary to design the fitness function for the following two constraint conditions: one is to find the shortest path to the target position, and the other is that the path cannot touch or cross the obstacles. Specifically, the fitness value is calculated using the following fitness function as follows:

[0049] In the formula, represents the position variable, represents the distance penalty coefficient, represents the obstacle collision penalty coefficient, represents the distance cost from the position variable to the target position, represents the obstacle penalty cost, represents the shortest distance from the obstacle in the environmental map to the path segment from the position variable to the target position, represents the preset minimum safety distance.

[0050] S3225. For each stag beetle among the stag beetles, update the corresponding position according to the fitness values of the corresponding left and right antennae, and calculate the fitness value of the corresponding new position according to the target position, and then execute step S3226.

[0051] In step S3225, the fitness function used to calculate the fitness value of the new position is the same as that of the left and right antennae, which will not be elaborated here.

[0052] S3226. Determine the best stag beetle according to the fitness values of the new positions of the stag beetles, and use the new position of the best stag beetle as the best position, and then execute step S3227.

[0053] S3227. Determine whether the optimal position is better than the position of the th non-starting and ending point. If so, update the position of the th non-starting and ending point to the optimal position, then execute step S3228; otherwise, execute step S3228.

[0054] S3228. Increment by 1, and then determine whether is less than the total number of points of the global feasible path. If so, return to execute step S3223; otherwise, update the movement step length of the longhorn beetle and the sensing length of the longhorn beetle's antennae, and then execute step S3229.

[0055] In step S3228, considering that the BAS algorithm needs to have a strong global search ability in the early stage to quickly approach the target position, and a strong local search ability in the later stage to accurately find the target position, this embodiment can adopt a decreasing strategy for the movement step length to adapt to this requirement of the algorithm, so as to improve the search efficiency of the algorithm while ensuring the search accuracy of the algorithm. That is, preferably, updating the movement step length of the longhorn beetle and the sensing length of the longhorn beetle's antennae includes, but is not limited to: first, update the movement step length of the longhorn beetle according to the following formula:

[0056] In the formula, represents the current search times of the longhorn beetle, represents the movement step length of the longhorn beetle at the th search, and when there is represents the initial value of the movement step length of the longhorn beetle, represents the movement step length of the longhorn beetle at the th search, represents the preset minimum value of the step length, represents the preset step length attenuation coefficient; then update the sensing length of the longhorn beetle's antennae according to the following formula:

[0057] In the formula, represents the sensing length of the antennae of the longhorn beetle at the th search, represents the preset antenna length factor.

[0058] S3229. Determine whether the number of path point updates has reached the maximum number of updates. If so, generate the shortest path between the pair of nodes according to the starting point, the ending point, and the current positions of all non-starting and ending points; otherwise, return to execute step S3222.

[0059] S33. Based on the shortest paths of the pairs of nodes, a recommended route with the shortest total distance and passing through all the inspection stop nodes in the environment map is obtained.

[0060] In step S33, in order to quickly obtain the recommended route, preferably, based on the shortest paths of each pair of nodes, a recommended route with the shortest total distance and passing through all inspection stop nodes in the environmental map in sequence is spliced, including but not limited to the following steps S331 to S333.

[0061] S331. The shortest paths of the pairs of nodes are combined and arranged to obtain a plurality of path queues, wherein each of the plurality of path queues complies with the following rules: the starting point of the first shortest path in the path queue is the starting point, and the end point of the adjacent preceding shortest path in the path queue is the same as the starting point of the adjacent succeeding shortest path, and the end points of all the shortest paths in the path queue include all the inspection stop nodes in the environmental map.

[0062] S332. For each path queue in the plurality of path queues, a corresponding total distance is obtained by superimposing and calculating all corresponding shortest paths.

[0063] S333. All the shortest paths of a certain path queue corresponding to the shortest total distance are concatenated to obtain a recommended route with the shortest total distance and passing through all the inspection stop nodes in the environment map in sequence.

[0064] S4. Based on the received displacement speed information, the PID algorithm is applied to control the bottom driving board in real time, so as to drive the robot to move along the recommended route through the DC motor and the Mecanum wheel, and stop at the inspection stop node.

[0065] In step S4, the PID algorithm is a closed-loop control algorithm based on proportional, integral, and derivative, which is used to eliminate system errors and achieve stable control. It adjusts the output in real time so that the controlled object can quickly and accurately reach the set value. It is widely used in industrial control, robotics, aerospace and other fields, so it can be conventionally modified to apply to this step.

[0066] S5. In the dwelling interval, a current indicator light image of the data center equipment is captured from the received video image, and then the number and color of the indicator lights in the current indicator light image are identified using computer vision technology to obtain the inspection data of the data center equipment.

[0067] In the step S5, the bounding box of the indicator light area of the data center device can also be detected in the received video image based on the existing object detection algorithm first, and then the image of the bounding box of the indicator light area is cropped as the current indicator light picture of the data center device. The computer vision technology for identifying the number and color of the indicator lights can be obtained by routine modification based on the existing technical means, which will not be elaborated here.

[0068] In addition, in order to achieve the purpose of triggering the device anomaly alarm in a timely manner, preferably, the data center device inspection robot further includes a wireless communication module communicatively connected to the industrial control computer; the industrial control computer is further configured to, after identifying the number and color of the indicator lights in the current indicator light picture by applying the computer vision technology, perform the following steps: judge whether the data center device is abnormal according to the identification result and the indicator light data and color of the data center device in the normal state, and if so, send an alarm message for the data center device to the data center device inspection platform or the data center device inspection personnel through the wireless communication module.

[0069] In summary, the ROS-based data center device inspection robot provided in this embodiment has the following technical effects: (1) This embodiment provides a new solution for a robot to autonomously inspect data center devices based on the ROS development environment, including a high-definition camera, a lidar, an encoder, Mecanum wheels, a DC motor, a bottom drive board, and an industrial control computer configured with the ROS development environment. Through their communication connection relationships and mutual cooperation, an environmental map can be constructed according to the received point cloud data during the full-coverage movement first, and the detected inspection stop nodes are added to the environmental map. Then, the Q-Learning algorithm is applied for inspection path planning to obtain a recommended route with the shortest total distance and passing through all the inspection stop nodes in the environmental map in sequence. Finally, the robot is driven to move along the recommended route and stop at the inspection stop nodes to obtain the indicator light inspection data of the data center devices. In this way, not only can the inspection task be automated, reducing the time and cost of manual inspection and improving work efficiency, but also the shortest inspection path can be autonomously planned to monitor the device operation status quickly in a cycle, thereby facilitating the timely discovery of potential faulty devices, early warning, avoiding large-scale failures caused by neglecting small faults, achieving the purpose of reducing the error rate and fast response, and being convenient for practical application and promotion.

[0070] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A ROS-based data center equipment inspection robot, characterized in that, It includes a high-definition camera, a lidar, an encoder, Mecanum wheels, DC motors, a bottom drive board, and an industrial computer configured with a ROS development environment. Among them, the high-definition camera is used to collect and transmit the video images of the robot's surrounding environment to the industrial computer in real time. The lidar is used to collect and transmit the point cloud data of the robot's surrounding environment to the industrial computer in real time. The encoder is used to collect and transmit the displacement speed information during the robot's movement to the industrial computer in real time. The bottom drive board is used to drive the DC motors under the control of the industrial computer to make the Mecanum wheels and the entire inspection robot move. The industrial computer is used to perform the following steps: Apply the keyboard control module in the ROS development environment to move the data center equipment layout site in a full-coverage manner under control. During the full-coverage movement, according to the received point cloud data, apply the RBPF-SLAM algorithm to construct an environmental map, and perform real-time data center equipment recognition processing on the received video images based on the target detection algorithm. When a data center equipment is recognized, add the position where the robot is located as an inspection stop node to the environmental map. After the full-coverage movement ends, starting from the current position of the robot, apply the Q-Learning algorithm in the ROS development environment for path planning to obtain a recommended route with the shortest total distance and passing through all the inspection stop nodes in the environmental map in sequence. According to the received displacement speed information, apply the PID algorithm to control the bottom drive board in real time, so as to drive the robot to move along the recommended route through the DC motors and Mecanum wheels and stop at the inspection stop nodes. During the stop interval, intercept the current indicator light picture of the data center equipment from the received video images, and then apply computer vision technology to identify the number and color of the indicator lights in the current indicator light picture to obtain the inspection data of the data center equipment.

2. The data center equipment inspection robot according to claim 1, wherein, Starting from the current position of the robot, apply the Q-Learning algorithm in the ROS development environment for path planning to obtain a recommended route with the shortest total distance and passing through all the inspection stop nodes in the environmental map in sequence, including: Take the current position of the robot as the starting point, and summarize the starting point and all the inspection stop nodes in the environmental map to obtain a node set. For each pair of nodes in the node set, with one corresponding node as the starting point and the other corresponding node as the ending point, apply the Q-Learning algorithm in the ROS development environment for path planning between the corresponding nodes to obtain the corresponding shortest path. According to the shortest paths of the pairs of nodes, splice them to obtain a recommended route with the shortest total distance and passing through all the inspection stop nodes in the environmental map in sequence.

3. The data center equipment inspection robot according to claim 2, characterized in that, For each pair of nodes in the node set, with one corresponding node as the starting point and the other corresponding node as the ending point, apply the Q-Learning algorithm in the ROS development environment for path planning between the corresponding nodes to obtain the corresponding shortest path, including: For a pair of nodes in the node set, taking one corresponding node as the starting point and the other corresponding node as the ending point, apply the Q-Learning algorithm in the ROS development environment to perform path planning between the corresponding nodes, and obtain the corresponding globally feasible path; Apply the beetle antenna search algorithm to optimize the path of the globally feasible path to obtain the shortest path of the pair of nodes.

4. The data center equipment inspection robot according to claim 3, wherein The Q-Learning algorithm has any one or any combination of the following improvement points (A) to (C): (A) An RBF neural network is used to obtain the Q-value function estimation result in the Q-Learning algorithm. Among them, the RBF neural network includes an input layer, a hidden layer, and an output layer. The input layer includes input nodes, the hidden layer includes hidden nodes, the output layer includes one output node, and the input nodes are used to input the -dimensional finite state variable and one-dimensional action variable in the Q-Learning algorithm one by one. represents a positive integer greater than or equal to 2. represents a positive integer and has . The hidden nodes in each of the -dimensional Gaussian functions, and the output expression of the th hidden node is . represents a positive integer greater than or equal to 2. represents a positive integer less than or equal to . represents a positive integer less than or equal to . represents the natural exponential function. represents the input value of the th input node among the input nodes. represents the th radial basis function's center value in the th dimension. represents the width value of the th radial basis function in the th dimension. The output expression of the output node is . represents the weight value between the th hidden node and the output node. (B) The greedy strategy in the Q-Learning algorithm is executed by dynamically adjusting the greedy factor, where the dynamically adjusted greedy factor is expressed as follows: In the formula, represents the maximum value function, represents the preset maximum value of the greedy factor, represents the preset minimum value of the greedy factor, represents the decay amount of the greedy factor in each training round, represents the current training round number; (C) The action set in the Q-Learning algorithm includes forward action, backward action, left action, right action, left front action, right front action, left rear action, and right rear action.

5. The data center equipment inspection robot according to claim 3, characterized in that, Applying the beetle antenna search algorithm to optimize the path of the globally feasible path to obtain the shortest path of the pair of nodes includes the following steps S3221 to S3229: S3221. Initialize the required parameters of the beetle antennae search algorithm, and then execute step S3222, where the required parameters include the number of beetle individuals , the initial value of the beetle movement step size, the initial value of the beetle antenna perception length, and the maximum number of updates; Initialize the integer variable to 1, and then execute step S3223; ​ S3223. Place a longhorn beetle at the th non-starting and non-ending point on the global feasible path in the direction from the starting point to the ending point, and use the th non-starting and non-ending point on the global feasible path in the direction from the starting point to the ending point as the target position, and then execute step S3224; S3224. Create one-to-one corresponding random vectors for each longhorn beetle, and calculate the fitness values of the left and right antennae of each longhorn beetle among the longhorn beetles according to the target position, and then execute step S3225; ​ S3225. For each longhorn beetle among the longhorn beetles, update the corresponding position according to the fitness value of the corresponding left and right antennae, calculate the fitness value of the corresponding new position according to the target position, and then execute step S3226; S3226. Determine the best beetle according to the fitness values of the new positions of the beetles, and use the new position of the best beetle as the best position, and then execute step S3227; S3227. Determine whether the optimal position is better than the position of the th non-starting and ending point. If so, update the position of the th non-starting and ending point to the optimal position, and then execute step S3228. Otherwise, execute step S3228; S3228. Increment it by 1, then judge whether it is less than the total number of points of the global feasible path. If so, return to execute step S3223. Otherwise, update the movement step of the longhorn beetle and the perception length of the longhorn beetle's antenna, and then execute step S3229; ​ S3229. Determine whether the number of path point updates reaches the maximum number of updates. If so, generate the shortest path of the pair of nodes according to the starting point, ending point, and the current positions of all non-start and end points. Otherwise, return to execute step S3222.

6. The data center equipment inspection robot according to claim 5, wherein The fitness value is calculated using the following fitness function as follows: In the formula, represents the position variable, represents the distance penalty coefficient, represents the obstacle collision penalty coefficient, represents the distance cost from the position variable to the target position, represents the obstacle penalty cost, represents the shortest distance from the obstacle in the environmental map to the path segment from the position variable to the target position, represents the preset minimum safety distance.

7. The data center equipment inspection robot according to claim 5, wherein, Updating the beetle movement step length and the beetle antenna sensing length includes: Update the beetle movement step length according to the following formula: In the formula, represents the current search times of the longhorn beetle, represents the movement step length of the longhorn beetle at the -th search, and when there is which represents the initial value of the movement step length of the longhorn beetle, represents the movement step length of the longhorn beetle at the -th search, represents the preset minimum step length, represents the preset step length attenuation coefficient; Update the beetle antenna sensing length according to the following formula: Wherein, represents the antenna sensing length of the longhorn beetle during the th search, represents a preset antenna length factor.

8. The data center equipment inspection robot according to claim 2, characterized in that, According to the shortest paths of the pairs of nodes, splice to obtain a recommended route with the shortest total distance and passing through all the inspection stop nodes in the environmental map in sequence, including: Perform a combined permutation on the shortest paths of the pairs of nodes to obtain multiple path queues. Among them, each path queue in the multiple path queues conforms to the following rules: the starting point of the first shortest path in the path queue is the starting point, and the ending point of the adjacent previous shortest path in the path queue is the same as the starting point of the adjacent subsequent shortest path, and the ending points of all the shortest paths in the path queue include all the inspection stop nodes in the environmental map; For each path queue in the multiple path queues, calculate the corresponding total distance by superimposing according to the corresponding all shortest paths; Perform a splicing process on all the shortest paths of a path queue corresponding to the shortest total distance to obtain a recommended route with the shortest total distance and passing through all the inspection stop nodes in the environmental map in sequence.

9. The data center equipment inspection robot according to claim 1, characterized in that, It also includes a wireless communication module communicatively connected to the industrial control computer; The industrial control computer is further configured to perform the following steps after identifying the number and color of the indicator lights in the current indicator light picture by using computer vision technology: According to the recognition result and the indicator light data and color of the data center device in the normal state, determine whether the data center device is abnormal. If so, send an alarm message for the data center device to the data center device inspection platform or the data center device inspector through the wireless communication module.

10. The data center equipment inspection robot according to claim 1, characterized in that, After the full-coverage movement ends, call the map storage package in the ROS development environment to save the environmental map.

Images