CN-121997179-A - Fault detection and diagnosis method for train purging robot

CN121997179ACN 121997179 ACN121997179 ACN 121997179ACN-121997179-A

Abstract

The invention discloses a fault detection and diagnosis method for a train purging robot, and relates to the technical field of rail transit maintenance. The method comprises the steps of collecting multisource operation data of a purging robot in real time, wherein the multisource operation data comprise equipment operation parameters, vibration data, temperature data and high-definition image data reflecting states of key parts of the equipment, extracting characteristics of the multisource operation data, inputting a characteristic data set into a convolutional neural network model which is pre-trained based on a multi-fault type sample and is adapted through a migration learning technology to conduct real-time reasoning, outputting fault types, positions and matching degrees, making decisions according to the matching degrees and triggering corresponding alarms, and forming closed-loop feedback after fault processing to optimize the model. The invention realizes real-time and accurate diagnosis and early warning of faults of the train purging robot, and remarkably improves the operation and maintenance efficiency and the equipment reliability.

Inventors

LIU SHAOYUAN
ZHANG XING
LUO GUANGXUN
WANG ZHIBIN
Luo guangan
HU ZHIKANG

Assignees

广东华能机电集团有限公司
湖大粤港澳大湾区创新研究院(广州增城)

Dates

Publication Date: 20260508
Application Date: 20260129

Claims (7)

1. The fault detection and diagnosis method for the train purging robot is characterized by comprising the following steps of: s1, acquiring multisource operation data of the purging robot in real time, wherein the multisource operation data comprise equipment operation parameters, vibration data, temperature data and high-definition image data reflecting the states of key parts of the equipment; S2, multi-mode feature extraction is carried out on the multi-source operation data to obtain a structured feature data set, wherein: extracting time domain statistical characteristics and frequency domain spectrum characteristics from the operation parameters of the equipment; performing wavelet packet transformation on the vibration data, and extracting energy characteristics of a fault characteristic frequency band associated with a preset fault type; identifying and positioning a device key component area in the image based on a semantic segmentation model for the high-definition image data, and extracting texture features and shape features in the area; S3, inputting the structured feature data set into a pre-trained fault recognition model for real-time reasoning, wherein the fault recognition model is a multi-fault parallel diagnosis model constructed based on a convolutional neural network, pre-trains through a labeling sample set containing a plurality of fault types, and adopts a transfer learning technology to finely tune a last full-connection layer of the model so as to adapt to target equipment; S4, carrying out hierarchical decision according to the characteristic matching degree, wherein if the matching degree is more than or equal to a first threshold value, determining that the fault is confirmed and triggering a primary alarm, and if the matching degree is between a second threshold value and the first threshold value, determining that the fault is suspected and triggering a secondary alarm, wherein the first threshold value is larger than the second threshold value; And S5, after triggering an alarm, generating and pushing alarm information comprising the fault position, the fault type and the processing suggestion, and after processing the fault, storing the characteristic data and the diagnosis result of the fault and the processing result in a correlated way to form a closed loop feedback sample for performing incremental learning optimization on the fault identification model.
2. The method according to claim 1, wherein in the step S1, the operation parameters of the device include fan motor current, fan rotation speed, and motor current and rotation speed of each joint of the mechanical arm, and the acquisition frequency of the high-definition image data is not lower than 15fps, and the resolution is not lower than 1920×1080.
3. The method according to claim 1, wherein in step S2, the semantic segmentation model is a U-net++ structure that introduces an attention mechanism, and the encoder section includes a channel attention module and a spatial attention module for enhancing feature weights of key component areas in a complex background.
4. The method of claim 1, wherein in step S3, the convolutional neural network comprises an input layer, at least three convolutional-pooling layer groups, a global average pooling layer and a fully-connected output layer which are sequentially connected, and wherein the migration learning technology specifically comprises freezing all weight parameters except the last fully-connected layer on the basis of a pre-training model, and performing fine tuning on the fully-connected layer by using normal operation data of target equipment.
5. The method according to claim 1, wherein in step S4, the first threshold is 95% and the second threshold is 80%.
6. The method according to claim 1 or 5, wherein in step S4, the triggering mode of the primary alarm includes starting a local audible and visual alarm, popping up a full-screen warning window at a control terminal, and pushing warning information to an operation and maintenance personnel mobile terminal, and the triggering mode of the secondary alarm includes controlling the flashing of a local rehmannia warning lamp and giving a text prompt at the control terminal, and if the degree of matching is not released or continuously increased within a preset period of 5 minutes, the primary alarm is automatically updated.
7. The method of claim 1, wherein in step S5, the incremental learning optimization specifically includes periodically adding closed loop feedback samples to a training data set, retraining full-connected layer weights of the failure recognition model, and retaining recognition capability of the original failure category by using a knowledge distillation technique.

Description

Fault detection and diagnosis method for train purging robot Technical Field The invention relates to the technical field of maintenance of rail transit vehicles, in particular to a fault detection and diagnosis method of a purging robot for rail transit vehicles such as subways and railways. Background In the running process of railway vehicles such as subways and railways, a large amount of pollutants such as dust, greasy dirt, metal scraps and the like can be accumulated on the vehicle bottom and the vehicle side. Traditionally, cleaning of these areas has relied primarily on manual purging with high pressure air streams, or the use of stationary, pass-through purging equipment. The mode has the defects of low operation efficiency, uneven cleaning effect, high energy consumption, potential safety hazard and the like. With the development of intelligent technology, intelligent purge robots are beginning to be applied to this field. However, the current intelligent purging device mainly focuses on cleaning path planning and dust removal execution, and the capability of real-time monitoring and fault early warning of the running state of the device is generally insufficient. Purge robots are typically operated in a humid, dusty, complex environment, and critical components (e.g., fans, robotic arms, piping, etc.) are subject to wear, blockage, or damage over extended periods of operation. Once a fault occurs, it often requires shutdown checks, affecting normal vehicle service plans, and possibly even causing greater loss due to the expansion of the fault. The prior art lacks an effective method capable of carrying out online fault detection and diagnosis on the purging robot in real time and accurately, and is difficult to realize the transition from 'post maintenance' to 'predictive maintenance'. Therefore, a method for accurately detecting and diagnosing the operation faults of the train purging robot in real time is needed to ensure the stable operation of equipment and improve the maintenance efficiency and the intelligent level. Disclosure of Invention Object of the invention The invention aims to overcome the defects of the prior art, provides a fault detection and diagnosis method for a train purging robot, and aims to realize real-time online monitoring of the running state of the purging robot and early and accurate identification and early warning of faults, so that the sudden fault rate of equipment is reduced, the unplanned downtime is reduced, and the operation and maintenance efficiency and safety are improved. (II) technical scheme In order to achieve the above purpose, the invention provides a fault detection and diagnosis method of a train purging robot, comprising the following steps: s1, acquiring multisource operation data of the purging robot in real time, wherein the multisource operation data comprise equipment operation parameters, vibration data, temperature data and high-definition image data reflecting the states of key parts of the equipment; S2, multi-mode feature extraction is carried out on the multi-source operation data to obtain a structured feature data set, wherein: extracting time domain statistical characteristics and frequency domain spectrum characteristics from the operation parameters of the equipment; performing wavelet packet transformation on the vibration data, and extracting energy characteristics of a fault characteristic frequency band associated with a preset fault type; identifying and positioning a device key component area in the image based on a semantic segmentation model for the high-definition image data, and extracting texture features and shape features in the area; S3, inputting the structured feature data set into a pre-trained fault recognition model for real-time reasoning, wherein the fault recognition model is a multi-fault parallel diagnosis model constructed based on a convolutional neural network, pre-trains through a labeling sample set containing a plurality of fault types, and adopts a transfer learning technology to finely tune a last full-connection layer of the model so as to adapt to target equipment; S4, carrying out hierarchical decision according to the characteristic matching degree, wherein if the matching degree is more than or equal to a first threshold value, determining that the fault is confirmed and triggering a primary alarm, and if the matching degree is between a second threshold value and the first threshold value, determining that the fault is suspected and triggering a secondary alarm, wherein the first threshold value is larger than the second threshold value; And S5, after triggering an alarm, generating and pushing alarm information comprising the fault position, the fault type and the processing suggestion, and after processing the fault, storing the characteristic data and the diagnosis result of the fault and the processing result in a correlated way to form a closed loop feedback sample for performing incremen