Rail brake device failure prediction method and system

Abstract

The application relates to the technical field of traffic failure prediction, in particular to a rail brake equipment failure prediction method and system, wherein the method comprises the following steps: acquiring rail brake equipment operation state data; constructing a dynamic failure correlation graph to obtain failure transmission relationships among brake equipment components; dynamically adjusting the edge weight of the dynamic failure correlation graph according to the rail brake equipment operation state data, calculating the PageRank value of the nodes of the dynamic failure correlation graph to identify key components, and finding a target failure propagation path by using a Dijkstra algorithm; constructing a multi-task predictive maintenance decision tree, inputting the rail brake equipment operation state data, the key components and the target failure propagation path into the multi-task predictive maintenance decision tree, and outputting multi-dimensional prediction results through multi-layer decision; and generating a priority maintenance suggestion according to the multi-dimensional prediction results. Thus, the problems of neglecting component dynamic changes, splitting component relationships and lacking maintenance decision support in the prior art are solved.

Description

Technical Field

[0001] This application relates to the field of traffic fault prediction technology, and in particular to a method and system for predicting faults in rail braking equipment. Background Technology

[0002] Currently, the maintenance model for track braking equipment still mainly relies on periodic inspections and reactive maintenance. Periodic inspections involve comprehensive checks and maintenance of the equipment according to fixed cycles. This approach ignores the differences in actual operation of individual equipment, which can easily lead to over-maintenance and waste of resources, or under-maintenance and the creation of safety hazards. Reactive maintenance is a remedial measure after a failure has occurred. Although it can solve the problem after the failure has occurred, train delays and operational interruptions are inevitable when a failure occurs, and it may even lead to serious safety accidents such as rear-end collisions and derailments, causing huge losses to society and the economy.

[0003] With the development of information technology and data science, data-driven fault prediction methods, including expert systems and machine learning, have begun to emerge in the rail transit field. However, some methods focus only on the detection of a single component or a certain type of fault, severing the complex fault propagation chain between braking equipment components and failing to accurately predict the development trend of faults. Some methods use static models, which are difficult to adapt to the dynamic process of frequent changes in operating conditions and continuous performance degradation during equipment operation. Moreover, most existing prediction results only provide fault warning information, lacking in-depth support for maintenance decisions and failing to accurately guide the rational allocation of maintenance resources and the scientific determination of maintenance priorities. Summary of the Invention

[0004] This application provides a method and system for predicting faults in rail braking equipment, in order to solve problems in the prior art such as ignoring the dynamic changes of components, severing the relationship between components, and lacking maintenance decision support.

[0005] The first aspect of this application provides a method for predicting faults in rail braking equipment, comprising the following steps: acquiring operational status data of the rail braking equipment; constructing a dynamic fault association graph to obtain the fault transmission relationship between components of the braking equipment, wherein constructing the dynamic fault association graph includes abstracting the physical components of the rail braking equipment as graph nodes, constructing directed edges based on physical connection relationships, energy transfer paths, and control signal flows, and setting initial weights according to historical fault statistics; dynamically adjusting the edge weights of the dynamic fault association graph according to the operational status data of the rail braking equipment, calculating the PageRank value of the nodes of the dynamic fault association graph to identify key components, and using the Dijkstra algorithm to find the target fault propagation path; constructing a multi-task predictive maintenance decision tree, inputting the operational status data of the rail braking equipment, the key components, and the target fault propagation path into the multi-task predictive maintenance decision tree, and outputting multi-dimensional prediction results through multi-level decision-making, wherein the multi-dimensional prediction results include a comprehensive risk index, faulty components, fault types and corresponding probabilities, expected occurrence time, and affected range; and generating priority maintenance suggestions based on the multi-dimensional prediction results.

[0006] Optionally, identifying key components by calculating the PageRank values ​​of the nodes in the dynamic fault association graph includes: in the dynamic fault association graph, the PageRank value represents the influence of a component in the fault propagation network; calculating the PageRank value of each component in turn; calculating the mean and standard deviation based on the PageRank values ​​of each component; if the PageRank value of a component is greater than the sum of the mean and 1.5 times the standard deviation, then the component is determined to be a core component; otherwise, if the PageRank value of a component is greater than the mean, then the component is determined to be an important component; otherwise, it is a general component; and the core components and important components are obtained as key components.

[0007] Optionally, the PageRank value is calculated using the following formula:

[0008]

[0009] in, This indicates the component node to be calculated. Indicates pointing to a node The set of nodes, Indicates from node To the node The edge weights, where d is the damping coefficient;

[0010] Optionally, the step of using Dijkstra's algorithm to find the target fault propagation path includes: using Dijkstra's algorithm, starting from the key component node, searching for the fault propagation path with the largest sum of weights as the target fault propagation path, wherein the specific steps are as follows:

[0011] S1. Remove edges whose weights are below the target threshold;

[0012] S2. Initialize the distance array, the visited set, and the predecessor node array. The distance array is used to store the cumulative weight of the maximum probability path from the starting node S to each node. Initially, the dist value of the starting node is set to 1, and the dist value of the other nodes is set to 0. The visited set is used to mark the nodes whose maximum probability path has been determined. The predecessor node array is used to store the predecessor node of each node on the maximum probability path.

[0013] S4. From all unvisited nodes, select the node U with the largest dist value and add it to the visited set;

[0014] S5. For all neighboring nodes V of node U, calculate the cumulative propagation probability from S through U to V. The specific calculation formula is as follows:

[0015]

[0016] in, Let U be the edge weight from node U to V;

[0017] S6. If the cumulative propagation probability is greater than the dist value of the adjacent node V, then update the dist value of the adjacent node V to the cumulative propagation probability, and record the predecessor node of V as U in the predecessor node array.

[0018] S7. Repeat steps S4-S6 until all reachable nodes have been processed.

[0019] S8. Starting from the last node, backtrack along the predecessor node array to the starting point, and after reversing, obtain the target fault propagation path.

[0020] Optionally, the multi-task predictive maintenance decision tree includes: an input layer, a shared feature layer, task-specific branches, and an output fusion layer. The input layer is used to receive track braking equipment operating status data, key component information, and fault propagation paths. The shared feature layer is used to extract common feature representations from the input data and filter key features. The task-specific branches are used to construct multiple branches to execute prediction tasks respectively. The output fusion layer is used to integrate the prediction results of each branch and apply decision rules to generate the final multi-dimensional prediction result.

[0021] Optionally, the task-specific branches include: a fault component prediction branch, a fault classification branch, a fault time prediction branch, and an impact range assessment branch.

[0022] Optionally, the step of inputting the operating status data of the track braking equipment, the key components, and the target fault propagation path into the multi-task predictive maintenance decision tree, and outputting multi-dimensional prediction results through multi-level decision-making, includes:

[0023] Faulty Component Prediction Branch: Based on threshold-detected anomaly parameters, calculate the anomaly score of the component and output a list of key anomaly components. The formula for calculating the anomaly score is as follows:

[0024]

[0025] Where v represents the component, and n represents the number of parameters of the component. These are the importance weights of the parameters. It is the deviation of the i-th parameter of component v. It is part v value;

[0026] Fault classification branch: Use a random forest classifier to output the fault type with the highest probability and its occurrence probability;

[0027] Failure time prediction branch: Extract parameter degradation features, use LSTM network to predict parameter evolution trends, and predict the expected time of failure;

[0028] Impact Scope Assessment Branch: Calculates the impact score based on the fault propagation path and outputs a list of affected components and their degree of impact;

[0029] A second aspect of this application provides a fault prediction system for track braking equipment, comprising: a data acquisition module, a fault association module, an association range calculation module, a multi-task prediction module, and a maintenance suggestion generation module. The data acquisition module acquires track braking equipment operating status data. The fault association module constructs a dynamic fault association graph to obtain the fault transmission relationship between braking equipment components and dynamically adjusts the edge weights of the dynamic fault association graph based on the track braking equipment operating status data. The association range calculation module calculates the PageRank values ​​of the nodes in the dynamic fault association graph to identify key components and uses the Dijkstra algorithm to find the target fault propagation path. The multi-task prediction module constructs a multi-task predictive maintenance decision tree, inputting the track braking equipment operating status data, the key components, and the target fault propagation path into the multi-task predictive maintenance decision tree, and outputs multi-dimensional prediction results after multi-level decision processing. The multi-dimensional prediction results include a comprehensive risk index, faulty components, fault type and corresponding probability, expected occurrence time, and affected area. The maintenance suggestion generation module generates priority maintenance suggestions based on the multi-dimensional prediction results.

[0030] A third aspect of this application provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being configured to perform a track braking device fault prediction method as described in the above embodiments.

[0031] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a method for predicting faults in a track braking device as described in the above embodiments.

[0032] The beneficial effects achieved by using the present invention described above are as follows:

[0033] This application's embodiments construct a dynamic fault association graph based on physical connections, energy transfer, and control signal flow by comprehensively collecting equipment operating status data. This graph accurately depicts the complex fault transmission relationships between components and dynamically adjusts edge weights based on real-time data, effectively reflecting dynamic changes during equipment operation. PageRank values ​​are used to quickly locate key components, and Dijkstra's algorithm is employed to efficiently pinpoint fault propagation paths, significantly improving the accuracy and efficiency of fault prediction. A multi-task predictive maintenance decision tree deeply integrates equipment operating data, key component information, and fault path characteristics. Through multi-layered decision output, it provides multi-dimensional results covering faulty components, types, probabilities, expected occurrence times, and affected areas, achieving a comprehensive and three-dimensional understanding of potential faults. Priority maintenance recommendations generated based on multi-dimensional predictions comprehensively consider the degree of fault risk, urgency, and impact scope, scientifically allocating maintenance resources and precisely planning maintenance tasks. This avoids the waste of resources from periodic maintenance and eliminates operational interruptions and safety hazards caused by post-maintenance, improving equipment reliability and effectively ensuring the safety and efficiency of rail transit operations. It also promotes the intelligent and refined operation and maintenance of rail braking equipment. This solves the problems in existing technologies, such as ignoring the dynamic changes of components, severing the relationships between components, and lacking support for maintenance decisions.

[0034] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0035] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0036] Figure 1 This is a flowchart of a fault prediction method for track braking equipment provided according to an embodiment of this application;

[0037] Figure 2This is a schematic diagram of a fault prediction system for a track braking device provided according to an embodiment of this application;

[0038] Figure 3 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0039] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0040] The following description, with reference to the accompanying drawings, illustrates a method and system for predicting faults in rail braking equipment according to an embodiment of this application. Addressing the problems mentioned in the background section regarding prior art, such as neglecting dynamic changes in components, severing relationships between components, and lacking maintenance decision support, this application provides a method for predicting faults in rail braking equipment. This method comprehensively collects equipment operating status data to construct a dynamic fault correlation graph based on physical connections, energy transfer, and control signal flow. This graph accurately depicts the complex fault transmission relationships between components and dynamically adjusts edge weights based on real-time data, effectively reflecting dynamic changes during equipment operation. It utilizes PageRank values ​​to quickly locate key components and leverages the Dijkstra algorithm to efficiently pinpoint fault propagation paths, significantly improving the accuracy and efficiency of fault prediction. Multi-task prediction is also included. The pre-maintenance decision tree deeply integrates equipment operation data, key component information, and fault path characteristics. Through multi-layered decision-making, it outputs multi-dimensional results covering faulty components, types, probabilities, estimated occurrence times, and affected areas, achieving a comprehensive and three-dimensional understanding of potential faults. Prioritized maintenance recommendations generated based on multi-dimensional predictions comprehensively consider the degree of fault risk, urgency, and impact, scientifically allocating maintenance resources and precisely planning maintenance tasks. This avoids the waste of resources from periodic inspections and eliminates operational interruptions and safety hazards caused by reactive maintenance, improving equipment reliability and effectively ensuring the safety and efficiency of rail transit operations. It also promotes the intelligent and refined operation and maintenance of rail braking equipment. Therefore, it solves the problems of existing technologies such as neglecting dynamic changes in components, severing relationships between components, and lacking maintenance decision support.

[0041] Specifically, Figure 1 This is a flowchart illustrating a method for predicting faults in a track braking device, as provided in an embodiment of this application.

[0042] like Figure 1 As shown, the fault prediction method for the track braking equipment includes the following steps:

[0043] In step S101, the operating status data of the track braking device is acquired.

[0044] The track braking equipment includes various brake valves, air compressors, air cylinders, brake cylinders, basic braking devices (brake levers, connecting rods, brake shoes), traction motors, brake resistor cabinets, converters, control contactors, brake control units, etc.

[0045] The operating status data of the rail braking equipment includes pressure data, temperature data, displacement data, and vibration data, which are acquired through corresponding sensors.

[0046] It is understood that the embodiments of this application construct a three-dimensional monitoring network for the operating status of the rail braking equipment by acquiring operating status data through sensors. Specifically, pressure data can monitor the braking power transmission status in real time, promptly detect abnormal pressure fluctuations, and avoid braking failure due to insufficient or excessive braking pressure; temperature data can keenly capture component temperature changes, quickly locate abnormal heat points, and prevent performance degradation or safety hazards caused by excessive component wear or poor heat dissipation; displacement data accurately presents the working position status of components, effectively avoiding abnormal braking force output due to component position deviations; vibration data, by analyzing subtle changes in frequency and amplitude, can identify potential faults such as component loosening and bearing wear at an early stage.

[0047] In step S102, a dynamic fault association graph is constructed to obtain the fault transmission relationship between the components of the braking equipment. The construction of the dynamic fault association graph includes abstracting the physical components of the track braking equipment into graph nodes, constructing directed edges based on physical connection relationships, energy transfer paths, and control signal flows, and setting initial weights according to historical fault statistics.

[0048] Specifically, the physical components of the track braking device are abstracted as graph nodes, where the node attributes include: component identifier, component type, current operating status, and real-time detection parameters;

[0049] Directed edges are constructed based on physical connections, energy transfer paths, and control signal flows. Physical connections represent direct mechanical connections, energy transfer paths represent energy flow paths, including hydraulic energy, electrical energy, and mechanical energy, and control signal flows represent the transmission paths of control commands or feedback signals. The attributes of directed edges include edge type and weight, where edge type includes three types: physical connections, energy transfer paths, and control signal flows.

[0050] Each edge is assigned an initial weight based on historical fault statistics, and the weight reflects the probability of fault propagation.

[0051] For example, components such as brake valves, air compressors, and brake cylinders are abstracted as nodes. For instance, a brake valve node records its identifier "BV-01," component type "control class," current "open state," and real-time pressure detection parameters. Based on physical connections, a directed edge is constructed between the brake cylinder and the basic braking device to represent the transmission of mechanical force; an edge is constructed between the air compressor and the main air cylinder according to the energy transmission path to represent the flow of compressed air; and an edge is constructed between the brake valve and the brake control unit according to the control signal flow to transmit braking commands. Based on historical statistics, there have been 10 cases where brake valve failure led to abnormal brake cylinder pressure. A high initial weight of 0.8 is assigned to the edge connecting these two components to reflect the probability of fault propagation.

[0052] It is understood that the embodiments of this application, by constructing a dynamic fault association graph, abstract the components of the track braking equipment into nodes and assign them rich attributes. Directed edges are constructed based on multi-dimensional relationships, and initial weights are set using historical data. This allows for a systematic and visual presentation of the complex fault transmission relationships between equipment components. Node attributes enable real-time tracking of component operating status and rapid anomaly location. Based on physical, energy, and signal edge relationships, fault propagation paths can be deeply explored, overcoming the limitations of traditional methods that only focus on single connections. Assigning initial weights using historical data provides a quantitative basis for fault propagation probability, making predictions more scientific.

[0053] In step S103, the edge weights of the dynamic fault association graph are dynamically adjusted based on the operating status data of the track braking equipment, the PageRank values ​​of the nodes in the dynamic fault association graph are calculated to identify key components, and the Dijkstra algorithm is used to find the target fault propagation path.

[0054] Among them, threshold detection is performed on the operating status data of the track braking equipment, the real-time operating status data is compared with the normal range of preset parameters, the current abnormal score is calculated, and the edge weights are dynamically adjusted according to the abnormal score.

[0055] It should be noted that the formula for calculating the anomaly score is:

[0056]

[0057]

[0058] Where n represents the number of operating status parameters of the track braking equipment components. For parameter weights, Let f be the anomaly score of the i-th parameter, and f be a non-linear mapping function. This is the real-time monitoring value of the i-th operating status parameter of this component. This represents the normal mean value of the i-th operating state parameter of the component. This represents the normal standard deviation of the i-th operating state parameter of the component.

[0059] The formula for dynamically adjusting edge weights is:

[0060]

[0061] in, The edge weight from component i to component j at time t. The overall anomaly score for component i. To adjust the coefficient, Edge type correction factor, different values ​​are assigned to different edge types.

[0062] It is understood that the embodiments of this application enable the fault association graph to reflect the equipment operating status in real time through a dynamic weight adjustment mechanism. By converting sensor data into edge weight changes, the dynamic characteristics of fault propagation are accurately captured. The PageRank algorithm is used to analyze the importance of nodes in the graph, quickly locate core components such as brake valves and control units, and prioritize maintenance resources to high-risk areas. The Dijkstra algorithm is used to efficiently find the most likely fault propagation path, providing data support for subsequently determining the fault impression range.

[0063] In this embodiment of the application, identifying key components by calculating the PageRank value of nodes in a dynamic fault association graph includes:

[0064] In a dynamic fault association graph, the PageRank value represents the influence of a component in the fault propagation network. The PageRank value of each component is calculated in turn, and the mean and standard deviation are calculated based on the PageRank values ​​of each component. If the PageRank value of a component is greater than the sum of the mean and 1.5 times the standard deviation, the component is determined to be a core component. Otherwise, if the PageRank value of a component is greater than the mean, the component is determined to be an important component. Otherwise, it is a general component.

[0065] Acquire core and important components as key components.

[0066] For example, a dynamic fault correlation diagram contains 20 component nodes, including brake valve, air compressor, brake cylinder, brake control unit, and basic braking device. By calculating the PageRank value of each node, the mean is 0.05 and the standard deviation is 0.03.

[0067] The PageRank value of the brake control unit is 0.12, which is significantly greater than the mean plus 1.5 times the standard deviation (0.05 + 1.5 × 0.03 = 0.095). Therefore, it is identified as a core component. The brake control unit is the control hub between the electrical signals and the air brakes, connecting multiple components such as the brake valve and anti-skid device. Its failure has the greatest impact. The PageRank value of the brake valve is 0.07, which is between the mean (0.05) and 0.095. Therefore, it is identified as an important component. As the core component for the driver to control the brakes, its failure will directly affect the transmission of braking commands. The PageRank value of the basic braking device is 0.04, which is lower than the mean. Therefore, it is identified as a general component. Failures in the basic braking device are usually local wear and have a small impact on the overall system.

[0068] It is understood that the embodiments of this application calculate the influence score of components in the dynamic fault association graph using the PageRank algorithm, and combine it with a statistical screening strategy based on the mean and standard deviation to systematically classify rail braking equipment components into core, important, and general components. This breaks through the limitations of traditionally relying on experience to judge key components, and objectively reflects the topological importance of components in the fault propagation network using a graph theory model. Core components, because they are located at the hub of fault propagation, have PageRank values ​​significantly higher than the mean and standard deviation thresholds, and their failure can easily trigger a chain of faults. Important components play a key role in local fault propagation paths. By classifying core and important components as key components, we can focus on high-risk components and invest more resources in links that have a significant impact on system safety.

[0069] In this embodiment of the application, the formula for calculating the PageRank value is:

[0070]

[0071] in, This indicates the component node to be calculated. Indicates pointing to a node The set of nodes, Indicates from node To the node The edge weights, where d is the damping coefficient;

[0072] In this embodiment of the application, finding the target fault propagation path using Dijkstra's algorithm includes: using Dijkstra's algorithm, starting from the critical component node, searching for the fault propagation path with the largest sum of weights as the target fault propagation path, wherein the specific steps are as follows:

[0073] S1. Remove edges whose weights are below the target threshold;

[0074] S2. Initialize the distance array, the visited set, and the predecessor node array. The distance array is used to store the cumulative weight of the maximum probability path from the starting node S to each node. Initially, the dist value of the starting node is set to 1, and the dist value of the other nodes is set to 0. The visited set is used to mark the nodes whose maximum probability path has been determined. The predecessor node array is used to store the predecessor node of each node on the maximum probability path.

[0075] S4. From all unvisited nodes, select the node U with the largest dist value and add it to the visited set;

[0076] S5. For all neighboring nodes V of node U, calculate the cumulative propagation probability from S through U to V. The specific calculation formula is as follows:

[0077]

[0078] in, Let U be the edge weight from node U to V;

[0079] S6. If the cumulative propagation probability is greater than the dist value of the adjacent node V, then update the dist value of the adjacent node V to the cumulative propagation probability, and record the predecessor node of V as U in the predecessor node array.

[0080] S7. Repeat steps S4-S6 until all reachable nodes have been processed.

[0081] S8. Starting from the last node, backtrack along the predecessor node array to the starting point, and after reversing, obtain the target fault propagation path.

[0082] For example, a dynamic fault association graph contains nodes A (brake valve), B (brake cylinder), C (brake control unit), and D (basic braking device). The edge weight from node A to node B is 0.8, the edge weight from node A to node C is 0.5, and the edge weight from node B to node D is 0.7.

[0083] Taking node A as the key component, the steps for calculating the target path are as follows:

[0084] Remove edges with weights below the target threshold of 0.3, and retain the remaining edges. , , D.

[0085] Initialize the distance array by setting the dist value of node A to 1 and the dist value of the other nodes to 0.

[0086] Initialize the visited set and the predecessor node array, setting them to empty.

[0087] Select node A with the largest dist value and add it to the visited set. Then calculate the cumulative propagation probability from node A to its neighboring nodes B and C, where the cumulative propagation probability from node A to neighboring node B is... The cumulative propagation probability from node A to its neighboring node C is Update the dist values ​​of nodes B and C respectively. And 0.5, record the predecessor node of B as A and the predecessor node of C as A in the predecessor node array.

[0088] Among the unvisited nodes, node B has the largest dist value, so node B is added to the visited set.

[0089] The cumulative propagation probability from node B to its neighboring node D is calculated as follows: Since 0.56 is greater than 0, the dist value of node D is updated to 0.56, and the predecessor node of D is recorded as B in the predecessor node array.

[0090] Tracing back from node D to node A, the target path A→B→D is obtained, with a cumulative propagation probability of 0.56.

[0091] It is understood that the embodiments of this application use Dijkstra's algorithm to start from the key components and search for the fault propagation path with the largest sum of weights as the target path. This can make full use of the quantitative expression of the fault propagation probability by the edge weights in the dynamic fault association graph and accurately locate the path most likely to spread the fault.

[0092] In step S104, a multi-task predictive maintenance decision tree is constructed. The operating status data of the track braking equipment and the propagation paths of key components and target faults are input into the multi-task predictive maintenance decision tree, and multi-dimensional prediction results are output after multi-level decision-making.

[0093] The multidimensional prediction results include a comprehensive risk index, faulty components, fault type and corresponding probability, expected occurrence time, and affected area.

[0094] It is understood that the embodiments of this application deeply integrate real-time equipment operating status data, key component information and fault propagation path. Through a multi-layer decision structure, it can not only accurately identify components and types that may fail, but also quantify the probability of failure, estimate the time of occurrence and define the scope of impact. It can grasp the full picture of the fault in advance, conduct systematic analysis from the root cause of the fault to the scope of impact, and thus formulate more targeted maintenance strategies and optimize the scheduling of maintenance resources.

[0095] In this embodiment, the multi-task predictive maintenance decision tree includes: an input layer, a shared feature layer, task-specific branches, and an output fusion layer. The input layer is used to receive track braking equipment operating status data, key component information, and fault propagation paths. The shared feature layer is used to extract common feature representations from the input data and filter key features. The task-specific branches are used to construct multiple branches to execute prediction tasks respectively. The output fusion layer is used to integrate the prediction results of each branch and apply decision rules to generate the final multi-dimensional prediction result.

[0096] The output fusion layer integrates the prediction results from each branch, calculates the comprehensive risk index, sorts them according to risk priority, and outputs a priority maintenance prediction list. Each maintenance prediction list includes the comprehensive risk index, the faulty component, the fault type and its corresponding probability, the expected occurrence time, and the affected area. The formula for calculating the comprehensive risk index is:

[0097]

[0098] in, The degree of urgency is determined by the scope of the affected area and by the expected time of occurrence.

[0099] For example, operating status data such as pressure and temperature, key component information of the brake cylinder and air compressor, and two fault propagation paths, "brake cylinder → brake line → brake shoe" and "air compressor → air cylinder → brake valve," are input into a multi-task predictive maintenance decision tree. After receiving the data, the input layer extracts common features such as pressure fluctuations and temperature change trends, and filters out key features such as abnormal pressure amplitude and temperature rise rate. In the task-specific branches, the fault component prediction branch determines that the brake cylinder has an abnormal score of 15 and the air compressor has an abnormal score of 8, adds them to the list of key abnormal components, and outputs them; the fault classification branch calculates the probability of "leakage" in the brake cylinder as 80% and the probability of "wear exceeding limit" in the air compressor as 75% using random forest; the fault time prediction branch shows that the remaining expected time of occurrence for the brake cylinder is 36 hours and the expected time of occurrence for the air compressor is 60 hours; the impact range assessment branch concludes that the brake cylinder failure will moderately affect the brake line and slightly affect the brake shoe, while the air compressor failure will slightly affect the air cylinder and brake valve. The output fusion layer integration results are substituted into the comprehensive risk index formula for calculation. The comprehensive risk index for the brake cylinder is 0.384, and the comprehensive risk index for the air compressor is 0.135. A maintenance prediction list is output according to risk priority. The brake cylinder is ranked first due to its higher risk, including a comprehensive risk index of 0.384, the faulty component being the brake cylinder, the fault type being a leakage probability of 80%, an estimated occurrence time of 36 hours, and an affected area of ​​(moderate) brake lines and (minor) brake shoes. The air compressor is ranked second, with its corresponding prediction results displayed.

[0100] It is understood that the embodiments of this application achieve centralized reception of multi-source data through the input layer, providing comprehensive information for subsequent analysis; the shared feature layer effectively reduces data redundancy and enhances the model's ability to capture core information by extracting common features and filtering key elements; the task-specific branch constructs independent and parallel analysis paths for different prediction targets such as faulty components, fault types, probabilities, times, and impact ranges, achieving specialized prediction; the output fusion layer integrates the results of each branch, comprehensively weighs them using decision rules, and outputs complete and accurate multi-dimensional predictions. The division of labor and cooperation among each layer avoids redundant calculations and reduces computational complexity, while improving prediction timeliness through multi-task parallelism. Simultaneously, the fusion mechanism ensures the accuracy and reliability of the results, providing solid technical support for the intelligent operation and maintenance of track braking equipment.

[0101] In this embodiment, the task-specific branches include: a fault component prediction branch, a fault classification branch, a fault time prediction branch, and an impact range assessment branch.

[0102] It is understood that by dividing the present application into multiple task-specific branches, each with a clear division of labor and handling prediction tasks of different dimensions, the complexity of track braking equipment fault prediction can be addressed in a refined manner, significantly improving the accuracy of prediction and the effectiveness of decision-making.

[0103] In this embodiment, the operating status data of the track braking equipment and the propagation paths of key components and target faults are input into a multi-task predictive maintenance decision tree. After multi-level decision-making, multi-dimensional prediction results are output, including:

[0104] Faulty Component Prediction Branch: Based on threshold-detected anomaly parameters, calculates the anomaly score for each component and outputs a list of key anomaly components. The anomaly score calculation formula is as follows:

[0105]

[0106] Where v represents the component, and n represents the number of parameters of the component. These are the importance weights of the parameters. It is the deviation of the i-th parameter of component v. It is part v value;

[0107] Fault classification branch: Use a random forest classifier to output the fault type with the highest probability and its occurrence probability;

[0108] Failure time prediction branch: Extract parameter degradation features, use LSTM network to predict parameter evolution trends, and predict the expected time of failure;

[0109] Impact Scope Assessment Branch: Calculates the impact score based on the fault propagation path and outputs a list of affected components and their degree of impact;

[0110] Specifically, in the random forest classifier, the average predicted probability of all decision trees for each fault type is calculated, and the fault type with the highest probability and its probability are output. Fault types include wear exceeding limits, deformation exceeding limits, loosening, leakage, aging, and failure. The prediction probability formula for the random forest classifier is:

[0111]

[0112] Where T represents the number of decision trees, Indicates the fault type of the t-th tree. The predicted probability, Fault type The integrated predicted probability.

[0113] Extracting parameter degradation features involves calculating the degradation rate of the operating parameters. The formula for calculating the degradation rate is:

[0114]

[0115] in, The current time represents the running parameter value, and the baseline value is the parameter value from the previous time window. The time window is the duration of the time period.

[0116] The impact score is calculated based on the fault propagation path, specifically by multiplying the path weight of each node in the fault propagation path, its corresponding PageRank value, and the cumulative propagation probability of the fault propagation path.

[0117] It is understood that the embodiments of this application integrate data and algorithms from different dimensions through a multi-task decision tree, capture equipment anomalies using real-time monitoring data, improve prediction accuracy by combining historical data patterns, and can detect potential faults and the scope of their impact in advance. Based on the maintenance ranking of the comprehensive risk index, resource allocation can be optimized, and equipment reliability and operational efficiency can be improved.

[0118] In step S105, priority maintenance recommendations are generated based on the multidimensional prediction results.

[0119] Specifically, the multidimensional prediction results are priority maintenance prediction lists, where each maintenance prediction list includes a comprehensive risk index, faulty component, fault type and corresponding probability, expected occurrence time, and affected scope.

[0120] Maintenance levels are determined based on a comprehensive risk index. If the comprehensive risk index is greater than 0.5, it is an emergency repair, requiring immediate shutdown for maintenance. If the comprehensive risk index is between 0.35 and 0.5, it is a high-priority maintenance, requiring maintenance to be arranged within 12 hours. If the comprehensive risk index is between 0.2 and 0.35, it is a medium-priority maintenance, requiring maintenance to be arranged within 24 hours. If the comprehensive risk index is less than 0.2, it is a low-priority maintenance, requiring maintenance to be arranged within 72 hours.

[0121] According to the set template, the comprehensive risk index, faulty components, fault type and corresponding probability, expected occurrence time and affected scope are filled into the template to generate a maintenance reference report, providing maintenance personnel with a basis for maintenance.

[0122] It is understood that the embodiments of this application use a quantitative comprehensive risk index to classify maintenance levels, which can clearly define the handling priority of different faults; the standardized report template integrates complex fault information into structured content, so that maintenance personnel can quickly obtain key information without repeatedly sorting through it, and can prepare spare parts and allocate professional personnel in advance, which can significantly shorten maintenance preparation time.

[0123] The following is a detailed description of a fault prediction method for track braking equipment through a specific embodiment:

[0124] The subway operation and maintenance center collects operational status data of the train's braking equipment components when braking occurs on a certain line of trains. The operational status data includes brake cylinder pressure, air cylinder pressure, hydraulic line pressure, brake disc temperature, motor temperature, hydraulic oil temperature, brake cylinder piston displacement, brake shoe clearance, air compressor vibration, and brake caliper vibration data.

[0125] The physical components of the train braking system are abstracted as graph nodes, including brake valves, air compressors, air cylinders, brake cylinders, basic braking devices (brake levers, connecting rods, brake shoes), traction motors, braking resistor cabinets, converters, control contactors, and brake control units. Directed edges are constructed based on physical connections, energy transfer paths, and control signal flows. For example, an edge "Air compressor → Air cylinder" is established to supply air to the air compressor; an edge "Brake control unit → Brake valve" is established to control the brake valve. Initial weights are assigned to edges based on historical fault statistics. For instance, the initial weight of "Brake valve → Brake cylinder" is set to 0.6 due to numerous historical associated faults.

[0126] Based on real-time data, the edge weights of the fault association graph are dynamically adjusted. Real-time data shows an abnormal drop in brake cylinder pressure; therefore, the weight of "brake valve → brake cylinder" is increased to 0.7, and the weight of "brake cylinder → brake line" is increased from 0.5 to 0.8. The PageRank values ​​of the nodes are calculated to identify the brake valve and brake cylinder as critical components. Using Dijkstra's algorithm, the target fault propagation path is determined to be "brake valve → brake cylinder → brake line".

[0127] The operational data, key components, and target fault propagation paths are input into a multi-task predictive maintenance decision tree. Among them, the fault component prediction branch calculates that the brake cylinder anomaly score exceeds the threshold and is listed as a key abnormal component; the fault classification branch judges that the probability of brake cylinder "leakage" fault reaches 80% through random forest; the fault time prediction branch uses LSTM network to predict that the fault is expected to occur in about 18 hours; the impact range assessment branch determines that the fault will have a serious impact on the brake line and a moderate impact on the brake shoe.

[0128] The output fusion layer calculates the comprehensive risk index of the brake cylinder to be 0.52. Since there is only one faulty component, the priority maintenance prediction list only contains the multidimensional prediction results of the brake cylinder component. The multidimensional prediction results include a comprehensive risk index of 0.52, the faulty component is the brake cylinder, the fault type is a leakage probability of 80%, the expected occurrence time is 18 hours, and the affected range is the brake line (severe) and the brake shoe (moderate).

[0129] Based on a comprehensive risk index of 0.52, the system is classified as requiring emergency repairs. A maintenance reference report is generated according to the multidimensional prediction results, sent to the maintenance personnel, and they are notified to immediately shut down the system for maintenance.

[0130] In summary, this application's embodiments construct a dynamic fault association graph based on physical connections, energy transfer, and control signal flow by collecting equipment operating status data. This graph accurately depicts the complex fault transmission relationships between components. By dynamically adjusting edge weights using real-time data, it effectively reflects the dynamic changes during equipment operation. It utilizes PageRank values ​​to quickly locate key components, leverages the Dijkstra algorithm to efficiently pinpoint fault propagation paths, and outputs multi-dimensional prediction results through deep fusion of multi-task predictive maintenance decision trees, achieving a comprehensive and three-dimensional understanding of potential faults. The priority maintenance recommendations generated based on multi-dimensional predictions comprehensively consider the degree of fault risk, urgency, and scope of impact, scientifically allocating maintenance resources and accurately planning maintenance tasks.

[0131] Next, referring to the accompanying drawings, a fault prediction system for track braking equipment according to an embodiment of this application is described.

[0132] Figure 2 This is a schematic diagram of the structure of a track braking equipment fault prediction system according to an embodiment of this application.

[0133] like Figure 2 As shown, the track braking equipment fault prediction system 10 includes: a data acquisition module 100, a fault association module 200, an association range calculation module 300, a multi-task prediction module 400, and a maintenance suggestion generation module 500.

[0134] The system includes the following modules: a data acquisition module 100 for acquiring track braking equipment operating status data; a fault association module 200 for constructing a dynamic fault association graph to obtain the fault transmission relationship between braking equipment components and dynamically adjusting the edge weights of the dynamic fault association graph based on the track braking equipment operating status data; a association range calculation module 300 for calculating the PageRank value of nodes in the dynamic fault association graph to identify key components and using the Dijkstra algorithm to find the target fault propagation path; a multi-task prediction module 400 for constructing a multi-task predictive maintenance decision tree, inputting track braking equipment operating status data, key components, and target fault propagation paths into the multi-task predictive maintenance decision tree, and outputting multi-dimensional prediction results after multi-level decision-making, including a comprehensive risk index, faulty components, fault type and corresponding probability, expected occurrence time, and affected area; and a maintenance suggestion generation module 500 for generating priority maintenance suggestions based on the multi-dimensional prediction results.

[0135] According to the embodiments of this application, a fault prediction system for rail braking equipment is proposed. By comprehensively collecting equipment operating status data, a dynamic fault association graph based on physical connections, energy transfer, and control signal flow is constructed to accurately depict the complex fault transmission relationships between components. The edge weights are dynamically adjusted in conjunction with real-time data to effectively reflect the dynamic changes during equipment operation. PageRank values ​​are used to quickly locate key components, and Dijkstra's algorithm is used to efficiently lock fault propagation paths, greatly improving the accuracy and efficiency of fault prediction. A multi-task predictive maintenance decision tree deeply integrates equipment operating data, key component information, and fault path characteristics. Through multi-layered decision output, it provides multi-dimensional results covering fault components, types, probabilities, expected occurrence times, and affected areas, achieving a comprehensive and three-dimensional understanding of potential faults. Priority maintenance suggestions generated based on multi-dimensional predictions can comprehensively consider the degree of fault risk, urgency, and impact scope, scientifically allocate maintenance resources, and accurately plan maintenance tasks. This avoids the waste of resources from regular maintenance and eliminates operational interruptions and safety hazards caused by post-maintenance, improving equipment reliability and effectively ensuring the safety and efficiency of rail transit operation, thus promoting the intelligent and refined operation and maintenance of rail braking equipment. This solves the problems in existing technologies, such as ignoring the dynamic changes of components, severing the relationships between components, and lacking support for maintenance decisions.

[0136] Figure 3A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0137] The memory 301, the processor 302, and the computer program stored on the memory 301 and capable of running on the processor 302.

[0138] When the processor 302 executes the program, it implements a fault prediction method for track braking equipment provided in the above embodiments.

[0139] Furthermore, electronic devices also include:

[0140] Communication interface 303 is used for communication between memory 301 and processor 302.

[0141] The memory 301 is used to store computer programs that can run on the processor 302.

[0142] The memory 301 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0143] If the memory 301, processor 302, and communication interface 303 are implemented independently, then the communication interface 303, memory 301, and processor 302 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0144] Optionally, in a specific implementation, if the memory 301, processor 302, and communication interface 303 are integrated on a single chip, then the memory 301, processor 302, and communication interface 303 can communicate with each other through an internal interface.

[0145] Processor 302 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0146] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for predicting faults in a track braking device.

[0147] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0148] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0149] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0150] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0151] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

Claims

1. A method for predicting faults in track braking equipment, characterized in that, Includes the following steps: Acquire operational status data of the track braking equipment; A dynamic fault association graph is constructed to obtain the fault transmission relationship between braking equipment components. The construction of the dynamic fault association graph includes abstracting the physical components of the track braking equipment into graph nodes, constructing directed edges based on physical connection relationships, energy transfer paths, and control signal flows, and setting initial weights based on historical fault statistics. The edge weights of the dynamic fault association graph are dynamically adjusted based on the operating status data of the track braking equipment. The PageRank values ​​of the nodes in the dynamic fault association graph are calculated to identify key components. In the dynamic fault association graph, the PageRank value represents the influence of a component in the fault propagation network. The PageRank value of each component is calculated in turn. The mean and standard deviation are calculated based on the PageRank values ​​of each component. If the PageRank value of a component is greater than the sum of the mean and 1.5 times the standard deviation, the component is determined to be a core component. Otherwise, if the PageRank value of a component is greater than the mean, the component is determined to be an important component. Otherwise, it is a general component. The core components and important components are obtained as key components. The Dijkstra algorithm is used to find the target fault propagation path. A multi-task predictive maintenance decision tree is constructed. The operating status data of the track braking equipment, the key components, and the target fault propagation path are input into the multi-task predictive maintenance decision tree. After multi-level decision-making, multi-dimensional prediction results are output. The multi-dimensional prediction results include a comprehensive risk index, faulty components, fault types and corresponding probabilities, expected occurrence time, and affected range. Priority maintenance recommendations are generated based on the multidimensional prediction results.

2. The method for predicting faults in track braking equipment according to claim 1, characterized in that, The formula for calculating the PageRank value is as follows: in, This indicates the component node to be calculated. Indicates pointing to a node The set of nodes, Indicates from node To the node The edge weights are given by d, where d is the damping coefficient.

3. The method for predicting faults in track braking equipment according to claim 1, characterized in that, The method of finding the target fault propagation path using Dijkstra's algorithm includes: using Dijkstra's algorithm, starting from the critical component node, searching for the fault propagation path with the largest sum of weights as the target fault propagation path, wherein the specific steps are as follows: S1. Remove edges whose weights are below the target threshold; S2. Initialize the distance array, the visited set, and the predecessor node array. The distance array is used to store the cumulative weight of the maximum probability path from the starting node S to each node. Initially, the dist value of the starting node is set to 1, and the dist value of the other nodes is set to 0. The visited set is used to mark the nodes whose maximum probability path has been determined. The predecessor node array is used to store the predecessor node of each node on the maximum probability path. S4. From all unvisited nodes, select the node U with the largest dist value and add it to the visited set; S5. For all neighboring nodes V of node U, calculate the cumulative propagation probability from S through U to V. The specific calculation formula is as follows: in, Let U be the edge weight from node U to V; S6. If the cumulative propagation probability is greater than the dist value of the adjacent node V, then update the dist value of the adjacent node V to the cumulative propagation probability, and record the predecessor node of V as U in the predecessor node array. S7. Repeat steps S4-S6 until all reachable nodes have been processed. S8. Starting from the last node, backtrack along the predecessor node array to the starting point, and after reversing, obtain the target fault propagation path.

4. The method for predicting faults in track braking equipment according to claim 1, characterized in that, The multi-task predictive maintenance decision tree includes an input layer, a shared feature layer, task-specific branches, and an output fusion layer. The input layer receives track braking equipment operating status data, key component information, and fault propagation paths. The shared feature layer extracts common feature representations from the input data and filters key features. The task-specific branches construct multiple branches to execute prediction tasks separately. The output fusion layer integrates the prediction results from each branch and applies decision rules to generate the final multi-dimensional prediction result.

5. The method for predicting faults in a track braking device according to claim 4, characterized in that, The task-specific branches include: fault component prediction branch, fault classification branch, fault time prediction branch, and impact range assessment branch.

6. A method for predicting faults in track braking equipment according to claim 1 or 5, characterized in that, The process of inputting the operating status data of the track braking equipment, the key components, and the target fault propagation path into the multi-task predictive maintenance decision tree, and outputting multi-dimensional prediction results through multi-level decision-making, includes: Faulty Component Prediction Branch: Based on threshold-detected anomaly parameters, calculate the anomaly score of the component and output a list of key anomaly components. The formula for calculating the anomaly score is as follows: Where v represents the component, and n represents the number of parameters of the component. These are the importance weights of the parameters. It is the deviation of the i-th parameter of component v. It is part v value; Fault classification branch: Use a random forest classifier to output the fault type with the highest probability and its occurrence probability; Failure time prediction branch: Extract parameter degradation features, use LSTM network to predict parameter evolution trends, and predict the expected time of failure; Impact Scope Assessment Branch: Calculates the impact score based on the fault propagation path and outputs a list of affected components and their degree of impact.

7. A fault prediction system for track braking equipment, which can implement the fault prediction method for track braking equipment as described in any one of claims 1-6, characterized in that, include: The module includes a data acquisition module, a fault correlation module, a correlation range calculation module, a multi-task prediction module, and a maintenance suggestion generation module. The data acquisition module is used to acquire the operating status data of the track braking equipment; The fault association module is used to construct a dynamic fault association graph to obtain the fault transmission relationship between braking equipment components, and to dynamically adjust the edge weights of the dynamic fault association graph according to the operating status data of the track braking equipment. The association range calculation module is used to calculate the PageRank value of the nodes in the dynamic fault association graph to identify key components and to use the Dijkstra algorithm to find the target fault propagation path. The multi-task prediction module is used to construct a multi-task predictive maintenance decision tree. The operating status data of the track braking equipment, the key components, and the target fault propagation path are input into the multi-task predictive maintenance decision tree. After multi-level decision-making, multi-dimensional prediction results are output. The multi-dimensional prediction results include a comprehensive risk index, faulty components, fault type and corresponding probability, expected occurrence time, and affected range. The maintenance suggestion generation module is used to generate priority maintenance suggestions based on the multidimensional prediction results.

8. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, the processor executing the program to implement a method for predicting faults in a track braking device as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement a method for predicting faults in a track braking device as described in any one of claims 1-6.

Images