Abrasion degree detection system and method based on image recognition

Abstract

The invention relates to the technical field of aircraft landing gear wear degree detection, and provides a wear degree detection system and method based on image recognition, and the system comprises an image collection unit which is used for obtaining image data of a target tire tread in a detection state, the image acquisition unit at least comprises a structured light projection module and a high-resolution color camera, the structured light projection module is used for projecting a coded light pattern to a tread, and the image processing and analyzing unit is in communication connection with the image acquisition unit and is used for receiving image data. Through combination of structured light three-dimensional reconstruction and high-resolution color imaging, the key safety parameter, namely the depth of a pattern groove, can be accurately measured, two-dimensional texture defects such as tread cuts and cracks can be synchronously identified, positioned and quantified, the fusion decision-making unit carries out correlation analysis on the two types of information, and the judgment result is more accurate. And a tire health state portrait which is more comprehensive and more stereoscopic than single depth measurement is provided.

Description

Technical Field

[0001] This invention relates to the field of aircraft landing gear wear detection technology, specifically to a wear detection system and method based on image recognition. Background Technology

[0002] As a critical component of the landing gear, aircraft tires directly bear the impact of landing, taxiing loads, and braking forces, and their wear condition directly affects flight safety and operational economy. Currently, the inspection of aircraft tire wear mainly relies on visual inspection by maintenance personnel on the ground and tread depth measurement using simple measuring tools (such as depth gauges). This traditional method largely depends on the experience and sense of responsibility of the personnel.

[0003] As the aviation industry increasingly demands higher operational efficiency and safety margins, existing manual inspection methods face several objective challenges. First, inspection efficiency and consistency need improvement. Within the short time windows of rapid flight transit, meticulously inspecting multiple tires presents time pressure and may lead to missed inspections. Second, for surface defects such as minor cuts and localized cracks on the tire tread, manual visual inspection struggles to provide standardized and quantitative assessments, resulting in highly subjective judgments. Furthermore, current measurement methods primarily focus on the single dimension of tread groove depth, lacking the ability to comprehensively correlate wear depth with surface texture damage, thus failing to fully assess the overall health of the tire. Therefore, a technological means is needed to achieve rapid, objective, and quantitative inspection. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an image recognition-based wear detection system and method, which solves the problems of reliance on manual labor, low efficiency, and limited evaluation dimensions in existing aircraft tire wear detection.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a wear detection system based on image recognition, comprising: An image acquisition unit is used to acquire image data of the target tire tread in a detection state. The image acquisition unit includes at least a structured light projection module and a high-resolution color camera. The structured light projection module is used to project an coded light pattern onto the tire tread. The image processing and analysis unit is communicatively connected to the image acquisition unit and is used to receive the image data and perform the following processing: calculate the three-dimensional point cloud data of the tire tread based on the coded light pattern, and extract two-dimensional texture features from the color camera image; The wear measurement calculation unit is connected to the image processing and analysis unit and is used to calculate the depth distribution and remaining depth of the tire tread grooves based on the three-dimensional point cloud data, and to identify and quantify tread cuts, cracks and foreign object embedding defects based on the two-dimensional texture features. The fusion decision unit is connected to the image processing and analysis unit and the wear measurement calculation unit, respectively, and is used to fuse the three-dimensional depth information and two-dimensional defect information to generate a comprehensive wear assessment report and maintenance recommendations.

[0006] Preferably, it also includes an offline calibration unit, which includes a standard stepped wear block with known accuracy, used to perform joint calibration of the image acquisition unit in a non-detection state to correct scale errors and camera distortion in 3D reconstruction.

[0007] Preferably, the offline calibration unit and the image processing and analysis unit are integrated. The calibration process is completed by automatically identifying specific marker points on the standard stepped wear block and calculating the reprojection error. The calibration parameters are stored and used for online correction of all subsequent on-machine inspection tasks.

[0008] Preferably, the image processing and analysis unit uses a multi-frequency heterodyne method combining phase shifting and Gray code to perform 3D reconstruction. This method is used to solve the phase unwrapping problem in high-reflectivity areas and tread edges of the tire tread. Its core formula involves absolute phase. Calculation: ,in The grayscale values ​​of the phase-shifted fringe image. The number of stripes is obtained by decoding Gray code.

[0009] Preferably, it also includes an intelligent sensing triggering unit, which includes a proximity sensor and a wheel hub speed sensor. When the proximity sensor detects that the tire has entered a preset detection area and the wheel hub speed sensor determines that the tire is stationary or rotating at low speed, the image acquisition unit is automatically triggered to work.

[0010] A wear detection method based on image recognition includes the following steps: Step 1, System Initialization and Calibration: Call the offline calibration unit and use the standard stepped wear block to perform joint calibration of the structured light module and color camera, and obtain and store high-precision internal and external system parameters; Step 2, Detection Condition Judgment and Triggering: The intelligent sensing triggering unit monitors the tire position and status, and automatically issues an image acquisition command when preset conditions are met; Step 3: Synchronous acquisition of multimodal images: The structured light projection module projects an coded light pattern onto the tread of a stationary or slowly rotating tire, while the color camera simultaneously acquires the modulated tread image. Step 4: Image Data Processing and Analysis: Decode the acquired image, calculate the absolute phase of each point on the tread and convert it into a 3D point cloud. At the same time, segment the tread region from the color image and extract texture features. Step 5: Quantitative calculation of wear features: In the 3D point cloud, locate the tread grooves along the tire circumference and axial direction, calculate the depth curve of the groove cross section and the minimum remaining depth, and in the 2D image, use the trained deep learning model to identify and select various surface defects, and calculate their area and length quantitative indicators. Step Six: Information Fusion and Report Generation: Establish the mapping relationship between three-dimensional depth data and two-dimensional defect information. Based on the preset weighted fusion algorithm, output a comprehensive evaluation report containing quantitative wear data, defect distribution map and specific maintenance suggestions.

[0011] Preferably, in step one, the joint calibration process includes: acquiring multiple sets of images of the standard wear block from different angles, and simultaneously optimizing the parameters of the structured light projection module and the color camera using the bundle adjustment method to ensure that the three-dimensional measurement results are consistent with the physical scale.

[0012] Preferably, in step four, the calculation of the three-dimensional point cloud further includes preprocessing the point cloud by denoising, filtering and hole repair, and using the iterative nearest point algorithm to register the current point cloud with a point cloud template of a standard new tire to eliminate measurement deviations caused by the tire mounting posture.

[0013] Preferably, in step five, the deep learning model is an improved U-Net network that introduces an attention mechanism in the encoder part to enhance the ability to extract features of minor cuts and fine cracks.

[0014] Preferably, in step six, the fusion decision algorithm is as follows: set a safety threshold pattern depth and a defect comprehensive index; when the measured minimum remaining depth is less than the pattern depth, the reporting priority is "immediate replacement"; when the measured minimum remaining depth is greater than or equal to the pattern depth, but the defect index is greater than the defect comprehensive index, the reporting priority is "close monitoring and shortening the inspection cycle"; otherwise, the report is "normal status".

[0015] This invention provides a wear detection system and method based on image recognition. It has the following advantages: 1. By combining structured light 3D reconstruction with high-resolution color imaging, this invention can not only accurately measure the key safety parameter of tread groove depth, but also simultaneously identify, locate and quantify two-dimensional texture defects such as tread cuts and cracks. The fusion decision unit analyzes these two types of information together, providing a more comprehensive and three-dimensional picture of tire health status than a single depth measurement, which helps to discover potential risks.

[0016] 2. This invention introduces a dual-mode design that combines "offline high-precision calibration" with "on-machine rapid measurement." During non-working hours, the system can complete precise calibration using standard parts to eliminate system errors. During working hours, it can perform rapid and automated online testing, balancing measurement accuracy and operational efficiency. The intelligent sensing trigger unit avoids invalid data acquisition and further optimizes the workflow. This method reduces the reliance on the professional skills of on-site operators and improves the standardization level and overall efficiency of testing operations.

[0017] 3. This invention automatically transforms quantified multi-source data into maintenance recommendations with clear priority ranking through preset fusion decision algorithms and rules. This provides aircraft maintenance personnel with direct and operable decision support, helps to promote the transition of maintenance work from "periodic inspection" to "condition-based predictive maintenance", and improves the accuracy and initiative of maintenance work. Attached Figure Description

[0018] Figure 1 This is a system diagram of the present invention; Figure 2 This is a flowchart of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example: Please see the appendix Figure 1 and attached Figure 2 This invention provides an image recognition-based wear detection system, comprising: The image acquisition unit is used to acquire image data of the target tire tread in the detection state. The image acquisition unit includes at least one DLP4500 digital micromirror device structured light projection module and one Baslerac A4112-20um high-resolution color camera. The structured light projection module uses a blue LED light source with a wavelength of 450nm to project a phase-shifted stripe pattern with a period of 30 pixels and a 5-bit Gray code pattern onto the tire tread. The color camera has a resolution of 4096×3000 pixels, is equipped with a 35mm fixed-focus lens, is installed at a distance of about 500mm from the tire tread, and forms a 15-degree angle with the optical axis of the structured light projection module. The acquisition is synchronized through a hardware synchronization line. The image processing and analysis unit is communicatively connected to the image acquisition unit. It is used to receive image data and perform the following processing: calculate the three-dimensional point cloud data of the tire tread based on the coded light pattern, and extract two-dimensional texture features from the color camera image. The wear measurement calculation unit, connected to the image processing and analysis unit, is used to calculate the depth distribution and remaining depth of tire tread grooves based on three-dimensional point cloud data, and to identify and quantify tread cuts, cracks and foreign object embedding defects based on two-dimensional texture features. The fusion decision unit is connected to the image processing and analysis unit and the wear measurement calculation unit, respectively, and is used to fuse three-dimensional depth information and two-dimensional defect information to generate a comprehensive wear assessment report and maintenance recommendations.

[0021] It also includes an offline calibration unit, which consists of a standard stepped wear block with known accuracy. The step heights of the wear block are 2mm, 4mm, and 6mm, with an accuracy of ±0.01mm. The surface of the wear block is covered with AprilTag visual tags, used for joint calibration of the image acquisition unit in non-detection mode to correct scale errors and camera distortion in 3D reconstruction. The offline calibration unit is integrated with the image processing and analysis unit. The calibration process is completed by automatically identifying the AprilTag tags on the standard stepped wear block and calculating the reprojection error. Calibration is considered successful when the average reprojection error is less than 0.2 pixels. The calibration parameters are stored and used for online correction in all subsequent on-machine detection tasks. The image processing and analysis unit uses a multi-frequency heterodyne method combining phase shifting and Gray code to perform 3D reconstruction, designed to solve the phase unwrapping problem in high-reflectivity areas and tread edges of the tire tread. Its core formula involves absolute phase. Calculation: ,in The grayscale values ​​of the phase-shifted fringe image. The stripe levels obtained by Gray code decoding also include an intelligent sensing trigger unit, which includes a SICKUM30-213111 ultrasonic proximity sensor and a Honeywell RTY series Hall effect wheel speed sensor. When the proximity sensor detects that the tire has entered the preset detection area and the wheel speed sensor determines that the tire is stationary or rotating at low speed, the image acquisition unit is automatically triggered to work.

[0022] A wear detection method based on image recognition includes the following steps: Step 1: System Initialization and Calibration: The offline calibration unit is invoked, and the structured light module and color camera are jointly calibrated using a standard stepped wear block. High-precision internal and external parameters of the system are acquired and stored. Specifically, multiple sets of images of the standard wear block are acquired from at least 10 different angles. The internal parameters (focal length, principal point, distortion coefficient) and external parameters (rotation matrix and translation vector) of the structured light projection module and color camera are simultaneously optimized using the bundle adjustment method to ensure that the three-dimensional measurement results are consistent with the physical scale. Finally, the three-dimensional measurement error is controlled within ±0.05mm. Step 2, Detection Condition Judgment and Triggering: The intelligent sensing triggering unit monitors the tire position and status, and automatically issues an image acquisition command when the preset conditions are met; Step 3: Synchronous acquisition of multimodal images: The structured light projection module projects an coded light pattern onto the tread of a stationary or slowly rotating tire, while the color camera simultaneously acquires the modulated tread image. Step 4, Image Data Processing and Analysis: The acquired image is decoded, the absolute phase of each point on the tread is calculated and converted into a 3D point cloud. Simultaneously, the tread region is segmented from the color image using a threshold segmentation method based on the HSV color space, and LBP texture features are extracted. The calculation of the 3D point cloud further includes statistical filtering (removing points with a mean standard deviation greater than 2) and radius filtering for noise reduction. Moving least squares is used for smoothing and hole repair. Subsequently, the iterative nearest point algorithm is used to register the current point cloud with a point cloud template of a standard new tire to eliminate measurement deviations caused by tire mounting posture. The registration error RMS value is less than 0.1mm. Step 5: Wear Feature Quantification Calculation: In the 3D point cloud, locate the tread grooves along the tire's circumference and axial direction, calculate the depth curve and minimum remaining depth of the groove cross-section, and use the trained deep learning model to identify and select various surface defects in the 2D image. The deep learning model is an improved U-Net network, which introduces channel attention modules in the 3rd and 4th layers of the encoder to enhance the extraction of micro-cuts and fine cracks. It can identify and calculate quantitative indicators such as the area and maximum length of defects. Step Six: Information Fusion and Report Generation: Establish the mapping relationship between three-dimensional depth data and two-dimensional defect information. Based on the preset weighted fusion algorithm, output a comprehensive evaluation report containing quantitative wear data, defect distribution map and specific maintenance suggestions.

[0023] In step one, the joint calibration process includes: acquiring multiple sets of images of the standard wear block from different angles, and simultaneously optimizing the parameters of the structured light projection module and the color camera using the bundle adjustment method to ensure that the three-dimensional measurement results are consistent with the physical scale.

[0024] In step four, the calculation of the 3D point cloud further includes preprocessing the point cloud by denoising, filtering and hole repair, and using the iterative nearest point algorithm to register the current point cloud with a point cloud template of a standard new tire to eliminate measurement deviations caused by the tire mounting posture.

[0025] In step five, the deep learning model is an improved U-Net network, which introduces an attention mechanism in the encoder part to enhance the ability to extract features of minor cuts and fine cracks.

[0026] In step six, the fusion decision algorithm is as follows: set a safety threshold for pattern depth and a comprehensive defect index. When the measured minimum remaining depth is less than the pattern depth, the reporting priority is "immediate replacement". When the measured minimum remaining depth is greater than or equal to the pattern depth, but the defect index is greater than the comprehensive defect index, the reporting priority is "close monitoring and shortening the inspection cycle". Otherwise, the report is "normal status".

[0027] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A wear detection system based on image recognition, characterized in that, include: An image acquisition unit is used to acquire image data of the target tire tread in a detection state. The image acquisition unit includes at least a structured light projection module and a high-resolution color camera. The structured light projection module is used to project an coded light pattern onto the tire tread. The image processing and analysis unit is communicatively connected to the image acquisition unit and is used to receive the image data and perform the following processing: calculate the three-dimensional point cloud data of the tire tread based on the coded light pattern, and extract two-dimensional texture features from the color camera image; The wear measurement calculation unit is connected to the image processing and analysis unit and is used to calculate the depth distribution and remaining depth of the tire tread grooves based on the three-dimensional point cloud data, and to identify and quantify tread cuts, cracks and foreign object embedding defects based on the two-dimensional texture features. The fusion decision unit is connected to the image processing and analysis unit and the wear measurement calculation unit, respectively, and is used to fuse the three-dimensional depth information and two-dimensional defect information to generate a comprehensive wear assessment report and maintenance recommendations.

2. The wear detection system based on image recognition according to claim 1, characterized in that, It also includes an offline calibration unit, which includes a standard stepped wear block with known accuracy, used to perform joint calibration of the image acquisition unit in a non-detection state to correct scale errors and camera distortion in 3D reconstruction.

3. The wear detection system based on image recognition according to claim 2, characterized in that, The offline calibration unit and the image processing and analysis unit are integrated. The calibration process is completed by automatically identifying specific marker points on the standard stepped wear block and calculating the reprojection error. The calibration parameters are stored and used for online correction of all subsequent on-machine inspection tasks.

4. The wear detection system based on image recognition according to claim 1, characterized in that, The image processing and analysis unit employs a multi-frequency heterodyne method combining phase shifting and Gray code to perform 3D reconstruction. This method is used to solve the phase unwrapping problem in highly reflective areas and tread edges of tires. Its core formula involves absolute phase. Calculation: ,in The grayscale values ​​of the phase-shifted fringe image. The number of stripes is obtained by decoding Gray code.

5. The wear detection system based on image recognition according to claim 1, characterized in that, It also includes an intelligent sensing triggering unit, which includes a proximity sensor and a wheel hub speed sensor. When the proximity sensor detects that the tire has entered a preset detection area and the wheel hub speed sensor determines that the tire is stationary or rotating at low speed, the image acquisition unit is automatically triggered to work.

6. A wear detection method based on image recognition, using an image recognition-based wear detection system as described in any one of claims 1-5, characterized in that, Includes the following steps: Step 1, System Initialization and Calibration: Call the offline calibration unit and use the standard stepped wear block to perform joint calibration of the structured light module and color camera, and obtain and store high-precision internal and external system parameters; Step 2, Detection Condition Judgment and Triggering: The intelligent sensing triggering unit monitors the tire position and status, and automatically issues an image acquisition command when preset conditions are met; Step 3: Synchronous acquisition of multimodal images: The structured light projection module projects an coded light pattern onto the tread of a stationary or slowly rotating tire, while the color camera simultaneously acquires the modulated tread image. Step 4: Image Data Processing and Analysis: Decode the acquired image, calculate the absolute phase of each point on the tread and convert it into a 3D point cloud. At the same time, segment the tread region from the color image and extract texture features. Step 5: Quantitative calculation of wear features: In the 3D point cloud, locate the tread grooves along the tire circumference and axial direction, calculate the depth curve of the groove cross section and the minimum remaining depth, and in the 2D image, use the trained deep learning model to identify and select various surface defects, and calculate their area and length quantitative indicators. Step Six: Information Fusion and Report Generation: Establish the mapping relationship between three-dimensional depth data and two-dimensional defect information. Based on the preset weighted fusion algorithm, output a comprehensive evaluation report containing quantitative wear data, defect distribution map and specific maintenance suggestions.

7. The wear detection method based on image recognition according to claim 6, characterized in that, In step one, the joint calibration process includes: acquiring multiple sets of images of the standard wear block from different angles, and simultaneously optimizing the parameters of the structured light projection module and the color camera using the bundle adjustment method to ensure that the three-dimensional measurement results are consistent with the physical scale.

8. The wear detection method based on image recognition according to claim 6, characterized in that, In step four, the calculation of the three-dimensional point cloud further includes preprocessing the point cloud by denoising, filtering and hole repair, and using the iterative nearest point algorithm to register the current point cloud with a point cloud template of a standard new tire to eliminate measurement deviations caused by the tire mounting posture.

9. The wear detection method based on image recognition according to claim 6, characterized in that, In step five, the deep learning model is an improved U-Net network, which introduces an attention mechanism in the encoder part to enhance the ability to extract features of minor cuts and fine cracks.

10. The wear detection method based on image recognition according to claim 6, characterized in that, In step six, the fusion decision algorithm is as follows: set a safety threshold for pattern depth and a comprehensive defect index. When the measured minimum remaining depth is less than the pattern depth, the reporting priority is "immediate replacement". When the measured minimum remaining depth is greater than or equal to the pattern depth, but the defect index is greater than the comprehensive defect index, the reporting priority is "close monitoring and shortening the inspection cycle". Otherwise, the report is "normal status".

Images