Swarm robot fault mutual diagnosis system and method based on cloud edge cooperation

Abstract

The invention discloses a swarm robot fault mutual diagnosis system and method based on cloud edge cooperation, and relates to the technical field of robot intelligent operation and maintenance. According to the invention, a cloud edge cooperation technology is adopted, efficient information sharing and cooperation are realized among a plurality of robots, and a distributed mutual monitoring and diagnosis mechanism is utilized, so that the states of the robots are monitored, the load capacity is reduced, and more functional modules can be carried; according to the Kalman filtering method based on multi-model adaptive estimation, the speed and displacement of the robot can be accurately estimated, a suspected fault robot can be more accurately found based on data consistency judgment deviation, and a cloud server further diagnoses according to historical data so that the judgment result can be more accurate. The cloud-side cooperation enables the maintenance personnel to find the fault more quickly and carry out maintenance in time, thereby guaranteeing the overall stability and reliability of the system.

Description

Technical Field

[0001] The present invention relates to the field of robot intelligent operation and maintenance technology, and specifically to a group robot fault mutual diagnosis system and method based on cloud-edge collaboration. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and may not constitute prior art.

[0003] Robotics has fundamentally changed the face of traditional industrial production and has had a profound impact on human society. Robot work is divided into single robots and swarm robots. Although a single robot has basic perception, movement and communication capabilities, the speed of information acquisition, processing and task execution of a single robot is limited, and it cannot cope with large-scale task requirements; while the swarm robot system can improve the overall perception coverage and enhance the parallel processing capability of tasks through the collaboration of multiple robots, effectively overcoming the limitations of a single robot. For swarm robot systems, with the increase in task complexity and number, efficient management and operation and maintenance of robots are particularly important. Compared with a single robot, swarm robots can achieve autonomous monitoring and fault handling through collective collaboration and information sharing. Each robot can monitor its own status in a timely manner, detect and repair faults, and coordinate the operation and maintenance tasks of other robots through information sharing within the group. This autonomous operation and maintenance method improves the reliability, efficiency and sustainability of the swarm robot system, and ensures the long-term stable operation of the system in a complex environment.

[0004] The swarm robot system has shown significant advantages in task execution, but its complex multi-robot coordination mechanism also brings new challenges. Robot failures are inevitable during operation. The failure of any robot in the swarm robot system may affect the efficiency and stability of the entire system. Therefore, the operation prediction and precise operation and maintenance of swarm robots become particularly important. Through effective robot operation prediction and operation and maintenance mechanisms, the robot faults can be effectively diagnosed and maintained, the normal operation of the system can be maintained, and the coordination of the group and the successful completion of the task can be ensured. A good fault diagnosis method can improve the reliability and stability of the swarm robot system.

[0005] Endogenous fault detection relies on the sensor data and internal state information of the individual robot to detect faults, such as hardware failure and software anomalies. It has the advantages of strong autonomy and rapid response, but its detection capability is limited. Exogenous fault detection detects individual faults, such as abnormal behavior or task execution problems, through the observation and information exchange of other robots in the group. It can use the collaboration of the group to improve the robustness and coverage of detection and give full play to the advantages of group robots. Existing exogenous fault detection methods include model prediction, mutual observation voting, machine learning, and network-based centralized monitoring.

[0006] Mathematical models are used to predict the expected state of each robot, and faults are detected by analyzing the residuals between the actual observed values ​​and the predicted values. This method relies on building an accurate system model and identifying faults by detecting residuals that are larger than expected. If the model is not accurate enough or does not cover all possible system behaviors, the prediction accuracy may be low, affecting the reliability of fault detection. Robots judge the state of individuals by observing each other or sharing information. For controversial fault judgments, the system can adopt a voting mechanism, and other robots decide whether a robot is faulty. In the presence of malicious robots, false information may lead to incorrect fault judgments. Machine learning algorithms (such as anomaly detection, classifiers, etc.) are used to analyze the behavioral data of the robot group to detect and classify faults. This method can learn complex behavior patterns and detect anomalies. Machine learning methods usually require a large amount of training data and have high requirements for the quality of training data. If there is data bias in the training data, the model may not be able to correctly learn all normal and abnormal behavior patterns, resulting in reduced detection accuracy. Network-based centralized monitoring collects the status data of all robots in the group through the network for centralized monitoring and analysis. The centralized monitoring system can detect the status of the entire group and detect anomalies or faults. This method relies on a reliable communication infrastructure, which may affect the real-time and accuracy of fault detection in the event of communication delays or data loss. Summary of the invention

[0007] The purpose of the present invention is to provide a group robot fault mutual diagnosis system and method based on cloud-edge collaboration to solve the problems existing in the prior art.

[0008] The technical solution of the present invention is as follows:

[0009] A group robot fault mutual diagnosis system based on cloud-edge collaboration, comprising: a data acquisition unit, an edge data processing and storage unit, a transfer data processing and storage unit, and a cloud data processing and storage unit;

[0010] The data acquisition unit is installed on each robot and is used to collect the position and speed data of other robots at each moment;

[0011] The edge data processing and storage unit is installed on each robot, and can accurately estimate the displacement and speed of other robots based on the data collected by the data acquisition unit using a multi-model adaptive Kalman filtering method, and compare it with the preset displacement and speed of other robots to determine whether it exceeds the error range;

[0012] The cloud data processing and storage unit is deployed together with the transit data processing and storage unit on the cloud, and the cloud data processing and storage unit and the edge data processing and storage unit transmit and receive data through the transit data processing and storage unit; the cloud data processing and storage unit sends the preset displacement and speed of all robots to each edge data processing and storage unit on the one hand, and receives the judgment results of the edge data processing and storage unit and the calculated displacement and speed on the other hand, and parses the processing results to perform further analysis and decision-making, and if a fault is found, an alarm is issued to notify the maintenance personnel.

[0013] Furthermore, the edge data processing and storage unit, the transit data processing and storage unit and the cloud data processing and storage unit adopt a publish and subscribe mode;

[0014] The publish and subscribe mode supports a one-to-many communication model, allowing a publisher to send messages to multiple subscribers, and supports message persistence. Even if the subscriber is temporarily unavailable, the message will not be lost. When the subscriber reconnects, the unprocessed message can still be received.

[0015] Furthermore, the cloud data processing and storage unit packages the preset displacement and speed data of all robots into a protocol message, and then publishes the packaged protocol message to topic A. The transit data processing and storage unit receives the published topic A and forwards it to all edge data processing and storage units that have subscribed to topic A.

[0016] When the edge data processing and storage unit starts, it subscribes to topic A, receives messages in topic A from the transit data processing and storage unit, parses the preset displacement and speed data of all robots, and stores them locally for subsequent use;

[0017] The edge data processing and storage unit uses a multi-model adaptive Kalman filtering method to accurately estimate the displacement and speed of other robots based on the data collected by the data acquisition unit, and compares it with the preset displacement and speed of other robots to determine whether it exceeds the error range and generate a judgment result; and packages the judgment result and the calculated displacement and speed into a protocol message and publishes it to topic B; the transit data processing and storage unit receives the published topic B and forwards it to the cloud data processing and storage unit that subscribes to topic B;

[0018] The cloud data processing and storage unit subscribes to topic B, receives the protocol message from the transit data processing and storage unit, parses the processing results, makes further analysis and decision, and issues an alarm to notify maintenance personnel if a fault is found.

[0019] Furthermore, it also includes: a human-computer interaction unit, which is connected to the cloud data processing and storage unit to provide a human-computer interaction function.

[0020] The present invention also proposes a group robot fault mutual diagnosis method based on cloud-edge collaboration, based on the above-mentioned group robot fault mutual diagnosis system based on cloud-edge collaboration, including:

[0021] Step S1: Each robot obtains the position and speed data of other robots at the current moment;

[0022] Step S2: Based on the acquired position and velocity data, the displacement and velocity of other robots are accurately estimated using a multi-model adaptive Kalman filter method;

[0023] Step S3: Compare the estimated result obtained in step S2 with the preset displacement and speed of other robots to determine whether it exceeds the error range; if it does not exceed the error range, jump to step S1 and reacquire data; if it exceeds the error range, continue to step S4;

[0024] Step S4: Based on historical data, diagnose the estimation result obtained in step S2 to determine whether a fault occurs.

[0025] Furthermore, the step S1 comprises:

[0026] The data acquisition unit carried by each robot collects the position and speed data of other robots;

[0027] And m, n, p, q... Kalman filter models are established for each robot respectively; wherein the numerical values ​​of m, n, p, q... are determined according to the possible motion trajectory and speed preset for each robot.

[0028] Furthermore, the step S2 comprises:

[0029] Step A: For a robot in a group of robots with m Kalman filter models, k∈(0,m), t i The data information at the moment is based on t i-1 The real state space model corresponding to the kth Kalman filter model, derived from the data information at the moment, contains noise and is set as:

[0030] x k (t i )=Fk x k (t i-1 )+B k u(t i-1 )+ω k (t i )

[0031] z k (t i )=H k x k (t i )+v k (t i )

[0032] x k (t i ) represents the kth Kalman filter model t i The state vector at the moment;

[0033] F k Represents the state transfer matrix of the kth Kalman filter model;

[0034] B k represents the control matrix of the kth Kalman filter model;

[0035] ω k (t i ) represents the process noise vector of the kth Kalman filter model, with a mean of 0 and a covariance of Q k ;

[0036] z k (t i ) represents the observation vector of the kth Kalman filter model;

[0037] H k (t i ) represents the observation matrix of the kth Kalman filter model;

[0038] v k (t i ) represents the observation noise vector of the kth Kalman filter model, with a mean of 0 and a covariance of R k ;

[0039] Step B: Calculate the predicted state vector based on the real state space model, that is, the prior predicted state vector The initial predicted state vector The initial observation value is used, that is, the state data collected by the data acquisition unit at the first sampling time point;

[0040]

[0041] Step C: Set the initial error covariance matrix P for the Kalman filter model k (t i-1 ), and then calculate the prior error covariance matrix

[0042]

[0043] Step D: Calculate the residual covariance matrix S k (t i );

[0044]

[0045] Step E: Calculate the Kalman gain K k (t i );

[0046]

[0047] Step F: Set t i The observed value at time z k (t i ) into the equation and calculate the residual vector r k (t i ), where the observed value z k (t i ) is calculated by the data acquisition unit;

[0048]

[0049] Step G: Update the state estimate, that is, obtain the posterior prediction state vector

[0050]

[0051] Step H: Update the error covariance matrix of the Kalman filter state to obtain the posterior error covariance matrix

[0052]

[0053] Step I: After obtaining the updated state estimate, calculate the probability of the assumed Kalman filter model; if the Kalman filter model matches the true system model, the residual follows a Gaussian distribution, so the conditional probability of each model is:

[0054]

[0055] f k is the Gaussian likelihood function:

[0056]

[0057] Step J: Multiply the state of each Kalman filter by the probability that the filter is correct, i.e., the conditional probability, and then sum them to obtain the final estimated state x k (t ik ) and the final error covariance matrix P k (t ik );

[0058]

[0059] Step K: The final estimated state x of step J k (t ik ) is the obtained t i Accurate displacement and velocity data at all times.

[0060] Furthermore, the step S4 comprises:

[0061] Step S41: Calculate the mean μ of the robot's normal historical data corresponding to the received estimation result t and standard deviation σ t ;

[0062] Step S42: Use the Z-Score to determine whether the new data point is abnormal. Set a threshold. If the Z-Score of a data point exceeds the threshold, the data point is considered abnormal.

[0063] For the new trajectory data, calculate the Z-Score of the data at each time point:

[0064]

[0065] Z tx : Z-Score of the x-axis coordinate value at time t in the new trajectory data;

[0066] Z ty : Z-Score of the y-axis coordinate value at time t in the new trajectory data;

[0067] Z tv : Z-Score of the velocity value at time t in the new trajectory data;

[0068] Step S43: If the Z-Score of the new trajectory data at a certain time point exceeds the set threshold, the status data is considered to be abnormal, and it is determined that the robot corresponding to the received estimation result is faulty. The cloud data processing and storage unit sends an alarm to the maintenance personnel and sends the faulty robot information, and the diagnosis ends here; if the Z-Score of any time point of the new trajectory data does not exceed the set threshold, the status data is considered to be normal, and the cloud data processing and storage unit stores the data, and then repeats steps S1 to S4.

[0069] Furthermore, the mean μ of the trajectory t The calculation formula is as follows:

[0070]

[0071] The standard deviation σ of the trajectory t The calculation formula is as follows:

[0072]

[0073] n t Indicates the total number of historical data at time t;

[0074] x tk Represents the x-axis coordinate value at time t in the k-th historical trajectory data;

[0075] y tk Represents the coordinate value of the y-axis at time t in the k-th historical trajectory data;

[0076] μ tx Represents the mean value of the x-axis coordinate value at time t in all historical trajectory data;

[0077] μ ty Represents the mean value of the y-axis coordinate value at time t in all historical trajectory data;

[0078] σ tx Represents the standard deviation of the x-axis coordinate value at time t in all historical trajectory data;

[0079] σ ty Represents the standard deviation of the y-axis coordinate value at time t in all historical trajectory data.

[0080] Furthermore, the mean value μ of the velocity tv The calculation formula is as follows:

[0081]

[0082] The standard deviation σ of the speed tv The calculation formula is as follows:

[0083]

[0084] n t Indicates the total number of historical data at time t;

[0085] v tk Indicates the speed value corresponding to time t in the k-th historical trajectory data;

[0086] μ tv Represents the mean speed corresponding to time t in all historical trajectory data;

[0087] σ tv Represents the standard deviation of speed corresponding to time t in all historical trajectory data.

[0088] Compared with the prior art, the present invention has the following beneficial effects:

[0089] 1. The present invention adopts cloud-edge collaborative technology, and realizes efficient information sharing and collaboration among multiple robots, and utilizes distributed mutual monitoring and diagnosis mechanisms to realize the monitoring of the robot status, while also reducing the load and being able to carry more functional modules; the Kalman filtering method based on multi-model adaptive estimation can accurately estimate the speed and displacement of the robot, and can more accurately detect suspected faulty robots based on data consistency judgment deviation. The cloud server can further diagnose based on historical data to make the judgment result more accurate. The cooperation of cloud and edge enables maintenance personnel to discover faults more quickly and perform repairs in a timely manner, ensuring the overall stability and reliability of the system.

[0090] 2. The present invention uses a multi-model adaptive Kalman filtering method at the edge to accurately estimate the displacement and speed of the group robot, stores historical data in the cloud, and diagnoses whether the real-time data of the robot is faulty based on the mean and standard deviation of the historical data at the corresponding time. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 This is a schematic diagram of the principle of a group robot fault mutual diagnosis system based on cloud-edge collaboration;

[0092] Figure 2 This is a schematic diagram of cloud-edge communication mode;

[0093] Figure 3 This is a flow chart of a group robot fault mutual diagnosis method based on cloud-edge collaboration.

[0094] Figure numerals: 1-data acquisition unit, 2-edge data processing and storage unit, 3-transit data processing and storage unit, 4-cloud data processing and storage unit, 5-human-computer interaction unit, 6-cable, 7-wireless communication link, 8-video cable. DETAILED DESCRIPTION

[0095] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0096] The features and performance of the present invention are further described in detail below in conjunction with the embodiments.

[0097] Embodiment 1

[0098] See also Figure 1 , a group robot fault mutual diagnosis system based on cloud-edge collaboration, specifically including the following modules:

[0099] Data collection unit 1, edge data processing and storage unit 2, transfer data processing and storage unit 3 (i.e. Figure 2 Message proxy server in the cloud), cloud data processing and storage unit 4 (i.e. Figure 2 in the cloud server), human-computer interaction unit 5;

[0100] The data acquisition unit 1 is connected to the edge data processing and storage unit 2 via a cable 6, the edge data processing and storage unit 2 is connected to the transit data processing and storage unit 3, and the transit data processing and storage unit 3 is connected to the cloud data processing and storage unit 4 via a wireless communication link 7; the cloud data processing and storage unit 4 is connected to the human-computer interaction unit 5 via a video cable 8;

[0101] It should be noted that the human-machine interaction unit is connected to the cloud data processing and storage unit to provide human-machine interaction functions; it should be noted that the human-machine interaction unit is connected to the cloud data processing and storage unit via a video cable; in this embodiment, the human-machine interaction unit is a large display screen, which displays the latest received motion status of each robot, including their displacement and speed data, and displays their movement trajectory based on these data, and displays the diagnosis results of each robot. When the robot is judged as "failure occurs", an alarm message is issued on the display screen to allow maintenance personnel to handle it;

[0102] It should be noted that this embodiment takes the inspection of a certain wheeled group robot as an example, assuming that the preset motion state of each wheeled robot is a linear change. The data acquisition unit (i.e., sensor) carried by the robot is a lidar, the edge data processing and storage unit is an NVIDIA Jetson processor, and the cloud data processing and storage unit is a cloud server. The sensor data transmits the data to the NVIDIA Jetson processor through an interface. The NVIDIA Jetson processor communicates with the cloud server via a wireless network to upload data and receive data messages. The publish / subscribe protocol is used for data transmission. Assume that a certain robot in the embodiment has a fault in that a differential wheel motor briefly stops rotating, causing the path to deviate to the left; it should be noted that the data acquisition unit is connected to the edge data processing and storage unit via cable 6;

[0103] The data acquisition unit is installed on each robot and is used to collect the position and speed data of other robots at each moment;

[0104] The edge data processing and storage unit is installed on each robot, and can use the multi-model adaptive Kalman filtering method to accurately estimate the displacement and speed of other robots based on the data collected by the data acquisition unit, and compare it with the preset displacement and speed of other robots to determine whether it exceeds the error range; it should be noted that compared with the general Kalman filtering method, the multi-model Kalman filtering adopts a method of parallel processing of multiple models. Each model is based on different assumptions (different system parameters or noise characteristics), which enables the filter to adapt to different system dynamics. By running multiple filters in parallel, the errors and deficiencies that may exist in a single model can be avoided to a certain extent. By weighted averaging the estimation results of multiple models, the accuracy of the overall estimation can be improved. Especially in the case of large noise or low quality of observation data, it can better filter out noise and provide more accurate state estimation;

[0105] The cloud data processing and storage unit is deployed together with the transit data processing and storage unit on the cloud, and the cloud data processing and storage unit and the edge data processing and storage unit transmit and receive data through the transit data processing and storage unit; the cloud data processing and storage unit sends the preset displacement and speed of all robots to each edge data processing and storage unit on the one hand, and receives the judgment results of the edge data processing and storage unit and the calculated displacement and speed on the other hand, and parses the processing results to perform further analysis and decision-making, and if a fault is found, an alarm is issued to notify the maintenance personnel.

[0106] In this embodiment, specifically, Figure 2As shown, the edge data processing and storage unit, the transit data processing and storage unit, and the cloud data processing and storage unit adopt the publish and subscribe mode; it should be noted that the use of the publish and subscribe mode protocol ensures low latency and high reliability of data transmission; in the publish and subscribe mode, there is no direct connection between the publisher and the subscriber, and the two parties communicate through an intermediate agent. This loosely coupled architecture allows publishers and subscribers to be independently developed and deployed, greatly improving the flexibility and scalability of the system;

[0107] The publish and subscribe mode supports a one-to-many communication model, allowing a publisher to send messages to multiple subscribers, and supports message persistence. Even if the subscriber is temporarily unavailable, the message will not be lost. When the subscriber reconnects, the unprocessed message can still be received.

[0108] In this embodiment, specifically, the cloud data processing and storage unit packages the preset displacement and speed data of all robots (these data include information such as the position, speed and timestamp of each robot on the inspection path) into a protocol message, and then publishes the packaged protocol message to topic A. The transit data processing and storage unit receives the published topic A and forwards it to all edge data processing and storage units that have subscribed to topic A. It should be noted that after the edge data processing and storage unit receives the preset displacement and speed data, it sends a control command to the single-chip microcomputer, and the single-chip microcomputer sends a message to the motor driver to drive the robot to move. The communication framework consists of multiple clients and a proxy server, and the client can be any device that can connect to the server. The client can subscribe to one or more topics to the server to receive messages under these topics. The client can also publish messages to the server, and these messages will be forwarded to clients that have subscribed to the relevant topics.

[0109] That is, when the edge data processing and storage unit starts, it subscribes to topic A, receives messages in topic A from the transit data processing and storage unit, parses and obtains the preset displacement and speed data of all robots, and stores them locally for subsequent use;

[0110] The edge data processing and storage unit uses a multi-model adaptive Kalman filtering method to accurately estimate the displacement and speed of other robots based on the data collected by the data acquisition unit, and compares it with the preset displacement and speed of other robots to determine whether it exceeds the error range and generate a judgment result; and packages the judgment result and the calculated displacement and speed into a protocol message and publishes it to topic B; the transit data processing and storage unit receives the published topic B and forwards it to the cloud data processing and storage unit that subscribes to topic B;

[0111] The cloud data processing and storage unit subscribes to topic B, receives the protocol message from the transit data processing and storage unit, parses the processing results, makes further analysis and decision, and issues an alarm to notify maintenance personnel if a fault is found.

[0112] Embodiment 2

[0113] See also Figure 3 This embodiment is based on a group robot fault mutual diagnosis system based on cloud-edge collaboration proposed in Embodiment 1, and further proposes a group robot fault mutual diagnosis method based on cloud-edge collaboration, including:

[0114] Step S1: Each robot obtains the position and speed data of other robots at the current moment;

[0115] Step S2: Based on the acquired position and velocity data, the displacement and velocity of other robots are accurately estimated using a multi-model adaptive Kalman filter method;

[0116] Step S3: Compare the estimated result obtained in step S2 with the preset displacement and speed of other robots to determine whether it exceeds the error range; if it does not exceed the error range, jump to step S1 and reacquire data; if it exceeds the error range, continue to step S4;

[0117] Step S4: Based on historical data, diagnose the estimation result obtained in step S2 to determine whether a fault occurs.

[0118] In this embodiment, specifically, step S1 includes:

[0119] The data acquisition unit carried by each robot collects the position and speed data of other robots;

[0120] And establish m, n, p, q... Kalman filter models for each robot; wherein, the values ​​of m, n, p, q... are determined according to the possible motion trajectory and speed preset by each robot; that is, the values ​​of m, n, p, q... are determined according to the possible motion trajectory and speed preset by each robot, including normal and faulty ones, and there are multiple possible faulty trajectories and speeds, so the number of Kalman filter models set is also multiple; for example, m Kalman filter models are set for one of the robots, and these m Kalman filter models contain normal preset linear transformation motion states and faulty motion states, and the faulty motion states include path deviation to the left, path deviation to the right, left turn, right turn, etc. Different parameters are set for the multiple Kalman filter models corresponding to each robot, and the Kalman filter algorithm defines the state estimation and measurement update equations and the covariance matrix of the estimation error based on these models. Compared with the general Kalman filter method, the multi-model Kalman filter adopts the method of parallel processing of multiple models. Each model is based on different assumptions (different system parameters or noise characteristics), which enables the filter to adapt to different system dynamics. By running multiple filters in parallel, the errors and shortcomings that may exist in a single model can be avoided to a certain extent.

[0121] In this embodiment, specifically, step S2 includes:

[0122] Step A: For a robot in a group of robots with m Kalman filter models, k∈(0,m), t i The data information at the moment is based on t i-1 The real state space model corresponding to the kth Kalman filter model contains the process noise ω k , observation noise v k , it is usually assumed that ω k 、v k is a zero-mean Gaussian white noise with covariance Q k , R k If the accuracy of the system model is high, a smaller Q k ; Otherwise, set a larger Q k According to the technical specifications of the sensor or the experimental calibration results, set R k . State transfer matrix F k , control matrix B k Setting based on physical kinematic equations, H k The matrix is ​​set according to the relationship between the state variables and the observed variables; the state space model corresponding to the kth Kalman filter model is set as:

[0123] x k (t i )=Fk x k (t i-1 )+B k u(t i-1 )+ω k (t i )

[0124] z k (t i )=H k x k (t i )+v k (t i )

[0125] x k (t i ) represents the kth Kalman filter model t i The state vector at the moment;

[0126] F k Represents the state transfer matrix of the kth Kalman filter model;

[0127] B k represents the control matrix of the kth Kalman filter model;

[0128] ω k (t i ) represents the process noise vector of the kth Kalman filter model, with a mean of 0 and a covariance of Q k ;

[0129] z k (t i ) represents the observation vector of the kth Kalman filter model;

[0130] H k (t i ) represents the observation matrix of the kth Kalman filter model;

[0131] v k (t i ) represents the observation noise vector of the kth Kalman filter model, with a mean of 0 and a covariance of R k ;

[0132] Step B: Calculate the predicted state vector based on the real state space model, that is, the prior predicted state vector The initial predicted state vector The initial observation value is used, that is, the state data collected by the data acquisition unit at the first sampling time point;

[0133]

[0134] Step C: Set the initial error covariance matrix P for the Kalman filter model k (t i-1 ), and then calculate the prior error covariance matrix

[0135]

[0136] That is, setting the initial error covariance matrix P for the Kalman filter model k (t i-1 ), the covariance matrix Q of the process noise k , the covariance matrix R of the observation noise k . Then calculate the prior error covariance matrix The initial error covariance matrix is ​​an important parameter of the Kalman filter, which represents the uncertainty of the state estimate at the initial moment. When setting it, it needs to be configured according to the uncertainty of the initial state of the system. If the measurement accuracy of the initial state is high, a larger initial error covariance matrix P should be set. k (t i-1 ); If the measurement accuracy of the initial state is low, a smaller initial error covariance matrix P can be set k (t i-1 );

[0137] Step D: Calculate the residual covariance matrix S k (t i );

[0138]

[0139] Step E: Calculate the Kalman gain K k (t i );

[0140]

[0141] It should be noted that the Kalman gain K k (t i ) is a key element in the Kalman filter and is used to update the state estimate. It determines the weight of the difference between the predicted state and the actual observation in updating the state estimate. The role of the Kalman gain is to optimize the estimate so that the error covariance of the new state estimate is minimized;

[0142] Step F: Set t i The observed value at time z k (t i ) into the equation and calculate the residual vector r k (t i ), where the observed value z k (t i) is calculated by the data acquisition unit;

[0143]

[0144] Step G: Update the state estimate, that is, obtain the posterior prediction state vector

[0145]

[0146] Step H: Update the error covariance matrix of the Kalman filter state to obtain the posterior error covariance matrix

[0147]

[0148] Step I: Get the updated state estimate Finally, the probability of the assumed Kalman filter model is calculated; if the Kalman filter model matches the real system model, the residual follows a Gaussian distribution, so the conditional probability of each model is:

[0149]

[0150] f k is the Gaussian likelihood function:

[0151]

[0152] Step J: Multiply the state of each Kalman filter by the probability that the filter is correct, i.e., the conditional probability, and then sum them to obtain the final estimated state x k (t ik ) and the final error covariance matrix P k (t ik );

[0153]

[0154]

[0155] The posterior prediction state vector of each Kalman filter is Posterior error covariance matrix Multiply it by the probability that the filter is correct, that is, the conditional probability, and then sum them up to get t i The final estimated state vector x at time k (t ik ) and the final error covariance matrix P k (t ik ), t i The final error covariance matrix P at the moment k (t ik ) as the next moment, ti+1 The initial error covariance matrix at the moment is used to calculate the prior error covariance matrix at the next moment

[0156] Step K: The final estimated state x of step J k (t ik ) is the obtained t i Accurate displacement and speed data at all times, which are used by edge data processing and storage units to preliminarily determine whether the data is deviated.

[0157] In this embodiment, specifically, step S3 includes:

[0158] The edge data processing and storage unit converts the robot t obtained above into i The accurate displacement and speed data at each moment are consistent with the robot’s preset t i The displacement and speed data at the moment are compared to determine whether they exceed the deviation range. In this embodiment, the deviation range of the displacement is set to: ±0.1m in the x-axis direction and ±0.1m in the y-axis direction; the deviation range of the speed is set to: ±10%. If t i If the status data at the moment does not exceed the deviation range, it is judged as t i No fault occurs at the time, and the estimated t i The accurate status data at the moment is uploaded to the transit data processing and storage unit according to the protocol format of the publish and subscribe mode, and then subscribed by the cloud data processing and storage unit, and finally stored in the cloud data processing and storage unit, and then step S1 is repeated; if t i If the state data at the moment exceeds the deviation range, the initial judgment is t i A fault occurs at time t i The accurate status data at each moment is uploaded to the transit data processing and storage unit in accordance with the protocol format of the publish and subscribe mode, and then to the cloud data processing and storage unit. After the cloud data processing and storage unit receives the data, it determines whether a fault occurs based on historical data.

[0159] In this embodiment, it should be noted that the diagnostic method used in step S4 is: the cloud data processing and storage unit uses a statistical analysis method based on historical data, determines whether such a fault has occurred by calculating the statistical difference between the actual data and the preset standard, uses the statistical characteristics of the historical data (mean, standard deviation) to set thresholds, and detects whether the real-time data deviates from these thresholds.

[0160] In this embodiment, specifically, step S4 includes:

[0161] Step S41: Calculate the mean μ of the robot's normal historical data corresponding to the received estimation result t and standard deviation σt ;

[0162] The mean μ of the trajectory t The calculation formula is as follows:

[0163]

[0164] The standard deviation σ of the trajectory t The calculation formula is as follows:

[0165]

[0166] n t Indicates the total number of historical data at time t;

[0167] x tk Represents the x-axis coordinate value at time t in the k-th historical trajectory data;

[0168] y tk Represents the coordinate value of the y-axis at time t in the k-th historical trajectory data;

[0169] μ tx Represents the mean value of the x-axis coordinate value at time t in all historical trajectory data;

[0170] μ ty Represents the mean value of the y-axis coordinate value at time t in all historical trajectory data;

[0171] σ tx Represents the standard deviation of the x-axis coordinate value at time t in all historical trajectory data;

[0172] σ ty Represents the standard deviation of the y-axis coordinate value at time t in all historical trajectory data;

[0173] The mean value μ of the velocity tv The calculation formula is as follows:

[0174]

[0175] The standard deviation σ of the speed tv The calculation formula is as follows:

[0176]

[0177] n t Indicates the total number of historical data at time t;

[0178] v tk Indicates the speed value corresponding to time t in the k-th historical trajectory data;

[0179] μ tvRepresents the mean speed corresponding to time t in all historical trajectory data;

[0180] σ tv Represents the standard deviation of speed corresponding to time t in all historical trajectory data;

[0181] Step S42: Use the Z-Score to determine whether the new data point is abnormal. Set a threshold (e.g., 3 or -3). If the Z-Score of a data point exceeds this threshold, the data point is considered abnormal.

[0182] For the new trajectory data, calculate the Z-Score of the data at each time point:

[0183]

[0184] Z tx : Z-Score of the x-axis coordinate value at time t in the new trajectory data;

[0185] Z ty : Z-Score of the y-axis coordinate value at time t in the new trajectory data;

[0186] Z tv : Z-Score of the velocity value at time t in the new trajectory data;

[0187] Step S43: If the Z-Score of the new trajectory data at a certain time point exceeds the set threshold, the state data is considered abnormal, and it is determined that the robot corresponding to the received estimation result is faulty, and the cloud data processing and storage unit sends an alarm to the maintenance personnel and sends the faulty robot information, and the diagnosis ends here; if the Z-Score of any time point of the new trajectory data does not exceed the set threshold, the state data is considered normal, and the cloud data processing and storage unit stores the data, and then repeats steps S1 to S4;

[0188] It should be noted that if the system has a low tolerance for faults, a lower Z-Score threshold (such as 2) can be selected; if the system is sensitive to false alarms, a higher Z-Score threshold (such as 3) can be selected. In this embodiment, the Z-Score threshold is set to 2.5. If the Z-Score of the new trajectory data at a certain time point exceeds the set threshold, the state data is considered abnormal, and it is judged that the robot corresponding to the data has a fault. The cloud data processing and storage unit sends an alarm to the maintenance personnel and sends the faulty robot information, and the diagnosis ends here; if the Z-Score of any time point of the new trajectory data does not exceed the set threshold, the state data is considered normal, and the cloud data processing and storage unit stores the data, and then repeats the above steps.

[0189] The above-mentioned embodiments only express the specific implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the protection scope of the present application. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the technical solution concept of the present application, and these all belong to the protection scope of the present application.

[0190] This background section is provided to generally present the context of the invention, and the work of the presently named inventors, the work to the extent described in this background section, and aspects of this section that did not constitute prior art at the time of application are neither explicitly nor implicitly admitted to be prior art to the present invention.

Claims

1. A group robot fault mutual diagnosis system based on cloud-edge collaboration, characterized in that: include: Data collection unit, edge data processing and storage unit, transit data processing and storage unit, cloud data processing and storage unit; The data acquisition unit is installed on each robot and is used to collect the position and speed data of other robots at each moment; The edge data processing and storage unit is installed on each robot, and can accurately estimate the displacement and speed of other robots based on the data collected by the data acquisition unit using a multi-model adaptive Kalman filtering method, and compare it with the preset displacement and speed of other robots to determine whether it exceeds the error range; The cloud data processing and storage unit is deployed together with the transit data processing and storage unit on the cloud, and the cloud data processing and storage unit and the edge data processing and storage unit transmit and receive data through the transit data processing and storage unit; the cloud data processing and storage unit sends the preset displacement and speed of all robots to each edge data processing and storage unit on the one hand, and receives the judgment results of the edge data processing and storage unit and the calculated displacement and speed on the other hand, and parses the processing results to perform further analysis and decision-making, and if a fault is found, an alarm is issued to notify the maintenance personnel.

2. According to the cloud-edge collaboration-based group robot fault mutual diagnosis system of claim 1, it is characterized in that: The edge data processing and storage unit, the transit data processing and storage unit and the cloud data processing and storage unit adopt a publish and subscribe mode; The publish and subscribe mode supports a one-to-many communication model, allowing a publisher to send messages to multiple subscribers, and supports message persistence. Even if the subscriber is temporarily unavailable, the message will not be lost. When the subscriber reconnects, the unprocessed message can still be received.

3. According to the cloud-edge collaboration-based group robot fault mutual diagnosis system of claim 2, it is characterized in that: The cloud data processing and storage unit packages the preset displacement and speed data of all robots into a protocol message, and then publishes the packaged protocol message to topic A. The transit data processing and storage unit receives the published topic A and forwards it to all edge data processing and storage units that have subscribed to topic A. When the edge data processing and storage unit starts, it subscribes to topic A, receives messages in topic A from the transit data processing and storage unit, parses the preset displacement and speed data of all robots, and stores them locally for subsequent use; The edge data processing and storage unit uses a multi-model adaptive Kalman filter method to accurately estimate the displacement and speed of other robots based on the data collected by the data acquisition unit, and compares them with the preset displacement and speed of other robots to determine whether they exceed the error range and generate a judgment result; The judgment result and the calculated displacement and velocity are packaged into a protocol message and published to topic B; the transit data processing and storage unit receives the published topic B and forwards it to the cloud data processing and storage unit that subscribes to topic B; The cloud data processing and storage unit subscribes to topic B, receives the protocol message from the transit data processing and storage unit, parses the processing results, makes further analysis and decision, and issues an alarm to notify maintenance personnel if a fault is found.

4. The group robot fault mutual diagnosis system based on cloud-edge collaboration according to claim 1 is characterized in that: Also includes: A human-computer interaction unit is connected to a cloud data processing and storage unit to provide a human-computer interaction function.

5. A group robot fault mutual diagnosis method based on cloud-edge collaboration, characterized in that: A group robot fault mutual diagnosis system based on cloud-edge collaboration according to any one of claims 1 to 4, comprising: Step S1: Each robot obtains the position and speed data of other robots at the current moment; Step S2: Based on the acquired position and velocity data, the displacement and velocity of other robots are accurately estimated using a multi-model adaptive Kalman filter method; Step S3: Compare the estimated result obtained in step S2 with the preset displacement and speed of other robots to determine whether it exceeds the error range; if it does not exceed the error range, jump to step S1 and reacquire data; if it exceeds the error range, continue to step S4; Step S4: Based on historical data, diagnose the estimation result obtained in step S2 to determine whether a fault occurs.

6. The method for mutual fault diagnosis of group robots based on cloud-edge collaboration according to claim 5 is characterized in that: The step S1 comprises: The data acquisition unit carried by each robot collects the position and speed data of other robots; And m, n, p, q... Kalman filter models are established for each robot respectively; wherein the numerical values ​​of m, n, p, q... are determined according to the possible motion trajectory and speed preset for each robot.

7. The method for mutual fault diagnosis of group robots based on cloud-edge collaboration according to claim 6 is characterized in that: The step S2 comprises: Step A: For a robot in a group of robots with m Kalman filter models, k∈(0,m), t i The data information at the moment is based on t i-1 The real state space model corresponding to the kth Kalman filter model, derived from the data information at the moment, contains noise and is set as: x k (t i )=F k x k (t i-1 )+B k u(t i-1 )+ω k (t i ) z k (t i )=H k x k (t i )+v k (t i ) x k (t i ) represents the kth Kalman filter model t i The state vector at the moment; F k Represents the state transfer matrix of the kth Kalman filter model; B k represents the control matrix of the kth Kalman filter model; ω k (t i ) represents the process noise vector of the kth Kalman filter model, with a mean of 0 and a covariance of Q k ; z k (t i ) represents the observation vector of the kth Kalman filter model; H k (t i ) represents the observation matrix of the kth Kalman filter model; v k (t i ) represents the observation noise vector of the kth Kalman filter model, with a mean of 0 and a covariance of R k ; Step B: Calculate the predicted state vector based on the real state space model, that is, the prior predicted state vector The initial predicted state vector The initial observation value is used, that is, the state data collected by the data acquisition unit at the first sampling time point; Step C: Set the initial error covariance matrix P for the Kalman filter model k (t i-1 ), and then calculate the prior error covariance matrix Step D: Calculate the residual covariance matrix S k (t i ); Step E: Calculate the Kalman gain K k (t i ); Step F: Set t i The observed value at time z k (t i ) into the equation and calculate the residual vector r k (t i ), where the observed value z k (t i ) is calculated by the data acquisition unit; Step G: Update the state estimate, that is, obtain the posterior prediction state vector Step H: Update the error covariance matrix of the Kalman filter state to obtain the posterior error covariance matrix Step I: After obtaining the updated state estimate, calculate the probability of the assumed Kalman filter model; if the Kalman filter model matches the true system model, the residual follows a Gaussian distribution, so the conditional probability of each model is: f k is the Gaussian likelihood function: Step J: Multiply the state of each Kalman filter by the probability that the filter is correct, i.e., the conditional probability, and then sum them to obtain the final estimated state x k (t ik ) and the final error covariance matrix P k (t ik ); Step K: The final estimated state x of step J k (t ik ) is the obtained t i Accurate displacement and velocity data at all times.

8. The method for mutual fault diagnosis of group robots based on cloud-edge collaboration according to claim 7 is characterized in that: The step S4 comprises: Step S41: Calculate the mean μ of the robot's normal historical data corresponding to the received estimation result t and standard deviation σ t ; Step S42: Use the Z-Score to determine whether the new data point is abnormal. Set a threshold. If the Z-Score of a data point exceeds the threshold, the data point is considered abnormal. For the new trajectory data, calculate the Z-Score of the data at each time point: Z tx : Z-Score of the x-axis coordinate value at time t in the new trajectory data; Z ty : Z-Score of the y-axis coordinate value at time t in the new trajectory data; Z tv : Z-Score of the velocity value at time t in the new trajectory data; Step S43: If the Z-Score of the new trajectory data at a certain time point exceeds the set threshold, the status data is considered to be abnormal, and it is determined that the robot corresponding to the received estimation result is faulty. The cloud data processing and storage unit sends an alarm to the maintenance personnel and sends the faulty robot information, and the diagnosis ends here; if the Z-Score of any time point of the new trajectory data does not exceed the set threshold, the status data is considered to be normal, and the cloud data processing and storage unit stores the data, and then repeats steps S1 to S4.

9. The method for mutual fault diagnosis of group robots based on cloud-edge collaboration according to claim 8 is characterized in that: The mean μ of the trajectory t The calculation formula is as follows: The standard deviation σ of the trajectory t The calculation formula is as follows: n t Indicates the total number of historical data at time t; x tk Represents the x-axis coordinate value at time t in the k-th historical trajectory data; y tk Represents the coordinate value of the y-axis at time t in the k-th historical trajectory data; μ tx Represents the mean value of the x-axis coordinate value at time t in all historical trajectory data; μ ty Represents the mean value of the y-axis coordinate value at time t in all historical trajectory data; σ tx Represents the standard deviation of the x-axis coordinate value at time t in all historical trajectory data; σ ty Represents the standard deviation of the y-axis coordinate value at time t in all historical trajectory data.

10. The method for mutual fault diagnosis of group robots based on cloud-edge collaboration according to claim 9 is characterized in that: The mean value μ of the velocity tv The calculation formula is as follows: The standard deviation σ of the speed tv The calculation formula is as follows: n t Indicates the total number of historical data at time t; v tk Indicates the speed value corresponding to time t in the k-th historical trajectory data; μ tv Represents the mean speed corresponding to time t in all historical trajectory data; σ tv Represents the standard deviation of speed corresponding to time t in all historical trajectory data.

Images