Interconnection communication system for fault detection of road berth robot

Abstract

The invention relates to the technical field of intelligent traffic and robots, in particular to an interconnection communication system for fault detection of a road berth robot, which is used for solving the problems that a fault recognition mechanism integrating various abnormal conditions cannot be constructed, key task information cannot be extracted for replacing judgment and the fault detection efficiency cannot be improved in the prior art. An optimal substitute cannot be selected through a comprehensive cost model, and the fault tolerance, the task completion rate and the operation robustness of the system cannot be improved. Lightweight distributed modeling is realized through a collaborative task recovery module, a fault recognition mechanism synthesizes various abnormal conditions, key task information is extracted for replacement judgment, candidate robots are screened according to distances, priorities and resource conditions, an optimal replacement person is selected through a comprehensive cost model, and the optimal replacement person is selected according to the optimal cost model. And the successor executes movement control, obstacle avoidance adjustment and state broadcasting according to the perception graph and the state of the successor, so that the fault tolerance, the task completion rate and the operation robustness of the system are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation and robotics, and more specifically, to an interconnected communication system for fault detection of road parking robots. Background Technology

[0002] In the construction of smart cities, road parking robots are key equipment in smart parking systems, undertaking tasks such as vehicle guidance, parking space management, and fee settlement. However, due to the complex environment, long-term operation, and multi-machine collaborative operation, they are prone to failure. Existing detection mainly relies on a single sensor or manual inspection, which makes it difficult to detect faults in a timely and comprehensive manner. Communication systems mostly adopt traditional methods, which have problems such as signal interference, short transmission distance, and poor reliability, resulting in poor information transmission, affecting processing efficiency, and may even cause traffic congestion.

[0003] Referring to patent application CN119229681A, a parking robot-assisted intelligent roadside parking scheduling method and system is disclosed. This method acquires multiple parking spaces and classifies and marks them according to their size and type. It then performs parking complexity analysis by combining multiple parking space type markings, parking space slope information, and neighborhood road information. The method predicts the parking skill level of the first driver and generates a first recommended parking space by combining multiple parking space size markings, the first parking skill prediction index, and multiple complexity indices. Based on the first recommended parking space, a first navigation route is generated, and the parking robot guides the driver to park according to the first navigation route. This solves the technical problem in existing technologies where parking robots do not consider differences in driver parking skill levels, resulting in excessively difficult recommended parking spaces and poor user experience. The method achieves the technical effect of matching the driver's parking skill level with the complexity of the parking space, thereby improving parking success rate and user experience.

[0004] However, while the aforementioned reference patents combine driver parking level prediction with parking space complexity analysis to achieve intelligent parking space matching and navigation guidance based on parking robots, thereby improving parking success rate and user experience, they cannot integrate fault classification, system health score, and task behavior sequence. They cannot identify fault states through multi-label models, cannot depict the causal chain of "fault → health score → task execution → completion," and cannot achieve full-link evaluation. Furthermore, they cannot construct a fault identification mechanism that integrates multiple abnormal situations, cannot extract key task information for succession judgment, cannot select the optimal successor through a comprehensive cost model, and cannot improve system fault tolerance, task completion rate, and operational robustness.

[0005] To address these issues, we propose an interconnected communication system for fault detection in road parking robots. Summary of the Invention

[0006] The purpose of this invention is to provide an interconnected communication system for fault detection of road parking robots. This system solves the problems of existing technologies, which cannot integrate fault classification, system health score, and task behavior sequence; cannot identify fault states through multi-label models; cannot depict the causal chain of "fault → health score → task execution → completion"; and cannot achieve full-link evaluation. Furthermore, these technologies cannot construct a fault identification mechanism that integrates multiple abnormal situations; cannot extract key task information for succession judgment; cannot select the optimal successor through a comprehensive cost model; and cannot improve system fault tolerance, task completion rate, and operational robustness.

[0007] The objective of this invention is achieved through the following technical solution:

[0008] An interconnected communication system for fault detection of road parking robots, applied to a parking robot communication management platform, includes:

[0009] The data preprocessing module is used to collect the status perception data of the parking robot and perform preprocessing operations on the collected status perception data.

[0010] The fault diagnosis module extracts time-series features from preprocessed state-aware data, integrates time-series and image information to identify various faults, and outputs multi-label fault classification results and system health scores.

[0011] The task impact analysis module, based on the fault classification results and system health score, combined with the task behavior sequence, uses a structural causal model to assess the degree of impact of the fault on the current task objective and outputs the task impact level assessment result.

[0012] The collaborative task recovery module constructs a local perception map based on the position and task status of neighboring robots, combines the task intent of the faulty robot with the system status to determine the task succession conditions, and outputs task allocation and collaborative action decisions.

[0013] In a preferred embodiment of the present invention, the process by which the fault diagnosis module extracts temporal features based on preprocessed state-aware data and fuses temporal and image information to identify various faults includes:

[0014] Acquire preprocessed state perception data, which includes lateral acceleration, rotational acceleration, encoder pulse count, motor temperature, battery voltage, signal strength, transmission delay time, and obstacle distance;

[0015] The preprocessed state-aware data is divided into fixed-length time window segments, and the window length and sliding step size are set.

[0016] Within each time window, the following seven statistics are calculated sequentially for the state-aware data: mean, standard deviation, maximum value, minimum value, skewness, kurtosis, and root mean square. Each statistic is used as a separate feature dimension, generating a total of seven feature dimensions.

[0017] The time-domain data is converted into a frequency-domain representation using Fourier transform, and the dominant frequency component and frequency band energy density are extracted.

[0018] An autoregressive model was used to model the historical dependencies of the time series. The model order was determined according to the AIC criterion, and the model parameters were estimated from historical observation data using the least squares method.

[0019] Feature encoding is performed on the data from the two modalities respectively:

[0020] Time series characteristics: Composed of statistical characteristics, frequency domain characteristics, and autoregressive coefficients, forming a fixed-length vector;

[0021] Image features: Spatial features are extracted using a convolutional neural network, and structured vectors are output.

[0022] After the two modalities have completed the initial encoding, they enter the fusion stage, where the temporal features and image features are vector-concatenated.

[0023] Establish cross-modal interaction relationships, define the first modality as query, the second modality as key and value, and calculate the weighted output;

[0024] After each of the two modalities has completed its independent prediction, the output results are weighted and combined.

[0025] In a preferred embodiment of the present invention, the process by which the fault diagnosis module outputs multi-label fault classification results and system health scores includes:

[0026] For each type of fault, an independent output unit is set up, and the Sigmoid function is used to map the output to the [0,1] interval:

[0027] This value represents the probability that the current sample belongs to the i-th type of fault. Multiple labels are allowed to be activated at the same time. The multi-label binary cross-entropy loss function is used to optimize the model.

[0028] Define the health score for each type of fault as the complement of its predicted probability, and sum the health scores of all fault types by weight to obtain the total health score of the system.

[0029] The system status level is determined by setting threshold ranges based on health scores:

[0030] If the score is greater than or equal to 0.8, it is marked as normal.

[0031] If the score is greater than 0.5 and less than 0.8, it is marked as a mild abnormal state;

[0032] If the score is less than or equal to 0.5, it is marked as a severe abnormal state.

[0033] In a preferred embodiment of the present invention, the process of the task impact analysis module based on fault classification results and system health scores, combined with task behavior sequences, includes:

[0034] To obtain fault classification results and system health scores, the system defines n preset types of faults during operation. The fault classification result at time point t is represented as a vector P(t), with the following form:

[0035] P(t) = [p1(t), p2(t), ..., p n (t)];

[0036] Each element p i (t) represents the probability of the i-th type of fault occurring;

[0037] Define a fault impact weight vector W, with the following form: W = [w1, w2, ..., w n ], where each element w i This represents the impact weight of the i-th type of fault, and its value ranges from [0,1].

[0038] The system's health score H(t) is an indicator of the overall health status at the current moment;

[0039] When the system runs continuously, it can obtain a set of health score sequences:

[0040] H = {H(t1), H(t2), ..., H(t)} k )};

[0041] A task action sequence is a record of actions performed by the system during task execution, arranged in chronological order. Each task action is 'a'. j Represented as a triple:

[0042] a j =(η j ,θ j ,t j ), where η j Indicates the behavior type, θ j Indicates the behavioral parameter, t j Indicates the timestamp of the action's execution;

[0043] The timestamps of all actions satisfy a strictly increasing relationship:

[0044] t1 < t2 < ... < t m ;

[0045] Each task behavior is linked to the health score and fault classification results at the corresponding time point, defining the task behavior set as:

[0046] B = {(a j ,P(t j ),H(t j ))|j=1,2,…,m}, where P(t j ) indicates at time point t j The fault classification results, H(t) j ) indicates at time point t j The system health score.

[0047] In a preferred embodiment of the present invention, the process by which the task impact analysis module uses a structural causal model to assess the degree of impact of a fault on the current task objective and outputs a task impact level assessment result includes:

[0048] Construct a structural causal model to describe the causal relationships between the following variables:

[0049] Various fault states F i System health score H, execution status of each task behavior S j and the degree of completion of the task objective, G;

[0050] The causal relationships in the model are as follows:

[0051] The system health score is determined by the probability of occurrence of various types of faults;

[0052] The execution status of each task action is determined by the system health score, as well as the type and parameters of that action;

[0053] The task objective completion rate is calculated by aggregating the execution status of all task behaviors and using a weighted average method.

[0054] Execution state S of task behavior j It is positively correlated with the system health score;

[0055] The task objective completion rate G is the weighted average of the execution status of all task behaviors;

[0056] Based on the numerical range of the task objective completion rate G, it is divided into four impact levels:

[0057] When G > 0.9, the task impact level is A;

[0058] When 0.7 < G ≤ 0.9, the task impact level is B;

[0059] When 0.4 < G ≤ 0.7, the task impact level is C;

[0060] When G≤0.4, the task impact level is D;

[0061] The final task impact assessment results include the following: the probability of occurrence p for each type of failure. i (t), system health score H(t), execution status S of each task behavior j Task objective completion rate (G) and task impact level.

[0062] In a preferred embodiment of the present invention, the process by which the collaborative task recovery module constructs a local perception map based on the positions and task states of neighboring robots includes:

[0063] In a multi-robot system, each robot constructs a local perception map based on information about other robots within its communication range;

[0064] A local perceptual graph consists of a set of nodes and a set of edges:

[0065] Node set: includes the current robot itself and all its neighboring robots;

[0066] Edge set: Represents a communication connection between two robots;

[0067] The weight w of each edge ij Defined as the Euclidean distance between two points;

[0068] Each node in the diagram corresponds to a robot entity and carries the following status information: current position, linear velocity, task status, current task number, target point coordinates, and remaining resource ratio.

[0069] In a preferred embodiment of the present invention, the process by which the collaborative task recovery module determines the task succession conditions by combining the task intent of the faulty robot with the system state includes:

[0070] A robot is marked as a faulty robot and a task handover process is triggered when any of the following conditions are met:

[0071] The communication interruption time is greater than or equal to a preset threshold.

[0072] Within a continuous time period, the distance between its current position and the task target point is greater than or equal to a set limit.

[0073] The task execution time has exceeded the set maximum allowed time;

[0074] Extract the task type, target point coordinates, task deadline, and task priority value from the historical state of the faulty robot.

[0075] Construct a candidate robot set, where robots in the set must simultaneously satisfy all of the following conditions:

[0076] The distance between the current location and the task target point is less than or equal to a set threshold;

[0077] The current task status is idle, or its task priority is less than or equal to the task to be replaced;

[0078] The remaining resource ratio is greater than or equal to the minimum resource ratio required for this task type.

[0079] In a preferred embodiment of the present invention, the process of the collaborative task recovery module outputting task allocation and collaborative action decision-making includes:

[0080] Select an optimal successor from the candidate set that minimizes the overall cost. The optimal successor is the robot that minimizes the overall cost.

[0081] After selecting a successor, perform the following operations:

[0082] Send the information about the failed task to the replacement;

[0083] Update the successor's task status to "Taking over";

[0084] Update the task status of the faulty robot to "Fault Confirmed";

[0085] Update the local perception map and synchronize it with neighboring robots;

[0086] The system scheduler records task status change information;

[0087] Based on the local perception map, its own state information, and the task intent, the successor generates the following action output through a policy function:

[0088] Control the movement to the target point;

[0089] Real-time obstacle avoidance and path adjustment;

[0090] Send a status broadcast.

[0091] Compared with the prior art, the advantages of this invention are:

[0092] (1) In this invention, the fault classification, system health score and task behavior sequence are integrated by the task impact analysis module. The fault state is identified by the multi-label model. The health score is calculated by weight. The structural causal model is constructed to depict the causal chain of "fault → health score → task execution → completion". The whole-link evaluation is realized. Historical data is bound by timestamps to enhance the temporal correlation. The task triple design is adopted to support diverse task modeling and improve flexibility. The task completion is divided into four impact levels after weighted aggregation, which quantifies the abstract impact and facilitates decision response.

[0093] (2) In this invention, lightweight distributed modeling is achieved through a collaborative task recovery module. Nodes include location, speed, task status and resource information. Edge weights reflect communication efficiency. The fault identification mechanism integrates multiple abnormal situations and extracts key task information for succession judgment. Candidate robots are selected based on distance, priority and resource conditions. The optimal successor is selected through a comprehensive cost model to complete task issuance, status update and information synchronization. The successor performs movement control, obstacle avoidance adjustment and status broadcasting based on the perception map and its own status, thereby improving the system's fault tolerance, task completion rate and operational robustness. Attached Figure Description

[0094] Figure 1 This is a system block diagram of Embodiment 1 of the present invention;

[0095] Figure 2 This is a system block diagram of Embodiment 2 of the present invention. Detailed Implementation

[0096] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0097] Example 1: As Figure 1 As shown, the present invention proposes an interconnected communication system for fault detection of road parking robots, which is applied to a parking robot communication management platform, and includes:

[0098] The data preprocessing module is used to collect the status perception data of the berthing robot and perform preprocessing operations on the collected status perception data. The preprocessing operations include data cleaning, outlier detection, timestamp alignment and format unification.

[0099] The data preprocessing module performs cleaning, outlier detection, timestamp alignment, and format standardization on the berth robot's state perception data, effectively improving data quality and removing noise and invalid information. It enables synchronization of multi-source heterogeneous data in the time dimension, supports data fusion between different sensors or systems, and ensures data format standardization, enhancing inter-system compatibility and the stability of subsequent algorithm processing, thereby improving the overall system reliability and operating efficiency.

[0100] The fault diagnosis module extracts time-series features from preprocessed state-aware data, integrates time-series and image information to identify various faults, and outputs multi-label fault classification results and system health scores.

[0101] The fault diagnosis module extracts temporal features from preprocessed state-aware data and fuses temporal and image information to identify various faults. The process includes:

[0102] Acquire preprocessed state perception data, which includes lateral acceleration, rotational acceleration, encoder pulse count, motor temperature, battery voltage, signal strength, transmission delay time, and obstacle distance;

[0103] The preprocessed state-aware data is divided into fixed-length time window segments. The window length and sliding step size are set. The data in each time window is represented as a local sequence. This process is used to transform non-fixed-length time series into discrete structures suitable for model processing.

[0104] Within each time window, the following seven statistics are calculated sequentially for the state-aware data: mean (average of all values), standard deviation (dispersion of values ​​relative to the mean), maximum (maximum observed value within the window), minimum (minimum observed value within the window), skewness (reflecting the degree of deviation of the numerical distribution from symmetry), kurtosis (reflecting the steepness or flatness of the tail of the numerical distribution), and root mean square (the square root of the average of all squared values). Each statistic is treated as a separate feature dimension, generating a total of seven feature dimensions. This set of features is provided as structured output for subsequent modules.

[0105] The time-domain data is converted into a frequency-domain representation through Fourier transform, and the dominant frequency component and frequency band energy density are extracted to characterize periodic and repetitive frequency changes.

[0106] An autoregressive model is used to model the historical dependencies of time series. The model order is determined according to the AIC criterion, and the model parameters are estimated from historical observation data using the least squares method. These parameters serve as structural features to characterize the dynamic evolution of the system.

[0107] Feature encoding is performed on the data from the two modalities respectively:

[0108] Time series characteristics: Composed of statistical characteristics, frequency domain characteristics, and autoregressive coefficients, forming a fixed-length vector;

[0109] Image features: Spatial features are extracted using a convolutional neural network, and structured vectors are output.

[0110] After the two modalities have completed the initial encoding, they enter the fusion stage, where the temporal features and image features are concatenated as vectors. Before concatenation, the features are normalized to eliminate interference caused by scale differences.

[0111] Establish cross-modal interaction relationships, define the first modality as query and the second modality as key and value, calculate weighted output, and adjust the influence weight of the second modality on the first modality through this mechanism;

[0112] After the two modalities have completed their independent predictions, the output results are combined by weighting, and the weights are automatically adjusted during the training process through the backpropagation algorithm.

[0113] The process by which the fault diagnosis module outputs multi-label fault classification results and system health scores includes:

[0114] For each type of fault, an independent output unit is set up, and the Sigmoid function is used to map the output to the [0,1] interval:

[0115] This value represents the probability that the current sample belongs to the i-th type of fault. Multiple labels are allowed to be activated at the same time. The model is optimized using a multi-label binary cross-entropy loss function, which evaluates the error for each label separately.

[0116] Define the health score for each type of fault as the complement of its predicted probability, and sum the health scores of all fault types by weight to obtain the total health score of the system. The weights are set according to the degree of influence of each type of fault on the overall system state.

[0117] The system status level is determined by setting threshold ranges based on health scores:

[0118] If the score is greater than or equal to 0.8, it is marked as normal.

[0119] If the score is greater than 0.5 and less than 0.8, it is marked as a mild abnormal state;

[0120] If the score is less than or equal to 0.5, it is marked as a severe abnormal state;

[0121] By fusing time-series and image information through the fault diagnosis module, extracting structured features and performing cross-modal fusion, multi-label fault classification and system health score output are achieved. Through time window processing, statistical and frequency domain modeling, image feature encoding and other means, the accuracy of fault identification and status assessment capabilities are improved, making it suitable for intelligent diagnosis and predictive maintenance in complex environments.

[0122] The task impact analysis module, based on the fault classification results and system health score, combined with the task behavior sequence, uses a structural causal model to assess the degree of impact of the fault on the current task objective and outputs the task impact level assessment result.

[0123] The task impact analysis module, based on fault classification results and system health scores, and combined with the task behavior sequence, includes the following processes:

[0124] To obtain fault classification results and system health scores, the system defines n preset types of faults during operation. Each type of fault exists independently or concurrently with other faults. A multi-label classification model is used to identify the fault state at the current moment. The fault classification result at time point t is represented as a vector P(t), with the following form:

[0125] P(t) = [p1(t), p2(t), ..., p n (t)];

[0126] Each element p i (t) represents the probability of occurrence of the i-th type of fault, satisfying the following condition:

[0127] 0≤p i (t)≤1, When multiple values ​​in the vector P(t) are simultaneously greater than zero, it indicates that multiple faults exist concurrently.

[0128] A fault impact weight vector W is defined to quantify the impact of various faults on the system state. The weight vector W has the following form: W = [w1, w2, ..., w n ], where each element w i This represents the impact weight of the i-th type of fault, and its value ranges from [0,1].

[0129] The system's health score H(t) is an indicator of the overall health status at the current moment, calculated using a weighted summation method: The score is within the range [0,1].

[0130] When the system runs continuously, it can obtain a set of health score sequences:

[0131] H = {H(t1), H(t2), ..., H(t)} k This sequence represents the change in the system's health status over time.

[0132] A task action sequence is a record of actions performed by the system during task execution, arranged in chronological order. Each task action is 'a'. j Represented as a triple:

[0133] a j =(η j ,θ j ,t j ), where η j Indicates the behavior type, θ j Indicates the behavioral parameter, t j Indicates the timestamp of the action's execution;

[0134] The timestamps of all actions satisfy a strictly increasing relationship:

[0135] t1 < t2 < ... < t m ;

[0136] Each task behavior is linked to the health score and fault classification results at the corresponding time point, defining the task behavior set as:

[0137] B = {(a j ,P(t j ),H(t j ))|j=1,2,…,m}, where P(t j ) indicates at time point t j The fault classification results, H(t) j ) indicates at time point t j The system health score, dataset B, provides a complete mapping between task behavior and system state, serving as the basis for subsequent causal modeling;

[0138] The process by which the task impact analysis module uses a structural causal model to assess the impact of a failure on the current task objective and outputs the task impact level assessment results includes:

[0139] Construct a structural causal model to describe the causal relationships between the following variables:

[0140] Various fault states F i System health score H, execution status of each task behavior S j and the degree of completion of the task objective, G;

[0141] The causal relationships in the model are as follows:

[0142] The system health score is determined by the probability of various types of failures occurring:

[0143] H = h(F1,F2,…,F) n );

[0144] The execution status of each task action is determined by the system health score, as well as the type and parameters of that action:

[0145] S j =g j (H,η j ,θ j );

[0146] The task objective completion rate is calculated by aggregating the execution status of all task actions using a weighted average method:

[0147] G = f(S1,S2,…,S) m );

[0148] Execution state S of task behavior jIt is positively correlated with the system health score, and the specific calculation method is as follows:

[0149] S j =λ j ·H(t j ), where λ j H(t) represents the importance coefficient of the j-th task behavior. j ) is at time point t j System health score;

[0150] The task objective completion rate G is the weighted average of the execution status of all task behaviors, and the specific formula is as follows:

[0151] Where ρ j S represents the weight of the j-th task action in relation to the overall task objective. j It represents the execution status of the j-th task action;

[0152] Based on the numerical range of the task objective completion rate G, it is divided into four impact levels:

[0153] When G > 0.9, the task impact level is A, which means that the deviation between the system's operating state and the normal state is within the allowable range, and the task completion rate is not affected by the fault.

[0154] When 0.7 < G ≤ 0.9, the task impact level is B, indicating that the system operation status is slightly affected and the task completion rate is slightly reduced.

[0155] When 0.4 < G ≤ 0.7, the task impact level is C, indicating that the system operation status is significantly affected and the task completion rate decreases significantly.

[0156] When G≤0.4, the task impact level is D, indicating that the system's operating status deviates significantly from expectations and the task cannot be completed;

[0157] The final task impact assessment results include the following:

[0158] The probability p of each type of failure i (t), that is, the vector P(t);

[0159] System health score H(t);

[0160] The execution state S of each task behavior j ;

[0161] Task objective completion rate G;

[0162] Task impact level (A / B / C / D);

[0163] By integrating fault classification results, system health scores, and task behavior sequences through a task impact analysis module, the current fault state is identified using a multi-label classification model. Combined with preset impact weights, a system health score is calculated, thereby constructing a structural causal model that clearly depicts the causal chain of "fault state → system health score → task behavior execution state → task completion rate," achieving end-to-end impact assessment from fault to task objective. Historical scores are bound to task behaviors using timestamps, forming a complete dynamic mapping dataset and enhancing the temporal correlation of the assessment. A task behavior triplet design is adopted to support diverse task modeling and improve system flexibility. Task completion rate is weighted and aggregated based on the execution states of each behavior, and then divided into four impact levels, transforming abstract impacts into actionable quantitative indicators for easier decision-making and response.

[0164] Example 2: The technical solution of this embodiment of the invention differs from that of Example 1 in that:

[0165] like Figure 2 As shown, the collaborative task recovery module constructs a local perception map based on the position and task status of neighboring robots, determines the task succession conditions by combining the task intent of the faulty robot and the system status, and outputs task allocation and collaborative action decisions.

[0166] The collaborative task recovery module constructs a local perception map based on the positions and task states of neighboring robots. The process includes:

[0167] In a multi-robot system, each robot constructs a local perception map based on information about other robots within its communication range. This map is used to support subsequent task redistribution and collaborative behavior planning. The local perception map is constructed autonomously by the current robot and only contains itself and other robot entities that can be connected within its communication range.

[0168] A local perceptual graph consists of a set of nodes and a set of edges:

[0169] Node set: includes the current robot itself and all its neighboring robots;

[0170] Edge set: Represents a communication connection between two robots;

[0171] The weight w of each edge ij Defined as the Euclidean distance between two points:

[0172] w ij =||p j -p i ||, where p i p represents the current position coordinates of robot i. j Represents the current position coordinates of robot j;

[0173] Each node in the diagram corresponds to a robot entity and carries the following state information:

[0174] Current location: 3D spatial coordinates;

[0175] Linear velocity: The velocity vector of the current movement;

[0176] Task status: can be one of the following: Idle, Running, Completed, or Faulted;

[0177] Current task number: zero if there is no task;

[0178] Target point coordinates: The target's location in three-dimensional space;

[0179] Remaining resource ratio: This represents the proportion of currently available resources to the maximum resources, ranging from [0,1].

[0180] This status information serves as the basic input for task judgment and decision-making;

[0181] The collaborative task recovery module determines the task succession conditions by combining the task intent of the faulty robot with the system status. The process includes:

[0182] A robot is marked as a faulty robot and a task handover process is triggered when any of the following conditions are met:

[0183] The communication interruption time is greater than or equal to a preset threshold.

[0184] Within a continuous time period, the distance between its current position and the task target point is greater than or equal to a set limit.

[0185] The task execution time has exceeded the set maximum allowed time;

[0186] Once any of the above conditions are met, its current task will be marked as available for replacement.

[0187] Extract the task type, target point coordinates, task deadline, and task priority value from the historical state of the faulty robot.

[0188] This information constitutes the mission intent and serves as a reference during the mission handover process;

[0189] Construct a candidate robot set, where robots in the set must simultaneously satisfy all of the following conditions:

[0190] The distance between the current location and the task target point is less than or equal to a set threshold;

[0191] The current task status is idle, or its task priority is less than or equal to the task to be replaced;

[0192] The remaining resource ratio is greater than or equal to the minimum resource ratio required for this task type;

[0193] The final candidate set consists of robots that meet all of the above conditions;

[0194] The collaborative task recovery module outputs the process of task allocation and collaborative action decision-making, which includes:

[0195] Select an optimal successor from the candidate set to minimize its overall cost. The overall cost is calculated by weighting the following four factors:

[0196] J j =α·d j +β·π j +γ·δ j +η·μ j ;

[0197] Where d j π represents the distance between the successor and the target point. j δ represents the difference between the priority of the successor's task and the priority of the task to be succeeded. j This indicates whether the substitute's task status is idle; a value of 1 indicates it is not idle, and a value of 0 indicates it is not idle. j This indicates the degree of matching between the successor's remaining resources and the resources required for the task; α, β, γ, and η are all weighting coefficients.

[0198] The optimal successor is to make J j The smallest robot;

[0199] After selecting a successor, perform the following operations:

[0200] Send the information about the failed task to the replacement;

[0201] Update the successor's task status to "Taking over";

[0202] Update the task status of the faulty robot to "Fault Confirmed";

[0203] Update the local perception map and synchronize it with neighboring robots;

[0204] The system scheduler records task status change information;

[0205] Based on the local perception map, its own state information, and the task intent, the successor generates the following action output through a policy function:

[0206] Control the movement to the target point;

[0207] Real-time obstacle avoidance and path adjustment;

[0208] Send status broadcasts to support coordinated adjustments;

[0209] The collaborative task recovery module constructs a lightweight local perception graph to achieve efficient modeling of distributed multi-robot systems. Nodes in the graph contain key information such as position, velocity, task status, and resources. Edge weights, based on Euclidean distance, reflect communication efficiency, providing a reliable basis for task redistribution. The fault identification mechanism integrates factors such as communication interruption, target deviation, and timeout execution, ensuring timely response. Task intent is extracted from faulty robots to guide the succession process. Candidate robot selection combines distance, priority matching, and resource guarantees to improve the success rate of succession. Task allocation employs a comprehensive cost model, integrating distance, priority difference, load status, and resource matching to select the optimal successor. A closed-loop process completes task distribution, status updates, and information synchronization. The successor can generate collaborative actions such as motion control, obstacle avoidance adjustments, and status broadcasting, significantly improving system fault tolerance, task completion rate, and operational stability.

[0210] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concept, should be covered within the scope of protection of the present invention.

Claims

1. An interconnected communication system for fault detection of road parking robots, applied to a parking robot communication management platform, characterized in that, include: The data preprocessing module is used to collect the status perception data of the parking robot and perform preprocessing operations on the collected status perception data. The fault diagnosis module extracts time-series features from preprocessed state-aware data, integrates time-series and image information to identify various faults, and outputs multi-label fault classification results and system health scores. The task impact analysis module, based on the fault classification results and system health score, combined with the task behavior sequence, uses a structural causal model to assess the degree of impact of the fault on the current task objective and outputs the task impact level assessment result. The collaborative task recovery module constructs a local perception map based on the position and task status of neighboring robots, combines the task intent of the faulty robot with the system status to determine the task succession conditions, and outputs task allocation and collaborative action decisions.

2. The interconnected communication system for fault detection of a road parking robot according to claim 1, characterized in that, The fault diagnosis module extracts temporal features from preprocessed state-aware data and fuses temporal and image information to identify various faults, including the following process: Acquire preprocessed state perception data, which includes lateral acceleration, rotational acceleration, encoder pulse count, motor temperature, battery voltage, signal strength, transmission delay time, and obstacle distance; The preprocessed state-aware data is divided into fixed-length time window segments, and the window length and sliding step size are set. Within each time window, the following seven statistics are calculated sequentially for the state-aware data: mean, standard deviation, maximum value, minimum value, skewness, kurtosis, and root mean square. Each statistic is used as a separate feature dimension, generating a total of seven feature dimensions. The time-domain data is converted into a frequency-domain representation using Fourier transform, and the dominant frequency component and frequency band energy density are extracted. An autoregressive model was used to model the historical dependencies of the time series. The model order was determined according to the AIC criterion, and the model parameters were estimated from historical observation data using the least squares method. Feature encoding is performed on the data from the two modalities respectively: Time series characteristics: composed of statistical characteristics, frequency domain characteristics, and autoregressive coefficients, forming a fixed-length vector; Image features: Spatial features are extracted using a convolutional neural network, and structured vectors are output. After the two modalities complete the initial encoding, they enter the fusion stage, where the temporal features and image features are vector-concatenated. Establish cross-modal interaction relationships, define the first modality as query, the second modality as key and value, and calculate the weighted output; After each of the two modalities has completed its independent prediction, the output results are weighted and combined.

3. The interconnected communication system for fault detection of a road parking robot according to claim 2, characterized in that, The process by which the fault diagnosis module outputs multi-label fault classification results and system health scores includes: For each type of fault, an independent output unit is set up, and the Sigmoid function is used to map the output to the [0,1] interval: This value represents the probability that the current sample belongs to the i-th type of fault. Multiple labels are allowed to be activated at the same time. The multi-label binary cross-entropy loss function is used to optimize the model. Define the health score for each type of fault as the complement of its predicted probability, and sum the health scores of all fault types by weight to obtain the total health score of the system. The system status level is determined by setting threshold ranges based on health scores: If the score is greater than or equal to 0.8, it is marked as normal. If the score is greater than 0.5 and less than 0.8, it is marked as a mild abnormal state; If the score is less than or equal to 0.5, it is marked as a severe abnormal state.

4. The interconnected communication system for fault detection of a road parking robot according to claim 1, characterized in that, The task impact analysis module, based on fault classification results and system health scores, and combined with the task behavior sequence, includes the following process: To obtain fault classification results and system health scores, the system defines n preset types of faults during operation. The fault classification result at time point t is represented as a vector P(t), with the following form: P(t)=[p1(t),p2(t),…,p n (t)]; Each element p i (t) represents the probability of the i-th type of fault occurring; Define a fault impact weight vector W, with the following form: W = [w1, w2, ..., w n ], where each element w i This represents the impact weight of the i-th type of fault, and its value ranges from [0,1]. The system's health score H(t) is an indicator of the overall health status at the current moment; When the system runs continuously, it can obtain a set of health score sequences: H={H(t1),H(t2),…,H(t k )}; A task action sequence is a record of actions performed by the system during task execution, arranged in chronological order. Each task action is 'a'. j Represented as a triple: a j =(η j ,θ j ,t j ), where η j Indicates the behavior type, θ j Indicates the behavioral parameter, t j Indicates the timestamp of the action's execution; The timestamps of all actions satisfy a strictly increasing relationship: t1<t2<…<t m ; Each task behavior is linked to the health score and fault classification results at the corresponding time point, defining the task behavior set as: B = {(a j ,P(t j ),H(t j ))|j=1,2,…,m}, where P(t j ) indicates at time point t j The fault classification results, H(t) j ) indicates at time point t j The system health score.

5. The interconnected communication system for fault detection of a road parking robot according to claim 4, characterized in that, The process by which the task impact analysis module uses a structural causal model to assess the degree of impact of a failure on the current task objective and outputs the task impact level assessment result includes: Construct a structural causal model to describe the causal relationships between the following variables: Various fault states F i System health score H, execution status of each task behavior S j and the degree of completion of the task objective, G; The causal relationships in the model are as follows: The system health score is determined by the probability of occurrence of various types of faults; The execution status of each task action is determined by the system health score, as well as the type and parameters of that action; The task objective completion rate is calculated by aggregating the execution status of all task behaviors and using a weighted average method. Execution status S of task behavior j It is positively correlated with the system health score; The task objective completion rate G is the weighted average of the execution status of all task behaviors; Based on the numerical range of the task objective completion rate G, it is divided into four impact levels: When G > 0.9, the task impact level is A; When 0.7 < G ≤ 0.9, the task impact level is B; When 0.4 < G ≤ 0.7, the task impact level is C; When G≤0.4, the task impact level is D; The final task impact assessment results include the following: the probability of occurrence p for each type of failure. i (t), system health score H(t), execution status S of each task behavior j Task objective completion rate (G) and task impact level.

6. The interconnected communication system for fault detection of a road parking robot according to claim 1, characterized in that, The process by which the collaborative task recovery module constructs a local perception map based on the positions and task states of neighboring robots includes: In a multi-robot system, each robot constructs a local perception map based on information about other robots within its communication range; A local perceptual graph consists of a set of nodes and a set of edges: Node set: includes the current robot itself and all its neighboring robots; Edge set: Represents a communication connection between two robots; The weight w of each edge ij Defined as the Euclidean distance between two points; Each node in the diagram corresponds to a robot entity and carries the following status information: current position, linear velocity, task status, current task number, target point coordinates, and remaining resource ratio.

7. The interconnected communication system for fault detection of a road parking robot according to claim 6, characterized in that, The process by which the collaborative task recovery module determines the task succession conditions by combining the task intent of the faulty robot with the system status includes: A robot is marked as a faulty robot and a task handover process is triggered when any of the following conditions are met: The communication interruption time is greater than or equal to a preset threshold. Within a continuous time period, the distance between its current position and the task target point is greater than or equal to a set limit. The task execution time has exceeded the set maximum allowed time; Extract the task type, target point coordinates, task deadline, and task priority value from the historical state of the faulty robot. Construct a candidate robot set, where robots in the set must simultaneously satisfy all of the following conditions: The distance between the current location and the task target point is less than or equal to a set threshold; The current task status is idle, or its task priority is less than or equal to the task to be replaced; The remaining resource ratio is greater than or equal to the minimum resource ratio required for this task type.

8. The interconnected communication system for fault detection of a road parking robot according to claim 7, characterized in that, The process of outputting task allocation and collaborative action decision-making by the collaborative task recovery module includes: Select an optimal successor from the candidate set that minimizes the overall cost. The optimal successor is the robot that minimizes the overall cost. After selecting a successor, perform the following operations: Send the information about the failed task to the replacement. Update the successor's task status to "Taking over"; Update the task status of the faulty robot to "Fault Confirmed"; Update the local perception map and synchronize it with neighboring robots; The system scheduler records task status change information; Based on the local perception map, its own state information, and the task intent, the successor generates the following action output through a policy function: Control the movement to the target point; Real-time obstacle avoidance and path adjustment; Send a status broadcast.

Images