Distributed fault detection method in swarm robot system based on machine learning

Abstract

The invention relates to a distributed fault detection method in a swarm robot system based on machine learning, and belongs to the technical field of robot control and fault detection. The method aims at solving the single-point fault problem existing in centralized fault detection, and fault detection and decision are achieved through a distributed architecture. The method specifically comprises the steps of information collection, feature extraction, machine learning classification, communication sharing, voting decision making and the like, the fault detection precision is improved through a machine learning classification algorithm, misjudgment is reduced through a weighted majority voting mechanism, and the robustness and fault-tolerant capability of the system are enhanced. The method can be widely applied to swarm robot system fault detection and cooperative work in a complex environment.

Description

Technical Field

[0001] The present invention relates to a fault detection technology in a multi-agent system, and in particular to a distributed fault detection method in a swarm robot system (SRS). Background Art

[0002] Swarm robotic systems consist of multiple intelligently interconnected nodes that exhibit complex group behaviors through collaborative cooperation. Although individual robots have limited capabilities in performing certain specific tasks, swarm robotic systems are able to leverage their distributed nature to accomplish complex tasks without central control through communication and interaction between nodes. This distributed architecture is often considered inherently robust. However, studies have shown that even failures in only a few robots can significantly affect the performance and stability of the entire system. Existing fault detection methods are generally divided into two categories: exogenous and endogenous. The former is used to detect faults in other robots, while the latter is used to detect self-faults. These methods can be divided into qualitative and quantitative methods based on different classification algorithms. Qualitative methods have simple structures and low resource consumption, but their detection accuracy is limited; quantitative methods rely on modeling of robot behavior and can provide higher detection accuracy. Quantitative methods are further subdivided into model-based and data-driven algorithms. The former relies on domain knowledge, while the latter autonomously learns through data analysis to distinguish between normal and faulty behaviors. With the continuous advancement of data storage technology and machine learning algorithms, data-driven methods have shown increasing potential in fault detection in swarm robotic systems. Summary of the invention

[0003] The technical problem to be solved by the present invention is how to implement distributed fault detection in a swarm robot system, thereby improving the robustness and fault tolerance of the system and avoiding the single point failure problem of centralized fault detection.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solution: The present invention designs a distributed fault detection method based on a swarm robot system, comprising the following steps:

[0005] Step 001: Each robot observes the neighboring robots around it through its own sensors and communication modules, and collects the behavior data of its neighboring robots. These data include information such as the distance, relative angle, speed and acceleration with the neighboring robots, so as to monitor and analyze their behaviors.

[0006] Step 002: The observed raw data is processed to generate features that represent the behavior of the neighboring robot. These features can be binary features (such as whether the robot is within a specified distance range) or numerical features (such as speed, acceleration, etc.). These features provide a basis for subsequent classification and help identify whether the robot has abnormal behavior.

[0007] Step 003: After feature extraction is completed, the robot conducts a preliminary assessment of its neighbors’ behaviors to identify possible anomalies. For example, if a robot is observed to remain motionless or move at an unreasonable speed, the system will mark the robot as “suspicious” to provide a reference for the subsequent classification stage.

[0008] Step 004: Within a time window, the robot accumulates feature data observed multiple times and passes these features as input to the machine learning classifier. By combining multiple observed features, the changing trend of the robot's behavior can be more comprehensively reflected.

[0009] In step 005, each robot uses a machine learning classifier (such as logistic regression or gradient boosting decision tree) to judge the fault status of its neighboring robots. The classifier classifies the robots according to the input feature data and outputs a judgment on whether there is a fault. Logistic regression is suitable for real-time detection with low resource consumption, while gradient boosting decision tree can handle more complex feature relationships and provide higher classification accuracy.

[0010] Step 006: After analyzing the feature data, the classifier generates a fault label. If the fault probability output by the classifier exceeds the set threshold, the robot is marked as a faulty robot; if it is lower than the threshold, it is marked as a normal working robot.

[0011] Step 007: After the classification is completed, the robot transmits the classification results of its neighboring robots to other surrounding robots through the local communication network. This information sharing enables each robot to obtain the classification results of multiple robots on the same target, providing basic data for the final voting decision.

[0012] Step 008: After receiving the classification results, each robot makes a collective decision through a majority voting mechanism. If the majority of robots determine that a robot is faulty, the robot is collectively identified as a faulty robot; if the majority of robots determine that it is normal, it is considered to be working normally. This distributed voting mechanism can effectively reduce the impact of a single robot's misjudgment on the overall system.

[0013] Step 009, after voting to determine whether a robot is faulty, the swarm robot system makes a corresponding decision based on the voting result. For a faulty robot, the system can choose to isolate or stop interacting with it, or maintain the normal operation of the overall system by reallocating tasks. In addition, it can also trigger self-recovery operations or send fault alarm signals based on the fault situation.

[0014] As a preferred technical solution of the present invention: each robot collects behavioral data such as distance, angle, speed, etc. with its neighboring robots through high-precision sensors (such as ultrasonic, infrared sensors or laser radar).

[0015] As a preferred technical solution of the present invention: in the machine learning classification stage of step 005, the system preferably uses a gradient boosting decision tree (GBDT) as the main classification algorithm. GBDT can efficiently process complex nonlinear features and achieve high-precision classification results by integrating multiple weak classifiers.

[0016] As a preferred technical solution of the present invention: in the voting mechanism of step 008, the weight of each robot's vote can be dynamically adjusted according to the accuracy of its sensor, data quality or the importance of its position.

[0017] As a preferred technical solution of the present invention: the system introduces a redundant communication mechanism, that is, each robot not only exchanges fault classification information with its direct neighbors, but also transmits the classification results to robots at a longer distance through multi-hop communication.

[0018] In order to solve the above technical problems, the present invention adopts the following technical solution: The present invention designs a distributed group robot fault detection system, which includes the following units:

[0019] Sensor unit: used to collect behavioral data such as distance, angle, speed and acceleration of neighboring robots to provide basic information for fault detection;

[0020] Data processing and feature extraction unit: responsible for processing sensor data and extracting binary or numerical features that can characterize the robot's behavior;

[0021] Machine learning classification unit: using the extracted feature data, the machine learning classifier is used to determine whether the robot has a fault;

[0022] Communication and sharing unit: responsible for the transmission and sharing of classification results between robots, ensuring that the information covers every robot in the entire system;

[0023] Voting decision unit: collectively decides whether a robot has a fault through a weighted majority voting mechanism;

[0024] Fault handling unit: After confirming the fault, take measures such as isolation, task reallocation or fault recovery to ensure the normal operation of the system.

[0025] The system of the present invention adopts distributed fault detection and weighted voting mechanism, which has higher robustness, real-time performance and detection accuracy compared with the prior art. It improves the accuracy of fault detection by introducing machine learning algorithm, and flexibly handles the fault according to its severity to ensure stable operation of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of a distributed group robot system fault detection method designed by the present invention. DETAILED DESCRIPTION

[0027] like Figure 1 As shown, the present invention designs a distributed group robot system detection method to perform distributed fault detection and decision-making on the system. In the actual application process, it specifically includes the following steps:

[0028] Step 001, each robot collects the behavior data of the neighboring robots through sensors to determine whether the data is successfully obtained: if yes, it goes to step 002, otherwise it collects data again.

[0029] Step 002: The robot processes the raw data to generate a feature vector.

[0030] In step 003, the robot makes a preliminary assessment of the extracted features, identifies whether there are any anomalies, and determines whether any abnormal behavior is found: if so, it is marked as "suspicious state" and proceeds to step 004; otherwise, it directly proceeds to step 4.

[0031] Step 004, input the feature data within a time window into the machine learning classifier to determine whether sufficient feature data has been collected: if yes, proceed to step 005, otherwise continue to accumulate data.

[0032] Step 005, analyze the features through a machine learning classifier to determine whether the robot has a fault.

[0033] Step 006, generate a fault label based on the analysis results of the classifier, and determine whether the fault probability exceeds the set threshold: if yes, mark it as a "faulty robot", otherwise mark it as a "normal working robot" and proceed to step 007.

[0034] Step 007, transmit the classification result to the neighboring robot through the local communication network to determine whether the result is shared successfully: if yes, proceed to step 008, otherwise retry sharing.

[0035] Step 008, collectively decide whether a robot has a fault through a weighted majority voting mechanism, and determine whether the majority voting results confirm a fault: if yes, proceed to step 009, otherwise end the detection process of the robot.

[0036] Step 009, take corresponding measures according to the voting results to determine the severity of the fault: if yes, isolate the faulty robot or reassign tasks; otherwise, continue monitoring or end detection.

[0037] In view of the above-mentioned specifically designed distributed group robot fault detection method, the present invention further designs a system based on the distributed fault detection method, which specifically includes a sensor unit, a data processing and feature extraction unit, a machine learning classification unit, a communication and sharing unit, a voting decision unit, and a fault processing unit, and makes specific designs for each of the above units. Among them, the sensor unit: is used to collect behavioral data between neighboring robots. The sensor unit can monitor the surrounding environment in real time and generate raw data to provide basic information for fault detection; data processing and feature extraction unit: responsible for processing the raw data collected by the sensor unit, extracting features that can characterize the behavior of neighboring robots, and providing analysis basis for subsequent classifiers; machine learning classification unit: responsible for inputting the extracted features into the machine learning classifier, using algorithms such as gradient boosting decision tree (GBDT) to classify and judge the fault and non-fault states of neighboring robots, and judging whether the robot is faulty according to the preset threshold; communication and sharing unit: used for sharing classification results between robots, using local communication or multi-hop communication to ensure that each robot can exchange information with neighboring robots and pass the classification results to more robots; voting decision unit: by collecting the classification results of other robots and using a weighted majority voting mechanism to determine whether a robot is faulty; fault handling unit: the fault handling unit takes corresponding measures according to the severity of the fault, such as isolating the faulty robot, reallocating tasks, or triggering fault recovery operations to ensure the normal operation of the overall function of the group system.

[0038] The method and system designed by the above technical solution have the following technical effects compared with the prior art: the method and system described in the present invention can realize efficient distributed fault detection in the group robot system. Each robot identifies faults through real-time observation, data processing and machine learning classifiers, and combines the voting mechanism to ensure the accuracy of the classification results. Through multi-level decision-making steps, the system can isolate faulty robots or reallocate tasks in time to ensure the continuity of group tasks and the robustness of the overall system. This distributed method improves the system's ability to respond to environmental changes and robot failures, and greatly improves the stability and reliability of group robots in complex environments.

[0039] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A distributed fault detection method in a swarm robot system based on machine learning, characterized in that: It includes an information collection module, a data processing and feature extraction module, a fault classification module, a communication and sharing module, a voting decision module and a fault processing module. Each module realizes fault detection and decision-making of the robot system through a distributed architecture.

2. According to the multi-agent attitude cooperative control system in a complex environment according to claim 1, it is characterized by: The information acquisition module includes a high-precision sensor for collecting data such as the distance, angle, speed and acceleration of neighboring robots, and the fault classification module uses a gradient boosting decision tree (GBDT) algorithm to classify faults of neighboring robots.

3. The distributed fault detection method in a swarm robot system based on machine learning according to claim 1, characterized in that: The voting decision module dynamically adjusts the voting weights according to the sensor accuracy and data quality of different robots through a weighted majority voting mechanism.

4. The distributed fault detection method in a swarm robot system based on machine learning according to claim 1, characterized in that: The communication and sharing module uses a redundant communication mechanism to enable each robot to not only exchange classification results with its direct neighbors, but also transmit information through multi-hop communication to ensure fault information sharing throughout the system.

5. The distributed fault detection method in a swarm robot system based on machine learning according to claim 1, characterized in that: After confirming the faulty robot, the fault handling module performs isolation, task reallocation or self-recovery operations to ensure the stable operation of the system.

6. A distributed fault detection system for a swarm robot system, characterized in that: The system comprises a sensor module, a data processing and feature extraction module, a fault classification module, a communication and sharing module, a voting decision module and a fault processing module. The system improves the robustness and real-time performance of fault detection through a distributed detection mechanism.

7. A distributed fault detection system for a swarm robot system according to claim 6, characterized in that: The sensor module includes an ultrasonic sensor, an infrared sensor or a laser radar, which is used to collect behavior data of neighboring robots.

8. The distributed fault detection system of a swarm robot system according to claim 6, characterized in that: The fault classification module classifies the extracted features through the gradient boosted decision tree (GBDT) algorithm and determines whether the robot has a fault according to a set threshold.

9. A distributed fault detection system for a swarm robot system according to claim 6, characterized in that: The communication and sharing module transmits the classification results to the remote robot through multi-hop communication, thereby realizing wide-area information sharing within the system.

Images