A fully automated in-depot acceptance system and method for locomotive monitoring devices

By introducing technologies such as industrial vision servo and flexible robotic arms, combined with edge computing and federated learning models, a fully automated in-warehouse acceptance system for locomotive monitoring devices has been implemented. This system solves the problems of low automation in acceptance, poor machine model compatibility, and delayed fault warning, improving acceptance efficiency and safety. It also enables multi-machine model compatibility and latent fault identification, meeting the application requirements of high-frequency locomotive turnover.

CN123067271APending Publication Date: 2026-11-10CHINA RAILWAY 16TH BUREAU GRP RAIL TRANSPORT ENG CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611295563.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-25
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

The existing locomotive monitoring device acceptance process suffers from problems such as low automation, poor model compatibility, delayed fault warnings, non-standard data management, and disconnected preparation processes. This results in low acceptance efficiency, inconsistent results, high labor intensity, and difficulty in identifying hidden faults.

Method used

The fully automated in-warehouse acceptance system, which employs locomotive monitoring devices, includes a vehicle number recognition and entry triggering unit, an adaptive alignment detection and docking unit, a multi-source detection execution unit, an edge computing processing unit, an intelligent judgment and early warning unit, and a data archiving unit. Combined with industrial vision servo, flexible robotic arms, multi-specification probe modules, edge computing architecture, and federated learning models, it achieves unmanned acceptance and closed-loop management.

Benefits of technology

Significantly improves the level of automation in the integration process, adapts to multiple models, identifies hidden faults, achieves efficient digitalization of data management, reduces labor intensity, forms a closed-loop control throughout the entire process, and improves acceptance efficiency and security.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123067271A_ABST
    Figure CN123067271A_ABST
Patent Text Reader

Abstract

This invention relates to the field of railway locomotive maintenance and automated testing technology, specifically a fully automated in-depot acceptance system and method for locomotive monitoring devices. The system includes a locomotive number recognition and entry triggering unit, an adaptive alignment detection docking unit, a multi-source detection execution unit, an edge computing processing unit, an intelligent judgment and early warning unit, as well as a data archiving unit and a maintenance and scheduling linkage unit. After the locomotive enters the depot, the system automatically identifies the locomotive number and equipment model. A visual servo flexible docking mechanism enables automatic docking of the detection interface under large parking deviations. Relying on a federated learning adaptive model to match corresponding detection standards, it integrates multi-source time-series data from electrical, vibration, and acoustic sources to complete full-item detection and latent fault early warning. An edge computing architecture enables local real-time processing and breakpoint resume, and it links with the maintenance and scheduling system to form a closed loop of detection-maintenance-re-inspection. This invention achieves fully unmanned operation of the entire in-depot acceptance process for locomotive monitoring devices.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of railway locomotive maintenance and automated testing technology, specifically a fully automated in-depot acceptance system and method for locomotive monitoring devices. Background Technology

[0002] Locomotive monitoring devices are core onboard equipment that ensures railway traffic safety. Their functional integrity and operational reliability are directly related to locomotive operation safety. According to railway maintenance regulations, the monitoring devices must be fully inspected and tested every time a locomotive enters the depot for maintenance. Only after confirming that the equipment is in good condition can it be put out of the depot for operation again.

[0003] Currently, the acceptance of monitoring devices in most locomotive depots in China still relies primarily on manual operations. Maintenance personnel must climb each locomotive individually, manually connect testing cables, perform each test item, manually verify parameters, and fill out acceptance records. This method has several significant drawbacks: Firstly, the acceptance efficiency is low. The full acceptance of a single locomotive takes a long time, which can easily cause congestion on the depot lines during peak preparation periods when train traffic is concentrated, and cannot meet the current demand for high-density locomotive turnover. Secondly, the testing standards are inconsistent, and the acceptance results are highly dependent on the professional skills and working conditions of the maintenance personnel. Different personnel have different judgment standards, which can easily lead to missed inspections, incorrect inspections, and misjudgments, posing potential hazards to driving safety. Third, the level of data management is lagging behind. Acceptance records are mostly filled out manually using paper ledgers, which is not only inefficient and prone to being contaminated and losing effectiveness, but also makes it difficult to summarize, analyze and trace the data, which does not meet the current management requirements of railway informatization and digital transformation. Fourth, the on-site work is labor-intensive, the working environment inside the warehouse is noisy and the space is limited, and the maintenance personnel have to perform repetitive high-intensity operations for a long time, resulting in high labor costs. The pressure of guarding at night and during peak hours is particularly prominent. Fifth, the fault early warning capability is lagging behind. Manual acceptance can only complete a single static qualification check, and cannot compare the equipment operation trend. It is difficult to identify intermittent and hidden early faults, resulting in equipment leaving the warehouse with faults, which affects driving safety.

[0004] To improve acceptance efficiency, some locomotive depots have introduced semi-automatic testing devices, which use fixed fixtures to automatically test certain items. However, these devices still have significant technical limitations: First, the docking compatibility is poor. Fixed docking fixtures have extremely high requirements for locomotive parking accuracy, while locomotives in the depot generally have large positional deviations when braking and parking, often requiring manual assistance for alignment, making it impossible to achieve truly unmanned docking. Secondly, the machine has weak compatibility. The detection parameters and processes are preset and fixed, and it can only be adapted to a single model of monitoring device. When facing scenarios where multiple models of equipment such as LKJ series and ATP series are mixed and operated, manual switching of configuration is required, which is inefficient. Third, the fault identification dimension is singular, only able to detect whether electrical parameters are qualified or not, unable to identify hidden or intermittent faults inside the device, and the early warning capability is insufficient; Fourth, the system is highly independent, and the detection data is disconnected from the preparation and scheduling system. Unqualified vehicles still need to be manually dispatched and transferred, which limits the overall efficiency improvement of the preparation process and makes it difficult to form a closed-loop management system.

[0005] Therefore, there is an urgent need for a fully automated warehouse acceptance solution that can achieve full-process automation, adapt to multi-machine operation, have the ability to identify hidden faults, and can be linked to preparation and scheduling. Summary of the Invention

[0006] To address the problems of low automation, poor model compatibility, delayed fault warning, non-standard data management, and disconnected maintenance processes in the existing locomotive monitoring device in-warehouse acceptance process, this invention provides a fully automated in-warehouse acceptance system and method for locomotive monitoring devices. Through the integrated application of cross-domain technologies, it achieves unmanned operation of the entire acceptance process, improves the consistency of testing and data management, strengthens the ability to identify hidden faults, and links maintenance scheduling to form a closed-loop control system, meeting the operational needs of high-frequency locomotive turnover.

[0007] The technical solution adopted by the present invention to solve its technical problem is: a fully automatic in-warehouse acceptance system for locomotive monitoring devices, including a vehicle number recognition and in-warehouse triggering unit, an adaptive alignment detection docking unit, a multi-source detection execution unit, an edge computing processing unit, an intelligent judgment and early warning unit, and a data archiving unit and a preparation scheduling linkage unit respectively connected to the intelligent judgment and early warning unit; The vehicle number recognition and entry triggering unit is used to collect images of locomotives entering the warehouse, identify the vehicle number and the monitoring device model, and generate a detection start signal. The adaptive alignment detection and docking unit includes an industrial vision module, a six-degree-of-freedom flexible robotic arm, and a multi-specification probe module. It is used to adjust the posture of the robotic arm according to the visual recognition results of the monitoring device interface panel and complete the automatic docking of the detection probe and the monitoring device interface within a preset parking deviation range. The multi-source detection execution unit is used to output detection excitation to the monitoring device through the docked probe, and simultaneously collect multi-source time-series data of electrical parameters, device vibration and acoustic signals; The edge computing processing unit is deployed locally at the inspection station to preprocess inspection data in real time, cache data when the network is interrupted, and resume transmission after the network is restored. The intelligent judgment and early warning unit has a built-in model adaptive detection model based on federated learning training, which is used to match the acceptance standard threshold of the corresponding model, compare the detection data and output the qualified judgment result, and at the same time integrate multi-source time series features to identify hidden faults and generate early warning information. The preparation and scheduling linkage unit is used to synchronize the qualified / unqualified results to the locomotive preparation and scheduling system, automatically assign maintenance work positions to unqualified locomotives, and trigger the re-inspection process after the maintenance is completed, thus forming an acceptance closed loop. The data archiving unit is used to store raw detection data, judgment results and alarm logs, and to generate electronic acceptance reports and equipment history.

[0008] Specifically, the adaptive alignment detection docking unit also includes a contact force sensor, which is installed at the end of the probe module and is used to provide real-time feedback of the probe contact pressure during the docking process, control the insertion force within a preset range, and avoid damage to the interface hardware.

[0009] Specifically, the training process of the aircraft type adaptive detection model is as follows: the edge nodes of the detection workstations of each locomotive depot train the local model based on local historical detection data, and only upload the model parameters to the central node. The central node aggregates the global model parameters using the FedAvg algorithm and then distributes them to each edge node. Under the premise that the original detection data does not leave the domain, the adaptation and iteration of multi-aircraft type detection standards is completed.

[0010] Specifically, the intelligent judgment and early warning unit identifies latent faults by extracting the time and frequency domain features of power supply ripple fluctuations, communication message delays, and interface contact resistance, combining the device housing vibration amplitude and acoustic spectrum features, performing similarity matching with the historical fault sample library, and outputting the early warning level and fault location of intermittent faults.

[0011] Specifically, the edge computing processing unit has a built-in local judgment engine, which can independently complete the qualification judgment and result storage of all inspection items without cloud connection, ensuring that the acceptance operation is not interrupted when the network is disconnected.

[0012] Specifically, the multi-specification probe module integrates a power interface probe, a communication bus interface probe, an IO interface probe, and an analog interface probe, and is compatible with the interface specifications of at least three different types of locomotive monitoring devices.

[0013] A fully automated in-depot acceptance method for locomotive monitoring devices, implemented using the aforementioned fully automated in-depot acceptance system for locomotive monitoring devices, includes the following steps: S1. Warehouse entry trigger: When the locomotive enters the inspection station, the vehicle number and the model of the on-board monitoring device are automatically identified, and the acceptance process is started. S2. Adaptive docking: Acquire images of the monitoring device interface panel, calculate the deviation between the actual position of the interface and the reference position, drive the flexible robotic arm to adjust its posture, and complete the automatic docking of the detection probe and the device interface. S3. Multi-source detection: Output test excitation according to the acceptance specifications of the corresponding model, and simultaneously collect electrical, vibration and acoustic multi-source time-series test data; S4. Local processing and judgment: The local edge node of the workstation preprocesses the detection data, calls the adaptive detection model to compare with the standard threshold, and outputs the qualified judgment result and hidden fault warning. S5. Closed-loop linkage: The test results are synchronized to the preparation and dispatch system. Qualified locomotives are automatically released, and unqualified locomotives are automatically dispatched to the maintenance station. After the maintenance is completed, a re-inspection is automatically triggered. S6. Data Archiving: Test data, judgment results and logs are automatically uploaded and archived to generate electronic acceptance reports and equipment history.

[0014] Specifically, in step S2, when the detected locomotive parking position deviation is within ±50mm, the robotic arm is automatically docked by visual servo control; if the deviation exceeds the threshold, a parking adjustment prompt is issued, and docking is re-executed after the locomotive is adjusted to the correct position.

[0015] Specifically, in step S4, if a network interruption is detected, the detection data and judgment results are temporarily stored on the local edge node. After the network is restored, the data will continue to be uploaded from the point of interruption without retransmitting the completed data, thus ensuring that the data is complete and not lost.

[0016] Specifically, in step S5, the re-inspection process automatically calls the fault location data from the first inspection, performs targeted inspection, updates the equipment history and releases it after passing the re-inspection, and transfers it for repair again if it fails, thus forming a closed-loop management system.

[0017] The beneficial effects of this invention are: The automation level of docking has been greatly improved. By introducing visual servo flexible docking technology from the field of industrial robots, it can adapt to locomotive parking deviations of ±50mm. The docking success rate has been increased from less than 60% with traditional fixed tooling to more than 99%, truly realizing unmanned automatic docking without the need for manual alignment.

[0018] It has strong multi-model adaptability and adopts a federated learning framework to train an adaptive detection model. Under the premise of meeting the railway data security requirements, it can automatically adapt to various types of monitoring devices and different maintenance process acceptance standards without manual configuration switching, thus improving detection efficiency in mixed operation scenarios.

[0019] The fault identification features a rich set of dimensions, integrating time-series data from multiple sources including electrical, vibration, and acoustic sources for analysis. This effectively identifies intermittent and latent early-stage faults, improving the latent fault identification rate compared to traditional single electrical testing and significantly reducing the risk of equipment leaving the warehouse with faults.

[0020] Data management is reliable and efficient, adopting an edge computing architecture. Local real-time processing ensures real-time detection, and the breakpoint resume mechanism ensures complete data archiving in complex network environments within the database. Electronic ledgers are automatically generated, supporting full lifecycle traceability and meeting the requirements of digital management.

[0021] The preparation process is a closed-loop linkage, with deep integration between the acceptance system and the preparation scheduling system. It automatically completes vehicle diversion and re-inspection triggering, improving the preparation turnover efficiency compared to the manual mode, greatly reducing the labor intensity of on-site personnel, and supporting 24-hour unattended operation. Attached Figure Description

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] Figure 1 This invention provides an architecture diagram of a fully automated in-warehouse acceptance system for locomotive monitoring devices; Figure 2 The flowchart illustrates a fully automated in-depot acceptance method for a locomotive monitoring device provided by this invention. Detailed Implementation

[0024] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0025] like Figure 1 As shown, the fully automated locomotive monitoring device in the present invention includes a locomotive number recognition and entry triggering unit, an adaptive alignment detection docking unit, a multi-source detection execution unit, an edge computing processing unit, an intelligent judgment and early warning unit, and a data archiving unit and a preparation scheduling linkage unit respectively connected to the intelligent judgment and early warning unit.

[0026] Vehicle Number Recognition and Entry Trigger Unit: This unit consists of a high-definition imaging camera and a vehicle number recognition algorithm module, installed at the entrance of the inspection station. When a locomotive enters the inspection station, the camera captures an image of the locomotive's side vehicle number. The locomotive number is then parsed using a character recognition algorithm. This parsing is combined with a pre-stored locomotive equipment ledger to match the model and maintenance information of the onboard monitoring device, generating an inspection start command and triggering the subsequent acceptance process. The entire process requires no manual intervention.

[0027] Adaptive Alignment Detection and Docking Unit: This unit is the core docking execution module, introducing visual servoing and flexible docking technologies from the field of industrial robots to solve the automatic docking problem caused by large parking deviations of locomotives. The unit includes an industrial vision module, a six-degree-of-freedom flexible collaborative robotic arm, multi-specification probe modules, and contact force sensors.

[0028] The industrial vision module is installed at the end of the robotic arm or on the side of the workstation to acquire images of the interface panels of the locomotive monitoring device. It uses feature recognition algorithms to locate the actual coordinates of each interface and calculates its three-dimensional deviation from the preset reference coordinates. The six-degree-of-freedom flexible robotic arm adjusts its posture in real time according to the deviation, moving the probe module closer to the interface. The multi-specification probe module integrates various specifications of spring-loaded probes, including power interfaces, communication bus interfaces, I / O interfaces, and analog interfaces, and is compatible with mainstream monitoring device interfaces such as the LKJ series and ATP series. A contact force sensor is installed at the connection end between the probe module and the robotic arm, providing real-time feedback on the contact pressure during docking, controlling the insertion force within a safe range of 5~10N, ensuring reliable contact while preventing damage to the device interface. This unit can achieve fully automatic docking within a locomotive parking position deviation of ±50mm, significantly reducing the requirements for locomotive parking accuracy.

[0029] Multi-source detection execution unit: This unit is a detection excitation and data acquisition module. In addition to traditional electrical parameter detection, it introduces vibration and acoustic detection dimensions to provide a data foundation for hidden fault identification. The unit includes a detection excitation module and a multi-channel data acquisition module: The detection excitation module outputs various detection excitations such as power on / off, communication commands, and operating condition simulations to the monitoring device according to the acceptance specifications of the corresponding model; The multi-channel data acquisition module simultaneously acquires electrical timing data such as power supply voltage, ripple, communication messages, and interface resistance of the monitoring device, and at the same time, it acquires vibration signals from the device shell and internal operating noise signals through miniature vibration sensors and acoustic sensors to form a multi-source timing detection dataset.

[0030] Edge Computing Processing Unit: This unit adopts an edge computing architecture from the industrial IoT field, deployed in the local edge gateway of each inspection station, undertaking data preprocessing, local judgment, and breakpoint resume functions. Inspection data first undergoes preprocessing operations such as filtering and feature extraction at the local edge node, reducing the amount of data transmitted to the cloud; a built-in local judgment engine can independently complete the pass / fail judgment of all inspection items without a cloud network connection, ensuring uninterrupted operation during network outages; configured with large-capacity local storage, when the wireless network is interrupted due to the shielded environment within the warehouse, all inspection data and results are automatically cached, and the upload continues from the breakpoint after the network is restored, without repeating the transmission of completed data, ensuring complete data archiving and solving the problem of unstable communication in the complex environment within the warehouse.

[0031] Intelligent Judgment and Early Warning Unit: This unit has a built-in adaptive detection model based on federated learning training, and has dual functions of qualification judgment and hidden fault early warning.

[0032] The model-adaptive detection model is trained using a federated learning framework: edge nodes of each locomotive depot train their local models based on local historical detection data, uploading only the updated model parameters to the central node. The original detection data is always stored locally, meeting the data security and confidentiality requirements of the railway industry. The central node uses the FedAvg average aggregation algorithm to weight and aggregate the parameters uploaded by each node, generating an updated global model and distributing it to each edge node. Through multiple iterations, the model can automatically adapt to the detection standards of monitoring devices of different models and maintenance procedures, without the need for manual parameter configuration.

[0033] The latent fault early warning function is achieved through the fusion of multi-source time-series features: extracting electrical time-domain and frequency-domain features such as power supply ripple fluctuation features, communication message delay distribution, and interface contact resistance change trends, and combining them with device vibration amplitude and acoustic spectrum features, and performing similarity matching with the historical fault sample library, it can identify latent problems that are difficult to detect by a single static test, such as intermittent power module faults, poor interface contact, and aging of internal components, and output the early warning level and suggested fault location.

[0034] The maintenance and scheduling linkage unit: This unit interfaces with the existing maintenance and scheduling system of the locomotive depot through a standard interface to achieve a closed-loop linkage between acceptance operations and the maintenance process. After the inspection is completed, the pass / fail results and fault location information are automatically synchronized to the scheduling system; for locomotives that pass the inspection, the maintenance status is automatically updated and they are released to the outbound waiting line; for locomotives that fail the inspection, alarm information is automatically triggered, and maintenance workstations are automatically allocated according to the current occupancy of maintenance workstations, and fault details are pushed to the maintenance terminal; after the maintenance is completed, the system automatically triggers the re-inspection process, and after the re-inspection is passed, the equipment status is updated and the locomotive is released, forming a closed-loop control of the entire process of "inspection - alarm - maintenance - re-inspection".

[0035] Data archiving unit: This unit connects to the cloud data center, automatically stores raw detection data, judgment results, alarm logs and operation records, generates standardized electronic acceptance reports, automatically links locomotive and equipment history, and supports multi-dimensional query, traceability and statistical analysis by car number, time, fault type and other dimensions, replacing traditional paper ledgers and meeting the requirements of digital management.

[0036] like Figure 2 As shown, the fully automated in-depot acceptance method for the corresponding locomotive monitoring device of the present invention includes the following steps: S1. Warehouse entry trigger: When the locomotive enters the inspection station, the vehicle number recognition unit collects images and parses the vehicle number, matches the monitoring device model and maintenance process, and automatically starts the acceptance process. S2, Adaptive docking: The industrial vision module acquires images of the interface panel of the monitoring device, locates the interface coordinates and calculates the position deviation, drives the six-degree-of-freedom flexible robotic arm to adjust its posture, and completes the automatic docking of the detection probe and the device interface under the contact force feedback control. S3, Multi-source detection: Calls the acceptance test script for the corresponding model, outputs various test excitations, and synchronously collects electrical parameters, vibration, and acoustic multi-source time-series test data; S4. Local processing and judgment: The edge node preprocesses the detection data locally, calls the adaptive detection model, compares it with the corresponding standard threshold, outputs the qualified judgment result, and integrates multi-source features to identify hidden faults and generate early warnings. S5. Closed-loop linkage: The test results are synchronized to the preparation and dispatching system. Qualified locomotives are automatically released, and unqualified locomotives are automatically dispatched to the maintenance station. After the maintenance is completed, the directional re-inspection is automatically triggered. S6. Data Archiving: The original detection data, judgment results and logs are automatically uploaded and archived, generating electronic acceptance reports and updating the locomotive equipment history.

[0037] Example 1: Fully Automated Acceptance System for Standard Workstations of a Single Locomotive Model: This example is applied to a fixed inspection workstation in a locomotive depot to perform fully automated acceptance of freight locomotives equipped with LKJ2000 monitoring devices.

[0038] The acceptance system in this embodiment includes a vehicle number recognition and entry triggering unit, an adaptive alignment detection docking unit, a multi-source detection execution unit, an edge computing processing unit, an intelligent judgment and early warning unit, a data archiving unit, and a preparation and scheduling linkage unit. The vehicle number recognition and entry triggering unit uses two 5-megapixel high-definition industrial cameras, which are installed on both sides of the entrance to the inspection station, with their height aligned with the position of the locomotive number mark. When the locomotive enters the station at a speed not exceeding 5 km / h, the cameras continuously acquire images, and the locomotive number is parsed through an OCR character recognition algorithm. The system then matches the information of the LKJ2000 monitoring device and the corresponding maintenance procedure on the locomotive from the locomotive depot's equipment ledger, generates a detection start command, triggers the system to power on for self-testing, and enters the docking-ready state.

[0039] The adaptive alignment detection and docking unit adopts a 6-degree-of-freedom collaborative robotic arm, which is mounted on the mounting bracket on the side of the detection station. An industrial vision camera and a multi-specification probe module are installed at the end of the robotic arm. The probe module is equipped with power probes, RS485 communication probes and IO interface probes adapted to the LKJ2000 device. The end of the probe is a spring-loaded structure, and a high-precision contact force sensor is installed at the root of the probe module. After the locomotive stops stably, the industrial vision camera captures the interface panel of the monitoring device host, locates the center coordinates of each interface through the feature point matching algorithm, and calculates the X, Y, and Z three-axis deviations from the system's preset reference coordinates; the robotic arm adjusts its pose in real time according to the deviation amount, driving the probe module to approach the interfaces. When the contact force sensor detects that the contact force reaches 3N, it decelerates and advances, and stops advancing and locks the position until the contact force reaches 8N to complete the docking. Tests show that when the locomotive parking position deviation is within the range of ±50mm, the single docking success rate can reach 99.2%, and the average docking time does not exceed 15 seconds.

[0040] After the docking is completed, the multi-source detection execution unit sequentially executes detection items in accordance with the depot inspection specifications for the LKJ2000 type monitoring device: including power supply voltage fluctuation test, communication interface connectivity test, program version verification, working condition simulation test, insulation performance test, etc. During the detection process, electrical data such as power supply ripple waveform, communication message timing, and interface contact resistance are collected synchronously, and the operation vibration signal of the device is collected through a micro vibration sensor attached to the device housing. All collected data is transmitted to the local edge computing processing unit in real time. After the edge gateway performs filtering, noise reduction and feature extraction on the original data, it calls the built-in LKJ2000 type detection model for real-time comparison and judgment, and automatically generates a pass / fail conclusion after all detection items are completed.

[0041] The detection results are synchronized to the depot servicing scheduling system through the servicing scheduling linkage unit: if the judgment is qualified, the system automatically updates the servicing status of the locomotive to "accepted and qualified", and the scheduling system automatically releases it to the outgoing waiting line; if the judgment is unqualified, the system automatically pops up an alarm message, pushes the fault items and positioning information to the maintenance team terminals at the same time, and automatically assigns an idle maintenance station. After the maintenance personnel complete the fault handling, they submit a maintenance completion instruction at the terminal, the system automatically triggers a re-inspection process. The re-inspection only conducts targeted detection for the fault items, and updates the status and releases after being qualified. All original detection data, judgment results and operation logs are automatically uploaded to the cloud data archiving unit to generate a standardized electronic acceptance report, which is associated with the equipment record file of the corresponding locomotive to support subsequent query, statistics and analysis.

[0042] Comparative Example 1: This comparative example uses the traditional manual acceptance method. The acceptance of a freight locomotive with the same LKJ2000 monitoring device was conducted in the depot, and all operations were completed by two qualified maintenance personnel. During the operation, the maintenance personnel carried the testing equipment to the locomotive cab, manually connected the testing cables, operated the testing software item by item, manually checked each parameter, and manually filled out a paper acceptance log after the test was completed.

[0043] On-site measurements showed that manual inspection of a single locomotive took an average of about 28 minutes. During peak preparation periods, due to insufficient personnel, the waiting and operation time for a single locomotive could exceed 40 minutes. The inspection results were affected by the condition of the personnel, and sampling statistics showed that the missed inspection rate was about 3.2%, while the consistency of judgments among different shifts was about 85%. The average time to fill out paper ledgers was about 5 minutes, and there were problems such as filling errors and lost ledgers. Data summarization and statistics required manual re-entry, which was inefficient. Moreover, manual inspection could only complete static qualification judgments and could not identify intermittent hidden faults.

[0044] Compared with Comparative Example 1, the total acceptance time for a single unit in this embodiment is reduced to less than 8 minutes, and the work efficiency is improved by more than 250%; the consistency of detection and judgment reaches 100%, and the missed detection rate is reduced to less than 0.1%; no on-site personnel are required to operate, and 24-hour unattended operation can be achieved; data is automatically archived, and the accuracy and traceability are significantly improved.

[0045] Example 2: Adaptive Acceptance in Mixed Operation Scenarios of Multiple Aircraft Types: This example is applied to the comprehensive inspection station of a passenger locomotive depot. This station is responsible for the acceptance of locomotives with multiple monitoring devices such as LKJ-15 and ATP, with mixed aircraft types and diverse repair processes.

[0046] Based on Embodiment 1, the system in this embodiment focuses on optimizing the multi-specification probe module and the federated learning adaptive detection model. The multi-specification probe module adopts a switchable probe head design, integrating interface probes adapted to various mainstream monitoring devices such as LKJ-15 and ATP. The robotic arm can automatically switch the corresponding probe head according to the identified device model, without the need for manual replacement.

[0047] The intelligent judgment and early warning unit's built-in adaptive detection model is built on a federated learning framework and deployed at the section-level central node and the edge nodes of each workstation. The model training and iteration process is as follows: The edge nodes of each detection workstation train locally based on the corresponding model's accumulated detection data. The training process is completed only locally, and the original detection data does not leave the workstation node. After every 10 rounds of local training, the edge nodes encrypt the updated model parameters and upload them to the section-level central node. The central node receives the parameters uploaded by each workstation, performs global aggregation using the FedAvg weighted average algorithm, and generates an updated global model. After the global model is verified, it is distributed to the edge nodes of each workstation, completing one model iteration. Through continuous iteration, the model can automatically adapt to the detection items, threshold standards, and maintenance requirements of different monitoring devices, without requiring technicians to manually configure parameters one by one.

[0048] During operation, the vehicle number recognition unit identifies the locomotive number and automatically matches the model of the monitoring device on the locomotive with the current maintenance procedure. The system automatically calls the corresponding detection script and judgment model. For example, when the locomotive is identified as having an ATP-type monitoring device, the robotic arm automatically switches the ATP interface probe and performs dedicated testing items such as speed sensor interface testing, brake interface testing, and wireless communication testing according to the acceptance specifications of the ATP device. The judgment model automatically uses the standard threshold corresponding to the ATP for comparison. During the detection process, edge nodes synchronously collect data. If special data characteristics of a new model of equipment are detected, they are automatically included in the local training set and regularly participate in global model iteration to continuously improve the model's adaptability.

[0049] This embodiment also supports adaptive adjustment of repair process. For different repair processes such as auxiliary repair, minor repair, and medium repair, the model automatically matches the corresponding test item range and pass threshold, without the need for manual switching of test schemes.

[0050] Comparative Example 2: This comparative example uses a traditional fixed-parameter automatic detection system. This system presets detection parameters and thresholds for a single model. When dealing with mixed operations of multiple models, technicians need to manually switch detection programs and change docking fixtures.

[0051] Actual testing revealed that this traditional system can only be adapted to one main monitoring device model. When dealing with other models, manual configuration and tooling switching takes an average of 12 minutes and is prone to parameter configuration errors, resulting in a false positive rate of 4.5%. For new models without preset parameters, automatic detection is completely impossible, requiring manual acceptance. Furthermore, because data cannot be shared between different locomotive depots, training new aircraft models requires the centralized collection of large amounts of data, which is time-consuming and poses data security risks.

[0052] Compared with Comparative Example 2, this embodiment can automatically adapt to at least 5 mainstream monitoring device models and 3 or more repair processes. The model switching takes zero time, the accuracy of detection parameter configuration is 100%, and the false judgment rate is reduced to below 0.3%. In the federated learning mode, the data does not leave the domain, which not only meets the data security requirements, but also can quickly improve the adaptability of new models through multi-node joint iteration, and shorten the model iteration cycle by more than 60%.

[0053] Example 3: Closed-loop acceptance control with latent fault early warning: This example is applied to the advanced maintenance and testing station of a high-speed railway locomotive depot, and focuses on the in-depot acceptance of the monitoring device of high-speed train, with an emphasis on strengthening the ability to identify latent faults and the closed-loop control of the whole process.

[0054] The multi-source detection execution unit in this embodiment is equipped with a higher-precision electrical acquisition channel, a high-sensitivity vibration sensor, and an acoustic sensor, enabling it to acquire multi-source time-series data with a higher sampling rate. During the detection process, in addition to conventional conformity testing items, 30 seconds of steady-state data of the device operation are simultaneously acquired, including the time-domain waveform of the power supply output ripple, the message delay sequence of the communication bus, the vibration spectrum of the internal cooling fan, and the operating acoustic signal of the power module.

[0055] The intelligent judgment and early warning unit incorporates a multi-source fusion fault early warning model. This model is pre-trained based on historical fault samples, extracting multi-dimensional features corresponding to various latent faults: for example, aging power module capacitors correspond to increased low-frequency ripple components, poor interface contact corresponds to increased communication latency jitter, and loose internal connectors correspond to abnormal peaks in the vibration spectrum. During detection, the model extracts features from the collected multi-source time-series data and performs cosine similarity matching between the extracted feature vectors and the fault sample library. When the similarity exceeds a preset threshold, it outputs a latent fault warning of the corresponding type, along with the warning level and suggested inspection points. Warnings are divided into three levels: Level 1 warning indicates a serious hidden danger, prohibiting removal from the warehouse; Level 2 warning indicates an early-stage fault, recommending planned repairs; and Level 3 warning indicates a minor anomaly, requiring close monitoring.

[0056] The preparation and scheduling linkage unit performs tiered scheduling based on the judgment results and warning levels: locomotives that pass acceptance and have no warnings are directly released from the depot; for locomotives with level two or three warnings, the warning information is automatically entered into the equipment history and pushed to the operation and maintenance departments, included in the tracking plan, without affecting the current release; for locomotives with level one warnings or that fail inspection, the release permission is automatically locked, and the scheduling system automatically diverts them to the maintenance station, pushing fault details and troubleshooting suggestions. After maintenance is completed, the system automatically triggers a full-item re-inspection, and after confirming that the fault has been eliminated and there are no hidden warnings, the lock is released and the history is updated. For level three warning locomotives included in the tracking, the corresponding items are automatically retested during each subsequent entry acceptance to track the fault development trend.

[0057] The data archiving unit establishes a full lifecycle equipment health record, integrates the test data and early warning records of each acceptance test, and generates equipment health trend curves to provide data support for condition-based maintenance.

[0058] Comparative Example 3: This comparative example uses a traditional automatic acceptance system that only has the function of determining whether a parameter is within the acceptable threshold range. This system only checks whether the parameters are within the acceptable threshold range and does not have the ability to identify hidden faults.

[0059] Statistics show that among locomotives that passed inspection and left the warehouse using the traditional system, approximately 1.8% of them triggered alarms en route due to latent faults in the monitoring devices. These were mostly intermittent power module failures and poor interface contact. Such faults were difficult to detect because parameters were still within acceptable ranges during a single static inspection. After a fault occurred, temporary repairs were required, disrupting transportation plans and posing safety hazards. Furthermore, the traditional system's inspection results were independent and disconnected from the dispatching process. Unqualified vehicles required manual coordination and dispatching, with an average turnaround time of approximately 25 minutes, resulting in low closed-loop efficiency in preparation.

[0060] Compared with Comparative Example 3, the accuracy rate of latent fault identification in this embodiment can reach over 85%, which can detect the vast majority of early latent faults in advance. The alarm rate caused by monitoring device failure during the journey is reduced by over 75%. The entire process of detection and scheduling is automatically linked, and the average turnover time of unqualified vehicles is shortened to less than 8 minutes. It truly realizes the closed-loop management of "detection-alarm-repair-re-inspection-tracking", which significantly improves the reliability of equipment operation and the efficiency of maintenance management.

[0061] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of protection claimed by the present invention. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A fully automated in-warehouse acceptance system for locomotive monitoring devices, characterized in that, It includes a vehicle number recognition and entry triggering unit, an adaptive alignment detection docking unit, a multi-source detection execution unit, an edge computing processing unit, an intelligent judgment and early warning unit, and a data archiving unit and a preparation scheduling linkage unit that are connected to the intelligent judgment and early warning unit respectively. The vehicle number recognition and entry triggering unit is used to collect images of locomotives entering the warehouse, identify the vehicle number and the monitoring device model, and generate a detection start signal. The adaptive alignment detection and docking unit includes an industrial vision module, a six-degree-of-freedom flexible robotic arm, and a multi-specification probe module. It is used to adjust the posture of the robotic arm according to the visual recognition results of the monitoring device interface panel and complete the automatic docking of the detection probe and the monitoring device interface within a preset parking deviation range. The multi-source detection execution unit is used to output detection excitation to the monitoring device through the docked probe, and simultaneously collect multi-source time-series data of electrical parameters, device vibration and acoustic signals; The edge computing processing unit is deployed locally at the inspection station to preprocess inspection data in real time, cache data when the network is interrupted, and resume transmission after the network is restored. The intelligent judgment and early warning unit has a built-in model adaptive detection model based on federated learning training, which is used to match the acceptance standard threshold of the corresponding model, compare the detection data and output the qualified judgment result, and at the same time integrate multi-source time series features to identify hidden faults and generate early warning information. The preparation and scheduling linkage unit is used to synchronize the qualified / unqualified results to the locomotive preparation and scheduling system, automatically assign maintenance work positions to unqualified locomotives, and trigger the re-inspection process after the maintenance is completed, thus forming an acceptance closed loop. The data archiving unit is used to store raw detection data, judgment results and alarm logs, and to generate electronic acceptance reports and equipment history.

2. The fully automated in-depot acceptance system for locomotive monitoring devices according to claim 1, characterized in that: The adaptive alignment detection docking unit also includes a contact force sensor, which is installed at the end of the probe module and is used to provide real-time feedback on the probe contact pressure during the docking process, so as to control the insertion force within a preset range and avoid damage to the interface hardware.

3. The fully automated in-depot acceptance system for locomotive monitoring devices according to claim 1, characterized in that: The training process of the adaptive detection model for the aircraft type is as follows: the edge nodes of the detection workstations of each locomotive depot train the local model based on local historical detection data, and only upload the model parameters to the central node. The central node aggregates the global model parameters using the FedAvg algorithm and then distributes them to each edge node. Under the premise that the original detection data does not leave the domain, the adaptation and iteration of the multi-aircraft detection standard is completed.

4. The fully automated in-depot acceptance system for locomotive monitoring devices according to claim 1, characterized in that: The intelligent judgment and early warning unit identifies latent faults by extracting the time and frequency domain features of power ripple fluctuations, communication message delays, and interface contact resistance, combining them with the vibration amplitude and acoustic spectrum features of the device casing, performing similarity matching with the historical fault sample library, and outputting the early warning level and fault location of intermittent faults.

5. The fully automated in-depot acceptance system for locomotive monitoring devices according to claim 1, characterized in that: The edge computing processing unit has a built-in local judgment engine, which can independently complete the qualification judgment and result storage of all test items without cloud connection, ensuring that the acceptance operation is not interrupted when the network is disconnected.

6. The fully automated in-depot acceptance system for locomotive monitoring devices according to claim 1, characterized in that: The multi-specification probe module integrates power interface probes, communication bus interface probes, IO interface probes and analog interface probes, and is compatible with the interface specifications of at least three different locomotive monitoring devices.

7. A fully automated in-depot acceptance method for locomotive monitoring devices, wherein the method is implemented using the fully automated in-depot acceptance system for locomotive monitoring devices as described in any one of claims 1-6, characterized in that, Includes the following steps: S1. Warehouse entry trigger: When the locomotive enters the inspection station, the vehicle number and the model of the on-board monitoring device are automatically identified, and the acceptance process is started. S2. Adaptive docking: Acquire images of the monitoring device interface panel, calculate the deviation between the actual position of the interface and the reference position, drive the flexible robotic arm to adjust its posture, and complete the automatic docking of the detection probe and the device interface. S3. Multi-source detection: Output test excitation according to the acceptance specifications of the corresponding model, and simultaneously collect electrical, vibration and acoustic multi-source time-series test data; S4. Local processing and judgment: The local edge node of the workstation preprocesses the detection data, calls the adaptive detection model to compare with the standard threshold, and outputs the qualified judgment result and hidden fault warning. S5. Closed-loop linkage: The test results are synchronized to the preparation and dispatch system. Qualified locomotives are automatically released, and unqualified locomotives are automatically dispatched to the maintenance station. After the maintenance is completed, a re-inspection is automatically triggered. S6. Data Archiving: Test data, judgment results and logs are automatically uploaded and archived to generate electronic acceptance reports and equipment history.

8. The fully automated in-depot acceptance method for locomotive monitoring devices according to claim 7, characterized in that: In step S2, when the detected locomotive parking position deviation is within ±50mm, the robotic arm is automatically docked by visual servo control; if the deviation exceeds the threshold, a parking adjustment prompt is issued, and docking is re-executed after the locomotive is adjusted to the correct position.

9. A fully automated in-depot acceptance method for locomotive monitoring devices according to claim 7, characterized in that: In step S4, if a network interruption is detected, the detection data and judgment results are temporarily stored on the local edge node. After the network is restored, the data will continue to be uploaded from the point of interruption without retransmitting the completed data, thus ensuring that the data is complete and not lost.

10. A fully automated in-depot acceptance method for locomotive monitoring devices according to claim 7, characterized in that: In step S5, the re-inspection process automatically calls the fault location data from the first inspection, performs targeted inspection, updates the equipment history and releases it after passing the re-inspection, and transfers it for repair again if it fails, thus forming a closed-loop management system.