CN-122016801-A - Automatic wheel surface image acquisition and damage analysis system for locomotive non-drop turning repair scene

CN122016801ACN 122016801 ACN122016801 ACN 122016801ACN-122016801-A

Abstract

An automatic wheel surface image acquisition and damage analysis system for a locomotive non-drop turning repair scene comprises a mechanical module unit, an image acquisition unit, an algorithm analysis unit and a main control unit, wherein the units work cooperatively through physical connection, electrical connection and data connection. The image acquisition and analysis system of the invention is a systematic scheme formed by a large number of researches, debugging and collaborative optimization aiming at the core pain point of the railway locomotive maintenance scene. Through image acquisition, light source adaptation and algorithm analysis and the deep coupling of the main control scheduling unit, the problem of scene adaptation of a general technology is solved, and the systematic efficiency is improved through unit cooperation.

Inventors

LIU TONG
PAN BILIN
SUN YUDUO
WANG FEIER
CHENG YAPING
FENG YING
YANG WENSONG
WANG FENG
SHI TIANBO
LI LI

Assignees

铁科金化检测中心有限公司
北京中铁科新材料技术有限公司
中国铁道科学研究院集团有限公司金属及化学研究所
铁科金化科技有限公司
中国铁道科学研究院集团有限公司

Dates

Publication Date: 20260512
Application Date: 20260114

Claims (14)

1. The automatic wheel surface image acquisition and damage analysis system for the locomotive non-drop turning repair scene is characterized by comprising a mechanical module unit, an image acquisition unit, an algorithm analysis unit and a main control unit, wherein the units work cooperatively through physical connection, electrical connection and data connection: The mechanical module unit comprises a system A and a system B, wherein the system A is arranged on the front side of the non-falling turning lathe through a mechanical module fixing device, the system B is arranged on the rear side of the non-falling turning lathe through a mechanical module fixing device, each set of system comprises 2 sets of mechanical modules which respectively correspond to left and right wheels, the mechanical modules drive an X-axis linear sliding table, a Y-axis linear sliding table and a Z-axis linear sliding table through servo motors, and the shafts are connected through synchronous transmission devices and mutually supported through fixing brackets; The image acquisition unit is rigidly connected with the tail end of the mechanical module unit through a camera fixing support and comprises a 3-axis array camera (15), a 2-axis array camera (151), a 3-axis fixed focus lens (16), a 2-axis fixed focus lens (161), a 3-axis main light source (17), a 2-axis main light source (171), a 3-axis auxiliary light source (18) and a 2-axis auxiliary light source (181), wherein the linear array camera is in threaded connection with the fixed focus lens through an M42 interface, the main light source and the auxiliary light source are in bolt connection with the camera fixing support through a support, and the light source angle is manually adjusted through an auxiliary light source adjuster; The algorithm analysis unit is deployed in the industrial personal computer, is in communication connection with the image acquisition unit and the main control unit through an internal bus of the main control unit, receives image data and returns a damage identification result; The main control unit integrates a PLC control system and a Windows system, and realizes automatic operation of the whole system by cooperative control of a LAN interface, a COM interface and an Ethercat master station, a mechanical module unit, an image acquisition unit and an algorithm analysis unit.
2. The automated wheel surface image acquisition and damage analysis system for a locomotive non-drop turning scenario of claim 1, wherein each set of mechanical modules in system a and system B of the mechanical module unit is configured as a 2-axis or 3-axis structure according to a non-drop turning machine type: The 2-axis mechanical module comprises an X-axis linear sliding table and a Z-axis linear sliding table, wherein the X-axis linear sliding table is driven by an X-axis servo motor (91) and realizes transverse movement by an X-axis mechanical module (111), the Z-axis linear sliding table is driven by a Z-axis servo motor (61) and realizes longitudinal movement by a Z-axis mechanical module (71), and the X-axis and the Z-axis are connected by an X-axis and Z-axis fixed bracket (51); The 3-axis mechanical module comprises an X-axis, a Y-axis and a Z-axis linear sliding table, wherein the Y-axis linear sliding table is driven by a Y-axis servo motor (2) and vertically moves through a Y-axis mechanical module (4), the X-axis linear sliding table is driven by an X-axis servo motor (9) and horizontally moves through an X-axis mechanical module (11), the Z-axis linear sliding table is driven by a Z-axis servo motor (6) and longitudinally moves through a Z-axis mechanical module (7), the Y-axis and the Z-axis are connected through a Y-axis and Z-axis fixed bracket (5), and the Z-axis and the X-axis are connected through a Z-axis and X-axis fixed bracket (8); The positioning error of the mechanical module is less than or equal to 1.0mm, and the size ranges of the three wheels of 1250-1150mm, 1050-975mm and 1050-950mm are adapted.
3. The automatic wheel surface image acquisition and damage analysis system for the locomotive non-drop turning scene is characterized in that 220V power supply is adopted by a servo motor in a mechanical module unit, band-type brakes are not arranged on an X-axis servo motor and a Y-axis servo motor, each servo motor drives a sliding table to move through a synchronous transmission device, the synchronous transmission device comprises a synchronous belt wheel, a synchronous belt and a protective cover, and the mechanical module unit further comprises a balancing weight (13) used for balancing movement inertia.
4. The automatic wheel surface image acquisition and damage analysis system for the locomotive non-drop turning scene of claim 1 is characterized in that a linear array camera is of the type of sea Conway vision MV-CL042-91GM and is provided with a 4096×2 line CMOS sensor, a prime lens focal length is 40mm, a maximum aperture F2.8, a main light source size is 200×200mm, power is 60.6W, an auxiliary light source size is 150×100mm, power is 26.4W, and light source brightness is adjusted steplessly through a main control unit to inhibit reflection of the wheel metal surface.
5. The automatic wheel surface image acquisition and damage analysis system for the locomotive non-drop turning scene is characterized in that a damage identification algorithm based on a YOLOv n-SMC model is arranged in an algorithm analysis unit, an industrial control integrated machine is configured with an i5-12400 processor, a 32G memory and 128G SSD+2T SSD hard disk, and the algorithm analysis unit realizes identification of four types of damage such as cracks, stripping, oblique cracks and scrap iron scratches by processing images in parallel through multiple threads.
6. The automated wheel surface image acquisition and damage analysis system for locomotive non-drop turning situations of claim 1, wherein the main control unit expands external devices through various interfaces, including a DP interface for display connection, an HDMI interface for standby display, a USB interface for data transmission, ensuring system scalability and stability.
7. The automated wheel surface image acquisition and damage analysis method for a locomotive non-drop turning scenario of claim 1, comprising the steps of: The system starting and enabling step, namely a main control unit sends an enabling signal to a servo motor of a mechanical module unit to enable the servo motor to enter a work ready state and reset the mechanical module to an initial station; Step two, a mechanical module moving step, in which a main control unit selects a system A or a system B of the mechanical module unit to work according to the received locomotive model, the locomotive warehouse-in end position and the turning and repairing axis sequence signal, and issues a displacement instruction to control the mechanical module to drive an image acquisition unit to move to a preset image acquisition position; The linear array camera obtains shooting signals of the main control unit, starts scanning according to preset frequency, supplements light by matching with a high-brightness light source, realizes full-coverage image acquisition of the surface of the wheel, and transmits the acquired images to the main control unit in real time; Step four, equipment resetting, namely after the image acquisition is completed, the main control unit sends an disabling signal to the servo motor to reset the mechanical module to an initial station; Step five, the main control unit transmits the collected image to the algorithm analysis unit, and the image is preprocessed, multithreaded parallel detected, lesion detected and filtered, priority judged and visualized marked by the customized lesion recognition model, and the position, type and confidence of the lesion are output; Step six, outputting the result, namely receiving the identification result by the main control unit, displaying a visual image in a display screen of the industrial control integrated machine, triggering an alarm if the injury is detected, and storing the injury image and data.
8. The method of claim 7, wherein the mechanical die set movement in step two specifically comprises: The main control unit controls a system A or a system B of the mechanical module unit to work according to the type of the non-drop turning lathe and the model of the locomotive, wherein the system A is arranged on the front side of the turning lathe, and the system B is arranged on the rear side of the turning lathe; in the mechanical module moving step, the main control unit automatically selects the moving track and the shaft number of the mechanical module according to the type of the non-drop turning lathe and the detection model: For the 2-axis mechanical module, the image acquisition unit is driven to move transversely and longitudinally by the cooperative movement of the X-axis linear sliding table and the Z-axis linear sliding table, so as to adapt to the size range of the wheel; For the 3-axis mechanical module, the vertical movement freedom degree is increased through the linkage of the X-axis linear sliding table, the Y-axis linear sliding table and the Z-axis linear sliding table, so that the spatial interference is avoided; Accurately positioning the image acquisition unit to the depth coverage range of the surface of the wheel, wherein the adaptive wheel size range is 1250-1150mm, 1050-975mm and 1050-950mm; The motion trail of the mechanical module is automatically matched based on a built-in model parameter library, and the positioning error is less than or equal to 1.0mm.
9. The method of claim 7, wherein in the image capturing step, the line camera cooperates with a highlighting light source to optimize imaging quality: The linear array camera scans the surface of the rotating wheel line by line at the line frequency of 80kHz at the highest, so that motion blur is avoided; the main light source irradiates the tread of the wheel with a low included angle, the auxiliary light source obliquely irradiates the root of the wheel rim, and the metal reflection is inhibited by diffuse light; The fixed focus lens is linked with the Z-axis linear sliding table through a manual focusing ring, so that quick and accurate focusing is realized.
10. The method according to claim 9, wherein the image acquisition specifically comprises: The linear array camera adopts a sea Conway to look at the MV-CL042-91GM model, the highlight light source comprises a main light source and an auxiliary light source, the main light source is used for supplementing light to the tread of the wheel, the auxiliary light source is used for supplementing light to the root of the wheel rim, the brightness of the light source is adjusted steplessly by the main control unit, and the angle is adjusted by the auxiliary light source adjuster so as to inhibit the reflection of the surface of the wheel and improve the damage contrast; In the image acquisition process, the linear array camera is synchronous with the rotation of the wheels, so that motion blur is avoided, and imaging definition is ensured.
11. The method of claim 7, wherein the lesion identification and classification step comprises the sub-steps of: Image preprocessing, namely converting an acquired BMP format image into a JPG format with 1024x1250 resolution, and keeping the image quality at 95%; Multi-thread parallel detection, namely splitting an image to be detected into a plurality of batches according to each 10 images, and realizing multi-batch parallel detection through a 3-thread pool; The flaw detection and filtration adopts a differential confidence coefficient threshold value, the detection confidence coefficient threshold value of cracks and inclined cracks is 0.4, the detection confidence coefficient threshold value of stripping and scrap iron scratches is 0.3, and an overlapped frame filtration threshold value of 0.3 is set; The damage priority judgment comprises the steps of outputting the damage result of a single image according to the priority of peeling crack inclined crack scrap iron scratch; and outputting visual labels and results, namely configuring special labeling colors for four types of injuries, generating a detection result image and returning the result in a JSON format.
12. The method of claim 7, wherein the YOLOv n custom model used in the lesion identification and classification step is optimized by: introducing a C2f-CCFM trans-scale feature fusion module into a backbone network layer 2 and a backbone network layer 4, and enhancing trans-scale feature interaction; and optimizing a head network structure, and adding a P2/4 scale detection head to form a reciprocating characteristic fusion path.
13. The method of claim 7, further comprising the step of collaborative management of lesion data: The main control unit is used for storing the damage identification result and the locomotive warehouse entry end position and axis sequence information in a correlated manner, generating a detection report and uploading the detection report to the server; The system supports damage history data query and statistical analysis and is used for turning repair quality traceability and operation and maintenance decision support.
14. The method of claim 7, wherein the method implements full-process automated closed-loop control by a master control unit, comprising: the main control unit monitors the turning lathe state in real time and dynamically adjusts the image acquisition parameters; The algorithm analysis unit cooperates with the mechanical module unit to ensure seamless connection of image acquisition and damage identification.

Description

Automatic wheel surface image acquisition and damage analysis system for locomotive non-drop turning repair scene Technical Field The invention relates to the technical field of railway maintenance, in particular to an automatic wheel surface image acquisition and damage analysis system for a locomotive non-drop turning repair scene. Background In the field of rail transit locomotive operation and maintenance, the non-drop wheel turning lathe is core equipment for locomotive wheel maintenance, and the wheel turning repair is a necessary procedure for removing existing damage on the surface of a wheel and repairing the standard profile of the wheel so as to ensure the running safety of the locomotive. However, in the current quality detection link of wheel surface damage after turning repair, the quality detection link still stays on the traditional manual operation level, no matched automatic image acquisition and intelligent damage identification and classification equipment exists, and a significantly short plate exists for the control of turning repair quality. In the prior art, turning repair operators mostly adopt non-professional equipment such as handheld mobile phones and portable cameras to collect images on the surfaces of wheels, so that not only are core indexes such as image resolution and contrast uneven due to equipment performance difference, but also the collected images are disordered in style and fuzzy in key damage characteristics due to non-uniform shooting angles and illumination conditions, a standardized damage image database cannot be formed, and subsequent unified and accurate damage judgment is difficult to support, so that hidden danger is generated for tracing and evaluating turning repair quality. The inspection of the surface quality of the wheel after turning is completed completely by manpower, the wheel turning lathe is required to be matched with the gradual angle rotation of the wheel turning lathe, and operators can check the damage of key parts such as the tread, the rim and the like of the wheel one by one through naked eyes, so that the process is complicated, the time consumption is long, and the operation and maintenance efficiency of the whole wheel turning is greatly reduced. The artificial judgment of the injury has strong subjective limitation. The detection result excessively depends on the experience level of operators, so that the fine damage is extremely easy to cause missed judgment and misjudgment due to factors such as fatigue, insufficient experience and the like of the operators, effective control of turning quality cannot be formed, even the on-line running of wheels with injuries can be caused, and the running safety risk is caused. Meanwhile, the condition that the judgment of the injury by each unit is not uniform is unfavorable for the injury statistics in the field of all railway engineering. At present, in the field of turning maintenance of rail transit locomotives without falling wheels, a set of special system capable of realizing standardized image acquisition and intelligent damage analysis is needed, the technical blank of automatic detection of wheel quality after turning maintenance is filled, and the overall operation and maintenance level and safety guarantee capability of turning maintenance procedures are improved. Disclosure of Invention In order to overcome the existing defects, the invention provides an automatic wheel surface image acquisition and damage analysis system for a locomotive non-drop turning scene. An automatic wheel surface image acquisition and damage analysis system for a locomotive non-drop turning repair scene comprises a mechanical module unit, an image acquisition unit, an algorithm analysis unit and a main control unit, wherein the units work cooperatively through physical connection, electrical connection and data connection: The mechanical module unit comprises a system A and a system B, wherein the system A is arranged on the front side of the non-falling turning lathe through a mechanical module fixing device, the system B is arranged on the rear side of the non-falling turning lathe through a mechanical module fixing device, each set of system comprises 2 sets of mechanical modules which respectively correspond to left and right wheels, the mechanical modules drive an X-axis linear sliding table, a Y-axis linear sliding table and a Z-axis linear sliding table through servo motors, and the shafts are connected through synchronous transmission devices and mutually supported through fixing brackets; The image acquisition unit is in rigid connection with the tail end of the mechanical module unit through a camera fixing support and comprises a 3-axis array camera, a 2-axis array camera, a 3-axis fixed focus lens, a 2-axis fixed focus lens, a 3-axis main light source, a 2-axis main light source, a 3-axis auxiliary light source and a 2-axis auxiliary light source, wherein the linear array camera is in threaded c