Vehicle fault sample library construction method and device, electronic equipment and storage medium
By automatically parsing and standardizing multi-source heterogeneous data from rail transit vehicles, a standardized data resource library is constructed. Data retrieval and association are performed based on the import of fault reports, which solves the problem of the difficulty in integrating multi-source heterogeneous data, improves the efficiency and accuracy of fault sample library construction, and supports large model training.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CRRC QINGDAO SIFANG CO LTD
- Filing Date
- 2025-12-29
- Publication Date
- 2026-04-10
Smart Images
Figure CN121833665A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of rail transit technology, and in particular to a vehicle fault sample library construction method and device, an electronic device and a storage medium. BACKGROUND
[0002] With the increasing intelligence of rail transit equipment, the amount and complexity of data generated by each system of the train have increased dramatically, including operation data, state data, video monitoring, vibration, audio, images and log records. These data play an important role in equipment state monitoring, health assessment and fault prediction, especially in the fields of intelligent modeling and anomaly detection.
[0003] However, the current industry faces the following challenges in constructing a high-quality vehicle fault sample library: high heterogeneity of multi-source data, difficulty in integration; sample construction process relies on manual screening and labeling, low efficiency and poor accuracy; lack of unified standards and platform support, inconsistent sample structure, difficult to share across projects; traditional databases are not suitable for time series and multi-modal mixed data management, slow query response, coarse retrieval granularity, and cannot meet the model training needs.
[0004] Therefore, an integrated fault sample construction and management system for multi-source heterogeneous data is urgently needed to realize intelligent, adaptive and extensible management of data analysis templates, improve sample quality, shorten modeling cycle, and enhance the reliability and promotion value of intelligent diagnostic systems. SUMMARY
[0005] The present application provides a vehicle fault sample library construction method, device, electronic device and storage medium to overcome the defects of current technology that multi-source heterogeneous data cannot be automatically fused, intelligently associated and standardized managed, resulting in low efficiency, poor accuracy and inability to support large model training for high-quality fault sample library construction, and comprehensively improving the organization efficiency, modeling value and sharing ability of train fault data.
[0006] In one aspect, the present application provides a vehicle fault sample library construction method, comprising: receiving multi-source heterogeneous data from a rail transit vehicle, and automatically analyzing and standardizing the multi-source heterogeneous data based on a preset data analysis template to construct a standardized data resource library; in response to the import of a current fault sheet, extracting the train number and fault time window in the current fault sheet, and retrieving the standard structured data segment and the standard unstructured data segment corresponding to the time period from the standardized data resource library based on the train number and the fault time window; associating the retrieved standard structured data segment and the standard unstructured data segment with the current fault sheet, and integrating them into a fault sample unit with a unified sample structure; storing the fault sample unit in a vehicle fault sample library, which is used to support large model training.
[0007] Further, the multi-source heterogeneous data includes structured data and unstructured data; accordingly, the automatic parsing and standardization processing of the multi-source heterogeneous data based on the preset data parsing template includes: automatically matching a corresponding data parsing template from a configurable parsing template library according to the data type, file extension and source system identifier of each type of data in the multi-source heterogeneous data; the data parsing template defines field mapping rules, unit conversion methods, timestamp extraction logic and exception handling strategies in the form of a meta-description file; according to the matched data parsing template, field alignment and numerical standardization are performed on the structured data, and modal adaptation parsing and feature processing are performed on the unstructured data; the processing results of the structured data and the processing results of the unstructured data are respectively stored in a time series database and an object storage system, while a unified time reference, a train number and a device identifier are bound for all data, to obtain a standardized data resource library supporting cross-modal retrieval.
[0008] Further, the retrieval of the standard structured data segment and the standard unstructured data segment corresponding to the time period from the standardized data resource library based on the train number and the fault time window includes: filtering all data records associated with the train number in the standardized data resource library, and expanding a preset tolerance interval centered on the fault time window; retrieving the standard structured data segment with a timestamp falling within the preset tolerance interval from the time series database, and retrieving the standard unstructured data segment with a timestamp intersecting with the preset tolerance interval from the metadata index of the object storage system, to form a multi-modal data set matching the current fault single in time and space.
[0009] Further, the association of the retrieved standard structured data segment and standard unstructured data segment with the current fault single includes: verifying the timestamps of the standard structured data segment and the standard unstructured data segment based on the unified time reference, and confirming that their train numbers are consistent with the train number in the current fault single; in the case of matching train numbers and timestamps falling within the fault time window or its preset tolerance interval, binding the unique identifier of the current fault single for each data segment, and recording its association with the current fault single in the metadata, to establish cross-modal data association centered on the fault event.
[0010] Further, it also includes: during the parsing and standardization processing of the multi-source heterogeneous data, in the case of file format identification failure, field missing or unit inconsistency, recording exception logs, skipping the current data item, and adding the current data item to a retry queue; if it still fails after continuous retries for a preset number of times, it is listed in the manual review list.
[0011] Further, the storing the fault sample unit into the vehicle fault sample library comprises: establishing a multi-dimensional index for the fault sample unit to support combined condition retrieval, permission control and direct calling of a model training platform; wherein the fields of the multi-dimensional index comprise vehicle model, column number, vehicle number, fault system name, fault type, time interval, data type and acquisition equipment.
[0012] In a second aspect, the present application further provides a vehicle fault sample library construction device, comprising: a data receiving and parsing module, configured to receive multi-source heterogeneous data from a rail transit vehicle, and automatically parse and standardize the multi-source heterogeneous data based on a preset data parsing template to construct a standardized data resource library; a fault-driven data retrieval module, configured to extract a vehicle number and a fault time window in a current fault sheet in response to import of the current fault sheet, and retrieve standard structured data segments and standard unstructured data segments corresponding to a time period from the standardized data resource library based on the vehicle number and the fault time window; a sample integration and association module, configured to associate the retrieved standard structured data segments and standard unstructured data segments with the current fault sheet, and integrate them into a fault sample unit with a unified sample structure; and a vehicle fault sample storage module, configured to store the fault sample unit into a vehicle fault sample library, the vehicle fault sample library being configured to support large model training.
[0013] In a third aspect, the present application further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the vehicle fault sample library construction method according to any one of the above aspects when executing the computer program.
[0014] In a fourth aspect, the present application further provides a non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the vehicle fault sample library construction method according to any one of the above aspects.
[0015] In a fifth aspect, the present application further provides a computer program product comprising a computer program, wherein the computer program is executable by a processor to implement the vehicle fault sample library construction method according to any one of the above aspects.
[0016] The vehicle fault sample library construction method provided by the application comprises the following steps: receiving multi-source heterogeneous data from a rail transit vehicle, automatically analyzing and standardizing the multi-source heterogeneous data based on a preset data analysis template, and constructing a standardized data resource library; then, in response to the import of a current fault sheet, extracting the vehicle number and fault time window in the current fault sheet, and retrieving the standard structured data segment and the standard unstructured data segment corresponding to the time period from the standardized data resource library based on the vehicle number and the fault time window; associating the retrieved standard structured data segment and the standard unstructured data segment with the current fault sheet, and integrating them into a fault sample unit with a unified sample structure; finally, storing the fault sample unit in the vehicle fault sample library, and using the vehicle fault sample library to support large model training. The method provides an automatic process integrating data access, analysis, time alignment, fault-driven matching, archiving and management, effectively solves the problem of difficulty in utilizing multi-source heterogeneous data in the rail transit field, realizes automatic analysis, standardization and fault-driven intelligent association of multi-source heterogeneous vehicle data, significantly improves the efficiency and accuracy of vehicle fault sample construction, and finally forms a vehicle fault sample library to effectively support large model training requirements. BRIEF DESCRIPTION OF DRAWINGS
[0017] In order to more clearly illustrate the technical solutions in the application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.
[0018] Figure 1 FIG. 1 is a flowchart of the vehicle fault sample library construction method provided by the embodiment of the application.
[0019] Figure 2 FIG. 2 is a structural diagram of the vehicle fault sample library construction device provided by the embodiment of the application.
[0020] Figure 3 FIG. 3 is a physical structure diagram of the electronic device provided by the embodiment of the application. DETAILED DESCRIPTION
[0021] In order to make the objects, technical solutions and advantages of the application clearer, the technical solutions in the application will be described clearly and completely below with reference to the drawings in the application. Obviously, the described embodiments are some embodiments of the application, but not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.
[0022] It needs to be explained that with the increasing intelligence of rail transit equipment, the amount and complexity of data generated by each system of the train are rapidly growing, such as operation data, state data, video monitoring, vibration, audio, image, log recording, etc. These data are widely used in equipment state monitoring, health assessment and fault prediction during operation and maintenance, especially in the fields of intelligent modeling and anomaly detection, and their value is increasingly prominent.
[0023] However, the current industry generally faces the following problems in building high-quality vehicle fault sample library: (1) high heterogeneity of multi-source data, difficult integration: different data formats, scattered sources, and parsing and alignment require a large amount of manual operation, especially between different manufacturers' equipment, the format standard is not unified, and the existing platform lacks flexible adaptive parsing mechanism; (2) sample construction process relies on manual screening and labeling: operation and maintenance personnel need to manually compare fault tickets with various data, which is low in efficiency and poor in accuracy; (3) lack of unified standards and platform support: inconsistent sample structure, poor cross-project sharing, not conducive to modeling reuse; (4) traditional database is not suitable for management of time series and multi-modal mixed data: slow query response, coarse retrieval granularity, not suitable for model training needs.
[0024] Therefore, it is urgent to develop an integrated fault sample construction and management system for multi-source heterogeneous data to realize intelligent, adaptive and extensible management of data parsing templates, improve sample quality, shorten modeling cycle, and improve the reliability and promotion value of intelligent diagnosis system.
[0025] In view of this, the present application provides a vehicle fault sample library construction method, specifically, Figure 1 The flowchart of the vehicle fault sample library construction method provided by the embodiment of the present application is shown.
[0026] As Figure 1 shown, the method comprises: S110, receiving multi-source heterogeneous data from a rail transit vehicle, and automatically parsing and standardizing the multi-source heterogeneous data based on a preset data parsing template to construct a standardized data resource library; S120, in response to the import of a current fault ticket, extracting the train number and fault time window in the current fault ticket, and retrieving the standard structured data segment and the standard unstructured data segment corresponding to the time period from the standardized data resource library based on the train number and the fault time window; S130, associating the retrieved standard structured data segment and the standard unstructured data segment with the current fault ticket, and integrating them into a fault sample unit with a unified sample structure; S140, storing the fault sample unit into a vehicle fault sample library, and the vehicle fault sample library is used to support large model training.
[0027] The steps S110-S140 and related steps will be described in detail below.
[0028] S110, receiving multi-source heterogeneous data from the rail transit vehicle, and automatically parsing and standardizing the multi-source heterogeneous data based on a preset data parsing template, and constructing a standardized data repository.
[0029] In actual application, the rail transit vehicle continuously generates a large amount of multi-source data from different systems during operation, including structured data (such as train operation records'speed, traction current', state data or diagnostic data 'PHM system alarm') with fixed fields and numerical formats and unstructured data (such as vehicle-mounted camera video, audio recording, axle box vibration signal, etc.) without fixed format. These data formats are different and sources are dispersed, which are recorded as multi-source heterogeneous data.
[0030] To solve the problem that multi-source heterogeneous data is difficult to be uniformly utilized, the embodiment pre-configures a set of extensible data parsing templates, each type of data corresponds to a data parsing template, and the data parsing template defines field mapping rules, unit conversion methods, timestamp extraction logic, etc.
[0031] When new multi-source heterogeneous data is accessed, the embodiment automatically identifies the data type, file extension and source system identifier of the multi-source heterogeneous data, matches the corresponding data parsing template, and completes automatic parsing and standardization processing. The processed structured data (standard structured data) is stored in a time series database, and the processed unstructured data (standard unstructured data) is stored in an object storage system, and is uniformly bound with a train number, a device identifier and a standard timestamp, and finally forms a standardized data repository supporting efficient query and cross-modal association.
[0032] On the basis of step S110 of receiving multi-source heterogeneous data from the rail transit vehicle and automatically parsing and standardizing the multi-source heterogeneous data based on a preset data parsing template to construct a standardized data repository, further, step S120 is executed.
[0033] S120, in response to the import of a current fault sheet, extracting a train number and a fault time window in the current fault sheet, and retrieving standard structured data segments and standard unstructured data segments corresponding to a time period from the standardized data repository based on the train number and the fault time window.
[0034] Specifically, when a staff or an automatic diagnosis system generates a current fault sheet (for example, XX column 03 car traction motor overheating) and imports the system, the step is triggered. First, two key information, train number and fault time window, are extracted from the current fault sheet. Among them, the train number is used to determine the target vehicle, and the fault time window is used to define the time range of the fault occurrence.
[0035] Subsequently, the relevant data is automatically retrieved from the constructed standardized data repository indexed by the two key information: on the one hand, all standard structured data segments such as temperature, current, and speed within the fault time window (including a reasonable tolerance interval) are obtained from the time series database; on the other hand, through metadata indexing, standard unstructured data segments such as video segments before and after the fault and vibration waveform records are found in the object storage system that have a timestamp intersecting with the fault time window. Finally, a multi-modal data set that accurately matches the current fault single in space and time is formed, including standard structured data segments and standard unstructured data segments.
[0036] Among them, the metadata index refers to a structured record describing the attributes of the standard unstructured data, including file path, timestamp, device identifier, etc.
[0037] The standard structured data segment refers to a subset of structured data that has completed standardization processing, and the standard unstructured data segment refers to a subset of unstructured data that has completed feature processing and is annotated with time.
[0038] After step S120, in response to the import of the current fault single, the vehicle number and fault time window in the current fault single are extracted, and the corresponding period of standard structured data segments and standard unstructured data segments are retrieved from the standardized data repository based on the vehicle number and fault time window, and then step S130 is further executed.
[0039] S130, the retrieved standard structured data segments and standard unstructured data segments are associated with the current fault single, and integrated into a fault sample unit with a unified sample structure.
[0040] The purpose of this step is to realize the logical binding and structured packaging of multi-source heterogeneous data and fault events. Specifically, first, verify whether the vehicle number of the retrieved standard structured data segments and standard unstructured data segments is consistent with the vehicle number in the current fault single, and confirm that the timestamps of the retrieved standard structured data segments and standard unstructured data segments indeed fall within the fault time window or its preset tolerance interval, to ensure the accuracy of the association.
[0041] After verification, the unique identifier (such as work order ID) of the current fault single is bound to each data segment (standard structured data segment and standard unstructured data segment), and the association with the current fault single is explicitly recorded in its metadata, thereby establishing a cross-modal data association centered on the fault event.
[0042] Subsequently, the fault basic information (such as fault type, fault system name, level) of the current fault order, the associated standard structured data segment, the storage path or reference identifier of the standard unstructured data segment (i.e. a logical link pointing to the actual file location instead of copying the original large file), and the above metadata are integrated into a logically complete fault sample unit according to a predefined specification. The fault sample unit has a unified sample structure, that is, all samples of the same type adopt the same organization format, facilitating subsequent batch processing and model training.
[0043] After associating the retrieved standard structured data segment and standard unstructured data segment with the current fault order and integrating them into a fault sample unit with a unified sample structure in step S130, further, step S140 is performed.
[0044] S140, store the fault sample unit into a vehicle fault sample library, the vehicle fault sample library is used to support large model training.
[0045] Specifically, the constructed fault sample unit is written into a dedicated vehicle fault sample library. The vehicle fault sample library can not only be used for long-term archiving, but also can be used for artificial intelligence applications, especially large model training scenarios. Since each fault sample unit contains multi-modal, high-alignment, and structured data, the vehicle fault sample library can be directly used for supervised learning, self-supervised pre-training, or fault pattern mining.
[0046] In addition, the vehicle fault sample library supports multi-dimensional indexing according to vehicle type, fault system name, time interval, etc. dimensions, facilitating staff to quickly screen specific type fault samples and batch export training sets, so as to significantly improve the efficiency and generalization ability of large model development.
[0047] In the embodiment, a standardized data resource library is constructed by receiving multi-source heterogeneous data from a rail transit vehicle, automatically parsing and standardizing the multi-source heterogeneous data based on a preset data parsing template, and then, in response to the import of a current fault single, extracting a vehicle number and a fault time window in the current fault single, and retrieving standard structured data segments and standard unstructured data segments corresponding to the time window from the standardized data resource library based on the vehicle number and the fault time window; the retrieved standard structured data segments and standard unstructured data segments are associated with the current fault single and integrated into a fault sample unit with a unified sample structure; finally, the fault sample unit is stored in a vehicle fault sample library, which is used to support large model training. The method provides an automatic process integrating data access, parsing, time alignment, fault-driven matching, archiving and management, effectively solves the problem of difficulty in fusion and utilization of multi-source heterogeneous data in the rail transit field, realizes automatic parsing, standardization and fault-driven intelligent association of multi-source heterogeneous vehicle data, significantly improves the efficiency and accuracy of vehicle fault sample construction, and finally forms a vehicle fault sample library to effectively support large model training requirements.
[0048] On the basis of the above embodiment, further, the construction process of the standardized data resource library in step S110 will be described in detail.
[0049] The multi-source heterogeneous data includes structured data and unstructured data; accordingly, the multi-source heterogeneous data is automatically parsed and standardized based on a preset data parsing template to construct a standardized data resource library, including: automatically matching a corresponding data parsing template from a configurable parsing template library according to the data type, file extension and source system identifier of each type of data in the multi-source heterogeneous data; the data parsing template defines field mapping rules, unit conversion methods, timestamp extraction logic and exception handling strategies in the form of meta-description files; according to the matched data parsing template, the structured data is subjected to field alignment and numerical standardization, and the unstructured data is subjected to modal adaptation parsing and feature processing; the processing results of the structured data and the unstructured data are respectively stored in a time series database and an object storage system, and a unified time reference, vehicle number and equipment identifier are bound for all the data, to obtain a standardized data resource library supporting cross-modal retrieval.
[0050] It is easy to understand that first, multi-source heterogeneous data from rail transit vehicles is received. The multi-source heterogeneous data includes structured data and unstructured data, which is specifically from on-board recording devices, PHM systems, video monitoring systems, MRO operation and maintenance platforms and other systems, and covers the following types: (1) train operation data, such as.dat or.csv files; (2) state diagnosis data, such as.tdms or.sbr files; (3) image data, such as.jpg or.png files; (4) video data, such as.mp4 files; (5) audio data, such as.wav files; and (6) vibration data, such as high sampling rate time series signal files.
[0051] To realize automatic parsing and standardized processing of multi-source heterogeneous data, the embodiment performs a unified processing flow based on a preset data parsing template. The preset data parsing template is stored in a configurable parsing template library (i.e., a configuration set for centrally managing various data parsing rules) and is defined in the form of a meta-description file to define field mapping rules, unit conversion methods, timestamp extraction logic and exception handling strategies.
[0052] Specifically, according to the data type, file extension and source system identifier of each type of data in the multi-source heterogeneous data, a corresponding data parsing template is automatically matched from the parsing template library, and different types of data correspond to different data parsing templates.
[0053] For example: for train operation data (.dat / .csv), the corresponding data parsing template defines how to map the original fields to standard operation parameters (such as speed, current, temperature) and unify the units, and this process can be realized through pandas / NumPy; for state diagnosis data (.tdms), the corresponding data parsing template specifies reading by channel, handling missing values and abnormal peak values, and outputting standard time series, and this process can be realized through NI TDMS library / pyTDMS; for image data (.jpg / .png), the corresponding data parsing template indicates extracting the timestamp in EXIF and associating the camera number, and this process can be realized through OpenCV / PIL; for video data (.mp4), the corresponding data parsing template configures key frame extraction, total duration recording and event frame identification logic, and this process can be realized through OpenCV / FFmpeg; for audio data (.wav), the corresponding data parsing template sets sampling rate conversion, segmentation strategy and start and end time labeling rules, and this process can be realized through librosa / scipy.io.wavfile; for vibration data (.json / .xls), the corresponding data parsing template defines resampling frequency, channel alignment method and signal standardization method, and this process can be realized through pandas / json.
[0054] In the process of parsing by using the data parsing template, field alignment and numerical standardization are performed on structured data; modal adaptation, parsing and feature processing are performed on unstructured data, and finally standard structured data segments and standard unstructured data segments with standard time information are generated to form a so-called standardized data resource library.
[0055] To ensure processing reliability, the embodiment is provided with a complete exception handling mechanism. When file format recognition fails, fields are missing, units are inconsistent or files are damaged, etc., error logs are automatically recorded, the current data item is skipped, and the task is added to the retry queue; in a multi-threaded scheduling environment, the failed task is automatically retried at most three times, and if it still fails, it is listed in the manual review list to ensure the integrity and traceability of the data processing process.
[0056] After parsing, the processing results are classified and stored, that is, the processing results of structured data are written into a time series database, and the processing results of unstructured data are stored in an object storage system. At the same time, a unified time reference, train number and equipment identifier are bound for all data segments, so that data of different sources and different modalities are aligned and associated in the time and space dimensions. The equipment identifier refers to the unique number of the sensor or device that collects the data.
[0057] In addition, the embodiment also supports version management and adaptive update of the data parsing template. Specifically, when a new data format is first accessed, candidate templates can be automatically generated based on file header features and field distribution, and after manual confirmation, they are included in the parsing template library; after the data parsing template is updated, automatic re-parsing of historical data can be triggered to ensure the consistency of the full amount of data. All data parsing templates support cross-model and cross-project reuse, and can be exported, imported or synchronized remotely through an interface to realize rapid expansion of parsing capability.
[0058] It is worth mentioning that the standardized data resource library constructed by the embodiment not only contains multi-source heterogeneous data that has been standardized, but also establishes the internal correlation between data through unified metadata, providing a reliable data foundation for subsequent precise retrieval based on fault single and high-quality fault sample unit construction.
[0059] On the basis of the above embodiment, further, the following will be described in detail with respect to the retrieval process and data correlation process based on the train number and the fault time window in step S120.
[0060] Retrieving the standard structured data segments and the standard unstructured data segments corresponding to the time period from the standardized data repository based on the vehicle number and the fault time window, including: screening all data records associated with the vehicle number in the standardized data repository, and expanding a preset tolerance interval centered on the fault time window; retrieving the standard structured data segments with timestamps falling within the preset tolerance interval from the time series database, and retrieving the standard unstructured data segments with timestamps intersecting with the preset tolerance interval from the metadata index of the object storage system, to form a multi-modal data set matching the current fault single in space and time.
[0061] Associating the retrieved standard structured data segments and the standard unstructured data segments with the current fault single, including: verifying the timestamps of the standard structured data segments and the standard unstructured data segments based on a unified time reference, and confirming that the vehicle number they belong to is consistent with the vehicle number in the current fault single; in the case of matching vehicle numbers and timestamps falling within the fault time window or its preset tolerance interval, binding the unique identifier of the current fault single to each data segment, and recording its association with the current fault single in the metadata, to establish a cross-modal data association centered on the fault event.
[0062] It is easy to understand that in the present embodiment, in response to the import of the current fault single, a field extraction operation is first performed. The current fault single is derived from an operation and maintenance work order system, and key information is extracted from the current fault single, including the vehicle number (i.e., used to uniquely identify the vehicle that has failed), the train number, the fault time window (i.e., the start and end time interval of the fault occurrence), the fault system name, the fault type, and the fault level. Among them, the vehicle number and the fault time window are the core index conditions for subsequent data retrieval.
[0063] Subsequently, based on the vehicle number and the fault time window, the standard structured data segments and the standard unstructured data segments corresponding to the time period are retrieved from the standardized data repository. Specifically, first, all data records associated with the vehicle number in the standardized data repository are screened, and a preset tolerance interval is expanded centered on the fault time window. The preset tolerance interval allows a certain time deviation buffer range, which is set to ±2 minutes by default, and can be configured according to the clock characteristics of different data sources to cope with real-world scenarios such as sensor clock drift, data reporting delay, or manual reporting errors.
[0064] Within the expanded preset tolerance interval, on the one hand, the standard structured data segments with timestamps falling within the preset tolerance interval are retrieved from the time series database, including train operation data, state diagnosis data, etc.; on the other hand, the standard unstructured data segments with timestamps intersecting with the preset tolerance interval are retrieved from the metadata index (i.e., a structured record describing the attributes of unstructured data, including file path, timestamp, device identifier, etc.) of the object storage system, including images, videos, audio, and vibration data, etc.
[0065] Next, the retrieved standard structured data segments and standard unstructured data segments are associated with the current fault order. Specifically, first, the timestamps of all data segments are checked based on a unified time reference, all data segments have been uniformly converted to UTC time format before being stored (time series database or object storage system), and aligned through a timestamp mapping table (a configuration table recording the collection frequency and time base offset relationship of various data sources). At the same time, it is confirmed that the train number of each data segment is consistent with the train number in the current fault order.
[0066] Under the condition that the train number is matched and the timestamp falls within the fault time window or its preset tolerance interval, the unique identifier (such as the work order ID) of the current fault order is bound for each data segment, and the association with the current fault order is explicitly recorded in its metadata.
[0067] To ensure the spatio-temporal consistency of multi-modal data (standard structured data segments and standard unstructured data segments), a time alignment strategy can also be further performed. Specifically, key frames are extracted from image data at a frame rate of seconds; audio data is sliced in units of sampling points; interpolation method is used for synchronous resampling of vibration data; all alignment results are quality verified to check whether the start and end times, data points or frame numbers of each modality data are coordinated and consistent; for slight offset (such as camera clock error) within ±3 seconds, automatic correction is performed; if effective alignment cannot be completed, the data is marked as alignment failed, and abnormal log is written for manual review.
[0068] The embodiment establishes cross-modal data association centered on fault events, ensures that each fault order can be associated with a set of spatio-temporally aligned, reliable and format-standardized multi-modal data, and provides high-quality input for subsequent construction of fault sample units with unified sample structure.
[0069] On the basis of the above embodiment, further, the integration process of the fault sample unit and the multi-dimensional index establishment process thereof will be described in detail as follows.
[0070] Storing the fault sample unit into the vehicle fault sample library, then including: establishing a multi-dimensional index for the fault sample unit to support combined condition retrieval, permission control and direct calling of the model training platform; wherein the fields of the multi-dimensional index include vehicle type, column number, train number, fault system name, fault type, time interval, data type and collection equipment.
[0071] It is easy to understand that after the association of the standard structured data segment and the standard unstructured data segment with the current fault sheet is completed, the associated data is further integrated into a fault sample unit with a unified sample structure. The fault sample unit contains the basic information of the current fault sheet (such as train number, car number, fault system name, fault type, fault time window, etc.), the reference identifier of the associated standard structured data segment and standard unstructured data segment (i.e. the logical path of the image, video, audio, vibration and other data in the object storage system), and the metadata describing the sample construction process (including data source system, acquisition device number, sample construction time, construction state and construction method, etc.).
[0072] Exemplarily, a fault sample unit is si={train_id: CRH-xxxx, car_id: xxx, type: 'brake abnormality', module: 'brake system', d_run: {WTD / PTU structured data}, d_img: [image slices], d_vid: [video frame sequence], d_aud: [audio slices (MFCC / time domain)], d_vib: [vibration data segment], metadata: {source: ['PTU', 'video monitoring', 'on-board recording'], build_time: datetime, labeler:'system automatic construction', status:'ready'}}.
[0073] Subsequently, the fault sample unit is stored in the vehicle fault sample library, i.e. the database that centrally manages the vehicle fault samples. The storage operation is bound according to the current fault sheet dimension, ensuring that each fault event corresponds to a unique and complete fault sample unit, avoiding data redundancy or mismatch.
[0074] After the storage is completed, a multi-dimensional index is further established for the fault sample unit. Through the multi-dimensional index, users can perform combined condition retrieval, such as filtering fault samples of 'CRHXXX series, brake system, 20XX year, containing video data'. It also supports identity-based permission control, different users or project groups can only access the samples within their authorized range. All access behaviors are recorded in the access log, realizing the traceability of operation.
[0075] Among them, the fields of the multi-dimensional index include vehicle type, train number, car number, fault system name, fault type, time interval, data type (such as running data, image, video, audio, vibration, etc.) and acquisition device (i.e. the specific sensor or recording device that generates the data).
[0076] At the application level, the vehicle fault sample library supports multiple sample calling methods. Specifically, a user can quickly query a target sample through a keyword or a combined condition; the library supports batch exporting of samples meeting the conditions as a model training set, and the export formats include a structured label file (such as a CSV file) and a multi-modal data package (such as a ZIP compressed package), which are compatible with the data loading interfaces of mainstream deep learning frameworks; and the model training platform can directly call the fault sample unit through an API for model training, verification, and evaluation without manual intervention.
[0077] In some other embodiments, the actual operation process of the vehicle fault sample library construction method provided in the embodiments of the present application is described by taking the automatic construction of abnormal fault samples of a certain type of intercity EMU braking system as an example.
[0078] On XX month XX day in 20XX, an operation and maintenance unit uploaded the following original multi-source heterogeneous data of a certain EMU to the data management platform: (1) WTD operation data: recording the vehicle number "CRH-XXX-XXX", and the fault time window is from 14:00 to 16:00 on XX month XX day in 20XX; (2) video monitoring data: video clips recorded by the car camera (format: MP4); (3) audio and vibration data: collected and recorded by the on-board sensor (format: TDMS); (4) fault single exported by the MRO system, recording contents including: a. fault number: BRK-20XX-XX; b. fault type: abnormal pressure delay of the braking system; c. vehicle number involved: CRH-XXX-XXX; d. fault time interval: 14:10:25-14:15:50 on XX-XX-XX.
[0079] After the uploading is completed, the task scheduling center automatically triggers the parsing task according to the file type. The parsing module first automatically matches the corresponding data parsing template from the parsing template library through the file extension, file header characteristics, and source system information. In the parsing process, the field mapping, unit conversion, channel selection, and abnormality filtering are completed according to the data parsing template definition, and a standardized structured table D' is automatically generated.
[0080] For train operation data (.dat / .csv), the structured fields (standard parameters such as speed, current, and pressure) are parsed according to the template rules; for state diagnosis data (.tdms), the multi-channel signals are extracted using the channel mapping and abnormal point filtering rules defined in the template; for video data (.mp4), the key frames are extracted and time-aligned according to the frame rate and timestamp rules defined in the template; for vibration data (.json / .xls), the resampling frequency, channel alignment method, and signal standardization method are defined.
[0081] For new file formats that appear for the first time (such as .datx files generated by different manufacturers' acquisition devices), the file header structure and field characteristics are automatically detected, candidate templates are generated and manual confirmation is prompted. After review, they can be automatically added to the template library to form a new version record.
[0082] After template parsing and verification, standard structured data is written into the time-series database (IoTDB), and standard unstructured data (images, audio and video) is written into the object storage system (MinIO). All data is then uniformly linked to the metadata index table to achieve standardized data entry and subsequent traceable retrieval.
[0083] The system automatically detects new data entry information for fault ticket "BRK-20XX-XX" during daily scheduling, triggering the sample construction process.
[0084] First, extract the key fields from the fault report, including vehicle number (CRH-XXX-XXX), fault time window (20XX-XX 14:10:25 ~ 14:15:50), fault type (brake system - delayed response), and fault level (level 1 fault).
[0085] Then, automatic matching and association are performed to automatically retrieve the following data within a fault time window of ±2 minutes: WTD runtime data slices (5-minute time intervals); video keyframe sequences; audio sampling waveforms; and vibration data segments. During the parsing phase, all types of data generate a unified format timestamp field based on a template mechanism, thereby achieving time alignment and synchronization between different modalities.
[0086] Next, a unified fault sample unit is constructed, generating a sample object structure with a unified format as follows: {"train_id": "CRH-XXX-XXX","fault_code": "BRK-20XX-XX","system":"Braking System","fault_type":"Delay Response","time_window":"20XX-XX-XX 14:10:25~14:15:50","run_data":{...},"video_frames":[...],"audio_clip":[...], "vibration_data":[...],"metadata":{"upload_time":"20XX-XX-XX 10:30:22","source":"PTU+video surveillance+TDMS","status":"already in the database","tag":["Level 1 fault","braking","template parsing"]}}.
[0087] Subsequently, the sample is warehoused and indexed. The fault sample unit is written into the vehicle fault sample library FS, and a multi-dimensional retrieval index is generated according to the following fields: vehicle model: CRH-XXX; fault system: braking system; fault level: level one; upload source: comprehensive upload; data coverage type: WTD + video + audio + vibration.
[0088] The operation and maintenance expert can quickly retrieve the sample set in the platform through conditional filtering (such as "vehicle model = CRH-XXX and fault system = braking system and fault level = level one"), view the sample details, or export for subsequent AI modeling and verification.
[0089] The vehicle fault sample library construction method provided by the embodiments of the present application can be deployed in a unified data management platform, and functions such as task creation, progress viewing, sample preview, and result exporting can be provided through a Web end or an API interface, so as to realize an efficient closed-loop management process from raw multi-source heterogeneous data access, to automatic parsing, fault-driven matching, sample construction, multi-dimensional index establishment, to permission-controllable sharing, thereby significantly improving the construction efficiency of high-quality fault samples in the field of rail transit and the AI model development support capability.
[0090] Corresponding to the vehicle fault sample library construction method described in the above embodiments, the present application also provides a vehicle fault sample library construction device. Specifically, Figure 2 A structural schematic diagram of the vehicle fault sample library construction device provided by the embodiments of the present application is shown.
[0091] As Figure 2 shown, the device comprises: a data receiving and parsing module 210, configured to receive multi-source heterogeneous data from a rail transit vehicle, and perform automatic parsing and standardization processing on the multi-source heterogeneous data based on a preset data parsing template, to construct a standardized data resource library; a fault-driven data retrieval module 220, configured to extract a vehicle number and a fault time window in a current fault sheet in response to import of the current fault sheet, and retrieve standard structured data segments and standard unstructured data segments corresponding to a time period from the standardized data resource library based on the vehicle number and the fault time window; a sample integration and association module 230, configured to associate the retrieved standard structured data segments and standard unstructured data segments with the current fault sheet, and integrate them into a fault sample unit with a unified sample structure; and a vehicle fault sample warehousing module 240, configured to store the fault sample unit into a vehicle fault sample library, the vehicle fault sample library being configured to support large model training.
[0092] In the embodiment, the multi-source heterogeneous data from the rail transit vehicle is received by the data receiving and analyzing module 210, and the multi-source heterogeneous data is automatically analyzed and standardized based on the preset data analysis template to construct a standardized data resource library; then, the fault-driven data retrieval module 220 responds to the import of the current fault single, extracts the train number and fault time window in the current fault single, and retrieves the standard structured data segment and the standard unstructured data segment corresponding to the time period from the standardized data resource library based on the train number and the fault time window; the sample integration and association module 230 associates the retrieved standard structured data segment and standard unstructured data segment with the current fault single, and integrates them into a fault sample unit with a unified sample structure; finally, the vehicle fault sample storage module 240 stores the fault sample unit into the vehicle fault sample library, and the vehicle fault sample library is used to support large model training. The device provides an automatic process integrating data access, analysis, time alignment, fault-driven matching, archiving and management, effectively solves the problem of difficult fusion and utilization of multi-source heterogeneous data in the field of rail transit, realizes automatic analysis, standardization and fault-driven intelligent association of multi-source heterogeneous vehicle data, significantly improves the efficiency and accuracy of vehicle fault sample construction, and finally forms a vehicle fault sample library to effectively support the training needs of large models.
[0093] It should be noted that the vehicle fault sample library construction device provided by the embodiment of the application can be correspondingly referred to the vehicle fault sample library construction method described in the above embodiments, which will not be described here.
[0094] Figure 3 An example of an entity structure diagram of an electronic device is shown in FIG. 1. Figure 3As shown, the electronic device can include a processor 310, a communications interface 320, a memory 330, and a communications bus 340, wherein the processor 310, the communications interface 320, and the memory 330 communicate with each other through the communications bus 340. The processor 310 can invoke the logical instructions in the memory 330 to execute the vehicle fault sample library construction method, which includes: receiving multi-source heterogeneous data from a rail transit vehicle, and automatically parsing and standardizing the multi-source heterogeneous data based on a preset data parsing template to construct a standardized data resource library; in response to the import of a current fault single, extracting the train number and fault time window in the current fault single, and retrieving the corresponding period of standard structured data segments and standard unstructured data segments from the standardized data resource library based on the train number and fault time window; associating the retrieved standard structured data segments and standard unstructured data segments with the current fault single, and integrating them into a fault sample single unit with a unified sample structure; and storing the fault sample single unit in a vehicle fault sample library, which is used to support large model training.
[0095] In addition, the logical instructions in the memory 330 described above can be implemented in the form of a software functional unit and sold or used as an independent product, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium, includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.
[0096] In another aspect, the present application also provides a computer program product comprising a computer program, which can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to perform the vehicle fault sample library construction method provided by the above-mentioned methods, which comprises: receiving multi-source heterogeneous data from a rail transit vehicle, and automatically parsing and standardizing the multi-source heterogeneous data based on a preset data parsing template to construct a standardized data resource library; in response to the import of a current fault sheet, extracting a car number and a fault time window in the current fault sheet, and retrieving standard structured data segments and standard unstructured data segments corresponding to a time period from the standardized data resource library based on the car number and the fault time window; associating the retrieved standard structured data segments and standard unstructured data segments with the current fault sheet, and integrating them into a fault sample sheet unit having a unified sample structure; and storing the fault sample sheet unit into a vehicle fault sample library, which is used to support large model training.
[0097] In another aspect, the present application also provides a non-transitory computer readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the vehicle fault sample library construction method provided by the above-mentioned methods, which comprises: receiving multi-source heterogeneous data from a rail transit vehicle, and automatically parsing and standardizing the multi-source heterogeneous data based on a preset data parsing template to construct a standardized data resource library; in response to the import of a current fault sheet, extracting a car number and a fault time window in the current fault sheet, and retrieving standard structured data segments and standard unstructured data segments corresponding to a time period from the standardized data resource library based on the car number and the fault time window; associating the retrieved standard structured data segments and standard unstructured data segments with the current fault sheet, and integrating them into a fault sample sheet unit having a unified sample structure; and storing the fault sample sheet unit into a vehicle fault sample library, which is used to support large model training.
[0098] The device embodiments described above are only schematic, wherein the units shown as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment scheme. Those skilled in the art can understand and implement it without creative labor.
[0099] Those skilled in the art can clearly understand the technical solutions of the various embodiments from the above description of the embodiments, and the various embodiments can be implemented by means of software with the necessary general hardware platforms, and of course, can also be implemented by hardware. Based on such understanding, the above technical solutions, essentially or in other words, the part of the prior art that makes a contribution, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, and the like, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0100] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features therein; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A method of constructing a vehicle failure sample library, characterized by, The method comprises the following steps: Receiving multi-source heterogeneous data from a rail transit vehicle, and automatically parsing and standardizing the multi-source heterogeneous data based on a preset data parsing template to construct a standardized data resource library; In response to the import of a current fault single, extracting the train number and fault time window in the current fault single, and retrieving the standard structured data segments and standard unstructured data segments of the corresponding time period from the standardized data resource library based on the train number and fault time window; Associating the retrieved standard structured data segments and standard unstructured data segments with the current fault single, and integrating them into a fault sample unit with a unified sample structure; Storing the fault sample unit in a vehicle fault sample library, which is used to support large model training.
2. The vehicle failure sample library construction method according to claim 1, characterized by, The multi-source heterogeneous data includes structured data and unstructured data; Accordingly, the automatic parsing and standardization of the multi-source heterogeneous data based on the preset data parsing template to construct a standardized data resource library comprises the following steps: According to the data type, file extension and source system identifier of each type of data in the multi-source heterogeneous data, automatically matching the corresponding data parsing template from a configurable parsing template library; the data parsing template defines field mapping rules, unit conversion methods, timestamp extraction logic and exception handling strategies in the form of meta-description files; According to the matched data parsing template, performing field alignment and numerical standardization on the structured data, and performing modal adaptation analysis and feature processing on the unstructured data; Storing the processing results of the structured data and the processing results of the unstructured data in a time series database and an object storage system respectively, and binding a unified time reference, train number and equipment identifier to all data to obtain a standardized data resource library supporting cross-modal retrieval.
3. The vehicle failure sample library construction method according to claim 2, characterized by, The retrieval of the standard structured data segments and standard unstructured data segments of the corresponding time period from the standardized data resource library based on the train number and fault time window comprises the following steps: Filtering all data records associated with the train number in the standardized data resource library, and expanding a preset tolerance interval centered on the fault time window; Retrieving the standard structured data segments with timestamps falling within the preset tolerance interval from the time series database, and retrieving the standard unstructured data segments with timestamps intersecting with the preset tolerance interval from the metadata index of the object storage system, to form a multi-modal data set matching the current fault single in space and time.
4. The vehicle failure sample library construction method according to claim 1, characterized by, The association of the retrieved standard structured data segments and standard unstructured data segments with the current fault single comprises the following steps: Based on the unified time reference, checking the timestamps of the standard structured data segments and the standard unstructured data segments, and confirming that their train numbers are consistent with the train number in the current fault single; In the case of matching train numbers and timestamps falling within the fault time window or its preset tolerance interval, binding the unique identifier of the current fault single to each data segment, and recording its association with the current fault single in the metadata to establish a cross-modal data association centered on the fault event.
5. The vehicle failure sample library construction method according to claim 1, characterized by, Further comprising: In the process of analyzing and standardizing the multi-source heterogeneous data, if file format recognition fails, fields are missing, or units are inconsistent, an abnormal log is recorded, the current data item is skipped, and the current data item is added to a retry queue; if it still fails after continuous retries for a preset number of times, it is listed in a manual review list.
6. The vehicle failure sample library construction method according to any one of claims 1 to 5, characterized by, The fault sample unit is stored in a vehicle fault sample library, and then includes: A multi-dimensional index is established for the fault sample unit to support combined condition retrieval, permission control, and direct calling of a model training platform. The fields of the multi-dimensional index include vehicle type, column number, vehicle number, fault system name, fault type, time interval, data type, and collection equipment.
7. A vehicle failure sample library construction apparatus characterized by comprising: It includes: A data receiving and parsing module is configured to receive multi-source heterogeneous data from a rail transit vehicle, and automatically analyze and standardize the multi-source heterogeneous data based on a preset data parsing template to construct a standardized data resource library; A fault-driven data retrieval module is configured to extract a vehicle number and a fault time window from a current fault sheet in response to import of the current fault sheet, and retrieve standard structured data segments and standard unstructured data segments corresponding to a time period from the standardized data resource library based on the vehicle number and the fault time window; A sample integration and association module is configured to associate the retrieved standard structured data segments and standard unstructured data segments with the current fault sheet, and integrate them into a fault sample unit with a unified sample structure; A vehicle fault sample storage module is configured to store the fault sample unit in a vehicle fault sample library, and the vehicle fault sample library is configured to support large model training.
8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, The processor executes the computer program to implement the vehicle fault sample library construction method of any one of claims 1 to 6. 9.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the vehicle fault sample library construction method of any one of claims 1 to 6.
10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the vehicle fault sample library construction method of any one of claims 1 to 6.