CN-121981939-A - Automatic switch machine switch defect detection method and system based on image recognition

CN121981939ACN 121981939 ACN121981939 ACN 121981939ACN-121981939-A

Abstract

The invention discloses a defect detection method and a defect detection system for an automatic switch machine shutter based on image recognition, which relate to the technical field of rail traffic detection and comprise the following steps of S1, collecting an automatic switch machine shutter image in real time; S2, preprocessing, feature extraction and defect identification are carried out on the automatic switch machine shutter image, in the feature extraction and defect identification, defect factors are detected based on a machine learning or deep learning algorithm, the defect type of the automatic switch machine shutter is identified, and a defect identification result is output, and S3, image data and the defect identification result are transmitted to a data center or a user terminal. The invention realizes the rapid, accurate and comprehensive detection of the defect state of the automatic switch of the switch machine, makes up the defects of the prior art in the aspects of efficiency, accuracy and detection range, improves the maintenance efficiency of switch equipment, reduces the labor cost and improves the safety of rail transit operation.

Inventors

ZHAN DONG
YAO QIN
SUN JIAJIE
CAO WEI
LIU HANG

Assignees

西南交通大学
成都唐源电气股份有限公司

Dates

Publication Date: 20260505
Application Date: 20251205

Claims (10)

1. The defect detection method for the automatic switch machine switch based on image recognition is characterized by comprising the following steps of: collecting an automatic switch machine shutter image in real time; preprocessing, feature extraction and defect identification are carried out on the automatic switch machine shutter image, wherein the feature extraction and defect identification adopt a machine learning or deep learning algorithm to detect defect factors, identify defect types of the automatic switch machine shutter, and output defect identification results to a data center or a user terminal for presentation and display; the defects of the automatic switch device comprise insufficient contact depth of the movable contact, contact clearance of the movable contact, missing of a movable contact ring, breakage or missing of a static contact, loosening of a mounting screw, incomplete cotter pin of the movable contact ring and arc discharge.
2. The image recognition-based automatic switch machine shutter defect detection method of claim 1, wherein the preprocessing, feature extraction and defect recognition of the automatic switch machine shutter image comprises: Graying, denoising and enhancing pretreatment are carried out on the collected original image of the automatic switch machine shutter; performing feature extraction and defect identification on the preprocessed automatic switch machine shutter image by adopting a machine learning or deep learning algorithm, and identifying various defect types on the automatic switch machine shutter image; and clearly and intuitively displaying and displaying the defect identification result on the data center or the user side, wherein the defect identification result comprises marking the defect position, the defect type and the defect severity on a display interface.
3. The automatic switch machine defect detection method based on image recognition as claimed in claim 1, wherein the preprocessing comprises the steps of: Converting the color original image of the automatic switch machine shutter into a gray image; Denoising, namely removing noise in the gray level image by adopting a method comprising median filtering, gaussian filtering or bilateral filtering; image enhancement, namely enhancing the contrast of the gray image by adopting a method comprising histogram equalization or adaptive contrast enhancement.
4. The image recognition-based automatic switch machine shutter defect detection method of claim 1, wherein the defect recognition comprises: For the defect recognition of insufficient contact depth and contact gap of the movable contact and the static contact, measuring the relative position of the movable contact and the static contact through image analysis, judging whether the contact depth and the gap are in a normal range or not, and measuring by adopting an edge detection and feature matching method; For the defect identification of the missing of the movable contact ring, the breakage of the static contact or the defect identification, judging whether the movable contact ring has the missing or breakage or not by identifying the shape and the outline of the movable contact ring, and adopting a template matching and outline analysis method for identification; for the loosening defect identification of the mounting screw, the position and the shape of the mounting screw are identified, whether the loosening condition exists or not is judged, and the characteristic point detection and the shape identification method are adopted for judging; and identifying incomplete cotter pins and arc discharge defects of the movable contact point ring by adopting an image analysis method.
5. The method for detecting defects of automatic switch machine shutters based on image recognition as recited in claim 1, wherein the defect recognition results at least comprise defect type, defect position and defect degree.
6. The image recognition-based automatic switch machine defect detection method of claim 2, wherein the defect severity is judged based on a preset threshold, comprising: when the contact depth of the movable contact and the fixed contact is lower than a first threshold value, the severity of the defect is urgent, and the switch machine needs to be immediately stopped and the movable contact and the fixed contact are replaced; When the contact gap of the dynamic contact and the static contact exceeds a second threshold, the defect severity is serious, and maintenance and adjustment are needed within 2 hours; when the missing area of the movable contact ring or the static contact piece exceeds a third threshold value, the defect severity is great, and an emergency maintenance plan needs to be started; When the loosening displacement of the mounting screw exceeds a fourth threshold, the defect severity is moderate, and the mounting screw is fastened within 24 hours; when the cotter pin defect of the movable contact ring exceeds a fifth threshold value, the severity of the defect is a warning, and the cotter pin of the movable contact ring needs to be replaced preferentially in the next maintenance period; when the arcing duration exceeds a sixth threshold, the defect severity is critical, requiring power outage investigation and replacement of the damaged component.
7. The automatic switch machine defect detection method based on image recognition according to claim 6, wherein the first threshold is 70% of a standard value, the second threshold is 0.5mm, the third threshold is 30%, the fourth threshold is 1mm, the fifth threshold is 50%, and the sixth threshold is 100ms; The deep learning algorithm is selected from at least one of the following: a convolutional neural network; A transducer architecture based on an attention mechanism; the target detection algorithm YOLO series; a residual neural network; regional convolutional neural network series algorithms.
8. The automatic switch machine switch defect detection system based on image recognition is characterized by comprising an image acquisition unit, an image processing unit and a data transmission unit; The image acquisition unit is used for acquiring images of the automatic switch machine shutter in real time and is provided with a light supplementing lamp to adapt to different illumination conditions; The image processing unit is in communication connection with the image acquisition unit and is used for preprocessing, extracting features and identifying defects of an automatic switch device image, detecting defect factors based on machine learning or deep learning algorithm in the feature extraction and defect identification, identifying defect types of the automatic switch device image and outputting defect identification results, wherein the defect types of the automatic switch device image comprise insufficient contact depth of a dynamic contact, contact clearance of the dynamic contact, missing of a dynamic contact ring, breakage or missing of the static contact, loosening of an installation screw, incomplete cotter pin of the dynamic contact ring and arc discharge; the data transmission unit is in communication connection with the image processing unit and is used for transmitting the image data and the defect identification result to a data center or a user terminal.
9. The automatic switch machine defect detection system based on image recognition according to claim 8, wherein the image processing unit comprises an image preprocessing module, a defect recognition module and a result output module; the image preprocessing module is used for carrying out graying, denoising and enhancing preprocessing on the collected original image of the automatic switch machine shutter; The defect identification module is in communication connection with the image preprocessing module, and is used for carrying out feature extraction and defect identification on the preprocessed automatic switch machine shutter image based on a machine learning or deep learning algorithm to identify various defect types on the automatic switch machine shutter image; The result output module is in communication connection with the defect identification module and is used for outputting and presenting the defect identification result to a user in a clear and visual mode, and the result output module comprises marking the defect position, displaying the defect type and displaying the defect severity on a display interface.
10. The defect detection system of automatic switch machine shutter based on image recognition as claimed in claim 8, wherein the image acquisition unit adopts a high resolution image sensor, and the image acquisition unit is installed above the automatic switch machine shutter and can adjust the shooting angle according to the actual situation; the image processing unit adopts an image processing unit based on an embedded platform, is provided with a high-performance processor and a memory and is used for image preprocessing, feature extraction and defect identification operation, and the embedded platform and the image sensor perform high-speed data transmission; The data transmission unit communicates with the data center or the user terminal through a wired or wireless communication interface, and the wireless communication interface comprises a Wi-Fi, 4G/5G or LoRa module.

Description

Automatic switch machine switch defect detection method and system based on image recognition Technical Field The invention relates to the technical field of railway/urban rail transit detection, in particular to the field of railway electric service/signal professional detection, and more particularly relates to a method and a system for detecting defects of an automatic switch machine based on image recognition. Background Currently, in the field of rail transit, in order to ensure reliable operation of turnout equipment, condition monitoring and fault diagnosis are required for key components of the turnout equipment. Aiming at the automatic switch of the turnout switch machine, the existing state monitoring means mainly depend on the following modes: 1. Manual periodic visual inspection This is the most widely used way today. The inspection personnel visually inspect the appearance, the connecting parts and the like of the automatic switch of the switch machine according to a preset period to find out whether the defects of crack, looseness, foreign matters, poor contact of the contact points, insufficient contact depth of the movable contact points and the static contact points and the like exist. However, manual visual inspection has a number of drawbacks that are difficult to overcome: The efficiency is low, a great deal of manpower and time are consumed, the inspection period is long, and the real-time monitoring of the equipment state is difficult to realize. The method is easily influenced by subjective factors, and the accuracy of the detection result is greatly dependent on the experience, responsibility and the current physical state of the inspection personnel, so that risks of missed detection and misjudgment exist. The labor intensity is high, and particularly in severe weather or areas with complex geographical environments, the inspection work is very hard. It is difficult to find hidden defects, i.e., defects which are internal and fine defects or defects which can be revealed only when the device is in a moving state, and it is difficult to find the hidden defects effectively by manual visual inspection. 2. Sensor-based monitoring To improve the automation level of monitoring, some sensor technologies are also applied to monitor the state of the switch machine: the vibration sensor is generally arranged in the switch machine and is used for monitoring vibration conditions in the running process of the switch machine, judging whether abnormal vibration exists or not and assisting in identifying the switch passing. And the temperature and humidity sensor is also integrated on the data acquisition board and used for monitoring the temperature and humidity inside the switch machine and judging whether the working environment is normal or not. The current sensor is arranged on a switch motor lead-out wire and used for collecting current data when the switch acts and analyzing the running load condition of the switch. The sensor-based monitoring means can reflect the operating state of the switch machine to a certain extent, for example, judge whether loosening or abrasion of mechanical parts exists or not through vibration data, and judge whether overload and other problems exist or not through current data. However, these sensors mainly monitor the overall operation state of the switch machine, and it is difficult to directly and effectively identify the defect states of specific components of the automatic switch, such as missing moving contact rings, broken fixed contact pieces, loose screws, small opening defects, breakage of the switch housing, arcing, and the like. At present, although other types of equipment are applied to monitoring scenes of outdoor equipment of railway traffic turnout, an image recognition technology is applied to an automatic switch of a turnout switch machine so as to realize automatic recognition of defect states of the turnout switch machine, and the prior art is still blank. The existing monitoring means have the defects of efficiency and accuracy, or are difficult to effectively identify specific defects of the automatic shutter directly. In summary, in the aspect of defect detection of the automatic switch of the switch machine in the prior art, the defect detection mainly depends on manual periodic visual inspection, and has low efficiency, is easily affected by personnel experience and responsibility, has high labor intensity, cannot realize real-time monitoring, and is difficult to find fine or hidden defects. The sensor-based monitoring method, although improving the monitoring automation level to some extent, makes it difficult to directly and effectively identify the defective state of a specific component of the automatic shutter. Therefore, a technical scheme capable of overcoming the defects of the prior art and efficiently and accurately identifying the defect state of the automatic switch of the switch machine is needed, so that the maintenance ef